Technical Report LBNL-2001090.Cite as: Felix Bunning, Michael Wetter, Marcus Fuchs and Dirk Muller.

Bidirectional low temperature district energy systems with agent-based control: Performance comparison andoperation optimization.

Applied Energy, Vol. 209, p. 502–515. 2017. DOI:10.1016/j.apenergy.2017.10.072

Bidirectional low temperature district energy systems with

agent-based control: Performance comparison and operation

optimization

Felix Bunning1,2, Michael Wetter

1, Marcus Fuchs

2, and Dirk Muller

2

11Lawrence Berkeley National Laboratory, Berkeley, USA12RWTH Aachen University, E.ON Energy Research Center, Institute for Energy Efficient

Buildings and Indoor Climate, Aachen, Germany

January 3, 2018

Abstract

Bidirectional low temperature networks are a novelconcept that promises more efficient heating andcooling of buildings. Early research shows theoret-ical benefits in terms of exergy efficiency over othertechnologies. Pilot projects indicate that the conceptdelivers good performance if heating and cooling de-mands are diverse. However, the operation of thesenetworks is not yet optimized and there is no quan-tification of the benefits over other technologies invarious scenarios. Moreover, there is a lack of un-derstanding of how to integrate and control multi-ple distributed heat and cold sources in such net-works. Therefore, this paper develops a control con-cept based on a temperature set point optimizationand agent-based control which allows the modular in-tegration of an arbitrary number of sources and con-sumers. Afterwards, the concept is applied to twoscenarios representing neighborhoods in San Fran-cisco and Cologne with different heating and coolingdemands and boundary conditions. The performanceof the system is then compared to other state-of-the-art heating and cooling solutions using dynamic simu-lations with Modelica. The results show that bidirec-tional low temperature networks without optimiza-

tion produce 26% less emissions in the San Franciscoscenario and 63% in the Cologne scenario in com-parison to the other heating and cooling solutions.Savings of energy costs are 46% and 27%, and reduc-tions of primary energy consumption 52% and 72%,respectively. The presented operation optimizationleads to electricity use reductions of 13% and 41%when compared to networks with free-floating tem-perature control and the results indicate further po-tential for improvement. The study demonstrates theadvantage of low temperature networks in differentsituations and introduces a control concept that isextendable for real implementation.

1 Introduction

1.1 Background and motivation

District heating systems have been used for spaceheating and domestic hot tap water since the 1880s.Since then the efficiency of these systems has beenimproved continuously and four different generationsof district heating systems can be distinguished, dif-fering in heat carrier, temperature levels, circulationsystems and substations [30].With respect to the four different generations of

1 INTRODUCTION 2

district heating systems, bidirectional low tempera-ture networks (LTN) can be viewed as a fifth gen-eration district heating and cooling system, as sug-gested by the European Commission [52]. They con-stitute an approach to further increase energy effi-ciency for heating and cooling of buildings by fur-ther lowering the fluid temperature in the networksto ambient temperature levels. The networks consistof two pipes. The warmer pipe has temperatures be-tween 12◦C and 20◦C, while the cold pipe has 8◦C to16◦C. Buildings equipped with heat pumps, chillersor direct cooling and individual circulation pumpsare connected to both lines. In the case of a heatingdemand, the circulation pump of the building with-draws water from the warm line, uses it in a heatpump to reach temperatures suitable for space heat-ing, and then discharges the cooled water to the coldline. In case of a cooling demand, the system worksin the other direction. Depending on the heating andcooling demands of the connected buildings, the fluidflow in the network can change direction. The totaldifference between heating and cooling flows needsto be balanced by external sources, e.g. central heatpumps and chillers, solar thermal plants or seasonalstorage facilities.The concept is in the early stages of research and

development. There is theoretical proof of the con-cept based on thermodynamic analysis [40]. Further-more, there are at least eight demonstration projects(unidirectional and bidirectional) in operation or un-der construction (e.g [48, 27, 9]). Research regardingthe optimal design in terms of diversity of cooling andheating loads has also been conducted lately [64].Despite the previous efforts, there is a significant

need for further research in multiple areas. It appearsthat no publications about control and operation op-timization of LTN are available. There is proof thatLTN are beneficial in terms of exergy efficiency, butthe full potential has not been exploited yet. More-over, there is a lack of understanding of how to in-tegrate and coordinate multiple source networks inwhich individual so called ”prosumers” can partici-pate in heat and cold supply. Furthermore, there isonly limited insight into the potential advantages abidirectional LTN has over other technologies in dif-ferent scenarios. To address these open questions is a

crucial prerequisite for the future of sustainable heat-ing and cooling systems through LTN technology.Therefore, this study introduces, to the authors’

knowledge, the first approach to operation optimiza-tion of low temperature networks. The approach con-sists of an optimization of the network temperature,which is the main influence on energy costs of suchnetworks, and a control concept based on a market-based multi-agent system. With proper multi-agentcontrol the electricity consumption of the heat pumpsand chillers in the network can be lowered. Further-more, the agent system allows the integration of mul-tiple heat and cold sources and energy storages intothe network, which makes it a practical approach forsmart-grid-like decentral networks. The approach isverified with the help of dynamic simulations basedon Modelica, for which the authors implemented thenecessary models [60, 31] and the agent-based controlsystem [2]. The case study examines two district sce-narios, one of which is situated in the USA and onein Germany, and compares low temperature networksto other heating and cooling options. The results al-low an insight into the behaviour of low temperaturenetworks, present a proof of concept for the controlapproach and provide a quantification of energy costsavings through the use of low temperature networksinstead of conventional heating and cooling technolo-gies.

1.2 Previous work on low tempera-ture networks

There is only a small number of peer-reviewed scien-tific work published in the field of bidirectional LTN.To give a complete overview of the topic, also reportson pilot projects by construction companies and othernon-scientific sources are cited in the following.One of the earlier examples of an LTN as described

in this study was introduced in [10] as a plan to createa district heating and cooling system for a developingarea in Visp, Switzerland. The unidirectional griduses waste heat from a sewage of a chemical plantas a heat source and de-centralized heat pumps areused in the individual residential buildings to lift thetemperatures sufficiently high to operate floor heat-ing systems. [27] presents an evaluation of the system

1 INTRODUCTION 3

installed in Visp. Completed in 2008, it has a totalthermal capacity in heating of 3.6 MW and is com-petitive with other technologies in terms of cost ofoperation at a domestic fuel oil price below $0.9/lt.In [48], an LTN for heating and cooling of parts ofthe campus of ETH Zurich is proposed. In order toreach the goals for a 2000-Watt-society [45], besidesother measures, plans for a grid consisting of mul-tiple ground heat storages, a geothermal field andde-central heat pumps are presented. Details on theprogress can be found in [9].Besides introducing the bidirectional concept, [46]makes several contributions to the topic of LTN: theauthor highlights the advantage of LTN in urban ar-eas, where it is difficult to install heat pumps with in-dividual ground or groundwater heat exchangers forresidential buildings. Moreover, heat and cold de-mand should be of a similar size in the ideal case.The role for the operator of the system is to keepthe temperatures at a sufficient level for cooling andheating and to balance the energy flows over the yearin all storage facilities. It is further pointed out thatLTN are modular and can be started as small projectsthat can be scaled up later on. A measure to furtherdecrease operation costs could be the provision of gridservices, as the used heat pumps constitute groups ofhigh electrical capacities.In [47], Sulzer examines the possibility of integrat-ing solar thermal panels into LTN. The author statesthat integration of such systems is favourable in thecase of seasonal heat storage as the demand for heatis higher in winter, but the production from solarthermal panels is high in summer. In [49], Sulzer etal. introduce different typologies of grids in termsof number of pipes between one and four. Whilethe previously introduced bidirectional networks op-erate with two pipes at different temperature levels,third and fourth pipes with higher or lower tempera-tures are possible in order to use direct floor heatingwithout a previous temperature lift via a heat pump,for example. Moreover, Sulzer now distinguishes be-tween unidirectional and bidirectional grids in termsof mass flow and energy flow. A grid can be unidi-rectional in mass flow but bidirectional in terms ofenergy flow (e.g. in the case of Visp). Additionally,he presents several problems and unanswered ques-

