Solar power forecasting using recurrent deep neural network

Abstract

Recent growth in the use of renewable energy has forced companies in the power sector to establish long-term plans. Forecasting is so essential, but as all renewable energy systems depend on nature, it also brings nonlinearity into the system. Machine learning and computation techniques are used to forecast non-linear data. This study compares a typical machine learning (ML) model, the artificial neural network, to a memory-based recurrent neural network (RNN). Two-month and one-month forecasts are provided by the models. The effectiveness of the model is assessed using statistical measures like the correlation of determination (R2), mean absolute error (MAE), mean square error (MSE) and mean absolute percent error (MAPE). According to calculations using statistical parameter data, the percentage improvement of the RNN model over the ANN model is 8.5412, 25.2926, 10.6301, and 5.8628 for predictions made one month in advance, and 12.5920, 26.3174, 32.3866, and 8.3180 for predictions made two months in advance. This finding is compelling evidence that RNN based prediction models are superior to ANNs and are suitable for long term planning.