Research On Comprehensive Supervision Of Urban Mobile Source Emissions Based On Spatiotemporal Data Mining

At present,the situation of China’s urban air quality is severe.The“China Eco-Environmental Status Bulletin(2018)”points out that 217 cities have exceeded the am-bient air quality,accounting for 64.2%among the 338 prefecture-level cities nation-wide.And according to”China Vehicle Environmental Management Annual Report(2019)”,the domestic vehicles volume is up to 307 million,and the total accounting of four emissions(CO,HC,NOx,PM)is up to 40.653 million tons.Mobile source emissions account for more than 80%of carbon monoxide and hydrocarbons,and more than 90%of nitrogen oxides and solid particles in the urban air pollutants.And the mobile source emissions have become the main source of urban air pollution,causing serious damage to the social ecological environment.Therefore,it is necessary to study the comprehensive supervision and analysis methods of urban mobile source emissions,which is of great significance for protecting public health and improving rational urban planning as well as traffic conditions.However,due to the high cost of construction and maintenance of mobile source emission remote sensing equipment stations,the monitoring stations deployed in the city are very sparse,and it is difficult to implement real-time monitoring of mobile source emissions in the entire city by deploying emission remote sensing equipments on ev-ery road segment.Meanwhile,the temporal and spatial distribution of urban mobile source emissions is affected by many complex factors.On the one hand,from the per-spective of long-term vehicle emission inventory calculation,it mainly depends on the city’s total vehicle volume and vehicle type composition.On the other hand,in terms of short-term and real-time variation of traffic emissions,it is mainly influenced by ur-ban road network topology,traffic flow conditions and external meteorological factors.This series of factors has led to great challenges in achieving full-time monitoring and comprehensive supervision of urban mobile source emissions.To address the aforementioned challenges,this thesis considers the spatial spar-sity and heterogeneity of mobile source emission measurement data,complex spatio-temporal dynamic characteristics,as well as the effects of road network connectivity and other external environmental factors,proposing a research on regulation of urban mobile source emissions based on spatio-temporal data mining.The main contents can be summarized as follows:(1)To tackle the spatial sparsity and heterogeneity of urban mobile source emission measurement data,we design a data augmentation method combining emission factor model and adversarial learning strategy.Firstly,we comprehensively consider the in-fluence of external environmental information such as traffic flow,road network struc-ture,vehicle driving conditions,etc.,and establish an amended emission factor model to determine the a priori distribution of emission data.Then,we design prior model loss and conditional constraint loss based on the spatial measurement distribution to construct a data generation network that integrates emission factor models and a two-class discrimination network to test whether the input data is pollution monitoring data.By introducing the adversarial learning strategy between the data generation network and the discrimination network,we can obtain a conditional measurement Generative Adversarial Networks(CMGAN)to generate realistic emission data.And the com-paring results demonstrate that our proposed model considers the spatial heterogeneity of mobile source emission data distribution and can mimic the real vehicle emission distribution.(2)Considering the spatiotemporal variation of mobile source emissions in urban areas and the influence of multi-source environmental factors,a spatiotemporal pat-tern aware network(STPAN)is proposed for the coarse-grained urban area emission prediction.Firstly,we use gridded data representation to construct the correlation be-tween mobile source emissions and external environmental factors.Then,the closeness,period,and trend temporal features of emission sequences are extracted by multiple time-dependent processing units.Finally,we can realize the coarse-grained urban grid area mobile source emission trend prediction in the deep spatiotemporal feature space by merging the spatiotemporal patterns and the external environmental features with early-late spatiotemporal residual unit.Experimental results show that this method is superior to existing methods in predicting the accuracy of different pollutants,and can effectively deal with the spatial and temporal heterogeneity of the pollution distribu-tion of mobile sources in the urban grid area,as well as the impact of complex external environmental factors.(3)To meet the requirement of refined spatiotemporal distribution of mobile source pollution in urban road networks,we propose a spatiotemporal graph convolution multi-fusion network(ST-MFGCN).Firstly,we leverage the graph data representation to cor-relate the emissions with road network inherent connectivity and traffic flow conditions.Then,we design a spatiotemporal graph convolution module to capture the spatiotem-poral interaction features between different road network nodes.And a multi-fusion strategy is used to merge the spatiotemporal patterns and the external features as well as fit the mutation of emission measurement data.By this way,we can realize the predic-tion of the spatiotemporal distribution of mobile source emissions in fine-grained urban road networks.Finally,the proposed model is evaluated on the practical monitoring data of vehicle emission data in Hefei,and the results demonstrate that the proposed model can address the spatiotemporal interactions between different road segments as well as different influence of complex external traffic environment factors.And it achieves low prediction errors on the prediction of different emissions in road networks,and has good adaptability to data size.(4)To deal with the problem of model dependence in mobile source emission management,we propose a urban road emission supervision strategy based on deep reinforcement learning(DRL)to learn an optimal traffic flow and speed management policy.The hybrid environmental states and combination return value function of emis-sions and traffic restrictions are designed to construct the deep long-term return value estimation DRL model(EFRL).Training the EFRL model with a sampling pool of emission sequence experience to establish the supervision strategy model.Realizing the online auxiliary decision-making of road mobile source emissions supervision with the dynamic road traffic restriction policy.Compared with the baseline methods,the proposed strategy can effectively reduce mobile source emissions on the target road segment.In summary,this thesis proposes a full technical chain of“Emission data augmenta-tion-Emission spatio-temporal characteristics analysis-Emission supervision strategy generation”a hierarchical methodology of urban mobile source emissions supervision based on spatio-temporal data mining,providing reliable data support and decision-making basis for the comprehensive supervision of urban mobile source emissions.