Highway ETC missing transaction data repairing method based on multi-scene features

Abstract

The application discloses a highway ETC missing transaction data repair method based on multi-scene characteristics, and establishes a trajectory data set through multi-source data fusion; and extracts an abnormal trajectory data set after abnormality detection on the trajectory data. Before data repair, the data in the service area section of the trajectory data needs to be marked according to an ETC topology data set; and a weighted FCM algorithm is cited to detect and mark whether the missing transaction gate section is in a congestion period. According to the detection result and algorithm requirement, relevant features are extracted to establish a repair data set. The repair problem of the path is converted into an interpolation problem of the missing transaction gate number. On the basis of the ETC missing transaction gate path repair method based on a topology constraint, the speed characteristics of the expressway traffic under different scenes are deeply analyzed, and a targeted time repair strategy is carried out. The average repair error of the method is within 18s, and the execution time is only 0.07s.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of highway management, and particularly relates to a highway ETC missing transaction data repair method based on multi-scene features. BACKGROUND

[0002] Due to the influence of ETC gantry on the adverse environment, there may be device failure and other problems. At the same time, the ETC vehicle-mounted device may appear wireless crosstalk, near large vehicle shielding and other phenomena when passing through the gantry. These problems and phenomena will cause the failure of information interaction between devices, and further produce abnormal data. These abnormal transaction data not only affect the operation and management of the highway; also increase the cost of data preprocessing, resulting in a large deviation between the research results of related expansion application and the expected effect. Therefore, repairing abnormal ETC transaction data to protect data quality not only helps to speed up the construction of intelligent highway, realizes the fine management of transportation industry, but also provides solid data support for ETC big data mining expansion application, which has very important significance.

[0003] At present, the research work on abnormal data repair mainly focuses on GPS trajectory data and traffic flow data collected by other traffic sensors. The unique OD constraint, spatio-temporal and semantic coupling of ETC transaction data make these researches cannot be directly applied to solve the data repair problem; and the researches on time repair are mostly on the restoration of time sequence, which cannot accurately repair the time quantitatively. SUMMARY

[0004] The purpose of the present application is to provide a highway ETC missing transaction data repair method based on multi-scene features.

[0005] The technical solution adopted by the present application is:

[0006] The highway ETC missing transaction data repair method based on multi-scene features comprises the following steps:

[0007] Step 1, a trajectory data set is established by multi-source data fusion; after the trajectory data of the trajectory data set is detected abnormally, an abnormal trajectory data set is extracted;

[0008] Step 2, a repair data set is established based on the data features extracted from the abnormal trajectory data detection results according to the ETC topology data set;

[0009] Step 3, the ETC missing transaction gantry path in the repair data set is repaired based on the topological constraint;

[0010] Step 4, the vehicle speed features in the corresponding scene are obtained according to the different scenes of the highway traffic flow, so as to repair the time of the highway ETC missing transaction data.

[0011] Further, the data in the trajectory data in the service area section in step 1 is marked according to the ETC topology data set; and the weighted FCM algorithm is used to detect and mark whether the section to which the missing transaction portal belongs is in the congestion period.

[0012] Further, step 2 specifically includes the following steps:

[0013] Step 2-1, extract the reference portal number for repairing the missing transaction portal path from the trajectory data, establish the portal number matrix F, then:

[0014]

[0015] F m =[f 11 f 21 … f n1 ] T (6)

[0016] Where F m represents the portal number column extracted from n vehicle trajectories, f nm represents the portal number, and n represents the number of vehicle trajectories in batch processing;

[0017] Step 2-2, according to the aggregation results of the portal number matrix F and the ETC topology data set, establish the section distance matrix D, then:

[0018] D=[D 12 D 23 … D (m-1)m ],D (m-1)m =[d 11 d 21 … d n1 ] T (7)

[0019] Where D (m-1)m is the distance of the section composed of the mth and m-1th adjacent portals;

[0020] Step 2-3, extract the transaction time matrix T using the same principle, then:

[0021] T=[T1 T2 … T m ],T m =[t 11 t 21 … t n1 ] T (8) Where t n is the portal transaction time, and T m represents the transaction time column extracted from n vehicle trajectories according to the algorithm requirements;

[0022] Step 2-4, establish the data repair matrix R, then:

[0023] R = [FTD] (9)

[0024] Wherein, F is the gantry number block matrix, which will be input as part of the path repair algorithm; T is the gantry transaction time block matrix, and D is the section distance block matrix, both of which will be input as part of the time repair algorithm.

[0025] Specifically, the repair of missing transaction trajectory data mainly includes path repair and transaction time repair. For the repair of the path, since the path in the trajectory data is represented as a sequence of gantry numbers, the problem is converted into the interpolation problem of missing transaction gantry numbers. F is the gantry number block matrix, which will be input as part of the path repair algorithm. For the repair of the transaction time, a corresponding repair algorithm needs to be designed according to the highway driving characteristics and ETC topological distribution characteristics. T is the gantry transaction time block matrix, and D is the section distance block matrix, both of which will be input as part of the time repair algorithm.

[0026] Further, step 3 specifically includes the following steps:

[0027] Step 3-1, establish a topological node table according to the ETC topological data;

[0028] Step 3-2, establish a gantry number matrix F through the adjacent two gantry numbers before and after the missing transaction gantry, then:

[0029] F = [f FOR F AFT ] (10)

[0030] Wherein, f FOR refers to the gantry number matrix established by the adjacent two gantries before the missing transaction gantry; f AFT refers to the gantry number matrix established by the adjacent two gantries after the missing transaction gantry.

[0031] Step 3-3, aggregate the topological table twice according to the gantry number matrix standard to obtain the node rear gantry list to_id_arr of the F FOR matrix and the node front gantry list from_id_arr of the F AFT matrix;

[0032] Step 3-4, calculate the intersection of the to_id_arr and from_id_arr lists to uniquely obtain the gantry number of the missing transaction gantry.

[0033] Further, the highway traffic flow in step 4 includes a non-saturated free flow scenario, a super-saturated congestion scenario, a service area parking scenario, and a missing transaction concurrency scenario.

[0034] Further, the specific steps of step 4 are as follows:

[0035] When in a non-saturated free flow state, the transaction time of the missing transaction gantry is repaired with reference to the average driving speed of the reference section, the transaction time of the missing transaction gantry in the free flow scenario of traffic is calculated, and the calculation formula is as follows:

[0036] T LOSE = T FOR + D FOR ⊙ [(D FOR + D AFT ) ⊙ (T AFT - T FOR ) -1 ] -1 (15)

[0037] Wherein, T FOR represents the transaction time of the adjacent previous gantry of the missing transaction gantry; T AFT represents the transaction time of the adjacent previous gantry of the missing transaction gantry; D FOR represents the section distance of the previous section of the section where the missing transaction gantry is located; D AFT represents the section distance of the next section of the section where the missing transaction gantry is located;

[0038] When in a super-saturated congestion state, the transaction time of the missing transaction gantry is repaired with reference to the speed characteristics of the same vehicle; that is:

[0039]

[0040] Wherein, represents the transaction time of the previous vehicle of the missing transaction vehicle at the previous gantry; D JAM represents the section distance of the congestion section; V JAM represents the average speed of the congestion section;

[0041] When in a service area parking state, the transaction time of the missing transaction gantry in the service area parking state is calculated based on the kinematic inertia principle method combined with ETC transaction data The corresponding calculation formula is as follows:

[0042]

[0043] Wherein, represents the transaction time of the previous two gantries of the missing transaction gantry; represents the transaction time of the previous two gantries of the missing transaction gantry; This indicates the distance between the two gates preceding the missed transaction gate and the gate preceding the missed transaction gate. This indicates the distance between the previous gantry of the missed transaction gantry and the missed transaction gantry;

[0044] When there are concurrent missed and erroneous transactions, the transaction time T is based on the erroneous transaction gantry. L ′ OSE Calculate the transaction time of the missed transaction gantry under concurrent conditions. The corresponding calculation formula is as follows:

[0045]

[0046] Among them, D LOSE T represents the error distance between the previous gantry of the missed transaction gantry and the missed transaction gantry; FOR Indicates the front gantry trading time of the missed trading gantry; D ERR Indicates the segment distance D of the segment where the missed transaction gantry is located. ERR .

