CN118013464B

CN118013464B - Geographic traffic data processing method and device and electronic equipment

Info

Publication number: CN118013464B
Application number: CN202410414256.0A
Authority: CN
Inventors: 方灵
Original assignee: Shanghai Hummingbird Instant Information Technology Co ltd; Zhejiang Koubei Network Technology Co Ltd
Current assignee: Shanghai Hummingbird Instant Information Technology Co ltd; Zhejiang Koubei Network Technology Co Ltd
Priority date: 2024-04-07
Filing date: 2024-04-07
Publication date: 2024-07-23
Anticipated expiration: 2044-04-07
Also published as: CN118013464A

Abstract

The embodiment of the invention discloses a geographic traffic data processing method, a geographic traffic data processing device and electronic equipment. The embodiment of the invention acquires the position characteristic information and the path characteristic information of a plurality of position path pairs by acquiring a position path pair data set comprising a plurality of pairs of position information and path information with corresponding relation, and carrying out corresponding characteristic coding on the position path pair data set according to a pre-trained position coding model and a road section coding model, and carrying out multi-mode training according to the position characteristic information and the path characteristic information of each position path pair to acquire a geographic traffic data processing model. Therefore, the embodiment can realize the processing of multiple downstream tasks through the multi-mode training geographic traffic data processing model, and uniformly express the position features and the path features in the geographic traffic data through the position coding model, the road section coding model and the geographic traffic data processing model, so that the effect and the performance of the multiple downstream tasks are improved, and the resource utilization rate is improved.

Description

Geographic traffic data processing method and device and electronic equipment

Technical Field

The present invention relates to the field of computer technologies, and in particular, to a method and an apparatus for processing geographic traffic data, and an electronic device.

Background

In the application scene of geographic traffic, the downstream tasks such as navigation, arrival time estimation, route determination and the like all need to process information such as geographic position, road network data and the like. For example, in an instant distribution scenario, navigation distance may be used to determine business turn divisions, generate distribution ranges for merchants, calculate arrival times, distribution fees, schedule shares, etc. based on navigation distance at task creation. In order to improve the performance of the downstream task, various algorithm models are often adopted in the prior art, and the accuracy of the independent algorithm models is not enough, so that unification is difficult to achieve in effect, and resource waste is caused.

Disclosure of Invention

In view of the above, embodiments of the present invention provide a method, an apparatus, and an electronic device for processing geographic traffic data, so as to implement processing of multiple downstream tasks by training a geographic traffic data processing model in multiple modes, and uniformly express location features and path features in geographic traffic data by using a location coding model, a road section coding model, and a geographic traffic data processing model, thereby improving effects and performances of multiple downstream tasks, and improving resource utilization.

In a first aspect, an embodiment of the present invention provides a method for processing geographic traffic data, where the method includes:

Acquiring a position path pair data set, wherein the position path pair data set comprises a plurality of pairs of position information and path information with corresponding relations;

Performing corresponding feature coding on the position path pair data set according to a pre-trained position coding model and a road section coding model to acquire position feature information and path feature information of a plurality of position path pairs;

And carrying out multi-mode training according to the position characteristic information and the path characteristic information of each position path pair to obtain a geographic traffic data processing model.

Optionally, the performing multi-mode training according to the location feature information and the path feature information of each location path pair, and obtaining the geographic traffic data processing model includes:

and pre-training by adopting a multi-mode geographic information mask and a position path matching method according to the position characteristic information and the path characteristic information of each position path pair after characteristic alignment to obtain the geographic traffic data processing model.

Optionally, the method further comprises:

acquiring a specific task data set corresponding to at least one specific task respectively;

and performing supervised training on parameters of the geographic traffic data processing model according to the at least one specific task data set.

Optionally, the at least one specific task includes a distance task, and the specific task data set includes a plurality of pairs of location information and distance information having a correspondence relationship.

Optionally, the at least one specific task includes a time determination task, and the specific task data set includes a plurality of pairs of location information and time information having a correspondence relationship.

Optionally, the at least one specific task includes a route scoring task, and the specific task data set includes a plurality of pairs of location information and route information having a correspondence relationship.

Optionally, the at least one specific task includes a road tag mining task, and the specific task data set includes a plurality of pairs of route information and tag information having a correspondence relationship.

Optionally, the training step of the position coding model includes:

acquiring a position data set;

Acquiring a characteristic information sequence of a position object in the position data set, wherein the characteristic information sequence is determined based on geographic information objects in a preset range of the position object;

and training and acquiring the position coding model according to each characteristic information sequence.

Optionally, the acquiring the characteristic information sequence of the position object in the position data set includes:

acquiring characteristic information of a geographic information object of the position object within a preset distance range;

and sorting the characteristic information of each geographic information object based on a preset condition to obtain a characteristic information sequence of the position object.

Optionally, the predetermined condition includes a distance between each of the geographic information objects and a corresponding location object.

Optionally, the characteristic information of the geographic information object includes one or more of the following: object identification, object shape, object position, relationship to a position object, position information of a relative position object.

Optionally, the geographic information object includes an AOI object, a POI object, and/or a road network object within a predetermined range of the corresponding task object.

Optionally, the training to obtain the position coding model according to each characteristic information sequence includes:

and pre-training by adopting a geographic contrast learning method according to each characteristic information sequence to obtain the position coding model.

Optionally, the training to obtain the position coding model according to each characteristic information sequence further includes:

acquiring a position road grabbing pair data set, wherein the position road grabbing pair data set comprises a plurality of pairs of position features and road grabbing information with corresponding relations;

and performing supervised training on the pre-trained position coding model according to the position road grabbing pair data set to obtain the position coding model.

Optionally, the training step of the road segment coding model includes:

acquiring a path sequence data set, wherein the path sequence data set comprises a plurality of path sequences formed by at least one road section;

and pre-training by adopting a masking training method according to the path sequence data set to obtain the road section coding model.

Optionally, the method further comprises:

Inputting target task information into a geographic traffic data processing architecture for processing, and obtaining a corresponding task result, wherein the target task information comprises position information and/or path information, and the task result comprises one or more of the following combinations: distance information, time prediction information, route information and road label information corresponding to the target task;

the geographic traffic data processing architecture comprises a position coding model, a road section coding model and a geographic traffic data processing model.

In a second aspect, an embodiment of the present invention provides a geographic traffic data processing method, where the method includes:

acquiring target task information, wherein the target task information comprises position information and/or path information;

inputting the target task information into a pre-trained geographic traffic data processing architecture for processing, and obtaining a corresponding task result;

The geographic traffic data processing architecture comprises a position coding model, a road section coding model and a geographic traffic data processing model, wherein the position coding model is used for carrying out feature coding on position information, the road section coding model is used for carrying out feature coding on route information, the geographic traffic data processing model carries out multi-mode training on a data set based on a position path, the position coding model and the road section coding model, and the position path pair data set comprises a plurality of pairs of position information and path information with corresponding relations.

