Electrical load forecasting using echo state network

An algorithm for half hourly electrical load forecasting based on echo state neural networks (ESN) is proposed in this paper. Electrical load forecasting is one of the most challenging real life time series prediction problems. This demands a dynamic network. ESN is a new epitome for using recurrent neural networks (RNNs) with a simpler training method. Several versions of ESN are discussed. The load profile is treated as time series signal. The forecasting performance of ESN is analysed on the basis of its key parameters. ESN is compared with feed forward neural network (FNN) and Bagged Regression trees. Simulation results demonstrate that the proposed ESN algorithms can obtain more accurate forecasting results than the FNN and Bagged Regression trees.