Photovoltaic power prediction under insufficient historical data based on dendrite network and coupled information analysis

Abstract

In recent years, the installed capacity of photovoltaic solar energy has been increasing year by year. The existing power prediction methods are difficult to achieve reliable power generation prediction for PV equipment that has just been put into service. This poses great difficulties for power system management and dispatching. Therefore, establishing a reliable PV power prediction model for this situation is important for the rapid integration of new-built PV installations into the power system for energy management and dispatching. In this paper, an ultra-short-term PV power prediction method based on coupled information analysis and a dendritic network is proposed. First, an improved Hampel filter is proposed to improve the accuracy of anomaly data processing by analyzing the information between strongly coupled variables. In addition, a dendritic network modeling method is introduced for PV generation prediction. Compared to other solutions, this prediction approach relies on very little historical data to achieve reliable predictions and also features a simple model structure and high generalization. Forty different sets of tests were conducted according to different weather conditions and data conditions. The experimental results show that the proposed method can achieve higher prediction performance and stability compared with the benchmark model.