A Novel Hybrid Deep Learning Model for Photovoltaic Power Forecasting Based on Feature Extraction and BiLSTM

Abstract

In this paper, a hybrid photovoltaic power forecasting model is proposed based on bidirectional long‐short‐term memory network. Firstly, the photovoltaic power and meteorological data are decomposed by ensemble empirical mode decomposition. Secondly, key features in the meteorological subsequences are extracted by kernel principal component analysis method to eliminate the correlation and redundancy in the original meteorological sequence and reduce the input dimension of the model, then the coupling time characteristics between meteorological subsequences and photovoltaic power subsequence are further explored. Thirdly, a bidirectional long‐short‐term memory network optimized by improved particle swarm optimization based on inertia weight of anti‐sine function is proposed, and then the eventual photovoltaic power forecasting results are obtained by superimposing the predicted values of each prediction component. The experiment is carried out with the operation data of the Inner Mongolia power grid, and the results show that the proposed forecasting model exhibits higher prediction accuracy. © 2023 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.