A NEURO-FUZZY MODEL FOR THE CONTROL OPERATION OF A WIND-DIESEL-BATTERY HYBRID POWER SYSTEM. P. S. Panickar, M. S. Rahman and S. M.

Transcription

1 A NEURO-FUZZY MODEL FOR THE CONTROL OPERATION OF A WIND-DIESEL-BATTERY HYBRID POWER SYSTEM Abstrac t P. S. Panickar, M. S. Rahman and S. M. Islam Centre for Renewable Energy and Sustainable Technologies Australia Curtin University of Technology, Australia This paper outlines the development of the control operation of a battery incorporated wind-diesel hybrid power system using the Neuro-Fuzzy modeling technique. The overall load dispatch scenario is controlled by the availability of wind power, total system load demand, diesel engine operational constraints and the proper management of the battery bank. The results of the preliminary study show that a considerable amount of energy can be saved in the system under study compared with the one without a battery bank. The system has also an added advantage in that it can substantiate a sudden drop in wind-turbine energy. The incorporation of a battery bank makes the control operation more practical and relatively easier.. INTRODUCTION In a remote hybrid power system, a combination of renewable energy sources with diesel generators and batteries are more preferable []. A wind-diesel hybrid power system could be better and more intelligently operated if an optimized control strategy is developed and incorporated. Quite a few articles [2, ] have been published earlier on the control strategy of a winddiesel hybrid system. However, these articles did not deal with the diesel dispatch strategy at all. An article was published on the adaptive control strategies in a wind-diesel hybrid system [4], which proposed a chrisp method of control using direct calculation of the required quantities from the predicted values of wind speed and system load. Also a paper has recently been published on a fuzzy logic model fo r the control operation of a wind-diesel hybrid system []. This study, however, did not include a storage battery bank in the system. Without the battery bank, the system appears to have certain drawbacks and lacks flexibility in operation. When the wind power availability is poor, the system requires an increased diesel dispatch (more than the required amount) and complicates the control operation. Also, when the wind power availability is high and the load demand is low, most of the power is wasted as dump load. A battery management system (BMS) has been proposed in a recent article for stand alone Photovoltaic Energy Systems [6]. This article suggests that the incorporation of a battery bank along with a proper battery management system may result in improved reliability of the system and a significant reduction in operating costs. A similar but suitable battery bank could also be incorporated in a winddiesel hybrid system, and an appropriate battery management system developed. The control operation would be based on the same fuzzy logic technique as described in reference []. This paper is an investigation into the battery-incorporated wind-diesel hybrid system, where the state of charge (SOC) of the battery bank would be considered as a new input variable for the control operation. 2. THE TEST SYSTEM A West Australian hybrid power system site, as illustrated in Figure, has been chosen for the study. Figure : Wind-diesel-battery hybrid system The ideal power balance equation for such a hybrid power system is given by [, 4], W + D + D2 ± B [ Dump Load] = L L L () where, L=system load, W=wind turbine power, D and D 2 are the power generated by the two diesel engines, +B is the power supplied by the battery bank and -B the power used by it during its charging.. PREDICTION MODELS For controlling the diesel dispatch from a wind-dieselbattery hybrid system, it is essential to forecast the system load and wind speed first. To predict the system load demand [, 8] and wind-speed [] from the historical data, various techniques are available, some of which have been used for the current investigation.

