Synergism of Recurrent Neural Network and Fuzzy Logic for Short Term Energy Load Forecasting

Abstract

Utility companies need to plan the generation capacity and resources for establishing an efficient energy management with reduced wastage. Load forecasting would become essential to a power system’s scheduling operations. Despite comprehensive work on short-term forecasting of loads using multiple machine learning models, there is room for improvement in prediction accuracy. A novel method of shortterm load forecasting based on the combination of fuzzy logic with recurrent neural network (RNN) model has been proposed in this paper. The Fuzzy logic model maps the highly nonlinear relationship between weather parameters like changes in temperature and their effect on the regular demand for electric charges. The Fuzzy logic model maps the highly nonlinear relationship between weather parameters such as temperature changes and their effect on the daily demand for electric loads. The proposed approach combines the advantages of fuzzy logic and neural networks to predict the next day’s load. Dataset of demand for electricity charging for a period of two years from 2013 to 2014 was obtained from ISO New England with a one hour resolution. It is observed that in terms of precision, the Fuzzy-RNN hybrid model outperforms its counterpart RNN. It is observed that in terms of precision, the Fuzzy- hybrid model outperforms its counterpart RNN. The proposed model was contrasted with other state-of - the-art methods for short-term load forecasting including artificial neural network (ANN), fuzzyANN, support vector machine (SVM), and generic neural regression network (GRNN). Overall, the computed results conclude that the synergetic use of fuzzy logic with RNN model is successful in achieving higher accuracy by efficiently mapping the effect of weather parameters with a change in load demand. Fuzzy-RNN has performed best with the highest accuracy in load forecasting amongst the six models considered.