CN-121980373-A - Bayonet travel time estimation method based on mixed integer optimization and space KNN

Abstract

The invention relates to the field of traffic control systems, in particular to a bayonet travel time estimation method based on mixed integer optimization and space KNN, which comprises the following steps of cleaning mapping bayonet data, measuring and calculating initial space-time parameters, and eliminating anomalies based on absolute medium bit differences; dividing a network according to a flow threshold, estimating a high-flow road section by adopting mixed integer optimization check sum Bayesian dynamic fusion, carrying out parameter deduction based on space KNN topological constraint and reverse solution of congestion characteristics on a low-flow road section, constructing a global parameter vector by fusing a double-source road section result, circularly optimizing by an iterative convergence mechanism, and outputting a high-precision global travel time field. The method utilizes the absolute intermediate level difference to construct a robust filtering mechanism to remove extreme anomalies, performs parameter deduction aiming at a low-flow road section, realizes high-precision estimation on the basis of ensuring physical interpretability, and accurately restores the dynamic running state of the road network.

Inventors

• GUAN DEYONG
• Zheng Lezhi
• WANG KE
• REN ZHIWEI
• ZHANG YUTING
• ZHU RUI
• ZHANG KE

Assignees

• 山东科技大学

Dates

Publication Date

20260505

Application Date

20260408

Claims (6)

1. 1. The bayonet travel time estimation method based on mixed integer optimization and space KNN is characterized by comprising the following steps of: S1, collecting card interface passing sequence data, carrying out cleaning mapping, measuring and calculating the driving mileage and the initial travel time, and constructing a stable filtering standard based on an absolute intermediate level difference to remove extreme abnormal values; S2, decomposing the travel time and stripping off the abnormal residence; s3, dynamically dividing the road network into a high-flow road section and a low-flow road section according to a flow threshold value, wherein the high-flow road section adopts a two-stage collaborative estimation strategy of mixed integer optimization check and Bayesian dynamic fusion, and the low-flow road section implements a parameter deduction strategy based on space KNN topological constraint and reverse solution of congestion characteristics; And S4, fusing the two-path segment results to generate a global parameter vector, and circularly optimizing or outputting a final high-precision travel time field according to the iterative convergence criterion.
2. 2. The bayonet travel time estimation method based on mixed integer optimization and space KNN according to claim 1, wherein the travel distance in S1 is calculated through Dijkstra shortest path algorithm, and the initial travel time is calculated through a timestamp difference value of a vehicle passing through an upstream bayonet and a downstream bayonet.
3. 3. The bayonet travel time estimation method based on mixed integer optimization and space KNN according to claim 1, wherein the robust filtering criteria in S1 is implemented by setting a dynamic threshold in the range of median ± 3 x 1.4826 x MAD, wherein MAD is an absolute median difference.
4. 4. The bayonet travel time estimation method based on mixed integer optimization and space KNN according to claim 1, wherein the travel time in S2 is allocated according to a weighted proportion of the road section prior travel time and the traffic frequency, and the calculation formula is as follows: , Wherein, the The allocation time of the trajectory i on the road segment j is represented, Indicating the number of times the trajectory i passes the road segment j, The prior travel time mean value of the road section j at the kth iteration is given, and M is the total number of road network sections.
5. 5. The bayonet travel time estimation method based on mixed integer optimization and space KNN according to claim 1, wherein the stripping off the abnormal resides in S2 comprises the steps of: (1) Calculating prior time mean and variance of each road section, and obtaining the whole path by aggregation based on the road section sequence passed by the track i Theoretical travel time distribution parameters of (1) including path expected mean value Sum path standard deviation ; (2) Construction based on Dynamic threshold of criteria, generation of binary indicator variable The decision rule is as follows: , Wherein, the Representing a binary indicator variable which is indicative of the value of the variable, The observed total travel time of the long distance trajectory i is represented, And Respectively represent paths traversed by the track i in the kth iteration The travel time of (1) desired mean and standard deviation, if Judging that said track is normal running, if It is determined that the trace contains an outlier.
6. 6. The bayonet travel time estimation method based on mixed integer optimization and space KNN according to claim 1, wherein the parameter deduction strategy in S3 comprises the following steps: (1) Applying double strict constraint of 'nearest space distance' and 'same road grade' to a road network by using a KNN algorithm, searching and constructing a neighbor high-flow source road section set, and marking the road section set as S; (2) For each source road segment S in the set S, calculating a congestion coefficient based on its corrected traffic data The calculation formula is as follows: , Wherein S represents an index of a source road section and belongs to a high-flow set S; the speed of the source road segment s; The legal speed limit value of the source road section s; Is the physical length of the source road segment s; The travel time average value of the source road section s after the k+1st iteration update is obtained; (3) Constructing a weight matrix based on the space distance between the source road section and the target blind area, wherein the calculation formula of the weight matrix is as follows: , Wherein, the To normalize the spatial weights, j represents the index of the target road segment, Representing a spatial distance between the target segment j and the source segment s; (4) Combining the congestion coefficient after weighted averaging with the speed limiting attribute of the target road section, and reversely deducing the travel time distribution parameter of the dead zone road section : , Wherein, the As the travel time average value of the target road section, For the physical length of the target blind zone segment j, Is the legal speed limit value of the target blind zone section j, Is the target road section congestion coefficient obtained by weighted summation.

