Prediction of Power System Load Using Echo State Network Based on Dropconnect Method

The application of artificial intelligence makes the rail transit power system more intelligent. Load forecasting has always been one of the main problems faced by the rail transit power system. The paper proposes an echo state network based on Dropconnect method for predicting rail transit power system load. This method subtly changes neuronal connection, optimizes connection structure of reservoir neurons, and improves efficiency of load forecasting in rail transit power system, ensuring stable operation of rail transit power system. For validation effectiveness of proposed method, the paper compared its predictive performance for rail transit power systems with classical ESN, Dropout ESN, and Dropconnect ESN from three aspects. At the same time, a comparison was made between the predictions of time series generated by Mackey- Glass system. Finally, this paper also compared the predictive performance of the proposed method with other neural network methods. The simulation results show Dropconnect ESN has better performance in load predictive task of rail transit power systems.

I. Introduction

Artificial intelligence has a wide range of applications in the power system, enabling it to operate more efficiently and intelligently. It not only improves the reliability, safety of the system, but also promotes the effective integration of renewable energy [1] - [3].