A new control method for grid-connected quasi-Z-source ...scientiairanica.sharif.edu/article_3800_6be28a5d9dc1b84570e3e1fb995c4a56.pdfScientia Iranica D (2015) 22(6), 2505{2515 Sharif

$Page 1: A new control method for grid-connected quasi-Z-source ...scientiairanica.sharif.edu/article_3800_6be28a5d9dc1b84570e3e1fb995c4a56.pdfScientia Iranica D (2015) 22(6), 2505{2515 Sharif$
Scientia Iranica D (2015) 22(6), 2505{2515

Sharif University of TechnologyScientia Iranica

Transactions D: Computer Science & Engineering and Electrical Engineeringwww.scientiairanica.com

A new control method for grid-connectedquasi-Z-source multilevel inverter based photovoltaicsystem

A. Akhavan� and H.R. Mohammadi

Department of Electrical Engineering, University of Kashan, Kashan, Iran.

Received 19 November 2014; received in revised form 5 June 2015; accepted 22 June 2015

KEYWORDSQuasi-Z-sourceinverter;Cascaded multilevelinverter;Photovoltaic system;Particle swarmoptimization;Arti�cial neuralnetwork.

Abstract. In this paper, a new control method for quasi-Z-source cascaded multilevelinverter based grid-connected photovoltaic (PV) system, using evolutionary algorithm andArti�cial Neural Network (ANN), is proposed. The proposed method is capable of boostingthe PV array voltage to a higher level and it solves the imbalanced problem of DC-linkvoltage in traditional cascaded H-bridge inverters using ANN. The proposed control systemadjusts the grid injected current in phase with the grid voltage and achieves independentMaximum Power Point Tracking (MPPT) for the separate PV arrays by Proportional-Integral (PI) controllers. For achieving the best performance, this paper presents anoptimum approach to design the controller parameters using Particle Swarm Optimization(PSO). The primary design goal is to obtain good response by minimizing the integralabsolute error. Also, the transient response is guaranteed by minimizing the overshoot,settling time, and rise time of the system response. E�ectiveness of the new proposedcontrol method has been veri�ed through simulation studies in a seven-level quasi-Z-sourcecascaded multilevel inverter.© 2015 Sharif University of Technology. All rights reserved.

1. Introduction

Photovoltaic (PV) power generation has a great po-tential to serve as a clean and inexhaustible renewableenergy source. However, output power of the PVarrays is greatly a�ected by environmental conditions,such as stochastic changes of the temperature andsolar irradiance. In PV systems, extracting the max-imum power of the PV array and current injectioninto the grid at unity power factor are necessary.In recent years, applying various multilevel invertertopologies to PV systems is getting more and moreattention due to the large power-scale and high voltagedemands. Among various topologies, Cascaded H-

*. Corresponding author. Tel.: +98 3155333908E-mail addresses: [email protected] (A.Akhavan); [email protected] (H.R. Mohammadi)

Bridge (CHB) inverter has unique advantages and hasbeen identi�ed as a suitable topology for transformer-less, grid-connected PV systems [1]. Applying CHBinverter in the PV systems has some advantages, suchas the independent Maximum Power Point Tracking(MPPT) of each array. However, the DC-link voltagein each inverter module is not constant, because PVarray voltage varies with changes of environmentalconditions, such as temperature and solar irradiation orpartial shadows. These cases will cause an unbalancedDC-link voltage among di�erent H-bridge modules.Furthermore, in the conventional Cascaded MultilevelInverter (CMI) based PV system, each module is a buckinverter because the �rst component of the output ACvoltage is always lower than the input DC voltage.Therefore, an additional DC-DC boost converter isnecessary to obtain the desired output voltage and alsoto balance the DC-link voltages. This DC-DC boost

2506 A. Akhavan and H.R. Mohammadi/Scientia Iranica, Transactions D: Computer Science & ... 22 (2015) 2505{2515

converter increases the complexity of the power andcontrol circuit and reduces the e�ciency [2-5].

