CN116295409A

CN116295409A - Route processing method, route processing device, computer readable medium and electronic equipment

Info

Publication number: CN116295409A
Application number: CN202310161446.1A
Authority: CN
Inventors: 刘雨亭
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2023-02-14
Filing date: 2023-02-14
Publication date: 2023-06-23

Abstract

The application can be applied to the technical fields of maps, traffic and the like, and particularly provides a route processing method, a route processing device, a computer readable medium and electronic equipment. The method comprises the following steps: generating a link expression corresponding to the planned route sample; extracting a plurality of initial embedded vectors corresponding to each element in the link expression through a pre-training model, and generating input feature vectors corresponding to each element; and carrying out multi-layer forward propagation processing according to the input feature vectors corresponding to the elements, and carrying out model training processing based on the planned route samples, training tasks corresponding to the pre-training models and intermediate result vectors corresponding to the elements so as to process the appointed route based on the converged pre-training modules. The technical scheme can solve the problem of huge workload caused by the need of model training by a large amount of labeling data, and can improve the accuracy of service processing.

Description

Route processing method, route processing device, computer readable medium and electronic equipment

Technical Field

The present invention relates to the field of computers and communication technologies, and in particular, to a route processing method, a route processing device, a computer readable medium, and an electronic device.

Background

The route planning is to find a route from a start point to an end point in a specified area, for example, in the field of map navigation, after a map user selects a start point position and an end point position on a map, a route from the start point position to the end point position can be planned. For example, in the field of robot control, path planning is to plan a collision-free safety route from a starting position to an end position of a robot. When a planned route is obtained, the planned route may typically be processed for related traffic, such as a rationality assessment, to improve the strategy of path planning based on the assessment results. In the related art, a developer is usually required to model based on industry experience and train a model based on feature engineering, and the scheme has the problem of huge workload and lower accuracy of service processing results.

Disclosure of Invention

The embodiment of the application provides a route processing method, a route processing device, a computer readable medium and electronic equipment, so that the problem of huge workload caused by the need of model training of a large amount of marked data can be solved, and the accuracy of business processing can be improved.

Other features and advantages of the present application will be apparent from the following detailed description, or may be learned in part by the practice of the application.

According to an aspect of the embodiments of the present application, there is provided a route processing method, including: generating a link expression corresponding to a planned route sample according to a plurality of links contained in the planned route sample and attribute information of each link; extracting a plurality of initial embedded vectors corresponding to each element in the link expression through a pre-training model, and generating input feature vectors corresponding to each element according to the plurality of initial embedded feature vectors; and carrying out multi-layer forward propagation processing according to the input feature vectors corresponding to the elements, and carrying out model training processing on the basis of the planned route samples, training tasks corresponding to the pre-training models and intermediate result vectors corresponding to the elements obtained by the multi-layer forward propagation processing so as to process the appointed route based on the converged pre-training modules.

According to an aspect of the embodiments of the present application, there is provided a route processing device including: the first generation unit is configured to generate a link expression corresponding to the planned route sample according to a plurality of links contained in the planned route sample and attribute information of each link; a feature extraction unit configured to extract a plurality of initial embedded vectors corresponding to respective elements in the link expression through a pre-training model; a second generating unit configured to generate input feature vectors corresponding to the respective elements according to the plurality of initial embedded feature vectors; and the model training unit is configured to perform multi-layer forward propagation processing according to the input feature vectors corresponding to the elements, perform model training processing on the basis of the planned route samples, training tasks corresponding to the pre-training models and intermediate result vectors corresponding to the elements obtained by the multi-layer forward propagation processing, and process the appointed route on the basis of the converged pre-training modules. .

In some embodiments of the present application, based on the foregoing solution, the first generating unit is configured to: generating elements which correspond to the links and are used for representing the attribute characteristics according to at least one attribute characteristic contained in the attribute information of the links; and combining elements respectively corresponding to the multiple links according to the sequence of the multiple links in the planned route sample to generate a link expression corresponding to the planned route sample.

In some embodiments of the present application, based on the foregoing solution, the first generating unit is configured to: according to the position of a specified link in the plurality of links in the planned route sample, adding a position marker bit corresponding to the specified link in the link expression; wherein the location identification bits include at least one of: a first flag bit added before an element corresponding to a start link in the planned route sample, a second flag bit added between elements corresponding to two specified links, and a third flag bit added after an element corresponding to a stop link in the planned route sample.

In some embodiments of the present application, based on the foregoing scheme, the specified two links include: links located at road intersections in the planned route samples, or any two adjacent links in the plurality of links.

In some embodiments of the present application, based on the foregoing scheme, the feature extraction unit is configured to: extracting word embedding vectors, position embedding vectors and route interval embedding vectors corresponding to all elements in the link expression through a pre-training model; the position embedding vector is used for representing the specific position of the element in the link expression, and the route interval embedding vector is used for representing the interval of the element in the link expression.

In some embodiments of the present application, based on the foregoing solution, the second generating unit is configured to: and based on a set merging processing mode, merging a plurality of initial embedded feature vectors corresponding to each element, and taking the merging processing result as an input feature vector corresponding to each element.

In some embodiments of the present application, based on the foregoing scheme, the model training unit is configured to: carrying out replacement operation on other links except the initial link in the planned route sample through set probability respectively to obtain a processed planned route sample; the replacing operation includes at least one of: the method comprises the steps of setting a flag bit, replacing a random link and keeping unchanged; and taking the intermediate result vector corresponding to each element as input, recovering the link represented by the set flag bit as an actual link in the planned route sample as an optimization target corresponding to a first training task, and performing model training processing.

In some embodiments of the present application, based on the foregoing solution, performing, by using a set probability, replacement operations on links other than the initial link in the planned route sample, respectively, includes: respectively determining whether replacement operation is to be performed on other links except the initial link in the planned route sample through a first probability; for the links requiring replacement operation, the replacement operation is performed by the second probability of replacing with the set flag bit, the third probability of replacing with the random link and the fourth probability of remaining unchanged.

In some embodiments of the present application, based on the foregoing scheme, the model training unit is configured to: replacing the links at the latter half of the planned route samples with random links through the set probability; the planned route sample is divided into a first half link and a second half link at two designated links; and taking an intermediate result vector corresponding to a first flag bit added before an element corresponding to an initial link in the link expression as input, recovering the random link to an optimization target corresponding to a second training task, wherein the second half link in the planned route sample is used as an optimization target, and performing model training processing.

In some embodiments of the present application, based on the foregoing solution, the route processing device further includes: the processing unit is configured to extract target embedded vectors corresponding to the elements through the converged pre-training model; generating a route embedding vector corresponding to the planning route sample according to the target embedding vector corresponding to each element; and taking the route embedded vector and target input information for adjusting the pre-training model as input, taking a set service as an optimization target, and adjusting model parameters of the pre-training model.

