A Hybrid Approach for Probabilistic mid-term Electricity Price Forecasting using deep learning

With deregulation of modern power systems, electricity market price forecasting become more and more important in managing electricity market which plays a key role in practical operation of electricity market and smart grids. In this paper, a hybrid approach is proposed for probabilistic mid-term electricity price forecasting. In proposed method, feature matrix is extracted using neighborhood component analysis while the seasonality of observed electricity prices is decomposed. After preparation of training dataset, each subset divided to training and validation subsets. Afterwards, using each training data subset and by utilizing a long short-term memory network configuration with three hidden layers, a prediction model will be trained and will be tested with training and validation data sets. The simulation results show that proposed hybrid method caused a decrease in the point forecasting error. Error reduction results to a decrease in final error of combined model and the best MAPE is related to Winsorized model with value of 9.2009 which shows a reduction in MAPE equal to 9.633 percent compared with the method presented for comparison. The results show that the proposed method for electricity price forecasting is an efficient method and can be used for mid-term probabilistic electricity price forecasting.