In a third aspect, an embodiment of the present invention provides a geographic traffic data processing apparatus, the apparatus including:

A data set acquisition unit configured to acquire a position path pair data set including a plurality of pairs of position information and path information having a correspondence relationship;

The feature coding unit is configured to perform corresponding feature coding on the position path pair data set according to a pre-trained position coding model and a road segment coding model, and obtain position feature information and path feature information of a plurality of position path pairs;

The training unit is configured to perform multi-mode training according to the position characteristic information and the path characteristic information of each position path pair, and obtain a geographic traffic data processing model.

In a fourth aspect, an embodiment of the present invention provides a geographic traffic data processing device, including:

An information acquisition unit configured to acquire target task information including position information and/or path information;

the processing unit is configured to input the target task information into a pre-trained geographic traffic data processing architecture for processing, and obtain a corresponding task result;

In a fifth aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, where the memory is configured to store one or more computer program instructions, where the one or more computer program instructions are executed by the processor to implement a method according to the first aspect or the second aspect of the embodiment of the present invention.

In a sixth aspect, embodiments of the present invention provide a computer readable storage medium having a computer program stored therein, which when executed by a processor implements a method according to the first or second aspect of embodiments of the present invention.

In a seventh aspect, embodiments of the present invention provide a computer program product which, when run on a computer, causes the computer to perform the method according to the first or second aspect of the embodiments of the present invention.

According to the embodiment of the invention, the position characteristic information and the path characteristic information of a plurality of position path pairs are obtained by obtaining the position path pair data set comprising the position information and the path information which have the corresponding relation, and corresponding characteristic coding is carried out on the position path pair data set according to the position coding model and the road section coding model which are trained in advance, and the multi-mode training is carried out according to the position characteristic information and the path characteristic information of each position path pair, so that the geographic traffic data processing model is obtained. Therefore, the embodiment can realize the processing of multiple downstream tasks through the multi-mode training geographic traffic data processing model, and uniformly express the position features and the path features in the geographic traffic data through the position coding model, the road section coding model and the geographic traffic data processing model, so that the effect and the performance of the multiple downstream tasks are improved, and the resource utilization rate is improved.

Drawings

The above and other objects, features and advantages of the present invention will become more apparent from the following description of embodiments of the present invention with reference to the accompanying drawings, in which:

FIG. 1 is a schematic diagram of a geographic traffic data processing architecture according to an embodiment of the present invention;

FIG. 2 is a flow chart of a training method of a position-coding model according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a determination process of a feature information sequence of a position object according to an embodiment of the present invention;

FIG. 4 is a flow chart of a method of fine tuning a position-coding model according to an embodiment of the present invention;

FIG. 5 is a flow chart of a training method of a segment coding model according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a training process of a segment coding model according to an embodiment of the present invention;

FIG. 7 is a flow chart of a method of geographic traffic data processing according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of a training process for a geographic traffic data processing model in accordance with an embodiment of the present invention;

FIG. 9 is a flow chart of a method of fine tuning a geographic traffic data processing model in accordance with an embodiment of the present invention;

FIG. 10 is a schematic diagram of a process for fine tuning a geographic traffic data processing model according to an embodiment of the present invention;

FIG. 11 is a flow chart of another method of geographic traffic data processing according to an embodiment of the present invention;

FIG. 12 is a schematic diagram of a geographic traffic data processing device according to an embodiment of the present invention;

FIG. 13 is a schematic diagram of another geographic traffic data processing device according to an embodiment of the present invention;

Fig. 14 is a schematic diagram of an electronic device according to an embodiment of the invention.

Detailed Description

The present application is described below based on examples, but the present application is not limited to only these examples. In the following detailed description of the present application, certain specific details are set forth in detail. The present application will be fully understood by those skilled in the art without the details described herein. Well-known methods, procedures, flows, components and circuits have not been described in detail so as not to obscure the nature of the application.

Moreover, those of ordinary skill in the art will appreciate that the drawings are provided herein for illustrative purposes and that the drawings are not necessarily drawn to scale.

Unless the context clearly requires otherwise, the words "comprise," "comprising," and the like throughout the application are to be construed as including but not being exclusive or exhaustive; that is, it is the meaning of "including but not limited to".

In the description of the present application, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. Furthermore, in the description of the present application, unless otherwise indicated, the meaning of "a plurality" is two or more.

The schemes described in the present specification and embodiments, if related to personal information processing, all perform processing on the premise of having a legal basis (for example, obtaining the authorized consent of the personal information body and the public body, or being necessary for executing a contract, etc.), and perform processing only within a prescribed or agreed range. The user refuses to process the personal information except the necessary information of the basic function, and the basic function is not influenced by the user.

FIG. 1 is a schematic diagram of a geographic traffic data processing architecture according to an embodiment of the present invention. In the embodiment of the invention, the position coding model, the road section coding model and the geographic traffic data processing model can be trained through various geographic traffic information in a specific geographic traffic application scene, so that the unified expression of the characteristics of various information in geographic traffic can be realized through the position coding model, the road section coding model and the geographic traffic data processing model, namely, various geographic traffic information is coded into a unified characteristic space, various downstream traffic tasks (such as downstream tasks of route planning, distance estimation, time estimation and the like) are realized, the effect and the performance of the various downstream traffic tasks are improved, and meanwhile, the various models in the data processing architecture of the embodiment are not independent algorithms, and compared with the prior art, the various independent algorithm models are adopted for respectively realizing the downstream traffic tasks, so that the resource waste is reduced, and the resource utilization rate is improved.

Specifically, as shown in fig. 1, the embodiment of the invention can encode various geographic traffic information (such as position data, road and route data, AOI and POI data) and the like into a unified feature space through a pre-training position coding model, a road section coding model and a geographic traffic data processing model so as to realize the processing of downstream traffic tasks.

In an optional implementation manner, the embodiment of the invention can perform supervised fine adjustment on the pre-training model based on historical downstream task data of various downstream traffic tasks, further improves the task processing effect of the various downstream traffic tasks, and simultaneously reduces the task training cost.