In some embodiments of the present application, based on the foregoing solution, the processing unit is configured to: calculating a target embedded vector mean value corresponding to the initial link according to a plurality of target embedded vectors corresponding to the initial link in the planned route sample; calculating to obtain a target embedded vector mean value corresponding to a plurality of links according to a plurality of target embedded vectors respectively corresponding to a plurality of links contained in the planned route sample; calculating to obtain a target embedded vector mean value corresponding to the terminated link according to a plurality of target embedded vectors corresponding to the terminated link in the planned route sample; and generating a route embedded vector corresponding to the planning route sample according to the target embedded vector average value corresponding to the initial link, the target embedded vector average values corresponding to the links and the target embedded vector average value corresponding to the termination link.

In some embodiments of the present application, based on the foregoing, the target input information for adjusting the pre-training model includes at least one of: route characteristics of the planned route samples obtained through characteristic engineering; and according to the route characteristics of the planned route sample, carrying out prediction processing of the setting business through a machine learning model to obtain a prediction result.

In some embodiments of the present application, based on the foregoing solution, the route processing device further includes: a first prediction unit; the first generation unit is further configured to: generating a link expression corresponding to a target planning route according to links contained in the target planning route to be processed and attribute information of each link; the feature extraction unit is further configured to: extracting target embedded vectors corresponding to all links in the target planning route through the converged pre-training model; the first prediction unit is configured to: and carrying out prediction processing of the setting service based on the target embedded vectors corresponding to the links in the target planning route.

In some embodiments of the present application, based on the foregoing scheme, the first prediction unit is configured to: and carrying out at least one of the following processing on the target planning route based on the target embedded vector corresponding to each link in the target planning route: abnormal link detection processing, completion processing of missing links in the target planned route, generation processing of a follow-up planned route and correlation processing of routes.

In some embodiments of the present application, based on the foregoing solution, the route processing device further includes: a second prediction unit; the first generation unit is further configured to: generating a link expression corresponding to a target planning route according to links contained in the target planning route to be processed and attribute information of each link; the second prediction unit is configured to: taking the link expression as the input of a converged pre-training model, and outputting a predicted result of the target planning route aiming at the set service through the converged pre-training model; wherein the setting service includes at least one of the following: the coverage rate of the driving track and the target planned route, the driving time of the target planned route, the planning rationality of the target planned route, the navigation completion rate of the target planned route and the flow of the target planned route.

According to an aspect of the embodiments of the present application, there is provided a computer readable medium having stored thereon a computer program which, when executed by a processor, implements a route processing method as described in the above embodiments.

According to an aspect of an embodiment of the present application, there is provided an electronic device including: one or more processors; and storage means for storing one or more computer programs which, when executed by the one or more processors, cause the electronic device to implement the route processing method as described in the above embodiments.

According to one aspect of embodiments of the present application, there is provided a computer program product comprising a computer program stored in a computer readable storage medium. The processor of the electronic device reads and executes the computer program from the computer-readable storage medium, so that the electronic device performs the route processing method provided in the above-described various alternative embodiments.

In the technical solutions provided in some embodiments of the present application, a link expression corresponding to a planned route sample is generated according to a plurality of links included in the planned route sample and attribute information of each link, then a plurality of initial embedded vectors corresponding to each element in the link expression are extracted through a pre-training model, and input feature vectors corresponding to each element are generated according to the plurality of initial embedded feature vectors, then multi-layer forward propagation processing is performed according to the input feature vectors corresponding to each element, and model training processing is performed based on the planned route sample, training tasks corresponding to the pre-training model, and intermediate result vectors corresponding to each element, so that a designated route can be processed based on a pre-training module after convergence, so that the problem of huge workload caused by the need of model training with a large amount of labeled data is solved, and as detailed features in the planned route sample can be learned through multi-layer forward propagation processing, the accuracy of service processing can be improved, meanwhile, different set services can be processed, and the effective application of multi-service fields is realized.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.

Drawings

FIG. 1 shows a schematic diagram of an exemplary system architecture to which the technical solutions of embodiments of the present application may be applied;

FIG. 2 illustrates a flow chart of a route processing method according to one embodiment of the present application;

FIG. 3 illustrates a flow chart of a route processing method according to one embodiment of the present application;

FIG. 4 illustrates a flow chart of a route processing method according to one embodiment of the present application;

FIG. 5 illustrates a route processing flow diagram according to one embodiment of the present application;

FIG. 6 illustrates a schematic diagram of breaking a route according to one embodiment of the present application;

FIG. 7 illustrates an unbedding feature schematic diagram corresponding to link in a route according to one embodiment of the present application;

FIG. 8 illustrates a Transformer forward propagation schematic according to one embodiment of the present application;

FIG. 9 illustrates a schematic diagram of the probability of processing for each link according to one embodiment of the present application;

FIG. 10 illustrates a schematic diagram of model fine tuning according to one embodiment of the present application;

FIG. 11 illustrates a block diagram of a route processing device according to one embodiment of the present application;

Fig. 12 shows a schematic diagram of a computer system suitable for use in implementing the electronic device of the embodiments of the present application.

Detailed Description

Example embodiments are now described in a more complete manner with reference being made to the figures. However, the illustrated embodiments may be embodied in various forms and should not be construed as limited to only these examples; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are set forth in order to provide a thorough understanding of embodiments of the present application. However, it will be recognized by one skilled in the art that the present application may be practiced without all of the specific details of the embodiments, that one or more specific details may be omitted, or that other methods, components, devices, steps, etc. may be used.

The block diagrams depicted in the figures are merely functional entities and do not necessarily correspond to physically separate entities. That is, the functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor devices and/or microcontroller devices.

The flow diagrams depicted in the figures are exemplary only, and do not necessarily include all of the elements and operations/steps, nor must they be performed in the order described. For example, some operations/steps may be decomposed, and some operations/steps may be combined or partially combined, so that the order of actual execution may be changed according to actual situations.

It should be noted that: references herein to "a plurality" means two or more. "and/or" describes an association relationship of an association object, meaning that there may be three relationships, e.g., a and/or B may represent: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship.

It will be appreciated that in the specific embodiments of the present application, related data such as planned routes are involved, and when the above embodiments of the present application are applied to specific products or technologies, user permissions or consents need to be obtained, and the collection, use and processing of related data need to comply with the relevant laws and regulations and standards of the relevant countries and regions.

As shown in fig. 1, in the map navigation field, after a map user selects a start position and an end position on a map, a route from the start position to the end position can be planned, specifically a planned route including Link1, link2, link3, and Link4 as shown in fig. 1. Link is the smallest unit of data describing a Link, and is a set of structured data including, but not limited to, link (Link) length, width, road class, etc.

After the planned route is obtained, the planned route may typically be processed for related traffic, such as a rationality assessment, to improve the strategy of path planning based on the assessment results. Machine learning algorithms in artificial intelligence (Artificial Intelligence, AI for short) may be employed in evaluating the planned path. Wherein. Artificial intelligence is the theory, method, technique and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend and expand human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and other directions.

Machine Learning (ML) is a multi-domain interdisciplinary, and involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and the like. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

In the related art, when processing related services such as rationality evaluation and the like on a planned route, a developer is usually required to model based on industry experience and train a model based on feature engineering, and the scheme has the problem of huge workload and lower accuracy of service processing results.

