Predicting photovoltaic power generation using double-layer bidirectional long short-term memory-convolutional network

Accurate photovoltaic (PV) power prediction is critical for PV power plant safety and stability. The main restrictions influencing the accuracy of the PV power forecast are the variability and intermittency of solar energy. Therefore, this study proposes a hybrid deep learning model for PV power forecast that is successfully developed using the combination of the bidirectional long short-term memory (BLSTM) and convolutional neural network (CNN) and is applied to the actual dataset collected in the DKASC PV system in Alice Springs, Australia. The proposed architecture is a structure of two major branches. BLSTM is used first to extract the bidirectional temporal characteristics of PV power. Next, CNN was used to capture the spatial characteristics. The prediction results of the hybrid model are compared with those of the single model LSTM, BLSTM, CNN, gated recurrent unit, recurrent neural network (RNN), and the hybrid network (LSTM–CNN, CNN–LSTM) in order to demonstrate the higher performance of the proposed hybrid prediction model. By comparing statistical performance indicators such as root mean square error (RMSE), mean absolute error (MAE), mean square error (MSE), and coefficient of determination (R2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document}) values with other existing deep learning models, the performance of the proposed BLSTM–CNN model has been demonstrated. The results indicate that the BLSTM–CNN model has the highest precision with the lowest MSE of 0.0089, MAE of 0.0531, RMSE of 0.0944, and highest R2 of 0.9993. BLSTM–CNN can enhance forecasting accuracy while also accurately capturing the various temporal–spatial characteristics of PV power.