Author(s): Boran Zhu; Di Zhang; Junqiang Lin; Qidong Peng; Tiantian Jin

Linked Author(s): Di ZHANG

Keywords: Reservoir operation; Long short-term memory network; Chaotic artificial electric field algorithm; Genetic algorithm; Flood season

Abstract: Scientific reservoir operation simulation is of great significance to ensure efficient and stable operation of the reservoir. The long-short-term memory (LSTM) model is widely used because it can accurately reflect the time sequence characteristics of reservoir operation. The application effect of the model is closely related to the parameter settings. However, traditional empirical settings and gradient-based optimization methods tend to cause the model training results to fall into local optimality and fail to achieve the expected results. This paper employs a Chaotic Artificial Electric Field Algorithm (CAEFA) for parameter training in LSTM models and verifies its effectiveness through simulations of the Xiluodu Reservoir operations. The results show that the CAEFA-enhanced LSTM model can better capture the characteristic processes of reservoir operation across different datasets, offering higher computational accuracy and lower uncertainty compared to classical Genetic Algorithms (GA) and Artificial Electric Field Algorithms (AEFA). Due to the variations in flood processes during the flood season, the model’s performance is slightly less effective in flood season datasets. Compared to conventional LSTM models, the model developed in this paper is more suitable for reservoir operation simulation.

DOI:

Year: 2024