Short-Term Solar Power Generation Forecasting Via Continuous Conditional Random Fields

Abstract

Solar power generation with highly variable mode brings adverse effects on the grid. In order to reduce the negative impact on the grid, we use continuous conditional random fields (CCRF) to forecast solar generation. The CCRF is a powerful tool for relationship learning, which can capture the interaction between predicted solar generation. The potential function of the CCRF is designed as quadratic forms, which can transform the learning problems of the CCRF to convex optimization problems. In addition, it can perform probabilistic forecasting. To avoid over-fitting, the regularization of the weight is added to the loss function. We conduct the experiments on the freely available dataset to evaluate the forecasting performance. Experimental results show that the CCRF forecasting model can further improve the forecasting accuracy, compared with benchmarking forecasting method.