Feature Selection Using Genetic Algorithm and Bayesian Hyper-parameter Optimization for LSTM in Short-Term Load Forecasting

Electricity load forecasting at nationwide level is important in efficient energy management. Machine learning methods using big data multi-time series are widely applied to solve this problem. Data used in forecasting are collected from advanced SCADA system, smart sensors and other related sources. Therefore, feature selection should be carefully optimized for machine learning models. In this study, we propose a forecasting model using long short-term memory (LSTM) network with input features selected by genetic algorithm (GA). Then, we employ Bayesian optimization (BO) to fine-tune the hyper-parameters of LSTM network. The proposed model are utilized to forecast Vietnam electricity load for two days ahead. Test results have confirmed the model has better accuracy in comparison with currently used models.