[0047] Furthermore, the calculation steps for the transaction time of the missed transaction gantry when in a non-saturated free-flow state are as follows:

[0048] The transaction time T of two adjacent gates using a missed transaction gate. FOR and T AFT Construct a time matrix T, and the segment distance D between the two segments preceding and following the segment where the missed transaction gantry is located. FOR and D AFT To construct the segment distance matrix D, and then to construct the data repair matrix R under the unsaturated free flow state. NOM The format is as follows:

[0049] R NOM =[F FOR F LOSE F AFT T FOR T AFT D FOR D AFT (11)

[0050] Among them, F LOSE This refers to the missing transaction gantry; F FOR This refers to the gantry numbering matrix established by the two adjacent gantry frames of the missed transaction gantry; F AFT It refers to the gantry numbering matrix established by the two gantry adjacent to the missed transaction gantry;

[0051] Obtain the time matrix T and the segment distance matrix D, and calculate the passage speed V of the gantry adjacent to the missed transaction gantry. FOR ,Right now:

[0052]

[0053] wherein V ALL is the average speed between the two preceding gantries adjacent to the missing transaction gantry,

[0054] According to the aggregation result of the gantry number block matrix and the ETC topology data table, the section distance D FOR of the section is obtained

[0055] ΔT=D FOR ⊙V FOR -1 (13)

[0056] wherein is the Hadamard product, representing the product of the elements of the two matrices;

[0057] Based on the transaction time T FOR of the preceding gantry adjacent to the missing transaction gantry and the travel time ΔT between the two preceding gantries adjacent to the missing transaction gantry, the transaction time T LOSE of the missing transaction gantry is calculated, and the corresponding calculation formula is as follows:

[0058] T LOSE = T FOR + ΔT (12)

[0059] That is, the transaction time of the missing transaction gantry in the free flow of traffic is obtained by the following formula:

[0060] T LOSE = T FOR + D FOR ⊙[(D FOR + D AFT )⊙(T AFT -T FOR ) -1 ] -1 (15)

[0061] Further, the calculation steps of the transaction time of the missing transaction gantry in the oversaturated congestion state are as follows:

[0062] First, the transaction time of the missing transaction gantry in the trajectories of the preceding and following vehicles of the missing transaction vehicle and the transaction time of the preceding gantry of the missing transaction gantry of the three vehicles and the section distance D JAM of the congestion section are extracted, and the data repair matrix R JAM in the oversaturated congestion state is established, and then:

[0063]

[0064] where F LOSE denotes the missing transaction gantry; F FOR denotes the gantry number matrix established by the adjacent two gantries in front of the missing transaction gantry;

[0065] the average speed V AVE of the preceding and following vehicle congestion sections of the missing transaction vehicle is used to represent the average speed V JAM of the corresponding congestion section, and then:

[0066]

[0067] The section distance D JAM of the congestion section is obtained according to the aggregation result of the gantry number block matrix and the ETC topology data table, and the travel time ΔT JAM of the congestion section is calculated, and the corresponding calculation formula is as follows:

[0068] ΔT JAM = D JAM ⊙ V JAM -1

[0069] where ⊙ is the Hadamard product, representing the product of the elements of two matrices;

[0070] The missing transaction gantry transaction time T of the congestion section is calculated based on the preceding vehicle transaction time T of the missing transaction vehicle and the travel time ΔT JAM of the congestion section, that is:

[0071]

[0072] Further, the calculation steps of the missing transaction gantry transaction time T when in the service area parking state are as follows:

[0073] The gantry transaction times T and T of the adjacent two gantries in front of the missing transaction gantry are obtained, and the section distance matrices D and D from the two gantries in front of the missing transaction gantry to the missing transaction gantry are obtained, and a data repair matrix R RSA in the service area parking state is constructed, and then:

[0074]

[0075] where T represents the transaction time of the two gantries in front of the missing transaction gantry; represents the transaction time of the gantry in front of the missing transaction gantry; and F LOSEis referred to as the missing transaction gantry; is referred to as the segment distance from the first two gantries of the missing transaction gantry to the front of the previous gantry of the missing transaction gantry; is referred to as the segment distance from the previous gantry of the missing transaction gantry to the missing transaction gantry;

[0076] The average speed of the segment from the first two gantries of the missing transaction gantry to the front of the previous gantry of the missing transaction gantry is calculated and The average speed of the segment from the previous gantry of the missing transaction gantry to the missing transaction gantry is calculated that is,

[0077]

[0078] The travel time ΔT of the segment in the service area stop state is calculated RSA , and the corresponding calculation formula is as follows:

[0079]

[0080] The transaction time of the missing transaction gantry in the service area stop state is calculated The corresponding calculation formula is as follows:

[0081]

[0082] Further, the transaction time in the missing and error transaction concurrent state The calculation steps are as follows:

[0083] The error distance D between each pair of gantries is calculated using the geographic position information of the gantries in the ETC topology data LOSE ;

[0084] The corresponding missing transaction gantry transaction time T′ is obtained LOSE , the front gantry transaction time T of the missing transaction gantry FOR , and the segment distance D of the segment ERR A data repair matrix is established under the missing and error transaction concurrent state, and then:

[0085] R SAM = [F FOR F LOSE F′ LOSE T FOR T′ LOSE D LOSE D ERR ] (22)

[0086] Based on the data repair matrix data under the missing and error transaction concurrent state, the transaction time of the missing transaction gantry is calculated The specific calculation formula is as follows:

[0087]

[0088] The application adopts the above technical scheme, and establishes a trajectory data set through multi-source data fusion; after performing abnormality detection on the trajectory data, an abnormal trajectory data set is extracted. Before data repair, the data in the service area section in the trajectory data needs to be marked according to the ETC topological data set; and the weighted FCM algorithm is referenced to detect and mark whether the missing transaction gantry belongs to the congestion period. Subsequently, related features are extracted according to the detection results and algorithm requirements to establish a repair data set. For the repair problem of the path, since the path in the trajectory data is in the form of a gantry number sequence, the problem is converted into an interpolation problem of the missing transaction gantry number. On the basis of the ETC missing transaction gantry path repair method based on topological constraints, four different scenes, i.e., a highway traffic flow non-saturated free flow scene, a supersaturated congestion scene, a service area parking scene and a missing and error transaction concurrent scene, are considered, and the speed characteristics of the vehicles in different scenes are deeply analyzed, and targeted time repair strategies are provided. The average repair error of the method is within 18s, and the execution time is only 0.07s. BRIEF DESCRIPTION OF DRAWINGS

[0089] The application will be further described in detail below in combination with the drawings and specific embodiments;

[0090] Figure 1 is a structural diagram of abnormal trajectory data;

[0091] Figure 2 is a schematic diagram of the principle of the ETC missing transaction gantry path repair method;

[0092] Figure 3 is a schematic diagram of the average speed change of part of vehicles in a non-saturated free flow state;

[0093] Figure 4 is a schematic diagram of the Speed Error of part of vehicles in a non-saturated free flow state;

[0094] Figure 5 is a schematic diagram of the average speed change of part of vehicles in a supersaturated congestion state;

[0095] Figure 6 is a schematic diagram of the Speed Error of part of vehicles in a supersaturated congestion state;

[0096] Figure 7 is a schematic diagram of the average speed change of part of vehicles in a supersaturated congestion state;

[0097] Figure 8 is a schematic diagram of the Time Error of the missing transaction and correct transaction gantries;

[0098] Figure 9 Repair effect comparison diagram for different strategies acting alone;

[0099] Figure 10 MAE result diagram for comparison experiment of different algorithms;

[0100] Figure 11 RMSE comparison diagram for comparison experiment of different algorithms;

[0101] Figure 12 Efficiency comparison diagram for different algorithms. Embodiments

[0102] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0103] As shown in Figures 1 to 12 The application discloses a highway ETC missing transaction data repair method based on multi-scene features, which comprises the following steps:

[0104] Step 1, a trajectory data set is established through multi-source data fusion; after performing abnormality detection on the trajectory data of the trajectory data set, an abnormal trajectory data set is extracted;

[0105] Step 2, a repair data set is established based on data features extracted from the abnormal trajectory data detection result according to an ETC topology data set;

[0106] Step 3, ETC missing transaction gantry paths in the repair data set are repaired based on topology constraints;

[0107] Step 4, vehicle speed features in corresponding scenes are obtained according to different scenes of highway traffic flow, so as to repair the time of the highway ETC missing transaction data in a targeted manner.

[0108] Further, the data in the trajectory data set in the service area section is marked according to the ETC topology data set in step 1; and whether the missing transaction gantry belongs to a congestion period is detected and marked by referring to a weighted FCM algorithm.

