A truck parking location prediction method based on historical GPS data semantic trajectory

Abstract

The application discloses a truck parking position prediction method based on historical GPS data semantic trajectory, which comprises the following steps: according to the vehicle number and time sequence, the original GPS trajectory points are matched to the road to generate the vehicle driving trajectory sequence; the vehicle parking position trajectory sequence is generated through spectral clustering, and semantic processing is performed on the vehicle parking position trajectory sequence to generate the vehicle parking semantic trajectory sequence; the individual frequent parking semantic trajectory mode tree and the group frequent parking position trajectory mode tree are respectively established to generate the individual similar parking semantic trajectory vehicle set and the group similar parking position trajectory vehicle set; the target trajectory sequence for prediction is input, the weighted score of the parking trajectory sequence mode tree after path matching is calculated, the vehicle parking position prediction candidate set is obtained, and the maximum probability value is selected as the predicted vehicle parking point position. The application improves the accuracy and timeliness of the truck parking position prediction, and is beneficial to effectively reducing the transport capacity cost of the truck operation enterprise.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method for predicting truck parking locations based on semantic trajectories of historical GPS data. Background Technology

[0002] In recent years, with the rapid development of my country's logistics industry, truck operation platforms have become increasingly important. These platforms typically serve most freight vehicles nationwide, providing functions such as in-transit order allocation and management, route planning, advance cargo matching, and gas station recommendations. The prerequisite for realizing these functions is the efficient and accurate prediction of vehicle parking locations.

[0003] Currently, location prediction technologies primarily rely on geographic trajectory similarity techniques. This approach has limitations and low accuracy in long-term predictions, or it simply uses historical trajectory data to build a Markov model, considering only the impact of the current location on future locations. It's worth noting that there is currently no method to predict truck parking locations. Different truck drivers may have different parking route preferences, and for vehicles with the same freight attributes, their parking locations are often highly similar and clustered. Therefore, for massive amounts of truck driving trajectory data, there is an urgent need to propose a truck parking location prediction method that simultaneously considers individual parking semantic trajectories and group parking location trajectories, providing technical support for truck operating companies to effectively reduce transportation costs and improve the large-scale organization of small and medium-sized transport capacity. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a truck parking location prediction method based on historical GPS data semantic trajectory, which comprehensively considers individual similar parking semantic trajectories and group similar parking location trajectories, and can reasonably and accurately predict the parking location of trucks.

[0005] Technical solution: The truck parking location prediction method of the present invention includes the following steps:

[0006] S1. Obtain historical GPS trajectory data of trucks, decode and clean the historical GPS trajectory data of trucks to obtain the original vehicle trajectory dataset D1, divide the trip and compress the trajectory according to the vehicle number and its time series to obtain the vehicle driving trajectory sequence dataset D2 for each trip of each vehicle, including latitude and longitude, driving speed and the north direction angle.

[0007] S2. Perform spectral clustering on the vehicle driving trajectory sequence dataset D2 to obtain the vehicle parking location trajectory sequence dataset D3. Call the Openstreet API to obtain surrounding POI information. Associate the POIs with vehicle IDs and combine them with vehicle parking duration and interval, vehicle remaining fuel and vehicle mileage information to semantically label the vehicle parking points in the vehicle parking location trajectory sequence dataset D3 to obtain the parking semantic trajectory sequence dataset D4 corresponding to the vehicle parking point locations.

[0008] S3, perform frequent pattern mining on individual stop semantic trajectory sequences in the stop semantic sequence dataset D4 to obtain a set of vehicles with similar individual stop semantic trajectories D5; merge the stop location trajectories of similar vehicles within the same cluster of the set of vehicles with similar individual stop semantic trajectories D5, and mine frequent patterns on group stop location trajectories to obtain a set of vehicles with similar group stop location trajectories D6, completing the initial training of the model; obtain the group frequent stop location trajectory pattern tree Z and the individual frequent stop semantic trajectory sequence pattern tree Y;

[0009] S4. The target vehicle trajectory sequence T used for prediction is matched with the group frequent stop location trajectory pattern tree Z and the individual frequent stop semantic trajectory sequence pattern tree Y respectively. The weighted score of the stop trajectory sequence pattern tree is calculated. The intersection of the obtained vehicle stop path prediction candidate set D7 and the group similar stop location trajectory vehicle set D6 is obtained to obtain the vehicle stop location prediction candidate set D8. The Markov truck stop location prediction model is established, and the maximum probability is selected as the predicted vehicle stop point location.

