Short-Term Solar PV Forecasting Based on Recurrent Neural Network and Clustering

With the large-scale deployment of solar photovoltaic (PV) installation, managing the efficiency of the generation system has become essential. One of the main challenges facing solar PV power output lies in the difficulty in managing solar irradiance fluctuation. Generally speaking, the power output is heavily influenced by solar irradiance and sky conditions which are consistently changing. Thus, the ability to accurately forecast the solar PV power is critical for optimizing the generation system and ensuring the quality of service. In this paper, we propose a solar PV forecasting model using Recurrent Neural Network (RNN) in a Cascade model combined with Hierarchical Clustering for improving the overall prediction accuracy of solar PV forecast. The proposed model, upon comparing with other learning algorithms, namely, Feed-forward Artificial Neural Network (FFNN), GRU, Support Vector Regression (SVR) and K Nearest Neighbors (KNN) using the cluster data from K-Means Clustering and Hierarchical Clustering, had the lowest average NRMSE of 8.88% using Hierarchical clustered data. According to the results, Hierarchical Clustering suits better for solar PV forecast than K-means clustering.