This research proposes a novel spatio-temporal approach that integrates ConvLSTM (Convolutional Long Short-Term Memory) networks and GNNs(Graph Neural Networks) to model and predict wind power generation and its impact on power flow. By using ConvLSTM models, the work achieves an R^2 value of 0.977 indicating high accuracy in forecasting wind generation dynamics across various temporal and spatial scales. Meanwhile, the GNN model, achieving an R^2 of 0.771, shows a viable approach to modeling power grid performance without the need for traditional iterative Newton-Raphson load flow methods. While the GNN model could be further optimized, it does mark a substantial improvement in scaling machine learning for real-time grid management. By harmonizing these models, this research addresses critical gaps in current approaches to the integration of renewable energy sources into power grids, aligning with ongoing efforts to enhance grid reliability and efficiency in the face of increasing renewable penetration.