Based on this, in one embodiment of the present application, a new route processing scheme is proposed, as shown in fig. 1, after the vehicle terminal 101 selects a start point and an end point on the electronic map, the server 102 may generate a planned route including Link1, link2, link3, and Link4 according to the start point and the end point. To perform business processes such as rationality assessment on the planned route, the server 102 may train the pre-training model based on planned route samples and then evaluate the planned route through the converged pre-training model.

In one embodiment of the present application, when the server 102 trains the pre-training model based on the planned route sample, a link expression corresponding to the planned route sample may be generated according to a plurality of links included in the planned route sample and attribute information of each link; extracting a plurality of initial embedded vectors corresponding to each element in the link expression through a pre-training model, and generating input feature vectors corresponding to each element in the link expression according to the plurality of initial embedded feature vectors; performing multi-layer forward propagation processing according to input feature vectors corresponding to all elements in a link expression, and performing model training processing based on the planned route samples, training tasks corresponding to the pre-training model and intermediate result vectors corresponding to all elements obtained by the multi-layer forward transmission processing; after the model converges, the specified route may be processed based on the converged pre-training module. For example, a target embedded vector corresponding to each element in the link expression can be extracted through the converged pre-training model, and the prediction processing of the setting service is performed based on the target embedded vector. Such as evaluating the rationality of the planned route samples, etc.

It should be noted that, the server 102 may be an independent physical server, or may be a server cluster or a distributed system formed by at least two physical servers, or may be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, a content distribution network (Content Delivery Network, CDN), and basic cloud computing services such as big data and an artificial intelligence platform. The vehicle terminal 101 may specifically refer to a smart phone, a smart speaker, a screen speaker, a smart watch, etc. with an in-vehicle function, but is not limited thereto, and for example, the vehicle terminal 101 may be replaced by a mobile terminal such as an aircraft. The respective vehicle terminals and servers may be directly or indirectly connected through wired or wireless communication, and meanwhile, the number of the vehicle terminals and servers may be one or at least two, which is not limited herein.

The implementation details of the technical solutions of the embodiments of the present application are described in detail below:

fig. 2 shows a flow chart of a route processing method according to an embodiment of the present application, which may be performed by a server, by a terminal device (e.g. the vehicle terminal 101 shown in fig. 1), or by both the server and the terminal device. Referring to fig. 2, the route processing method at least includes steps S210 to S230, and is described in detail as follows:

In step S210, a link expression corresponding to the planned route sample is generated according to the links included in the planned route sample and the attribute information of each link.

In some alternative embodiments, the Link is the minimum data unit describing the road, and the attribute information of the Link includes, but is not limited to, the ID, length, width, road grade, road condition, traffic, and other attributes of the Link.

In some alternative embodiments, according to the links included in the planned route sample and the attribute information of each link, the process of generating the link expression corresponding to the planned route sample may be: and generating elements corresponding to each link and used for representing the attribute characteristics according to at least one attribute characteristic contained in the attribute information of each link, and then combining the elements corresponding to each link according to the sequence of the links in the planned route sample to generate a link expression corresponding to the planned route sample. For example, if the link expression is generated according to the ID of the link, the planned route sample R may be expressed as: route r= { link ₁ ,link ₂ ,link ₃ … }; wherein link is ₁ ,link ₂ ,link ₃ … each represents an ID of a corresponding link, i.e., an element corresponding to a link in a link expression.

If the link expression is generated according to the road class, road condition state and flow of the link, the planned route sample R is expressed as: route r= { rc_rs_flpw ₁ ,rc_rs_flow ₂ ,rc_rs_flow ₃ … }; wherein rc_rs_flow ₁ Representing link ₁ The road class, road condition state and flow of the link 1 are elements corresponding to the link 1 in the link expression; rc_rs_flow ₂ Representing link ₂ The road class, road condition state and flow of the link 2 are elements corresponding to the link in the link expression; rc_rs_flow ₃ Representing link ₃ The road class, road condition status and flow of (3) are elements corresponding to link 3 in the link expression.

In some optional embodiments, when generating the link expression corresponding to the planned route sample, a position flag bit corresponding to the specified link may be further added to the link expression according to the position of the specified link in the planned route sample in the multiple links; wherein the location identification bits include at least one of: a first flag bit added before an element corresponding to a start link in the planned route sample, a second flag bit added between elements corresponding to two specified links, and a third flag bit added after an element corresponding to a stop link in the planned route sample. I.e. the link expression is divided into a first half and a second half by means of a second flag bit. Alternatively, the first flag bit may be a CLS, and both the second flag bit and the third flag bit may be SEP.

In some alternative embodiments, the link expression may be divided into a first half and a second half at road intersections in the planned route sample, or may be divided at any two adjacent links. Namely, the two links specified in the foregoing embodiment include: links at road intersections in the planned route samples, or any two adjacent links of the plurality of links.

With continued reference to fig. 2, in step S220, a plurality of initial embedded vectors corresponding to respective elements in the link expression are extracted by the pre-training model, and input feature vectors corresponding to respective elements are generated according to the plurality of initial embedded feature vectors.

It should be noted that, the elements in the link expression include elements corresponding to each link and used for representing attribute features, and may also include elements corresponding to the position flag bits in the above embodiment.

In some alternative embodiments, the plurality of initial embedded vectors corresponding to the respective elements may include: word embedding vectors (token embedding), position embedding vectors (positioning embedding) and route interval embedding vectors (segment embedding) corresponding to the respective elements; the position embedding vector is used for representing the specific position of the element in the link expression, and the route interval embedding vector is used for representing the interval (i.e. in the first half or the second half) of the element in the link expression. Optionally, the plurality of initial embedded vectors corresponding to the respective elements may also include: word embedding vectors and location embedding vectors.

In some alternative embodiments, the process of generating the input feature vector corresponding to each element from the plurality of initial embedded feature vectors may be: based on a set merging processing mode, merging processing is carried out on a plurality of initial embedded feature vectors corresponding to each element, and the merging processing result is used as an input feature vector corresponding to each element. For example, a plurality of initial embedded vectors corresponding to the respective elements may be superimposed, and the superimposed result may be used as the corresponding input feature vector.

In step S230, multi-layer forward propagation processing is performed according to the input feature vectors corresponding to the elements, and model training processing is performed based on the planned route samples, training tasks corresponding to the pre-training model, and intermediate result vectors corresponding to the elements obtained by the multi-layer forward transmission processing, so as to process the specified route based on the converged pre-training module.

In some alternative embodiments, the process of performing model training processing based on the planned route sample, the training task corresponding to the pre-training model, and the intermediate result vector corresponding to each element obtained by the multi-layer forward transmission processing may include: carrying out replacement operation on other links except the initial link in the planned route sample through set probability respectively to obtain a processed planned route sample; the replacing operation includes at least one of: the method comprises the steps of setting a flag bit, replacing a random link and keeping unchanged; and then taking the intermediate result vector corresponding to each element as input, recovering the link represented by the set flag bit into an actual link in the planned route sample as an optimization target corresponding to the first training task, and performing model training processing. The technical scheme of the embodiment is that links except the initial link in the original planned route sample are subjected to probability replacement, and then model training is carried out by recovering the planned route sample as an optimization target.