Alternatively, taking the instant distribution field as an example, the location data may include a static user location, a merchant location, and may also include a dynamic distribution resource location and status. The road and route data may include attribute information on the road (e.g., name, status, lane attributes, etc.), historically employed route data, and the like. AOI (Area of Interest) represents a geographic Area on a map with a range or boundary, such as a business district, industrial park, etc. The POI (Point of Interest, points of interest) characterizes locations with punctual entities, such as shops, bus stops, etc. It should be understood that the specific content of the above various data may obtain corresponding traffic geographic data according to different application scenarios, and the embodiment of the present invention is not limited to the instant distribution field, and other fields, for example, geographic traffic related application fields such as network bus, express delivery, etc. may all adopt the geographic traffic data processing architecture of the embodiment of the present invention.

In an alternative implementation, the position-coding model is pre-trained based on a position dataset, which may encode the geographic position into corresponding feature vectors. Alternatively, the position coding model of the present embodiment may use an absolute position coding, a relative position coding, or a position mask coding mode, and further may use a converter (transducer) original position coding, a learnable position coding mode, a coding mode based on a convolutional neural network, a model with a relative position coding introduced by a transducer-XL, or the like, a recursive position coding, a periodic position coding, or a dynamic position coding as a basic model to perform model training.

In another alternative implementation, the position-coding model of the present embodiment implements model pre-training based on geo-contrast learning.

FIG. 2 is a flow chart of a training method of a position-coding model according to an embodiment of the present invention. Optionally, as shown in fig. 2, the training method of the position coding model of the present embodiment includes the following steps:

Step S110, a location data set is acquired.

In an alternative implementation, the location data set may include information of any location of the corresponding region. Further alternatively, the information of the locations in the location dataset may be determined based on the corresponding application domain related locations. For example, in an instant distribution application scenario, the location data set may be determined based on merchant location, distribution resource stay location (e.g., location where stay frequency is greater than a frequency threshold or stay time is greater than a time threshold, etc.), and/or other location (e.g., a scratch pad location, etc.). For another example, in a network taxi application scenario, the location data set may be determined based on a get-on/off point within each area, a start/end point location where the amount of tasks is greater than the amount of tasks threshold, and/or a vehicle stay location, etc.

In an alternative implementation, the location in the location dataset may be determined based on information such as a location, a track, etc. in a historical task, or may be determined by location information in a road network, which is not limited by the present embodiment.

Step S120, a feature information sequence of the position object in the position data set is acquired. Wherein the sequence of characteristic information is determined based on geographic information objects within a predetermined range of the location object.

Alternatively, the geographic information object may be an AOI, POI, and/or road network object within a predetermined range of the corresponding location object. Wherein the road network object is used for characterizing an object representing a road segment or a combination of road segments in the road network, which may include a path object, a road segment object, etc. Alternatively, the embodiment does not limit the division manner and the range size of the predetermined range, and may be set according to a specific application scenario. For example, in this embodiment, a circle may be drawn with a predetermined distance as a radius around the position object to determine a predetermined range, or an entire polygon may be drawn with the position object as the center to determine a predetermined range.

In an alternative implementation manner, the embodiment obtains the feature information sequence of the position object by acquiring the feature information of the geographic information objects of the position object within the predetermined distance range and sorting the feature information of each geographic information object based on the predetermined condition. Optionally, the predetermined condition includes a distance between each of the geographic information objects and a corresponding location object. For example, geographic information objects g1, g2, and g3 exist within a predetermined distance range of the position object P1, and feature information of the geographic information objects g1, g2, and g3 is f1, f2, and f3, respectively. The distances from the geographic information objects g1, g2 and g3 to the position object P1 are d1, d2 and d3, respectively, and if d2> d3> d1, the characteristic information sequence of the position object P1 is { f1, f3, f2}.

In other alternative implementations, the present embodiment may also use the category of the geographic information object and the distance from the location object as the predetermined condition. Specifically, the geographic information objects can be classified, the geographic information objects in the same category are ranked according to distance, and each category of geographic information objects is ranked based on a predetermined category sequence to obtain a corresponding characteristic information sequence. For example, the category of the geographic information objects present within the predetermined distance range of the position object P1' includes C1 and C2, wherein the geographic information objects g1', g2' within the predetermined distance range of the position object P1' belong to the category C1, and the geographic information objects g3, g4' belong to the category C2. The characteristic information of the geographic information objects g1', g2', g3', and g4' are f1', f2', f3', and f4', respectively. The geographic information objects g1', g2', g3', and g4' are separated from the position object P1' by distances d1', d2', d3', and d4', respectively. If d2' > d1', d3' > d4' and the predetermined class sequence is { C1, C2}, the feature information sequence of the position object P1' is { f1', f2', f4', f3' }. It should be understood that other predetermined conditions for sorting may be applied in the present embodiment, and the present embodiment is not limited thereto.

Further, in an alternative implementation, the characteristic information of the geographic information object includes a combination of one or more of the following: object identification, object shape, object position, relationship to a position object, position information of a relative position object. The object shape may be a shape corresponding to a contour of the geographic information object, such as an contour of AOI or the like. Alternatively, the object shape may be further characterized by using maximum coordinate values of the geographic information object in multiple directions, or using coordinate values of multiple vertices, which is not limited by the embodiment. The object location may be a latitude and longitude representation of the geographic information object. The positional relationship with the positional object may be an inclusive or exclusive relationship. If the position object is located inside the geographic information object, the relationship is included, or if the position object is located outside the geographic information object, the relationship is not included. Further alternatively, the location of the geographic information object may be characterized by the latitude and longitude of the center point of the geographic information object. Alternatively, the center point of the geographic information object may be the center point of an inscribed image or an circumscribed image of the geographic information object, which is not limited in this embodiment.

In an alternative implementation manner, the embodiment obtains the feature information by encoding the information of the geographic information object in a predetermined manner. For example, the characteristic information of the geographic information object may be encoded by a label encoding method, a single-hot encoding method, an ordinal encoding method, a binary encoding method, a count encoding method, and/or a hash encoding method, which is not limited to the specific encoding method in this embodiment.

In an alternative implementation manner, when the information of the geographic information object is encoded, a predetermined masking method may be used to mask a part of features in the original information, so as to perform hiding, blurring or desensitizing processing on feature areas or features of the corresponding geographic space data. Alternatively, the present embodiment may use a geographic information mask manner, for example, a pixelation manner, a perturbation manner, an aggregation manner, a position offset manner, a geometric deformation manner, a void filling manner, and the like, which is not limited in this embodiment. Further, the embodiment may perform random content masking on the original information of the geographic information object according to a certain proportion, for example, replace the original information with a mask. In other alternative implementations, the location of the content to be masked may be selected first, then the selected location may be partially replaced with a mask, partially randomly replaced with other content, and the remaining location content unchanged. It should be understood that the present embodiment is not limited to a specific masking operation, and can be adapted to the corresponding application field.