Description

Bayonet travel time estimation method based on mixed integer optimization and space KNN Technical Field The invention relates to the field of traffic control systems, in particular to a bayonet travel time estimation method based on mixed integer optimization and space KNN. Background Road section travel time and passing speed are core indexes for representing the running state and service level of the urban road network. In recent years, road network travel time deduction based on truck passing sequence data has become a mainstream technical means. However, due to the spatial sparse layout of the bayonet equipment, adjacent snapshot points often span multiple road sections, abnormal delay noise caused by non-driving behaviors such as vehicle midway residence and detouring is easily mixed in a long-distance track, and meanwhile, a dead zone road section in a low flow area is difficult to directly obtain an effective estimated value due to lack of observation samples. Aiming at the difficult problems of data quality and space coverage, the prior art mainly adopts two typical schemes to estimate and deduce the travel time. The first category of solutions focuses on purely data driven and driver crowd characterization. The method comprises the steps of firstly constructing coarse-granularity path travel time based on the snapshot time difference of the same vehicle at adjacent bayonets, then constructing a tensor model by fusing multidimensional features such as vehicle identifications, travel time periods, travel time durations, frequency and the like, and describing behavior portraits of driving groups by using a clustering algorithm. Aiming at the missing or abnormal items of the travel time, the technology relies on a strict history sample matching mechanism, namely, only a history observation value with completely consistent clustering category, travel time period and road section attribute is selected, and the reconstruction and estimation of missing data are carried out by adopting segment spline interpolation. The second scheme focuses on solving the problem of track residence identification and blind area completion under the sparse layout of equipment by utilizing a combination optimization theory. The method comprises the steps of firstly constructing a mixed integer programming model, introducing binary indicating variables to judge non-driving residence behavior in a vehicle track, and deducing the total path duration by combining the prior distribution of road section travel time, so that the abnormal residence track is effectively removed. Based on data cleaning, the method adopts an alternate iterative algorithm based on the fixed point theory to reasonably decompose the total travel time of the purified path to each road section along the way. Aiming at the problem of lack of samples of a low-flow road section, the technology assumes that the average value and the standard deviation of a high-flow road section have a linear proportional relationship, utilizes a least square method to fit the rule, and directly extrapolates the distribution parameters of the low-flow road section through a linear equation so as to realize the estimation of the travel time of the whole road network. On one hand, partial technology directly adopts adjacent bayonet time subtraction method when calculating the travel time, can not effectively identify and strip abnormal delay generated by parking, working or bypassing and other 'stay' behaviors of vehicles in the middle of the vehicle mechanically, causes a large amount of long tail noise mixed in a basic sample, and seriously influences the accuracy of estimating the speed of the road section of the vehicle. On the other hand, for dead zone sections with low flow or no detection data, the prior art mostly adopts spline interpolation method based on the history data of the same road section or linear regression calculation method based on the distribution parameter fitting of the high flow road section. The former can be completely disabled when the road section is not used for a long time; the method is used as a purely mathematical numerical fitting means, so that the real topological structure of the urban road network and the spatial physical conduction characteristic of traffic jam are severely split, the strong spatial traffic correlation between adjacent or equal road segments is ignored, the dead zone travel time of the deduced performance lacks practical physical significance, and the real dynamic running state of the road network cannot be accurately reflected. Disclosure of Invention In order to solve the defects of the existing bayonet travel time estimation technology, the invention provides a collaborative estimation method based on mixed integer optimization and space KNN. The method not only effectively solves the problem of distortion of speed estimation caused by non-elimination of abnormal resident noise in the road section of the vehicle, but al