Load forecasting model consisting of data mining based orthogonal greedy algorithm and long short-term memory network

Abstract

To reduce the waste of electricity, load forecasting is essential for power scheduling and system management. However, when the external environment experiences unexpected changes, most of the existing load forecasting models have no capability to adjust the predicted values, accordingly. Therefore, in this paper, we propose forecasting model consisting of data mining based orthogonal greedy algorithm and long short-term memory (DM-OGA–LSTM) network. It utilizes DM-OGA algorithm to excavate the correlation between factors of various industries and electricity consumption, and meanwhile, make the selected features orthogonal. Then, the LSTM network is adopted to achieve prediction of future electricity consumption under the consideration of time factor and selected features. The simulation results show that the second-order features strongly correlated to the electricity consumption can be found from the factors of various industries. Meanwhile, DM-OGA–LSTM​ forecasting model can achieve more accurate predictions with the relevant second-order features.

Keywords

• Load forecasting model
• Feature selection
• Data mining based orthogonal greedy algorithm
• Long short-term memory network