Fig. 3 is a schematic diagram of a determining process of a feature information sequence of a location object according to an embodiment of the present invention. As shown in fig. 3, for the position object P, geographic information objects O1, O2, O3, and O4 exist within a predetermined distance range of the position object P. Wherein, the geographic information objects O1 and O2 are AOI objects, the geographic information object O3 is a POI object, and the geographic information object O4 is a road network object.

Further, information of the geographic information objects O1, O2, O3 and O4 is acquired and encoded, and corresponding feature information f _O1、f_O2、f_O and f _O4 is obtained. Wherein the feature information f _O1、f_O2、f_O and f _O4 include corresponding features f1-f5, respectively. Optionally, the feature f1 characterizes an object identification feature, f2 characterizes an object shape feature, f3 characterizes an object position feature, f4 characterizes a relationship feature of a corresponding geographic position object and a position object P, and f5 characterizes a position feature of a relative position object P. The feature f3 corresponding to the geographic information object O2 and the features f3 and f5 corresponding to the geographic information object O4 are subjected to mask processing to ensure data privacy. Thus, the present embodiment can obtain the feature information sequence f= { F _O1,f_O2,f_O3,f_O4 } of the position object P.

Step S130, training and obtaining a position coding model according to the characteristic information sequence of each geographic object.

In an alternative implementation manner, the embodiment adopts a geographic contrast learning method to pretrain according to the characteristic information sequence of each geographic object to obtain the position coding model GeoEncoder.

Geographic contrast learning (Geographic Contrastive Learning, GCL) is an unsupervised training method that learns a characteristic representation of data by comparing similarities or differences between different data samples (e.g., different location objects in this embodiment). According to the embodiment of the invention, spatial relation and attribute information in the geographic data are adopted, and the representation of the ground clearance characteristics is learned through a comparison learning method, so that the tasks of automatic extraction, classification, identification and the like of the geographic information are realized.

Further alternatively, the position coding model of the embodiment may perform parameter tuning by learning the corresponding loss function through geographic comparison. Alternatively, the corresponding loss function may be determined by the difference of the predicted result of the similarity or difference between the objects at different positions from the real tag. It should be understood that the present embodiment is not limited to a specific loss function. Therefore, the embodiment adjusts the characteristic information of each position object through the similarity and/or the difference of each position object so as to regress the characteristic information of each position object accurately.

In another alternative implementation manner, the embodiment may further train the position coding model by using a pre-training mode that combines a geographic information mask mode with a geographic contrast learning method.

According to the embodiment, the position coding model can be subjected to parameter adjustment according to the geographic information mask loss and the geographic comparison learning loss, and the position coding model is obtained.

Further, since the geographic information mask is adopted to cause the loss of the geographic information to a certain extent, the embodiment can compensate the performance loss caused by the data loss through the geographic information mask loss, or can determine whether the data is subjected to the masking operation or not through balancing the performance difference between the front and the rear of the data mask, so that the accuracy of the position coding model is further improved.

In an optional implementation manner, the position coding model obtained by pre-training adopts an unsupervised training manner, and in this embodiment, the pre-training position coding model may be further subjected to supervised training by using a small amount of training data, so as to fine tune the pre-training position coding model, thereby further improving accuracy of the position coding model.

Fig. 4 is a flowchart of a method for fine tuning a position-coding model according to an embodiment of the present invention. Optionally, as shown in fig. 4, the present embodiment performs fine tuning on the position-coding model after the unsupervised pre-training by the following method steps:

Step S140, acquiring a location road-holding pair dataset. The position road grabbing pair data set comprises a plurality of pairs of position features and road grabbing information which have corresponding relations. The road grabbing means grabbing or identifying the corresponding road or road section according to the corresponding position. Optionally, the road or road segment corresponding to the location may include a road or road segment where the location is located, and/or an available road or road segment corresponding to the route plan corresponding to the location, and so on.

In an alternative implementation, such as an application scenario with a start and end point, the location features in the location grip pair include a start feature and an end point feature, and the corresponding grip information includes spot grip information and end point grip information. It should be understood that the position road-grabbing pair of the present embodiment is not limited to the starting point, and other positions where road-grabbing operation is required can be applied in the present embodiment.

And step S150, performing supervised training on the pre-trained position coding model according to the position road grabbing pair data set to obtain a final position coding model.

The embodiment can input a plurality of pairs of position characteristics and road grabbing information with corresponding relations in the position road grabbing data set into a pre-trained position coding model for processing, obtain corresponding position characteristic information and road section characteristic information, determine model loss through similarity of the position characteristic information and the road section characteristic information with the corresponding relations, and adjust the position coding model based on the loss until the performance required by the model is met.

Taking an application scene with a starting point and a finishing point as an example, inputting information of the starting point position, information of the finishing point position, starting point road grabbing information and finishing point road grabbing information into a position coding model for processing, obtaining starting point characteristic information and road segment characteristic information corresponding to road grabbing, determining model loss by calculating similarity of the starting point characteristic information and the road segment characteristic information corresponding to the road grabbing, and adjusting the position coding model based on the loss until the performance required by the model is met.

According to the embodiment of the invention, the unsupervised pre-training of the position coding model is realized through geographic comparison learning, and the position coding is finely adjusted through supervised training, so that the coding model can not only perform feature coding on the position information more accurately, but also perform sorting scoring on recalled roads in a road grabbing task, and the model complexity and the calculation amount of different tasks are further reduced.

In an alternative implementation, the segment coding model is obtained based on training of the path sequence data set, which may encode the information of the segment into corresponding feature vectors. The information of the link may include a link identification, a link type, each lane type of the link, and auxiliary information (e.g., speed limit information, construction information, etc.).

Fig. 5 is a flowchart of a training method of a link coding model according to an embodiment of the present invention. As shown in fig. 5, the training method of the road segment coding model according to the embodiment of the invention includes the following steps:

Step S210, a path sequence data set is acquired. Wherein the path sequence data set comprises a plurality of path sequences consisting of at least one road segment. Each path sequence may include information of a corresponding one or more road segments.

In an alternative implementation manner, the embodiment obtains the feature information by encoding the information of the road section in a predetermined manner. For example, the path segment information may be encoded using a label encoding, a single-hot encoding, an ordinal encoding, a binary encoding, a count encoding, and/or a hash encoding, which is not limited in this embodiment.

In an alternative implementation manner, when the information of the road section is encoded, a predetermined masking method may be used to mask a part of features in the original information, so as to perform hiding, blurring or desensitizing processing of feature areas or features on the corresponding geospatial data. Alternatively, the present embodiment may use a geographic information mask manner, for example, a pixelation manner, a perturbation manner, an aggregation manner, a position offset manner, a geometric deformation manner, a void filling manner, and the like, which is not limited in this embodiment. Further, the embodiment may perform random content masking on the original information of the road section according to a certain proportion, for example, replace the original information with a mask. In other alternative implementations, the location of the content to be masked may be selected first, then the selected location may be partially replaced with a mask, partially randomly replaced with other content, and the remaining location content unchanged. It should be understood that the present embodiment is not limited to a specific masking operation, and can be adapted to the corresponding application field.

