Training method and device of path prediction model and electronic equipment

Abstract

The invention provides a training method and device of a path prediction model and electronic equipment. The method comprises the steps of obtaining sample origin-destination data of a sample vehicle driving from a sample starting point to a sample ending point in a road network, wherein the sample origin-destination data comprises sample starting point data, sample ending point data and sample driving time of the sample vehicle driving from the sample starting point to the sample ending point; based on the road network, at least one sample candidate path between the sample starting point and the sample ending point is determined, each sample candidate path comprises at least two sample nodes, the at least two sample nodes comprise the sample starting point and the sample ending point, and a sample road section is formed between every two adjacent sample nodes; determining the sample road section passing time of the sample vehicle in each sample road section and the standard deviation of the sample road section passing time based on the sample origin-destination data by using a Gaussian mixture model; constructing a training sample based on the sample road section passing time and the sample road section passing time standard deviation; and training a path prediction model based on the training sample.

Description

Technical Field

[0001] The present application relates to path restoration technology, and in particular to a training method, device and electronic device for a path prediction model. Background Art

[0002] Road vehicle driving path restoration plays an important role in many current scenarios, such as smart traffic control such as road intelligent expansion, active control, and highland linkage collaborative control. Highway vehicle driving path restoration is a necessary technical link for the current implementation of networked toll collection, and is also the data support for the analysis of highway network traffic supply and demand characteristics.

[0003] In the absence of continuous vehicle positioning coordinates, the basic data for path restoration is usually the vehicle passing records identified by cameras or ETC on the road. However, when the ETC gantry data is missing, the accuracy of path restoration is significantly reduced, especially in the extreme case where there are only origin / destination (OD) records and all vehicle passing data during driving is missing. The difficulty of path restoration is significantly increased, and the shortest path connecting the OD points is usually used as the vehicle's driving path. However, this technical solution is difficult to meet the demand for accurate restoration of vehicle paths. For this reason, it is necessary to seek a more reliable technical solution to achieve vehicle path restoration in scenarios where only OD data exists. Summary of the invention

[0004] The embodiments of the present application provide a training method, device, electronic device, storage medium and computer program product for a path prediction model, which can improve the reliability of vehicle path restoration.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present application provides a method for training a path prediction model, including:

[0007] Obtaining sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point;

[0008] Based on the road network, at least one sample candidate path between the sample starting point and the sample end point is determined, each sample candidate path includes at least two sample nodes, the at least two sample nodes include the sample starting point and the sample end point, and a sample road section is formed between two adjacent sample nodes;

[0009] Using a Gaussian mixture model, based on the sample start and end data, determine the sample section travel time and the sample section travel time standard deviation of the sample vehicle in each sample section;

[0010] Constructing training samples based on the sample road section travel time and the sample road section travel time standard deviation;

[0011] Based on the training samples, training a path prediction model;

[0012] The path prediction model is used to predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's starting and ending data, so as to determine the vehicle's driving path.

[0013] In the above scheme, the Gaussian mixture model is used to determine the sample travel time of the sample vehicle in each sample road section and the standard deviation of the sample road section travel time based on the sample start and end data, including:

[0014] Constructing a travel time probability model for a sample road segment in the at least one sample candidate path;

[0015] The Gaussian mixture model is used to solve the parameters of the travel time probability model based on the sample travel time in the sample start-end data to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

[0016] In the above scheme, the Gaussian mixture model is used to solve the parameters based on the travel time probability model to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including:

[0017] Determining a sample shortest path of the at least one sample candidate path;

[0018] Based on the path length of the sample shortest path and the sample travel time, the sample section travel time and the sample section travel time standard deviation of the sample section in the sample shortest path are initialized to obtain the sample initial time and the initial standard deviation;

[0019] Based on the travel time probability model, the sample initial time and the initial standard deviation, the Q function of the Gaussian mixture model is used to iteratively calculate the sample section travel time and the sample section travel time standard deviation of each sample section.

[0020] In the above scheme, based on the path length of the sample shortest path and the sample travel time, the sample section travel time and the sample section travel time standard deviation of the sample section in the sample shortest path are initialized to obtain the sample initial time and the initial standard deviation, including:

[0021] Determine the average travel speed of the sample vehicle on each sample road section on the sample shortest path based on the path length of the sample shortest path and the sample travel time;

[0022] Obtaining the section length of the sample section in the sample shortest path;

[0023] Determining the road section travel time of the sample vehicle on each sample road section based on the road section length and the average travel speed;

[0024] Based on the road segment travel time on each sample road segment, calculate the expected and standard deviation of the travel time of each road segment;

[0025] The calculated expectation is used as the sample initial time, and the calculated standard deviation is used as the initial standard deviation.

[0026] In the above scheme, based on the travel time probability model, the sample initial time and the initial standard deviation, the Q function of the Gaussian mixture model is used to perform iterative calculation to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including:

[0027] Get the speed limit interval of the road network;

[0028] Calculate the time interval of the road section travel time based on the speed limit interval and the section length of each sample road section;

[0029] Based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval, the Q function of the Gaussian mixture model is used to perform iterative calculations to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

[0030] In the above scheme, the travel time probability model is:

[0031]

[0032] Where y is the observation time, is the average travel time vector of all sample sections in at least one sample candidate path; is the standard deviation vector of travel time of all sample road segments in at least one sample candidate path; is a distance set of at least one candidate path; R is a set consisting of at least one sample candidate path; λ k The probability of selecting path k from at least one candidate path; is the selection probability of path k in at least one sample candidate path corresponding to the sample start and end data corresponding to the observation time y;

[0033]

[0034] Where C() is the utility function; is the parameter of the utility function; d k is the length of path k; dj is the length of path j;

[0035]

[0036] in, is the expected function of path k; is the time variance of path k;

[0037]

[0038] Among them, μ O is the average driving time of the sample vehicle on the starting sample road section of path k that includes the sample starting point O; μ D μ is the average driving time of the sample vehicle on the terminal sample section of path k that includes the sample terminal D; s is the average driving time of the sample vehicle on the complete sample road segment s in path k;

[0039]

[0040] Among them, σ O is the standard deviation of the driving time of the initial sample road section; σ D is the standard deviation of the travel time of the terminal sample section; σ s is the standard deviation of the travel time for the complete sample road segment.