In recent years, the Z-Source Inverter (ZSI) andQuasi-Z-Source Inverter (QZSI) have been employedfor the PV power generation system due to some uniqueadvantages and features. The ZSI has a discontinuousinput current during the shoot-through state due topresence of the blocking diode [6-8]. Thus, this con-�guration is not appropriate for photovoltaic applica-tions. Nowadays, Quasi-Z-Source Cascaded MultilevelInverter (QZS-CMI) based PV systems were proposedwhich have continuous input current and boostingcapability and also inherit the advantages of traditionalCMI, while overcoming issues with imbalanced DC-link voltages among independent modules [9,10]. Liuet al. [2] use PI controllers for designing the wholecontrol system; but in the presented control method,each QZS-CMI module needs a control loop for DC-link peak voltage control which causes the complexityof the control system. Several works are done relatedto QZS-CMI, for example: Liu et al. [11], Shajithand Kamaraj [12], and Sun et al. [13] present thevarious multi-carrier bipolar PWM techniques for QZS-CMI and Liu et al. [14], and Sun et al. [15] focus onparameter design of the QZS-CMI. The Phase ShiftedSinusoidal Pulse Width Modulation (PS-SPWM) isused as a modulation scheme for QZS-CMI in [16].

In this paper, a new control method for a QZS-CMI based PV system is proposed. The controlobjectives are independent DC-link voltage control,independent MPPT control, and current injection intothe grid at unity power factor. The Arti�cial NeuralNetwork (ANN) and PI controllers are employed tocontrol QZS-CMI. The independent DC-link voltagecontrol is achieved using ANN. Applying the ANN hassome advantages in comparison to using PI controllersin [2]. In this method, the DC-link peak voltage controlloop for each module is removed and only one ANN isused for all modules, which causes simplicity in thecontrol system and cost reduction. Also, PI controllershave a delay time due to integral term. Consequently,by removing the PI controllers in the DC-link peakvoltage control loop and using ANN, the delay timedecreases in the transient conditions. Besides, thereliability of the control system enhances using ANN.In fact, if any input data is out of range due tofailing in the measurements, the ANN gets acceptableoutputs. In this paper, the independent MPPT andcurrent injection are controlled using PI controllers.There have been many technical methods for designingthe control parameters in the time and frequencydomain. Mostly, design of control parameters usingthese methods are time-consuming. To achieve a highand fast performance, this paper presents an optimumapproach to design PI parameters using Particle SwarmOptimization (PSO). In the PSO method, the pri-

mary goal for the design of control parameters is toobtain good steady state response by minimizing theintegral absolute error. Also, the transient responseis guaranteed by minimizing the overshoot, settlingtime, and rise time of the system response. In fact,the objective function is de�ned so that both thesteady state and transient response improve. Also, thePS-SPWM modulation scheme is used for the QZS-CMI. This paper is organized as follows: Section 2consists of an overview of the system with the proposedcontrol strategy; Section 3 focuses on the systemmodeling and grid-connected control; design of the PIparameters using PSO is presented in Section 4; the PS-SPWM modulation scheme is presented in Section 5;e�ectiveness of the proposed control strategy is veri�edby simulation in PSCAD/EMTDC software and theresults are compared with the results of reference [2] inSection 6; and �nally, a conclusion is made in Section 7.

2. QZS-CMI and its control strategy

The QZS-CMI based grid-connected PV system withthe proposed control strategy is shown in Figure 1.Comparing to the conventional CMI module, aninductor-capacitor impedance network is included inthe input stage of each module. This structure is usedto synthesize DC voltage sources to generate 2n + 1staircase output waveform, wherein n is the number ofindependent PV array.

2.1. Quasi-Z-source inverter operationThe QZS-CMI combines the QZS network into eachCMI module. The QZSI can be operated in twomodes, i.e. the non-shoot-through and the shoot-through [16]. Figure 2 shows the QZSI equivalentcircuits operating in the two modes and de�nes thepolarities of all voltages and currents. If the switchingperiod, shoot-through period, and non-shoot-throughperiod are considered as Ts, Tsh, and Tnsh, respectively,the Eq. (1) can be written as:

Ts = Tsh + Tnsh: (1)

Therefore, the shoot-through duty ratio is D = Tsh=Ts.When the kth module is in non-shoot-through state,the power is transmitted from the DC side to the ACside and the inverter operates as a traditional CMI. Insteady state, the following relations can be obtained [2]:

VDCk =1

1� 2DkVPV k = BkVPV k;

VHk = SkVDCk: (2)

While in a shoot-through mode, there is no powertransmission, because the DC-link voltage is zero. Inthis mode, there are:

VDCk = 0; VHk = 0: (3)

A. Akhavan and H.R. Mohammadi/Scientia Iranica, Transactions D: Computer Science & ... 22 (2015) 2505{2515 2507

Figure 1. (a) QZS-CMI based grid-connected PV power system. (b) ANN structure for dc-link peak voltage control.

Figure 2. The equivalent circuit of the QZSI: (a) Non-shoot-through state; and (b) shoot-through state.


