Microgrid Forecasting Using Multiple Deep learning Models

To achieve electrical grid decentralization, microgrids and distributed energy resources are widely used. However, these systems are dependent on inherently unstable renewable energy sources such as wind and solar energy and thus, forecasting electrical generation and demand is crucial for decision-making and energy control. The effectiveness of the three variants of Recurrent Neural Networks (RNNs): Long Short-Term Memory (LSTM), standard-Recurrent Neural Network (s-RNN), and Gated Recurrent Unit (GRU) models for short-term forecasting of solar and wind energy is investigated in this study. Using the root mean square error (RMSE) to evaluate accuracy, it was identified that GRU performed best with RMSE of 0.035 kW. LSTM followed in terms of accuracy with an RMSE of 0.038 kW. The s-RNN had the poorest results with RMSEs being measured at 0.042 kW in photovoltaic (PV) forecasting. In the case of wind forecasting the GRU led the best result with 0.134 kW of RMSE, 0.138 kW for LSTM and 0.142 kW for s-RNN. In forecasting of load demand, the GRU and LSTM got the same result with RMSE of 0.029 kW and 0.032 kW for s-RNN. Nevertheless, the LSTM and GRU had longer computational times of 10 seconds compared to the s-RNN's of 4 seconds.