[0109] Further, step 2 specifically comprises the following steps:

[0110] Step 2-1, a reference gantry number for repairing the missing transaction gantry path is extracted from the trajectory data, a gantry number matrix F is established, and the following equation is established:

[0111]

[0112] F m = [f 11 f 21… f n1 ] T (6)

[0113] where F m represents the gantry number column extracted from n vehicle trajectories, f nm represents the gantry number, and n represents the number of vehicle trajectories for batch processing;

[0114] Step 2-2, according to the results of the aggregation of the gantry number matrix F and the ETC topology data, a section distance matrix D is established, and then:

[0115] D=[D 12 D 23 … D (m-1)m ],D (m-1)m =[d 11 d 21 … d n1 ] T (7)

[0116] where D (m-1)m is the distance of the section composed of the mth gantry and the m-1th gantry;

[0117] Step 2-3, using the same principle, a transaction time matrix T is extracted, and then:

[0118] T=[T1 T2 … T m ],T m =[t 11 t 21 … t n1 ] T (8) where t n is the gantry transaction time, and T m represents the transaction time column extracted from n vehicle trajectories according to the algorithm requirements;

[0119] Step 2-4, a data repair matrix R is established, and then:

[0120] R=[F T D] (9)

[0121] where F is the gantry number block matrix, which will be used as the input of the path repair algorithm part; T is the gantry transaction time block matrix, and D is the section distance block matrix. Both T and D will be used as the input of the time repair algorithm part.

[0122] Specifically, the repair of missing transaction trajectory data mainly includes path repair and transaction time repair. For the repair of the path, since the path in the trajectory data is represented as a sequence of gantry numbers, the problem is converted into the interpolation of missing transaction gantry numbers. F is a gantry number block matrix, which will be used as the input of the path repair algorithm. For the repair of the transaction time, a corresponding repair algorithm is designed according to the highway driving characteristics and ETC topological distribution characteristics. T is a gantry transaction time block matrix, and D is a section distance block matrix, both of which will be used as the input of the time repair algorithm.

[0123] Further, step 3 specifically includes the following steps:

[0124] Step 3-1, establishing a topological node table according to the ETC topological data;

[0125] Step 3-2, establishing a gantry number matrix F through the adjacent two gantry numbers before and after the missing transaction gantry, so that:

[0126] F = [F FOR F AFT ] (10)

[0127] Wherein, F FOR refers to the gantry number matrix established by the adjacent two gantries before the missing transaction gantry; F AFT refers to the gantry number matrix established by the adjacent two gantries after the missing transaction gantry.

[0128] Step 3-3, aggregating the topological table twice according to the gantry number matrix standard to obtain the node rear gantry list to_id_arr of the F FOR matrix and the node front gantry list from_id_arr of the F AFT matrix;

[0129] Step 3-4, calculating the intersection of the to_id_arr and from_id_arr lists to uniquely obtain the gantry number of the missing transaction gantry.

[0130] Further, the highway traffic flow in step 4 includes a non-saturated free flow scenario, an oversaturated congestion scenario, a service area parking scenario, and a missing and error transaction concurrent scenario.

[0131] Further, the specific steps of step 4 are as follows:

[0132] When in a non-saturated free flow state, the transaction time of the missing transaction gantry is repaired by referring to the average driving speed of the reference section, and the transaction time of the missing transaction gantry in the free flow scenario of traffic is calculated, and the calculation formula is as follows:

[0133] T LOSE = T FOR+D FOR ⊙[(D FOR +D AFT )⊙(T AFT -T FOR ) -1 ] -1 (15)

[0134] wherein, T FOR represents the transaction time of the previous gantry adjacent to the missing transaction gantry; T AFT represents the transaction time of the previous gantry adjacent to the missing transaction gantry; D FOR represents the section distance of the previous section of the section where the missing transaction gantry is located; D AFT represents the section distance of the next section of the section where the missing transaction gantry is located;

[0135] When in the supersaturated congestion state, the transaction time of the missing transaction gantry is repaired with reference to the speed characteristics of the same lane vehicle; that is:

[0136]

[0137] wherein, represents the transaction time of the previous vehicle of the missing transaction vehicle; D JAM represents the section distance of the congestion section; V JAM represents the average speed of the congestion section;

[0138] When in the service area parking state, the transaction time of the missing transaction gantry in the service area parking state is calculated based on the kinematic inertia principle method combined with ETC transaction data The corresponding calculation formula is as follows:

[0139]

[0140] wherein, represents the transaction time of the previous two gantries of the missing transaction gantry; represents the transaction time of the previous two gantries of the missing transaction gantry; represents the section distance from the previous two gantries of the missing transaction gantry to the front of the previous gantry of the missing transaction gantry; represents the section distance from the previous gantry of the missing transaction gantry to the missing transaction gantry;

[0141] When in the missing and false transaction concurrent state, the transaction time of the missing transaction gantry in the missing and false transaction concurrent state is calculated based on the false transaction gantry transaction time T′ LOSE The corresponding calculation formula is as follows:

[0142]

[0143] wherein, D lOSE ​Error distance between the previous gantry and the missing transaction gantry; T FOR Transaction time of the previous gantry of the missing transaction gantry; D ERR Section distance D of the section where the missing transaction gantry is located ERR .

[0144] Further, when in the non-saturated free flow state, the transaction time of the missing transaction gantry is calculated as follows:

[0145] The transaction time T of the two adjacent gantries before and after the missing transaction gantry is used FOR and T AFT The time matrix T is constructed, and the section distance D of the two sections before and after the section where the missing transaction gantry is located FOR and D AFT The section distance matrix D is constructed, and the data repair matrix R in the non-saturated free flow state is constructed NOM , which is in the form as follows:

[0146] R NOM = [F FOR F LOSE F AFT T FOR T AFT D FOR D AFT ] (11)

[0147] Wherein, F LOSE refers to the missing transaction gantry; F FOE refers to the gantry number matrix established by the two adjacent gantries before the missing transaction gantry; F AFT refers to the gantry number matrix established by the two adjacent gantries after the missing transaction gantry;

[0148] The time matrix T and the section distance matrix D are obtained, and the passing speed V of the previous gantry adjacent to the missing transaction gantry is calculated FOR , that is:

[0149] V FOR ≒V ALL = (D FOR +D AFT )⊙(T AFT -T FOR ) -1 (14)

[0150] Wherein, V ALL is the average speed between the two adjacent gantries before the missing transaction gantry,

[0151] According to the aggregation results of the gantry number block matrix and the ETC topology data table, the section distance D of the section is obtained FORThe passage time ΔT between the two adjacent gantries of the missed transaction gantry is calculated using the following formula:

[0152] ΔT=D FOR ⊙V FOR -1 (13)

[0153] Where ⊙ represents the Hadamard product, which is the product of the elements of two matrices;

[0154] Based on the transaction time T of the previous gate adjacent to the missed transaction gate FOR The transaction time T of the missed transaction gantry is calculated by taking the passage time ΔT between the two adjacent gantry ... LOSE The corresponding calculation formula is as follows:

[0155] T LOSE =T FOR +ΔT (12)

[0156] That is, the transaction time of the missed transaction gantry in the traffic free-flow scenario is obtained by the following formula:

[0157] T LOSE =T FOR +D FOR ⊙[(D FOR +D AFT )⊙(T AFT -T FOR ) -1 ] -1 (15).

[0158] Furthermore, the calculation steps for the transaction time of the missed transaction gantry in the case of oversaturation and congestion are as follows:

[0159] First, extract the transaction time of the missed transaction gantry from the preceding and following vehicle trajectories of the vehicles that missed transactions. The time of the front gantry transaction of the three vehicles missing the transaction gantry. And the segment distance D of the congested section JAM Establish a data repair matrix R under oversaturated congestion conditions. JAM ,but:

[0160]

[0161] Among them, F LOSE This refers to the missing transaction gantry; F FOR It refers to the gantry numbering matrix established by the two gantry adjacent to the gantry with the missing transaction;

[0162] The average speed V of the congested sections before and after the vehicle with the missed transaction is used. AVE The average speed V represents the corresponding congested section.JAM Then we have:

[0163]

[0164] According to the aggregation result of the gantry number block matrix and the ETC topology data table, the section distance D of the congestion section is obtained JAM , and the travel time ΔT of the congestion section is calculated JAM The corresponding calculation formula is as follows:

[0165] ΔT JAM =D JAM ⊙V JAM -1

[0166] Where ⊙ is the Hadamard product, indicating the product of the elements of the two matrices;

[0167] Based on the front vehicle transaction time of the missing transaction vehicle at the front gantry And the travel time ΔT of the congestion section JAM The missing transaction gantry transaction time of the congestion section is calculated That is:

[0168]

[0169] Further, when the missing transaction gantry is in the service area parking state, the transaction time of the missing transaction gantry The calculation steps are as follows:

[0170] Get the gantry transaction time of the first two gantries adjacent to the missing transaction gantry And And the section distance matrix of the first two gantries of the missing transaction gantry to the missing transaction gantry And Construct the data repair matrix R of the missing transaction gantry in the service area parking state RSA Then we have:

