To address the issues of insufficient feature extraction and inadequate prediction accuracy of single models in wind turbine condition forecasting, this study proposes a wind turbine operational condition prediction method that integrates the CatBoost algorithm with the Long Short-Term Memory network (LSTM). Firstly, based on the wind turbine sensor features and temporal features, the SVFE (a feature extraction method, assume its full name is known in the specific context) and MVFE (another feature extraction method, assume its full name is known in the specific context) methods are employed for cross-fusion to generate global composite features. Additionally, feature dimension reduction is achieved by incorporating grey relational analysis improved with the entropy weight method. Secondly, the Sparrow Search Algorithm (SSA) enhanced by chaotic mapping, termed CSSA, is introduced to conduct global optimization of the hyperparameters of the LSTM model, enabling adaptive screening and precise determination of the optimal parameter combination. Finally, the CatBoost model and the optimized LSTM model are deeply fused using an optimal weighted combination strategy to enhance prediction accuracy and model generalization capability. Taking the wind turbines of a phosphorus chemical enterprise in Yichang, China, as an example, the proposed CSSA-CatBoost-LSTM wind turbine condition prediction method was validated. The validation results demonstrate significant improvements in both accuracy and reliability of this method.