Research On Traffic Volume Classification And Prediction Algorithm In The Scenario Of Road Emission Monitoring

With the growth of car ownership in cities,exhaust emissions from the road network have become a major source of air pollution in cities.Therefore,the monitoring and control of emissions in the road network is an essential part of urban environmental management.Because of the high price and maintenance cost of exhaust gas detection equipment,and the close relationship between automobile traffic and exhaust emissions and easy to obtain,the study of road network traffic volume can reveal the law of exhaust gas emissions in the road network and strengthen the monitoring and control of road network exhaust gas.Therefore,it is of great practical significance to study the traffic volume application algorithm in the road network emission monitoring scenario.In this paper,we focus on two scenarios: traffic emission level classification at intersections and traffic volume prediction on road sections,and propose a new traffic volume emission level classification algorithm and traffic volume prediction algorithm by combining traditional machine learning and deep learning methods.The main research work includes the following two aspects:(1)Traffic volume emission level classification algorithm research.For the problems of large data volume of unlabeled samples at intersections and large cost of sample labeling,this paper proposes a traffic volume emission level classification algorithm KLDM-LGC based on active learning and semi-supervised learning,which evaluates the labeling value of each sample by calculating the KL scatter and variance between samples,finds samples with high labeling value,and builds a high-quality pool of labeled samples,and then utilizes the label propagation algorithm to mine the potential feature information of unlabeled samples.The experimental results show that the KLDM-LGC algorithm improves the classification accuracy by 24.57%,11.22% and 3.06% compared with the fully supervised algorithm,semi-supervised algorithm and active learning algorithm,respectively,which verifies that the algorithm has good classification performance in the traffic volume emission level classification scenario.(2)Traffic volume prediction algorithm research.To address the problem of time variance drift of traffic volume data in road sections and how to predict traffic volume more accurately,this paper proposes a traffic volume prediction algorithm Ada-TGCGRU based on adaptive time-domain distribution matching and spatio-temporal map convolution.the algorithm first maps the road network structure to the topological map,then divides the traffic volume data set into multiple segments using the timedomain distribution partitioning algorithm,and then uses the The temporal feature extraction module is used to learn the spatial features of the road network with the time-domain features of each segment of data,and finally the Boosting-based algorithm is used to match the distribution information learned in the network,to mine the common information among different distribution segments and to make predictions.The experimental results show that Ada-TGC-GRU has reduced 61.50%35.80%,33.50% and 36.30% in the four types of error indicators MSE,RMSE,MAE and MAPE,respectively,compared with the comparison algorithm,which verifies that the algorithm can fully exploit the spatio-temporal features in the road network and reduce the prediction error caused by the time variance drift,and has a high prediction performance in traffic prediction The algorithm has a high prediction performance in traffic prediction scenarios.