2 . Load Prediction The Neural network technique has been used to predict the system load [2]. A 24-hour ahead load prediction strategy has been adopted here and a feed forward neural net, with a learning rate of.6 and hidden nodes, has been used. The actual and predicted loads are shown in Figure 2. The accuracy of the load forecast lies within ±% of the actual load demand [] Actual Load Predicted Load 64 Figure 2: Actual and predicted load pattern.2 Wind Speed Prediction An autoregressive statistical model has been used for wind forecasting [4, ], L L (2) where, y(t) is the future (one step ahead) prediction of the wind speed, n is the coefficient index, a n are the linear prediction coefficients, T is the sample time interval and y(t-nt) are the past values of the wind speed corresponding to delays of nt. The coefficients are calculated from a selected modeling window of m-number of past observations using the popular least square error method. Two methods of wind speed prediction using statistical models are analyzed [4]. The first method considered only the previous wind speed values to predict the wind speed value of an instant, which is one sample point ahead. The second method predicted future wind speed values from the previous same hour of a 24-hour data set giving a full day prediction horizon. The sample results of these two cases are shown in Figures and 4, respectively. 8 y t) = a t y( t T ) a t y( t 2T ) L an y( t nt) ( 2 Use has been made of these predicted values for the current fuzzy implementation of the model. The predicted results were compared with the actual values of wind speed by estimating the figures of merit, such as, Correlation Coefficient (r), Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) [], which are included on top of Figures and 4. 8 Wind Speed m/s 2 Corr. Coef. =.868, MAPE =.8, RMSE=.2m/s 2 Actual WS 6 Predicted WS One hour ahead prediction: n=,m= 2 Figure : One-hour ahead prediction 24 hour ahead prediction : n=, m= Corr. Coeff. =.66, MAPE = 2.6, RMSE =.6m/s 2 Actual WS Predicted WS 6 Wind Speed m/s 2 Figure 4: Twenty-four hour ahead prediction 4. OPERATIONAL CONSTRAINTS In planning the control strategy, the following two major operational constraints are considered. 4. Diesel Generator Constraints The operational constraints for a diesel generator are the minimum operating load, the minimum run time for the diesel engines, and the spinning reserve regulations for the system. The diesel engines are never allowed to run below a certain percentage (usually 4%) of its rated power, to avoid critical maintenance problems. Then there are the starting and idling cost aspects, contributing to the total system operational economics []. The operational constraints may thus override the diesel dispatch strategy that is based on the predicted scenario alone. 4.2 The Battery Bank Constraints The battery bank in renewable energy systems mostly include gel type lead-acid batteries as energy storage device, as they require very little maintenance. The battery state of charge (SOC) is the cumulative sum of the daily charge/discharge energy transfers. The life of a battery depends on its charging and discharging cycles as shown in Figure, which suggests that the battery should not be operated at a low SOC.

3 Figure : Depth of discharge versus cycles. BATTERY MANAGEMENT SYSTEM The battery management system (BMS) in a winddiesel-battery hybrid system would monitor the battery SOC by keeping track of the charging and discharging cycle, and would supply the information to the controller every hour. The BMS would help prevent continuous operation of the battery at a low SOC, and thereby ensure longer battery life. ( I ( t) I ) t SOC bat gas t t + = SOCt + L L CD where: SOC t is the SOC at a defined starting point, SOC t+ is the SOC after the first calculation, I bat (t) is the battery current, and t is the time interval between calculations. The sign of I bat (t) in equation () is positive if net current is flowing in the battery, and negative if flowing out. The loss factor I gas, is defined as [6] CD I = [ Cv ( Vbat / cell 2.2) + Ct ( Tbat 2) ] gas I go e L (4) NB where: I go is the normalized gassing current in Ah, C v is the voltage coefficient, C t is the temperature coefficient, V bat/cell is the battery voltage per cell, T bat is the battery temperature, N B is the nominal battery capacity in Ah, and C D is the battery capacity at the nominal D hour discharge rate. Figure shows a plot of I gas. (). Size of the Battery Bank Two major factors are considered in determining the size of the battery bank: (i) the load demand to be met when wind power availability is poor, and (ii) the amount of surplus power available for charging the battery bank when the load demand is low. Finally, using the Hybrid2 simulation program, the size of the battery bank has been chosen to be 6Ah (2V, series x 2 parallel)..2 Battery SOC Calculation The most common method to determine the battery SOC in a dynamic system is the Ampere-hour (Ah) balancing technique that includes a variable loss term I gas [6]. The block diagram in Figure 6 shows the variables used for measurement of the battery SOC. Here, I in and I out represent respectively, the current flowing in and out of the battery, and I gas represents the battery losses. I in (t) t SOC I gas (t) Loss Current I (t) out Figure 6: Variables for battery SOC measurement Equation () is used to calculate the SOC: Figure : I gas as a function of V bat and T bat. Battery Management Strategies The following strategies have been adopted in implementing the current BMS. To ensure a longer battery life, the battery bank should be brought in to supply part of the total load demand only when the battery SOC is more than %. Although, the battery bank may be in a state to supply a certain amount of load; there are other factors, viz., the total load demand, wind power availability, current diesel dispatch along with the diesel engine constraints, that would finally dictate whether or not the battery bank should be brought in. When the SOC is less than %, it should be taken out and kept ready for charging in the next available opportunity. The charging scheme should be carefully chosen so the battery can attain a maximum level of SOC possible, once it is taken for charging. Usually, the battery bank should be taken for charging when the total generation becomes greater than the total system load demand. This is the case when the