tions related to the networks: So far, technologicalexpertise is limited to a very small group of people.There are no standardized ways to calculate the costof such a network for planners who were not involvedin the pilot projects. Furthermore, the operation ofLTN has not been optimized yet.In [32], two different types of ownership for LTN aresuggested. Bigger LTN could be owned by a gridoperator, similarly to electrical grids or common dis-trict heating grids. Smaller grids could be ownedand operated by the real estate owner. An exampleof a real-estate owned system in Chur, Switzerland,is given. A comparison between investment cost fora centralized heat pump system and an LTN is madeand the LTN is found favourable. Furthermore, acomparison between yearly energy costs for the LTNin Chur and the local heating system installed beforeis made. A reduction of 62% of the annual energycosts could be observed.Although focusing on heating grids with higher sup-ply temperatures, Li et al. [29] point out thatthe development of future district heating grids willmove away from hierarchical, fossil-based, large-scalestructures towards future decentralized, multiple re-newable and waste-heat-dominated small structures.Moreover, they postulate the idea that individual”prosumers” in the network who have the capacityto produce surplus heat from building installed so-lar collector, heat pump, micro-CHP and individualthermal storage, should be able to participate in thefuture grids. Pietra et al. [5] investigated the ben-efit of a grid connection for owners of solar thermalpanels. The bidirectional grid operates as a virtualheat storage for the prosumer with the effect thatthe collectors can be operated throughout the wholesummer instead of being stopped because of low heatdemand during that period. Results show that thesolar collector is able to supply more than 100% ofthe total yearly heat demand for an apartment whenconnected to the grid in comparison to 27% before,which leads to an operational cost reduction of 60%.Gautschi [9] presents long-term experience with LTN.According to the building company, which built threeLTN in the past, carbon emissions can be lowered by70 to 80% and primary energy demand by 30 to 50%can be achieved. However, a reference technology

1 INTRODUCTION 4

for these results is not given in the source. Further-more, seasonal coefficients of performance from thesystem at ETH Zurich are reported with 7 for heatingand 26.5 for space cooling. Two additional examplesof LTN are presented. The system ”Richti-Areal inWallisellen” conditions offices and residential build-ings with the help of 220 borehole heat exchangerswith a depth of 220 m. The grid ”Familienheim-Genossenschaft Zurich” uses wasteheat from a Swiss-com data-center. Further heat producers are goingto be added in the future.Sulzer et al. [50] introduce a rapid-prototyping testbed, which allows to investigate the behaviour andcontrol strategies of LTN. The system is designed as ahardware-in-the-loop test bed, which means that theintegration of simulation-based methods is possible.First results of the test bed show that a ring-typologyis favourable and could reduce pump energy by morethan 50%. Furthermore, it is pointed out that a de-central pump design is more energy-efficient than acentralized one.Vetterli et al. [57] show monitoring data from an LTNin operation in comparison to previous simulations ofthe same grid in the planning phase. The results showthat the heating demand was underestimated and thecooling demand was overestimated, which leads to adecrease of temperature in the ground energy storage.Also the measured coefficient of performance (COP)is lower and circulation pump electricity consumptionhigher than predicted. The false predictions are as-cribed to different user behaviour in terms of indoortemperature and ventilation (open windows) than ex-pected. By adapting the model to the different userbehaviour, reasonable accordance between measure-ment and simulation could be achieved.In [16], the government of the canton St. Gallen inSwitzerland presents financial measures to facilitatethe building of LTN. The measures include the fi-nancial support for 5% of the building costs of thenetworks and a provision of $ 80 to 120 per MWh en-ergy delivered during the first two years of operation.The government explicitly forbids the generation ofheat with combined heat and power units.Henchoz et al. [12] compared LTN to two differentdistrict heating and cooling concepts based on the useof latent heat with refrigerants (CO2 and R1234yf)

as the heat carrying fluids. Simulation results showthat none of the investigated grids has significant ad-vantages in terms of energy efficiency over the othergrids. It is also pointed out that no significant dif-ferences in terms of investment costs were present.The decision for one particular technology has there-fore to be made based on soft characteristics, such ascompactness or safety.Summermatter et al. [51] introduce an LTN for avillage in the Alps of Switzerland. Two possibilities,one based on the integration of solar thermal pan-els and one based on the integration of photovoltaic(PV) panels, are compared. The option featuring PVpanels is found to be more cost efficient. The systemuses the electricity from the PV panels to operateair-water heat pumps during the summer to heat upwater. The heat is stored in a ground heat storageand can then be used with water-water heat pumpsduring the winter months. Moreover, the authorsplan to offer grid services to the electricity grid inorder to reduce costs. The grid will continuously beextended until the whole village is integrated.Zach [63] presents a concept for an LTN in the dis-trict ”Nordbahnhof” in Wien with 13,000 residentsand 5,000 workplaces. The grid consists of one centralLTN with a ground heat storage and five de-centralgrids connected to it. Photovoltaic thermal hybridsolar collectors are suggested as renewable heat andelectricity sources after an annualized analysis of thesystem costs.Zarin Pass et al. [64] explore the question of whenand why bidirectional systems outperform their in-dividual counterparts. The thermodynamic perfor-mance of bidirectional LTN is analyzed and a di-versity criterion is developed to understand when itmay be a more energy-efficient alternative to mod-ern, high-efficiency individual-building systems. Thiscriterion is then applied to standardized referencebuilding hourly load profiles in three cities to lookfor promising building diversities. It is found that abidirectional system has benefits when the ratio ofheating to cooling loads on average is at least 1 to5,7 or vice versa.Research in modelling of LTN includes the follow-

ing: Krauchi et al. [22, 21] present a model in theobject-oriented IDA-ICE simulation software. It uses

1 INTRODUCTION 5

different objects for the components heat consumer,heat supplier, earth heat storage, pipes and pumps.Heat consumers and suppliers are modelled with thehelp of time series of demand and supply. Based onthe desired temperature and the temperature of thegrid, COPs for the heat pumps are calculated. Thereare models for the ground energy storage availablethat do not describe any temperature distribution inthe ground, but act as a kind of fully mixed storage.The pipes are also modelled as fully mixed fluid vol-umes, but compute pressure losses. The pumps cal-culate mechanical and electrical power to compensatethe pressure loss within the LTN. The model allowsbidirectional flow. However, this feature is imple-mented by duplicating parts of the model for eachdirection of flow and activating or deactivating therelevant parts based on the current state of the sys-tem.In [23] the above named model components are aggre-gated to a planned actual system. The results showgood agreement with results from the planning phaseof the grid. The authors therefore propose to use themodels for the planning of LTN. Schluck et al. [40]use the described models in order to compare unidi-rectional and bidirectional grids. The results favourthe bidirectional grid in terms of exergy efficiency andoperating costs.Prasanna et al. [35] present a model of parts of theSuurstoffi LTN [57] formulated as a mixed integerlinear problem. The model is calibrated with moni-toring data from parts of the grid. An optimizationalgorithm is then used to identify possible improve-ments of the system operation in order to reducethe carbon footprint of the network. Furthermore,possible scenarios regarding the extension of the sys-tem with components like additional photovoltaic orphotovoltaic thermal hybrid solar collectors, batterystorage and air-water heat pumps are simulated andcompared.The Modelica Buildings library [60] contains modelsto simulate LTN, including a water heat exchanger,central heat pumps and consumer substations. Otherincluded models such as water storage tanks and so-lar thermal panels can also be used in an LTN districtmodel. The models allow bidirectional flow.Heissler et al. [11] present a modelling approach for

the simulation of a low temperature district heatingnetwork with seasonal heat storage, collectors andbuildings. The approach combines the simulation en-vironments Dymola with the Modelica Buildings li-brary and TRNSYS [33] with the help of the BuildingControls Virtual Testbed [59]. The system model is asimple network with buildings with only heating de-mand, solar thermal panels and a water storage tank.A neighborhood in Munich is used as a case study.

1.3 Previous work on agent-basedcontrol of distributed energy sys-tems

In this work a method to find an optimized temper-ature set point for a LTN is presented. In order totrack the set point, feedback control is needed. Sucha control algorithm needs to be able to orchestratedistributed heat and cold sources in an effective andstable way, in presence of distributed heat sourcessuch as solar thermal roof panels. It should further-more be flexible and expandable if additional build-ings, storage or sources are added. Agent-based con-trol is a concept which allows to control such complexsystems by splitting the main control objective intosmaller objectives which so-called agents try to fulfilby interacting with each other. Others used it to con-trol electrical smart grids ([15], [20], [25], [62], [17],[19], [36], [37], [61] and [1]) and complex building sys-tems ([14], [13], [42], [3] and [41]).