Step S220, pre-training by adopting a masking training method according to the path sequence data set to obtain a road section coding model.

The embodiment of the invention adopts a masking training method to mask part of road segments of the input path sequence so as to predict the masked road segments based on the road segments of the unmasked path sequence. Therefore, the road section coding model provided by the embodiment of the invention can fully learn the information of the road section and the information of the adjacent road section, and improves generalization capability and robustness.

In an alternative implementation manner, the embodiment of the invention performs dense embedding (Dense Embedding) on the characteristic information of the sparse path sequence, so as to obtain a corresponding hidden representation (Hidden Representation), so as to obtain an output result, calculates corresponding loss based on the output result, and adjusts parameters of the road section coding model based on the loss, so as to obtain the trained road section coding model.

In a further alternative implementation manner, since the same or similar road segments may exist in different path sequences, the embodiment of the invention samples by sampling the softmax classifier to sample and obtain a part of the training samples in all categories to calculate the loss, so that the calculation amount of each training iteration can be reduced while the capability of distinguishing different categories is ensured.

Fig. 6 is a schematic diagram of a training process of a road segment coding model according to an embodiment of the present invention. As shown in fig. 6, in this embodiment, after feature encoding is performed on the segments in each path sequence, corresponding Sparse Features (spark Features) are obtained, the Sparse Features are input to a dense embedding layer (Dense Embedding) to compress the high-dimensional Sparse Features into dense data, so as to improve model performance, reduce computational complexity, and extract corresponding Features or information from the dense data, obtain a representation of hidden space (Hidden Representation), further convert the feature representation of hidden space, obtain an Output result (Output Matrix), and calculate a loss of a corresponding training period based on the Output result and a predetermined loss function, so as to adjust parameters of the segment encoding model based on the loss, so as to obtain the segment encoding model meeting the predetermined performance.

Further alternatively, as shown in fig. 6, since the same or similar road segments may exist in different path sequences, the embodiment of the present invention performs sample sampling by sampling the softmax classifier to sample and obtain a part of the samples from the training samples in all the classes to calculate the loss, so that the calculation amount of each training iteration can be reduced while the capability of distinguishing different classes is ensured.

In this embodiment, the geographic traffic data processing model processes information in the data set based on the position coding model and the road section coding model, and performs multi-modal training based on the processed result. The multi-modal training can process and understand various different types of data or 'modes', such as different types of geographic traffic data, and can capture and fuse the inherent connection and information consistency among different modes in the training process, so that geographic traffic information can be better understood and expressed.

FIG. 7 is a flow chart of a method of processing geographic traffic data in accordance with an embodiment of the present invention. Specifically, as shown in fig. 7, the geographic traffic data processing method according to the embodiment of the invention includes the following steps:

Step S310, a position path pair dataset is acquired. Wherein the position path pair data set includes a plurality of pairs of position information and path information having a correspondence relationship.

In an alternative implementation, the position path pair dataset of the present embodiment may be an OD-path sequence pair dataset. That is, the position path pair dataset may have multiple pairs of starting and ending points and corresponding path sequences. Alternatively, the embodiment may construct the position path pair dataset with the starting and ending points in the historical task and the corresponding actual routes. It should be understood that the location path pair data set of this embodiment may be constructed based on a corresponding application scenario, for example, the starting and ending point may be a rider location, a merchant location, or a merchant location and a user location in the instant distribution field, and the path sequence may be a route traveled by the distribution resource.

Step S320, corresponding feature coding is carried out on the position path pair data set according to the pre-trained position coding model and the road section coding model, and position feature information and path feature information of a plurality of position path pairs are obtained.

In this embodiment, the position information in the data set is feature-coded on the position path according to the position coding model obtained through training in the training process, so as to obtain position feature information. Alternatively, the position feature information may include a start position feature and an end position feature.

And carrying out feature coding on the path sequence in the data set on the position path according to the road section coding model obtained through training in the training process, and obtaining path feature information. The path characteristic information comprises characteristic information of each road section in the corresponding path sequence.

And step S330, performing multi-mode training according to the position characteristic information and the path characteristic information of each position path pair to acquire a geographic traffic data processing model.

In an optional implementation manner, the embodiment of the invention adopts a multi-mode geographic information mask and a position path matching method to pretrain according to the position characteristic information and the path characteristic information of each position path pair after characteristic alignment so as to acquire a geographic traffic data processing model.

In this embodiment, the multimodal geographic information mask (Multimodal Geospatial Information Masking) may enable multiple different types of geographic traffic information data to be processed and analyzed using specific masking techniques. Therefore, the geographic traffic data processing model of the embodiment of the invention can extract and highlight the geographic data characteristics required by the corresponding downstream tasks, so that other downstream traffic tasks can be processed, and the accuracy of various downstream traffic tasks is ensured.

Further, a location path matching Method (MATCH) is used to implement the method of cross-modal information integration and transfer. Optionally, the location path matching method of the present embodiment may enable the model to dynamically allocate weights according to the correlation between different modalities through an attention mechanism, so as to focus on key areas or features in the data of different modalities.

Further optionally, the embodiment may implement spatial alignment, fusion and conversion of different modality features by constructing a multi-modality interaction layer (Muti-modal Interaction Layer) to capture fine-grained information related to geographic traffic data of different modalities.

Specifically, the embodiment of the invention aligns the characteristic information of a plurality of modes through the multi-mode interaction layer, namely, the position characteristic information and the path characteristic information of the position path pair are aligned in characteristics, so that unified expression of multiple types of geographic traffic information in the same characteristic space is realized. Alternatively, the multi-modal interaction layer may be designed by means of attention mechanisms, element level interactions, gating mechanisms, cross-modal pooling, depth separable convolution, joint embedding space, or bilinear pooling/transformation. The embodiment of the invention can design the multi-mode interaction layer by adopting a corresponding design mode based on a specific application scene, and the embodiment is not limited to the design mode.

Further alternatively, the present embodiment may also process and optimize traffic data flows from multiple modalities simultaneously through joint training, so that the segment coding model can understand and correlate different types of input data as a whole.

Further optionally, the embodiment may further set a matching target (for example, a starting point and a destination are set as the matching target) or a loss function through cross-modal matching, so that the geographic traffic data processing model can establish an effective mapping relationship between multiple modalities.

