Research On Grain Condition Monitoring And Early Warning Model Based On Deep Learning

In recent years,grain production has continued to increase,and grain security issues have become increasingly prominent.Among the problems encountered in grain storage,temperature has the greatest impact on it.If the existing granary data can be used to effectively predict the temperature of grain,it is possible to find grain storage safety problems in advance and reduce the loss of grain during storage.Firstly,BP neural network and improved particle swarm optimization algorithm are used to optimize BP neural network to predict grain temperature.Then,the recurrent neural network and its variants are used to predict grain temperature.In order to further improve the prediction effect,different optimization methods are used to improve forecast model.This paper has done the following work to establish an effective grain temperature prediction model:(1)Using BP neural network to fit almost any function and automatically update its own parameters,establish a grain prediction model based on BP neural network;according to the shortcomings of BP neural network,it is improved by particle swarm optimization(PSO)algorithm,and the improved BP neural network is applied to the grain prediction model.At the same time,aiming at the shortcomings of particle swarm optimization,such as the slow convergence speed of particle swarm optimization algorithm,a new inertia weight is proposed and the genetic algorithm cross-over and mutation operation is used to improve the particle swarm optimization algorithm.Finally,the improved PSO-BP neural network is applied to the grain prediction model.The neural network prediction results of BP,PSO-BP and improved PSO-BP are tested.The BP mean square error is 0.02472,the PSO-BP mean square error is 0.01970,and the improved PSO-BP mean square error is 0.01592.The improved PSO-BP neural network has better stability than the other two,and can predict the change of grain temperature well.(2)Using recurrent neural network to process the characteristics of time series,establish a grain temperature prediction model based on recurrent neural network;according to LSTM,it can solve the characteristics of the disappearance of cyclic neural network gradient,and establish a LSTM-based grain prediction model;in order to get better prediction results,first optimize its network structure,and then use the improved activation function for comparison test;in order to solve the over-fitting phenomenon in the model,the regularization method is used to solve the over-fitting problem,then different optimization algorithms are applied in LSTM,and the prediction results are compared;then compare the improved BP grain forecasting model with the improved LSTM grain forecasting model.Finally,based on the predicted temperature and corresponding standards,a grain condition warning model is established.In order to verify the availability of the improved LSTM network structure,RNN and the three variants are predicted in the same situation,and the mean square error of RNN is 0.02840,the mean square error of LSTM is 0.02463,the mean square error of GRU is 0.01626,the mean square error of CLSTM is 0.01536,which can be seen that the predicted effect of the improved CLSTM preferred.At the same time,in order to verify the effect of using improved activation function,adding regularization and selecting better optimization algorithm,it predicts in different ways,and the final mean squared error of CLSTM is 0.00384.Furthermore,it can be seen that the predicted effect of CLSTM after optimization is better than that before optimization.