[0041] In the above scheme, based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval, the Q function of the Gaussian mixture model is used to perform iterative calculation to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including:

[0042] Construct an expected model in which the path k is selected, and the expected model is:

[0043]

[0044] The sample initial time is taken as The initial standard deviation is taken as Substitute the value of into the expected model to obtain the expectation of path k being selected;

[0045] The Q function is calculated based on and The maximum value of , the Q function is:

[0046]

[0047] The maximum value of the Q function is:

[0048]

[0049] in, is the lower limit value in the time interval, is the upper limit value in the time interval;

[0050] Iteratively calculate the maximum value and update it during the iterative calculation process and The value of until the Q function converges;

[0051] The last updated The value is determined as the travel time of the sample road section, and the last updated The value is determined as the standard deviation of the travel time of the sample road section.

[0052] In the above solution, the construction of training samples based on the sample road section travel time and the sample road section travel time standard deviation includes:

[0053] Determine at least one training sample path based on the road network;

[0054] Based on the sample road section travel time and the sample road section travel time standard deviation, determining the distribution of the training sample travel time of each training sample road section in the training sample path;

[0055] Based on the distribution of the training sample driving duration, a driving simulation is performed on each training sample path to obtain a simulated driving arrival time corresponding to each training sample node in the corresponding training sample path;

[0056] The training samples are obtained based on the training sample nodes corresponding to the training sample paths and the corresponding simulated travel arrival times.

[0057] In the above solution, the training path prediction model based on the training samples includes:

[0058] Inputting the first training sample node at the first position and the second training sample node at the last position in each of the training sample paths into the path prediction model;

[0059] Predicting the intermediate nodes between the first training sample node and the second training sample node by using the path prediction model to obtain predicted intermediate nodes;

[0060] Determine an error between a training sample intermediate node and the prediction intermediate node, the training sample intermediate node being a training sample node between the first training sample node and the second training sample node;

[0061] Based on the error, model parameters of the path prediction model are updated.

[0062] In the above scheme, the method further comprises:

[0063] Obtain the starting and ending data of the vehicle traveling from the starting point to the end point;

[0064] Based on the road network, determining a plurality of candidate paths between the starting point and the end point;

[0065] Using the path prediction model, based on the start and end data, predict the intermediate nodes between the start point and the end point to obtain a target number of predicted intermediate nodes;

[0066] Obtaining a predicted path of the vehicle based on the starting point, the end point, and the target number of predicted intermediate nodes;

[0067] The similarities between the predicted path and a plurality of candidate paths are respectively determined, and the candidate paths whose similarities meet a similarity condition are determined as the driving paths of the vehicle.

[0068] In the above solution, determining the similarity between the predicted path and multiple candidate paths includes:

[0069] Calculating the similarity between the predicted path and multiple candidate paths by using a Levenshtein similarity function to obtain the similarity;

[0070] Wherein, the Levenshtein similarity function is:

[0071]

[0072] Among them, S1 is the predicted path, S2 is the candidate path, is the Levenshtein distance between S1 and S2, and ξ are the labels of the sections in the corresponding paths.

[0073] In the above solution, determining multiple candidate paths between the starting point and the end point in the road network includes:

[0074] Obtaining a road node set and a connection attribute data set of a road network, wherein the road node set includes a plurality of road nodes, and the connection attribute data set includes connection relationships between the road nodes;

[0075] The relationship between the starting point, the end point, the road node set and the connection attribute data set is utilized, and the multiple candidate paths are determined using a shortest path algorithm.

[0076] In the above scheme, the method for determining the target quantity includes:

[0077] Determining the number of intermediate nodes of each candidate path among the multiple candidate paths;

[0078] A maximum number among the numbers of intermediate nodes is determined as the target number.

[0079] The present application provides a training device for a path prediction model, including:

[0080] An acquisition module, used to obtain sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point;

[0081] A path determination module, configured to determine, based on the road network, at least one sample candidate path between the sample starting point and the sample end point, wherein each sample candidate path includes at least two sample nodes, wherein the at least two sample nodes include the sample starting point and the sample end point, and a sample road section is formed between two adjacent sample nodes;

[0082] A time determination module, for determining the sample section travel time and the sample section travel time standard deviation of the sample vehicle in each sample section based on the sample start and end data by using a Gaussian mixture model;

[0083] A construction module, used to construct a training sample based on the sample road section travel time and the sample road section travel time standard deviation;

[0084] A training module, used for training a path prediction model based on the training samples;

[0085] The path prediction model is used to predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's starting and ending data, so as to determine the vehicle's driving path.

[0086] An embodiment of the present application provides an electronic device, including:

[0087] A memory for storing executable instructions;

[0088] The processor is used to implement the path prediction model training method provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0089] An embodiment of the present application provides a computer-readable storage medium storing executable instructions for causing a processor to execute and implement the path prediction model training method provided in the embodiment of the present application.

[0090] An embodiment of the present application provides a computer program product storing a computer program for implementing the path prediction model training method provided in the embodiment of the present application when executed by a processor.

[0091] The embodiment of the present application obtains sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point. Based on the road network, at least one sample candidate path between the sample start point and the sample end point is determined, wherein each sample candidate path includes at least two sample nodes, wherein the at least two sample nodes include the sample start point and the sample end point, and a sample road section is formed between two adjacent sample nodes. A Gaussian mixture model is used to determine the sample road section travel time and the sample road section travel time standard deviation of the sample vehicle in each sample road section based on the sample start-end data. Based on the sample road section travel time and the sample road section travel time standard deviation, a training sample is constructed. Based on the training sample, a path prediction model is trained to obtain a path prediction model with high accuracy. By using the path prediction model, the vehicle's travel path can be determined based only on the vehicle's start-end data without relying on an identification device, thereby avoiding the risk of the identification device being missing or not being able to restore the path on time, thereby improving the reliability of the path restoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 is an optional structural diagram of an electronic device 100 provided in an embodiment of the present application;

[0093] Figure 2 This is an optional flowchart of a path prediction model training method provided in an embodiment of the present application;

[0094] Figure 3 This is an optional detailed flow chart of step 203 of the path prediction model training method provided in the embodiment of the present application;

[0095] Figure 4 It is an optional schematic diagram of a candidate path provided in an embodiment of the present application. DETAILED DESCRIPTION

[0096] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0097] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0098] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0099] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0100] The embodiments of the present application provide a training method, device, electronic device and computer-readable storage medium for a path prediction model, which can improve the reliability of path restoration.

