A novel scenario generation and forecasting method for multiple future states at source side based on combinatorial models

Abstract

This paper proposes a novel method for scenario generation and scenario forecasting considering wind-photovoltaic-hydro-thermal power sources. With the rapid development and application prospects of energy internet and grid modernization, accurate source-side multiple output modeling and forecasting is essential to observe and estimate the multiple curves. It is becoming more and more important to optimize the operation efficiency of power grid. The traditional methods mainly focus on multi-scene generation of a single power source. Grid scheduling requirements that could not take into account multiple time correlations and multiple scenario forecasting. The paper proposes a scenario forecasting method considering scenario generation. Firstly, mutual Information (MI) and depth-first traversal (DFS) are combined to analyze the factors associated with power generation and build the MI-DFS-K2-Bayes network to generate multiple scenarios of wind- photovoltaic-hydro-thermal power. Secondly, the improved k-means algorithm is proposed to obtain the center curve of each power source. The algorithm improves the accuracy of the multivariate clustering center curve. It solves the problem that the initial clustering center and the number of clusters are difficult to select, and builds the multivariate running state. Finally, the upper and lower interval of point forecasting constrained by confidence parameters is proposed to make the running state scene match the point forecasting. The future state curves of each power source in short time scale are obtained. The future state scenario is implemented. Numerical examples demonstrate that the proposed method reduces the modeling error of wind- photovoltaic-hydro-thermal power curves by 0.13%–0.4%. The overall forecasting accuracy is significantly better than the state of the study.