Figure 3. Block diagram of the proposed control grid-connected system for the QZS-CMI based PV system.

For the QZS-CMI, the output synthesized voltage isequal to:

VH =nXk=1

VHk =nXk=1

SkVDCk: (4)

In the above equations, VDCk is the kth module DC-link peak voltage; VPV k is the output voltage of thekth PV array; Dk and Bk are the shoot-through dutyratio and boost factor of the kth module, respectively;VHk is the output voltage of the kth module and Sk 2f�1; 0; 1g is the switching function of the kth module.

2.2. Principle of control strategyEach QZS-CMI module has two independent controlcommands: shoot-through duty ratio, Dk, and modu-lation signal, Vmk. Dk is used to adjust the DC-linkvoltage to a desired reference value, while Vmk is usedto control the grid injected power. The main goals ofthe control system of QZS-CMI based grid-connectedPV system are: 1) Independent MPPT for each moduleto ensure the maximum power extraction from each PVarray; 2) Current injection into the grid at unity powerfactor; and 3) Balanced DC-link peak voltage for allQZS-CMI modules.

To achieve these goals, the PI controllers andANN are employed. The total PV array voltage loopadjusts the sum of n PV array voltages using a PIcontroller, PIt. Each PV array voltage reference iscalculated by its MPPT block. Also, the current loopachieves a sinusoidal grid-injected current in phasewith the grid voltage. A Proportional-Resonant (PR)controller makes the actual grid-injected current totrack its desired value [17]. The n� 1 independent PVarray voltage loops control the other n � 1 PV arrayvoltages to achieve their own MPPTs through the n�1PI controllers, named as PI2 to PIn, respectively. Also,as shown in Figure 1(b), the DC-link peak voltage ofeach module is controlled by means of shoot-throughduty ratio which is determined by an ANN. The inputsof ANN are the real PV array voltages and its outputs

are the shoot-through duty ratios for the DC-linkvoltage control system. Finally, the shoot-through dutyratio Dk and modulation signal Vmk for kth moduleare combined into the PS-SPWM modulation schemeto achieve the desired goals.

3. System modeling

The block diagram of the QZS-CMI based grid-connected PV system is shown in Figure 3. The detailswill be explained as follows.

3.1. Independent PV voltage and injectedcurrent control

In the kth QZS-CMI module, the current of theinductor L1 is:

iL1k = iPV k � Cp dVPV kdt; (5)

where iL1k is the current of inductor, L1; iPV k is thecurrent of kth PV array; and Cp is the shunt capacitorwith the PV array. The total output voltage of theQZS-CMI can be written as:

VH = Vg + Lfdisdt

+ rf is; (6)

where Vg is the grid voltage and is is the grid injectedcurrent; rf is parasitic resistance and Lf is the �lterinductance. Consequently, the transfer function of thegrid injected current can be obtained by:

Gf (s) =Is(s)

VH(s)� Vg(s) =1

Lfs+ rf: (7)

To make the actual grid injected current track thedesired reference, a PR controller GPRi(s) = kiP +kiR!0s2+!2

0is used, where !0 is the resonant frequency, i.e.

314 rad/s.In the next step, a grid voltage feed-forward

control loop is applied. Therefore, the kth module hasthe following modulation signal:

Vmk = V 0mk + Vg(s)Gvfk(s); (8)


where, Vmk is the kth module modulation signal; V 0mkis output of the PI controller in the kth module, andGvfk(s) is as follows [2]:

Gvfk(s) =1

nGinvk(s);

Ginvk(s) =VHk(s)Vmk(s)

= VDCk: (9)

Due to DC-link peak voltage balance control, the DC-link peak voltages are equal. Therefore:

Ginvk(s) = Ginv(s); k 2 f1; 2; :::;mg: (10)

According to Figure 3, the closed-loop transfer functionof the grid injected current control system can bewritten in Eq. (11) as shown in Box I. According toFigure 1, each PV array voltage can be obtained by:

VPV k(s) =1Cps

[IPV k(s)� IL1k(s)] : (12)

In the non-shoot-through mode, the output power isequal to input power [2]. Therefore:

isvHk2

= vDCk�iDCK = �PV k�iL1k nsh: (13)

In the above equation, vHk is the output peak voltage ofthe kth module; �iDCk is the average current of the DC-link in the kth module; �iL1k nsh is the average currentof inductor L1 in non-shoot-through mode. Eq. (13)can be rewritten using Eq. (2) as follows:

�iL1k nsh =isvHk

2vDCk(1� 2Dk): (14)

Also, in the shoot-through mode, the average currentof the inductor L1 is:

�iL1k sh = iPV k: (15)

Therefore, the average current of the inductor L1 inthe one switching cycle can be obtained as follows:

�iL1k = Dk�iL1k sh + (1�Dk)�iL1k nsh

= DkiPV k +is(1�Dk)vHk

2vDCk(1� 2Dk): (16)

Figure 4. Block diagram of the total PV voltage loop.