[0101] First, an electronic device for implementing the training method of the above-mentioned path prediction model provided in an embodiment of the present application is described.

[0102] See also Figure 1 , Figure 1 It is an optional structural diagram of the electronic device 100 provided in the embodiment of the present application. In practical applications, the electronic device 100 can be implemented as a terminal or a server. Among them, the terminal can be a laptop, a tablet computer, a desktop computer, a smart phone, a dedicated messaging device, a portable gaming device, a smart speaker, a smart watch, etc., but is not limited to this. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution network (CDN, Content Delivery Network) services, and big data and artificial intelligence platforms. Figure 1 The electronic device 100 shown includes: at least one processor 101, a memory 105, at least one network interface 102 and a user interface 103. The various components in the electronic device 100 are coupled together via a bus system 104. It is understood that the bus system 104 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 104 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 104 is not described in detail. Figure 1 Various buses are labeled as bus system 104 .

[0103] The processor 101 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0104] The user interface 103 includes one or more output devices 1031 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 103 also includes one or more input devices 1032, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0105] The memory 105 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 105 may optionally include one or more storage devices that are physically remote from the processor 101.

[0106] The memory 105 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 105 described in the embodiments of the present application is intended to include any suitable type of memory.

[0107] In some embodiments, the memory 105 can store data to support various operations. Examples of these data include programs, modules, and data structures, or subsets or supersets thereof. In the embodiment of the present application, the memory 105 stores an operating system 1051, a network communication module 1052, a presentation module 1053, an input processing module 1054, and a training device 1055 for a path prediction model; specifically,

[0108] Operating system 1051, including system programs for processing various basic system services and performing hardware-related tasks, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0109] A network communication module 1052, used to reach other computing devices via one or more (wired or wireless) network interfaces 102, exemplary network interfaces 102 include: Bluetooth, Wireless Compatibility Certification (WiFi), and Universal Serial Bus (USB);

[0110] a presentation module 1053 for enabling presentation of information (e.g., a user interface for operating peripherals and displaying content and information) via one or more output devices 1031 (e.g., a display screen, a speaker, etc.) associated with the user interface 103;

[0111] The input processing module 1054 is configured to detect one or more user inputs or interactions from one of the one or more input devices 1032 and to translate the detected inputs or interactions.

[0112] In some embodiments, the training device of the path prediction model provided in the embodiments of the present application can be implemented in software. Figure 1 The training device 1055 of the path prediction model stored in the memory 105 is shown, which can be software in the form of a program and a plug-in, etc., including the following software modules: an acquisition module 10551, a path determination module 10552, a time determination module 10553, a construction module 10554 and a training module 10555. These modules are logical, so they can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be explained below.

[0113] In other embodiments, the training device of the path prediction model provided in the embodiments of the present application can be implemented in hardware. As an example, the training device of the path prediction model provided in the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the training method of the path prediction model provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can adopt one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs) or other electronic components.

[0114] The following will describe the training method of the path prediction model provided in the embodiment of the present application in combination with the exemplary application and implementation of the terminal provided in the embodiment of the present application.

[0115] See also Figure 2 , Figure 2 is an optional flow chart of the training method of the path prediction model provided in the embodiment of the present application, which will be combined with Figure 2 The steps shown are explained.

[0116] Step 201, obtaining sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point;

[0117] Step 202: determining at least one sample candidate path between the sample starting point and the sample end point based on the road network, each sample candidate path includes at least two sample nodes, the at least two sample nodes include the sample starting point and the sample end point, and a sample road section is formed between two adjacent sample nodes;

[0118] Step 203, using a Gaussian mixture model, based on the sample start and end data, determining the sample section travel time and the sample section travel time standard deviation of the sample vehicle in each sample section;

[0119] Step 204, constructing a training sample based on the sample road section travel time and the sample road section travel time standard deviation;

[0120] Step 205: training a path prediction model based on the training samples;

[0121] The path prediction model is used to predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's starting and ending data, so as to determine the vehicle's driving path.

[0122] In actual implementation, the sample start and end data of the sample vehicle traveling from the sample starting point to the sample end point are obtained, and the sample start and end data include the sample starting point data, the sample end point data and the sample travel time of the sample vehicle traveling from the sample starting point to the sample end point. Next, the topological structure of the road network where the vehicle is located is obtained. Here, the road network can be a highway, or a city road, etc. Determine multiple sample candidate paths between the sample starting point and the sample end point in the road network. Here, based on the road network topological structure, the road node set P and the connection attribute data set S between the nodes are obtained. For any OD data in the road network, according to the relationship between the starting and ending points and P, S, a candidate path set is constructed using the k-shortest path algorithm.

[0123] Then, based on the topological structure of the road network, at least one sample candidate path between the sample starting point and the sample end point is determined. Specifically, at least one sample candidate path between the sample starting point and the sample end point can be determined based on the candidate path set. Each sample candidate path includes at least two sample nodes, and the at least two sample nodes include the sample starting point and the sample end point, and a sample road section is formed between two adjacent sample nodes.

[0124] Next, the Gaussian mixture model is used to determine the sample section travel time and the standard deviation of the sample section travel time of the sample vehicle in each sample section based on the sample start and end data. Here, the Gaussian mixture model is to accurately quantify things using the Gaussian probability density function (normal distribution curve). It is a model that decomposes things into several models based on the Gaussian probability density function (normal distribution curve). In the embodiment of the present application, the Gaussian mixture model is used to calculate the sample section travel time and the standard deviation of the sample section travel time of the vehicle in each sample section based on at least one sample candidate path between the sample starting point and the sample end point in the sample start and end data, and the sample travel time of the sample vehicle from the sample starting point to the sample end point.

[0125] After obtaining the sample road section travel time and the sample road section travel time standard deviation, a training sample is constructed. Here, the training sample includes multiple training sample paths. Each training sample path is composed of multiple training sample nodes, and the training sample also includes the sample time point corresponding to each training sample node (i.e., the simulated driving arrival time described later). Based on the training samples, a path prediction model is trained. The trained path prediction model can predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's start and end data, thereby determining the vehicle's driving path. Here, the vehicle's start and end data include the vehicle's starting point, end point, and the driving time for the vehicle to travel from the starting point to the end point. In one embodiment, the vehicle's start and end data can also report the time when the vehicle arrives at the starting point and the time when the vehicle arrives at the end point.

