With the accelerated global energy structure transformation towards low-carbonization

the continuous expansion of grid-connected wind power capacity has made its power volatility and spatio-temporal coupling characteristics to be a critical challenge for power grid safety and economy. To address the issues of unclear spatio-temporal dependencies and inter-wind farm collaboration mechanisms in wind power cluster forecasting

this paper proposes a prediction method based on dual attention mechanisms and spatio-temporal gating fusion. Firstly

the maximal information coefficient is utilized to quantify the implicit influence of nonlinear meteorological features; combined with Pearson correlation matrix analysis of inter-wind farm cooperative relationships

the spatio-temporal coupling characteristics are decoupled. Secondly

an improved convolutional network with attention reinforcement and parameter redundancy suppression is developed to establish a spatio-temporal dual-driven feature extraction architecture with dual attention mechanisms in temporal and channel dimensions

and adaptively adjusts feature weights. Finally

a spatio-temporal gating mechanism module is designed to enhance spatio-temporal feature capturing capability

thereby effectively solving the problem of insufficient spatio-temporal feature fusion. Experimental results demonstrate that the proposed model significantly outperforms baseline models on wind farm cluster datasets

exhibiting strong generalization ability and practical applicability.