Short-term load forecasting using a long short-term memory network

Abstract

Load forecasting is an essential part of a power system. It enhances the energy-efficiency and reliable operation of the power system. As depicted in the proposal of the smart grid, an increasing number of smart meters have been being installed in many utilities on a global scale. Thus, a large number of historical residential consumption data now can be obtainable easily which were not available in the past. However, traditional forecasting techniques may not satisfy the much higher demand of precision in load forecasting. In this paper, a novel approach to short-term load forecasting using a LSTM (long short-term memory) network based on RNNs (recurrent neural networks) is proposed. RNNs have powerful nonlinear mapping capabilities, especially in field of time series, and LSTM models take advantage of memory units to make better abstract for long sequences. Test results show that the method can obtain high precision.