[0126] The embodiment of the present application obtains sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point. Based on the road network, at least one sample candidate path between the sample start point and the sample end point is determined, wherein each sample candidate path includes at least two sample nodes, wherein the at least two sample nodes include the sample start point and the sample end point, and a sample road section is formed between two adjacent sample nodes. A Gaussian mixture model is used to determine the sample road section travel time and the sample road section travel time standard deviation of the sample vehicle in each sample road section based on the sample start-end data. Based on the sample road section travel time and the sample road section travel time standard deviation, a training sample is constructed. Based on the training sample, a path prediction model is trained to obtain a path prediction model with high accuracy. By using the path prediction model, the vehicle's travel path can be determined based only on the vehicle's start-end data without relying on an identification device, thereby avoiding the risk of the identification device being missing or not being able to restore the path on time, thereby improving the reliability of the path restoration.

[0127] In some embodiments, see Figure 3 , Figure 3 This is an optional detailed flow chart of step 203 of the path prediction model training method provided in an embodiment of the present application. Step 203 includes:

[0128] Step 2031, constructing a travel time probability model for a sample road segment in the at least one sample candidate path;

[0129] Step 2032, using a Gaussian mixture model, based on the sample travel time in the sample start-end data, solve the parameters of the travel time probability model to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

[0130] In some embodiments, the transit time probability model is:

[0131]

[0132] Where y is the observation time, is the average travel time vector of all sample sections in at least one sample candidate path; is the standard deviation vector of travel time of all sample road segments in at least one sample candidate path; is a distance set of at least one candidate path; R is a set consisting of at least one sample candidate path; λ k The probability of selecting path k from at least one candidate path; is the selection probability of path k in at least one sample candidate path corresponding to the sample start and end data corresponding to the observation time y;

[0133]

[0134] Where C() is the utility function; is the parameter of the utility function; d k is the length of path k; d j is the length of path j;

[0135]

[0136] in, is the expected function of path k; is the time variance of path k;

[0137]

[0138] Among them, μ O is the average driving time of the sample vehicle on the starting sample road section of path k that includes the sample starting point O; μ D μ is the average driving time of the sample vehicle on the terminal sample section of path k that includes the sample terminal D; sis the average driving time of the sample vehicle on the complete sample road segment s in path k;

[0139]

[0140] Among them, σ O is the standard deviation of the driving time of the initial sample road section; σ D is the standard deviation of the travel time of the terminal sample section; σ s is the standard deviation of the travel time for the complete sample road segment.

[0141] In actual implementation, specifically, for multiple sample start-end data in the road network, a travel time probability model for the i-th data in period t is constructed:

[0142]

[0143] Among them, R i is the path set involved in the OD corresponding to the i-th data. represents the OD corresponding to the i-th data, the probability of the vehicle choosing path k, and the calculation formula is:

[0144]

[0145] The utility function calculation formula is:

[0146] Among them, β1 and β2 are obtained by estimating the parameters of the multi-level discrete choice model (MNLogit) through historical observation results. The historical observation results here include but are not limited to the road network itself, and can also be other highway networks; The calculation formula is:

[0147]

[0148] in, This formula represents the path k corresponding to data i, and its expected function is equal to the sum of the expected time of the complete section corresponding to k and the partial time of the section where the starting and ending points are located.

[0149] For example, see Figure 4 , Figure 4 It is an optional schematic diagram of a candidate path provided in an embodiment of the present application, where the starting point of the vehicle is point O and the end point is point D. For the path k connecting point OD, its waypoints are B, C, and E, among which BC, CE, and EF are complete sections, and OB and FD are parts of sections AB and FG respectively; a complete section refers to a road connecting two adjacent nodes. is the average driving time from point O to point B in period t, is the average driving time from point F to point D in period t, is the average driving time of vehicles on the complete road section s in time period t; s is the route section corresponding to k;

[0150]

[0151] The above formula indicates that the variance of path k corresponding to data i is equal to the sum of the variances of the corresponding road sections; represents the standard deviation vector of travel time for all sections on the road network during period t, represents the standard deviation of the driving time from point O to point B in period t, represents the standard deviation of the driving time from F to D in period t, represents the standard deviation of the vehicle's driving time on the complete road section s during the t period; where η o The value is the ratio of the length of OB to that of AB, η D The value is the ratio of the length of FD to that of FG.

[0152] In some embodiments, the use of a Gaussian mixture model, the use of a Gaussian mixture model, performing parameter solving based on the travel time probability model to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, includes: determining the sample shortest path of the at least one sample candidate path; initializing the sample section travel time and the sample section travel time standard deviation of the sample section in the sample shortest path based on the path length of the sample shortest path and the sample travel time to obtain the sample initial time and the initial standard deviation; based on the travel time probability model, the sample initial time and the initial standard deviation, iteratively calculating using the Q function of the Gaussian mixture model to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

[0153] Here, specifically, it can be an iterative calculation of the maximum value of the Q function of the Gaussian mixture model. The sample shortest path can be the shortest path among at least one sample candidate path. In the embodiment of the present application, the sample section travel time and the sample section travel time standard deviation obtained based on the sample shortest path are used as the sample initial time and the initial standard deviation, and the values ​​of the two in the travel time probability model are initialized. In actual implementation, after determining the sample shortest path, based on the travel time probability model, the sample initial time and the initial standard deviation, the maximum value of the Q function of the Gaussian mixture model is used for iterative calculation, and the section travel time and the travel time standard deviation of each section obtained after the last iterative calculation are used as the sample section travel time and the sample section travel time standard deviation.

[0154] In some embodiments, based on the path length of the sample shortest path and the sample travel time, the sample section travel time and the standard deviation of the sample section travel time of the sample section in the sample shortest path are initialized to obtain the sample initial time and the initial standard deviation, including: determining the average travel speed of the sample vehicle on each sample section on the sample shortest path based on the path length of the sample shortest path and the sample travel time; obtaining the section length of the sample section in the sample shortest path; determining the section travel time of the sample vehicle on each sample section based on the section length and the average travel speed; calculating the expectation and standard deviation of the travel time of each section based on the section travel time on each sample section; using the calculated expectation as the sample initial time, and using the calculated standard deviation as the initial standard deviation.

