Prediction Of Regional Air Temperature And Humidity Based On Deep Learning

The differences in space and altitude between meteorological monitoring stations and agricultural parks have a certain impact on the prediction effect and prediction range of meteorological data.It is difficult to achieve the accurate prediction of short-term regional air temperature and humidity in agricultural parks.Such problems still restrict the development of agricultural production in various regions.Deep learning algorithm is an effective method for meteorological prediction in recent years,but an independent deep learning prediction model is difficult to achieve stable and high-precision prediction of the combination model.Given the above problems,this paper takes the meteorological data collected by the experimental demonstration station as the research object,and studies the regional air temperature and humidity prediction method based on deep learning.The main work is summarized as follows:(1)By analyzing the characteristics of meteorological data,aiming at the problems of noise and missing values in the original meteorological data,the k-nearest neighbor algorithm and data weighted supplement is used to process the missing values,and the statistical analysis and outlier detection is used to detect the abnormal values.Finally,210,000 valid data are obtained.Given the problem that meteorological data will affect the experimental results due to different dimensions and different sizes and ranges of data,the normalization formula is used to preprocess the data.Finally,the processed data are reconstructed by a sliding window,which lays the foundation for the subsequent prediction experiment of meteorological data.(2)Aiming at the problem that the prediction accuracy of a single time series prediction model is not high,the combined model is used to predict the air temperature and humidity.Cavity causal convolution is used to extract the feature of meteorological data.GRU and Skip-GRU methods are used to extract the long-term and long-periodic dependence of meteorological data.The attention mechanism algorithm is used to extract data correlation,and the time correlation dependence is obtained to improve the prediction accuracy.Finally,the linear correlation of data is learned by combining the autoregressive model.Given the problem that it is difficult to achieve the optimal combination of the hyper-parameters of the artificially selected model,the improved particle swarm algorithm is used to find the optimal solution vector of the hyper-parameters.The results show that in the prediction experiments of air temperature and humidity in different seasons in different regions for 1–24 hours and the next 2–3 days,the average absolute error of the combined model proposed in this paper is 17.34 % lower than that of the traditional prediction model.(3)To better apply the prediction method proposed in this paper to practice,a small program system based on temperature and humidity prediction is designed and implemented in this paper.Based on the analysis of the requirements of the small program,the model is transplanted to the background of the small program,and the small program client with two main functional modules is designed and developed.Finally,the prediction accuracy of air temperature and humidity in a week of the system is tested.The test results show that the average absolute error of air temperature in 1–24 hours is 1.95? and 10.55%RH,respectively,which verifies the effectiveness,real-time performance and prediction accuracy of the system.