ACCURATE LOAD PREDICTION BASED ON HYBRID ARTIFICIAL NEURAL NETWORK AND MODIFIED BAT ALGORITHM

Abstract

Precise prediction of load forecasting problem is important for the new competitive electricity market. Due to the deregulation of the electricity industry, load pattern is almost complicated. Therefore, an essential task for a specific electricity network is finding an appropriate prediction model. So, this article proposes a novel hybrid method using cascade artificial neural network (ANN) and bat algorithm (BA) for short term electric load forecasting. The proposed method makes use of BA and Levenberg-Marquardt to modify the training process of cascade ANN and reduce the forecast error. The input features are the load data, temperature and humidity of the test region. To improve the training procedure, a new modification approach is developed to avoid premature convergence and reduce the chance of trapping in local optimum. Several prediction criteria are employed to compare the performance of the proposed method with some current methods in this field. The experimental results on the practical electric load dataset confirm the high accuracy and satisfying the proposed model performance. The experimental results on the practical dataset show that the proposed hybrid method has better the high accuracy and satisfying performance than the other methods, such as BRF, MLP-BR, MLP-BFGS, MLP-LM and SVM.