[0155] In actual implementation, the parameters Initialization, the initialization method is as follows:

[0156] For the starting and ending points OD of data i, obtain the corresponding shortest path, whose length is d i , the observation time is Δt i , ignoring the speed fluctuation, the average speed of i on each section of the shortest path is The length of any section s along its route is Among them l s is the length of road section s; all data are processed according to the processing method of data i to obtain the duration set of any road section s; for the vehicle driving duration data corresponding to each road section, the 3σ strategy is repeatedly implemented to clean up the outliers until all retained data are within a reasonable range; based on the cleaned data, the expectation and standard deviation of the duration of each road section are calculated, which is used as the initial and The value of .

[0157] In some embodiments, based on the travel time probability model, the sample initial time and the initial standard deviation, the Q function of the Gaussian mixture model is used for iterative calculation to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including: obtaining the speed limit interval of the road network; calculating the time interval of the section travel time based on the speed limit interval and the section length of each sample section; based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval, the Q function of the Gaussian mixture model is used for iterative calculation to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

[0158] In actual implementation, the speed threshold of the road section can be determined as a lower limit of 20km / h and an upper limit of 135km / h based on the speed limit range of 60-120km / h on expressways, taking into account speeding when the road conditions are good and slow driving when the road is congested. The lower limit of the road section driving time can be calculated based on the length of the road section. And upper limit The upper limit value and the lower limit value constitute the time interval of the road section travel time.

[0159] In some embodiments, the Q function of the Gaussian mixture model is used to perform iterative calculation based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including:

[0160] Construct an expected model in which the path k is selected, the expected model is:

[0161]

[0162] The sample initial time is taken as The initial standard deviation is taken as Substitute the value of into the expected model to obtain the expectation of path k being selected;

[0163] The Q function is calculated based on and The maximum value of , the Q function is:

[0164]

[0165] The maximum value of the Q function is:

[0166]

[0167] in, is the lower limit value in the time interval, is the upper limit value in the time interval;

[0168] Iteratively calculate the maximum value and update it during the iterative calculation process and until the Q function converges;

[0169] The last updated The value is determined as the travel time of the sample road section, and the last updated The value is determined as the standard deviation of the travel time of the sample road section.

[0170] In actual implementation, for data i, the probability of selecting path k is Under these conditions, the time of observation The corresponding probability is Then the expectation of k being selected is expressed as:

[0171]

[0172] This parameter can be initialized by and Directly calculated.

[0173] Then, find the function Q for and The maximum value of . Among them, the Q function is expressed as:

[0174]

[0175] In actual situations, due to speed limit regulations, the travel time range of the road section is bounded, that is, the above time interval. The time corresponding to the highest speed is The time required to pass the minimum speed is The task of this step can be expressed as:

[0176]

[0177] Next, initialize point x0, set the allowable error ε>0, store the most recent number of iterations u=6, the number of iterations j=0, and the maximum number of iterations j max , the Hessian matrix is ​​initialized to the unit matrix, that is, H0 = I, and the initial value of the gradient change when When , return the optimal solution x j+1 , otherwise calculate the feasible direction p of this iteration j =-r j , and based on the step size ρ j > 0, perform a one-dimensional search for the following formula:

[0178]

[0179] Then, according to x j+1 =x j +ρ j p j Update weight x; when j>u, keep the vector pairs with the most recent u times and delete (z j-u ,c j-u ) range information; update z j =x j+1 -x j and Calculate the inverse of the Hessian matrix, that is Further calculation to obtain the gradient change Update j=j+1, when j>j maxWhen , the algorithm stops, otherwise it returns to the previous step. Finally, update and Repeat the above steps until the Q value converges.

[0180] In some embodiments, constructing training samples based on the sample road section travel time and the sample road section travel time standard deviation includes: determining at least one training sample path based on the road network; determining the distribution of training sample travel time of each training sample road section in the training sample path based on the sample road section travel time and the sample road section travel time standard deviation; performing driving simulation on each training sample path based on the distribution of the training sample travel time to obtain the simulated driving arrival time corresponding to each training sample node in the corresponding training sample path; obtaining the training sample based on each training sample node corresponding to each training sample path and the corresponding simulated driving arrival time.

[0181] In the embodiment of the present application, after obtaining the sample section travel time and the standard deviation of the sample section travel time of each section, a training sample is generated. Specifically, for the elements in the road node set P, a pairwise combination operation is performed. Note that there are len(P) elements in P. After the pairwise combination operation, combinations.

[0182] For any combination, the dynamic multi-path planning (KSP) algorithm is executed, K=3, if the number of paths is less than 3, the first 2 are taken, if less than 2, only 1 is taken; thus, Paths, K i ≤3; For each path, in order to reduce the impact of outliers on the results, the driving time d is randomly generated on the road section s. s When setting s It obeys uniform distribution, that is μ s , σ s are the mean and standard deviation of the travel time of road section s respectively; in order to increase the amount of data, the travel time of each path is simulated M times, and finally The data are used for training and testing, and each data is as follows:

[0183] [(node_id_1,time_1),(node_id_2,time_2),…,(node_id_N,time_N)];

[0184] Among them, N is the number of nodes on this path, and the N of each data is not necessarily equal. time_1=0, time_j-time_j-1 is the driving time of the section between node_id_j-1 and node_id_j, j∈[1,N].

[0185] In some embodiments, the training path prediction model based on the training samples includes: inputting the first training sample node at the first position and the second training sample node at the last position in each training sample path into the path prediction model; predicting the intermediate node between the first training sample node and the second training sample node through the path prediction model to obtain a predicted intermediate node; determining the error between the training sample intermediate node and the predicted intermediate node, the training sample intermediate node being the training sample node between the first training sample node and the second training sample node; and updating the model parameters of the path prediction model based on the error.

[0186] The embodiment of the present application realizes the prediction of intermediate nodes of the driving path by constructing a BI-LSTM model, wherein the training set data of the model is obtained by constructing a road section observation time probability model and performing statistical analysis based on a large amount of historical data; this method can solve the problem of low path restoration accuracy caused by missing driving process data.

