Temperature Prediction for Stored Grain: A Multi-model Fusion Approach Based on Machine Learning
Temperature fluctuations significantly affect microorganism growth and pest activities in grain pile, precise monitoring and forecasting temperature of stored grain are essential for maintaining the quality and safety of grain storage. This paper proposes a multi-model fusion approach to predict grain temperature using historical temperature data of stored grain and meteorological data from the region. Firstly, four distinct machine learning models, namely Adaboost, decision tree, extra trees, and random forest, are fine-tuned through parameter optimization to enhance their predictive capabilities respectively; Subsequently, these optimized models are fused to form different ensemble models, which are compared for predidction accuracy to obtain the optimal fusion model. In essence, the fusion process integrates the predictions of each individual model as new feature inputs into the fusion models. Furthermore, random forest is utilized to identify the key factors influencing grain temperature, providing insights into the importance of different influencing factors. The experimental results demonstrate that the proposed fusion models can achieve higher prediction accuracy and robustness compared with the single-model prediction methods. Additionally, the analysis of feature importance also offers empirical evidence for understanding the factors influencing grain temperature.