FIG. 8 is a schematic diagram of a training process for a geographic traffic data processing model in accordance with an embodiment of the present invention. Optionally, as shown in fig. 8, in the training process of the geographic traffic data processing model according to the embodiment of the present invention, a location path pair data set D is first acquired, and a plurality of training samples (i.e., a plurality of location data pairs) are acquired from the location path pair data set D. The position data pair comprises a starting point position o_l, an ending point position d_l and a path sequence path.

Further, the start position o_l and the end position d_l are input into a pre-trained position coding model 81, OD position characteristic information including start characteristic information and end characteristic information is obtained, and the path sequence path is input into a pre-trained link coding model 82, so as to obtain link path characteristic information.

Further, in this embodiment, the obtained OD position feature information and link path feature information are input into the geographic traffic data processing model 83 to be trained, and the output of the geographic traffic data processing model 83 is obtained through a multi-layer neural network architecture (for example, including a neural network layer for implementing a multi-mode interaction layer, etc.), so that parameters of the geographic traffic data processing model 83 are adjusted based on the output, and the MATCH loss and the multi-mode mask loss until a model satisfying the performance of the application scenario is obtained. Further alternatively, the present embodiment may use a weighted sum of MATCH penalty and multi-mode mask penalty to obtain a final penalty, and make the tuning based on the final penalty. It should be understood that the present embodiment is not limited to the calculation method of the final loss, and may be determined by direct addition or the like.

In an alternative implementation manner, the embodiment may maintain the parameters of the location coding model and the road segment coding model in the training period of the geographic traffic data processing model 83, or further fine tune the location coding model and the road segment coding model in the training period of the geographic traffic data processing model 83, so as to further improve the accuracy of the geographic traffic data processing architecture. The embodiment of the invention does not limit the specific parameter adjustment mode.

Optionally, the multi-layer neural network architecture of the geographic traffic data processing model 83 includes a multi-Head Attention mechanism layer (Muti-Head Attention) 831, an addition and normalization layer 832 (Add & Norm) 832, a Feed-Forward layer 833, and an addition and normalization layer 834 (Add & Norm).

The multi-head attention mechanism layer 831 can capture different subspace characteristics of data, enhance the ability of model understanding and learning complex dependency relationships, and can perform parallel processing, thereby effectively improving the computing efficiency. Further, the present embodiment may design a corresponding multi-modal interaction layer based on the multi-head attention mechanism layer 831 to achieve spatial alignment of multi-modal features.

An addition and normalization layer 832 (Add & Norm) and an addition and normalization layer 834 (Add & Norm) are disposed after the multi-headed attention mechanism layer 831 and the feed forward layer 833 for improving stability and efficiency in the model training process.

The feed-forward layer 833 can implement the processing of the geographic traffic data by linear transformation and activation functions, and outputs the final feed-forward result.

It should be understood that the multi-layer neural network architecture of the geographic traffic data processing model 83 in this embodiment is merely exemplary, and other network architectures suitable for embodiments of the present invention, such as providing a plurality of feedforward layers, etc., can be applied in this embodiment, which is not limited thereto.

In an alternative implementation manner, the embodiment of the invention can also perform unsupervised training on the model by using training data (namely, the position path pair data set) in a pre-training stage through migration learning, and perform fine tuning on a specific downstream task, so that the geographic traffic data processing model can migrate knowledge learned from one mode to another mode.

FIG. 9 is a flow chart of a method of fine tuning a geographic traffic data processing model in accordance with an embodiment of the present invention. Optionally, as shown in fig. 9, the method for fine tuning a geographic traffic data processing model of the present embodiment includes the following steps:

step S340, acquiring specific task data sets corresponding to at least one specific task respectively.

Step S350, performing supervised training on parameters of the geographic traffic data processing model according to at least one specific task data set to obtain a final geographic traffic data processing model.

In an alternative implementation, taking instant messaging application scenarios as an example, the at least one specific task comprises a distance task. Wherein the specific task data set includes a plurality of pairs of position information and distance information having a correspondence relationship. Optionally, the end-distance dataset is constructed from historical data in the historical delivery tasks. Further alternatively, the present embodiment constructs the corresponding data set by the historical delivery task satisfying the predetermined condition. The predetermined condition may be a condition such as time distribution, and the present embodiment is not limited thereto.

In an embodiment, the distance task may be a distance-of-delivery task, a distance-of-delivery-based merchant distance-of-delivery range task, or the like.

Further optionally, in this embodiment, the geographic traffic data processing model is finely tuned by the difference between the predicted distance (i.e. the output of the geographic traffic data processing model) and the real distance corresponding to the regression starting point, until the difference between the predicted distance and the real distance output by the geographic traffic data processing model meets the performance of the specific application scenario.

In another alternative implementation, the at least one particular task includes a time determination task. Wherein the specific task data set includes a plurality of pairs of position information and time information having a correspondence relationship.

Optionally, the end-arrival time data set is constructed from historical data in the historical delivery tasks. Further alternatively, the present embodiment constructs the corresponding data set by the historical delivery task satisfying the predetermined condition. The predetermined condition may be a condition such as time distribution, and the present embodiment is not limited thereto.

Further optionally, in this embodiment, the geographic traffic data processing model is finely tuned by the difference between the predicted arrival time (i.e. the output of the geographic traffic data processing model) and the actual arrival time corresponding to the regression starting end point until the difference between the predicted arrival time and the actual arrival time output by the geographic traffic data processing model meets the performance of the specific application scenario.

In another alternative implementation, the at least one specific task includes a route scoring task. Wherein the specific task data set includes a plurality of pairs of location information and route information having a correspondence relationship.

Optionally, the data set of the end-delivery route is constructed from historical data in the historical delivery tasks. Further alternatively, the present embodiment constructs the corresponding data set by the historical delivery task satisfying the predetermined condition. The predetermined condition may be a condition such as time distribution, and the present embodiment is not limited thereto.

Further optionally, in this embodiment, the geographical traffic data processing model is finely adjusted by returning the matching degree of the starting point and the corresponding route until the errors of the route and the actual route determined by the matching degree of each starting point and the route output by the geographical traffic data processing model meet the performance of the specific application scenario.

In another alternative implementation, the at least one specific task includes a road tag mining task. Wherein the specific task data set includes a plurality of pairs of route information and tag information having a correspondence relationship.

Alternatively, the present embodiment may construct the data set of the route and the road label by the route of the historical delivery task and the road label corresponding to the route. Further alternatively, the present embodiment constructs the corresponding data set by the historical delivery task satisfying the predetermined condition. The predetermined condition may be a condition such as time distribution, and the present embodiment is not limited thereto.