[0171]

[0172] Where, Indicates the first two gantry transaction times of the missing transaction gantry; Indicates the first gantry transaction time of the missing transaction gantry; F LOSE Refers to the missing transaction gantry; Indicates the section distance from the first two gantries of the missing transaction gantry to the front of the first gantry of the missing transaction gantry; Indicates the section distance from the first gantry of the missing transaction gantry to the missing transaction gantry;

[0173] Calculate the average speed of the section from the first two gantries of the missing transaction gantry to the front of the first gantry of the missing transaction gantry and Average speed of the previous gantry to the section of the missing transaction gantry That is

[0174]

[0175] Calculate the passage time ΔT of the section in the service area stop state RSA , the corresponding calculation formula is as follows:

[0176]

[0177] Calculate the transaction time of the missing transaction gantry in the service area stop state The corresponding calculation formula is as follows:

[0178]

[0179] Further, when the missing and error transaction concurrent state, the transaction time The calculation steps are as follows:

[0180] Use the geographic position information of the gantry in the ETC topology data to calculate the error distance D between each pair of gantries LOSE ;

[0181] Obtain the corresponding missing transaction gantry transaction time T LOSE , the previous gantry transaction time T of the missing transaction gantry FOR And the section distance D of the section ERR Establish the data repair matrix under the missing and error transaction concurrent state, then:

[0182] R SAM = [F FOR F LOSE F′ LOSE T FOR T′ LOSE D LOSE D ERR ] (22)

[0183] Based on the data repair matrix data under the missing and error transaction concurrent state, the transaction time of the missing transaction gantry is calculated The specific calculation formula is as follows:

[0184] The specific principle of the present application will be described in detail as follows:

[0185] Through the anomaly detection, it is known that the abnormal transaction events are mainly three kinds of missing transaction, error transaction and repeated transaction. Figure 1The gantry transaction sequence and generation mode of different abnormal trajectory data are shown. Among them, the abnormal problem of trajectory data caused by false transactions and repeated transactions is essentially data redundancy, and the solution is relatively simple, which can be directly deleted. The data missing problem caused by missing transactions is the key problem to be solved by the application. According to statistics, the probability of missing transactions in the trajectory data of two or more consecutive gantries is less than 5%. Therefore, the application only repairs most of the missing transaction trajectory data. Here, the consecutive missing transaction gantries in the subsequent missing transaction trajectory data are limited to only one, that is, the front and rear adjacent two gantries of the missing transaction gantry have normal transaction records. Since the ETC transaction data has rich features, the key features are mainly repaired, and the rest of the vehicle related information, gantry name and road number, etc. can be repaired according to the condition constraints.

[0186] In order to realize the rapid repair of massive data, batch and hierarchical processing of data sets is needed. First, the trajectory data set is established by multi-source data fusion; after the trajectory data is detected, the abnormal trajectory data set is extracted. It should be noted that before data repair, the data in the service area section of the trajectory data needs to be marked according to the ETC topology data set; and the weighted FCM algorithm is used to detect and mark whether the section to which the missing transaction gantry belongs is in the congestion period. Then, according to the detection results and algorithm requirements, the relevant features are extracted to establish the repair data set. Specifically, first, the reference gantry number used to repair the missing transaction gantry path is extracted from the trajectory data, and the gantry number matrix F is established, that is:

[0187]

[0188] F m =[f 11 f 21 … f n1 ] T (6)

[0189] Among them, F m represents the gantry number column extracted from n vehicle trajectories according to the algorithm requirements, f nm represents the gantry number, and n represents the number of vehicle trajectories for batch processing. Then, according to the aggregation results of the gantry number matrix F and the ETC topology data set, the section distance matrix D is established, that is:

[0190] D=[D 12 D 23 … D (m-1)m ],D (m-1)m =[d 11 d 21 … d n1 ] T (7)

[0191] where D (m-1)m is the distance of the section composed of two adjacent gantries numbered m and m-1. Finally, the transaction time matrix T is extracted using the same principle, and then we have:

[0192] T = [T1 T2 … T m ], T m = [t 11 t 21 … t n1 ] T (8)

[0193] where t n is the transaction time of the gantry, and T m represents the transaction time column extracted from n vehicle trajectories according to the algorithm requirements. At this point, the data repair matrix R can be established according to the algorithm requirements, and then we have:

[0194] R = [F T D] (9)

[0195] The repair of missing transaction trajectory data mainly includes path repair and transaction time repair. For the repair of the path, since the path in the trajectory data is represented by a sequence of gantry numbers, the problem is converted into the interpolation of missing transaction gantry numbers. F is the gantry number block matrix, which will be used as the input of the path repair algorithm part. For the repair of the transaction time, corresponding repair algorithms need to be designed according to the characteristics of highway driving and ETC topological distribution, etc. T is the gantry transaction time block matrix, and D is the section distance block matrix, both of which will be used as the input of the time repair algorithm part.

[0196] ETC missing transaction gantry path repair based on topological constraints

[0197] According to the ETC gantry topology and the numbering rules of the background terminal system, when there is only one continuous gantry missing transaction, there will be no ambiguous trajectory problem, and the missing transaction gantry number can be uniquely determined. First, a topological node table is established according to the ETC topological data. The relevant features and descriptions are shown in Table 1.

[0198] Table 1 ETC topological node table

[0199] Serial number Field name Field attribute Example 1 Flag_id Topological node 35**03 / 27** 2 to_id_arr Node rear gantry list [35**03,…,34**05] 3 from_id_arr Node front gantry list [35**03,…,34**05]

[0200] where the topological node Flag_id contains all nodes in the ETC topological data; the node rear gantry list to_id_arr contains all gantry numbers that have topological connectivity and are located behind it; and from_id_arr contains all gantry numbers that have topological connectivity and are located in front of it. Subsequently, a gantry number matrix F is established through the adjacent front and rear two gantry numbers of the missing transaction gantry, and then we have:

[0201] F=[F FOR F AFT ] (10)

[0202] The topology table is aggregated twice in the gantry number matrix standard, respectively F FOR The node rear gantry list to_id_arr of the matrix, and F AFT The node front gantry list from_id_arr of the matrix. Finally, the list intersection of to_id_arr and from_id_arr can uniquely obtain a gantry number as the missing transaction gantry. The specific operation is shown in Figure 2

[0203] This method can not only accurately find the missing transaction gantry number in the missing transaction track data, ensure the integrity of the vehicle track path, and also has very excellent performance in repair efficiency. Experiments prove that in the case of data set magnitude of 100k, the algorithm only needs 30.6ms to complete the path repair task, which can meet the repair work requirements of ETC transaction data.

[0204] 2. Freeway ETC missing transaction data time repair strategy: For the ETC missing transaction gantry transaction time repair problem, the present application considers four different scenes of freeway traffic flow non-saturated free flow scene, oversaturated congestion scene, service area parking scene and missing error transaction concurrent scene. By deeply mining and analyzing the vehicle speed characteristics in different scenes, a targeted time repair strategy is designed.

[0205] (1) Non-saturated free flow state: As we all know, the traffic flow of most highways is in a non-saturated state, and the vehicle runs relatively smoothly under good road conditions, especially the driving speed fluctuation of adjacent sections is smaller. Such driving characteristics are called non-saturated free flow driving characteristics. Under this characteristic, the vehicle average driving speed V ExSec can describe its driving speed at a certain time in the section. The present application randomly selects a road section consisting of three consecutive sections on a highway, and depicts the V ExSec change curve of each section of all vehicles running on the road section during the free flow driving period, and the vehicle order is arranged according to the transaction time of the section front gantry of the first section, as shown in Figure 3 It can be seen that the section average driving speed of most vehicles basically remains at 70-110km / h, which is consistent with the driving speed characteristics of the highway in real life. At the same time, the section average driving speeds of all vehicles in the three consecutive sections are basically similar.

[0206] ​Further, in order to more intuitively describe the same vehicle in free flow driving characteristics, the V ExSec fluctuation of the continuous several sections, the present application calculates the average driving speed V ExRos of each vehicle in the section during the period ExSec , and depicts their respective V ExRos and the average absolute error of V FOR . From Figure 4 it can be seen that the average driving speed of these vehicles in the continuous several sections fluctuates less, and the average absolute error of most vehicles is below 5km / h.

