Electricity Price Forecasting using Multilayer Perceptron Optimized by Particle Swarm Optimization

In this paper, electricity price forecasting using a hybrid multilayer perceptron, back propagation and modified particle swarm optimization is implemented. Here modified particle swarm optimization technique is used to improve the performance of the backpropagation algorithm while training the multilayer perceptron. Two different MLP are used for electricity price forecasting one MLP is with a single hidden layer and another MLP is with three hidden layers, both the neural networks are trained by BP and initial parameters such as weights between different layers, the bias of the layers, and activation function of each layer except input layer are selected by MPSO. Normally MLP trained by BP uses linear activation functions for all layers and neurons, but in this case, we use three different functions namely linear function, sigmoid function, and tangent function as activation functions. These three different activation functions are independently selected for each neuron by MPSO based on the data set which is used. Because of the independent selection of activation function to each neuron the overall performance, convergence time, and convergence efficiency of the BP are greatly improved. The proposed method is implemented to predict Austria and Northern Italy electricity price.