Complex Network Based Charateristics Analysis And Hybrid Predictionof Of Traffic Flow

Currenly,traffic congestion,traffic accidents,environmental pollution have become a main obstacle to restrict the rapid development of the city.These pr oblems seriously affect the life of urban residents and the attractiveness of the city.Considering increasingly serious traffic problems,scholars and traffic engineers obtain abundant results in following three aspects: traffic data quality control,dynamic analysis of traffic flow characteristics and traffic flow short-term prediction.On the basis of previous research,data quality control is conducted to guarantee the rationality of analysis.Meanwhile,through in-depth analysis on traffic flow dynamic characteristics,the charateristiscs including periodicity,fluctuation and randomness are extracted.Accordingly,a hybrid traffic flow prediction model considering weekly similarity is established.This study contains following several main research achievements.(1)Proposed an approach to integrate Fuzzy C-means based imputation method for missing traffic volume data estimationAlthough various innovative traffic sensing technologies have been widely employed,incomplete sensor data is one of the most major problems to significantly degrade traffic data quality and integrity.In this study,a hybrid approach integrating the Fuzzy C-means(FCM)-based imputation method with the Genetic Algorithm(GA)is develop for missing traffic volume data estimation based on inductance loop detector outputs.By utilizing the weekly similarity among data,the conventional vector-based data structure is firstly transformed into the matrix-based data pattern.Then,the GA is applied to optimize the membership functions and centroids in the FCM model.The experimental tests are conducted to verify the effectiveness of the proposed approach.The traffic volume data collected at different temporal scales were used as the testing dataset,and three different indicators,incl uding root mean square error,correlation coefficient,and relative accuracy,are utilized to quantify the imputation performance compared with some conventional methods(Historical method,Double Exponential Smoothing,and Autoregressive Integrated Moving Average model).The results show the proposed approach outperforms the conventional methods under prevailing traffic conditions.(2)Characterizing traffic time series based on complex network theoryFirstly,the phase space,which describes the evolution of the behavior of a nonlinear system,is reconstructed using the delay embedding theorem.Secondly,in order to convert the new time series into complex network,the critical threshold is estimated by the characteristics of complex network,which include degree distribute,cumulative degree distribute,density and clustering coefficient.We find that degree distribution of associated complex network can be fitted with a Gaussian function,and cumulative degree distribute can be fitted with an exponential function.Density and clustering coefficient are then researched to reflect the change of connections between nodes in complex network,and the results accord with the observation to the plot of adjacent matrix.Consequently,based on complex network analysis,the proper range of the critical threshold is determined.(3)Dynamic analysis of traffic time series at different temporal scales based on complex networks theoryIn this study,we first measure the complexity of traffic flow data by Lempel-Ziv algorithm at different temporal scales,and the data are collected from loop detectors on freeway.Second,to obtain more insight into the complexity and periodicity in traffic time series,we then construct complex networks from traffic time series by considering each day as a cycle and each cycle as a single node.The optimal threshold value of complex networks is estimated by the distribution of density and its derivative.In addition,the complex networks are subsequently analyzed in terms of some statistical properties,such as average path length,clustering coefficient,density,average degree and betweenness.Finally,take 2min aggregation data as example,we use the correlation coefficient matrix,adjacent matrix and closeness to exploit the periodicity of weekdays and weekends in traffic flow data.(4)Exploring dynamic property of traffic flow time series in multi-states based on complex networksWe use the data collected from loop detectors on freeway to establish traffic flow model and classify the flow into three states based on K-means method.We then introduced two widely used methods to convert time series into networks: phase space reconstruction and visibility graph.Furthermore,in phase space reconstruction,we discuss how to determine delay time constant and embedding dimension and how to select optimal critical threshold in term of cumulative degree distribution.In the visibility graph,we design a method to construct network from multi-variables time series based on logical OR.Finally,we study and compare the statistic features of the networks converted from original traffic time series in three states based on phase space and visibility by using the degree distribution,network structure,correlation of the cluster coefficient to betweenn ess and degree-degree correlation.(5)Proposed a hybrid predicting approach based on double exponential smoothing and support vector machineTraffic flow prediction is considered as a key technology of ITS(Intelligent Transportation System).First,in the hybrid model,DES is applied to predict the future data and its smoothing parameters are determined by Levenberg-Marquardt algorithm.Second,support vector machine(SVM)is employed to estimate the residual series between the prediction results by DES model and actual measured data.In SVM model,we use crossing correlation rule to optimize its parameters.Finally,a case study is performed to test the proposed model by using the data at different temporal scales.Furthermore,data smoothing strategies i ncluding difference and ratio scheme based on weekly similarity are applied as data preprocess before prediction.The proposed hybrid model along with the preprocessing scheme demonstrates superiority in prediction accuracy comparing with autoregressive integrated moving average(ARIMA),DES and DES-SVM models.