Research On Deduction Of Signalized Intersections Operation State And Timing Optimization Method Based On Floating Car Data

Signalized intersections play an important role in the road network.Urban road intersection bottleneck has become an important factor in causing traffic delay and restricting traffic efficiency.It is essential to formulate control strategies to alleviate intersection delay.However,because of the sophisticated intersection traffic condition,it is difficult to identify the operation performance,diagnose the causes of time delay,predict the traffic evolution pattern,and optimize signal timing scheme at intersections by the traditional data and prediction methods.The development of big data technology and the deep learning model provides us a good chance to address this challenge.Among these technologies,the floating car data have gradually become popular due to its wide coverage,low cost,large amount,and informative Spatio-temporal features.Floating car data can detect the operation status of signalized intersections in real-time,and it is considered as a "Stethoscope" to the road network.However,further studies remain to be done on how to better use floating car data at intersections.In order to understand,sort out and dig out the operation rules of floating cars at intersections,the floating car data are matched to the grid model in this dissertation to extract the Spatio-temporal features and evolution mechanism effectively.The accurate evaluation of the operation state and the automatic diagnosis of delay causes is the premise for prediction at intersections.Therefore,this dissertation proposes a multi-task fusion deep learning(MFDL)model based on massive floating car data to effectively predict the passing time and speed at intersections,which provides the technical basis for the adaptive timing optimization scheme.The signal timing optimization scheme includes two parts: the construction of a deep reinforcement learning model and the extraction of the traffic status of floating car data,in order to improve the operation efficiency of intersections.The main contents and findings of this dissertation are divided into four parts:(1)Based on the massive floating car data,the basic attribute features,Spatiotemporal features,environmental characteristics,and evolution rules of the floating car data are deeply explored.The traffic characteristics are extracted by developing the intersection grid model to match the floating car data.Moreover,the fuzzy C-means(FCM)clustering method is developed to identify the intersection area and accurately depict the influence areas of different intersections and directions.At the same time,the direction of the trajectory data of the floating cars is identified by the grid model.The results show that the Spatio-temporal features of floating car data can effectively support the research of the operation state at intersections.(2)Based on the grid model,the characteristics of floating car data were extracted,and the traffic parameters and signal timing parameters were estimated to construct the operation state evaluation system and delay diagnosis index of signalized intersections,to realize the operation state perception and automatic identification of delay problems at signalized intersections.The case study finds that the operation evaluation method constructed in this dissertation can effectively evaluate the overall and internal service level of signalized intersections.The method of signal timing parameter estimation can accurately estimate the signal timing parameters of the fixed time scheme,and the delay diagnosis index based on this method can effectively identify the cause of the delay.The results and findings can help to construct the prediction method and signal timing optimization strategy of intersections.(3)The Multi-task fusion deep learning(MFDL)model is constructed to predict the travel time and speed of the large range of signalized intersections at the same time.The model fully considers the Spatio-temporal feature,topology feature,and weather characteristics of traffic parameters,and uses a residual neural network to improve the depth of the model,release the potential of model prediction,and realize the automatic distribution of variable weight by using attention mechanism to show the advantages of feature fusion.The case study shows that the model has high accuracy compared with the basic models.It is found that the model has strong robustness by changing the variable group of the model.Compared with the single-task model,it is found that the model can take the advantage of information sharing among variables,and has the advantages of higher prediction accuracy and shorter training time.The associated results and techniques can effectively support the research on timing optimization of deep reinforcement learning.(4)The deep adaptive control scheme of signal intersection is realized by constructing 3DQN-PSTER(Double Dueling DQN Priority Sum Tree Experience Replay).The model combines Double DQN,Dueling DQN technology,and Priority Sum Tree Experience Replay strategy to improve the performance of the model.We evaluate our model via simulation in the Simulation of Urban Mobility(SUMO)in a vehicular network based on the overall data,floating car data,and induction loop data.The case study shows that 3DQN-PSRER has fast convergence,strong model stability,and high accuracy of evaluation results.In different traffic volumes and different entrance road scenarios,compared with different timing optimization schemes,it is found that the reinforcement learning timing scheme(RLSC)based on 3DQN-PSRER has unique advantages in the real-time,unbalanced dynamic traffic flow environment.In the traffic environment with floating car data,the higher the permeability of floating car data,the more stable the learning process of the model,the better the performance effect of the model.This dissertation adheres to the research concept of discovery delay,diagnosing delay,forecasting situation,and alleviating delay.The paper uses floating car data to identify the operation situation,diagnose the cause of delay,predict the operation status and optimize the signal timing scheme.The traffic big data processing technology,deep learning method,and reinforcement learning method are used to deal with the complex traffic states of intersections.This study has great theoretical and practical significance to reduce the delay of signalized intersections and improve the operation efficiency of the whole road network.