254 IEEE Transactions on Power Systems, Vol. 3 , No. 1, February 1988
AN EXPERT SYSTEM FOR THE DESIGN OF A POWER PLANT ELECTRICAL AUXILIARY SYSTEM
Hans B. Pnttgen School of Electrical Engineering Georgia Institute of Technology Atlanta, GA 30332
Senior Member IEEE
m m m The ASDEP Expert System, which is oriented toward
the electric power plant auxiliary system design problem, is presented. The Artificial Intelligence techniques incorporated into ASDEP are reviewed along with the limitations of the expert system. An actual design session is stepped through for a nuclear power plant auxiliary system design. Finally, some possible extensions of the methods used to other power system design problems are alluded to.
Artificial Intelligence has received considerable attention during the past few years. Particularly, several so-called Expert Systems have been designed, implemented, and tested to carry out tasks normally carried out by human experts [l] . Expert Systems have primarily been successful for diagnostics applications where the problem is to arrive at correct diagnostics, or analysis, results based on a number of facts and measurements. Examples of such Expert Systems are PROSPECTOR, a computer-based consultation system developed at SRI International to assist geologists working on mineral exploration, and Gen-Aid, developed by Westinghouse Electric CO to diagnostic failures in turbine generators [ 2 ] . Only rarely have Expert Systems been implemented to handle design oriented tasks. One such system is the R1, or XCON, Expert System [3 ] which is capable of configuring a DEC VAX 111780 computer system based on a customer order.
The general field of power systems analysis and design has received little attention as far as Artificial Intelligence techniques are concerned, with the exception of the turbine generator diagnostics package mentioned above. In this paper a particular Expert System, called ASDEP (Auxiliary System Design And Evaluation Program), which is specifically oriented toward the power plant electric auxiliary system design problem, is described. The auxiliary system design problem was chosen in view of the fact that such systems are generally of limited and well defined size, that approximate analysis techniques can be relied upon, and that a considerable body of prior expertise is available from a number of human experts. The paper first summarizes a few design principles as they relate
This paper was sponsored by the IEEE Power Engineering Society for presentation at the IEEE Power Industry Computer Applica- tion Conference, Montreal, Canada, May 18-21, 1987. Manuscript was published in the 1987 PICA Conference Record.
John F. Jansen Department of Electrical Engineering Viriginia Polytechnic and State University Blacksburg, VA 24060
to the particular auxiliary system design problem. Next, the ASDEP methodology is outlined. An actual design is stepped through, using the specifications of the auxiliary system of a nuclear power plant as starting point. Finally, the possible extensions of the ASDEP methodolgy to other power system design problems are discussed.
SOME KEY OBSERVATIONS RELATED TO ASDEF'
The presently used design methodolgy related to power plant electric auxiliary systems usually evolves around a group of engineers who, after having examined the load and performance requirements of the particular plant, propose a selection of possible configurations. Each configuration is further analyzed by performing various short-circuit, load-flow, transient, and economic studies. A subjective evaluation is carried out based on reliability, flexibility, maintainability, and expandability [ 4 ] . Based on this information, a particular configuration is chosen, equipment specifi- cations are produced and the completed design is once again analyzed. It is important to note here that very little feedback is used during the entire design. ThiE means that generally an entire design is carried out before any "what if" types of questions are raised. Also, in view of the time involved, it is rather rare that entire design procedures be repeated to answer any such "what if" questions. The ASDEP Expert System also operates using the philosophy that an entire design is first completed before fundamental "what if" questions are raised. This is refered to as the no back-tracking assumption. However, in view of the speed of the design process provided by ASDEP, it is indeed feasible and convenient to repeat the entire design under different fundamental design options. This latter observation highlights one of the major benefits of the ASDEP Expert System.
Past experience shows that auxiliary system designs developed by human experts are generally adequate. However, a number of areas are in need of improvements:
- Clearly stated design criteria and system performance criteria are often not properly and explicitely stated, in writing, prior to the design process.
- Power system analysis is often performed based on non-systematic procedures.
- Very few fundamental "what if" questions are addressed in view of the time delavs involved.
- Due to the large staff involved in any design effort, overall system objectives can become unclear and conflicting.
The use of a very systematic approach to the design problem, as required to implement an expert
system, will help alleviate some of the problems mentioned above.
The goal of the ASDEP Expert System is to design a good electrical auxiliary system for a power plant which meets its operational and regulatory requirements. The quality of the design is measured against the criteria of operational performance, reliability, maintainability, flexibility, expandability, and cost. ASDEP was conceived in an effort to demonstrate the feasibility of using Artificial Intelligence techniques for a particular power systems design problem. The objective was not the implementation of a full-scale industrial level program but rather the development of a prototype Expert System and the demonstration of its viability by use of practical design cases.
Limitations of ASDEF'
Since the primary goal of the ASDEP project was the demonstration of feasibility, a few limitations were placed on the auxiliary systems to be designed:
- The functional boundaries of the auxiliary system to be designed need to be clearly defined beforehand.
- Low voltage systems are not explicitely considered (below 1000 V); however, such loads are implicitely considered as lumped loads at the intermediate voltage levels.
- ASDEP is designed to handle only single unit power plants.
- The final output generated by ASDEP is not intended to be a detailed engineering drawing but rather a document describing the fundamental topology of the auxiliary system.
ASDEP M E l ' H O D O ~ Y
The general software organization of the ASDEP Expert System is illustrated in Figure 1. As indicated by the Figure, ASDEP is structured around seven distinct software blocks which are called upon to initiate, update, modify, and finalize the design of the electric auxiliary system for any particular power plant. In this Section, a functional description of each block is provided.
