Spatio-Temporal Probabilistic Forecasting of Photovoltaic Power Based on Monotone Broad Learning System and Copula Theory
Nan Zhou, Xiaoyuan Xu, Zheng Yan, Mohammad Shahidehpour
Abstract
Probabilistic forecasting of photovoltaic (PV) power provides system operators with pertinent information on the uncertainty of PV power generation. This paper proposes a spatio-temporal probabilistic forecasting model based on monotone broad learning system (MBLS) and Copula theory. MBLS is a novel neural network structure for providing an efficient quantile regression solution. MBLS guarantees the monotonicity between quantiles and their probability for thoroughly avoiding the quantile crossing problem. The historical PV data are then clustered using the self-organizing map and samples in each cluster are used for Copula parameter estimations. The proposed approach provides an efficient spatio-temporal forecast of multiple PV plants by combining marginal distributions predicted by MBLS with Copula functions. The real-world data of PV plants in Australia and USA are used to the validate the superiority of the proposed method through detailed comparisons with existing methods using comprehensive evaluation criteria. The presented results demonstrate that the proposed method can provide high-quality probabilistic forecasts corresponding with PV power scenarios.