[0207] After the above analysis of the free flow driving characteristics, the present application repairs the transaction time of the missing transaction gantry by referring to the average driving speed of the section. Specifically, the method uses the transaction time T FOR and T AFT of the two adjacent gantries before and after the missing transaction gantry to construct a time matrix T, and uses the section distance matrix D FOR and D AFT of the two sections before and after the missing transaction gantry to construct a section distance matrix D, which is in the following form:

[0208] R NOM =[F FOR F LOSE F AFT T FOR T AFT D FOR D AFT ] (11)

[0209] In order to realize the repair of the transaction time T LOSE of the missing transaction gantry, on the basis of the known transaction time T FOR of the previous gantry adjacent to the missing transaction gantry, it is necessary to first confirm the passage time ΔT of the section, that is:

[0210] T LOSE =T FOR +ΔT (12)

[0211] According to the aggregation result of the gantry number block matrix and the ETC topology data table, the section distance D FOR of the section can be obtained, so the section passage time can be obtained by the following formula:

[0212] ΔT=D FOR ⊙V FOR -1 (13)

[0213] Where ⊙ is Hadamard product, which represents the product of the elements of two matrices. V FORAs an unknown parameter, according to the above principle, the average speed V of the link is used ALL Description, that is:

[0214] V FOR ≒V ALL =(D FOR +D AFT )⊙(T AFT -T FOR ) -1 (14) In summary, the transaction time of the missing transaction portal in the free traffic flow scenario is obtained by the following formula:

[0215] T LOSE =T FOR +D FOR ⊙[(D FOR +D AFT )⊙(T AFT -T FOR ) -1 ] -1 (15)

[0216] (2) Over-saturation congestion state: However, due to the increase of travel demand during morning / evening peak hours / holidays, or sudden situations such as traffic accidents, road maintenance, etc., traffic congestion often occurs on expressways. Severe traffic congestion may affect a long section of road, causing the driving speed of two or even more consecutive sections of vehicles to fluctuate greatly. The present application randomly selects a section of expressway consisting of three consecutive sections, and depicts the V ExSec variation curve of all vehicles in each section during the congestion period, as shown in Figure 5 It can be seen that during the congestion period, the vehicles in the section far from the congestion event point are all in free flow state and are not affected by the congestion event. As for the congestion section affected by the congestion event, the average driving speed of the vehicles passing through the portal of the congestion section in the early and middle periods of congestion, i.e. 8:18-11:50, gradually decreases and then remains stable at below 5km / h; however, due to the setting of the emergency lane on the expressway, some vehicles may occupy the emergency lane during the congestion period, resulting in some mutations in the speed characteristic curve. Finally, with the solution of the congestion event, the vehicles quickly recover to the standard driving speed, and the vehicles affected by the congestion have a low traffic volume in the early stage of recovering to the free flow driving characteristic.

[0217] Through the above analysis, it can be seen that during the congestion period, the V ExSecIt is impossible to describe the speed characteristics of a vehicle within a specific segment at any given moment. However, during congested periods, the average speed characteristics of vehicles that are adjacent to each other in terms of transaction time at the front gantry of the segment are quite similar. To more intuitively verify this idea, this invention extracts the V values ​​of all vehicles in the segment during both congested and non-congested periods. ExSec ; Calculate the average speed V of the two vehicles in front and behind each other in the section. AVE With V ExSec The absolute difference in Speed ​​Error, such as Figure 6 As shown, the Speed ​​Error is low during congested periods, mostly remaining below 3 km / h, while during free-flow driving, the driving characteristics of adjacent vehicles differ significantly.

[0218] As the above analysis shows, vehicles entering a congested section sequentially and whose transaction times at the preceding gantry of the section are adjacent share a high degree of similarity in their segment travel characteristics. Therefore, for time restoration in this case, the speed characteristics of vehicles traveling in the same segment are referenced. Adjacent vehicles are defined as those adjacent in transaction time at the preceding gantry of the congested section. Based on the location of the missed transaction gantry within the congested section, the missed transaction gantry is identified as the following gantry of the congested section. First, this invention extracts the transaction time of the missed transaction gantry from the trajectories of vehicles before and after the missed transaction vehicle. The time of the front gantry transaction of the three vehicles missing the transaction gantry. And the segment distance D of the congested section JAM If a data repair matrix is ​​established, then:

[0219]

[0220] Based on the above analysis, this invention uses the average speed V of the congested sections before and after the missed transaction vehicle. AVE To describe its average velocity V in this section JAM Specifically, these include:

[0221]

[0222] Combining equations (12) and (13), the transaction time of the missed transaction gantry under congestion conditions is obtained by the following formula:

[0223]

[0224] (3) Service area stop state: Relevant research shows that in order to reasonably allocate rest time for drivers and avoid fatigue driving caused by long driving, the highway network will set up service areas or parking areas at intervals in consideration of available resources. According to the statistical analysis of the ETC topology data set, the ETC gantry system divides the highway network in Fujian Province into 2,948 sections, of which 1,455 sections have gantries at both nodes; further, of these sections composed of gantries, 1,123 sections are service area-free sections, and 332 sections are service area sections, that is, on average, there is one service area section in every four sections. If the missing transaction gantry is the front gantry or the rear gantry of a service area section, since it cannot be determined whether the vehicle enters the service area or not, and if it stays in the service area, the specific stay time, the use of V ExSec may have a large deviation in describing the traffic characteristics of the vehicle in the section. Further, the present application randomly selects a highway section containing a service area section, and depicts the V ExSec variation curves of all vehicles in each section in some time periods under the free flow traffic characteristics, as shown in Figure 7 . Among them, section 1 is a service area section, and sections 2 and 3 are the two adjacent sections without service areas. It can be seen that the V ExSec curve fitting degree of sections 2 and 3 is high, that is, the V ExSec fluctuation of most vehicles in the adjacent service area-free sections is small; and the V ExSec curve of section 1 is quite different from that of sections 2 and 3, which is due to the possibility of some vehicles stopping at the service area.

[0225] According to the above analysis, it can be seen that the vehicle speed characteristics of the service area section are greatly affected by whether the vehicle enters the service area or not. And the present application cannot accurately determine whether the vehicle enters the service area or not and its stay time according to the ETC related transaction data. Therefore, in view of the above situation, combined with the analysis of the vehicle driving characteristics in the free flow state, the kinematic inertia principle method is proposed. It should be noted that the repair principles of the front gantry and the rear gantry of the missing transaction gantry in the service area section are the same, and the transaction time of the front gantry missing transaction in the service area section is repaired below, and the other case is not described. For convenience of description, it is defined that in the road section composed of the service area section and its two preceding sections, the service area section is section 3, and the two preceding sections are section 2 and section 1 in turn. This method uses the gantry transaction time and of the two adjacent gantries of the missing transaction gantry to establish a time matrix T, and the section distance matrix and of section 1 and section 2 to establish a section distance matrix D, that is:

[0226]

[0227] The service area is located in section 3. Vehicles traveling in sections 1 and 2 are in free-flow mode and are not affected by the possibility of stopping at the service area. According to the road segment definition, the missed transaction gantry is also the rear gantry of section 2. In order to realize the transaction time T of the missed transaction gantry... LOSE The repair, in the known section 2 front gantry trading time Based on equations (12) and (13), it is necessary to first determine the travel time ΔT and the average speed of the section. in, As an unknown parameter, based on the above principle, the average speed of segment 1 is used. Description, namely:

[0228]

[0229] Therefore, in this scenario, the transaction time of the missed transaction gate is obtained by the following formula:

[0230]

[0231] (4) Concurrent Missed and Erroneous Transactions: During the statistical analysis of anomaly detection results, this invention discovered that a portion of the vehicle trajectory data in the abnormal vehicle trajectories is both missed and erroneous transaction trajectory data for a certain gantry. That is, the gantry with the transaction time to be repaired simultaneously experienced both missed and erroneous transactions. To address this situation, fully utilizing the transaction time data of the erroneous transaction gantry for repair work can improve repair accuracy. To verify this idea, this invention obtained the transaction time difference (Time Error) between the erroneous transaction gantry and the correct transaction gantry in a portion of the erroneous transaction trajectory data. For example... Figure 8 It can be seen that the passage time difference is basically maintained at around 8 seconds. Therefore, when missed transactions and erroneous transactions occur at the same time, the transaction time of the missed transaction gantry can be corrected based on the erroneous transaction time of the gantry and the distance error between a pair of gantries.