Alternatively, when the links other than the initial link in the planned route sample are respectively replaced by the set probabilities, whether the links other than the initial link in the planned route sample are to be replaced or not may be determined respectively by the first probabilities, and for the links requiring replacement, the replacement may be performed respectively by the second probability of replacing with the set flag bit, the third probability of replacing with the random link, and the fourth probability of remaining unchanged. For example, the first probability may be 50%, the second probability may be 50%, the third probability may be 5%, and the fourth probability may be 45%. I.e. the sum of the second probability, the third probability and the fourth probability is 1.

In some alternative embodiments, the process of performing model training processing based on the planned route sample, the training task corresponding to the pre-training model, and the intermediate result vector corresponding to each element obtained by the multi-layer forward transmission processing may include: replacing the links in the latter half of the planned route samples with random links through the set probability; the method comprises the steps that a planned route sample is divided into a first half link and a second half link at two designated links; and taking an intermediate result vector corresponding to a first flag bit added before an element corresponding to the initial link in the link expression as input, and recovering the random link to an optimization target corresponding to a second training task by taking a second half link in the planned route sample as an optimization target, so as to perform model training processing. The technical scheme of the embodiment is that probability replacement is carried out on the latter half part of the original planned route sample, and then model training is carried out by recovering the planned route sample as an optimization target.

Alternatively, the specified route in step S230 may be a planned route sample, a route planned during actual application, or any selected route.

In some optional embodiments, after the pre-training model converges, the specified route is processed based on the converged pre-training module, which specifically may be that the target embedded vector corresponding to each element is extracted through the converged pre-training model; generating a route embedding vector corresponding to the planned route sample according to the target embedding vector corresponding to each element, then taking the route embedding vector and target input information for adjusting the pre-training model as input, taking the set business as an optimization target, and adjusting model parameters of the pre-training model.

In the embodiment of the present application, the target embedding vector corresponding to each element may also include a word embedding vector (token), a position embedding vector (positioning), and a route section embedding vector (segment embedding).

The technical scheme of the embodiment is that the pretraining model is subjected to fine adjustment processing through target embedded vectors and target input information corresponding to each element. Optionally, the setting service in this embodiment includes at least one of: the method comprises the steps of estimating the coverage rate of a driving track and a planned route sample, estimating the driving time of the planned route sample, estimating the planned rationality of the planned route sample, estimating the navigation completion rate of the planned route sample, and estimating the flow of the planned route sample.

In some alternative embodiments, according to the target embedded vector corresponding to each element, the process of generating the route embedded vector corresponding to the planned route sample may be: calculating to obtain a target embedded vector mean value corresponding to the initial link according to a plurality of target embedded vectors corresponding to the initial link in the planned route sample; according to a plurality of target embedded vectors respectively corresponding to a plurality of links contained in the planned route sample, calculating to obtain a target embedded vector average value corresponding to the links; calculating to obtain a target embedded vector mean value corresponding to the terminated link according to a plurality of target embedded vectors corresponding to the terminated link in the planned route sample; and generating a route embedded vector corresponding to the planned route sample according to the target embedded vector average value corresponding to the initial link, the target embedded vector average values corresponding to the links and the target embedded vector average value corresponding to the ending link. For example, the target embedded vector average value corresponding to the start link, the target embedded vector average values corresponding to the plurality of links, and the target embedded vector average value corresponding to the end link may be used as the route embedded vector corresponding to the planned route sample.

In some alternative embodiments, the target input information for adjusting the pre-training model may include at least one of: route characteristics of the planned route samples obtained through the characteristic engineering; and according to the route characteristics of the planned route sample, carrying out prediction processing of the setting business by a machine learning model to obtain a prediction result. Alternatively, the machine learning model may be a decision tree model.

Alternatively, the route characteristics may include one or more of mined characteristics, real-time characteristics, and static characteristics. The mined features may include historical arrival times, historical travel speeds, etc. of the planned route samples; the real-time characteristics may include real-time road conditions, real-time travel speeds, etc. of the planned route samples; static features may include road attributes of the planned route samples, such as road width, road type (e.g., high speed, provincial or national roads), etc.

Fig. 3 shows a flow chart of a route processing method according to an embodiment of the present application, which may be performed by a server, by a terminal device (e.g. the vehicle terminal 101 shown in fig. 1), or by both the server and the terminal device. Referring to fig. 3, the route processing method at least includes steps S310 to S330, and is described in detail as follows:

in step S310, a link expression corresponding to the target planned route is generated according to the links included in the target planned route to be processed and the attribute information of each link.

The specific implementation details of this step are similar to those of step S210, and will not be described again.

In step S320, the target embedded vectors corresponding to the links in the target planned route are extracted through the converged pre-training model.

In the embodiment of the present application, the pre-training model may be trained by the technical solution of the embodiment shown in fig. 2, and the target embedding vector corresponding to each link may include a word embedding vector (token), a location embedding vector (location embedding), and a route interval embedding vector (segment embedding).

In step S330, the prediction process of the setting service is performed based on the target embedded vectors corresponding to the links in the target planned route.

In some alternative embodiments, based on the target embedded vector corresponding to each link in the target planned route, at least one of the following processes may be performed on the target planned route: abnormal link detection processing, completion processing of missing links in a target planning route, generation processing of a follow-up planning route and correlation processing of routes.

Specifically, for the abnormal link detection service, a predicted value may be generated for each link in the target planned route by using the target embedded vector, where the predicted value represents a probability that each link belongs to the target planned route, and if the predicted value corresponding to a certain link is smaller, it may be stated that the link may have an abnormality. The completion process of the missing link may be to predict which link should be complemented by the missing link by the target embedding vector when a certain link is missing in the target planned route. The generation process of the subsequent planned route may be to predict a missing link or a link that should be complemented between the start position and the end position from the target embedding vector after the start position and the end position are known. The correlation processing of the routes may be to calculate the distance between the routes according to the target embedded vector corresponding to the routes, and then determine the correlation according to the distance value.

Fig. 4 shows a flow chart of a route processing method according to an embodiment of the present application, which may be performed by a server, by a terminal device (e.g. the vehicle terminal 101 shown in fig. 1), or by both the server and the terminal device. Referring to fig. 4, the route processing method at least includes steps S410 to S420, and is described in detail as follows:

in step S410, a link expression corresponding to the target planned route is generated according to the links included in the target planned route to be processed and the attribute information of each link.

In step S420, the link expression is used as an input of the converged pre-training model, and a prediction result of the target planning route for the set service is output through the converged pre-training model.

In some alternative embodiments, the setup service includes at least one of: coverage rate of the running track and the target planned route, running time of the target planned route, planning rationality of the target planned route, navigation completion rate of the target planned route and flow rate of the target planned route.

According to the technical scheme, the problem of huge workload caused by the fact that a large amount of marked data are needed for model training is solved by processing the unmarked planned route samples, and the accuracy of service processing can be improved because the detailed characteristics in the planned route samples can be learned through multi-layer forward propagation processing, meanwhile, different setting services can be processed, and effective application in the multi-service field is achieved.

