The Design And Implementation Of Intelligent Monitoring System Of Grain Condition Based On Neural Network

As China is a big country with large demand for grain,the grain problem is related to the national prosperity and people's livelihood.Scientific grain storage is an important part after grain production.In the process of food storage,insect pests and mildew are the important hidden dangers of food loss.In order to reduce the loss of stored grain and ensure the safety of stored grain,it is necessary to predict the change of grain condition in time,and take measures such as ventilation and fumigation as early as possible before the occurrence of problems.With the development of artificial intelligence technology,machine learning and neural network technology can replace the traditional method based on individual experience to judge the safety of stored grain.The prediction model based on neural network can improve the accuracy and be real-time when predict stored grain safety problem,The main work of this thesis as follows:In this thesis,an intelligent monitoring system of grain situation is designed.The hardware design of the system realizes the monitoring of environmental parameters of grain storage.The GRU network and Attention mechanism are used to extract the time series characteristics of grain situation data:1?This thesis designs a feature extraction model for temporal series data based on GRUAttention network.According to the temporal characteristics of the food situation data,attention mechanism was added to the GRU network to assign weight to the food situation features,and Dropout algorithm was used in the hidden layer to prevent the network model from overfitting.The experiment shows that the RMSE error of the model was reduced by 18% after Dropout and 4.5% after attention mechanism was added.2?An IPSO-GRU-Attention algorithm for predicting grain temperature was proposed.Because GRU network is trapped in extreme value and can not converge during training,particle swarm optimization algorithm(PSO)is added during network initialization to adjust the initial weight and threshold of network for the global optimal solution interval.Linear differential decrement strategy inertia factor and asynchronous learning factor are used to optimize the particle search strategy in PSO to avoid premature convergence of particles.After Adam optimization training,the final grain condition temperature prediction model was obtained.The experiment showed that the RMSE error of the model was 0.070,the MAE error was 0.058,and the MAPE error was 0.332%.After optimization,the RMSE error was reduced by 13%.3?The grain situation monitoring system was designed and implemented.The hardware of the system adopted wireless terminal equipment,serial port relays,sensors and Linux cloud server to realize parallel collection and upload of temperature and humidity data and multichannel carbon dioxide data collection and upload.The system background receives and parses the data packet,and stores it in the My SQ L database for early warning analysis.The system is based on B/S structure.Users can query the grain situation data,the analysis of the warning results,and manage system and personnel data through the front end.After testing,the average accuracy of the system's early warning is 89.9%,which meets the demand of food situation early warning and meets the requirements of the system.