Figure 1. ASDEP Organizational Flowchart
BLOCK 1: Language Processor
The Language Processor provides the user friendly interface between the computer program and the design engineer. It relies on a fixed and problem-oriented vocabulary to allow the user to conveniently input,
display, and modify design data as well as to select and apply the appropriate rule groups during the design procedure. At the present time, the Language Processor, which clearly is an interactive module, uses an alphanumeric display terminal. Future developments will provide for a fully graphical user-machine interface.
BLOCK 2: Input Procedure
Before the actual design of any auxiliary system can be attempted, three specific types of input data must be provided to ASDEP:
Load and system data, which is specific to the particular power plant of interest. This data describes the functional features of the power plant itself as well as the terminal characteris- tics of the network to which the plant is to be connected. For example, the input data required to describe a large induction motor encompasses: motor type, KV rating*, horsepower, demand factor*, distance to the switchyard", efficiency" , running power factor", motor KVA*, high and low voltage limits*, locked-rotor current*, locked-rotor power factor*, running RPM*, X/R ratio*, where the * indicates input data which can be initially omitted by the user and for which default data is subsequently to be provided by the Knowledge Base. All input data provided by the Knowledge Base can be later refined at the request of the user.
Equipment inventory, which provides a list of the types of equipment (breakers, cables, transformers, ..) which may be used during the design stage of the auxiliary system. This list is typically specific for each utility since it reflects the type of equipment which is standardized throughout the utility.
Goal and constraint data, which is specific to the particular power plant considered. This data provides the key reliability indices which must be met by the design (for example, three out of four reactor coolant pumps are needed to ensure normal system operation) as well as the minimum performance specifications to be satisfied (for example, acceptable voltage levels). As a result, this data also reflects the operational philosophy of the particular utility while it must also be such that all regulatory require- ments are fully satisfied by the final design generated by ASDEP.
3: Initial Design Procedure
Contrary to other power system design problems, such as transmission and generation system planning, where the object is to expand and modify an existing system, the auxiliary system design problem starts from a blank piece of paper since the design is to be carried out from "scratch". This situation calls for a rather carefully considered strategy for the generation of the initial design proposal which is to be provided by the Expert System. Two extreme methodologies can be relied upon when considering the initial design:
A highly reliable and very costly initial design is proposed. In this case, the subsequent design refinement stages would consist of "tearing" actions where the system is gradually simplified, to reduce its cost, while still meeting all stipulated reliability and performance constraints. The process would terminate when no further tearing actions can be implemented without violating some stipulated constraints.
1. DIVIDE LOADS INTO M R C W C Y L NON-EKERCENCY LOADS
- A low cost and low reliability and performance design is proposed. In this case, the subsequent design refinement stages would consist of "addition" actions where the system is consolidated, to improve its reliability and performance, while keeping the capital cost requirements at reasonable levels. The process would terminate when all stipulated reliability and performance constraints are met.
The approach taken by human experts is similar to the second one mentioned above and which is also adopted for the Initial Design Procedure of ASDEP.
The Initial Design Procedure flowchart is shown in Figure 2. Some additional comments are appropriate:
- All loads are initially divided into emergency and non-emergency loads based on the information obtained from BLOCK 2. This classification is relied upon during the reliability evaluation procedure discussed in BLOCK 7.
- The load buses are initially created based on the load goals (for example, three circulation pumps must be functional at all times) and based on the KVA rating of the related switchgear.
- The actual loads are distributed on the load buses to balance the KVA ratings, as much as possible, among previously created similar buses. The KVA load on each bus is obtained by simple addition of the actual KVA load connected to that bus.
6. CONNECT DISTRIBUTION EQUIP. IN ICCSIAL CONFIGUIUTION
BLocg 4: Knowledge Base
The Knowledge Base (KB) constitutes the core of the Expert System. A knowledge base is a particular data base where human knowledge is transcribed into computer recognizable rules. This transcription requires a complete and analytical understanding of the expertise accumulated by one or several human experts in the field of interest. This knowledge must be encoded in a form which is amenable to digital processing. Generally, the latter step is handled using the LISP computer language as it is done in ASDEP
The Knowledge Base used in ASDEP is organized around so-called "production rules" where bindings are sought between certain conditions or situations which may occur during the design process. If a particular binding is found, in the KB, between two conditions or situations, then a particular action is executed as called for by the rule which was "triggered". The rules in the KB of ASDEP are classified into two major categories depending on the actions resulting from a successful binding:
- Equipment default rules, which are used to supplement all incomplete data obtained from the user in BLOCK 2. These rules also include the voltage rating selections. At the present time, the KB contains over 60 equipment default rules.
Design change rules, which are used to update and modify any particular defective intermediate design to bring it closer to a satisfactory final design based on the performance and reliability indices. At the present time, the KB contains over 90 design change rules.
ASDEP's evolution with time will primarily focus on the expansion and enhancement of its Knowledge Base. Indeed, as further experience is acquired, additional rules will be added to the KB while existing rules will be updated to better reflect existing design practices and to better address new design problems.
A key portion of the KB is devoted to the crucial voltage selection problem. Several factors affect voltage selections [ 5 ] : load magnitude, distance from the main power supply, utilization device availability (as a function of voltage ratings and limitations), safety, codes and standards, cost.
BLOCK 5: Design Storage
In order to allow the user to follow the evolution of the design, the specifications of each intermediate design stage are stored in three different data sections relating to the loads, the branches (starting and ending buses, along with KVA ratings), and the buses (list of all loads connected to the bus along with KVA ratings). As each design stage is completed, the user is provided with a complete description of the intermediate design, ideally using a graphical display, and a summary of the reliability and/or performance indices which are still violated.
BLOCK 6: Critics
To propose effective design changes, in...