Further optionally, in this embodiment, the classification task performs fine adjustment on the geographic traffic data processing model based on the difference between the route road label class and the actual road label class output by the model, until the errors of the route road label class and the actual road label class output by the geographic traffic data processing model meet the performance of the specific application scenario.

It should be understood that specific tasks corresponding to different geographic traffic application scenarios are different, and this embodiment is only exemplified by instant delivery tasks, and specific tasks in other application scenarios may also be fine-tuned to the geographic traffic data processing model based on migration learning to adapt to the corresponding specific tasks.

FIG. 10 is a schematic diagram of a process for fine tuning a geographic traffic data processing model according to an embodiment of the present invention. As shown in fig. 10, in the process of fine tuning the geographic traffic data processing model, a specific task data set of a specific task is acquired, a start point o_l and an end point d_l in the specific task data set are input into the position coding model 101, start-end point position feature information is acquired, route data in the specific task data set is input into the road segment coding model 102 to acquire corresponding route feature information, the start-end point position feature information and the route feature information are input into the geographic traffic data processing model 103 for processing, and output states h_o and h_d, a route output state h_path and a road segment output state h_link corresponding to the start-end points are acquired. Further, in this embodiment, the output states h_o, h_d, the route output state h_path and/or the road section output state h_link are decoded to obtain the output result of the corresponding specific task, and corresponding losses, such as a Loss DISLoss corresponding to the distance task, a Loss ETA Loss corresponding to the time determination task, a Loss LABEL Loss corresponding to the road LABEL mining task, and the like, are calculated based on the actual data in the specific task data set, so as to adjust the parameters of the geographic traffic data processing model 103 based on the corresponding losses, and obtain the fine-tuned geographic traffic data processing model 103.

In an alternative implementation, the embodiment of the present invention further includes: and inputting the target task information into a geographic traffic data processing architecture for processing, and obtaining a corresponding task result. Wherein the target task information includes location information and/or route information. The task results include a combination of one or more of the following: distance information, time prediction information, route information, and road tag information corresponding to the target task. The geographic traffic data processing architecture comprises a position coding model, a road section coding model and a geographic traffic data processing model. It should be appreciated that the target task may be any particular task involved in the model tuning process, which may be determined based on the corresponding geographic traffic application scenario.

FIG. 11 is a flow chart of another method of processing geographic traffic data in accordance with an embodiment of the present invention. As shown in fig. 11, the geographic traffic data processing method according to the embodiment of the invention includes the following steps:

Step S410, obtain the target task information. The target task information includes location information and/or path information.

Step S420, inputting the target task information into a pre-trained geographic traffic data processing architecture for processing, and obtaining a corresponding task result.

It should be understood that the pre-training and fine tuning of the position coding model, the road segment coding model and the geographic traffic data processing model in the geographic traffic data processing architecture are similar to those of the above embodiments, and will not be repeated herein.

According to the embodiment of the invention, the traffic data frame comprising the position coding model, the road section coding model and the geographic traffic data processing model in the architecture is adopted to process the target task information, so that the corresponding task processing result is obtained. Therefore, the embodiment can realize the processing of multiple downstream tasks through the multi-mode training geographic traffic data processing model, and uniformly express the position features and the path features in the geographic traffic data through the position coding model, the road section coding model and the geographic traffic data processing model, so that the effect and the performance of the multiple downstream tasks are improved, and the resource utilization rate is improved.

Fig. 12 is a schematic diagram of a geographic traffic data processing device according to an embodiment of the present invention. As shown in fig. 12, the geographic traffic data processing device 12 of the embodiment of the present invention includes a data set acquisition unit 121, a feature encoding unit 122, and a training unit 123.

The data set acquisition unit 121 is configured to acquire a position path pair data set including a plurality of pairs of position information and path information having a correspondence relationship. The feature encoding unit 122 is configured to perform corresponding feature encoding on the position path pair data set according to a pre-trained position encoding model and a road segment encoding model, and obtain position feature information and path feature information of a plurality of position path pairs. The training unit 123 is configured to perform multi-modal training according to the location feature information and the path feature information of each of the location path pairs, and obtain a geographic traffic data processing model.

In an alternative implementation, the training unit 123 is further configured to: and pre-training by adopting a multi-mode geographic information mask and a position path matching method according to the position characteristic information and the path characteristic information of each position path pair after characteristic alignment, and acquiring a geographic traffic data processing model.

In an alternative implementation, the geographic traffic data processing device 12 further includes a supervised training unit configured to obtain task-specific data sets corresponding to at least one task-specific respectively, and to perform supervised training on parameters of the geographic traffic data processing model according to the task-specific data sets.

In an alternative implementation, the at least one specific task includes a distance task, and the specific task data set includes a plurality of pairs of location information and distance information having a correspondence relationship.

In an alternative implementation, the at least one specific task includes a time determination task, and the specific task data set includes a plurality of pairs of location information and time information having a correspondence relationship.

In an alternative implementation, the at least one specific task includes a route scoring task, and the specific task data set includes a plurality of pairs of location information and route information having a correspondence.

In an alternative implementation, the at least one specific task includes a road tag mining task, and the specific task data set includes a plurality of pairs of route information and tag information having a correspondence relationship.

In an alternative implementation, the geo-traffic data processing device 12 further includes a training unit of the location coding model configured to:

acquiring a position data set;

In an alternative implementation, the training unit of the position-coding model is further configured to:

In an alternative implementation, the predetermined condition includes a distance of each of the geographic information objects relative to a corresponding location object.

In an alternative implementation, the characteristic information of the geographic information object includes a combination of one or more of: object identification, object shape, object position, relationship to a position object, position information of a relative position object.

In an alternative implementation, the geographic information objects include AOI objects, POI objects, and/or road network objects within a predetermined range of the corresponding task object.

In an alternative implementation, the geographic traffic data processing device 12 further comprises a training unit of the road segment coding model configured to:

In an alternative implementation, the geographic traffic data processing device 12 further comprises a task processing unit configured to:

Fig. 13 is a schematic diagram of another geographic traffic data processing device according to an embodiment of the present invention. As shown in fig. 13, the geographic traffic data processing device 13 of the embodiment of the present invention includes an information acquisition unit 131 and a processing unit 132.

The information acquisition unit 131 is configured to acquire target task information including position information and/or path information. The processing unit 132 is configured to input the target task information into a pre-trained geographic traffic data processing architecture for processing, and obtain a corresponding task result. The geographic traffic data processing architecture comprises a position coding model, a road section coding model and a geographic traffic data processing model, wherein the position coding model is used for carrying out feature coding on position information, the road section coding model is used for carrying out feature coding on route information, the geographic traffic data processing model carries out multi-mode training on a data set based on a position path, the position coding model and the road section coding model, and the position path pair data set comprises a plurality of pairs of position information and path information with corresponding relations.

