Multi-Step Ahead Photovoltaic Power Forecasting Model Based on Timegan, Soft Dtw-Based K-Medoids Clustering, And a Cnn-Gru Hybrid Neural Network

Accurate PV (Photovoltaic) power output forecasting is very important for making economic and reliable power dispatching plans. At present, the PV power prediction technology is still in the primary stage, especially for the forecasting cases with large forecasting time and when the weather changes suddenly, the forecasting accuracy is hard-pressed to meet demand of power grid dispatching. This study proposes a multi-step ahead PV power forecasting model, which combines time-series generative adversarial networks (TimeGAN), DTW-based K-medoids clustering algorithms, and a mixed neural network model composed by a convolutional neural network (CNN) and gated recurrent units (GRU). Given the uneven distribution of PV power data for different weather conditions and insufficient training samples, it is proposed to use TimeGAN to realize the enhancement and expansion of the historical PV power data. The K-medoids method is used to effectively classify the expanded PV power data, in which the soft-DTW is used as the time series similarity measure of the K-medoids clustering method to enhance the clustering accuracy. Finally, the mixed neural network of CNN-GRU is applied to forecast the PV power output for different weather conditions. The outcomes demonstrate that this method has significantly improved the accuracy of multi-step forecasts compared with other benchmark models