[0010] Furthermore, in step S2, the specific implementation steps for spectral clustering of the vehicle driving trajectory sequence dataset D2 are as follows:

[0011] S21, the graph construction process of spectral clustering, measures the similarity S between the p-th and q-th travel trajectory sequences in the vehicle trajectory sequence dataset D2. pq When S pq When the similarity exceeds a set threshold, a similarity weight W is defined. pq =S pq The similarity weight W is obtained. pq The adjacency matrix W is formed in which the edge weight between two trajectory sequences that are far apart is low, while the edge weight between two trajectory sequences that are close together is high.

[0012] S22, Calculate based on similarity weights The degree matrix D is obtained as diag(d1, d2, ..., d Q ), thus obtaining the Laplace matrix L = DW;

[0013] S23, the spectral clustering process, includes calculating the feature matrix F based on the Laplacian matrix L, obtaining the target cluster through K-means clustering, and determining the vehicle stop location for each vehicle in each trip.

[0014] Furthermore, the specific implementation steps for obtaining the set D6 of vehicles with similar parking location trajectories in step S3 are as follows:

[0015] S31, using the PrefixSpan algorithm on the stop semantic trajectory sequence dataset D4, mine the frequent patterns of individual stop semantic trajectory sequences of the a-th vehicle on the b-th trip to obtain the set of individual stop semantic frequent trajectories D. 41 ;

[0016] S32, for the set of frequent trajectories D of individual docking semantics 41 Construct a prefix tree to obtain the individual frequent stop semantic trajectory sequence pattern tree Y for the a-th vehicle on the b-th trip. ab Repeat step S31 and use the PrefixSpan algorithm to mine the frequent patterns of the individual vehicle stopping semantic trajectory sequences, and finally obtain the individual frequent stopping semantic trajectory sequence pattern tree Y.

[0017] S33, cluster vehicles with similar parking location selection behavior preferences based on individual similar parking semantic trajectories to obtain the vehicle set D5 with individual similar parking semantic trajectories;

[0018] S34, merge the parking location trajectories of similar vehicles within the same cluster obtained by clustering from the individual similar parking semantic trajectory vehicle set D5, and use the PrefixSpan algorithm to mine frequent patterns of the group parking location trajectory sequence to obtain the group frequent parking location trajectory set D. 51 ;

[0019] S35, for the set of frequent trajectories of group stopping locations D 51 Construct a prefix tree to obtain the pattern tree Z of the trajectory sequence of the group's frequent stopping positions;

[0020] S36. Based on the similar parking location trajectories of the group, cluster vehicles with similar parking location locations to obtain the group set of vehicles with similar parking location trajectories D6.

[0021] Furthermore, in step S4, the specific steps for obtaining the predicted vehicle parking location by establishing the Markov truck parking location prediction model are as follows:

[0022] S41, perform path matching between the target vehicle trajectory sequence T used for prediction and the individual stop semantic trajectory pattern tree Y, and calculate the individual stop semantic trajectory sequence pattern tree score. semantic (T,Y);

[0023] S42, perform path matching between the target vehicle trajectory sequence T used for prediction and the group frequent stop location trajectory sequence pattern tree Z, and calculate the group stop location trajectory sequence pattern tree score. location (T,Z);

[0024] S43, calculate the weighted score Score(T) of the group docking location trajectory sequence pattern tree and the individual docking semantic trajectory sequence pattern tree, using the following formula:

[0025] Score(T) = α × Score location (T,Z)+(1-α)×Score semantic (T,Y)

[0026] In the formula:

[0027] α is the weight value, and satisfies 0 < α < 1;

[0028] Score semantic (T,Y) represents the individual docking semantic trajectory sequence pattern tree score after path matching;

[0029] Score location (T,Z) represents the pattern tree score of the group docking location trajectory sequence after path matching;

[0030] S44. Sort the candidate vehicle parking routes from largest to smallest according to the weighted score values ​​to obtain the candidate set D7.

