Day-Ahead Photovoltaic Power Forecasting Using a Hybrid BiLSTM-CNN Model

Weather conditions can significantly impact the performance of Photovoltaic (PV) systems, resulting in erratic and sporadic power output. Hence, this paper presents a hybrid model integrating Bidirectional Long-Short Term Memory (BiLSTM) with a Convolutional Neural Network (CNN) to predict day-ahead PV power production under diverse weather conditions. The proposed hybrid model uses a BiLSTM layer to capture temporal relationships, whereas a CNN layer analyzes spatial features within the dataset. To validate the performance of the proposed model, we employed historical weather and PV power data from the BP solar plant in Alice Springs. In addition, the BiLSTM-CNN model's performance is assessed against three standalone models (BiLSTM, LSTM, and CNN) and two hybrid models (LSTM-CNN and CNN-BiLSTM). Five evaluation metrics are used to compare all these forecasting models. The findings show that the proposed model can provide accurate day-ahead forecast results for sunny and cloudy days. The BiLSTM-CNN achieved an average Root Mean Squared Error (RMSE) of 77.73, a Mean Absolute Error (MAE) of 41.32, and a Mean Absolute Percentage Error (MAPE) of 5.085. Consequently, the proposed model can precisely capture complex attributes and patterns in weather data, resulting in accurate PV power forecasting.