[0232] This invention uses the geographical location information of gantry in the ETC topology data to calculate the error distance D between each pair of gantry. LOSE ; and combined with the erroneous transaction gantry transaction time T L ′ OSE The front gantry transaction time T of the missing transaction gantry FOR And the segment distance D of this section ERR By establishing a data restoration matrix, we have:

[0233] R SAM =[F FOR F LOSE FL ′ OSE T FOR T L ′ OSE D LOSE D ERR ](22))

[0234] In combination with the above known parameters, the transaction time of the missing transaction gantry is obtained by the following formula:

[0235] T LOSE =T L ′ oSE ±D LOSE ⊙[D ERR ⊙(T L ′ OSE -T FOR ) -1 ] -1 (23)

[0236] The verification experiment of the technical effects of the present application: based on Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz and 64GB RAM, using CentOS Linux release 7.9.2009 (Core) operating system, using Python 3.8.8 programming language on open source Web application Jupyter Notebook, and using SQL language on Clickhouse21.11.4.14.

[0237] 1. Experimental data set construction: the ETC related data set and selected features are described; the process of constructing trajectory data set and test data set by multi-data fusion is introduced in detail.

[0238] 2. Data description: ETC transaction data: in the highway network, any vehicle passing through any gantry, the background terminal of the ETC system will produce an ETC transaction data, containing transaction time, vehicle information, gantry number and entrance toll station, etc. A total of 103 related information. Since the same vehicle has no uniqueness in the journey of the highway, in order to prevent the mixed influence of multiple journeys on toll management, the system also has an identification field for the same vehicle in different journeys. Table 2 is the selected features and their related descriptions.

[0239] Table 2 ETC transaction data

[0240] Serial number Field name Field attribute Example 1 TradeID Transaction number 3502*****3256 2 TradeTime Transaction time 2021-05-01 00:00:00 3 FlagID Gantry number 34**17 4 OBUPlate Vehicle identification ******** 5 VehClass Vehicle type 1 6 EnTime Entry time 2021-05-01 01:00:00 7 EnStation Entry toll station 1**1 8 PassID Trip number 013************5529

[0241] Toll station transaction data: The highway is a closed road network, and vehicles must enter and exit the road network at toll stations. Toll station transaction data records 24 pieces of relevant information about vehicles, including the time of entry and exit, toll station number, and other information. The purpose of using toll station transaction data is to determine the exit toll station of the vehicle's travel trajectory, making the ETC vehicle trajectory data complete and closed. Table 3 shows the selected features and their descriptions.

[0242] Table 3 Toll station transaction data

[0243] Serial number Field name Field attribute Example 1 TradeID Transaction number 3502*****3256 2 OBUPlate Vehicle identification ******** 3 VehClass Vehicle type 1 4 EnTime Entry time 2021-05-01 00:00:00 5 EnStation Entry toll station 1**1 6 ExTime Exit time 2021-05-01 00:00:00 7 ExStation Exit toll station 8**8 8 PassID Trip number 013************5529

[0244] ETC topology data: This data mainly records the connectivity and section distance between any two gantries or toll stations in the highway road network, as well as the geographic location, type, and number of service areas. Table 4 shows the selected features and their descriptions.

[0245] Table 4 ETC topology data

[0246] Serial number Field name Field attribute Example 1 FromID Section front gantry number 35**03 2 ToID Section rear gantry number 35**03 3 From_gro Section front gantry geographic information [26.5*19,118.7*5] 4 To_gro Section rear gantry geographic information [26.56*9,118.4*5] 5 Distance Section distance (m) **** 6 Service Service area name **Service area / no service area

[0247] 3. Data set construction: After simple filtering of the original data set features according to experimental requirements, further processing is needed to form the ETC vehicle trajectory data set. Through analysis of the data set, it was found that the PassID feature cannot completely distinguish vehicle travel because it records errors when a mistransaction occurs. After further research, a unique vehicle travel identification feature ID was established by combining the PassID feature and other relevant features, which can completely distinguish vehicle travel. According to this feature, table association and group aggregation were performed to complete the fusion of ETC transaction data and toll station transaction data, and to establish the trajectory data set. Table 5 shows a sample of the data set.

[0248] Table 5 ETC vehicle trajectory data set

[0249]

[0250]

[0251] The list feature is a specific description of the vehicle trajectory feature, the gantry number list Flagid Arr contains the gantry number of the vehicle trajectory that completes the transaction, and is arranged according to the topological rule constraint of the ETC highway network; the transaction time list Tradetime Arr and the trip number list PassID Arr contain the passing time of each gantry and the trip number, respectively, and the arrangement rule is the same as Flagid Arr. The algorithm aims to solve the abnormal transaction data repair work in the province. The highway network covers a wide range, so the data involved is large. In order to be as close to the actual demand as possible, the ETC transaction data of 3 different sections of the highway network in Fujian Province on May 1, 2021 is selected, and the three do not have any characteristic correlation, which can better reflect the general applicability of the algorithm. Due to the surge in highway traffic during holidays, selecting holiday data sets will include more driving features than other date sets, such as different degrees of congestion, more service area stops, etc. This basically includes all possible situations on the highway, and can better reflect the stability of the algorithm. Through the statistics of the real situation of the 3 sections, the experimental data set contains a total of 26,241 ETC vehicle trajectories, of which 7,545 vehicle trajectories involve service area sections, which basically meet the section distribution proportion of ETC topological data; 3,859 vehicle trajectories involve gantry mis-transaction; since the 3 sections all contain congestion sections, but the congestion time span does not reach 24 hours, the data extracted during the congestion period is used as the test data of the oversaturated congestion state, with a data amount of 5,504; the data of the non-congestion period is used as the test data of the non-saturated free flow state, with a data amount of 9,333.

[0252] On the basis of the above trajectory data set, the gantry number of a certain gantry and the transaction time of the gantry are separated and recorded as data labels; and the related features used in the test experiment are extracted from the trajectory data, and the ETC topological data set is aggregated to form the repair algorithm test data set. It is worth noting that when constructing the test data of the mis-transaction concurrent state, the trajectory data selected is only the trajectory data with mis-transaction, and the gantry as a data label needs to be paired with the mis-transaction gantry to ensure the verifiability of the experiment. Table 6 is a data set example.

[0253] Table 6 Repair algorithm test data set

[0254]

[0255]

[0256] Wherein, the missing transaction gantry number and its transaction time are data tags, and the two-bit repair reference gantry number is used for the implementation of the path repair algorithm. Five-bit reference gantry transaction time is set for time repair, and it is worth noting that not all repair strategies have the same meaning of reference gantry transaction time, and not all reference gantry transaction times are used in the execution process of the time repair algorithm. Specifically, according to the setting order, for the time repair of the missing and concurrent state of the transaction, only the missing transaction gantry transaction time Time_con_1 is used as the transaction time of the time repair reference gantry; when the vehicle is in the non-saturated free flow state, the transaction time repair reference missing transaction gantry is the first one gantry transaction time Time_con_1 and the last one gantry transaction time Time_con_2 of the missing transaction gantry; in the service area parking state, the time repair reference missing transaction gantry is the first one gantry transaction time Time_con_1 and the second one gantry transaction time Time_con_2 of the missing transaction gantry; when the vehicle is in the oversaturated congestion state, the time repair reference missing transaction gantry is the first one gantry transaction time Time_con_1, the two gantry transaction times Time_con_2 and Time_con_3 of the front vehicle, and the two gantry transaction times Time_con_4 and Time_con_5 of the rear vehicle.

[0257] 4. Evaluation index: The effectiveness of the method is verified by using real ETC transaction data, and the algorithm evaluation indexes used are mean absolute error (MAE), root mean squared error (RMSE), and algorithm execution efficiency. MAE can reflect the actual situation between the repair value and the missing value, and the smaller the value, the higher the accuracy of the repair value. RMSE is used to measure the absolute deviation between the repair value and the missing value, and the smaller the RMSE, the closer the repair value is to the original value. The following are the calculation formulas of the two indexes:

[0258]

[0259]

[0260] 5. Ablation experiment: The time repair strategy is split, and each module is tested separately on the entire data set to test the influence on the repair accuracy. When using a repair strategy for a test data set, other test data uses the repair strategy under the non-saturated free flow state, and the experimental results are shown in Table 7.

