Search

CN-117116032-B - Information fusion optimization method based on fixed and mobile vehicle detection data

CN117116032BCN 117116032 BCN117116032 BCN 117116032BCN-117116032-B

Abstract

The invention discloses an information fusion optimization method based on fixed and mobile vehicle detection data, which comprises the steps of decomposing a transportation system into a plurality of input and output subsystems to obtain multi-source detection data of each subsystem, extracting traffic information from a historical transportation data set, establishing a combined optimization model, solving a joint matching probability model of the mobile detection data and the fixed detection data by dynamic programming, directly matching the upstream and downstream fixed detection data marked by the same piece of mobile detection data through data fusion, and solving the joint matching probability model of the upstream and downstream fixed detection data by a K-M algorithm to realize joint matching between the upstream and downstream unmarked fixed detection data. By using the method, the globally optimal matching result can be obtained in a very short time, and even if the mobile detection data has larger inaccuracy or the fixed detection data has a certain loss, the data matching accuracy is still maintained at a higher level, so that the microscopic traffic flow state can be reflected more accurately.

Inventors

  • SUN ZHANBO
  • HUANG ZHIHANG
  • HUANG TIANYU

Assignees

  • 成都格林希尔德交通科技有限公司

Dates

Publication Date
20260505
Application Date
20210427

Claims (4)

