276

IEEE Transactions on Energy Conversion, Vol. 13, No. 3, September 1998

WO-Loop Con

aximizing ~ e r f o ~ m a n c o e Grid-Connected -Fuel Cell IHybrid Power

Kyoungsoo Ro, Student Member, IEEE Saifur Rahman, Senior Member, IEEE Center for Energy and the Global Environment Virginia Polytechnic Institute and State University Blacksburg, VA 2406 1, U. S.A e-mail: [email protected]
Abstract: Maximizing performance of a grid-connected photovoltaic (PV)-fuel cell hybrid system by use of a two-loop controller is discussed. One loop is a neural network controller for maximum

power point tracking, which extracts maximum available solar power from PV arrays under varying conditions of insolation, temperature, and system load. A reallreactive power controller (2) is the other loop. The RRFT achieves the systems requirements for real and reactive powers by controlling incoming fuel to fuel cell stacks as well as switching control signals to a power conditioning subsystem. Results of time-domain simulations prove not only effectiveness of the proposed computer models of the two-loop controller but also its applicability for use in stability analysis of the hybrid power plant. Keywords: photovoltaics, fuel cells, neural network, maximum Dower Doint tracking controller. real/reactive Dower control
Y

I. INTRODUCTION As conventional fossil-fuel energy sources diminish and the worlds environmental concern about acid deposition and global warming increases, renewable energy sources (solar, wind, tidal, and geothermal, etc.) are attracting more attention as alternative energy sources. Among them solar photovoltaic (PV) energy has been widelyutilized in small-size applications and is the most promising candidate for research and development for large-scale use, as the fabrication of less costly photovoltaic devices becomes a reality. PV power generation, which directly converts solar radiation into electricity, has lots of significant advantages. For example, one positive environmental effect of photovoltaics is replacing electricity generated in more polluting way. Another is to provide electricity where none was available before. With increased penetration of solar photovoltaic devices, various antipollution apparatus can be

PE-800-EC-0-06-1997A paper recommended and approved by the IEEE Energy Development and Power Generation Committee of the IEEE Power Engineering Society for publication in the lEEE Transactions on Energy Conversion. Manuscript submitted December 20, 1996, made available for printing June 6, 1997

operated by solar PV power; for example, water purification through electrochemical processing and stopping desert expansion by photovoltaic water pumping with tree plantaI tion Ill]. From an operational point of view, a photovoltaic array may experience large variances of its output power under variable weather conditions resulting in control problems. Fig. 1 illustrates a sample of PV power variations for ope day. The data was measured at a time interval of 10 minutes from a PV test facility in the Virginia Tech campus on July 18, 1993. When photovoltaic penetrations approach about 10% of the system load, it is dfiicult for the conventional power generating system to keep track of rapid photovoltaic output changes [XI. One method to overcome the above problem is to integrate PV power plants with other power sources; for example, diesel backup [14], fuel cell backup [1,2], battery backup [6,9,12], and superconductive magnetic energy storage (SMES) backup [lo]. The diesel backup for PV systems is able to make a continuous 24-hour power supply possible, but its severe drawbacks are that its efficiency decreases significantly at low levels of power output, and diesel power generation is environmentally detrimental as well. SMES technology is many years from commercialization, and there are significant potential health risks associated with this technology because of strong magnetic fields. Fuel cells are a very attractive option to be used with intermittent sources of generation like the photovoltaics because of high efficiency, fast load-response, modularity and fuel flexibility. Their feasibility in coordination with photovoltaic systems has been successfully demonstrated for both grid-connected and stand-alone applications [ 11. Due to the fast ramping capabilities of fuel cells, the PV-fuel cell hybrid system is able to smooth out the PVs inherent yroblem of intermittent power generation. Unlike a storage battery, which also contains attractive attributes such as fast response rate, modular construction, and flexibility for site selection, fuel cells can produce electricity continuously to support the PV power generation so that the quality of overall power fed into the utility grid would be improved. l The purpose of this study is to design an efficient controller for a PV-fuel cell hybrid power plant so that the combination can be used as a reliable power source. Such a source can be used to serve design loads either on a standalone or a grid-interactive mode. The controller, consisting

0885-8969/98/$10.00

1997 IEEE

277
300 250

g
s

200

control signals to throttle the valve regulating the fuel fed into the fuel cell stacks and to adjust the switching control scheme for the inverter connected to the fuel cell system in order to control real and reactive output powers.