[0261] Table 7 Ablation experiment

[0262] Miss and error strategy Service area strategy Congestion strategy MAE RMSE Test 1 140.38 451.89 Test 2 √ 120.04 353.09 Test 3 √ 103.75 368.88 Test 4 √ 74.84 381.75 Test 5 √ √ √ 26.34 56.95

[0263] It can be seen that different modules have different degrees of improvement on the accuracy of the repair algorithm. Specifically, the repair strategy of the oversaturated congestion state alone has a greater improvement in algorithm accuracy, especially the MAE near 1 times the improvement effect. This confirms that when the highway traffic is congested, this strategy makes an outstanding contribution to the repair of the gantry transaction time. But it can be found that its RMSE has not been improved to the corresponding degree, which may be due to the high proportion of data in the overall test data during the congestion period. At the same time, the complex driving characteristics during the congestion period result in a large error interval for time repair. The repair strategy of the missing and false transaction concurrency alone has a greater improvement on the repair accuracy of the algorithm than the congestion state, which may be due to the small proportion of the amount of false transaction data in the data set, but the excellent improvement effect of its RMSE value is sufficient to prove its stability. In order to exclude the influence of data proportion on the repair accuracy of the algorithm, the different strategies are used alone on the corresponding data set, and the repair effect is compared with the repair effect of the repair strategy in the non-saturated free flow state. Among them, in the non-saturated free flow state, the time repair effect of the strategy is relatively good, with MAE of 4.619 and RMSE of 6.238. The repair effect of the remaining strategies is as shown in Table 5. Figure 10

[0264] It can be seen that the repair strategy in the missing and false transaction concurrency state performs best, with an average error of less than 4s and a small error fluctuation interval. The repair strategy of the service area stop state is also relatively good, with MAE and RMSE both remaining at 15-20s. Both of them have a greater improvement than the repair method using the average driving speed of the road section. In terms of horizontal comparison of several strategies, the repair effect of the repair strategy in the oversaturated congestion state seems to be poor, but this is also due to the fact that when the section is in the congestion period, the traffic flow increases dramatically, resulting in a very low average driving speed of the vehicle in the section. Therefore, in the same time, there is a shorter driving distance than during the free flow period. Further combining the actual driving speed of different vehicles in the section, the repair error of the gantry transaction time is analyzed. The results are shown in Table 8, where MAE-1 and MAE-2 are the repair error distances of the missing transaction gantry before and after the section.

[0265] Table 8

[0266] MAE-1 MAE-2 Congestion strategy 485.591 489.815 Service area strategy 391.548 393.093

[0267] It can be seen that from the dimension of distance error, the average repair error of both is basically maintained at 500-400m. Combined with the section distance statistics of each section in the topology data, such error distance can guarantee 99.98% of the topology accuracy. This means that the repair error of the gantry passing time will not cause the deviation of the gantry position in the trajectory data of the ETC vehicle, so as to affect the expansion application of ETC big data mining.

[0268] ​6. Algorithm comparison experiment: After different test experiments and in-depth analysis of the repair strategy proposed in this paper, according to the ETC transaction data features used by the algorithm in this paper, some classic algorithms commonly used in repair work are selected. These algorithms include average speed interpolation method, linear regression algorithm, KNN algorithm, decision tree algorithm and random forest algorithm. The test data is processed according to the feature requirements of different algorithms, and 70% of the data set is used as the training set of machine learning, and 30% is used as the test set for comparison experiment, to further prove the effectiveness of the algorithm.

[0269] 7. Precision evaluation: Evaluate the repair effect of each algorithm with algorithm precision. Specifically, randomly cause part of the data to be missing, and obtain 5% to 40% of 8 different missing rate data sets. In order to obtain relatively stable and high reliability experimental results, all algorithms are executed on different data sets for 100 times, and the mean value of each evaluation index is taken. As shown in Figure 11 The average speed interpolation method and the linear regression model, which are simple linear missing data imputation methods, perform poorly when dealing with more complex highway driving characteristics, with MAE values remaining around 100s; other machine learning algorithms perform well, especially the ensemble learning algorithm based on the random forest model, with a repair average error of about 40s. The algorithm proposed in this application performs well in different missing rate test data sets, with an error mean of 18s.

[0270] From the stability, with the increase of data missing rate, the repair effect of the selected contrast algorithm presents a downward trend, and the growth rate of RMSE value remains between 11-108. Among them, the linear regression algorithm and the average speed interpolation method are easily affected by data missing, with high RMSE and large repair effect fluctuation; while the RMSE of decision tree algorithm and random forest algorithm remains between 100-110, but both are less affected by data missing. The algorithm proposed in the application is almost not affected by the data missing rate, and the RMSE is the lowest, which remains at 40. According to the above analysis, due to the limitation of feature diversity, the machine learning algorithm does not show good characteristics when solving the data repair problem. The reason for this phenomenon may be that the ETC data in the time domain is different, and the spatial span of the highway network is large, which does not show good data correlation. According to the topological data, the average span between highway gantries is 2km, combined with the general speed limit interval of highway (80-120km / h), if the transaction time repair error is large, it may affect the use of highway ETC system data, and further affect the highway operation management. The algorithm proposed in this paper controls the average repair error of time within 18s, combined with the average driving speed of the vehicle in the section before and after the missing transaction gantry, the average absolute error distance caused by the repair of the missing transaction gantry transaction time is 257.16m, which means that this algorithm not only can complete the quantitative repair of time, and the repair error caused will not cause the phenomenon that the order of gantry arrangement and the order of transaction time in the ETC vehicle trajectory data are deviated; while the machine learning algorithm will make about 14%-31.2% of the topology wrong, which means that the repair error of the transaction time of the gantry by these algorithms may cause the order of the transaction time in the trajectory data arranged according to the transaction sequence to be arranged in disorder, which has a great influence on the expansion application of ETC transaction data.

[0271] 8. Efficiency evaluation: In order to evaluate the execution efficiency of the proposed method, different data quantities are set for experiments, which are 1k, 10k, 100k, 1000k (1M) and so on.

[0272] After executing 100 repair operations on different magnitude data sets for all algorithms, and taking the average of the execution time, the change curve of the execution efficiency of each algorithm with the data magnitude is obtained, as shown in Figure 12It can be seen that the average speed interpolation method is the most efficient among all algorithms and is almost not affected by the change of data volume; while the machine learning algorithm has a relatively excellent performance in accuracy, but its time complexity shows a trend of improvement with the increase of data volume. For the repair work of less data volume, the execution efficiency of each algorithm remains within 0.1s, and the difference can be ignored. When the data volume reaches 100k, the execution time of the RF and KNN algorithms which perform relatively well in repair accuracy begins to show a large difference with other algorithms; when dealing with 1M magnitude data volume, the execution time of the RF algorithm is 2.05s, while the execution efficiency of the algorithm in the present application remains the same level as the average speed interpolation method, and the repair time only needs 0.07s. At the same time, the algorithm in the present application does not need model pre-training, while the machine learning algorithm needs model training and part of parameter optimization, which greatly reduces the repair efficiency of the algorithm.

[0273] The present application adopts the above technical scheme, and establishes a trajectory data set through multi-source data fusion. After detecting the trajectory data, the abnormal trajectory data set is extracted. Before data repair, the data in the service area section of the trajectory data needs to be marked according to the ETC topology data set; and the weighted FCM algorithm is used to detect and mark whether the missing transaction gantry belongs to the congestion period. Then, the related features are extracted according to the detection results and algorithm requirements to establish a repair data set. For the repair problem of the path, since the path in the trajectory data is represented as a sequence of gantry numbers, the problem is converted into the interpolation problem of the missing transaction gantry number. On the basis of the ETC missing transaction gantry path repair method based on topological constraints, considering four different scenes of highway traffic flow non-saturated free flow scene, oversaturated congestion scene, service area parking scene and missing and error transaction concurrent scene, the speed characteristics in different scenes are deeply analyzed, and the corresponding time repair strategy is adopted. The average repair error of the method is within 18s, and the execution time is only 0.07s.