  1. 1. An information fusion optimization method based on fixed and mobile vehicle detection data is characterized by comprising the following steps: S1, decomposing a transportation system into a plurality of input and output subsystems to obtain multi-source detection data of each subsystem, wherein the multi-source detection data comprises fixed detection data and mobile detection data, the fixed detection data comprises time stamps, lane information and detected vehicle length information of fixed detectors on the upper and lower sides of a vehicle passing subsystem, and the mobile detection data comprises a vehicle ID, time stamps and vehicle position information recorded at fixed frequency and is used for verifying and calibrating the fixed detection data; s2, extracting traffic information from the historical traffic data set, wherein the traffic information comprises the probability of each possible lane selection or confluence selection, the corresponding travel time distribution probability under each possible lane selection or confluence selection and the error distribution probability of the vehicle detection vehicle length; s3, establishing a combined optimization model, wherein the combined optimization model comprises a combined matching probability model of mobile detection data and fixed detection data and a combined matching probability model of upstream and downstream fixed detection data; S4, solving a joint matching probability model of the mobile detection data and the fixed detection data by using dynamic programming to realize fusion of the upstream mobile detection data and the upstream fixed detection data and fusion of the downstream mobile detection data and the downstream fixed detection data; S5, solving a joint matching probability model of the upstream and downstream fixed detection data by using a K-M algorithm, and realizing joint matching between the upstream and downstream fixed detection data which are not marked by the mobile detection data so as to find a matching result with maximized probability between the upstream and downstream fixed detection data; The step S5 specifically includes: s5-1, converting a joint matching probability model of upstream and downstream fixed detection data into a linear mathematical form so as to solve by using a K-M algorithm; the joint matching probability model of the upstream and downstream fixed detection data is converted into a linear mathematical form as follows: Equation (13.1) is the joint matching probability of the upstream and downstream fixed detection data In the form of summation of (a) by Taking the logarithm to obtain, wherein As a probability function In logarithmic form, as shown in formula (13.2), lower case letters Representing downstream first Sequence number in the upstream fixed detection data set of the strip fixed detection data, formula (13.3) As decision variables, when downstream Strip fixed detection data and upstream first 1 When the bar-fixed detection data is matched, otherwise 0, formula (13.4) Representing downstream first Sequence number set of fixed detection data in upstream fixed detection data set, namely downstream The set of upstream stationary detection data matched with the strip stationary detection data, which is determined by the range of the historical travel time, this range is used in equation (13.4) Indicating that if downstream is Strip fixed detection data and upstream first The travel time between the fixed test data is outside this range, so that the probability of the two fixed test data being generated in the same vehicle is small and negligible, thus for the downstream first The strip-fixed detection data can only be matched with The method comprises the steps of collecting upstream fixed detection data in a set, matching the upstream fixed detection data in the set, wherein the formula (13.5) is used for ensuring that each piece of downstream fixed detection data is matched for the upstream fixed detection data only once, and the formula (13.6) is used for ensuring that the matching of the upstream fixed detection data and the downstream fixed detection data is carried out one by one; s5-2, solving a joint matching probability model of upstream and downstream fixed detection data by using a K-M algorithm, wherein the joint matching probability model comprises the following steps: s5-2-1, marking the fixed detection data set which is not marked by the movement detection data at the downstream as Aggregation, in which records are called Recording, marking the fixed detection data set which is not marked by the movement detection data at the upstream as Aggregation, in which records are called Record, will be Aggregation and collection Matching weights between collection records Conversion to a contiguous weight matrix According to the adjacency weight matrix For each of Recording and recording Recording labels, wherein The recorded reference numerals are denoted as , The recorded reference numerals are denoted as ; Equal to the first Strip Maximum matching weight of records, i.e. adjacency weight matrix First, the Maximum value of row All 0, the sum of these numbers being equal to Aggregation and collection The maximum matching weight possible between the collection records; S5-2-2, generating a set LES of label feasible sides based on the labels of all records in order to find the match with the maximum weight, wherein each side in the set LES of label feasible sides All satisfy the first Strip Record and first Strip The matching weights between records are equal to the sum of their labels, i.e This condition is used to ensure that if a match is made in the set LES of label-feasible edges, then the weight of this match is equal to the sum of all labels, i.e. the maximum matching weight, and thus for each in the set LES of label-feasible edges Record one by one Matching records; S5-2-3 in the set LES of the label feasible edges for matching Recording and recording The recorded edges are called active edges, the rest edges are called inactive edges, and a non-matched edge is selected based on the label feasible edge set LES Record from this Recording, namely alternately passing through active edges and inactive edges to form an alternate tree; s5-2-4, searching an alternative tree to start from the unmatched tree Recording ending with a non-matched piece The recorded path is called a successful path, and if at least one successful path exists in the alternate tree, the step S5-2-5 is switched to, and if no successful path exists in the alternate tree, the step S5-2-6 is switched to; S5-2-5, changing the inactive edge in the successful path into an active edge, and conversely changing the active edge in the successful path into an inactive edge to decompose the successful path to form a new matching state, and judging whether all the new matching states are in the new matching state The records are matched, if yes, the step S5-2-7 is carried out, and if not, the step S5-2-3 is carried out; S5-2-6, modification on alternate trees Recording and recording The recorded label is added with a new label feasible edge to expand the alternate tree, which comprises firstly finding the minimum cost for expanding the alternate tree, secondly, adding the alternate tree to the alternate tree The recorded labels minus the minimum cost will alternate between trees Finally, generating a set LES of feasible edges of the new label according to the new label, and returning to the step S5-2-3; S5-2-7, outputting the matching with the largest weight.
  2. 2. The method for optimizing information fusion based on stationary and mobile vehicle detection data according to claim 1, wherein the step S2 specifically comprises extracting the following four probability information from the historical traffic data set: s2-1, lane selection probability I.e. through downstream roadways Also pass through the upstream road Wherein, capitalized symbols And Representing random variables, lowercase symbols And A value representing a random variable; Through downstream roadways And an upstream roadway Is a historical data sample number; Through downstream roadways Is a historical data sample count of (1); s2-2, vehicle confluence probability By downstream flow direction Is upstream of Wherein, capitalized symbols And Representing random variables, lowercase symbols And A value representing a random variable; is through downstream flow direction And upstream flow direction Is a historical data sample number; is through downstream flow direction Is a historical data sample count of (1); the lane selection probability and the vehicle convergence probability have the same mathematical expression form, the former is applied to the input subsystem, the latter is applied to the output subsystem, and in the following description, the lane selection probability and the vehicle convergence probability are adopted It is indicated that, for the input subsystem, / Representing the lane, and for the output subsystem, / Representing a lane group; s2-3, probability of Stroke time distribution I.e. upstream lanes or groups of lanes To downstream lanes or lane groups Is the travel time of Wherein: , a time stamp indicating the vehicle passing the downstream stationary detector, A time stamp representing the vehicle passing the upstream stationary detector; representing sequential passage through upstream lanes or lane groups in the history data And downstream lanes or lane groups Is a vehicle travel time; for the number of historical data samples, each historical travel time is recorded as ; Representing bandwidth, for a given pair of upstream and downstream fixed detection data, can be at run time Finding out the corresponding probability density; S2-4, error distribution probability of vehicle length detection I.e. upstream stationary detector detects length recordings And a downstream stationary detector detecting a length record Probability of being generated for the same vehicle, wherein: The standard deviation can be determined according to historical data or expert in researching the detection precision of the vehicle length.
