<p>In recent years, with the continuous expansion of wind power installed capacity, the risks of grid security hazards caused by the volatility and intermittency of wind power during grid connection have also increased accordingly, and high-accuracy wind power forecasting can effectively support the safe operation of power grids to address this issue. To this end, aiming at the problems of insufficient spatiotemporal information capture and multi-step forecasting homogenization existing in wind turbine clusters of wind farms, this study proposes a spatiotemporal graph-based forecasting method for wind turbine clusters that considers spatiotemporal errors and iterative forecasting to improve forecasting performance: first, a directed graph of spatiotemporal errors is constructed based on the graph convolution forecasting errors derived from dual-channel global information; second, a three-channel spatiotemporal graph convolutional network (GCN), integrated with the directed spatiotemporal error graph, undirected geographical location graph, and undirected power correlation graph, is used to extract spatiotemporal feature information and generate the initial forecasting feature sequence; finally, the initial forecasting feature sequence is sliced and segmented to enable the iterative multi-step forecasting module, thereby obtaining the final forecasting results. Verification with actual operational data shows that the proposed method can improve the forecasting performance of wind farms, effectively support scheduling departments in advance planning for the grid connection of wind farm power generation, and provide a reliable technical basis for the safe and stable operation of power grids.</p>