9 150

B 2

5 100
50

111. NEURAL NETWORK CONTROLLER FOR MPPT A typical interconnection of the grid-connected PV system using two, 6-pulse dc-ac inverter is shown in Fig. 3. The PV array is made up of 80 series-connected and 480 parallel-connected ARC0 Solar M5S modules and produces 2 MW at 1400 V when solar insolation is 100 mW/cm2 and cell temperature is 25 O C . Photovoltaic power generator requires so much larger initial cost compared to other power generation sources that it is imperative to extract as much available solar power as possible from the PV array; if not, the system would lose valuable solar energy. Maximum power output of the PV array varies when solar insolation, temperature, and (or) load level change. Control is, therefore, needed for the PV generator to always keep track of the maximum power points. Such a controller is implemented by a neural network (NN) approach. Nonlinear I-V characteristics of a PV module match well with a neural network application. In Fig. 4, a configuration of a multi-layer feedforward perceptron-type NN is proposed for the MPPT control. The network consists of an input layer, one hidden layer and an output layer. The number of neurons in the hidden layer was determined by trial and error. The neurons in the hidden layer for the neural network

0
-50 Time [hr]

Fig. 1. S'unple of PV power output v'uiations.

of two loops, is able not only to achieve maximum utilization of available solar power but also to satisfy operational requirements of the hybrid plant for real and reactive power control. 11. SYSTEM DESCRIPTION AND METHODOLOGY Fig. 2 illustrates a simplified diagram of a gridconnected PV-fuel cell hybrid power plant including two loops of controllers, that is, a neural network controller for maximum power point tracking (MPPT) and a real/reactive power controller (RRPC). The hybrid system consists of a photovoltaic array, a fuel cell module and reformer, and power conditioning subsystems (PCSs) which consist of dcac inverters. The photovoltaic generator operates independently and is controlled to produce maximum available solar power by the neural network controller. The fuel cell generator is used as a supplement to the photovoltaic system and is controlled by the RRPC so as to meet the system's required power demand. The PV array, comprised of a set of series-parallel connected solar cells, generates dc power that is then converted to useful ac power by the PCS, which acts as the interface between the photovoltaic system and the ac network. The output ac power of the PCS is provided to the load through transformers and interconnecting reactances. The neural network controller for MPPT receives as inputs the available insolation level and cell temperature. Then it sends out output control signals to the dc-ac inverter connected to the PVs, and hence the photovoltaic generator always generates the maximum available solar power. The fuel cell reforming system processes raw natural gas fuel through desulfurization, steam reforming and the CO shift process to produce a hydrogen-rich fuel. About SO% H2, 20% COz and a small amount of residuals are supplied to the fuel cell stacks. The reformed fuel and oxidant or O2 provoke the electrochemical reaction inside the stacks to generate the dc power. The inputs to the RRPC are real and reactive power commands, output powers of the PVs and the fuel cells. The outputs of the RFWC are used as

Power Controller
Reformer

Pam

Qcom

Fig. 2. Sitiiplifiedoverall diagram of PV-fuel cell hybrid system.

E
Array

Fig. 3. Diagram of the grid-connected PV system.

278

receive outside inputs, calculate their outputs using the sigmoid activation function, then pass them to those in the output layer. The network outputs are generated by the neurons in the output layer. The input-output training data of the neural network are obtained from the characteristics of the solar cells and interconnected system values. Updating the weights is performed using error back-propagation algorithm with the steepest descent method in order to minimize the sum of the mean squared errors. The neural network uses solar insolation and cell temperature to generate switching control signals to the dc-ac inverter. The outputs of the network are used to adjust the inverters switching scheme so that the PV system not only provides its maximum available power to a utility, but also maintains its reactive power transferred to the utility constant. In this study that value is set to be zero. Overall dynamics of the grid-connected PV system are investigated via an appropriate model of the PV and its MPPT control system, which is represented in Fig. 5 . A mathematical model of the NN can be obtained by looking into the relation between its inputs and outputs. The outputs are determined for different solar insolations, h , and cell temperatures, T, , through an off-line curve fitting method as
Vs, = 0.844 / f i + 0.0008h + 0.0026Tc + 0.57075 V41 = -0.00108h

control signals and provides the utility grid with required real and reactive powers.

