Intelligent Identification And Improvement Of Traffic Congestion Based On Multi-source Data Mining

The identification and improvement of traffic congestion has important theoretical significance and engineering application value in the field of transportation.The in-depth research on key issues such as low accuracy and inability to intelligently identify the current methods is proposed.A method for intelligently identifying and improving traffic congestion based on multi-source data mining is proposed.The main research work is as follows:(1)For the short-term prediction method of traffic flow due to incomplete consideration of traffic flow characteristics,the prediction accuracy is low.A short-term prediction method of urban road traffic flow based on the spatio-temporal node wrapped feature selection and BPNN is proposed.The space-time traffic flow corresponding to the road traffic flow spatial nodes in the study area is used as the input,the future traffic flow of the road space nodes to be predicted is used as the output,and the historical data is used as the training set to convert the problem into data-driven Multi-input single-output regression prediction problem.First,the mechanism and data correlation analysis is performed on the characteristics of the traffic flow to obtain the spatiotemporal characteristics of the traffic flow.Second,the candidate spatiotemporal node set is determined according to the reachable range of the traffic flow,and the fitness is calculated using the inverse of the sum of squared errors as the objective function.We use the wrapped feature selection method,genetic algorithm,and back propagation neural network(BPNN)to solve the spatiotemporal node selection to obtain the final spatiotemporal node and the trained BPNN.Finally,we will select the spatiotemporal nodes on the working set.The actual measured value is input into the trained BPNN,and the short-term prediction value of urban road traffic flow is obtained.This method more rationally selects and uses the spatiotemporal traffic flow data,and the accuracy will be higher.On Tianan North Road,Quanzhou,based on actual measurement results on a total of 5 working days in a week in August 2019,the results show that the accuracy of the method in this paper is higher than that of using only time-space node data,using other time node ranges,and support vector machines and gradient boosting tree methods.(2)The causes of urban road traffic congestion include problems such as dynamic real-time,complex and changeable,subjective identification methods,poor real-time,and unable to identify automatically.This paper proposes a multi-cause automatic identification of urban road traffic congestion based on causal Bayesian network.Observable variables related to urban road traffic conditions within the study area are used as inputs,and 0-1 discrete types of traffic congestion causes are used as outputs.Historical data is used as a training set to transform the problem into a data-driven multi-classification problem.First,the method performs systematic mechanism analysis and simulation verification on the relationship between the dynamic observable variables of urban road traffic and multiple congestion causes,thereby constructing a causal Bayesian network structure.Secondly,the obtained measured historical data is used for parameter learning and training to obtain a complete causal Bayesian network model.Finally,by inputting the observable variables of traffic under road working conditions into the Bayesian network model,multiple causes of traffic congestion can be automatically and simultaneously identified in real time.This method has high flexibility,can better express the correlation of nodes,has strong interpretability,can make full use of expert experience and knowledge,and can achieve automatic real time.The research results on Quanxiu Street in Quanzhou City show that the construction of a multi-cause automatic real-time identification of urban road traffic congestion based on causal Bayesian network is reasonable.Pedestrian influence,peak traffic,parking occupied road,unreasonable signal timing,and impact of traffic crossing the road,the recognition accuracy of the five congestion causes is better than that of the comparison method experiments.(3)Aiming at the current frequent traffic jams at intersections,and the accuracy and flexibility of signal control methods are poor,a real-time logic open-loop control method of traffic lights based on the measured vehicle flow and combined control strategy is proposed.The problem is transformed into a time-varying optimization problem.An open-loop control system is designed to solve this problem.The green light duration and signal state are used as controllable variables to improve the traffic efficiency at the intersection.This method first considers three control strategies: full service control,threshold limit control,and priority control based on actual traffic problems.A real-time traffic signal logic control model is constructed by combining the strategies.Secondly,simulation methods are used to verify the correctness of the model and the effectiveness of the control method.The method can deal with emergencies in a timely manner under the conditions met,improve the traffic efficiency at the intersection,and realize the refined and intelligent control of the intersection signal.At the Tianan intersection in Fengze,Quanzhou,the results of experiments on September 5,2019 show that the proposed method is more effective than the actual fixed timing signal control scheme and adaptive green light delay algorithm is good.In addition,there are places in this article where we can continue our research.For example,the number of hidden layers and nodes of BPNN in traffic flow prediction can be tested more times to find the optimal structure.The accuracy of Bayesian network methods to identify multiple causes of traffic congestion can be more multi-classification model indicators are evaluated;the proposed intersection control method can be considered for practical implementation to obtain measured data,thereby further verifying the accuracy and effect of the method.