Controller Optimization Approach Using LSTM-Based Identification Model for Pumped-Storage Units

Abstract

In this paper, the controller optimization problem of the pumped-storage unit (PSU) was examined. The objectives of this paper were to identify the dynamic model of the PSU according to the deep learning model through training of the input-output data and to optimize the parameters of the controller on the basis of this identified model. To achieve the objectives, a novel pump-turbine model based on the B-spline surface was employed to precisely simulate the PSU for data measurement and identification. Next, the long short-term memory (LSTM) network architecture was applied to identify the dynamic model of the PSU. Then, gain tuning of the proportional-integral-derivative (PID) controller was conducted by applying particle swarm optimization on the basis of the identified model. To verify the effectiveness of the proposed method, a simulation platform based on the pumped-storage hydropower plant in China was chosen as the experimental object, and the comparative experiments were conducted. The results show the following: 1) the LSTM model performed better compared with the autoregressive model with exogenous variables, support vector machine, and feedforward neural network inaccuracy; 2) the PID controller tuned with the identified LSTM model has excellent control ability compared with the other identified models, and; 3) the identified LSTM model and optimized controller have very good robustness under different conditions.