RSNN: Rate encoding mechanism-based spiking neural network for renewable energy forecasting

The smart grid enhances the integration of renewable energy sources into power generation, providing sustainable energy solutions. Accurate forecasting is essential for integrating variable renewable sources like solar and wind to ensure reliable grid operations. However, the inherent variability in solar and wind data poses significant challenges for time-based forecasting accuracy. Traditional regression and machine learning models struggle to manage non-linear data and fail to capture temporal features effectively, leading to suboptimal forecasting performance. To overcome this limitation, this paper employs a rate-encoded spiking neural network designed to improve forecasting accuracy. The training methodology focuses on addressing a 24-step-ahead forecasting problem, utilizing various values for the data-splitting random_state parameter to enhance model robustness. The rate encoding mechanism is developed to generate the time of spike for encoding information in terms of spike trains. The recommended models achieve superior performance to deep learning forecasting models, such as RNN, LSTM, GRU, CNN-LSTM, CNN-GRU, CNN-LSTM-ATTENTION, and CNN-GRU-ATTENTION for different selected case studies. The average value of performance MSE and R² score of the proposed model is 55% and 1.08% for wind, 122.5%, and 20% for solar forecasting. The proposed model's robustness analysis is conducted with the 30 simulation runs. In addition, the computational analysis is carried out by recording the model training time for solar and wind data. The accuracy of the proposed model depends on the data properties and locations. The proposed model examines three climate conditions to evaluate its performance under intraday data variability for all seven selected climate zones.