[0187] In some embodiments, the method also includes: obtaining the start and end data of the vehicle traveling from the starting point to the end point; based on the road network, determining multiple candidate paths between the starting point and the end point; using the path prediction model, based on the start and end data, predicting the intermediate nodes between the starting point and the end point, and obtaining a target number of predicted intermediate nodes; based on the starting point, the end point and the target number of predicted intermediate nodes, obtaining the predicted path of the vehicle; respectively determining the similarity between the predicted path and multiple candidate paths, and determining the candidate path whose similarity meets the similarity condition as the driving path of the vehicle.

[0188] In some embodiments, determining multiple candidate paths between the starting point and the end point in a road network includes: obtaining a road node set and a connection attribute data set of the road network, wherein the road node set includes multiple road nodes, and the connection attribute data set includes a connection relationship between each road node; utilizing the relationship between the starting point, the end point, the road node set and the connection attribute data set, and utilizing a shortest path algorithm to determine the multiple candidate paths.

[0189] In actual implementation, the starting and ending data of the vehicle traveling from the starting point to the end point are obtained. Then, the topological structure of the road network where the vehicle is located is obtained. Here, the road network can be a highway or a city road. Multiple candidate paths between the starting point and the end point in the road network are determined.

[0190] In some embodiments, the method for determining the target number includes: determining the number of intermediate nodes of each candidate path in the plurality of candidate paths; and determining the maximum number of the intermediate nodes as the target number. Identify the maximum number of intermediate nodes N included in the candidate path set, denoted as N max , N max Determine the target quantity.

[0191] In actual implementation, construct max The node sequence {Node} and time series {Time}, for the number of nodes less than N max In the case of n, the filling method is used, that is, the number of nodes is n<N max When , the first n nodes are consistent with the original sequence, and the n+1 to N max The nodes are all set to the nth node Node n ; The time series setting method is similar to this, just replace the node with time. On this basis, construct a BI-LSTM network, and use training samples to train the network to obtain a path intermediate node prediction model. When making predictions, based on the start and end data, the node where the starting point is located is taken as node_id_1, and the time after the starting point is taken as time_1; the node where the end point is located is taken as node_id_N, and the time after the end point is taken as time_N; (node_id_1, time_1) and (node_id_N, time_N) are input into the path intermediate node prediction model, and the intermediate node prediction value is output, and the predicted path is generated in combination with the start and end points.

[0192] In some embodiments, determining the similarity between the predicted path and multiple candidate paths includes: calculating the similarity between the predicted path and multiple candidate paths using a Levenshtein similarity function to obtain the similarity.

[0193] In order to further determine the actual driving path of the vehicle, the KSP algorithm is executed, k=10, and the predicted path obtained in the previous step is compared with the k paths by Levenshtein similarity;

[0194] The Levenshtein similarity function is:

[0195]

[0196] Among them, S1 is the predicted path, S2 is the candidate path, is the Levenshtein distance between S1 and S2, and ξ are the indexes of the path segment array respectively. The minimum value of both is 0, and the maximum value is the number of segments contained in the corresponding path - 1, and the value is an integer.

[0197] In actual implementation, the candidate path whose similarity meets the similarity condition is determined as the driving path of the vehicle. Here, the similarity condition is The minimum value of The path corresponding to the minimum value of is determined as the actual driving path of the vehicle.

[0198] The following continues to describe the exemplary structure of the training device 1055 of the path prediction model provided in the embodiment of the present application implemented as a software module. In some embodiments, such as Figure 1 As shown, the software modules stored in the training device 1055 of the path prediction model in the memory 105 may include:

[0199] The acquisition module 10551 is used to obtain sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in the road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point;

[0200] A path determination module 10552 is used to determine at least one sample candidate path between the sample starting point and the sample end point based on the road network, each sample candidate path includes at least two sample nodes, the at least two sample nodes include the sample starting point and the sample end point, and two adjacent sample nodes constitute a sample road section;

[0201] The time determination module 10553 is used to determine the sample section travel time and the sample section travel time standard deviation of the sample vehicle in each sample section based on the sample start and end data by using a Gaussian mixture model;

[0202] A construction module 10554 is used to construct a training sample based on the sample road section travel time and the sample road section travel time standard deviation;

[0203] A training module 10555 is used to train a path prediction model based on the training samples;

[0204] The path prediction model is used to predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's starting and ending data, so as to determine the vehicle's driving path.

[0205] In some embodiments, the time determination module 10553 is also used to construct a travel time probability model for the sample sections in the at least one sample candidate path; and to solve the parameters of the travel time probability model based on the sample travel time in the sample start and end data using a Gaussian mixture model to obtain the sample section travel time and the standard deviation of the sample section travel time for each sample section.

[0206] In some embodiments, the time determination module 10553 is also used to determine the sample shortest path of the at least one sample candidate path; based on the path length of the sample shortest path and the sample travel time, the sample section transit time and the sample section transit time standard deviation of the sample section in the sample shortest path are initialized to obtain the sample initial time and the initial standard deviation; based on the transit time probability model, the sample initial time and the initial standard deviation, the Q function of the Gaussian mixture model is used to perform iterative calculations to obtain the sample section transit time and the sample section transit time standard deviation of each sample section.

[0207] In some embodiments, the time determination module 10553 is also used to determine the average driving speed of the sample vehicle on each sample section on the sample shortest path based on the path length of the sample shortest path and the sample driving time; obtain the section length of the sample section in the sample shortest path; determine the section driving time of the sample vehicle on each sample section based on the section length and the average driving speed; calculate the expectation and standard deviation of the driving time of each section based on the section driving time on each sample section; use the calculated expectation as the sample initial time, and use the calculated standard deviation as the initial standard deviation.

[0208] In some embodiments, the time determination module 10553 is also used to obtain the speed limit interval of the road network; based on the speed limit interval and the section length of each sample section, the time interval of the section travel time is calculated; based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval, the Q function of the Gaussian mixture model is used to perform iterative calculations to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

[0209] In some embodiments, the transit time probability model is:

[0210]

[0211] Where y is the observation time, is the average travel time vector of all sample sections in at least one sample candidate path; is the standard deviation vector of travel time of all sample road segments in at least one sample candidate path; is a distance set of at least one candidate path; R is a set consisting of at least one sample candidate path; λ k The probability of selecting path k from at least one candidate path; is the selection probability of path k in at least one sample candidate path corresponding to the sample start and end data corresponding to the observation time y;

[0212]

[0213] Where C() is the utility function; is the parameter of the utility function; d k is the length of path k; d j is the length of path j;

[0214]

[0215] in, is the expected function of path k; is the time variance of path k;

[0216]

[0217] Among them, μ O is the average driving time of the sample vehicle on the starting sample road section of path k that includes the sample starting point O; μ D μ is the average driving time of the sample vehicle on the terminal sample section of path k that includes the sample terminal D; s is the average driving time of the sample vehicle on the complete sample road segment s in path k;

[0218]

[0219] Among them, σ O is the standard deviation of the driving time of the initial sample road section; σ D is the standard deviation of the travel time of the terminal sample section; σ s is the standard deviation of the travel time for the complete sample road segment.

