Research On Intelligent Control Method Of Mushroom House Environment Based On Deep Learning

The high-efficiency and precise regulation of the greenhouse mushroom house plays a key role in maintaining the indoor microclimate environmental factors in the optimal range of edible fungus growth.But the main type of mushroom house is greenhouse at present,and it is an environmental system with large time delay,strong coupling and nonlinearity.The lag of data acquisition and control measures will cause adverse effects on the stability of greenhouse mushroom house control.Therefore,timely warning and taking corresponding control measures based on the scientific and effective prediction model are the important premise for the normal growth of edible fungi,which is very significant for the stable production of edible fungi in greenhouse.In recent years,artificial intelligence technology has penetrated into the agriculture step by step due to its huge advantages.In this paper,we takes the cultivation of pleurotus ostreatus during the fruiting period in the winter greenhouse as the test object,use deep learning prediction model to carry out environmental early warning for mushroom house,and we build the control strategy on this basis.The main work contents are as follows:(1)Experimental platform construction and data preprocessing.First,the experimental platform is built by the environmental monitoring and control system developed by Beijing Academy of agricultural and Forestry Sciences to realize data acquisition.Then,the linear interpolation and mean value method were used to fill and smooth repair the abnormal data.The key environmental factors affecting the temperature and humidity in mushroom house were determined by correlation analysis.Finally,the environmental distribution characteristics of mushroom house were analyzed,which laid the foundation for the construction of the prediction model of mushroom house.(2)Research on multi-point temperature and humidity prediction method of greenhouse mushroom house based on CNN-GRU.According to the characteristics of time series,different spatial distribution and non-linear of temperature and humidity in mushroom house,a multi-point prediction method of temperature and humidity for the mushroom house based on convolutional neural network(CNN)and gated recurrent unit neural network(GRU)was proposed.According to the experimental results,the average RMSE values for each point temperature in the mushroom house are 0.211 ℃ and 2.731%,respectively;The average MAE values of temperature and humidity in the mushroom house were 0.140 ℃ and 2.731%,respectively.Comparing with conventional BP neural network,long short-term memory neural network(LSTM),and gated recurrent unit neural network(GRU),the prediction model proposed had higher prediction accuracy.(3)Development of multi-point temperature and humidity early warning system for greenhouse mushroom house based on Py Qt5.The multi-point temperature and humidity warning client of greenhouse mushroom house was developed by using Python language,including landing interface,registration interface and warning interface.The data display,data preprocessing,temperature and humidity prediction,temperature and humidity warning,data storage and other core module functions were integrated.Each interface can be reasonably converted.Finally,it can be converted into executable file to facilitate production management personnel The operation of.(4)The design of the intelligent control mode of the greenhouse mushroom room.First,analyze the combination of internal control equipment in the mushroom house.According to the winter mushroom cultivation experience,the control strategy is formulated on the basis of the forecast and early warning system,and the experiment is verified in the mushroom house of the Beijing Academy of Agriculture and Forestry Sciences.This article focuses on the control of air relative humidity.The test results show that the maximum deviation of the average relative humidity of each area below the lower limit is 1.56%,and the maximum deviation above the upper limit is 0.5%,which is basically maintained It is between85%～95%,indicating that the control method based on deep learning designed in this paper is feasible.