The following details of implementation of the technical solutions in the embodiments of the present application are described in detail with reference to fig. 5 to 10 by taking estimated route rationality as an example:

in the related art, the schemes for estimating the route rationality mainly include the following schemes: scheme 1, a route is manually modeled, a machine learning model is used to perform rationality estimation on a route sample with a fixed length of a feature, and the machine learning model can be selected from a logistic regression model, a GBDT (Gradient Boosting Decision Tree, gradient descent tree) model, a lambdaRank (a sort algorithm) model, and the like. And 2, modeling the fixed length features of the route sample through a method combining deep learning with artificial modeling, for example, through artificial feature extraction, and combining with a route rationality prediction scheme of a deep model structure. Scheme 3, through the method of embedding, deep learning, artificial modeling, for example through the structure of a double-tower model, the artificial feature extraction is carried out on the route and the user, meanwhile, the data unit of the route and the user are respectively subjected to embedding, and then the reasonability of the route is estimated through the structure of a deep network.

However, the manual modeling in the

schemes

1 and 2 in the related art strongly depends on industry experience and business understanding of developers, and it is difficult to achieve completeness of business modeling, so that it is difficult to achieve an optimal effect under a business objective. The route unbinding scheme in the scheme 3 is developed based on a route link topological structure, the information of the unbinding result mainly comprises the topological attribute of a route network formed by routes, the topology attribute is difficult to match with an expected business target, and the application scene is small.

Based on this, in the technical solution provided in the present application, as shown in fig. 5, the adopted data includes a high-quality navigation route, and the high-quality navigation route refers to a route formed by link strings, which is reasonable and acceptable. The manner of quantification of rationality generally includes: coverage rate of historical driving track and navigation route, navigation completion rate under the navigation route, rationality probability of navigation route, etc.

It should be noted that: the navigation route is formed by continuous links in a series of network topologies, and the link data unit comprises: 1. geographical position information of the road, such as longitude and latitude coordinates, length, road grade, speed limit condition, administrative dependent information and the like; 2. road topology information such as upstream and downstream link, peripheral facilities, etc.; 3. real-time information such as the traffic speed of the vehicle at a certain moment, road condition status, the traffic speed of the vehicle at a historical moment, etc.

After obtaining the high quality navigation route, processing may be performed to obtain pre-trained route data.

Specifically, the following design may be made as input samples of the pre-training model:

first, link id strings are used as expressions for route samples. For example, if the link expression is generated according to the ID of the link, the route sample R may be expressed as: route r= { link ₁ ,ink ₂ ,ink ₃ … }; wherein link is ₁ ,ink ₂ ,ink ₃ … each represent the ID of the corresponding link.

The link attribute combination mode can also be used as the expression of the route sample for different business scenes, for example, the route sample r= { road class_road condition_flow … }, specifically expressed as: route r= { rc_rs_flow ₁ ,c_rs_flow ₂ ,c_rs_flow ₃ … }; wherein rc __ flow ₁ Representing link ₁ Road class, road condition status and flow; rc __ flow ₂ Representing link ₂ Road class, road condition status and flow; rc __ flow ₃ Representing link ₃ Road class, road condition status and traffic volume.

It should be noted that: for route R of different expression patterns, the pre-training will produce different emmbedding results.

Next, as shown in fig. 6, the route R is interrupted at any intersection of the route R, or the route R may be interrupted between any two links, so as to form a first half and a second half of the route. And then adding a flag bit CLS before the initial link of the route R, and adding a flag bit SEP after the break and the end link. Then an input sample R of the pre-trained model is obtained _sample ：

R _sample ＝{CLS,link ₁ ,link ₂ ,link ₃ ,SEP,link ₄ ,link ₅ ,link ₆ ,SEP}

After obtaining the input samples of the pre-trained model, a pre-training process of the model may be performed, as shown in fig. 5 at step S501.

In one embodiment of the present application, a BERT (Bidirectional Encoder Representations from Transformers, transducer-based bi-directional encoder representation) model may be employed in pre-training. The input training samples of the BERT model are the first half and the second half of a sentence, and may correspond to the first half and the second half of the route in the embodiment of the present application, and links in the route may correspond to a word in a sentence. Specifically, as shown in fig. 7, links in a route may have three types of unbinding: tokenEmbedding, segmentEmbedding and positioning. Wherein Segment Embeddings identifies whether a link is in the first half or the second half, and if so, can be denoted as E _A The method comprises the steps of carrying out a first treatment on the surface of the If located in the latter half, it can be denoted as E _B . Position Embeddings are used to represent the location characteristics of each link in the route.

The forward propagation of the BERT model in the embodiment of the present application is divided into two phases: the forward propagation of the transducer structure and the forward propagation of the two subtasks are described separately below.

The forward propagation process of the transducer structure is specifically as follows: in the Embedding stage, for each link and flag bit in the route sample, searching corresponding Token Embe in the Embedding datadding; searching corresponding Segment Embedding according to the route part of each link and the marker bit; and searching corresponding Position Embedding according to the route position of each link and the marker bit. Then the vectors of Token references, segment Embedding, position Embedding found by each link and flag are accumulated (in other embodiments of the present application, other arithmetic processing based on Token references, segment Embedding, position Embedding may be used) to obtain references of the link or flag in the route sample _input I.e. E shown in FIG. 8 ₁ 、E ₂ 、…、E _N 。

With continued reference to FIG. 8, an Embedding is used _input Multi-layer transducer forward propagation as input (two-layer example is shown in FIG. 8, and more layers are possible in other embodiments), with the last layer transducer output yielding an intermediate result vector transducer _out I.e. T as shown in FIG. 8 ₁ 、T ₂ 、…、T _N Embedding relies on the attentional mechanisms of the transducer structure _input Intermediate result transfer of corresponding position _out Link information of other positions on the route sample is fused.

The forward propagation process of the two subtasks is specifically: since the BERT model designed two subtasks, MLM (Mask Language Model ) and NSP (Next Sentence Prediction, next sentence prediction), respectively.

For the MLM task, first, for the input samples R _sample The treatment is specifically as follows: the first link of the sample does not operate; for each of the remaining links, as shown in FIG. 9, at P _op Is operated on by a link, the operated link is operated on by P _MASK The probability of (1) is set as a flag bit MASK, with P _RAND Is replaced by random link, with P _KEEP The probability of link is kept unchanged. Alternatively, P _op ＝50％，P _MASK ＝50％，P _RAND ＝5％，P _KEEP ＝45％。

The processed sample contains a marker bit MASK and a random link _RAND Each link and mark in MLM task input routeBit intermediate result vector transfer _out And calculating softmax for each MASK zone bit through a DNN (Deep Neural Network ) model to obtain probability distribution of all links on the MASK zone bit, and then comparing true value links set as the MASK zone bit, calculating gradient and back-propagating. Alternatively, in other embodiments of the present application, the DNN model may be replaced with other deep learning network models.