[0220] In some embodiments, the time determination module 10553 is further used to construct an expected model for the path k to be selected, and the expected model is:

[0221]

[0222] The sample initial time is taken as The initial standard deviation is taken as Substitute the value of into the expected model to obtain the expectation of path k being selected;

[0223] The Q function is calculated based on and The maximum value of , the Q function is:

[0224]

[0225] The maximum value of the Q function is:

[0226]

[0227] in, is the lower limit value in the time interval, is the upper limit value in the time interval;

[0228] Iteratively calculate the maximum value and update it during the iterative calculation process and The value of until the Q function converges;

[0229] The last updated The value is determined as the travel time of the sample road section, and the value of the last update is determined as the standard deviation of the travel time of the sample road section.

[0230] In some embodiments, the construction module 10554 is also used to determine at least one training sample path based on the road network; determine the distribution of the training sample driving time of each training sample section in the training sample path based on the sample section travel time and the standard deviation of the sample section travel time; perform driving simulation on each training sample path based on the distribution of the training sample driving time to obtain the simulated driving arrival time corresponding to each training sample node in the corresponding training sample path; obtain the training sample based on each training sample node corresponding to each training sample path and the corresponding simulated driving arrival time.

[0231] In some embodiments, the training module 10555 is also used to input the first training sample node located at the first position and the second training sample node located at the last position in each of the training sample paths into the path prediction model; predict the intermediate node between the first training sample node and the second training sample node through the path prediction model to obtain a predicted intermediate node; determine the error between the training sample intermediate node and the predicted intermediate node, the training sample intermediate node being the training sample node between the first training sample node and the second training sample node; and update the model parameters of the path prediction model based on the error.

[0232] In some embodiments, the device also includes: a path restoration module, which is used to obtain the start and end data of the vehicle traveling from the starting point to the end point; based on the road network, determine multiple candidate paths between the starting point and the end point; use the path prediction model to predict the intermediate nodes between the starting point and the end point based on the start and end data to obtain a target number of predicted intermediate nodes; based on the starting point, the end point and the target number of predicted intermediate nodes, obtain the predicted path of the vehicle; determine the similarity between the predicted path and multiple candidate paths respectively, and determine the candidate path whose similarity meets the similarity condition as the driving path of the vehicle.

[0233] In some embodiments, the path restoration module is also used to obtain a road node set and a connection attribute data set of a road network, wherein the road node set includes multiple road nodes, and the connection attribute data set includes a connection relationship between each road node; and the relationship between the starting point, the end point, the road node set and the connection attribute data set is used to determine the multiple candidate paths using a shortest path algorithm.

[0234] In some embodiments, the method for determining the target number includes: determining the number of intermediate nodes of each candidate path among the multiple candidate paths; and determining the maximum number of the intermediate nodes as the target number.

[0235] It should be noted that the description of the device of the embodiment of the present application is similar to the description of the above-mentioned method embodiment, and has similar beneficial effects as the method embodiment, so it will not be repeated.

[0236] The embodiment of the present application provides a computer program product, which includes computer instructions, which are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the training method of the path prediction model described in the embodiment of the present application.

[0237] An embodiment of the present application provides a computer-readable storage medium storing executable instructions, wherein executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the training method of the path prediction model provided in the embodiment of the present application.

[0238] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0239] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0240] As an example, executable instructions may, but do not necessarily, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (for example, files storing one or more modules, subroutines, or code portions).

[0241] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0242] In summary, the reliability of path restoration can be improved through the embodiments of the present application.

[0243] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A method for training a path prediction model, characterized in that: include: Obtaining sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point; Based on the road network, at least one sample candidate path between the sample starting point and the sample end point is determined, each sample candidate path includes at least two sample nodes, the at least two sample nodes include the sample starting point and the sample end point, and a sample road section is formed between two adjacent sample nodes; Using a Gaussian mixture model, based on the sample start and end data, determine the sample section travel time and the sample section travel time standard deviation of the sample vehicle in each sample section; Constructing training samples based on the sample road section travel time and the sample road section travel time standard deviation; Based on the training samples, training a path prediction model; The path prediction model is used to predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's starting and ending data, so as to determine the vehicle's driving path.

2. The method according to claim 1, characterized in that The Gaussian mixture model is used to determine the sample travel time of the sample vehicle in each sample road section and the standard deviation of the sample road section travel time based on the sample start and end data, including: Constructing a travel time probability model for a sample road segment in the at least one sample candidate path; The Gaussian mixture model is used to solve the parameters of the travel time probability model based on the sample travel time in the sample start-end data to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

3. The method according to claim 2, characterized in that The Gaussian mixture model is used to solve parameters based on the travel time probability model to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including: Determining a sample shortest path of the at least one sample candidate path; Based on the path length of the sample shortest path and the sample travel time, the sample section travel time and the sample section travel time standard deviation of the sample section in the sample shortest path are initialized to obtain the sample initial time and the initial standard deviation; Based on the travel time probability model, the sample initial time and the initial standard deviation, the Q function of the Gaussian mixture model is used to perform iterative calculations to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

4. The method according to claim 3, characterized in that The sample section travel time and the sample section travel time standard deviation of the sample section in the sample shortest path are initialized based on the path length of the sample shortest path and the sample travel time to obtain the sample initial time and the initial standard deviation, including: Determine the average travel speed of the sample vehicle on each sample road section on the sample shortest path based on the path length of the sample shortest path and the sample travel time; Obtaining the section length of the sample section in the sample shortest path; Determining the road section travel time of the sample vehicle on each sample road section based on the road section length and the average travel speed; Based on the road segment travel time on each sample road segment, calculate the expected and standard deviation of the travel time of each road segment; The calculated expectation is used as the sample initial time, and the calculated standard deviation is used as the initial standard deviation.

