Application Research Of Change Point Detection Of Traffic Flow Based On FPOP

Excessive change points can be usually detected by using many traditional change point detection methods in the presence of outliers.In this paper,a class of dynamic pro-gramming change point detection algorithms based on functional pruning are concerned,which can solve the penalised cost optimization problem effectively.Further,the detec-tion of mean change point is studied from three aspects:loss function,penalty parameter and state constraint.The specific research contents are summarized as follows.For the problem of on-line detection of mean change point in the presence of outliers,the efficient functional pruning optimal partitioning(FPOP)algorithm and the robust functional pruning optimal partitioning(R-FPOP)algorithm are studied.A travel time prediction method for road section is proposed based on R-FPOP algorithm.The simu-lation results show that the R-FPOP algorithm with the biweight loss using three times the standard deviation of the noise has better detection performance under various noise distributions.The case analysis shows that the prediction intervals with an average cov-erage ratio of 83.30%are obtained by the proposed method,and the prediction effect is excellent.For the problem of adaptive selection of penalty parameters about FPOP and R-FPOP algorithms,the algorithm of adaptive selection of data-driven penalty parameters by combining CROPS algorithm with dimension jump algorithm(CROPS-D)is proposed for the piecewise constant model.The CROPS is a change point detection algorithm that can find the optimal segmentation for a range of penalties.The simulation results show that the mean change point can be detected more effectively by combining the proposed algorithm with the R-FPOP algorithm.The case analysis shows that the mean change point of section traffic flow can be better identified by the proposed algorithm,and it has certain practicability in the mining of road intersection traffic demand features.For the problem of peak change point detection with mean jump up and down in the presence of outliers,the generalized functional pruning optimal partitioning(GFPOP)algorithm and the search algorithm under the specified number of peak are studied.The robust generalized functional pruning optimal partitioning(R-GFPOP)algorithm is proposed by combining the biweight loss.The simulation results show that the proposed algorithm has better detection performance under the Student~?s t-distribution,no matter the number of peak is specified or not specified.The case analysis shows that it is useful to estimate the peak time of road section travel time when the number of peak is specified.