Figure 5. Block diagram of the separate PV voltage loop.

In addition, a PI controller GPIt(s) = kPt+ kIts is used

to track the total reference voltage coming from MPPT.The block diagram of the total PV array voltage loopis shown in Figure 4. Also, for modules 2 to n, a PIcontroller GPIk(s) = kPk + kIt

s is applied to separatePV voltage to achieve their own MPPTs. The blockdiagram of the separate PV array voltage loop is shownin Figure 5.

3.2. Independent DC-link voltage control usingANN

Arti�cial neural network is a model which is inspiredfrom the biological neurons. It has a series of nodeswith interconnections where mathematical functionsand bias factors are applied to do an input/outputmapping. An important feature of ANN that makesit proper for this kind of problems is its exibility tolead in its domain and outside it, as well as work withthe non-linear nature of the problem [18]. Althoughthe data set for the ANN training is not complete andpossibly all combinations are not considered, the ANNwill be able to interpolate and extrapolate the results.These features make ANN a good and reliable choicefor use as a controller.

Generally, the computational speed of ANN ishigh, because only simple mathematical functions ap-ply to each layer. The ANN topology, which is usedin this paper, is shown in Figure 1(b). It is a feed-forward structure which has three layers. The �rstlayer has 20 neurons and also, second layer has 10neurons. The number of neurons in the output layeris equal to the number of outputs. This topology has

Gic =Is(s)I�s (s)

=GPRi(s)Gf (s)Ginv(s)

1 +GPRi(s)Gf (s)Ginv(s)

=VDCk

�kiPS2 + kip!2

0 + kiR!0�

Lfs3 + (rf + VDCkkip)s2 + Lf!20s+ VDCkkip!2

0 + VDCkkiR!0 + rf!20

(11)

Box I


tangent-sigmoid activation function hidden layers and,also, a linear activation function output layer. ThisANN takes the real value of PV array voltages valuesas inputs and gives the shoot-through duty ratios forthe control system as outputs. If the range of PVarray voltage variation is considered 110 to 130 V anda step of 1V for its variation, then we have 21 points foreach array leading to a data-set of a size equal to 213.For input data, the output of the ANN, which is theshoot-through duty ratio of PV modules, is calculatedby using Eq. (2). After determining the inputs andoutputs, the back propagation learning algorithm canbe used for ANN training [19].

4. Design of PI controller parameters usingparticle swarm optimization

The PI is a well-known controller for industrial pro-cesses. This is due to its good performance and simplestructure in a wide range of operating conditions.Tuning such a controller requires speci�cation of twoparameters: proportional gain Kp and integral gainKi [20]. In the past, this problem has been solved bya trial and error technique. In this paper, the problemof PI parameters tuning is formulated as an opti-mization problem. The problem formulation employsfour performance indexes, i.e. the overshoot, settlingtime, rise time, and integral absolute error of the stepresponse as the objective function terms to tune thePI parameters for getting the best performance in agiven plant. In this study, the primary design goal isto obtain good steady state response by minimizingthe integral absolute error. At the same time, thetransient response is guaranteed by minimizing theovershoot, settling time, and rise time of the stepresponse. For solving this optimization problem, theparticle swarm optimization algorithm is used in thispaper.

The PSO is a stochastic optimization techniquewhich uses swarming behaviors observed in ock ofbirds. In fact, the PSO was inspired by the sociologicalbehavior associated with swarms. PSO was developedby James Kennedy and Russell Eberhart in 1995 asa new heuristic method [21]. It uses a number ofparticles that constitute a swarm moving around inone N -dimensional search space looking for the bestposition. The individuals in the swarm are calledparticles. Each particle in the PSO algorithm is apotential solution for the optimization problem andkeeps track of its coordinates in the problem space andtries to search for the best position through ying ina multidimensional space, which are associated withthe best solution (called best �tness) it has achievedso far and is called \pbest". Another \best" valuecalled \gbest", that is tracked by the global versionof the particle swarm optimizer, is the overall best