5. The method according to claim 4, characterized in that The Q function of the Gaussian mixture model is used to perform iterative calculation based on the travel time probability model, the sample initial time and the initial standard deviation to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including: Get the speed limit interval of the road network; Calculate the time interval of the road section travel time based on the speed limit interval and the section length of each sample road section; Based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval, the Q function of the Gaussian mixture model is used to perform iterative calculations to obtain the sample section travel time and the sample section travel time standard deviation of each sample section.

6. The method according to claim 5, characterized in that The travel time probability model is: Where y is the observation time, is the average travel time vector of all sample sections in at least one sample candidate path; is the standard deviation vector of travel time of all sample road segments in at least one sample candidate path; is a distance set of at least one candidate path; R is a set consisting of at least one sample candidate path; λ k The probability of selecting path k from at least one candidate path; is the selection probability of path k in at least one sample candidate path corresponding to the sample start and end data corresponding to the observation time y; Where C() is the utility function; is the parameter of the utility function; d k is the length of path k; d j is the length of path j; in, is the expected function of path k; is the time variance of path k; Among them, μ O is the average driving time of the sample vehicle on the starting sample road section of path k that includes the sample starting point O; μ D μ is the average driving time of the sample vehicle on the terminal sample section of path k that includes the sample terminal D; s is the average driving time of the sample vehicle on the complete sample road segment s in path k; Among them, σ O is the standard deviation of the driving time of the initial sample road section; σ D is the standard deviation of the travel time of the terminal sample section; σ s is the standard deviation of the travel time for the complete sample road segment.

7. The method according to claim 6, characterized in that The Q function of the Gaussian mixture model is used to perform iterative calculation based on the travel time probability model, the sample initial time, the initial standard deviation and the time interval to obtain the sample section travel time and the sample section travel time standard deviation of each sample section, including: Construct an expected model in which the path k is selected, the expected model is: The sample initial time is taken as The initial standard deviation is taken as Substitute the value of into the expected model to obtain the expectation of path k being selected; The calculation of Q function is based on and The maximum value of , the Q function is: The maximum value of the Q function is: in, is the lower limit value in the time interval, is the upper limit value in the time interval; Iteratively calculate the maximum value and update it during the iterative calculation process and The value of until the Q function converges; The last updated The value is determined as the travel time of the sample road section, and the last updated The value is determined as the standard deviation of the travel time of the sample road section.

8. The method according to claim 1, characterized in that The constructing of training samples based on the sample road section travel time and the sample road section travel time standard deviation includes: Determine at least one training sample path based on the road network; Based on the sample road section travel time and the sample road section travel time standard deviation, determining the distribution of the training sample travel time of each training sample road section in the training sample path; Based on the distribution of the training sample driving duration, a driving simulation is performed on each training sample path to obtain a simulated driving arrival time corresponding to each training sample node in the corresponding training sample path; The training samples are obtained based on the training sample nodes corresponding to the training sample paths and the corresponding simulated travel arrival times.

9. The method according to claim 8, characterized in that The training of the path prediction model based on the training samples includes: Inputting the first training sample node at the first position and the second training sample node at the last position in each training sample path into the path prediction model; Predicting the intermediate nodes between the first training sample node and the second training sample node by using the path prediction model to obtain predicted intermediate nodes; Determine an error between a training sample intermediate node and the prediction intermediate node, the training sample intermediate node being a training sample node between the first training sample node and the second training sample node; Based on the error, model parameters of the path prediction model are updated.

10. The method according to claim 1, characterized in that The method further comprises: Obtain the starting and ending data of the vehicle traveling from the starting point to the end point; Based on the road network, determining a plurality of candidate paths between the starting point and the end point; Using the path prediction model, based on the start and end data, predict the intermediate nodes between the start point and the end point to obtain a target number of predicted intermediate nodes; Obtaining a predicted path of the vehicle based on the starting point, the end point, and the target number of predicted intermediate nodes; The similarities between the predicted path and a plurality of candidate paths are respectively determined, and the candidate paths whose similarities meet a similarity condition are determined as the driving paths of the vehicle.

11. The method according to claim 10, characterized in that The determining the similarity between the predicted path and a plurality of candidate paths includes: Calculating the similarity between the predicted path and multiple candidate paths by using a Levenshtein similarity function to obtain the similarity; Wherein, the Levenshtein similarity function is: Among them, S1 is the predicted path, S2 is the candidate path, is the Levenshtein distance between S1 and S2, and ξ are the labels of the sections in the corresponding paths.

12. The method according to claim 1, characterized in that Determining a plurality of candidate paths between the starting point and the end point in the road network includes: Obtaining a road node set and a connection attribute data set of a road network, wherein the road node set includes a plurality of road nodes, and the connection attribute data set includes connection relationships between the road nodes; The relationship between the starting point, the end point, the road node set and the connection attribute data set is utilized, and the multiple candidate paths are determined using a shortest path algorithm.

13. The method according to claim 1, characterized in that Methods for determining the target quantity include: Determining the number of intermediate nodes of each candidate path among the multiple candidate paths; A maximum number among the numbers of intermediate nodes is determined as the target number.

14. A training device for a path prediction model, characterized in that: include: An acquisition module, used to obtain sample start-end data of a sample vehicle traveling from a sample start point to a sample end point in a road network, wherein the sample start-end data includes sample start point data, sample end point data, and a sample travel time of the sample vehicle traveling from the sample start point to the sample end point; A path determination module, configured to determine, based on the road network, at least one sample candidate path between the sample starting point and the sample end point, wherein each sample candidate path includes at least two sample nodes, wherein the at least two sample nodes include the sample starting point and the sample end point, and a sample road section is formed between two adjacent sample nodes; A time determination module, for determining the sample section travel time and the sample section travel time standard deviation of the sample vehicle in each sample section based on the sample start and end data by using a Gaussian mixture model; A construction module, used to construct a training sample based on the sample road section travel time and the sample road section travel time standard deviation; A training module, used for training a path prediction model based on the training samples; The path prediction model is used to predict the intermediate nodes of the vehicle between the starting point and the end point based on the vehicle's starting and ending data, so as to determine the vehicle's driving path.

15. An electronic device, characterized in that: include: A memory for storing executable instructions; A processor, configured to implement the path prediction model training method described in any one of claims 1 to 13 when executing the executable instructions stored in the memory.

Images