where
Vpn =

*
7I

NVp,V,, cos 15

and

N : turn ratio of the transformers X : interconnecting reactance

Effectiveness of the model developed in the above is evaluated through programming the corresponding equations on a digital computer. Transient responses of the PV system following an insolation increase from 30 to 100 dW/cm2 is depicted in Fig. 6, which shows plots of the fdllowing variables: I,, - PV array current, Vpv- PV array voltage, Ppv - DC power output of the PV array, Pat,] - PV real power supplied to the grid, and Qac,l - PV reactive phwer supplied to the grid. At the instant the insolation increases, the PV array current increases instantaneously whereas the array voltage varies continuously because of the existence of the shunt capacitor, C d ,in parallel with the PV array. From Fig. 6 (c), the PV output power follows very accurately the theoretical maximum power line, which pkoves applicability of the proposed neural network controlldr to real PV power plants.

+ 0.000476T, + 1.0124

(1) (2)

The dc-ac inverter receives those NN outputs as switching

N N Controller

Fig. 4. NN controller for MPPT of the photovoltzic power generator.

IV. REAL/REACTIVE POWER CONTROLLER ( W C ) A 2-MW phosphoric acid fuel cell (PAFC) power plant aCcompanies the PV power pkmt discussed above in order to sktisfj the systems demand for real and reactive powers, The fuel cell power plant consists of 8 sets (4 in series, 2 in parallel) of 260-kW fuel cell stacks, and each stack is comprised of 5 16 single cells connected in series. Fig. 7 shows voltage descriptions for the PV-fuel cell hybrid system connected to the power system grid through interconnecting reactors. When the interconnecting reacare assumed to be equal, the real and tors, X, and X f ,
I

Fig. 5 . Computer model of the PV ,and MPPT control system.

reactive powers supplied to the grid are written as

279
(9)

$pn

In the neural network controller for MPPT, Vpnand were set to certain values for maximum power point

tracking of the PV array. Those variables are regarded as constant for the purpose of realheactive power control. It is Qnand 4 f i that remain control variables in this situation.

A diagram of the shown in Fig. 8. If the W control mode, the reactive tor, Qat,,, is added to Qac,2.

realheactive power controller is C operates in the reactive power power output of the PV generaThe sum of those reactive power

outputs is compared to the reactive power command and the error is sent to the angle controller to generate an appropriate control signal to the dc-ac inverter, If the RRPC operates in the real power control mode, is compared to the real power the sum of Pat,, and command and the error goes to the phase controller for generating a switching control signal to the inverter. The real power error is also supplied to the valve controller to control the fuel flow into the fuel cell stacks. Fig. 9 illustrates a computer model of the fuel cell stacks and RRPC system. Real and reactive power commands, in the simulation, are set to 2.4 MW and 1 MVar, respectively. Control signals for the inverter are generated through PI-type control loops. The real and reactive powers supplied to the grid are calculated using the control signals and the fuel cell voltage.

1150

4
I
1.17 149 1.52 1.6 17 1.79 1.88 1.97

where

1600

B 800
O 2

- I 400
1 I17 149 152 16 17 179 188 197

Transient responses of the fuel cell system after the corresponding insolation change in the PV array is displayed in Fig. 10, which shows plots of the following vari-

@I

* 50E+06
a

--

Theoretical Pmax

Fig. 7. Voltilge notiitionsfor the PV-fuel cell hybrid system.

g 150E+06 -~
100E+06 --

0 00E+00 -I

2.00E+06

Angle

.-.
0

1.50+06

6. 1.00+06 5.00E+05

O.OOE+OO

1
f

1.17

1.49

1.52

1.6

1.7

1.79

1.88

1.97

Time (sec)

Fig. 6. Transient responsesof the PV power plant for insolation increase 6om 30 to 100 mW/cm2.

