Solar Radiation Intensity Probabilistic Forecasting Based on K-Means Time Series Clustering and Gaussian Process Regression

Solar radiation intensity is intermittent and uncertain under the influence of meteorological conditions. Clustering them and obtaining high-precision and reliable probabilistic forecasting results play a vital role in the planning and management of solar power. In this study, a novel K-means time series clustering (K-MTSC) algorithm is first proposed to cluster solar radiation intensity and compared with astronomy method and K-means. Then, different feature inputs for different categories of solar radiation intensity are screened. Afterwards, the different kernel functions of Gaussian process regression (GPR) are compared and optimal kernel function is selected in terms of deterministic forecasting and probabilistic forecasting for different categories. Finally, the case study in Tibet province, China are performed to verify the validity and practicability of this research model and method. In this experiment, the average accuracy of GPR is 44% higher than that of Artificial Neural Network ANN, and 17% higher than that of Support Vector Regression. The experiments show that (1) the clustering results obtained by the K-MTSC algorithm have a larger inter-group distance and a smaller intra-group distance, and at the same time, it will not destroy the continuity of the time series. (2) The probability forecast results obtained by GPR are reliable and high-accuracy.