Research On Regional Traffic Flow Prediction Based On Deep Learning And Multi-Objective Joint Optimization Of Single Intersection Signals

With the automobile industry thriving,the gradual increase in car ownership per capita has led to a more severe contradiction between road supply and scale of vehicles.To solve this problem,a formula to estimate average delay and the capacity of vehicles is proposed,based on the traffic prediction using Tensor Flow design structure,which has quality of deep learning.Aiming at the timing plan of the SCATS system,this thesis proposes the calculation formula of the objective function of the average delay and capacity of vehicles.The main work is as follows:(1)At the first,this thesis concludes the current achievements of the predicting traffic situation and signal timing for intersections,exemplifies some applications based on deep learning in traffic controlling,and introduces the theory of signal controlling.(2)Secondly,realize a regional traffic prediction system based on CNN,RNN/LSTM and Attention mechanism.In the experiment,RNN structure is chosen to extract and model the timing and spatial features,combining with LSTM and CNN;the Attention mechanism is used to analyze the components of the module The factor performs the differential distribution of the attention component,and realizes the uneven distribution of the weights of different CNN and RNN modules to meet the adaptive matching of the flow weights of the traffic blocks at different times and in different regions.(3)Referencing the timing formula from classic traffic managing system,SCATS,this thesis proposes a new signal timing model using muti-object for evaluation,which based on two indexes of average delay and traffic capacity.To generalize the value,it is chosen as the output that the margin between the two indexes multiplying their relevant weight.For optimizing the parameters,this thesis adopts three meta-heuristic algorithms,GA,SA and PSO,to optimizes the objective function.The experimental results prove that the three algorithms above all shows great quality of exploring optimal value.(4)Finally,Regional traffic flow forecast uses NYC-Taxi data set,the data is divided into 10×20 grids in latitude and longitude,a certain intersection is selected in the area as the research object,and the actual traffic data within one hour of the evening rush hour is substituted into three optimization algorithms.Use Spyder software to obtain the result of the objective function value after algorithm optimization,evaluate and analyze the three sets of data,and draw relevant conclusions.