LTN are similar to electrical smart-grids and com-plex building energy systems. They feature dis-tributed sources for heat and cold as well as dis-tributed consumers of heat and cold. A control sys-tem for an LTN needs to maintain the temperature inboth temperature lines within a usable range for heatpumps and chillers. This needs to be done by balanc-ing supply and demand. The principle is analogousto the control mechanisms implemented in electricalgrids, where supply and demand need to be coordi-nated in order to keep the voltage within a range.Similar analogies can be found in complex buildingenergy systems, such as the building presented in[8]. As agent-based control has been shown to bean efficient way to implement control for smart-grids

2 METHODOLOGY 6

and complex building energy systems, and as LTNprovide control problems which are similar to thosefound in smart-grids and complex building energysystems, agent-based control is used in the course ofthis work to control the LTN.

2 Methodology

We will now describe the methodology used in ourstudy.In order to optimize the operation of LTN, we used

a cascading control as shown in Fig. 1. It consists ofan optimization of the temperature set point profilefor the network and a following sequence of controlsteps, including agent-based control, to maintain theoptimized temperature set point in the network withmultiple heat and cold sources. The optimization andcontrol steps will be presented in the following sec-tion.

2.1 Temperature set point optimiza-tion

Fig. 2 and Fig. 3 show the Carnot COP of a heatpump and a chiller, respectively, as a function of thetemperature of the fluid in the pipe. A constant 4Tof 2K between water and refrigerant is assumed incondensers and evaporators. As for the efficiency ofheat pumps, high pipe temperatures are favourablewhereas for chillers low temperatures are favourable,there is a potential for optimization to find the opti-mum temperature. However, in previous research, afree floating temperature approach was usually used,in which the warm line temperature is kept betweena lower and a higher limit. Between these limits itchanges depending on the heating and cooling de-mands of the consumers. Finding a suitable set pointthrough numeric optimization is our first step.The system is optimized for a cost function that

represents a performance metric. As the cold linetemperature depends on the warm line temperature,the optimization only searches for a set point for thewarm line. The cost function can for example be pri-mary energy consumption, electricity consumption orcarbon emissions. As the only costs are generated by

the heat pumps and chillers, these cost functions areproportional to each other if constant emission andprimary energy factors are assumed. Therefore, theoutcome of the optimization is independent of this se-lection. For example, the cost function for electricityconsumption follows

Cost =

∫ τ

0

(Pchi + Php) dt, (1)

where Cost is the value of the cost function, Pchi

is the electrical power of all chillers and Php is theelectrical power of all heat pumps.In order to maintain comparability with the pre-

viously used free floating approach, the set point isconstraint by 12◦C and 20◦C.The optimization problem was formulated as

minimize{a,b,c}∈<

f(Tset,wl),

subject to Tset,wl = a+ b

(Qh

Qh,max

)n

+ c

(Qc

Qc,max

)n

,

Tset,wl ∈ [12◦C, 20◦C],(2)

where Tset,wl is the set point for the warm line andn ≥ 0 is a constant. For n = 0, Tset,wl is a constant.For n = 1, the set point is linear in the load, unlessthe constraint Tset,wl ∈ [12◦C, 20◦C] is active. As wewill see below, it turns out that a constant set pointsuffices, and hence we will below only compare thecases for n ∈ {0, 1, 4}, with n = 4 selected to showthe effect of a large exponent. The evaluation of thecost function f : < → < involves a simulation of aModelica model which is a simplified model of theones used for the case study. It uses hourly load datafor heating and cooling loads for a district and mod-els the electricity consumption of heat pumps andchillers for individual building types with the help ofa Carnot approach. The heat pumps use an idealwater source at Tset,wl on the evaporator side, whilethe chillers use an ideal water source of Tset,wl − 4Kon the condenser side. A different model is used forthe optimization than in the case study. In the opti-mization model the pipe network and the electricityconsumption of the circulation pumps is not modelledto increase optimization speed.

2 METHODOLOGY 7

Figure 1: Optimization and control scheme for low temperature networks

8 10 12 14 16 18 2010

15

20

Temperature of the source [◦C]

COP

C

Figure 2: Carnot Coefficient of Performance of a heat pump withThea = 30◦C with 4T = 2K in heat exchangers

8 10 12 14 16 18 200

50

100

150

Temperature of heat rejection [◦C]

COP

C

Figure 3: Carnot Coefficient of Performance of a chiller withTcoo = 10◦C with 4T = 2K in heat exchangers

The optimization was done in Python 2.7. To auto-mate the optimization, the standard Dymola-Pythoninterface, which is provided with Dymola, was used.The simplex method Nelder-Mead with default pa-rameters and tol = 1E-6 from the Python SciPy pack-age was used to solve the optimization problem.

Fig. 4 shows the results for Tset,wl of the optimiza-tion for the San Francisco case (see section 3). In thefree floating case, one can see that the temperatureis lowest during winter, as heating dominates cool-ing, which results in a temperature drop of the whole

0 50 100 150 200 250 300 350

10

15

20

25

30

Time in d

Tem

perature

in◦C n=1 n=4

n=0 free-floating

Figure 4: Temperature set points of the warm line after optimiza-tion

network. The highest temperatures are reached inlate summer when cooling is high. Consequently, thetemperatures in the network rise. Contrasting thisbehaviour to the COP of heat pumps and chillers, itbecomes clear that a free floating temperature causesthe grid to establish temperatures that are oppositeof the ideal case. The grid is warm when it should becold and is cold when it should be warm.

It can be noticed that all optimized results havea higher average temperature than the free floatingcase. Besides a slightly lower temperature in thelate summer days, the linear approach does not dif-fer much from the constant approach. For n=4, thetemperature during summer is lowered by 6 K. This isexpected because the cooling load is higher in sum-mer and lower temperatures are favourable for theperformance of chillers.

However, the cost function values are all within 1%.Hence, n has little effect on the performance. Thus,

2 METHODOLOGY 8

whether the line temperature during summer is at20 ◦C or at 14 ◦C has little impact on performance.However, the free floating case shows a 15% higherelectricity consumption.The surprising result can by explained by further

examination of the network performance at differentload ratios. Neglecting the power consumption of thepumps, assuming that non-balanced heat in the dis-tribution network can be made up with renewableenergy or storage that shifts loads, and assuming aconstant Carnot efficiency η, the normalized electric-ity consumption is

pel =1

η

(qh

COPh+

qcCOPc

), (3)

where qh is the normalized heating load, qc is thenormalized cooling load, COPh is the Carnot coef-ficient of performance for heating and COPc is theCarnot coefficient of performance for cooling. UsingCOPh = Th/(Th − Tc) and COPc = Tc/(Th − Tc),where Th and Tc are the refrigerant temperatures inthe condenser and evaporator, we can write (3) as

pel =1

η

(qh

Th − Tc

Th+ qc

Th − Tc

Tc

). (4)

Let Twl and Tcl be the water temperatures in thewarm and cold pipe, let Thea be the space heatingsupply temperature and Tcoo be the space coolingsupply temperature,let ∆Tl , Twl−Tcl and let ∆Tr >0 be the temperature difference between water andrefrigerant. Then, we can write (4) as

pel =1

η(qh

(Thea +∆Tr)− (Twl −∆Tr)

Thea +∆Tr

+ qc(Twl −∆Tl +∆Tr)− (Tcoo −∆Tr)

Tcoo −∆Tr).

(5)Assuming η = 1, Thea = 30 ◦C, Tcoo = 10 ◦C, ∆Tr

= 2 K in all heat exchangers and ∆Tl = 4 K , thenormalized electrical consumption is as shown in Fig.5. It turns out that pel,network is linear in the warmline temperature.The plot shows that electricity consumption for

higher heating ratios (red) is generally higher than forhigher cooling ratios (blue). This can be explained

8 10 12 14 16 18 200

0.5

1

Temperature of the warm line Twl in◦C

pel,netw

ork

qh = 5, qc = 5

qh = 9, qc = 1

qh = 1, qc = 9

Figure 5: Relative electricity consumption for different heating tocooling ratios

by the generally higher coefficient of performance ofchillers compared to heat pumps. Furthermore, itcan be seen that the performance curve of the net-work at equal heating and cooling load is nearly flat,which explains the results of the optimization. Asheating and cooling loads are of similar magnitude inthe summer months for the San Francisco scenario,low temperature set points during summer performequally well as high ones. For this case, it can bestated that a well-chosen temperature set point dur-ing winter with high heating loads is more importantto the network performance than the set point duringsummer.

