Reservoir Computing-based Ensemble Photovoltaic Power Generation Forecasting

Abstract

PV power forecasting has been studied in several methods, ML has been popular in recent years. Studies have reviewed that autoregressive moving average, SVM and ANN have been common in predicting data-driven time-series forecasting in the renewable energy domain, with 40% on solar predictions. In the recent decade as a viable alternative for training RNNs, reservoir computing was developed. However, these works do not access improvement on ESN when compared to uncertainty predictions in PV power generation. This research focuses on introducing a novel ensemble reservoir computing model using echo state networks. The proposed method uses the multivariate time series sequenced generated inputs to create the forecasting results by using the AdaBoost ensemble model which is integrated with reservoir computing. To evaluate the author has used two datasets with multivariate time series datasets and Bagging and single reservoir computing model has been used. The error values for the proposed model were lower than the benchmarked models.