Prediction of Environment Temperature for Grain Warehouses Based on LightGBM and Transformer

With the increasing prominence of global climate change and food security issues, the monitoring and control of grain warehouse environments have become particularly important. The external environment temperature has a direct impact on the internal temperature of grain warehouses and the safety of stored grain. Therefore, accurately predicting the external environment temperature of grain warehouses is crucial for formulating reasonable warehouse management strategies. Existing research mainly focuses on traditional statistical methods and simple machine learning models, which have limited performance in handling complex time series data. This study employs advanced deep learning models and optimizes input features through feature engineering to improve the accuracy of temperature predictions. The study proposes a method to generate new feature variables through feature fusion and validates its effectiveness. By adding new time series feature variables to the dataset, the model's understanding of time series data is enhanced, effectively reducing experimental errors. A comparative analysis of the predictive performance of various models identifies the best model and feature combination.