As all set point optimizations showed similar costfunction values, only the most simple one, the con-stant temperature approach, was chosen for the casestudies. For a transfer to a real system, this approachis more suitable than the others as no load data needto be measured. The same approach was chosen forthe Cologne case, as the optimization showed similarresults.

2.2 Control steps with agent-basedcontrol

In the above section, a method to find an optimizedtemperature set point for the network was presented.In order to track the set point, agent-based control isused.

Fig. 6 shows the LTN equipped with the agent-based control system. The two lines with warm and

2 METHODOLOGY 9

BrokerConsumer

Agent

Producer

Agent

Producer

Agent

Producer

Agent

Temperature

measurement

Figure 6: Bidirectional low temperature network with agent-based control

cold temperature level are represented by the redand blue ellipses. The network has heat and coldconsumers represented by the orange and light bluebuildings connected to the network. Besides theseconsumers, also three producers are connected to thenetwork: a solar thermal panel, a water heat ex-changer and a water storage tank. It should be men-tioned that consumers generally are also producers ina bidirectional LTN, for example, a heat pump con-sumes heat while producing cold. However, we meanby producer a facility whose main task is to maintainthe right network temperature, and by consumer afacility whose main task is to heat or cool buildings.

The network is controlled as follows. The temper-ature of the network is measured at multiple pointsthat are evenly spread along the warm circular pipe(in this case four points) and an average temperatureis computed. This temperature is compared to thetemperature set point Tset,wl after a low pass filter. API controller with anti-windup converts the differencebetween the current average warm line temperatureof the network and Tset,wl into a demand for heatingor cooling. The consumer agent sends its capacity de-

mand to the broker. The broker then proceeds to callfor a proposal from each producer agent in the sys-tem. With the help of a cost function, each produceragent computes a proposal for the requested capac-ity adjustment. If a producer is not able to make theadjustment, it can refuse the call for proposal by thebroker. When the broker has collected the proposals(and refusals) of all producer agents, the proposalsare compared and the most cost-efficient combinationis selected. Notifications of accepted and refused of-fers are sent to the producer agents. The produceragents then adjust the capacity of the producer theyrepresent.In contrast to a building energy system, the re-

newable producers in this example of an LTN donot produce real costs during their operation be-cause they produce heat or cold close to zero marginalcost. Therefore, a virtual cost function is necessaryas a measure of cost. The cost function is based onthe temperature difference of the water that the pro-ducer can supply and the current network tempera-ture. The cost function for a heat supplier is

C = Cap (Tset,wl − Tsource), (6)

3 CASE STUDY 10

where C is cost, Cap is the capacity of the source,Tset,wl is the set point temperature of the hot lineand Tsource is the supply temperature of the source.

Similarly, for cold suppliers, the cost is

C = Cap (Tsource − Tset,cl), (7)

where Tset,cl = Tset,wl−4K is the set point of the coldline. In (6) and (7), the temperature gives lower costthe closer a source is to the set point. For example,if Tset,wl = 20◦C and heat sources at 18◦C and 19◦Care available, the source at 19◦C is chosen in orderto optimize the network performance.

In the last step of the control sequence, the ca-pacity requested from a producer agent needs to betranslated into an enthalpy change in the networkfluid. In the case of low exergy sources such as waterheat exchangers and seasonal storage facilities, this isdone with a P controller and a valve which regulatesthe resulting mass flow induced by the individual con-sumer pumps. In the case of a high exergy source,such as solar thermal panel, a pump which inducesadditional mass flow instead of a valve is used in orderto obtain better network penetration and as a resultmore even network temperatures. In both cases, theabsolute value of the requested capacity is used as theset point for the P controller and the absolute valueof the enthalpy difference upstream and downstreamof the source as the measurement input. Absolutevalues are necessary in order to handle the possiblechange of flow direction in the source, and to keepthe control stable.

3 Case study

To quantify the benefits of LTN and to verify thepresented control strategy, different scenarios wereanalyzed using Modelica. Two different example dis-tricts,the Shipyard neighborhood in San Francisco,USA, and the Rheinauhafen district in Cologne, Ger-many, were selected. The scenarios differ from eachother in terms of type and amount of buildings, in-sulations standards, and climate leading to differentheating and cooling loads. For both scenarios, a com-parison of the following technologies was made:

• Conventional gas-fired district heating andstand-alone chillers

• Stand-alone heat pumps and chillers

• Unidirectional LTN with free floating tempera-tures

• Bidirectional LTN with free floating tempera-tures

• Bidirectional LTN with agent-based control andmultiple sources

3.1 Load data and boundary condi-tions

In order to model heating and cooling loads for theSan Francisco scenario, information on the distri-bution of buildings in the neighborhood was gath-ered from [28]. With the help of the informationon the type of buildings and the area occupied bythese buildings, load profiles were created using De-partment of Energy reference models [53]. Becauseearly simulations indicated that the given numberand types of buildings would lead to an unbalancednetwork, not all residential buildings were connectedto the network. The resulting floor area consists of72% for offices, 16% for housing and 12% for retail.Fig. 7 shows the integrated heating and cooling loadsfor the neighborhood per m2 floor area. One cansee that the heating load is bigger than the coolingload during winter, but the system is well balancedduring the summer months. The annual diversity in-dex calculated after [64] is divannual = 0.70, whichis a value that indicates that a bidirectional LTNshould perform well in this scenario. Weather datafor the simulation was taken from TMY3 San Fran-cisco International Airport. Water temperatures forthe ocean heat exchanger were taken from [55] for thenearby city of Alameda, which is 8 km north-east ofthe neighborhood. The temperatures are shown inFig. 8. Tab. 4 shows the main parameters of the SanFrancisco location.For the Cologne scenario, information on the distri-

bution of building types was obtained from [39]. Thetotal floor area consists of 60% office space and 40%

3 CASE STUDY 11

housing. To generate load curves, the Python pack-age TEASER (Tool for Energy Analysis and Simu-lation for Efficient Retrofit) [38] and Modelica sim-ulations were used. TEASER allows the creationof archetype building model records for a low-orderbuilding model of the AixLib [31] library and Build-ings [60] library . Fig. 9 shows the integrated loadsper m2 floor area for the Cologne scenario. One cansee that there is only a limited amount of overlappingbetween heating and cooling loads. Additionally, inthe periods in which both types of loads occur, theydo not occur during the same hours (heating at night,cooling during the day). This leads to an annual di-versity index of divannual = 0.04. As there is no dataon Rhine water temperatures available for the city ofCologne, data from the Dusseldorf-Flehe station from2015 was used instead [26], which is 40km away fromCologne. TRY2012 data from Aachen (63km fromCologne) was used to simulate the weather. Fig. 10shows water and ambient air temperatures. Tab. 5shows the main parameters of the Cologne location.

Operating temperatures are 30 ◦C for supply and25 ◦C for return in case of space heating, 16 ◦C forsupply and 20 ◦C for return in case of space coolingand 60 ◦C for supply in case of domestic hot tap waterin all models and both locations.

0 50 100 150 200 250 300 350

−50

0

50

100

Time in d

Loadin

kW

hm

2

Qh Qc

Figure 7: Heating and cooling load in San Francisco scenario

0 50 100 150 200 250 300 350

0

20

40

Time in d

Tem

perature

in◦C

Tair,drybulb Twater

Figure 8: Air and water temperatures in San Francisco scenario

0 50 100 150 200 250 300 350

−50

0

50

100

Time in d

Loadin

kW

hm

2

Qh Qc

Figure 9: Heating and cooling load in Cologne scenario

3.2 Models for the investigated tech-nologies

For each heating and cooling technology, a dy-namic Modelica model was created. All thermaland hydraulic models were built and aggregatedusing the Modelica Buildings Library (branch is-sue653 carnot dt, commit add1439) [60] and theAixLib library (master branch, commit 4b7047c) [31].The agent-based control elements were realized withthe Modelica HVACAgentBasedControl library (alsoaccessible through AixLib) [2]. The aggregated sys-tem models can be requested from the authors. Eachmodel will be described in the following section.