[0031] S45, find the intersection of the set of vehicles with similar parking location trajectories D6 and the candidate set of vehicle parking paths D7 to obtain the candidate set of vehicle parking location prediction D8 = {L1, L2, ..., L...} N};

[0032] S46. Establish a Markov truck parking location prediction model for the candidate set D8 of vehicle parking location prediction, and select the maximum probability as the predicted next parking location of the vehicle.

[0033] Furthermore, the specific implementation of step S45 is as follows:

[0034] If the number of stops in the intersection of the candidate set of vehicle stopping routes D7 and the set of vehicles with similar stopping location trajectories D6 in a single trip is less than a threshold, then the stopping location is added to the candidate set of vehicle stopping location prediction D8.

[0035] If the number of stopping points in the intersection of the candidate set of vehicle stopping paths D7 and the set of similar stopping location trajectories of vehicles D6 in a single trip is greater than or equal to a threshold, then the data is re-input into the vehicle target trajectory sequence T used for prediction.

[0036] Furthermore, the specific implementation steps of step S46 are as follows:

[0037] S461, take the position of the first vehicle stop in the target vehicle trajectory sequence T as the initial state probability vector A(0);

[0038] S462, predict the candidate set D8 = {L1, L2, ..., L...} from the vehicle parking locations. N In the calculation, the distance from the docking position L is calculated. i To docking position L j State transition probability P ij The calculation formula is as follows:

[0039] P ij =P(E i →E j )=P(E j / E i )

[0040] In the formula:

[0041] E i Parking location for vehicles at L i The probability values, where i = 1, 2, ..., N;

[0042] E j Parking location for vehicles at L j The probability values, where j = 1, 2, ..., N;

[0043] S463, Construct the state transition probability matrix P:

[0044]

[0045] Where N1≥i, N2≥j;

[0046] S464, based on the initial state probability vector A(0) and the state transition probability matrix P, predicts the vehicle's stopping position as L. k The probability value A(k) j The calculation formula is as follows:

[0047]

[0048] In the formula:

[0049] A(0) is the initial state probability vector;

[0050] The vehicle parking location is L k j state transition probability values;

[0051] S465, Select The maximum value in the range is the prediction result.

[0052] Compared with the prior art, the significant advantages of this invention are as follows:

[0053] 1. This invention achieves trajectory compression, and when the compression ratio reaches 98% or more, the compressed trajectory data can still reflect the shape characteristics of the original trajectory. While preserving complete feature values, it can reduce the amount of computation and save computation time and space.

[0054] 2. This invention determines the location of vehicle stops through spectral clustering, which is easy to implement on a computer. At the same time, it avoids the shortcomings of clustering algorithms such as DBSCAN, which cannot consider the overall shape features of the trajectory and are difficult to effectively extract the entire trajectory.

[0055] 3. This invention mines frequent patterns of individual parking semantic trajectories and frequent patterns of group parking location trajectories, respectively. It performs path matching between the target trajectory sequence for prediction and the trajectory sequence pattern tree. By finding the intersection of the set of vehicles with similar parking location trajectories and the candidate set of vehicle parking paths, a candidate set of vehicle parking location predictions is obtained. A Markov model is then established to accurately and efficiently predict vehicle parking locations, providing technical support for truck operating companies to effectively reduce transportation costs and improve the large-scale organization of small and medium-sized transportation capacity. Attached Figure Description

[0056] Figure 1 This is a flowchart of the present invention;

[0057] Figure 2 This is a schematic diagram of the truck parking trajectory sequence;

[0058] Figure 3 (a) is a schematic diagram of the semantic trajectory pattern tree of an individual docking, and (b) is a schematic diagram of the trajectory pattern tree of the docking location of a group. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0060] like Figure 1 As shown, the method for predicting truck parking locations based on historical GPS data semantic trajectories according to the present invention includes the following steps:

