Research On Detection And Forecasting Algorithm Of Motor Vehicle Exhaust Concentration Based On Remote Sensing Monitoring Data

The emission of pollutants from motor vehicle exhaust has emerged as a primary contributor to the decline in urban air quality.The swift expansion of China’s economic output and the concurrent elevation of living standards have engendered a proliferation of motor vehicles,which has furnished greater convenience for daily commuting and general mobility.However,this widespread adoption has also significantly exacerbated the issue of motor vehicle exhaust emissions,thereby rendering it an increasingly conspicuous and pressing environmental concern.Hence,it is imperative to implement an efficient system for monitoring and controlling the emission of motor vehicle exhaust on urban roads,in order to safeguard atmospheric ecological integrity and promote public health.However,the self-developed remote sensing detection system has the problems of incomplete accuracy and instability,and the detection results of the existing machine learning model are not high.It is difficult to adapt to complex field environment changes to realize real-time and accurate detection of urban road vehicle exhaust pollutants.The inherent randomness and unbalance of complex meteorological factors and traffic flow lead to strong nonlinear characteristics of vehicle exhaust pollutants,which is difficult to ensure the stable forecasting of pollution.Moreover,the spatial and temporal distribution of vehicle exhaust pollution in urban areas is easily affected by the topological structure of urban roads,traffic flow conditions and other factors.This series of problem lead to the realization of urban vehicle pollution emission detection and forecasting is very challenging.To address the aforementioned challenges,this thesis considers the inherent nonlinear and linear characteristics of vehicle pollutants,complex temporal and spatial dynamic trends as well as the influence of external environmental factors,proposing the research on the detection and forecasting algorithm of urban vehicle exhaust concentration based on remote sensing monitoring data,so as to help effectively control urban air pollution and protect human ecological environment.The main work and research findings of this thesis are as follows:(1)To address the issues related to deficient stability in detection and inaccuracies in results arising from utilizing combustion equation inverse solution to obtain exhaust gas concentration in stationary motor vehicle exhaust remote sensing monitoring system,as well as the shortcomings of the current machine learning model that shows low accuracy in detection results,an improved Stacking inversion estimation algorithm is proposed to improve the detection accuracy of motor vehicle exhaust pollution.The improved Stacking model adopts the three-layer stacking model to build the accuracy improvement model.The first layer is the main learner,and the second layer is the remote sensing monitoring feature data,which is transformed and combined by the nonparametric machine learning method with high detection accuracy.The second layer is a secondary learner,in which the remote sensing monitoring feature data is trained using a simple statistical learning model to obtain the preliminary estimation results of the remote sensing monitoring data.The first two layers actually constitute a simple Stacking model;The third layer integrates the preliminary estimation results of the second layer with the exhaust detection results of other strong machine learning models to obtain the detection results of the entire model.Experimental results show that the proposed method is superior to the popular neural network model and the traditional machine learning model in the detection accuracy of vehicle exhaust concentration,and can effectively improve the detection accuracy.(2)Considering the meteorological factors and the inherent randomness and unbalance of traffic flow,vehicle exhaust pollution has strong non-stationarity and nonlinear characteristics.A new hybrid deep learning framework,namely a multi-component vehicle exhaust forecasting method based on dual attention fusion network,was proposed.Multiple exhaust forecasting was conducted using the effective combination of temporal convolutional network,convolutional neural network,skip-based long and short term memory network,dual attention mechanism and autoregressive decomposition model.Considering the interference of meteorological factors,multi-layer perceptron is used to capture the nonlinear characteristics of external factors to compensate for pollution to effectively improve the accuracy of forecasting.The forecasting model is composed of three main components.Firstly,for the nonlinear part of pollution,the temporal convolutional network,convolutional neural network and skip-based long and short term memory network are used to capture the dynamic long and short term time dependence of multiple exhaust effectively.Secondly,for the linear characteristics of pollution,the autoregressive decomposition model is used to solve the problem of scale insensitivity caused by the neural network model.Thirdly,for the interference of meteorological factors on pollution,multi-layer perceptron is used to compensate the influence of external factors on pollution emission.The experimental results show that the proposed method is superior to the existing methods in both short and long term forecasting accuracy of multiple vehicle exhaust pollution,and can effectively deal with the nonlinear and linear characteristics of exhaust pollution,as well as the influence of various complex external environmental factors.(3)Considering the spatio-temporal variation characteristics of road vehicle exhaust pollution in urban areas,the strong nonlinearity of emission data,the time correlation between different road segments and the spatial interaction and other factors,an attention-based global and local spatio-temporal graph convolution network for pollution spatio-temporal variation was proposed.A spatio-temporal attention mechanism was used to capture the dynamic spatio-temporal correlation of vehicle pollutants by integrating the components of hourly,daily and weekly time series.Global and local spatial graph convolutional networks are designed to capture the hidden global and local spatial dependencies in regional motor vehicle pollution.The experimental results show that this method can effectively capture the spatial and temporal distribution trend of vehicle pollutants in the road network by using the connected structure information,and has good adaptability to different types of data sets.Drawing upon the aforementioned research,this thesis successfully explores the two fundamental dimensions associated with detecting and forecasting motor vehicle exhaust pollution emissions on urban roads.To this end,a remote sensing monitoringbased approach for monitoring and forecasting such emissions is developed,thereby furnishing technical support for facilitating effective surveillance and scientific mitigation of urban motor vehicle exhaust pollution.This approach holds immense promise for amplifying the efficacy of urban air quality management and national air pollution management.