3.2.1 LTN with agent-based control and mul-tiple sources

Fig. 11 shows an overview of the hydraulic set up andthe control signals of the model for the bidirectional

3 CASE STUDY 12

0 50 100 150 200 250 300 350

0

20

40

Time in d

Tem

perature

in◦C

Tair,drybulb Twater

Figure 10: Air and water temperatures in Cologne scenario

LTN with agent-based control. The schematics showsas an example one consumer with the evaporator ofa heat pump and the condenser of a chiller. The ac-tual model has more than one consumer and separateheat pumps for space heating and domestic hot tapwater. The heat and cold consumers were modelledas substations based on hourly load data for heatingand cooling demands. Heat pumps and chillers use aCarnot-efficiency-based approach with η = 0.45 and atemperature difference in the heat exchangers depen-dent on the load ratio with 4Tr,max = 2K in orderto determine their electricity consumption. The totalsubstation electricity consumption includes the heatpumps, chillers and circulation pumps. Also shownare the three supply plants, consisting of water heat-exchanger that is connected to a river, lake or ocean,storage tank and solar thermal plant. The two lat-ter ones are equipped with producer agents, whichcontrol the valve and the pump. The valve of thewater heat exchanger acts as a bypass for the stor-age tank. The hydraulic network is modelled withpipe models that account for pressure losses based onflow-rate dependent flow friction and thermal lossesor gains. (For more details on the models please referto [58] and the documentation within the open sourceModelica Buildings Library [60]). The total length ofthe network in San Francisco and Cologne are 8 kmand the pipes are sized to accommodate the nomi-nal mass flow at a flow velocity of 1.5 m

s . The pipesare insulated with a 20 cm layer with a heat transfercoefficient of 0.04 W

mK . For the boundary conditionof the thermal losses, a undisturbed ground tempera-

ture model which accounts for seasonal change of theground temperature is used [24]. The pipe networkis divided in one hundred control volumes, each witha fully mixed temperature.Important parameters of the system are given in

Tab. 6 for San Francisco and in Tab. 7 for Cologne.Apart from that, the models for San Francisco andCologne only differ in heating and cooling demands,weather and water temperatures for the water heatexchangers, as described in the previous section.In order to assess the network performance, the

electricity consumption of all relevant appliances inthe system is

Eprim =

∫ τ

0

PEFel (Ppumps + Pchi + Php) dt, (8)

where Eprim is the primary energy consumption,PEFel is the primary energy factor for electric-ity, Ppumps is the electrical power of all circulationpumps, Pchi is the electrical power of all chillers andPhp is the electrical power of all heat pumps.

3.2.2 Bidirectional LTN with free-floatingtemperatures

Fig. 12 (A) shows the hydraulic system for the bidi-rectional LTN with free floating temperatures. Theonly heat and cold source is a water heat exchanger.The water heat exchanger is controlled to keep thewarm line temperature between 12◦C and 20◦C andthe cold line temperature between 8◦C and 16◦C.The pipe network and substations are identical to theLTN with agent-based control and multiple sources.

3.2.3 Unidirectional LTN with free-floatingtemperatures

Fig. 12 (B) shows the hydraulic system for theunidirectional LTN with free floating temperatures.Again, the only heat and cold source is a water heatexchanger. It is controlled to keep the water tem-perature within the same limits as in the bidirec-tional LTN. In the unidirectional LTN, heat pumpsand chillers draw water from the same supply lineand discharge to the same return line. Therefore,

3 CASE STUDY 13

water

storage

water

HEX

-

1

solar

plant

consumer

control

signal

control

signal

bypass

signal

= valve = heat exchanger = storage = pump= solar thermal

panel

Figure 11: Bidirectional LTN with multiple sources model schematic

the pump of the chillers in the substations is turnedaround. The pipe network has the same length, di-ameter and insulation as in the bidirectional LTN.Also the efficiency of the heat pumps and chillers inthe substations is computed with the same approachas in the bidirectional systems.

3.2.4 District heating with stand-alone cool-ing

Fig. 13 (A) shows the hydraulic scheme for the dis-trict heating with stand-alone cooling on the left-hand side and the solution with stand-alone heatingand cooling on the right-hand side. The district heat-ing system uses a central natural gas heater with athermal efficiency of 0.95. The heater is controlledto supply a constant water temperature of 90 ◦C. In-dividual pumps at the consumer substations controlthe mass-flow in order to reach return temperaturesof 60 ◦C. The network is also dimensioned to accom-modate the nominal mass-flow rate at a velocity of1.5 m

s . This leads to smaller diameters (26 cm) thanin the LTN case (71 cm), as the temperature spreadbetween supply and return is much bigger. Further-more, an insulation layer of 40 cm with a heat transfercoefficient of 0.035 W

mK (compare [43]) is used in orderto reduce thermal losses. In the substations, a simple

heat exchanger with 100% efficiency is used insteadof the heat pump for space heating and domestic hottap water in the substations. To meet the coolingdemand, a chiller that is connected to the ambientair is used. The main parameters of the system aregiven in Tab. 8.

To assess the performance of the conventional dis-trict heating scenario, the primary energy is

Eprim =

∫ τ

0

[PEFgas Wgas

+PEFel (Ppumps + Pchi)]dt,

(9)

where PEFgas is the primary energy factor for natu-ral gas and Wgas is the chemical work of the naturalgas.

3.2.5 Stand-alone solution

Fig. 13 (B) shows the stand-alone solution. Thesubstations use heat pumps and chillers which areboth connected to the ambient air. There is no pip-ing network in this scenario. Parameters of the heatpumps and chillers are identical to the LTN cases.The performance evaluation for the pure stand-alonecase follows (8).

4 RESULTS AND DISCUSSION 14

consumer

water

HEX

consumer

water

HEX

(A) (B)

Figure 12: (A): bidirectional LTN with water heat exchanger schematic, (B): unidirectional LTN with water heat exchanger schematic

4 Results and discussion

4.1 Primary energy usage, emissionsand energy costs

Fig. 14 (A) and Fig. 14 (B) show the relative pri-mary energy consumption, normalized by the maxi-mum primary energy consumption for the case stud-ies San Francisco and Cologne respectively. Primaryenergy factors for electricity and natural gas, takenfrom [4] and [6], were used as stated in Tab. 1.

Table 1: Primary energy factors used in simulation

Energy carrier San Francisco CologneNatural gas 1.09 1.10Electricity 2.89 1.80

Both scenarios show that solutions with heatpumps generally perform better than the districtheating solution. In the Cologne scenario the differ-ence between district heating and stand-alone solu-tion is of higher magnitude than in the San Franciscoscenario because the primary energy factor for elec-tricity in Germany is lower than for the western USA.However, the difference is not proportional to thedifference in primary energy factors, as the climatein San Francisco is milder than the one in Cologne,which leads to better heat pump and chiller perfor-mance. In comparison to the stand-alone solution, a

bidirectional LTN with free-floating temperature con-trol decreases the primary energy consumption by 9%in San Francisco and 13.5% in Cologne. The selec-tion of a bidirectional network instead of a unidirec-tional network leads to an improvement of 5% in SanFrancisco but only 1.5% in Cologne. This can be ex-plained by the shape of the heat and cold demands(compare Fig. 7 and Fig. 9) and the exergy destruc-tion of mixing fluid streams of different temperatures.In the case of San Francisco, there is a higher occur-rence of heat and cold demands at the same time thanin the Cologne case, where heat is mainly used dur-ing winter and cold during summer. The advantageof bidirectional networks in terms of exergy efficiencyis the avoidance of mixing the warm return streamsfrom the chillers and the cold return streams from theheat pumps in one common return line, as discussedin [40]. In the case where heat and cold demandsare separated, the exergy destruction in the mixingnodes does no longer take place. Between the casesbidirectional LTN with free-floating temperature andbidirectional LTN with agent-based control, the pri-mary energy consumption could be further loweredby 13% in the San Francisco scenario and 41% inthe Cologne scenario. The higher improvement inthe Cologne case can again be explained by the dif-ferent boundary conditions. Both bidirectional mod-els use water heat exchangers. In the case of SanFrancisco, the heat exchanger exchanges heat with

4 RESULTS AND DISCUSSION 15

consumer

central

gas

heater

environment

environment

= boiler

consumer

environment

environment

(A) (B)

Figure 13: (A): district heating and stand-alone cooling schematic, (B): stand-alone schematic

the San Francisco Bay, which has relatively constanttemperatures between 10 ◦C and 19 ◦C. This temper-ature range is a level at which both heat pumps andchillers can operate with high efficiency. The waterheat exchanger in the Cologne case exchanges heatwith the Rhine, whose temperature ranges from 6 ◦Cin winter to 24 ◦C in summer. Together with the factthat high heat demands occur only during winter andhigh cold demands during summer in Cologne, thisleads to a large improvement potential through theinclusion of other heat and cold sources such as solarthermal panels and seasonal heat and cold storage.