Fig. 14 is a schematic diagram of an electronic device according to an embodiment of the invention. In the present embodiment, the electronic device 14 includes a server, a terminal, and the like. As shown in fig. 14, the electronic device 14: at least one processor 141; and a memory 142 communicatively coupled to the at least one processor 141; and a communication component 143 communicatively connected to the scanning device, the communication component 143 receiving and transmitting data under the control of the processor 141; the memory 142 stores instructions executable by the at least one processor 141, and the instructions are executed by the at least one processor 141 to implement the geographic traffic data processing method.

Specifically, the electronic device includes: one or more processors 141, and a memory 142, one processor 141 being illustrated in fig. 14. The processor 141, memory 142 may be connected by a bus or other means, for example in fig. 14. The memory 142 serves as a non-volatile computer-readable storage medium for storing non-volatile software programs, non-volatile computer-executable programs, and modules. The processor 141 executes various functional applications of the device and data processing, i.e., implements the above-described geographic traffic data processing method, by running non-volatile software programs, instructions, and modules stored in the memory 142.

Memory 142 may include a storage program area that may store an operating system, at least one application program required for functionality, and a storage data area; the storage data area may store a list of options, etc. In addition, memory 142 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 142 optionally includes memory located remotely from processor 141, which may be connected to an external device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

One or more modules are stored in the memory 142 that, when executed by the one or more processors 141, perform the geographic traffic data processing method of any of the method embodiments described above.

The product may perform the method provided by the embodiment of the present application, and has the corresponding functional module and beneficial effect of the performing method, and technical details not described in detail in the embodiment of the present application may be referred to the method provided by the embodiment of the present application.

Another embodiment of the present invention is directed to a non-volatile storage medium storing a computer readable program for causing a computer to perform some or all of the method embodiments described above.

That is, it will be understood by those skilled in the art that all or part of the steps in implementing the methods of the embodiments described above may be implemented by a program stored in a storage medium, where the program includes several instructions for causing a device (which may be a single-chip microcomputer, a chip or the like) or a processor (processor) to perform all or part of the steps in the methods of the embodiments of the present application. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, and various modifications and variations may be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method of processing geographic traffic data, the method comprising:

Performing corresponding feature coding on the position path pair data set according to a pre-trained position coding model and a road segment coding model, obtaining position feature information and path feature information of a plurality of position path pairs, wherein the position coding model is obtained by training a feature information sequence of a position object in the pre-obtained position data set, the feature information sequence is determined based on a geographic information object in a preset range of the position object, the road segment coding model is obtained by training a masking training method according to a pre-obtained path sequence data set, and the path sequence data set comprises a plurality of path sequences formed by at least one road segment;

2. The method of claim 1, wherein the multi-modal training based on the location feature information and the path feature information of each of the location path pairs, the obtaining the geographic traffic data processing model comprises:

3. The method according to claim 1, wherein the method further comprises:

4. A method according to claim 3, wherein the at least one specific task comprises a distance task and the specific task data set comprises a plurality of pairs of location information and distance information having a correspondence.

5. A method according to claim 3, wherein the at least one specific task comprises a time determination task and the specific task data set comprises a plurality of pairs of location information and time information having a correspondence.

6. The method of claim 3, wherein the at least one specific task comprises a route scoring task and the specific task data set comprises a plurality of pairs of location information and route information having a correspondence.

7. The method of claim 3, wherein the at least one task-specific data set includes a plurality of pairs of route information and tag information having a correspondence, and wherein the at least one task-specific data set includes a road tag mining task.

8. The method of claim 1, wherein the training step of the position-coding model comprises:

acquiring a position data set;

acquiring a characteristic information sequence of a position object in the position data set;

9. The method of claim 8, wherein the obtaining the sequence of characteristic information of the location objects in the location dataset comprises:

10. The method of claim 9, wherein the predetermined condition includes a distance of each of the geographic information objects relative to a corresponding location object.

11. The method of claim 9, wherein the characteristic information of the geographic information object comprises a combination of one or more of: object identification, object shape, object position, relationship to a position object, position information of a relative position object.

12. The method of claim 8, wherein the geographic information objects comprise AOI objects, POI objects, and/or road network objects within a predetermined range of the corresponding task object.

13. The method of claim 8, wherein said training to obtain said position-coding model from each of said characteristic information sequences comprises:

14. The method of claim 13, wherein said training to obtain said position-coding model from each of said characteristic information sequences further comprises:

15. The method of claim 1, wherein the training step of the segment coding model comprises:

acquiring a path sequence data set;

16. The method according to claim 1, wherein the method further comprises:

Inputting target task information into a geographic traffic data architecture for processing, and obtaining a corresponding task result, wherein the target task information comprises position information and/or path information, and the task result comprises one or more of the following combinations: distance information, time prediction information, route information and road label information corresponding to the target task;

The geographic traffic data architecture comprises a position coding model, a road section coding model and a geographic traffic data processing model.

17. A method of processing geographic traffic data, the method comprising:

The geographic traffic data processing architecture comprises a position coding model, a road section coding model and a geographic traffic data processing model, wherein the position coding model is used for carrying out feature coding on position information, the road section coding model is used for carrying out feature coding on route information, the geographic traffic data processing model is obtained by carrying out multi-mode training on a data set based on a position path, the position coding model and the road section coding model, the position path data set comprises a plurality of pairs of position information and path information with corresponding relations, the position coding model is obtained by training according to a feature information sequence of a position object in the position data set, the feature information sequence is determined based on a geographic information object in a preset range of the position object, the road section coding model is obtained by training according to a preset path sequence data set by adopting a masking training method, and the path sequence data set comprises a plurality of path sequences formed by at least one road section.

18. A geographic traffic data processing device, the device comprising:

The feature coding unit is configured to perform corresponding feature coding on the position path pair data set according to a pre-trained position coding model and a road segment coding model, obtain position feature information and path feature information of a plurality of position path pairs, wherein the position coding model is obtained by training a feature information sequence of a position object in a pre-obtained position data set, the feature information sequence is determined based on a geographic information object in a preset range of the position object, the road segment coding model is obtained by training a masking training method according to a pre-obtained path sequence data set, and the path sequence data set comprises a plurality of path sequences formed by at least one road segment;

19. A geographic traffic data processing device, the device comprising:

20. An electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method of any of claims 1-17.

21. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program which, when executed by a processor, implements the method according to any of claims 1-17.

22. A computer program product, characterized in that the computer program product, when run on a computer, causes the computer to perform the method according to any of claims 1-17.