A Hybrid LSTM-CNN Model for Short-Term Photovoltaic Power Forecasting in Italy

Accurate photovoltaic (PV) power forecasts are increasingly crucial for managing and controlling integrated energy systems. Hence, this paper presents a hybrid model that combines long short-term memory (LSTM) and convolutional neural networks (CNN) to forecast PV power output accurately. The proposed approach uses a parallel processing technique, where datasets are fed into the LSTM and CNN branches simultaneously. Each branch analyses the data separately. CNN layers focus on spatial properties, while LSTM layers capture temporal correlations. The proposed model is evaluated against three single models: CNN, LSTM, and gated recurrent unit (GRU), and three hybrid models: GRU-CNN, CNN-LSTM, and GRU-LSTM. In addition, to facilitate comparison and assessment, we included three commonly used evaluation metrics: root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Notably, LSTM-CNN stands out as the best performer, showing the lowest RMSE, MAE, and MAPE values compared to other models. The proposed model achieved RMSE, MAE, and MAPE values of 25.65W, 11.94W, and 0.71%, respectively, on the training dataset and 12.85W, 5.94W, and 0.37%, respectively, on the testing dataset. Thus, the proposed LSTM-CNN model efficiently captures complex features and patterns in weather data, resulting in accurate PV power forecasts.