Fig. 14 (C) and Fig. 14 (D) show the normalizedrelative CO2 emissions for the San Francisco case andCologne case respectively, with the emission factorsas in Tab. 2, taken from [4], [54] and [18]. Comparedto the primary energy results, the CO2 results showa much smaller gap between the district heating andthe stand-alone case for San Francisco. This is dueto the relatively high electricity CO2 emission factor.For Cologne, the difference in emission is much closerto the difference in primary energy. For San Fran-cisco, a reduction of 18% from district heating to thestand-alone solution could be achieved. For Cologne,this value is 47%. As options (2) to (5) are all electricsystems the relative changes are proportional for pri-

mary energy consumption, CO2 emissions and energycosts.

Table 2: CO2 emission factors used in simulation inkgCO2,eq

kWh

Energy carrier San Francisco CologneNatural gas 0.181 0.250Electricity 0.788 0.565

For both locations, the results show that bidirec-tional LTN offer potential for primary energy andCO2 emission reductions, especially when used withmultiple heat and cold sources and agent-based con-trol in order to achieve beneficial network tempera-tures.Fig. 14 (E) and Fig. 14 (F) show the normalized

energy costs for both scenarios. Tab. 3 shows theenergy costs that are used for the simulation, takenfrom [34], [7], [56] and [44].

Table 3: Energy prices used in simulation incentEuro/Dollar

kWh

Energy carrier San Francisco CologneNatural gas 5.08 5.80Electricity 15.34 29.16

The San Francisco results show that technologies

4 RESULTS AND DISCUSSION 16

San Francisco Cologne

(A) (B)

(1) (2) (3) (4) (5)

0

50

100100

53.07 50.83 48.2 41.96

PE

consumptionin

%

(1) (2) (3) (4) (5)

0

50

100100

31.85 28.09 27.6216.32

PE

consumptionin

%

(C) (D)

(1) (2) (3) (4) (5)

0

50

100100

82.45 78.95 74.1565.19

CO

2em

issionsin

%

(1) (2) (3) (4) (5)

0

50

100100

42.69 37.66 37.0321.87

CO

2em

issionsin

%

(E) (F)

(1) (2) (3) (4) (5)

0

50

100100

59.73 57.19 54.24 47.23

Energycostsin

%

(1) (2) (3) (4) (5)

0

50

100100

8474.11 72.86

43.04

Energycostsin

%

Figure 14: Case study results: (A) and (B) normalized primary energy consumption, (C) and (D) normalized CO2 emissions, (E)and (F) normalized cost of operationScenarios: (1) District heating with stand-alone cooling, (2) stand-alone heating and cooling, (3) unidirectional LTN with free-floatingtemperature, (4) bidirectional LTN with free-floating temperature, (5) bidirectional LTN with agent-based control

using mainly electricity for heat generation are muchmore cost effective than gas-fired district heatingwithout cogeneration. By using stand-alone heatpumps instead of district heating, the energy costscan be reduced by 40% because the price for elec-tricity is low in California. The best scenario witha bidirectional LTN and agent-based control offers atotal reduction of 53%. For Cologne, the difference inenergy costs between district heating and the stand-

alone solution is much lower due to the significantlyhigher electricity prices in Germany. However, thebest technology with a bidirectional LTN with agent-based control achieves a reduction of 57% comparedto the district heating case. The savings of 53% and57% are in good agreement with [32], who describesa 62% reduction of operating cost for the LTN inChur compared to the previously installed heatingand cooling system.

4 RESULTS AND DISCUSSION 17

For a complete economic evaluation, the netpresent value analysis of the competing technologiesneeds to be made. For this analysis, the investmentcosts need to be known. However, in the case ofbidirectional LTN these investments are only knownby a limited number of people as there are only afew demonstration projects available so far. Further-more, other economic benefits, such as maintenanceand usage of roof and floor space should be consid-ered.The results should be viewed keeping in mind

that one focus of this work is operation optimiza-tion of bidirectional LTN. However, the control ofall evaluated technologies could be further developed.Such developments could lead to further reductions ofprimary energy consumption, emissions and energycosts in all cases. Using combined heat and powerunits or waste heat instead of gas-fired boilers in thedistrict heating case would also lead to different re-sults that would benefit the performance of the con-ventional district heating system. Another point ofdiscussion is the selection of primary energy factorsand emission factors. Such factors always carry anuncertainty. As these values are usually calculatedusing life cycle analysis, in which many assumptionsare made, factors can vary depending on the partythat conducted the analysis. A further aspect is thedate of publication, which varies a lot in the case ofprimary energy factors. Also, due to the increasedintegration of renewables in the electricity grid, theprimary energy factors are changing.

4.2 Performance of agent-based con-trol

As the functionality of the control approach is inde-pendent of the chosen location, only the San Fran-cisco results are discussed in the following in order toavoid redundancy.Fig. 15 shows the average of the four tempera-

ture measurement points in the agent-based controlapproach in comparison to the optimized set point of20 ◦C and the temperature trajectory in the case of abidirectional network with free-floating temperature.The figure shows that the system does not have

enough heating capacity to keep the temperature at

0 50 100 150 200 250 300 350

10

15

20

25

Time in d

Tem

perature

in◦C Tset,wl Tagents Tfree−floating

Figure 15: Average temperature trajectory with agent-based con-trol and with free-floating temperatures for San Francisco

the optimal set point in winter. The control erroris up to 3 K. This problem could be solved by im-plementing more solar thermal panels or other heatsources. This is, however, a question of design op-timization and is out of scope of this paper. Theagent-based control is able to keep the average sys-tem temperature within a margin of 2 K around theset point during times in which the heat and cold de-mand in the system is more balanced. The error of2 K during summer only occurs when the solar ther-mal plant is activated, which indicates that it couldbe lowered by improving the control of the solar ther-mal plant. During most times, the control error is lessthan 1 K.

In general the agent-based control in combinationwith multiple sources keeps the network temperaturemuch closer to the optimal set point than the free-floating approach, which leads to the improved per-formance presented above.

Fig. 16 shows the temperature in the warm line ofthe network at the four points of measurement. Thetemperatures vary widely and the assumption thata single temperature for each line is sufficient to de-scribe the state of a bidirectional network is proven tobe wrong. The temperatures T1,T3 and T4 appear tobe relatively stable whereas T2 shows high frequencyfluctuations of up to 9 K. Fig. 17 shows the mass flowrates at the points of temperature measurement. Thehighest mass flow rates and also the highest variationin mass flow rates are at the T2 temperature measure-

5 CONCLUSION 18

0 50 100 150 200 250 300 350

15

20

25

Time in d

Tem

perature

in◦C

T1 T2 T3 T4

Figure 16: Temperature variation with agent-based control for SanFrancisco

0 50 100 150 200 250 300 350

−500

0

500

Time in d

Mass

flow

inkg s

m1 m2 m3 m4

Figure 17: Mass flow variation with agent-based control for SanFrancisco

ment. This indicates that the point of measurementis a place where a large amount of fluid is bypassedbetween the warm and cold line, as large mass flowsshould mainly appear at bypass points in a networkwith reasonably well distributed heat and cold de-mands. A comparison to a mass flow rate measure-ment directly at the water heat exchanger and thefact that the water heat exchanger is used as the by-pass whenever storage or solar thermal plants are notactive support this observation.The observed temperature variation does howevernot constitute a serious problem for the performanceof the LTN. By comparing the bidirectional agent-based control case with the free-floating temperatureapproach and the optimization minimum (the case

(1) (2) (3)

0

50

100

150

10087.1 84

100

59.147.5

Norm

alizedprimary

energyconsumptionin

%

San Francisco Cologne

Figure 18: Agent-based control performance with (1) Free floatingtemperature (2) Bidirectional agent-based control (3) Optimiza-tion minimum

in which heat pumps are used with an ideal heatsource at Tset,wl and chillers with an ideal heat sinkat Tset,cl = Tset,wl − 4K), it becomes clear that theagent-based approach achieved for San Francisco a13% reduction at a maximum possible reduction of16% (see Fig. 18). The remaining 3% could possiblybe achieved by optimizing the design of the system,such as adding more heat sources. For Cologne, areduction of 41% out of possible 56% was achieved.