For NSP tasks, first for input sample R _sample Processing, in particular to the second half part of the route marked by the SEP marker bit, by P _op Probability replacement with random lower part route, optionally P _op =50%. Intermediate result vector transfer of NSP task input route start position CLS flag bit _out Calculating softmax through DNN model to obtain the route of the second half part belonging to R _sample Is compared to determine if it is replaced, and the gradient is calculated and back-propagated.

Through pre-training, the converged Token Embedding, segment Embedding, position Embedding results, and each layer of transducer structure can be optimized in a large number of route samples.

The fine tuning procedure of step S502 shown in fig. 5 specifically makes the following processing for the route service: first, the route-level ebedding expression is composed of the ebedding of the starting link, the route average ebedding, and the ebedding of the ending link in the route. The method is specifically as follows:

Embedding _start ＝∑Token Embedding _start ,Segment Embedding _start ,Position Embedding _start

Embedding _end ＝ΣToken Embedding _end ,Segment Embedding _end ,Position Embedding _end

wherein, tokenEmbedding _start 、SegmentEmbedding _start And a positioning element _start Is the initial linkembedding；TokenEmbedding _end 、SegmentEmbedding _end And a positioning element _end Is the ebedding of terminating link; tokenEmbedding _i 、SegmentEmbedding _i And a positioning element _i Is the ebedding of the ith link; n represents the number of links in the route. In other embodiments of the present application, route level ebedding may also be based on other arithmetic processing results of Token Embedding, segment Embedding, position Embedding vectors.

And secondly, obtaining route characteristics of the artificial characteristic engineering design and a pre-estimated result output by a machine learning method of the decision tree model, and taking the two parts as input information of the fine tuning model.

Finally, the historical driving track and the coverage rate T of the navigation route are used as quantification standards of route rationality to be used as optimization targets of the model for fine adjustment, and the more reasonable route has higher coverage rate, wherein T is E (0, 1). The fine tuning process is shown in FIG. 10, where the route level is the Embedding vector (i.e., embedding _start 、Embedding _avg And Embedding _end ) Respectively through multi-layer perceptron MLP ₁ Transformed to a D1 vector. Linking Embedding _start Embeddinggavg and Embeddenged vectors through multi-layer perceptron MLP2, MLP ₃ 、MLP ₄ Transformed to a D2 vector, a D3 vector and a D4 vector. And then taking the manually extracted features and the prediction result of the tree model as input, and linking D4 to construct a D5 vector. Through multilayer perceptron MLP ₅ 、MLP ₆ And MLP ₇ The transformation results in a D6 vector, a D7 vector and a D8 vector. Optionally, d1=128, d2=1536, d3=384, d4=8, d5= 8+N (features) +1, D6=64, d=16, d8=1. And D8 vector is a 1-dimensional vector, namely, outputting a route rationality estimation result. The number of layers of the multi-layer perceptron shown in fig. 10 is merely exemplary, and more layers are possible in other embodiments of the present application.

In the above technical scheme of the application, the result of Token encapsulation, segment Embedding, position Embedding obtained by pre-training can make the reasonable composition of the whole route fully considered in the process of route rationality estimation. In a machine learning scheme different from general manual feature extraction, route detail information can be lost in a prediction process due to route global feature design. According to the technical scheme, the higher-rationality route is provided with a better route part based on the attention mechanism in the transducer structure. In addition, the training result obtained through pre-training in the embodiment of the application can be used for abnormal route detection, route completion, automatic route generation, road correlation analysis, urban road planning and other businesses, and the function of predicting the subsequent route based on the preamble track can be realized. In addition, the technical scheme of the embodiment of the application not only can be used for carrying out route rationality estimation tasks, but also can be used for the services of route time estimation, route rationality quantification, route probability estimation, route flow estimation and the like by modifying the optimization targets of the pre-training subtasks.

The technical scheme of the embodiment is an unsupervised learning scheme in the pre-training stage, route data is data which is difficult to collect user feedback in a navigation service scene, and the scheme of the embodiment of the application can extract information of the navigation service facing the non-labeled data by using a pre-training method through massive non-labeled data. By pre-training the acquired information, a route rationality prediction model with better effect can be obtained by using smaller scale labeling data in a fine tuning model stage, and the problems of data under labeling, data noise, position-Bias and the like in the service field are effectively solved. In addition, the information acquired in the pre-training stage is rich and diversified, and the information trend depends on the structural design of the pre-training scheme, so that the pre-training model structure can be designed in a targeted manner according to different route learning targets. For example: time estimation of a route, quantification of route rationality, route probability estimation, route flow estimation and the like.

The following describes an embodiment of an apparatus of the present application, which may be used to perform the route processing method in the above-described embodiment of the present application. For details not disclosed in the embodiments of the apparatus of the present application, please refer to the embodiments of the route processing method described in the present application.

Fig. 11 shows a block diagram of a route processing device according to an embodiment of the present application.

Referring to fig. 11, a route processing device 1100 according to an embodiment of the present application includes: a first generation unit 1102, a feature extraction unit 1104, a second generation unit 1106, and a model training unit 1108.

The first generating unit 1102 is configured to generate a link expression corresponding to a planned route sample according to a plurality of links included in the planned route sample and attribute information of each link; the feature extraction unit 1104 is configured to extract a plurality of initial embedded vectors corresponding to respective elements in the link expression through a pre-training model; the second generating unit 1106 is configured to generate input feature vectors corresponding to the respective elements according to the plurality of initial embedded feature vectors; the model training unit 1108 is configured to perform multi-layer forward propagation processing according to the input feature vectors corresponding to the elements, and perform model training processing based on the planned route samples, training tasks corresponding to the pre-training models, and intermediate result vectors corresponding to the elements obtained by the multi-layer forward propagation processing, so as to process the specified route based on the converged pre-training modules.

In some embodiments of the present application, based on the foregoing solution, the first generating unit 1102 is configured to: generating elements which correspond to the links and are used for representing the attribute characteristics according to at least one attribute characteristic contained in the attribute information of the links; and combining elements respectively corresponding to the multiple links according to the sequence of the multiple links in the planned route sample to generate a link expression corresponding to the planned route sample.

In some embodiments of the present application, based on the foregoing solution, the first generating unit 1102 is configured to: according to the position of a specified link in the plurality of links in the planned route sample, adding a position marker bit corresponding to the specified link in the link expression; wherein the location identification bits include at least one of: a first flag bit added before an element corresponding to a start link in the planned route sample, a second flag bit added between elements corresponding to two specified links, and a third flag bit added after an element corresponding to a stop link in the planned route sample.

In some embodiments of the present application, based on the foregoing scheme, the feature extraction unit 1104 is configured to: extracting word embedding vectors, position embedding vectors and route interval embedding vectors corresponding to all elements in the link expression through a pre-training model; the position embedding vector is used for representing the specific position of the element in the link expression, and the route interval embedding vector is used for representing the interval of the element in the link expression.

In some embodiments of the present application, based on the foregoing scheme, the second generating unit 1106 is configured to: and based on a set merging processing mode, merging a plurality of initial embedded feature vectors corresponding to each element, and taking the merging processing result as an input feature vector corresponding to each element.