Fig. 8. Real/reactivepower control system of the firel cell power plant

280

V. CONCLUSION The paper contributes in i d e n a i n g the need for solving control problems for integrating a PV-fuel cell hybrid power plant to a utility grid by using a two-loop controller. The positive aspect that power generation of a PV plant mostly takes after a systems peak demand is less appealing due to its feature of intermittent power outputs on a fluctuating weather condition. Combining a fuel cell power plant with a PV plant had already been proven to be feasible in order to attack such a PVs inherent problem [ 1J , A neural network based controller is adopted to keep track of maximum power points of the PV array because its nonlinear I-V characteristics match well with a NN application. Test simulations revealed that the proposed method is able to make the PV array accurately track the peak power conditions even at times of rapidly changing insolation. A PI-type controller is proposed for the fuel cell power plant to satisfy the systems requirements for real and reactive powers. Satisfactory dynamic responses verify effectiveness of the developed control scheme as can be seen from the computer simulation results.

Fig. 9. Computer model ofthe he1 cell plant and real/ reactive power control.

1380

4
1 1.17 149
152

I
16

17

179

186

197

(a)
2 00E+06

VI. APPENDIX A. MODEL OF A PV ARRAY

Qac2

150E+06 -I .00E+06 -

5 OOE+OS
0 00E+00

The characteristics of a solar cell relating the cells voltage to current are expressed by the following equations 131.

--

Pac2
1.52

4
1
117 149 1.6 17 179
188

197

(b)
3 00E+06 2 40E+06

or

where
6 00E+05
0 00E+00

117

149

152

16

17

179

188

1.97

Time (sec)

Fig. 10. Transient responses ofthe fuel cell power plant for the corresponding insolation change in the PV system.

and PV cell output current (A)

Terminal voltage for a PV cell (V)

ables : V, - fuel cell plant voltage, Ifc fuel cell plant current, Pac,2- fuel cell real power supplied to the grid, fuel cell reactive power supplied to the grid, Pa, total real power from PV and fuel cell plants, and Qoc total reactive power from PV and fuel cell plants. Fig. 10 (c) reveals that the fuel cell power plant responds very well to the instant change of insolation in the PV array and the PV-fuel cell hybrid system could be combined to run very effectively.

light-generated current in a PV cell (A) PV cell saturation current (A) Ideality factor (1.60) PV cell temperature ( K ) Boltzmanns constant ( 1 . 3 8 0 5 ~ 1 0 / ~ O K ~~ ) Electron charge ( l.GxlO- @) Series resistance of a PV cell (0.0) PV cell short-circuit current at 25 C and 100

281 mW /an2 (3.27 A) Short-circuit current temperature coefficient at I,. (0.0017 A loC )
B.K. Bose, P.M. Szczesny and R.L. Steigerwald, Microcomputer Control o f a Residential Photovoltaic Power Conditioning System, IEEE Transactions on Industry Applications, Vol. IA-21, No. 5, SeptembedOctober 1985, pp. 1182-1191. 0. Wasynczuk, Modeling and Dynamic Performance of a LmeCommutated Photovoltaic Inverter System , I E E E Transactions on Energy Conversion, Vol. 4, No. 3, September 1989, pp. 337-343. 0. Wasynczuk and N.A Anwah, Modeling and Dynamic Performance of a Self-Commutated Photovoltaic Inverter System, I E E E Transactions on Energy Conversion, Vol. 4, No. 3, September 1989, pp. 322-328. K.C. Kalaitzakis and G.J. Vachtsevanos, On the Control and Stability of Grid Connected Photovoltaic Sources, IEEE Transactions on Energy Conversion, Vol. 2, No. 4, December 1987, pp. 556562. U. Boegli and R. Ulmi, Realization of a New Inverter Circuit far Direct Photovoltaic Energy Feedback into the Public Grid, IEEE Transactions on Industry Applications, Vol. IA-22, No. 2, MarcWApril 1986, pp. 255-258. S.M. Chalmers, P.M. Anderson and P.L. Vogt, The Effect of Photovoltaic Power Generation on Utility Operation I8 IEEE Transactions on Power Apparatus and Sysrems, Vol. PAS-104, No. 3, March 1985, pp. 524-530. B.H. Chowdhury and S. Rahman, Analysis of Interrelationships Between Photovoltaic Power and Battery Storage for Electric Utility Load Management, IEEE Transactions on Power Systems, Vol. 3, NO.3, August 1988, pp. 900-907. K. Tam,P. Kumar and M. Foreman, Enhancing the Utilization of Photovoltaic Power Generation by Superconductive Magnetic EnE E E Transactions on Energy Conversion, Vol. 4, ergy Storage, I No. 3, September 1989, pp. 314-321. M. Ohnishi and A Takeoka, Advanced Photovoltaic Technologies and Residential Applications, Renewable Energy, Vol. 6, No. 3, pp. 275-282, 1995. A. Chaurey and S. De,mbi, Battery Storage for PV Power Systems: An Overview, Renewable Energy, Vol. 2, No. 3, pp. 227235,1992. L.J. Blomen and M.N. Mugerwa, Fuel Cell Systems, Plenum Press, New York, 1993 T.M., Autonomous Photovoltaic-Diesel Power System Design, Proceedings of IEEE PV Specialists Conference, Las Vegas, Nevada, October 1985, pp. 280-284. T. Hiyama, S. Kouzuma and T. Imakubo, Identification of Optimal Operating Point of PV Modules using Neural Network for Real Time Maximum Power Tracking Control, IEEE Transactions on Energy Conversion, Vol. 10, No. 2, June 1995, pp. 360-367. T. Hiyama, S . Kouzuma and T. lmakubo, Evaluation of Neural Network Based Real Time Maximum Power Tracking Controller for PV System, IEEE Transactions on Energy Conversion, Vol. 10, No. 3, September 1995, pp. 543-548.