[0061] Step 1: Obtain historical GPS trajectory data of trucks, including vehicle number, latitude and longitude, collection time, driving speed, north-direction angle, mileage and fuel level, etc. Decode and clean the historical GPS trajectory data to obtain the original vehicle trajectory dataset D1. Based on the vehicle number and its time series, divide the journey and compress the trajectory to obtain the vehicle driving trajectory sequence dataset D2 for each vehicle, including latitude and longitude, driving speed and north-direction angle.

[0062] Step 2: Perform spectral clustering on the vehicle driving trajectory sequence dataset D2 to obtain the vehicle parking location trajectory sequence dataset D3. Call the OpenStreet API to obtain surrounding POI (Point of Information) information. Associate the POIs with vehicle IDs, and combine this with vehicle parking duration and intervals, remaining fuel, and mileage information to semantically label the vehicle parking points in the vehicle parking location trajectory sequence dataset D3, resulting in the parking semantic trajectory sequence dataset D4 corresponding to each vehicle parking point location. The implementation process is as follows:

[0063] Step 21, the graph construction process of spectral clustering. Measure the similarity S between the p-th and q-th travel trajectory sequences in the vehicle trajectory sequence dataset D2. pq When S pq When the similarity exceeds a set threshold, a similarity weight W is defined. pq =S pq The similarity weight W is obtained. pq The adjacency matrix W is formed in which the edge weights between two trajectory sequences that are far apart are lower, while the edge weights between two trajectory sequences that are close together are higher.

[0064] Step 22: Calculate based on similarity weights. The degree matrix D is obtained:

[0065] D = diag(d1, d2, ..., d Q (1)

[0066] This leads to the Laplace matrix:

[0067] L = DW (2)

[0068] Step 23, the spectral clustering process. Based on the Laplacian matrix L, the feature matrix F is calculated. K-means clustering is then used to obtain the target cluster, determining the vehicle stop location for each vehicle in each trip.

[0069] Step 3: Considering the transportation characteristics, driving characteristics, external environment characteristics, time characteristics, driver attributes, and traffic regulations of trucks, perform frequent pattern mining on the individual stopping semantic trajectory sequence dataset D4 to obtain the set of vehicles with similar individual stopping semantic trajectories D5. Merge the stopping position trajectories of similar vehicles in the same cluster of the set of vehicles with similar individual stopping semantic trajectories D5, and mine the frequent patterns of group stopping position trajectories to obtain the set of vehicles with similar group stopping position trajectories D6. The initial training of the model is completed. The implementation process is as follows:

[0070] Step 31: Use the PrefixSpan algorithm to mine the frequent patterns of individual stop semantic trajectory sequences of vehicle a in the b-th trip on the stop semantic trajectory sequence dataset D4, to obtain the set of frequent individual stop semantic trajectories D. 41 .

[0071] Step 32, for the set D of frequent individual docking semantic trajectories 41 Construct a prefix tree to obtain the individual frequent stop semantic trajectory sequence pattern tree Y for the a-th vehicle on the b-th trip. ab Repeat step 31 and use the PrefixSpan algorithm to mine the frequent patterns of the individual vehicle stopping semantic trajectory sequences, and finally obtain the individual frequent stopping semantic trajectory sequence pattern tree Y.

[0072] Step 33: Based on the individual similar parking semantic trajectories, cluster vehicles with similar parking location selection behavior preferences to obtain the individual similar parking semantic trajectory vehicle set D5.

[0073] Step 34: Merge the parking location trajectories of similar vehicles within the same cluster obtained by clustering in the individual similar parking semantic trajectory vehicle set D5, and use the PrefixSpan algorithm to mine frequent patterns in the group parking location trajectory sequence to obtain the group frequent parking location trajectory set D. 51 .

[0074] Step 35, for the set of frequent trajectories D of the group's stopping locations 51 Construct a prefix tree to obtain the pattern tree Z of the trajectory sequence of the group's frequent stopping positions.

