Multi-channel attention mechanism graph convolutional network considering cumulative effect and temporal causality for day-ahead wind power prediction

Under the current situation where the proportion of wind power generation in the existing energy structure is gradually increasing, accurate short-term power prediction of wind farm cluster (WFC) plays an important auxiliary role in the formulation of regional power generation plans. However, it is difficult to clarify the correlation between wind farms in the power prediction of WFC, which leads to the difficulty of stable and accurate prediction. Therefore, this paper proposes a short-term power prediction method for WFC considering wind power cumulative effect (CE) and temporal causality (TC). Firstly, the two relationships (CE and TC) between wind farms are clarified and used to construct an adjacency matrix separately, then the error characteristics (EC) are used to form an adjacency matrix as well. Secondly, the adjacency matrix information with different attributes is fused by the multi-channel attention mechanism to obtain the wind power prediction results. Finally, the proposed method is applied to three WFCs in China, by which the universality and effectiveness of the method is verified. Compared with the traditional prediction method, its RMSE is reduced by 1.55%, 1% and 3.4% on average. • clarified two correlations between wind farms; • applied the cumulative effect to day-ahead power prediction of wind farm cluster; • proposed a novel method for utilizing error characteristics; • proposed a multi-channel attention mechanism for graph neural network;