Solar Power Prediction Using Iterative Network Pruning Technique for Microgrid Operation

This paper proposes an efficient day-ahead Photovoltaic (PV) forecasting method by improving our previous method. The proposed approach uses only public weather data of a local area and is applied to a community microgrid. In our previous works, the weather-clustered feedforward neural networks (WFNNs) have been created from different weather patterns and time zones. In this paper, an iterative pruning (IP) algorithm is introduced in the WFNNs structure to achieve efficient computation for learning process. The authors utilize confidence intervals (CIs) for secure the reliability of the solar prediction. The obtained CIs are effectively used for microgrid system operations and estimated of robust power system security. The computation times are improved about 90% against our previous method.