Ki

A
I,,

PV cell illumination ( mW / cm2

Saturation current at

( 2.0793~10-~ )

T, Em

Reference temperature (30 1.18 OK ) Band gap for silicon (1.10 eV)

VII. APPENDIX B. MODEL OF A FUEL CELL The performance of a fuel cell that operates at low and intermediate temperatures of 25-200 C is illustrated by a typical cell voltage versus current density plot, which is shown in Fig. B.l [13]. The following relation may represent the relating equations fiom a current density of 0 to the end of the linear region.

where
[ll] [12]

Terminal voltage of a fuel cell (V) Cell current density (A/cm2) Tafef slope (0.11) Exchange current density for oxygen reduction reaction (0.003) Internal resistance of a fuel cell (0.5) Reversible potential of the cell (1.0 V)

[13]

[ 141
[ 151

(161

VIII. REFERENCES
S. Rahman and K. Tam,A Feasibility Study of Photovolhic-Fuel I . Transactions on Energy ConCell Hybrid Energy System, version, Vol. 3, No. 1, March 1988, pp. 50-55. K. Tam and S . Rahman, System Performance Improvement Provided by a Power Conditioning Subsystem for Central Station PhoE E E Transactions on Energy tovoltaic-Fuel Cell Power Plant, I Conversion, Vol. 3, No. 1, March 1988, pp. 64-70.
1.2 1
1A

IX. BIOGRAPHIES
Kyoungsoo Ro (S95) obtained his B.S.E.E. and M.S E.E. from Seoul National University, Seoul, Korea in 1985 and 1987, respectively. He is currently a Ph.D candidate at the Virginia Polytechnic Institute and State University. His main areas of interest are power system operation, control ;and stability, renewable energy systems (photovoltaics, fuel cells and batteries) and artificial intelligence applications to power systems.
Saifur Rahnim (S75, M78, SM83) graduated fiom the Bangladesh LJniversity of Engineering and Technology in 1973 with a B.Sc. degree in Electrical Engineering. He obtained his M.S E.E. from the State University of New York at Stone Brook in 1975. His Ph.D degree (1978) is in Electrical Engineering from Virginia Polytechnic Institute and State LJniversity. Saifur Rahman directs the Center for Energy and Global Environment at the VPI&SU. He serves on the System Planning and Demand Side Management Subcommittee, and the Long Range Plannin& the Load Forecasting and the Photovoltaics working groups of IEEE Power Engineering Society. He has authored more thm 160 technical papers and reports in these areas

0.05

0.1

015

02

025

current density (2)

Fig. B. 1. Typical plot of cell voltage vs. current density for a fuel cell.