Data Driven Control For Urban Traffic Network

Nowdays,traffic congestion happens frequently in urban area,seriously affecting urban economic development and reducing residents’ life experience.With the growing number of vehicles,traffic congestion has spread from first-tier cities to second-tier and third-tier cities.The carrying capacity of urban traffic facilities has been greatly tested.As a unique method of relieving congestion,traffic signal control improves the overall traffic efficiency of urban road network only by coordinating the signal lights of each intersection in the traffic network,which has a high cost-effective ratio.In recent years,with the development and popularization of traffic acquisition technology,traffic data provided by traffic equipment has increased explosively.Due to big data technology and artificial intelligence,the traditional transportation science has ushered in a new turning point.Traffic engineering has changed from the integration of mathematics to the integration of data.Traffic control uses the rules of traffic flow evolution in data mining and generates control rules by learning methods are the main research directions at present.Based on the above background,this paper proposes a data-driven urban traffic network control strategy.Using big data and artificial intelligence technology to model traffic flow,mining the hidden relationship between different data,and replacing traditional mechanism model with data model,makes the model closer to reality,based on which traffic flow control is carried out.The main work of this paper is as follows:1)An algorithm for estimating the degree of road congestion in the future time period by using traffic flow is proposed.Using a large number of historical data and hidden Markov model,the state transition model of traffic flow and congestion degree is established.To solve the problem of insufficient hidden state historical data,a training method combining supervised learning with unsupervised learning is proposed.At the same time,according to the Viterbi algorithm,an algorithm for estimating future hidden state is proposed,and the accuracy of the model is verified in different traffic scenarios.2)A data-driven distributed control algorithm is proposed to calculate the timing allocation of each intersection by using the traffic flow of each lane.The Hidden Markov estimation model is improved by incorporating the information of control dimension,and the distributed optimization problem of state transition is constructed by reducing traffic congestion degree as an index to coordinate signal timing between itself and adjacent intersections.In addition,an algorithm is proposed to solve the optimization problem,and the control effect is verified under different traffic demands.3)A road network control method based on historical decision learning is proposed.Hidden Markov model is used as data model to mine the relationship between traffic flow and signal timing.Different evaluation indicators are matched with different data,and the training model is used to construct the state transition model between traffic flow and signal using historical data instead of rolling optimization of traffic control method.According to the input characteristics of the traffic state generation model,the output is the corresponding signal lamp timing.On this basis,feedback and model updating mechanism are integrated to self-tune the output of the data model,which makes the timing parameters obtained only by data matching more suitable to the current state.In this process,new data are constantly screened for model training to increase the adaptability of the data model and realize the selflearning process of the data model.4)Based on LSTM neural network,the flow timing transfer model of the upper layer of regional control is constructed.In order to alleviate large-scale congestion,a congestion diffusion model of the whole road network is established,which is based on the congestion state of hidden Markov model,and the output of LSTM is corrected in real time.In order to enhance the adaptability of the model,the self-learning strategy of the regional model is designed.