[0075] Step 36: Cluster vehicles with similar parking location locations based on the similar parking location trajectories of the group to obtain the group set of vehicles with similar parking location trajectories D6.

[0076] Step 4, as follows Figure 3 As shown in Figures (a) and (b), the target vehicle trajectory sequence T used for prediction is path matched with the group's frequent stop location trajectory pattern tree Z and the individual's frequent stop semantic trajectory sequence pattern tree Y, respectively. The weighted score of the stop trajectory sequence pattern tree is calculated to obtain the vehicle stop path prediction candidate set D7. The intersection of the vehicle stop path prediction candidate set D7 and the group's similar stop location trajectory vehicle set D6 is obtained to obtain the vehicle stop location prediction candidate set D8. A Markov truck stop location prediction model is established, and the vehicle stop point location with the maximum probability is selected. The implementation process is as follows:

[0077] Step 41: Perform path matching between the target vehicle trajectory sequence T used for prediction and the individual stop semantic trajectory pattern tree Y, and calculate the individual stop semantic trajectory sequence pattern tree score. semantic(T,Y).

[0078] Step 42: Perform path matching between the target vehicle trajectory sequence T used for prediction and the group's frequent stop location trajectory sequence pattern tree Z, and calculate the group stop location trajectory sequence pattern tree score. location (T,Z).

[0079] Step 43: Calculate the weighted score Score(T) of the group docking location trajectory sequence pattern tree and the individual docking semantic trajectory sequence pattern tree. The calculation formula is as follows:

[0080] Score(T) = α × Score location (T,Z)+(1-α)×Score semantic (T,Y) (3)

[0081] In the formula:

[0082] α is the weight value, and satisfies 0 < α < 1;

[0083] Score semantic (T,Y) represents the individual docking semantic trajectory sequence pattern tree score after path matching;

[0084] Score location (T,Z) represents the pattern tree score of the group docking location trajectory sequence after path matching.

[0085] Step 44: Sort the candidate vehicle parking routes from largest to smallest according to the weighted score values ​​to obtain the vehicle parking route candidate set D7.

[0086] Step 45: Find the intersection of the set of vehicles with similar parking location trajectories D6 and the candidate set of vehicle parking paths D7 to obtain the candidate set of vehicle parking location prediction D8 = {L1, L2, ..., L...} N},like Figure 2 As shown, specifically:

[0087] (1) If the number of stops in the intersection of the candidate set of vehicle stopping paths D7 and the set of vehicles with similar stopping location trajectories D6 in a trip is less than the threshold, then the stopping location is added to the candidate set of vehicle stopping location prediction D8.

[0088] (2) If the number of stops in the intersection of the candidate set of vehicle stopping paths D7 and the set of similar stopping location trajectories of vehicles D6 in a trip is greater than or equal to the threshold, then the vehicle target trajectory sequence used for prediction is re-inputted and the algorithm is re-entered.

[0089] Step 46: Establish a Markov truck parking location prediction model for the candidate set D8 of vehicle parking location predictions, and select the location with the highest probability as the predicted next parking location of the vehicle. The implementation process is as follows:

[0090] Step 461: Take the position of the first vehicle stop in the target vehicle trajectory sequence T as the initial state probability vector A(0).

[0091] Step 462, predict the candidate set D8 = {L1, L2, ..., L...} from the vehicle parking locations. N In the calculation, the distance from the docking position L is calculated. i To docking position L j State transition probability P ij The calculation formula is as follows:

[0092] P ij =P(E i →E j )=P(E j / E i (4)

[0093] In the formula:

[0094] E i Parking location for vehicles at L i The probability values, where i = 1, 2, ..., N;

[0095] E j Parking location for vehicles at L j The probability value of , where j = 1, 2, ..., N.

[0096] Step 463, construct the state transition probability matrix P:

[0097]

[0098] Where N1≥i, N2≥j.

