Research On Atmospheric Visibility Prediction Based On Deep Learning

Atmospheric visibility is an important meteorological parameter that can significantly reflect the degree of air pollution.It is not only a key indicator of the transparency of the atmosphere,but also an important basis for evaluating air quality.Atmospheric visibility has important research significance in transportation,navigation,aviation,and national defense and military activities.Therefore,the precise prediction and forecast of atmospheric visibility play a pivotal role and significance in the control of urban air pollution,the protection of public traffic safety,and the protection of people's lives and property.At present,deep learning has received widespread attention and applications in computer vision,speech recognition,natural language processing and other fields.This thesis focuses on the deep learning prediction method of atmospheric visibility.First,using the data from the visibility meter,small weather station and particle size spectrometer of the Atmospheric Remote Sensing Center of Xi'an University of Technology,we carried out a study on the correlation between visibility and various meteorological and environmental parameters,and combined the principal component analysis method to analyze the various parameters.The impact on visibility provides a theoretical basis for the establishment of the input data set of the subsequent visibility network model.Secondly,based on the deep learning theory,according to the process design of the atmospheric visibility prediction model of the deep belief network,the experimental simulation and analysis of data preprocessing,the setting of the number of hidden layers and the number of nodes,the activation function and the weight function and other structural parameters are carried out.,Through the comparison and analysis of the atmospheric visibility prediction results,it is found that the Z-Score method is used for data normalization,using a double-layer hidden layer and a node number structure of(70,50),and the Tanh activation function and Adam optimization are selected.It can obtain the best visibility prediction effect,thus establishing a DBN network model suitable for atmospheric visibility prediction.Furthermore,the atmospheric visibility prediction and result analysis based on the deep belief network were carried out,and the prediction effects of the BP neural network and the convolutional neural network were compared.The results found that the DBN-based atmospheric visibility prediction results are significantly better than BP and CNN networks,with the best prediction results,with a prediction accuracy rate of 84%and a prediction correlation of 96%.The analysis of the contribution of various meteorological parameters to visibility prediction was carried out,and it was found that parameters such as radiation quantity have little effect on visibility prediction,while parameters such as PM2.5 and relative humidity have a greater impact on visibility prediction.Among them,PM2.5 The contribution is the highest.Finally,the forecasting method of the future trend of visibility based on the deep confidence network based on time series is adopted,and the short-term and long-term future trend forecast results of atmospheric visibility are obtained.The prediction results show that in the short-term visibility prediction,the prediction result after 2 hours of comprehensive judgment is better,and the prediction accuracy rate can reach 94%;in the long-term visibility prediction,the prediction result after 3 days is better and accurate rate can reach 79%,and it can cover the visibility change judgment of different weather conditions.