value and its location obtained so far by each particlein the swarm. The sociological behavior, which ismodeled in the PSO, is used to guide the particles.Each particle is determined by two vectors in the N -dimensional search space: The position vector andthe velocity vector [22]. Each particle investigates itssearch space through previous experience, its presentvelocity, and the experience of the neighboring parti-cles. The PSO concept is the change in the velocityof each particle toward its pbest and gbest positionsat each iteration. In the PSO, acceleration of eachparticle is weighted by a random term [23]. PSO isa history-based algorithm such that in each iteration,particles use their own behavior associated with theprevious iterations. Let X and V denote the particle'sposition vector and its corresponding velocity vectorin the search space. In this algorithm, at iterationh, each particle i has its position de�ned by Xh

i =[Xi;1; Xi;2; :::; Xi;N ] and also, its velocity vector de�nedby V hi = [Vi;1; Vi;2; :::; Vi;N ] in the search space. In thenext iteration, position and velocity of each particlecan be calculated as:

V h+1i;n = W � V hi;n + C1 � rand1 � �pbesti;n �Xh

i;n�

+C2 � rand2 � �gbestn �Xhi;n�;

i = 1; 2; :::;m; n = 1; 2; :::; N; (17)

Xh+1i;n =Xh

i;n + V h+1i;n if Xmin;i;n � Xh+1

1 � Xmax;i;n;

= Xmin;i;n if Xh+1i;n < Xmin;i;n;

= Xmax;i;np if Xh+1i;n > Xmax;i;n: (18)

The transient response is very important, therefore,both the overshoot and time duration of the transientresponse must be kept within tolerable limits. Hence,four indexes of the transient response are utilized tocharacterize the performance of PI control systems.

The objective function (fTotal) for optimal designof PI controller parameters can be formulated asfollows:

fTotal = fo + fRT + fST + fIAE ; (19)

where fo, fRT , fST and fIAE are the overshoot, risetime, settling time, and integral absolute error ofthe step response, respectively. To apply PSO fortuning the PI controller parameters, the closed-looptransfer function of the total PV voltage loop andseparate PV voltage loop are necessary. These closed-loop transfer functions which are calculated usingblock diagram of Figures 4 and 5, are included inAppendix A. Also, the parameters used in PSO aregiven in Appendix B.


Figure 6. Modulation scheme for the proposed system.

5. The PS-SPWM for QZS-CMI

The modulation technique applied in the proposedsystem is a phase shifted sinusoidal pulse width mod-ulation which is shown in Figure 6. The shoot-through states are inserted by the simple boost controlmethod [16]. In this control method, two straightlines, which are denoted as 1 � Dn and Dn � 1,envelops equal to or greater than the peak value of thesinusoidal reference signals are used to control shoot-through duty ratio. If the triangular carrier signal issmaller than Dn � 1 or bigger than 1 � Dn, the twoswitches of one leg in H-bridge module are turned onsimultaneously. In PS-SPWM schemes, the number oftriangular carrier waves is equal to m � 1, where m isequal to level number. Also, the required phase shiftsamong di�erent carriers is given as:

' =360�m� 1

: (20)

Therefore, in this case (m = 7), triangular carriers ofdi�erent H-bridge modules are shifted 60� with respectto each other.

6. Simulation results

Performance evaluation of the proposed control strat-egy was carried out by simulation in PSCAD/EMTDCsoftware. The parameters of QZS-CMI are shown inTable 1. Also, the codes of PSO algorithm are written

Table 1. QZS-CMI parameters.

QZS-CMI parameters Value

QZS inductance, L1 and L2 1.8 mHQZS capacitance, C1 and C2 3300 �F

PV array parallel capacitance, Cp 1100 �FFilter inductance, Lf 1 mHCarrier frequency, fc 5 kHz

Grid voltage 220V/50 Hz

Table 2. The parameters of the control system obtainedby PSO algorithm.

Parameters Value Parameters Value

kiP 0.00491 kiR -0.01673kPt 1.1511 kIt 1.5348kPk 0.0263 kIk 0.0017

in MATLAB for the best results to be obtained for thecontrol parameters. As mentioned earlier, the closed-loop transfer function of the total PV voltage controlloop and the separate PV voltage control loop are usedfor the optimization problem. The results of the PSOalgorithm are shown in Table 2. E�ectiveness of theproposed method is shown by comparing the resultsof the new proposed method with the results of [2] inTable 3. In this table, the overshoot, settling time, risetime, and integral absolute error of the step responsefor these closed-loop transfer functions are given. Asshown in this table, the overshoot, settling time, risetime, and integral absolute error of step response for allthe closed-loop transfer functions, obtained using PSOalgorithm, are smaller than the results of [2]. Also, thenntool toolbox in MATLAB is used for ANN training.