[0099] Step 464: Based on the initial state probability vector A(0) and the state transition probability matrix P, predict the vehicle's parking position as L. k The probability value A(k) j The calculation formula is as follows:

[0100]

[0101] In the formula:

[0102] A(0) is the initial state probability vector;

[0103] The vehicle parking location is L k The j state transition probability values.

[0104] Step 465, select The maximum value in the range is the prediction result.

[0105] As can be seen from the above embodiments, the present invention establishes a Markov model based on historical GPS data and its semantic trajectory, taking into account both individual similar parking semantic trajectories and group similar parking trajectories, which can accurately predict the parking location of trucks.

[0106] The embodiments in this specification focus on the differences from other embodiments. Similar or identical parts between the embodiments can be referred to each other. The operation of the embodiments of this invention is simpler to describe since it corresponds to the methods disclosed in the embodiments; relevant parts can be referred to the method section.

[0107] The above description is only a preferred embodiment of the present invention. The present invention is not limited to the embodiments shown herein. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting truck parking locations based on historical GPS data semantic trajectories, characterized in that, The steps include the following: S1. Obtain historical GPS trajectory data of trucks, decode and clean the historical GPS trajectory data of trucks to obtain the original vehicle trajectory dataset D1, divide the trip and compress the trajectory according to the vehicle number and its time series to obtain the vehicle driving trajectory sequence dataset D2 for each trip of each vehicle, including latitude and longitude, driving speed and the north direction angle. S2. Perform spectral clustering on the vehicle driving trajectory sequence dataset D2 to obtain the vehicle parking location trajectory sequence dataset D3. Call the Openstreet API to obtain surrounding POI information. Associate the POIs with vehicle IDs and combine them with vehicle parking duration and interval, vehicle remaining fuel and vehicle mileage information to semantically label the vehicle parking points in the vehicle parking location trajectory sequence dataset D3 to obtain the parking semantic trajectory sequence dataset D4 corresponding to the vehicle parking point locations. S3. Perform frequent pattern mining on individual parking semantic trajectory sequences in the parking semantic trajectory sequence dataset D4 to obtain the set of vehicles with similar parking semantic trajectories D5. Merge the parking location trajectories of similar vehicles within the same cluster of the set of vehicles with similar parking semantic trajectories D5, and mine the frequent patterns of group parking location trajectories to obtain the set of vehicles with similar group parking location trajectories D6, thus completing the initial training of the model. Obtain the trajectory pattern tree of the group's frequent docking locations and individual frequent docking semantic trajectory sequence pattern tree ; S4 will be the target vehicle trajectory sequence used for prediction. Trajectory pattern tree of frequent stops of the group and individual frequent docking semantic trajectory sequence pattern tree Perform path matching, calculate the weighted score of the stop trajectory sequence pattern tree, find the intersection of the obtained vehicle stop path prediction candidate set D7 with the group similar stop location trajectory vehicle set D6 to obtain the vehicle stop location prediction candidate set D8, establish the Markov truck stop location prediction model, and select the maximum probability as the predicted vehicle stop point location. In step S2, the specific implementation steps for spectral clustering of the vehicle driving trajectory sequence dataset D2 are as follows: S21, the graph construction process of spectral clustering, measures the number of nodes in the vehicle trajectory sequence dataset D2. Second and third Similarity between sub-trajectory sequences ,when When the similarity exceeds a set threshold, a similarity weight is defined. The similarity weights are obtained. The adjacency matrix W is formed in which the edge weight between two trajectory sequences that are far apart is low, while the edge weight between two trajectory sequences that are close together is high. S22, Calculate based on similarity weights The degree matrix is ​​obtained. This leads to the Laplace matrix L=DW; S23, the spectral clustering process, includes calculating the feature matrix F based on the Laplacian matrix L, obtaining the target cluster through K-means clustering, and determining the vehicle stop location for each vehicle in each trip; In step S4, the specific steps for obtaining the predicted vehicle parking location by establishing the Markov truck parking location prediction model are as follows: S41, the target vehicle trajectory sequence used for prediction Individual docking semantic trajectory pattern tree Perform path matching and calculate the individual stop semantic trajectory sequence pattern tree score. ; S42, the target vehicle trajectory sequence used for prediction Pattern tree of frequent docking location trajectories of the group Perform path matching and calculate the pattern tree score of the group's docking location trajectory sequence. ; S43, Calculate the weighted score of the group docking location trajectory sequence pattern tree and the individual docking semantic trajectory sequence pattern tree. The calculation formula is as follows: = + ， In the formula: Let be the weight value, and satisfy... ; The individual docking semantic trajectory sequence pattern tree score is generated after path matching. The pattern tree score is calculated for the group docking location trajectory sequence after path matching. S44. Sort the candidate vehicle parking routes from largest to smallest according to the weighted score values ​​to obtain the vehicle parking route candidate set D7. S45. Find the intersection of the set of vehicles with similar parking location trajectories D6 and the candidate set of vehicle parking paths D7 to obtain the candidate set of vehicle parking location predictions. , This is the Nth parking position; S46. Establish a Markov truck parking location prediction model for the candidate set D8 of vehicle parking location prediction, and select the maximum probability as the predicted next parking location of the vehicle. The specific implementation of step S45 is as follows: If the number of stops in the intersection of the candidate set of vehicle stopping routes D7 and the set of vehicles with similar stopping location trajectories D6 in a single trip is less than a threshold, then the stopping location is added to the candidate set of vehicle stopping location prediction D8. If the number of stop points in the intersection of the candidate set of vehicle stop routes D7 and the set of vehicles with similar stop location trajectories D6 in a single trip is greater than or equal to a threshold, then the data is re-input into the vehicle target trajectory sequence used for prediction. .

