Research on Weather Prediction Methods based on Hybrid Machine Learning Models

Accurate weather forecasting is vital for socio-economic development and people’s daily lives. However, the complex and dynamic nature of meteorological systems poses significant challenges to prediction tasks. To enhance forecasting accuracy and robustness, this paper proposes an intelligent weather prediction system based on multi-model integration. The system integrates various machine learning models including Long Short-Term Memory networks (LSTM), XGBoost, and Random Forest, achieving precise temperature forecasting through a dynamic weight fusion strategy. A deep learning LSTM model with attention mechanisms was constructed to effectively capture long-term dependencies in meteorological data. Additionally, a dynamic weight allocation method based on prediction errors was proposed, significantly improving the accuracy and robustness of ensemble forecasting. Experimental results demonstrate that the system achieves excellent performance in 7-day temperature forecasting tasks, with both Mean Absolute Error (MAE) and Mean Square Error (MSE) outperforming single-model approaches.