An improved neural network prediction model for load demand in day-ahead electricity market

Abstract

Load demand prediction is vital for maintaining stability and controlling risks of electricity market. An improved model which combines neural network with genetic algorithm is proposed to accurately predict load demand at equilibrium situation of day-ahead electricity market. In the proposed model, load demand prediction problem is converted into optimization problem of error minimization between the actual output and the desired output. Optimal topology and initial weights of neural network are obtained by using hybrid genetic operation of selection, crossover and mutation. Next, gradient learning algorithm with momentum rate is used to train neural network and optimal connection weights are obtained. The proposed model is tested on load demand prediction in California electricity market. The test results show that the proposed model can effectively approximate input/output mapping of training samples and can obtain more accurate load demand prediction values than BP neural network.