  3. 3. The method for optimizing information fusion based on stationary and mobile vehicle detection data according to claim 1, wherein the step S3 specifically includes: s3-1, establishing a joint matching probability model of mobile detection data and fixed detection data; By means of And The mathematical expressions of the joint matching probabilities between the downstream fixed detection data set and the upstream fixed detection data set and the mobile detection data set are shown as the formula (4) and the formula (5) respectively: In the formulas (4) and (5), the vehicles in the data set are detected for downstream movement The corresponding sequence numbers in the downstream fixed detection data set are defined as Wherein As a variable of the sequence number, For vehicles in the upstream movement detection dataset The corresponding sequence numbers in the upstream fixed detection data set are defined as Wherein As a variable of the sequence number, The movement detection data quantity of the upper and the lower stream is equal to the value of the sequence number variable , And For the same vehicle, consider that there may be overtaking in the road section or intersection, which may be different in location in the upstream and downstream movement detection data sets, so different subscripts are used And To represent the serial numbers of vehicles in the downstream and upstream movement detection data sets, for And Must be ensured , Thereby ensuring one-to-one matching between the mobile detection data and the fixed detection data; And Representing the discrete matching probabilities between the stationary detection record and the mobile detection record at the downstream and upstream positions, respectively, the two probabilities being calculated according to equations (6) and (7): Formula (6) represents the downstream first Strip movement detection record and downstream item The probability of matching a fixed detection record, which is the timestamp for two records And The probability density function of the differences follows a normal distribution, in which Representing the downstream movement detection dataset The vehicle passes the time stamp of the downstream stationary detection point, Representing the first of the downstream stationary detection data sets Time stamp of the bar record, formula (7) represents upstream first Strip movement detection record and upstream first The probability of matching a fixed detection record, which is the timestamp for two records And The probability density function of the differences follows a normal distribution, in which Representing the upstream movement detection dataset The vehicle passes the timestamp of the upstream fixed detection point, Representing the first of the upstream stationary detection dataset A timestamp of the bar record; s3-2, establishing a joint matching probability model of upstream and downstream fixed detection data; By means of The mathematical expression of the joint matching probability between the fixed detection data of which the downstream and the upstream are not marked by the movement detection data is shown as the formula (8): In formula (8), the random variable A first set of fixed detection data representing downstream non-moving detection data markers Serial numbers in fixed detection data sets of vehicles which are not marked upstream by movement detection data, wherein Representing the total number of samples of the fixed detection data set that are not marked downstream by the movement detection data, for any Then there is Ensuring that each downstream fixed detection data and each upstream fixed detection data are matched one to one; Representing downstream stationary detection data And upstream fixed detection data The matching probability between the two is expressed as the following formula (9): the (9) is the lane selection probability or the vehicle convergence probability Probability of travel time distribution Error distribution probability for detecting length of vehicle Is represented by the product of the three, Is a normalization item; S3-3, establishing an objective function and constraint conditions of the combined optimization model, wherein the objective function and constraint conditions are represented by formulas (10.1) to (10.5): Equation (10.1) is an objective function, the optimization result requires the maximization of the overall matching probability, equation (10.2) and equation (10.3) are constraint conditions, the relative sequence numbers of the two mobile detection records in the mobile detection data set and the fixed detection data set are ensured to be consistent, equation (10.4) is constraint conditions, the matching between the upstream fixed detection data and the downstream fixed detection data is ensured to be carried out only for the fixed detection data which are not marked by the mobile detection data, and equation (10.5) is constraint conditions, and the one-to-one matching between the fixed detection data is ensured.
  4. 4. The method for optimizing information fusion based on stationary and mobile vehicle detection data according to claim 1, wherein the step S4 specifically includes: s4-1, solving a joint matching probability model of upstream mobile detection data and upstream fixed detection data by using dynamic programming to realize fusion of the upstream mobile detection data and the upstream fixed detection data, wherein the upstream dynamic programming solving formula is as follows: equation (11.1) is an objective function in which Representing the upstream first Strip to the first The maximum probability product of a match between the strip movement detection data and the fixed detection data set, Indicating the upstream first Strip movement detection data matching upstream first The bars fix the probability of detecting the data, Representing the upstream first Strip to the first The maximum probability product of matching the strip movement detection data and the fixed detection data set, the formula (11.2) is the constraint condition, namely the boundary condition of dynamic programming, and the formula (11.3) is the constraint condition, wherein For decision variables, determine upstream first Strip movement detection data and upstream first Equation (11.4) is a constraint condition, and for the two upstream mobile detection records, the matching of the two upstream mobile detection records with the upstream fixed detection data set is ensured according to the condition that the relative sequence numbers are consistent; S4-2, solving a joint matching probability model of downstream mobile detection data and downstream fixed detection data by using dynamic programming to realize fusion of the downstream mobile detection data and the downstream fixed detection data, wherein a downstream dynamic programming solving formula is as follows: equation (12.1) is an objective function in which Representing downstream first Strip to the first The maximum probability product of a match between the strip movement detection data and the fixed detection data set, Indicating the downstream first Strip movement detection data matching to the first The bars fix the probability of detecting the data, Representing downstream first Strip to the first The maximum probability product of matching the strip movement detection data and the fixed detection data set, the formula (12.2) is the constraint condition, namely the boundary condition of dynamic programming, and the formula (12.3) is the constraint condition, wherein For decision variables, downstream first Strip movement detection data and downstream item And (3) matching the fixed data, wherein the formula (12.4) is a constraint condition, and the two downstream mobile detection records are matched with the downstream fixed detection data set according to the condition that the relative sequence numbers are consistent.

