Development of an Intelligent System for Short-Term Electrical Power Load Forecasting in Maharashtra State

In this paper an Elman RNN is developed to forecast Maharashtra's total real time electrical power load reasonably one day in advance. This paper compares the performance of two NNs configurations namely a well known RBF and the proposed Elman RNN. Load data are clustered according to the differences in their characteristics. Special days are extracted from the normal training sets. In this way, solution is provided for different load types, including working days, weekend and, special days. RBFNN is constructed for the same data as a benchmark. It is shown that, the proposed Elman NN clearly outperforms the RBFNN. Results show that the optimal Elman NN model has average MSE as low as 0.00408; average correlation coefficient has as high as 0.8266 and, average prediction error 6.21% obtained for the year 2006. With the developed optimal Elman NN, the average of MAPE on all three clusters has been reduced by 1% which is the major outcome of this research.