The P-V characteristic of the PV array is shownin Figure 7. The measured voltage and current foreach PV array are used to apply the MPPT searchalgorithm for determining the PV voltage reference atthe MPP. Incremental conductance algorithm is usedfor tracking the maximum power point of a PV arrayin this paper [24].

At �rst, all modules are working at 1000 W/m2

and 25�C conditions and all the initial voltage refer-ences of MPPT algorithm are given at 120 V. The peakvalue of DC-link voltage in all modules is controlled at145V. After 1 second, the third module's irradiationdecreases to 700 W/m2. Hence, according to Figure 7,the reference voltage of this module is decreases toabout 119.2V by MPPT algorithm. It should be notedthat the change of temperature mostly a�ects thevoltage of maximum power point, so that temperature

Figure 7. PV array power-voltage characteristic.


Table 3. Comparison of the results of the new proposed method with the results of [2].

Transferfunctions

Overshoot(%)

Settlingtime (sec)

Risetime (sec)

Integralabsolute error

Usingcontroller

parametersobtainedby PSO

Usingcontroller

parametersin [2]

Usingcontroller


Usingcontroller

parametersin [2]

Usingcontroller


Usingcontroller

parametersin [2]

Usingcontroller


Usingcontroller

parametersin [2]

Total PVvoltage loop

8.62 56.1 0.362 2.84 0.0395 0.177 0.0384 0.5052

Separate PVvoltage loop

12.9 33.3 0.438 0.511 0.0578 0.0651 0.0584 0.0988

Figure 8. Simulation results: (a) Total PV array voltage; (b) PV array voltage of the �rst module; (c) PV array voltageof the second module; and (d) PV array voltage of the third module.

rising causes the voltage of maximum power point todrop. While, the change of solar irradiation mainlya�ects the current injection.

The total PV voltage (sum of all three PV arrayvoltages) and other PV array voltages are shown inFigure 8(a)-(d), respectively. It can be seen that duringthe change in the MPPT reference value due to thechange in environmental condition, the proposed con-trol method has excellent tracking performance aftera very short transient. It can be seen in Figure 8(d)that after a change in the environmental condition ofthird module, the controller of this module tracks thenew reference. As shown in Figure 8(b)-(d), PV arrayvoltage of the �rst module is di�erent with respectto the second and third modules. This is due to thefact that the modulation signal generation of the �rstmodule is di�erent from that of other modules.

The inverter output voltage is shown in Fig-ure 9. It can be indicated that the solar irradiationor temperature does not a�ect the seven-level staircaseoutput voltage of the inverter due to shoot-throughoperation. It is no matter if the temperature and thesolar irradiation change or not; the QZS-CMI always

Figure 9. Inverter output voltage.

outputs a constant voltage which veri�es the voltagebalancing capability.

Figure 10 shows the grid-injected current which isexactly in phase with the grid voltage and ensures unitypower factor. It should be noted that lower irradiationcauses reduction of the grid-injected current. The DC-link voltage of the third module is shown in Figure 11.Low DC-link peak voltage uctuation is because ofcapacitor voltages uctuation due to inverter statuschange from the shoot-through mode to non-shoot-through mode and vice versa. It can be seen thatwith the ANN, DC-link peak voltages are kept atthe reference value (145 V). It should be noted that,


Figure 10. Grid voltage and current.

Figure 11. DC-link voltage of the third module.

Figure 12. Shoot-through duty ratio of the third module.

according to Eq. (2), after decrement of the referencevoltage of the third module by MPPT algorithm, longershoot-through time interval is necessary for this moduleto �x the DC-link peak voltage at 145 V. The shoot-through duty ratio of third module, during the changein the MPPT reference value, is shown in Figure 12.As shown in this �gure, ANN adjusts new shoot-through duty ratio after a very short transient, sothat third module's DC-link peak voltage has a verylow distortion after a change in the solar irradia-tion.

7. Conclusion

In this paper, a new control method for QZS-CMIbased single-phase grid-connected PV system is pro-posed. The proposed system is a combination of QZSIand CHB multilevel topology and has both of theiradvantages. This enables independent MPPT controlfor each PV array in di�erent conditions. Moreover,independent DC-link voltage control enforces all QZS-HBI modules to have balanced voltages and, also, grid-injected current is ful�lled at unity power factor. Thecontrol parameters are designed using PSO algorithm.It was shown that the system has fast and accurate re-

sponse. Also, an ANN used for DC-link voltage controlhas some advantages such as simplicity, cost reduction,fast response, and more reliability. Performance of theproposed control system for a seven level QZS-CMIis evaluated under variable environment conditions.The simulation results show the e�ectiveness of theproposed control strategy for a QZS-CMI based single-phase grid-connected PV system.

