A Hybrid Deep Learning Model with Evolutionary Algorithm for Short-Term Load Forecasting

Abstract

Load forecasting is a pivotal part of the power utility companies. To provide load-shedding free and uninterrupted power to the consumer, decision-makers in the utility sector must forecast the future demand of electricity with the least amount of error percentage. Load prediction with less percentage of error can save millions of dollars to the utility companies. There are many techniques to amicably forecast the demand of electricity. Amongst which the hybrid models show the best result. In this study, a hybrid method integrating Genetic Algorithm (GA), which is an evolutionary algorithm, and long short-term memory (LSTM) network is laid down. For the LSTM network, heuristical trial and error is usually employed to choose the best window size, neuron number, and other architectural factors. This study proposes a systematic method for electrical load forecasting by determining the time lags, neuron number, and batch size using GA. Very few research work has been done to increase the accuracy of the electrical load forecasting by selecting the best batch size for LSTM model. To evaluate the proposed hybrid model, the model is tested on half-hourly load data, collected from the Australian Energy Market Operator (AEMO). The experimental results show that the proposed hybrid model of GA-LSTM network surpasses the other standard models such as support vector machine (SVM), multilayer perceptron (MLP) and traditional LSTM model with the least MAE and RMSE value of 87.304 and 118.007 respectively. The proposed model shows 5.89% and 8.19% error reduction with respect to LSTM model in both MAE and RMSE respectively.