Abstract: With the increasing proportion of wind power installations, accurately predicting the output power of wind turbines is crucial for ensuring power quality and enhancing the stability of power systems. To address the multi-modal characteristics, nonlinear features, and temporal correlation issues in wind farm turbine data, this study introduces a grouping scheme based on the Gaussian Mixture Model (GMM) and develops a combined day ahead wind power forecasting model integrating Convolutional Neural Networks (CNN), Bidirectional Long Short Term Memory (BiLSTM), and the Attention mechanism. First, the GMM is employed to categorize wind turbines into several types based on the characteristics of historical turbine data. Subsequently, a grouped prediction model, CNN-BiLSTM-Attn, is constructed for each subgroup to perform day-ahead wind power forecasting. In this model, CNN extracts local features from the nonlinear data of turbines, BiLSTM captures long-term dependencies, and the Attention mechanism assigns weights to the features extracted by BiLSTM. Validation using data from a wind farm demonstrates that the proposed method outperforms traditional single prediction algorithms and other grouped prediction approaches, providing an accurate and efficient solution for day-ahead wind power forecasting.