Title: Day-ahead hourly electricity load forecasting based on long short-term memory neural networks: a comparison study

Authors: Qianying Zhang; Junqiu Yuan; Xiang Liu; Li Sun

Addresses: Changzhou Power Supply Branch, State Grid Jiangsu Electric Power Co., Ltd., Changzhou 213004, China ' Changzhou Power Supply Branch, State Grid Jiangsu Electric Power Co., Ltd., Changzhou 213004, China ' Changzhou Power Supply Branch, State Grid Jiangsu Electric Power Co., Ltd., Changzhou 213004, China ' Changzhou Power Supply Branch, State Grid Jiangsu Electric Power Co., Ltd., Changzhou 213004, China

Abstract: Electricity load forecasting (ELF) is crucial for the economic planning of power systems. Due to its time-series essence, long short-term memory neural network (LSTM) is considered as a promising solution for ELF. This paper investigates the LSTM algorithms using different model inputs and structures. First, a univariate LSTM model is developed to train the relationship between the historical load and future load, based on which ELF is carried out. Second, a multi-variable LSTM model is proposed by incorporating the temperature data as another input, thus making the prediction results more reliable if day-ahead weather forecast is available. Finally, this multi-variable LSTM model is further extended to be of double layers for each cell. The error statistics show that the more complex two-layer LSTM reduces the RMSE, MAE, MAPE, IAE, and SD by 14.17%, 20.06%, 21.43% 20.08%, and 6.53% respectively, compared to the P_LSTM_T model.

Keywords: electricity load forecasting; ELF; long short-term memory; LSTM; day-ahead prediction.

DOI: 10.1504/IJSCC.2025.144538

International Journal of Systems, Control and Communications, 2025 Vol.16 No.1, pp.45 - 61

Received: 22 Oct 2024
Accepted: 04 Dec 2024
Published online: 18 Feb 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article