In some embodiments of the present application, based on the foregoing scheme, the model training unit 1108 is configured to: carrying out replacement operation on other links except the initial link in the planned route sample through set probability respectively to obtain a processed planned route sample; the replacing operation includes at least one of: the method comprises the steps of setting a flag bit, replacing a random link and keeping unchanged; and taking the intermediate result vector corresponding to each element as input, recovering the link represented by the set flag bit as an actual link in the planned route sample as an optimization target corresponding to a first training task, and performing model training processing.

In some embodiments of the present application, based on the foregoing scheme, the model training unit 1108 is configured to: replacing the links at the latter half of the planned route samples with random links through the set probability; the planned route sample is divided into a first half link and a second half link at two designated links; and taking an intermediate result vector corresponding to a first flag bit added before an element corresponding to an initial link in the link expression as input, recovering the random link to an optimization target corresponding to a second training task, wherein the second half link in the planned route sample is used as an optimization target, and performing model training processing.

In some embodiments of the present application, based on the foregoing solution, the route processing device 1100 further includes: a processing unit 1110, configured to extract target embedded vectors corresponding to the elements through the converged pre-training model; generating a route embedding vector corresponding to the planning route sample according to the target embedding vector corresponding to each element; and taking the route embedded vector and target input information for adjusting the pre-training model as input, taking a set service as an optimization target, and adjusting model parameters of the pre-training model.

In some embodiments of the present application, based on the foregoing scheme, the processing unit 1110 is configured to: calculating a target embedded vector mean value corresponding to the initial link according to a plurality of target embedded vectors corresponding to the initial link in the planned route sample; calculating to obtain a target embedded vector mean value corresponding to a plurality of links according to a plurality of target embedded vectors respectively corresponding to a plurality of links contained in the planned route sample; calculating to obtain a target embedded vector mean value corresponding to the terminated link according to a plurality of target embedded vectors corresponding to the terminated link in the planned route sample; and generating a route embedded vector corresponding to the planning route sample according to the target embedded vector average value corresponding to the initial link, the target embedded vector average values corresponding to the links and the target embedded vector average value corresponding to the termination link.

In some embodiments of the present application, based on the foregoing solution, the route processing device 1100 further includes: a first prediction unit; the first generating unit 1102 is further configured to: generating a link expression corresponding to a target planning route according to links contained in the target planning route to be processed and attribute information of each link; the feature extraction unit 1104 is further configured to: extracting target embedded vectors corresponding to all links in the target planning route through the converged pre-training model; the first prediction unit is configured to: and carrying out prediction processing of the setting service based on the target embedded vectors corresponding to the links in the target planning route.

In some embodiments of the present application, based on the foregoing solution, the route processing device 1100 further includes: a second prediction unit; the first generating unit 1102 is further configured to: generating a link expression corresponding to a target planning route according to links contained in the target planning route to be processed and attribute information of each link; the second prediction unit is configured to: taking the link expression as the input of a converged pre-training model, and outputting a predicted result of the target planning route aiming at the set service through the converged pre-training model; wherein the setting service includes at least one of the following: the coverage rate of the driving track and the target planned route, the driving time of the target planned route, the planning rationality of the target planned route, the navigation completion rate of the target planned route and the flow of the target planned route.

It should be noted that, the computer system 1200 of the electronic device shown in fig. 12 is only an example, and should not impose any limitation on the functions and the application scope of the embodiments of the present application.

As shown in fig. 12, the computer system 1200 includes a central processing unit (Central Processing Unit, CPU) 1201 which can perform various appropriate actions and processes, such as performing the methods described in the above embodiments, according to a program stored in a Read-Only Memory (ROM) 1202 or a program loaded from a storage section 1208 into a random access Memory (Random Access Memory, RAM) 1203. In the RAM 1203, various programs and data required for the system operation are also stored. The CPU 1201, ROM 1202, and RAM 1203 are connected to each other through a bus 1204. An Input/Output (I/O) interface 1205 is also connected to bus 1204.

The following components are connected to the I/O interface 1205: an input section 1206 including a keyboard, a mouse, and the like; an output portion 1207 including a Cathode Ray Tube (CRT), a liquid crystal display (Liquid Crystal Display, LCD), and a speaker, etc.; a storage section 1208 including a hard disk or the like; and a communication section 1209 including a network interface card such as a LAN (Local Area Network ) card, a modem, or the like. The communication section 1209 performs communication processing via a network such as the internet. The drive 1210 is also connected to the I/O interface 1205 as needed. A removable medium 1211 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is installed as needed on the drive 1210 so that a computer program read out therefrom is installed into the storage section 1208 as needed.

In particular, according to embodiments of the present application, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising a computer program for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication portion 1209, and/or installed from the removable media 1211. When executed by a Central Processing Unit (CPU) 1201, performs the various functions defined in the system of the present application.

It should be noted that, the computer readable medium shown in the embodiments of the present application may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-Only Memory (ROM), an erasable programmable read-Only Memory (Erasable Programmable Read Only Memory, EPROM), flash Memory, an optical fiber, a portable compact disc read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present application, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with a computer-readable computer program embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer program embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. Where each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer programs.

The units involved in the embodiments of the present application may be implemented by means of software, or may be implemented by means of hardware, and the described units may also be provided in a processor. Wherein the names of the units do not constitute a limitation of the units themselves in some cases.

As another aspect, the present application also provides a computer-readable medium that may be contained in the electronic device described in the above embodiment; or may exist alone without being incorporated into the electronic device. The computer readable medium carries one or more computer programs which, when executed by the electronic device, cause the electronic device to implement the methods described in the above embodiments.

It should be noted that although in the above detailed description several modules or units of a device for action execution are mentioned, such a division is not mandatory. Indeed, the features and functions of two or more modules or units described above may be embodied in one module or unit, in accordance with embodiments of the present application. Conversely, the features and functions of one module or unit described above may be further divided into a plurality of modules or units to be embodied.

From the above description of embodiments, those skilled in the art will readily appreciate that the example embodiments described herein may be implemented in software, or may be implemented in software in combination with the necessary hardware. Thus, the technical solution according to the embodiments of the present application may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (may be a CD-ROM, a usb disk, a mobile hard disk, etc.) or on a network, and includes several instructions to cause a computing device (may be a personal computer, a server, a touch terminal, or a network device, etc.) to perform the method according to the embodiments of the present application.

Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains.

It is to be understood that the present application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims

1. A route processing method, comprising:

generating a link expression corresponding to a planned route sample according to a plurality of links contained in the planned route sample and attribute information of each link;

extracting a plurality of initial embedded vectors corresponding to each element in the link expression through a pre-training model, and generating input feature vectors corresponding to each element according to the plurality of initial embedded feature vectors;

and carrying out multi-layer forward propagation processing according to the input feature vectors corresponding to the elements, and carrying out model training processing on the basis of the planned route samples, training tasks corresponding to the pre-training models and intermediate result vectors corresponding to the elements obtained by the multi-layer forward propagation processing so as to process the appointed route based on the converged pre-training modules.

