Document Type: Article
Department of Industrial Engineering, Science and Research Branch, Islamic Azad university, Tehran, Iran
Department of Mathematics, Science and Research Branch, Islamic Azad university, Tehran, Iran
Department of Industrial Engineering, Mazandaran University of Science and Technology Branch, Babol, Iran
In this paper, a novel method is proposed to predict the cost of short-term hourly electrical energy based on combined neural networks. In this method, the influential parameters that play a key role in the accuracy of these systems are identified and the most prominent ones are selected. Due to the fluctuations of electricity prices during various seasons and days, these parameters do not adhere to the same pattern. In the proposed method, initially, using the SOM network, similar days are placed in close clusters. In the next stage, the temperature parameter and prices pertaining to similar days are trained separately in two MLP neural networks because of their differences concerning the range of changes and their nature. Finally, the two networks are merged with another MLP network. In the proposed hybrid method, an evolutionary search method is used to provide an appropriate initial weight for neural network training. Given the price data changes, the price amidst the previous hour has a significant effect on the prediction of the current state. In this vein, in the proposed method, the predicted data in the previous hour is considered as one of the inputs of the next stage.