Description

Information fusion optimization method based on fixed and mobile vehicle detection data Technical Field The invention relates to the field of traffic engineering, in particular to a data and information fusion optimization method based on fixed and mobile vehicle detection data. Background The acquisition of traffic information of vehicles mainly depends on fixed detectors and mobile detectors. The fixed detector can collect large-scale vehicle data, but can only record the data at fixed points, the obtained data is discrete in space and has a certain error (such as that some vehicles pass the detector but cannot be detected), the mobile detector can continuously record more detailed information of the vehicles, but the whole traffic condition of a road section can not be reflected only for the vehicles provided with the equipment, and related data can relate to user privacy, so that the use of the mobile detector is limited. Thus, there is a certain deficiency in each of these two types of vehicle detector data. In recent years, by adopting a method of multi-source data fusion taking fixed detection data and mobile detection data into consideration, two types of detection data can be made complementary, thereby improving accuracy of acquired information. However, at present, a large number of multi-source data fusion methods acquire macroscopic and integrated traffic parameters, such as average journey time, average speed and the like, and only a small number of multi-source data fusion methods can acquire microscopic traffic states, such as individual journey time and the like. However, the multi-source data fusion method mostly adopts an exhaustion method or a heuristic algorithm (such as a tabu search algorithm) to fuse the multi-source data, and the information integration lacks consideration of the characteristic information of the vehicle, and the multi-source data fusion of the long road section and the short road section needs to be separately assumed whether a first-in first-out (FIFO) condition limit needs to be added. Firstly, the result can be globally optimal by fusing mobile data and fixed data through an exhaustion method, but the calculation time is too long, and the result can be locally optimal only through a heuristic algorithm although the calculation time is short. Secondly, because the characteristic information of the vehicle is not considered, the integrated traffic information is incomplete and cannot well correspond to the real traffic flow state, and therefore, the result error is increased by using the incomplete traffic integrated information. Thirdly, the first-in first-out condition is not considered for the long road section, the first-in first-out condition is considered for the short road section, the short road section is determined based on normal traffic flow, and the short road section is not applicable to abnormal traffic flow (such as frequent lane change caused by too many vehicles driving on the short road section), so that the first-in first-out condition is artificially set, the result error is increased, and the description of the traffic flow state is more distorted. The data and information fusion method based on the fixed and mobile vehicle detection data disclosed in CN107886192B still has some of the above-mentioned technical problems. Therefore, a data fusion and information integration method which can fuse mobile data and fixed data in a short time, consider the characteristic information of the vehicle, does not need to assume first-in first-out (FIFO) conditions, and can solve by analysis so that the result reaches global optimum is lacking at present. Disclosure of Invention The invention aims to provide an information fusion optimization method based on fixed and mobile vehicle detection data, which can fuse the mobile detection data and the fixed detection data in a short time, and can solve the problems by integrating the detection vehicle length information of a vehicle and using an analytic method capable of enabling the result to reach global optimum, so that the result precision of data fusion and information integration is greatly improved, and the microscopic traffic flow state can be more accurately described by using the method. The invention discloses an information fusion optimization method based on fixed and mobile vehicle detection data, which comprises the following steps: S1, decomposing a traffic transportation system (road network) into a plurality of input (road section) and output (intersection) subsystems to obtain multi-source detection data of each subsystem, wherein the multi-source detection data comprises fixed detection data and mobile detection data, the fixed detection data comprises time stamps of vehicles passing through upstream and downstream fixed detectors of the subsystems, lane information and detection vehicle length information, and the mobile detection data comprises vehicle IDs, time stamps recorded at fixed