Study On Urban Traffic Status Prediction Based On Data Drive

With the acceleration of urbanization,road traffic flow is affected by more and more complex factors,and changes randomly in time and space.The wide application of intelligent sensor devices provides a large number of traffic data with diverse sources and rich types for dynamic perception of traffic flow changes in road network.Comprehensive utilization of these massive traffic data can provide more comprehensive information for analyzing and grasping the change trend of traffic state.In the face of the massive traffic data with the characteristics of nonlinearity,randomness,temporal and spatial gradientness,the traditional data analysis methods based on expert experience or accurate mathematical model have the disadvantages of too many conditions and fixed parameters,which are difficult to adapt to the complex characteristics of massive data and meet the refined demand of traffic state trend prediction.Therefore,combining with the advantages of data driven idea,its strong adaptability and learning ability,the paper describes and forecasts the dynamic change process of road traffic flow by using relevant data driven technology,which is conducive to fully mining abundant traffic information and improving the accuracy of urban traffic state prediction.Firstly,according to the demand of basic data for road traffic state prediction,the paper introduces the traffic data acquisition technology and its advantages and disadvantages,analyzes the basic characteristics of dynamic traffic data,introduces the calculation method of common representation indexes according to the principle of selecting traffic state representation,expounds the application advantages of data driving theory in traffic state prediction,and introduces the relevant data driving technology The technique and its advantages and disadvantages.Secondly,aiming at the problem of fixed detection data missing,error and so on,based on the combination of threshold method and traffic mechanism method to identify the fault data,combined with the strong adaptability of k-nearest neighbor algorithm,a fault data repair model based on improved KNN algorithm is proposed.According to the multi-source characteristics of traffic flow parameters and the selflearning ability of neural network,a traffic data fusion model based on GAPSO-WNN is proposed.Example analysis shows that the data-driven repair and fusion model can effectively improve the reliability of state index dataThen,according to the fuzziness of traffic state description and the similarity of state representation indexes,combined with the objectivity of fuzzy clustering analysis,a traffic state classification model based on GA-WFCM is proposed,which divides the traffic state into five different level intervals;based on the selected multiple traffic state representation indexes,the traffic state is classified by using the intuitionistic fuzzy entropy theory starting from the data itself The comprehensive evaluation scheme of traffic state classification based on data-driven is established,which is used as the evaluation basis of traffic state characterization index prediction value.Finally,according to the physical static structure of road network and the dynamic characteristics of traffic flow,the spatial adjacency matrix of road network is established.Combined with the correlation theory,the spatial-temporal correlation measurement function of road network traffic flow is established.On this basis,the reasonable spatial-temporal series factors of traffic state prediction are selected as the input of the prediction model.Combined with the long-term and short-term memory recurrent neural network with the advantages of "memory" and "storage" A traffic state index prediction model based on spatiotemporal correlation and LSTM RNN is proposed.Through the example analysis,the feasibility and stability of the data-driven traffic state classification evaluation scheme and prediction model are verified.