Research On Adaptive Traffic State Robust Recognition Algorithm And Its Application

At present,most of the identification methods of near-stationary traffic state use visual inspection,which is not only time-consuming and laborious,but also not suitable for traffic state identification of massive traffic flow data,so there is a lack of effective automatic identification methods.In the process of near-stationary traffic states identification,the key is how to segment data accurately and effectively.Dynamic programming change point detection algorithm can effectively solve this problem.In this paper,the robust identification algorithms of near-stationary traffic state under the piecewise mean model and piecewise linear model are considered,respectively.The specific research contents are as follows:Aiming at the problem of near-stationary traffic state identification and adaptive selection of penalty parameters under the piecewise mean model,the algorithms of efficient and robust function pruning optimal partitioning(RFPOP)and the data-driven change points for a range of penalties(CROPS)are studied.Combined with the modified Cassidy criterion,a robust near-stationary traffic state automatic identification(RNSAI-1)algorithm is proposed.The simulation results show that under different noise distribution,the RFPOP algorithm after wavelet transform has more stable performance and better detection effect in the change point detection,especially when the noise follows heavy tail distribution,it is basically not affected by the noise interference in the data.The example analysis shows that the RNSAI-1 algorithm can well identify the nearstationary traffic state from the traffic flow data.The triangular fundamental diagram calibrated based on the identification results fits well and has physical significance.It has a certain application prospect in the estimation of traffic capacity of urban road network.Aiming at the problem of near-stationary traffic state identification under piecewise linear model,the continuous-piecewise linear pruned optimal partitioning(CPOP)algorithm and pruning conditions are studied.Using the characteristic that the bounded Biweight loss function is insensitive to outliers,a robust continuous-piecewise linear pruned optimal partitioning(RCPOP)algorithm is proposed.Further,combined with the augmented Dickey-Fuller(ADF)test in time series analysis,a novel and robust nearstationary traffic state automatic identification(RNSAI-2)algorithm is proposed for the near-stationary traffic states identification of traffic flow time series.For RCPOP algorithm,under certain conditions,the consistency of estimating the location and number of change points is given.For the corresponding variance estimation problem,it is found that the median absolute deviation estimator using differential time series has the best estimation effect.The simulation results show that RCPOP algorithm is more stable and better than CPOP algorithm in the detection performance of the location and number of change points.The example analysis shows that the RNSAI-2 algorithm can well identify the near-stationary traffic state of the cross-sectional traffic flow time series,and the identified near-stationary traffic state is effective.The proposed algorithm has a certain reference value for exploring the calibration and estimation of the macroscopic fundamental diagram in the urban road network under the near-stationary traffic state.