4 diesel generators are running at their prescribed level and the available wind power is high; or when the wind power is normal and the diesel constraints have lifted the generation up (by the minimum 4% load limit). However, when the total generation is nearly equal to the total load demand and the battery bank has been waiting for a long time to be charged, it is wise to increase the generation so the battery could be charged and made ready for future use. Proper care should be taken not to allow deep discharge in a single cycle. The total number of charging and discharging cycles should also be kept as minimum as possible. 6. THE NEURO-FUZZY CONTROL The controller for the wind-diesel-battery hybrid system makes use of a combination of neural net, fuzzy logic and other traditional logical techniques. 6. The Control Strategy Figure 8 shows a flow diagram of the control strategy. 6.2 The Fuzzy Model Figure shows the fuzzy model for diesel dispatch. Figure : The fuzzy model for diesel dispatch 6.2. Fuzzy Membership Functions The power generated from the wind turbine generator is computed from the corresponding power characteristic curve [4]. The wind turbine power output is then fuzzified, the membership functions of which are shown in Figure (a). In a similar way, the system load is fuzzified, and the membership functions are shown in Figure (b).. µ(w) very low. cut-off low medium high very high highest 2 2 W, kw Figure (a): Membership functions for wind power µ(l).. very low low medium high very high highest L, kw Figure (b): Membership functions for system load Figure 8: Flow diagram of the control strategy. At first the fuzzy inference system is run. The defuzzified output from the controller is then checked against the diesel constraints and diesel start-stop timing sequence []. This is a primary diesel dispatch plan. Based on this 24-hour prediction scenario, a battery bank loading and charging plan has been formulated, and an estimate of the battery SOC is obtained from the BMS. Next the suggested diesel dispatch and battery contribution are computed. Appropriate corrective measures may be taken to obtain the final diesel dispatch D, D 2, and the battery bank contribution B, as and when the hourly predicted data become available. Now, based on the input scenario the output diesel contributions have to be determined. This is achieved by defining a set of output membership functions for D and D 2 and developing the appropriate fuzzy rule base. The membership functions for D and D 2 are shown in Figures (c) and (d), respectively. The fuzzification, rule base and defuzzification are based on heuristic expert knowledge from a study of the crisp model [4]. µ(d).. cut-off minimum low medium high maximum 2 2 D, kw Figure (c): Membership functions for D

5 µ(d2).. cut-off minimum low medium high very high maximum D2, kw Figure (d): Membership functions for D Fuzzy Rule base Now, knowing the two fuzzy sets representing wind power and system load demand, a fuzzy rule matrix is then constructed, which looks like as shown in Table-. Each entry in the matrix may represent one rule, and thus forms the basis of a rule set. For example, IF wind power is very high and system load demand is low (Rule ) THEN D is high and D 2 is low. As there are two diesel engines D and D 2, each IF statement would have two sets of rules. Employing a set of fuzzy rule base, the diesel power contribution in crisp terms has been quantified using the centroid method of defuzzification. Then, a diesel-timing diagram is planned from the crisp output D D2 B L W D+D2+W+B Fig. (a): Output with the Battery Bank model D D2 L W D+D2+W Fig. (b): Output without the Battery Bank System Load Demand Table-: Fuzzy Rule Matrix Wind Turbine Power Output Cut-off Very low Low Medium High Very High Highest Very low Low Medium High Very High Highest In a similar way, the outputs were computed for a day when the wind power contribution is low. Figure 2(a) shows the diesel dispatch computed from the new model and Figure 2(b) without the battery bank D D2 B L W D+D2+W+B. RESULTS AND DISCUSSION The Neuro-Fuzzy controller, as described in section 6, has now been applied for the simulation of the control strategy. The estimated battery SOC has been applied to obtain the modified dispatch. Two examples have been considered for the case study. The first case is for a day when the wind power contribution is high. The diesel dispatch is computed from the new model with the battery bank and the results are shown in Figure (a). The same data were fed to the fuzzy model without the battery bank [] and the computed diesel dispatch results are shown in Figure (b). The results for the two models look the same as it was not required to bring the battery bank in to supply part of the load demand. Fig. 2 (a): Output with the Battery Bank model D D2 L W D+D2+W Fig. 2 (b): Output without the Battery Bank This time the output from the fuzzy model with the battery bank looks different in the initial part from the model without the battery bank. The summary of the results is shown in Table-2.