References

1. Sastry, J., Bakas, P., Kim, H., Wang, L. andMarinopoulos, A. \Evaluation of cascaded H-bridgeinverter for utility-scale photovoltaic systems", Renew-able Energy, 69(1), pp. 208-218 (2014).

2. Liu, Y., Ge, B., Abu-Rub, H. and Peng, F.Z. \Ane�ective control method for quasi-Z-source cascademultilevel inverter-based grid-tie single-phase photo-voltaic power system", IEEE Transactions on Indus-trial Informatics, 10(1), pp. 399-407 (2014).

3. Liu, Y., Ge, B., Abu-Rub, H. and Sun, D. \Compre-hensive modeling of single-phase quasi-Z-source pho-tovoltaic inverter to investigate low-frequency voltageand current ripple", IEEE Transactions on IndustrialElectronics, 62(7), pp. 4194-4202 (2015).

4. Patidar, K. and Umarikar, A.C. \High step-up pulse-width modulation DC-DC converter based on quasi-Z-source topology", IET Power Electronics, 8(4), pp.477-488 (2015).

5. Liu, S., Ge, B., Jiang, X., Abu-Rub, H. and Peng, F.Z.\Comparative evaluation of three Z-source/quasi-Z-source indirect matrix converters", IEEE Transactionson Industrial Electronics, 62(2), pp. 692-701 (2015).

6. Peng, F.Z. \Z-source inverter", IEEE Transactions onIndustry Applications, 39(2), pp. 504-510 (2003).

7. Loh, P.C., Vilathgamuwa, D.M., Lai, Y.S., ChuaG.T. and Li, Y. \Pulse-width modulation of Z-sourceinverters", IEEE Transactions on Power Electronics,20(6), pp. 1346-1355 (2005).

8. Miaosen, S., Joseph, A., Jin, W., Peng, F.Z. andAdams, D.J. \Comparison of traditional invertersand Z -source inverter for fuel cell vehicles", IEEETransactions on Power Electronics, 22(4), pp. 1453-1463 (2007).

9. Yuan, L., Anderson, J., Peng, F.Z. and Dichen, L.\Quasi-Z-source inverter for photovoltaic power gener-ation systems", Applied Power Electronics Conferenceand Exposition (APEC 2009), pp. 918-924 (2009).

10. Nguyen, M., Jung, Y. and Lim, Y. \Single-phase AC-AC converter based on quasi-Z-source topology", IEEETransactions on Power Electronics, 25(8), pp. 2200-2210 (2010).

11. Colak, I., Kabalci, E. and Bayindir, R. \Review ofmultilevel voltage source inverter topologies and con-


trol schemes", Energy Conversion and Management,52(2), pp. 1114-1128 (2011).

12. Liu, Y. Abu-Rub, H. Ge, B. and Peng F.Z. \Phase-shifted pulse-width-amplitude modulation for quasi-Z-source cascade multilevel inverter based PV powersystem", Energy Conversion Congress and Exposition(ECCE), pp. 94-100 (2013).

13. Shajith, A. and Kamaraj, V. \Bipolar multicarrierPWM techniques for cascaded quasi-Z-source multi-level inverter", Circuits, International Conference onPower and Computing Technologies (ICCPCT 2013),pp. 236-240 (2013).

14. Sun, D., Ge, B., Yan, X., Bi, D., Zhang, H., Liu,Y., Abu-Rub, H., BenBrahim, L. and Peng, F.Z.\Modeling, impedance design, and e�ciency anal-ysis of quasi-Z-source module in cascade multilevelphotovoltaic power system", IEEE Transactions onIndustrial Electronics, 61(11), pp. 6108-6117 (2014).

15. Liu, Y., Abu-Rub, H., Ge, B. and Peng, F.Z.\Impedance design of 21-kW quasi-Z-source H-bridgemodule for MW-scale medium-voltage cascaded multi-level photovoltaic inverter", International Symposiumon Industrial Electronics (ISIE 2014), pp. 2490-2495(2014).

16. Sun, D., Ge, B., Peng, F., Abu-Rub, H., Bi, D. andLiu, Y. \A new grid-connected PV system based oncascaded H-bridge quasi-Z source inverter", IEEE In-ternational Symposium on Industrial Electronics (ISIE2012), pp. 951-956 (2012).