5 Conclusion

Bidirectional LTN lead to an improvement in opera-tion efficiency and can effectively be controlled withagent-based control, which aids the modular exten-sion of such LTN systems. The agent-based con-trol successfully coordinated various heat and coldsources to keep the network temperature around aspecified set point. For both locations, the bidirec-tional LTN with agent-based control is the most ef-ficient technology. In comparison to a conventionalgas-fired district heating system, the optimized LTNleads to primary energy consumption reductions of58% and 84% in the US and German scenario, re-spectively. Reductions of CO2 emissions are 35%and 78%, and reductions of energy costs 53% and57%. The savings are in good agreement with the re-sults from existing demonstration projects. Anotherconclusion is that a constant set point for the pipetemperatures is sufficient for the analyzed systems.

A LOCATION PARAMETERS 19

6 Acknowledgements

This research was supported by the Assistant Sec-retary for Energy Efficiency and Renewable Energy,Office of Building Technologies of the U.S. Depart-ment of Energy, under Contract No. DE-AC02-05CH11231. This work was also funded by LawrenceBerkeley National Laboratory through the Labora-tory Directed Research and Development (LDRD)ProgramThis work emerged from the Annex 60 project, an

international project conducted under the umbrellaof the International Energy Agency (IEA) within theEnergy in Buildings and Communities (EBC) Pro-gramme. Annex 60 developed and demonstratednew generation computational tools for building andcommunity energy systems based on Modelica, Func-tional Mockup Interface and BIM standards.We gratefully acknowledge the financial support

for Felix Bunning from the German Academic Ex-change Service (DAAD) through its Thematic Net-work ”ACalNet”.

A Location parameters

Table 4: Parameters for the location San Francisco

Parameter Value UnitAfloor 170, 000 m2

Qnom 10, 000 kWQh 24.204E9 kWhQc 10.348E9 kWh

Table 5: Parameters for the location Cologne

Parameter Value UnitAfloor 170, 000 m2

Qnom 10, 000 kWQh 24.204E9 kWhQc 10.348E9 kWh

B MODEL PARAMETERS 20

B Model parameters

Table 6: Parameters for bidirectional agent-based control LTNmodel of San Francisco

Parameter Value UnitHeat pumps

ηc 0.45 −4Teva 2 K4Tcon 2 K

Tsup,space 30 ◦CTret,space 25 ◦CTsup,hotwat 60 ◦CChiller

ηc 0.45 -4Teva 2 K4Tcon 2 K

Tsup,space 16 ◦CTret,space 20 ◦CNetwork4Tnominal 4 K

mnominalQnominal

cpw·4Tnominal

kgs

vnominal 1.5 ms

Pipesltotal 8 km

Rnominal 100 Pam

di 71 cmdins 20 cm

λins 0.04 WmK

PI controller

P 2.5E6 WK

Ti 1000 symax 1.2E7 -ymin −1.2E7 -

Seasonal StorageV 1.0E6 m3

Solar thermal plantA 2.0E4 m2

Temperature set pointsTset,wl 20 ◦CTset,cl 16 ◦C

Table 7: Deviating parameters for bidirectional agent-based con-trol LTN model Cologne

Parameter Value UnitSeasonal Storage

V 2.0E6 m3

Table 8: Parameters for district heating model

Parameter Value UnitCentral heat supplier

Tsup 90 ◦CTret 60 ◦Cη 0.95 −

Pipesltotal 8 km

Rnominal 100 Pam

di 26 cmdins 40 cm

λins 0.035 WmK

REFERENCES 21

References

[1] J. Ansari, A. Gholami, and A. Kazemi.Multi-agent systems for reactive power controlin smart grids. International Journal ofElectrical Power & Energy Systems, 83:411–425, 2016.

[2] F. Bunning, R. Sangi, and D. Muller. Amodelica library for the agent-based control ofbuilding energy systems. Applied Energy, 193:52–59, 2017.

[3] J. Cai, D. Kim, R. Jaramillo, J. E. Braun, andJ. Hu. A general multi-agent control approachfor building energy system optimization.Energy and Buildings, 127:337 – 351, 2016.ISSN 0378-7788.

[4] M. Deru and P. Torcellini. Source energy andemission factors for energy use in buildings.Technical Report NREL/TP-550-38617,National Renewable Energy Laboratory, 2007.

[5] B. Di Pietra, F. Zanghirella, and G. Puglisi.An evaluation of distributed solar thermal “netmetering” in small-scale district heatingsystems. Energy Procedia, 78:1859–1864, 2015.

[6] DIN Deutsches Institut fur Normung e. V. DINV 18599 2016, 2016.

[7] Electricity Local. California electricity rates &consumption.http://www.electricitylocal.com/states/california/, 2017. Accessed on: 03-03-2017.

[8] J. Fuetterer and A. Constantin. Energyconcept for the e.on erc main building. E.ONEnergy Research Center Series, 4, Issue 9, 2014.

[9] T. Gautschi. Anergienetze in Betrieb. TechnicalReport 66, Amstein + Walthert AG, 2015.

[10] Gemeinde Visp, CH. Anergienetz, eineerdolunabhangige Energieversorgung.Medienkonferenz, 2007.

[11] K. M. Heissler, L. Franke, I. Nemeth, andT. Auer. Modeling low temperature districtheating networks for the utilization of localenergy potentials. Bauphysik, 38(6):372–377,2016.

[12] S. Henchoz, P. Chatelan, F. Marechal, andD. Favrat. Key energy and technologicalaspects of three innovative concepts of districtenergy networks. Energy, 117(2):465–477, 2016.

[13] M. Huber, S. Brust, T. Schutz, A. Constantin,R. Streblow, and D. Muller. Purely agent basedcontrol of building energy supply systems. InECOS - International Conference onEfficiency, Cost, Optimization, Simulation andEnvironmental Impact of Energy Systems, 2015.

[14] L. Hurtado, P. Nguyen, and W. Kling. Smartgrid and smart building inter-operation usingagent-based particle swarm optimization.Sustainable Energy, Grids and Networks, 2:32–40, 2015.

[15] Z. Jiang. Agent-based control framework fordistributed energy resources microgrids. InIEEE/WIC/ACM International Conference onIntelligent Agent Technology, 2006.

[16] Kanton St. Gallen. Amtsblatt Nr.7, M15, 2016.

[17] C.-S. Karavas, G. Kyriakarakos, K. G.Arvanitis, and G. Papadakis. A multi-agentdecentralized energy management system basedon distributed intelligence for the design andcontrol of autonomous polygenerationmicrogrids. Energy Conversion andManagement, 103:166–179, 2015.

[18] KEA Klimaschutz- und EnergieagenturBaden-Wrttemberg GmbH.Co2-emissionsfaktoren. http://www.kea-bw.de/service/emissionsfaktoren/, 2016.Accessed on: 03-03-2017.

[19] M. R. B. Khan, R. Jidin, and J. Pasupuleti.Multi-agent based distributed controlarchitecture for microgrid energy management

REFERENCES 22

and optimization. Energy Conversion andManagement, 112:288–307, 2016.

[20] K. Kok, B. Roossien, P. MacDougall, O. vanPruissen, G. Venekamp, R. Kamphuis,J. Laarakkers, and C. Warmer. Dynamicpricing by scalable energy managementsystemsfield experiences and simulation resultsusing powermatcher. In Power and EnergySociety General Meeting, 2012 IEEE, pages1–8. IEEE, 2012.

[21] P. Krauchi and M. Kolb. Simulationthermischer arealvernetzung mit ida-ice. InFourth German-Austrian IBPSA Conference -Berlin University of the Arts, pages 205–211,2012.

[22] P. Krauchi and M. Kolb. Modellbildung furthermische Arealvernetzung mit IDA-ICE. InFifth German-Austrian IBPSA Conference -RWTH Aachen University, pages 160–165,2014.

[23] P. Krauchi, T. Schluck, and M. Sulzer.Modelling of low temperature heating networkswith IDA-ICE. In Proceedings of InternationalConference CISBAT 2015 Future Buildings andDistricts Sustainability from Nano to UrbanScale, number EPFL-CONF-213420, pages827–832. LESO-PB, EPFL, 2015.

[24] T. Kusuda and P. R. Achenbach. Earthtemperature and thermal diffusivity at selectedstations in the united states. Technical report,DTIC Document, 1965.

[25] E. Kuznetsova, Y.-F. Li, C. Ruiz, and E. Zio.An integrated framework of agent-basedmodelling and robust optimization formicrogrid energy management. Applied Energy,129:70–88, 2014.