[0274] Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

Claims

1. A method for repairing missing transaction data of an expressway ETC based on multi-scene features, characterized in that: It comprises the following steps: Step 1, establishing a trajectory data set by multi-source data fusion; Step 2, based on the ETC topology data set, extracting data features according to the abnormal trajectory data detection results to establish a repair data set; Step 3, repairing the ETC missing transaction gantry path in the repair data set based on topological constraints; Specifically: Step 3-1, establishing a topological node table according to the ETC topology data; Step 3-2, establishing a gantry number matrix F through the adjacent front and rear two gantry numbers of the missing transaction gantry, then: Step 3-4, calculating the list intersection of to_id_arr and from_id_arr to uniquely obtain a gantry number as the missing transaction gantry number; F = [F FOR F AFT ] (10) where F FOR Fref refers to the matrix of gate numbers established by the two adjacent gates in front of the missed transaction gate; F AFT Fref refers to the matrix of gate numbers established by the two adjacent gates in front of the missed transaction gate; Step 3-3, Topology table is aggregated twice with portal number matrix standard, respectively F FOR Node post-portal list to_id_arr of matrix and F AFT Node pre-portal list from_id_arr of matrix Step 4, obtaining the vehicle speed characteristics of the corresponding scene according to different scenes of highway traffic flow to repair the time of highway ETC missing transaction data; Specifically: When in the non-saturated free flow state, the transaction time of the missing transaction gantry is repaired by referring to the average driving speed of the reference section, and the formula is: When in the over-saturated congestion state, the transaction time of the missing transaction gantry is repaired by referring to the speed characteristics of the same vehicle; That is: T LOSE = T FOR + D FOR ⊙ [(D FOR + D AFT ) ⊙ (T AFT - T FOR ) -1 ] -1 (15) Wherein, T FOR Indicates the transaction time of the previous gantry adjacent to the missing transaction gantry; T AFT Indicates the transaction time of the next gantry adjacent to the missing transaction gantry; D FOR Indicates the section distance of the previous section of the section where the missing transaction gantry is located; D AFT Indicates the section distance of the next section of the section where the missing transaction gantry is located; In step 1, the data in the service area section in the trajectory data is marked according to the ETC topology data set; and the weighted FCM algorithm is used to detect and mark whether the section where the missing transaction gantry belongs is in the congestion period. wherein, refers to the front vehicle of the leak transaction vehicle at the front gantry transaction time; D JAM refers to the segment distance of the congestion segment; v JAM refers to the average speed of the congestion segment; When in the service area parking state, based on the kinematic inertia principle method combined with ETC transaction data to calculate the transaction time of the missing transaction gantry in the service area parking state The formula is: wherein, refers to the first two gantry transaction times of the missing transaction gantry; refers to the first one gantry transaction time of the missing transaction gantry; refers to the segment distance from the first two gantries of the missing transaction gantry to the first one gantry of the missing transaction gantry; refers to the segment distance from the first one gantry of the missing transaction gantry to the missing transaction gantry; When the miss transaction is concurrent, the transaction time T' of the miss transaction gate is calculated based on the transaction time T of the miss transaction gate LOSE The transaction time T' of the miss transaction gate in the concurrent state is calculated The calculation formula is: wherein D LOSE Error distance between the previous gate of the missing transaction gate and the missing transaction gate; T FOR Transaction time of the previous gate of the missing transaction gate; D ERR Section distance of the section where the missing transaction gate is located.

2. The method for repairing missing transactions of expressway ETC based on multi-scene features according to claim 1, characterized in that: Step 2 specifically comprises the following steps: 3.The method of claim 1, wherein the method further comprises: Step 2-1, extracting the reference gantry number used to repair the missing transaction gantry path from the trajectory data, establishing a gantry number matrix F, then: Step 2-2, according to the aggregation results of the gantry number matrix F and the ETC topology data set, establishing a section distance matrix D, then: F m = [f 1m f 2m … f nm ] T (6) where F m represents the column of gantry numbers extracted from n vehicle trajectories, f nm represents the gantry number, and n represents the number of vehicle trajectories for batch processing; Step 2-3, using the same principle to extract a transaction time matrix T, then: D = [D 12 D 23 … D (m-1)m ], D (m-1)m = [d 1m d 2m … d nm ] T (7) where D (m-1)m is the distance of the section consisting of the two adjacent gantries numbered m and m-1. Step 2-4, establishing a data repair matrix R, then: T = [T1 T2... T m ], T m = [t 1m t 2m ... t nm ] T (8) where T m represents the transaction time column extracted from n vehicle trajectories according to the algorithm requirements; R=[F T D] (9) Wherein, F is the gantry number matrix, which will be used as the input of the path repair algorithm part; T is the gantry transaction time matrix, and D is the section distance matrix, T and D will be used as the input of the time repair algorithm part. In step 4, the highway traffic flow includes non-saturated free flow scene, over-saturated congestion scene, service area parking scene and missing transaction concurrent scene.

4. The method of claim 1, wherein the method is characterized by: When in the non-saturated free flow state, the calculation steps of the transaction time of the missing transaction gantry are as follows:

5. The method for repairing missing transactions of expressway ETC based on multi-scene features according to claim 1, characterized in that: Wherein, ⊙ is Hadamard product, which represents the product of the elements of two matrices; The transaction time T of the two adjacent gantries before and after the missing transaction gantry FOR and T AFT The time matrix T is constructed, and the section distance D of the two sections before and after the section where the missing transaction gantry is located FOR and D AFT The section distance matrix D is constructed, and then the data repair matrix R in the non-saturated free flow state is constructed NOM , in the form as follows: R NOM = [F FOR F LOSE F AFT T FOR T AFT D FOR D AFT ] (11) where F LOSE is the missing transaction gantry; F FOR is the matrix of gantry numbers established by the two gantries adjacent to the missing transaction gantry; F AFT is the matrix of gantry numbers established by the two gantries adjacent to the missing transaction gantry; Obtain time matrix T and section distance matrix D, calculate the passing speed V of the previous gate adjacent to the missing transaction gate FOR That is: where V ALL is the average velocity between the two adjacent gantries preceding the leaky one, According to the aggregation result of the gantry number block matrix and the ETC topological data table, the section distance D of the section is obtained FOR The passage time AT between the two adjacent gantries before the missing transaction gantry is calculated, and the corresponding calculation formula is as follows: ΔT = D FOR ⊙V FOR -1 (13) That is, the transaction time of the missing transaction gantry in the traffic free flow scene is obtained by the following formula: Based on the transaction time T of the previous gantry adjacent to the missing transaction gantry FOR and the transit time ΔT between the previous two gantries adjacent to the missing transaction gantry LOSE The corresponding calculation formula is as follows: T LOSE = T FOR + ΔT (12) When in the over-saturated congestion state, the calculation steps of the transaction time of the missing transaction gantry are as follows: T LOSE = T FOR + D FOR ⊙ [(D FOR + D AFT )⊙ (T AFT - T FOR ) -1 ] -1 (15).

6. The method of claim 1, wherein the method is characterized by: Wherein, ⊙ is Hadamard product, which represents the product of the elements of two matrices; First, the transaction time of the missing transaction gantry in the front and rear vehicle trajectories of the missing transaction vehicle is extracted and the front gantry transaction time of the missing transaction gantry of the three vehicles and the section distance D of the congestion section JAM , a data repair matrix R under the condition of oversaturated congestion is established JAM Then: where F LOSE denotes the missing transaction gantry; F FOR denotes the gantry number matrix established by the adjacent first two gantries of the missing transaction gantry; the average speed V AVE characterizes the average speed V JAM then: According to the aggregation result of the gantry number block matrix and the ETC topological data table, the section distance D of the congestion section is obtained JAM , the passing time ΔT of the congestion section is calculated JAM , and the corresponding calculation formula is as follows: ΔT JAM = D JAM ⊙ V JAM -1 ​ Based on the missing transaction vehicle's front vehicle's front gantry transaction time and the congestion segment's travel time ΔT JAM Calculate the congestion segment missing transaction gantry transaction time That is:

7. The method of claim 1, wherein the method is characterized by: Transaction time of a missed transaction gantry when in a service area stop state The calculation steps are as follows: obtaining the gantry transaction times of the two gantries adjacent to the missed transaction gantry and and the segment distance matrix of the two gantries before the missed transaction gantry to the missed transaction gantry respectively and constructing the data repair matrix R in the service area stop state RSA then: wherein, represents the time of the first gate transaction of the missing gate transaction; represents the time of the previous gate transaction of the missing gate transaction;F LOSE refers to the missing gate transaction; represents the segment distance from the first gate transaction of the missing gate transaction to the previous gate of the missing gate transaction; represents the segment distance from the previous gate transaction of the missing gate transaction to the missing gate transaction; Computing the average speed of the section from the previous gantry to the previous gantry before the missing transaction gantry for the first two gantries to the missing transaction gantry and Computing the average speed of the section from the previous gantry to the missing transaction gantry for the first two gantries to the missing transaction gantry i.e. Calculate the travel time ΔT of the section in the service area stop state RSA The corresponding calculation formula is as follows: The transaction time of the missing transaction portal in the service area stop state is calculated The corresponding calculation formula is as follows: 8.The method of claim 1, wherein the method further comprises: determining whether the vehicle is in the first lane or the second lane based on the first lane information and the second lane information; and determining whether the vehicle is in the first lane or the second lane based on the first lane information and the second lane information. Transaction time when a miss occurs in a concurrent state The calculation steps are as follows: Using the geographic position information of gantries in the ETC topology data to calculate the error distance D between each pair of gantries LOSE ; Obtaining corresponding false transaction gantry transaction time T' LOSE , the front door of the missing transaction gantry transaction time T FOR And the section distance D ERR of the section where it is located. Establish the data repair matrix under the missing false transaction concurrent state, then R SAM = [F FOR F LOSE F′ LOSE T FOR T′ LOSE D LOSE D ERR ] (22) Based on the data repair matrix data under the concurrent state of the missing transaction, the transaction time of the missing transaction gantry is calculated The specific calculation formula is as follows:

Images