Machine Learning And Numerical Simulation Based Grain Pile Temperature Field Prediction And Application

As grain production continues to increase,grain storage security has become an important cornerstone to ensure the country’s economic development.In the storage process,the grain storage pile as a complex ecosystem,factors affecting food security are temperature,humidity,moisture and pests,etc.,of which the most influential is the grain temperature.Once the temperature of the grain pile exceeds the safe storage value,it will lead to the rapid reproduction of pests and molds,jeopardizing food security.Therefore,accurate analysis,judgment and prediction of grain pile temperature is of great significance for safe grain storage.The main contents of this paper are as follows:(1)According to the ecological environment of grain storage,the main factors affecting grain pile temperature are determined based on gray correlation analysis as temperature inside the grain bin,humidity inside the grain bin,temperature outside the grain bin,humidity outside the grain bin,average temperature of the grain bin,and surface temperature,and based on the historical monitoring data affecting grain pile temperature,the Lagrangian interpolation method is carried out to process the missing and abnormal values in the data,and for the processed data,considering the complexity of the grain pile temperature field and The BP neural network,radial basis function(RBF),random forest(RF),and support vector regression(SVR)prediction models were established for the processed data,and the SVR model with the best prediction effect was determined by the prediction results.(2)To address the problem that the prediction model is easy to fall into the local optimum,two optimization algorithms,whale optimization algorithm(WOA)and gray wolf optimization algorithm(GWO),are invoked to optimize the SVR model with better prediction effect;for the problem of data volume and model accuracy in the study,the optimized WOASVR model is integrated using the guided aggregation algorithm(Bagging)integration method,and by comparing prediction results and evaluation indexes,the constructed Bagging-WOASVR model has higher accuracy,which improves the prediction accuracy and convergence speed of the base model.(3)Based on the heat transfer theory of porous medium and heat mass transfer mechanism,the numerical simulation model in the static storage process of japonica rice pile is constructed according to the conservation equation,and the temperature transfer process and change law in the stored grain pile are analyzed based on the respiration of the grain itself.The combination of the real bin data and the numerical simulation model shows that: during the static storage process,the temperature of the grain pile at the bin wall is influenced by the external environment,and the internal temperature changes relatively slowly;the temperature of the bin wall rises significantly in summer and transfers heat to the central area,and a hightemperature gathering area is formed inside the bin in winter.(4)A grain temperature prediction module was designed according to the machine learning prediction model to realize the prediction of grain pile temperature in different layers of the silo,and the grain temperature prediction module was applied to the grain storage safety system to provide technical support for the mechanical ventilation work.