Deep learning-based integrated XAI for photovoltaic power forecasting considering actual power production period

Abstract

This paper proposes a deep learning (DL)-based integrated explainable artificial intelligence (XAI) framework for photovoltaic (PV) power forecasting, explicitly considering the actual power production period to improve operational reliability. The framework uses solar irradiance, ambient temperature, and relative humidity as input features and evaluates nine DL architectures, including artificial neural networks (ANN), recurrent neural networks (RNN), convolutional neural networks (CNN), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), CNN-LSTM, CNN-BiLSTM, RNN-LSTM, and RNN-BiLSTM. Model performance is evaluated using mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE). The results show that the residual-based RNN-LSTM model provides highest forecasting accuracy, achieving MAE of 1.21 kW, MAPE of 5.12%, and RMSE of 2.24 kW. In comparison, the LSTM and BiLSTM models exhibit substantially higher prediction errors, with MAPEs exceeding 21%, while hybrid convolutional models show moderate improvements but remain inferior. To enhance model transparency, XAI techniques are integrated to interpret feature contributions. The analysis confirms that solar irradiance is the dominant influencing factor, while temperature and humidity introduce secondary nonlinear effects captured effectively by recurrent architectures. The proposed framework provides a high-accuracy and interpretable solution for PV power forecasting, supporting reliable energy management and smart grid applications.