[26] Landesamt fur Natur, Umwelt undVerbraucherschutz - Nordrhein-Westfalen.Wassertemperatur der Station D.-Flehe imJahr 2015. http://luadb.lds.nrw.de/LUA/hygon/pegel.php?stationsname t=

D.-Flehe&yAchse=Standard&hoehe=468&breite=724&jahr=2015&jahreswerte=

ok&nachSuche=&meifocus=&neuname=, 2016.Accessed on: 11-10-2016.

[27] Lauber IWISA AG. Eine erdolunabhangigeEnergieversorgung. Technical report, 2012.

[28] Lennar. The San Francisco Shipyard.https://www.lennar.com/new-homes/california/san-francisco-bay-area/san-

francisco/the-san-francisco-shipyard,2016. Accessed: 11-10-2016.

[29] H. Li and S. J. Wang. Challenges in smartlow-temperature district heating development.Energy Procedia, 61:1472–1475, 2014.

[30] H. Lund, S. Werner, R. Wiltshire, S. Svendsen,J. E. Thorsen, F. Hvelplund, and B. V.Mathiesen. 4th generation district heating(4gdh): Integrating smart thermal grids intofuture sustainable energy systems. Energy, 68:1–11, 2014.

[31] D. Muller, M. R. Lauster, A. Constantin,M. Fuchs, and P. Remmen. AixLib - AnOpen-Source Modelica Library within theIEA-EBC Annex 60 Framework. In BauSIM -Dresden, 2016.

[32] M. Nani. Anergienetze und Warmepumpen,HOVAL. http://www.fws.ch/tl files/

download d/Downloads/FWS-Tagung%202014/

Marco-Nani-Anergienetze.pdf, 2014.Accessed on: 05-06-2017.

[33] U. of Wisconsin-Madison. SolarEnergy Laboratory and S. A. Klein. TRNSYS,a transient system simulation program. SolarEnergy Laborataory, University ofWisconsin–Madison, 1979.

[34] Pacific Gas and Electric Company. Residentialaverage gas rate, february 2017 forecast, 2017.

[35] A. Prasanna, N. Vetterli, V. Dorer, andM. Sulzer. Modelling the suurstoffi districtbased on monitored data to analyse future

REFERENCES 23

scenarios for energy self-sufficiency. In 19.Status-Seminar Forschen fur den Bau imKontext von Energie und Umwelt, 2016.

[36] B. M. Radhakrishnan and D. Srinivasan. Amulti-agent based distributed energymanagement scheme for smart gridapplications. Energy, 103:192–204, 2016.

[37] M. Rahman, M. Mahmud, A. Oo, H. Pota, andM. Hossain. Agent-based reactive powermanagement of power distribution networkswith distributed energy generation. EnergyConversion and Management, 120:120–134,2016.

[38] P. Remmen, M. Lauster, M. Mans, M. Fuchs,T. Osterhage, and D. Muller. Teaser: an opentool for urban energy modelling of buildingstocks. Journal of Building PerformanceSimulation, pages 1–15, 2017.

[39] RVG Rheinauhafen VerwaltungsgesellschaftmbH. Die Eckdaten des Rheinauhafens. http://www.rheinauhafen-koeln.de/Uebersicht,2016. Accessed on: 11-10-2016.

[40] T. Schluck, P. Krauchi, and M. Sulzer.Non-linear thermal networks - how can ameshed network improve energy efficiency? InProceedings of International ConferenceCISBAT 2015 Future Buildings and DistrictsSustainability from Nano to Urban Scale,number EPFL-CONF-213421, pages 779–784.LESO-PB, EPFL, 2015.

[41] P. H. Shaikh, N. B. M. Nor, P. Nallagownden,I. Elamvazuthi, and T. Ibrahim. Intelligentmulti-objective control and management forsmart energy efficient buildings. InternationalJournal of Electrical Power & Energy Systems,74:403–409, 2016.

[42] S. Skarvelis-Kazakos, P. Papadopoulos, I. G.Unda, T. Gorman, A. Belaidi, and S. Zigan.Multiple energy carrier optimisation withintelligent agents. Applied Energy, 167:323–335,2015.

[43] STROBEL-VERLAG GmbH & Co. KG .Dammstandards von Fernwarmeleitungen. InIKZ-Haustechnik, volume 8, 2014.

[44] Strompreise.de. Strompreis pro kilowattstunde.https:

//www.strompreise.de/strompreis-kwh/,2017. Accessed on: 03-03-2017.

[45] R. Stulz, S. Tanner, R. Sigg, and F. Sioshansi.Swiss 2000-watt society: A sustainable energyvision for the future. Energy, Sustainability andthe Environment. Oxford, UK: Elsevier Inc,pages 477–496, 2011.

[46] M. Sulzer. Effizienzsteigerung mitAnergienetzen: Potentiale – Konzepte –Beispiele. Technical report, INRETIS - Energieund Gebaudetechnik, 2011.

[47] M. Sulzer. Solarenergie und Anergienetze -sinnvoll oder unsinnig? - Grundlagen,Konzepte und Beispiele, 2013.

[48] M. Sulzer and T. Gautschi. ETH Zurich,Honggerberg Masterplan Energie. In 15.Schweizerisches Status-Seminar ”Energie- undUmweltforschung im Bauwesen”, pages431–443, 2008.

[49] M. Sulzer and D. Hangartner. Kalte Fernwarme(Anergienetze) - Grundlagen-/Thesenpapier.Technical report, Lucerne University of AppliedSciences and Arts, 2014.

[50] M. Sulzer, U. Menti, and R. Sporri. Maschenund knoten. TEC21, 34/2015:31–33, 2015.

[51] S. Summermatter and M. Sulzer. SolareFernwarme im alpinen Raum –Wirtschaftlichkeitsanalyse. In 19.Status-Seminar Forschen fur den Bau imKontext von Energie und Umwelt, 2016.

[52] The European Comission. Fifth generation, lowtemperature, high exergy district heating andcooling networks. http://cordis.europa.eu/project/rcn/194622 en.html, 2015. Accessedon: 26-09-2017.

REFERENCES 24

[53] U.S. Department of Energy. Commercialreference buildings. http://en.openei.org/datasets/files/961/pub/,2016. Accessed on: 11-10-2016.

[54] US Environmental Protection Agency.Emission Factors for Greenhouse GasInventories. https://www.epa.gov/sites/production/files/2015-07/documents/

emission-factors 2014.pdf, 2014. Accessedon: 06-05-2017.

[55] U.S. National Oceanic and AtmosphereAdministration. Water temperature table of allcoastal regions. http://www.nodc.noaa.gov/dsdt/cwtg/all meanT.html, 2016. Accessedon: 11-10-2016.

[56] Verivox. Verivox-verbraucherpreisindex gas.http://www.verivox.de/verbraucherpreisindex-gas/, 2017. Accessedon: 03-03-2017.

[57] N. Vetterli and M. Sulzer. Dynamic analysis ofthe low-temperature district network”suurstoffi” through monitoring. In Proceedingsof International Conference CISBAT 2015Future Buildings and Districts Sustainabilityfrom Nano to Urban Scale, numberEPFL-CONF-213373, pages 517–522.LESO-PB, EPFL, 2015.

[58] M. Wetter. Modelica library for buildingheating; ventilation and air-conditioningsystems. In Proceedings of the 7th InternationalModelica Conference; Como; Italy; 20-22September 2009, number 43, pages 393–402.Linkping University Electronic Press; Linkpingsuniversitet, 2009.

[59] M. Wetter. Co-simulation of building energyand control systems with the building controlsvirtual test bed. Journal of BuildingPerformance Simulation, 4(3):185–203, 2011.

[60] M. Wetter, W. Zuo, T. S. Nouidui, andX. Pang. Modelica buildings library. Journal ofBuilding Performance Simulation, 7(4):253–270, 2014.

[61] E. Xydas, C. Marmaras, and L. M. Cipcigan. Amulti-agent based scheduling algorithm foradaptive electric vehicles charging. AppliedEnergy, 177:354–365, 2016.

[62] D. Ye, M. Zhang, and D. Sutanto.Decentralised dispatch of distributed energyresources in smart grids via multi-agentcoalition formation. Journal of Parallel andDistributed Computing, 83:30–43, 2015.

[63] F. Zach. Anergienetze - Optimierte Nutzunglokaler erneuerbarer Energietrager in urbanenNeubaugebieten – Beispiel Nordwestbahnhofe.Austrian Energy Agency, 2016.

[64] R. Zarin Pass, M. Wetter, and M. Piette. Athermodynamic analysis of a novel bidirectionaldistrict heating and cooling network. In review,2017.