6 Table-2: Comparative Analysis With Battery Bank Without Battery Bank D D2 B D D2 No. of Start 2 On Time (h) KWh Fuel (Lt.) 62 6 x 2 22 Total Fuel 2 (Lt.) 222 (Lt.) Table-2 shows that use of the battery bank has saved 2. liters of fuel for a particular day. However, there may be many such and other cases where the battery bank would share a substantial amount of load. Thus the battery bank can save fuel, and at times, it can also help prevent simultaneous operation of both the diesel engines, thereby saving other operational costs. The ultimate goal of the implementation is to apply the battery SOC as a fuzzy input variable to develop a unified fuzzy model as shown in Figure. This model may be implemented in the final block of the flow diagram shown in Figure 8. Figure : Fuzzy model for further study In this case, an estimate of the battery SOC for the next hour should be obtained from the BMS. The ranges of the membership functions for the battery SOC may be defined as follows: Highest (%- %), Very High (8%-%), High (6% -8%), Medium (%-6%) and Low (<%). Also, an output membership function µ(b) for the battery bank should be defined and a new set of fuzzy rule base developed. Finally, the fuzzy logic model should be trained and tuned to achieve the optimum performance. It should, however, be noted here that the preliminary dispatch, as proposed in Figure 8, was based on a 24- hour prediction horizon. As the SOC of the battery bank could not be predicted similarly for the next 24 hours, it was not possible to apply the unified fuzzy model in the primary dispatch. 8. CONCLUSIONS and the estimated battery SOC combined with fuzzy logic controlled diesel dispatch simplify the control algorithm and ensure intelligent operation of the hybrid system. Preliminary results indicate that the proposed model may prove more cost-effective than the former mo dels when considered for a long-term operation scenario.. REFERENCES [] C. V. Nayer, S. J. Phillips, W.L. James, T. L. Pryor and D. Remmer. Novel Wind/Diesel/Battery Hybrid Energy System. Solar Energy. Vol., pp. 6-8,. [2] P. S. Panickar, S. M. Islam and C. V. Nayar. A New Quasi-Optimal Control Algorithm for a Wind-Diesel Hybrid System. Wind Engineering, Vol. 22, No., pp. -, 8. [] R. Chedid, S. Karaki and C. Chemali. Adaptive Fuzzy Control for Wind-Diesel Weak Power Systems. IEEE Trans. Power Engineering. PE- -EC (-),. [4] P. S. Panickar, M. S. Rahman, S. M. Islam and T. L. Pryor. Adaptive Control Strategies in Wind Diesel Hybrid Systems. Proc. Australian Universities Power Engineering Conference. Australia, 2. [] P. S. Panickar, M. S. Rahman and S. M. Islam. Fuzzy Logic Control of a Wind-Diesel Hybrid Power System. Presented at the International Power Engineering Conference, IPEC, - May, 2, Singapore. [6] S. Duryea, S. Islam, W. Lawrence. A Battery management System for Stand-Alone Photovoltaic Energy Systems. IEEE Industry Applications Magazine, May/June 2. [] T. M. Peng, N. F. Hubele and G. G. Karady. Advancement in the Application of Neural Networks for Short Term Load Forecasting. IEEE Trans. Power System. Vol. No., pp. 2-2, 2. [8] D. K. Ranaweera, N. F. Hubele and G. G. Karady. Fuzzy Logic for Short Term Load Forecasting. Journal of Electrical Power and Energy Systems, Elsevier Science. Vol. 8, No. 4, pp , 6. [] P. Panicka r. Wind Speed Prediction using Statistical Methods. Proc. Inter-University Postgraduate Electrical Engineering Symposium. Western Australia, 2. [] L. L. J. Mahon. Diesel Generator Handbook. Butterworth-Heinemann, Oxford, 2. A fuzzy logic controller has been simulated for realizing the control strategies of a wind-diesel-battery hybrid system. The predicted load and wind speed,