2. The route processing method according to claim 1, wherein generating a link expression corresponding to a planned route sample based on a plurality of links included in the planned route sample and attribute information of each link, comprises:

generating elements which correspond to the links and are used for representing the attribute characteristics according to at least one attribute characteristic contained in the attribute information of the links;

and combining elements respectively corresponding to the multiple links according to the sequence of the multiple links in the planned route sample to generate a link expression corresponding to the planned route sample.

3. The route processing method according to claim 2, wherein generating a link expression corresponding to a planned route sample based on a plurality of links included in the planned route sample and attribute information of each link, further comprises:

according to the position of a specified link in the plurality of links in the planned route sample, adding a position marker bit corresponding to the specified link in the link expression;

wherein the location identification bits include at least one of: a first flag bit added before an element corresponding to a start link in the planned route sample, a second flag bit added between elements corresponding to two specified links, and a third flag bit added after an element corresponding to a stop link in the planned route sample.

4. A route processing method according to claim 3, wherein the specified two links include: links located at road intersections in the planned route samples, or any two adjacent links in the plurality of links.

5. The route processing method according to claim 1, wherein extracting a plurality of initial embedded vectors corresponding to respective elements in the link expression by a pre-training model includes: extracting word embedding vectors, position embedding vectors and route interval embedding vectors corresponding to all elements in the link expression through a pre-training model;

the position embedding vector is used for representing the specific position of the element in the link expression, and the route interval embedding vector is used for representing the interval of the element in the link expression.

6. The route processing method according to claim 1, wherein generating the input feature vector corresponding to the respective element from the plurality of initial embedded feature vectors includes:

and based on a set merging processing mode, merging a plurality of initial embedded feature vectors corresponding to each element, and taking the merging processing result as an input feature vector corresponding to each element.

7. The route processing method according to claim 1, wherein performing model training processing based on the planned route sample, the training task corresponding to the pre-training model, and the intermediate result vector corresponding to each element obtained by the multi-layer forward transmission processing includes:

carrying out replacement operation on other links except the initial link in the planned route sample through set probability respectively to obtain a processed planned route sample; the replacing operation includes at least one of: the method comprises the steps of setting a flag bit, replacing a random link and keeping unchanged;

and taking the intermediate result vector corresponding to each element as input, recovering the link represented by the set flag bit as an actual link in the planned route sample as an optimization target corresponding to a first training task, and performing model training processing.

8. The route processing method according to claim 7, wherein the replacing operation is performed by a set probability on links other than the start link in the planned route sample, respectively, comprising:

respectively determining whether replacement operation is to be performed on other links except the initial link in the planned route sample through a first probability;

For the links requiring replacement operation, the replacement operation is performed by the second probability of replacing with the set flag bit, the third probability of replacing with the random link and the fourth probability of remaining unchanged.

9. The route processing method according to claim 1, wherein performing model training processing based on the planned route sample, the training task corresponding to the pre-training model, and the intermediate result vector corresponding to each element obtained by the multi-layer forward transmission processing includes:

replacing the links at the latter half of the planned route samples with random links through the set probability; the planned route sample is divided into a first half link and a second half link at two designated links;

and taking an intermediate result vector corresponding to a first flag bit added before an element corresponding to an initial link in the link expression as input, recovering the random link to an optimization target corresponding to a second training task, wherein the second half link in the planned route sample is used as an optimization target, and performing model training processing.

10. The route processing method according to claim 1, wherein the processing of the specified route based on the converged pre-training module includes:

Extracting target embedded vectors corresponding to the elements through the converged pre-training model;

generating a route embedding vector corresponding to the planning route sample according to the target embedding vector corresponding to each element;

and taking the route embedded vector and target input information for adjusting the pre-training model as input, taking a set service as an optimization target, and adjusting model parameters of the pre-training model.

11. The route processing method of claim 10, wherein generating the route embedding vector corresponding to the planned route sample according to the target embedding vector corresponding to each element comprises:

calculating a target embedded vector mean value corresponding to the initial link according to a plurality of target embedded vectors corresponding to the initial link in the planned route sample;

calculating to obtain a target embedded vector mean value corresponding to a plurality of links according to a plurality of target embedded vectors respectively corresponding to a plurality of links contained in the planned route sample;

calculating to obtain a target embedded vector mean value corresponding to the terminated link according to a plurality of target embedded vectors corresponding to the terminated link in the planned route sample;

And generating a route embedded vector corresponding to the planning route sample according to the target embedded vector average value corresponding to the initial link, the target embedded vector average values corresponding to the links and the target embedded vector average value corresponding to the termination link.

12. The route processing method of claim 10, wherein the target input information for adjusting the pre-training model comprises at least one of:

route characteristics of the planned route samples obtained through characteristic engineering;

and according to the route characteristics of the planned route sample, carrying out prediction processing of the setting business through a machine learning model to obtain a prediction result.

13. The route processing method according to any one of claims 1 to 12, characterized in that the processing of the specified route based on the converged pre-training module includes:

generating a link expression corresponding to a target planning route according to links contained in the target planning route to be processed and attribute information of each link;

extracting target embedded vectors corresponding to all links in the target planning route through the converged pre-training model;

And carrying out prediction processing of the setting service based on the target embedded vectors corresponding to the links in the target planning route.

14. The route processing method according to claim 13, wherein the predicting process of the setting service based on the target embedded vector corresponding to each link in the target planned route includes:

and carrying out at least one of the following processing on the target planning route based on the target embedded vector corresponding to each link in the target planning route: abnormal link detection processing, completion processing of missing links in the target planned route, generation processing of a follow-up planned route and correlation processing of routes.

15. The route processing method according to any one of claims 1 to 12, characterized in that the processing of the specified route based on the converged pre-training module includes:

taking the link expression as the input of a converged pre-training model, and outputting a predicted result of the target planning route aiming at the set service through the converged pre-training model;

Wherein the setting service includes at least one of the following: the coverage rate of the driving track and the target planned route, the driving time of the target planned route, the planning rationality of the target planned route, the navigation completion rate of the target planned route and the flow of the target planned route.

16. A route processing device, characterized by comprising:

the first generation unit is configured to generate a link expression corresponding to the planned route sample according to a plurality of links contained in the planned route sample and attribute information of each link;

a feature extraction unit configured to extract a plurality of initial embedded vectors corresponding to respective elements in the link expression through a pre-training model;

a second generating unit configured to generate input feature vectors corresponding to the respective elements according to the plurality of initial embedded feature vectors;

and the model training unit is configured to perform multi-layer forward propagation processing according to the input feature vectors corresponding to the elements, perform model training processing on the basis of the planned route samples, training tasks corresponding to the pre-training models and intermediate result vectors corresponding to the elements obtained by the multi-layer forward propagation processing, and process the appointed route on the basis of the converged pre-training modules.

17. A computer readable medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the route processing method according to any one of claims 1 to 15.

18. An electronic device, comprising:

one or more processors;

a memory for storing one or more computer programs that, when executed by the one or more processors, cause the electronic device to implement the route processing method of any of claims 1-15.

19. A computer program product, characterized in that the computer program product comprises a computer program stored in a computer-readable storage medium, from which computer-readable storage medium a processor of an electronic device reads and executes the computer program, causing the electronic device to perform the route processing method according to any one of claims 1 to 15.