2. The method for predicting truck parking locations based on historical GPS data semantic trajectories according to claim 1, characterized in that, The specific implementation steps for obtaining the set D6 of vehicles with similar parking location trajectories in step S3 are as follows: S31, Apply the PrefixSpan algorithm to the docking semantic trajectory sequence dataset D4, and then... The vehicle number By mining the frequent patterns of individual stop semantic trajectory sequences in the next trip, a set D of frequent individual stop semantic trajectories is obtained. 41 ; S32, for the set of frequent trajectories D of individual docking semantics 41 Build a prefix tree to obtain the first... The vehicle number Semantic trajectory sequence pattern tree of frequent stops of individuals in the next journey Repeat step S31 and use the PrefixSpan algorithm to mine the frequent patterns of the individual vehicle stopping semantic trajectory sequences, and finally obtain the individual frequent stopping semantic trajectory sequence pattern tree Y. S33, cluster vehicles with similar parking location selection behavior preferences based on individual similar parking semantic trajectories to obtain the vehicle set D5 with individual similar parking semantic trajectories; S34, merge the parking location trajectories of similar vehicles within the same cluster obtained by clustering from the individual similar parking semantic trajectory vehicle set D5, and use the PrefixSpan algorithm to mine frequent patterns of the group parking location trajectory sequence to obtain the group frequent parking location trajectory set D. 51 ; S35, for the set of frequent trajectories at group stopping locations D 51 Construct a prefix tree to obtain the pattern tree of the trajectory sequence of the group's frequent stops. ; S36. Based on the similar parking location trajectories of the group, cluster vehicles with similar parking location locations to obtain the group set of vehicles with similar parking location trajectories D6.

3. The method for predicting truck parking locations based on historical GPS data semantic trajectories according to claim 1, characterized in that, The specific implementation steps of step S46 are as follows: S461, using the target vehicle trajectory sequence The position of the first vehicle stop is the initial state probability vector. ; S462, Predicting candidate set from vehicle parking locations In the middle, calculate from the docking position Arrive at the parking location State transition probability The calculation formula is as follows: ， In the formula: For the vehicle parking location The probability value, where, ; For the vehicle parking location The probability value, where, ; S463, Construct the state transition probability matrix : ， in, ; S464, based on the initial state probability vector and state transition probability matrix The predicted parking location is probability value The calculation formula is as follows: ， In the formula: This is the initial state probability vector; For vehicle parking locations of Each state transition probability value; S465, Select The maximum value in the range is the prediction result.

Images