Short-Term Load Power Prediction Based Deep Learning Gated Recurrent Unit in Solar Power Plant

Abstract

Forecasting the load power in the solar plant is essential to maximize the available profit from the solar plant. This project develops the Gated Recurrent Unit (GRU) based deep learning machine to predict hourly load power in a solar plant. The weather parameters influence the load power, which affects the user behavior of load power. All selected features and the target variable are collected for more than one year in a solar plant installed in Taiwan. The collected data are utilized in the simulation of the Gated Recurrent Unit to evaluate the accuracy and performance in predicting the short-term load power. The performances of GRU are compared with the RNN model by statistical benchmarks. The experiment results prove that the GRU-based deep learning machine could achieve higher accuracy and better stability in predicting the short-term load power in solar plants.