17. Zmood, D. and Holmes, D. \Stationary frame currentregulation of PWM inverters with zero steady-state er-ror", IEEE Transactions on Power Electronics, 18(3),pp. 814-822 (2003).

18. Bose, B. \Neural network applications in power elec-tronics and motor drives - An introduction and per-

spective", IEEE Transactions on Industrial Electronic,54(1), pp. 14-33 (2007).

19. Karnin, E. \A simple procedure for pruning back-propagation trained neural networks", IEEE Transac-tions on Neural Networks, 1(2), pp. 239-242 (1990).

20. Hsiao, Y., Chuang, C. and Chien, C. \Ant colonyoptimization for designing of PID controllers", IEEEInternational Symposium on Computer Aided ControlSystems Design, pp. 321-326 (2004).

21. Kennedy, J. and Eberhart, R. \Particle swarm opti-mization", IEEE International Conference on NeuralNetworks, pp. 1942-1948 (1995).

22. Taghizadeh, H. and Hagh, M.T. \Harmonic elimi-nation of cascade multilevel inverters with nonequalDC sources using particle swarm optimization", IEEETransactions on Industrial Electronics, 57(11), pp.3678-3684 (2010).

23. Clerc, M. and Kennedy, J. \The particle swarm - explo-sion, stability, and convergence in a multidimensionalcomplex space", IEEE Transactions on EvolutionaryComputation, 6(1), pp. 58-73 (2002).

24. Salas, V., Ol�as, E., Barrado, A. and L�azaro, A.\Review of the maximum power point tracking algo-rithms for stand-alone photovoltaic systems", SolarEnergy Materials and Solar Cells, 90(11), pp. 1555-1578 (2006).

Appendix A

Closed-loop transfer functions are calculated usingblock diagram of Figures 4 and 5 as shown in Box II.

Appendix B

The parameters used in PSO algorithm is shown inTable B.1.

PVPV kPV �PV k

=

�kPtkiP s3 + kItkiP s2 + kPt(kiP!2

0 + kiR!0)s+ kIt�kiP!2

0 + kiR!0�� 1�D1

2(1�2D1) + 1�D22(1�2D2) + 1�D3

2(1�2D3)

�CPLfs

5 + CP (VDCkiP + rf )s4 +�CPLf!

20 + kPtkiP � (

1�D1

2(1� 2D1)+

1�D2

2(1� 2D2)+

1�D3

2(1� 2D3))�s3+

+�CP (VDCkip!2

0 + VDCkiR!0 + rf!20) + kItkiP

�1�D1

2(1� 2D1)+

1�D2

2(1� 2D2)+

1�D3

2(1� 2D3)

��s2+

+ kPt�kiP!2

0 + kiR!0

�� 1�D1

2(1� 2D1)+

1�D2

2(1� 2D2)+

1�D3

2(1� 2D3)

�s

+ kIt�kiP!2

0 + kiR!0

�� 1�D1

2(1� 2D1)+

1�D2

2(1� 2D2)+

1�D3

2(1� 2D3)

�

(A.1)

VPV kV �PV k

=kPk

�1�Dk1�2Dk

�is2 s+ kIk

�1�Dk1�2Dk

�is2

CP s2 + kPk�

1�Dk1�2Dk

�is2 s+ kIk

�1�Dk1�2Dk

�is2

(A.2)

Box II


Table B.1. PSO parameters.

Particle no 100NPar 5

MaxIter 80w 0.9

KEI 0.1n1 2n2 2

Biographies

Ali Akhavan was born in Kashan, Iran in 1990. Hereceived the BSc and MSc degrees in Electrical Engi-neering from the University of Kashan, Kashan, Iran,

in 2012 and 2014, respectively. He is currently workingtoward the PhD degree in Electrical Engineering, in theUniversity of Kashan. His research interests includepower electronics, renewable energy power generation,power quality, and microgrids.

Hamid Reza Mohammadi was born in Qom, Iran.He received the BSc degree in Electrical Engineeringfrom Sharif University of Technology, Tehran, Iran, in1993, the MSc degree in Electrical Engineering from theUniversity of Tabriz, Tabriz, Iran, in 1995, and the PhDdegree in Electrical Engineering from Tarbiat ModaresUniversity, Tehran, in 2008. His research interestsinclude power quality, digital signal processing, andpower electronics.

Documents

A new control method for grid-connected quasi-Z-source ...scientiairanica.sharif.edu/article_3800_6be28a5d9dc1b84570e3e1fb995c4a56.pdfScientia Iranica D (2015) 22(6), 2505{2515 Sharif