Research On Analysis And Prediction Method And Visualization Of Stored Grain Condition

China is a big grain country and a big population,the food security has a direct relation with our country’s development,but our country every year grain loss in circulation link can reach 70 billion catties,accounted for about 5% of the total output.As the most critical link of grain circulation,storage is directly related to grain security.Grain condition includes temperature and humidity of grain pile,gas and pests,etc.Grain temperature is the most important index of grain condition,which is taken as an important index to measure grain condition in this paper.Therefore,this paper proposes a grain temperature prediction model based on GRU,and designs and implements an analysis and visualization system of grain situation to improve the accuracy of grain situation judgment.The main content of this paper is as follows:(1)Analyze and preprocess test data.Firstly,the annual temperature change of Yushu City in 2021,the structure of grain depot and the layout of sensors were analyzed,and the2021 grain situation data of Yushu City direct warehouse was used for analysis and preprocessing.The outliers were removed and interpolation of missing values were carried out,and the overall data were analyzed by descriptive statistics.It was found that grain was a bad conductor of heat.The grain temperature near the outside is easy to be changed by the change of external conditions,and the grain temperature inside is more stable.Grain temperature in winter there is a "cold skin and hot core" phenomenon and in summer there is a "hot skin and cold core" phenomenon,and the overall existence of the same change trend with the outside temperature,but there is a certain lag phenomenon.Finally,BP neural network was used for 3D interpolation of the data.On the basis of 480 pieces of data at one time,a total of 283101 pieces of data were obtained by interpolation at 0.1m interval for future use of 3D visualization system.(2)A grain temperature prediction model based on GRU is proposed.In order to reduce the complexity of the GRU model and improve the prediction accuracy,the algorithm combining genetic algorithm(GA)and particle swarm optimization(PSO)is used to optimize the superparameters of the GRU neural network model,obtain the hidden layer neurons and batch size of the GRU model,and solve the model overfitting problem by using Dropout.Adam optimization algorithm was used to solve the problem that the fixed learning rate caused the model to fall into the local optimal advantage.The activation function and deactivation rate of the GRU model were determined by comparing multiple experiments,and the optimal structure of the GRU model was finally determined by the optimization algorithm.Finally,the attention mechanism layer was added after the GRU layer to improve the prediction accuracy of the model.Comparison tests were conducted with GRU model,LSTM model,Lstmattention model and GA-PSO-GRU-Attention model proposed in this paper.The tests showed that GA-PSO-GRU-Attention model had the best prediction effect,and MAE was 0.046.MSE was 0.003;R2 was 0.914,which proved that the grain temperature prediction model proposed in this paper was effective.(3)Build a front-end visualization system of B/S architecture,including personnel login,data processing module,three-dimensional interpolation module and visual display module.The two-dimensional visual display module mainly draws the contour map of grain temperature data,and the three-dimensional visualization module carries out the threedimensional visualization of the data after 3D interpolation using BP neural network.The system can carry out the three-dimensional display of grain temperature on the web side,with the functions of zooming,dragging and other functions.It can make the staff grasp the grain temperature data more accurately.After testing,this system can reduce the workload of the staff in the grain depot,improve the stability of monitoring,and has practical value.