Method, device, electronic device and storage medium for determining trip arrival time

Abstract

This application provides a method, device, electronic device, and storage medium for determining a trip arrival time. These methods relate to the field of artificial intelligence technology and can be applied to various scenarios, including cloud technology, artificial intelligence, smart transportation, and assisted driving. The method includes generating a segment feature sequence for a to-be-estimated trip route based on multidimensional feature information; then, extracting features from the segment feature sequence based on a target feature generation network to obtain target route features for the to-be-estimated trip route; and performing regression prediction on the target route features based on a target time prediction network to obtain an estimated arrival time for the to-be-estimated trip route. The target feature generation network and target time prediction network are obtained by iteratively training an initial generation network, an initial discriminant network, and an initial prediction network through adversarial generation and regression prediction based on historical segment feature sequences, multidimensional route statistics, and actual arrival times for historical trip routes. This application can effectively improve the accuracy of predictions.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, electronic device, and storage medium for determining a trip arrival time. Background Art

[0002] Estimating trip arrival times is a fundamental function of map applications, playing a crucial role in many scenarios, such as route navigation and operational order allocation. Existing methods for estimating arrival times primarily include rule-based, segment-by-segment cumulative methods and tree-based methods. The rule-based, segment-by-segment cumulative method relies on manual experience to formulate estimation rules, which is highly subjective. Furthermore, under complex road conditions, the manually set rules have low scenario coverage and poor time estimation accuracy. The tree-based method, a recently emerging arrival time estimation method, extracts features from the entire route and feeds these features into a tree-based machine learning algorithm for training, such as the Gradient Boosting Decision Tree (GBDT) algorithm. However, this method only considers overall features, resulting in large prediction errors under complex road conditions and during special time periods. Summary of the Invention

[0003] The present application provides a method, device, electronic device and storage medium for determining a trip arrival time, which are applicable to various road conditions and time periods and can significantly improve the accuracy and reliability of determining a trip arrival time.

[0004] In one aspect, the present application provides a method for determining a trip arrival time, the method comprising:

[0005] Obtain multi-dimensional feature information of the road segments in the estimated travel route;

[0006] Generating a road segment feature sequence of the to-be-estimated travel route according to the multi-dimensional feature information;

[0007] Extracting features from the road segment feature sequence based on a target feature generation network to obtain target route features of the route to be estimated;

[0008] Performing regression prediction processing on the target route features based on the target time prediction network to obtain an estimated arrival time of the route to be estimated;

[0009] Among them, the target feature generation network and the target time prediction network are obtained by iteratively training the initial generation network, the initial discriminant network and the initial prediction network through adversarial generation and regression prediction based on the historical section feature sequence of the historical travel route, multi-dimensional route statistical information and the actual arrival time.

[0010] Another aspect provides a device for determining a trip arrival time, the device comprising:

[0011] Multi-dimensional feature acquisition module: used to obtain multi-dimensional feature information of the road sections in the estimated travel route;

[0012] A road segment feature generation module is configured to generate a road segment feature sequence of the to-be-estimated travel route based on the multi-dimensional feature information;

[0013] Route feature extraction module: used for extracting features from the road segment feature sequence based on the target feature generation network to obtain target route features of the route to be estimated;

[0014] Regression prediction module: used to perform regression prediction processing on the target route characteristics based on the target time prediction network to obtain the estimated arrival time of the route to be estimated;

[0015] Among them, the target feature generation network and the target time prediction network are obtained by iteratively training the initial generation network, the initial discriminant network and the initial prediction network through adversarial generation and regression prediction based on the historical section feature sequence of the historical travel route, multi-dimensional route statistical information and the actual arrival time.

[0016] On the other hand, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the above-mentioned method for determining the trip arrival time.

[0017] On the other hand, a computer-readable storage medium is provided, wherein the storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by a processor to implement the above-mentioned method for determining the trip arrival time.

[0018] On the other hand, a server is provided, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the above-mentioned method for determining the trip arrival time.

[0019] On the other hand, a terminal is provided, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the above-mentioned method for determining the trip arrival time.

[0020] On the other hand, a computer program product or a computer program is provided, which includes computer instructions, and when the computer instructions are executed by a processor, the method for determining the trip arrival time as described above is implemented.

[0021] The method, apparatus, device, storage medium, server, terminal, computer program, and computer program product provided in this application have the following technical effects:

[0022] The present application obtains multi-dimensional feature information of the sections passed through in the route to be estimated, and generates a section feature sequence of the route to be estimated based on the multi-dimensional feature information; then, based on the target feature generation network, feature extraction is performed on the section feature sequence to obtain the target route features of the route to be estimated; then, based on the target time prediction network, regression prediction processing is performed on the target route features to obtain the estimated arrival time of the route to be estimated; based on the multi-dimensional feature information of each section passed through in the route, a section feature sequence of the entire route is generated, and the target feature generation network and the target time prediction network are combined to perform time prediction on the section feature sequence, thereby significantly improving the prediction accuracy. In addition, the target feature generation network and the target time prediction network are obtained by iteratively training the initial generation network, the initial discriminant network and the initial prediction network through adversarial generation and regression prediction based on the historical segment feature sequence of the historical travel route, multi-dimensional route statistical information and the actual arrival time. In this way, the adversarial training method is used to assist in optimizing the feature extraction process, and the actual arrival time of the historical travel and the statistical information generated by the target object passing through each route segment are combined as the supervision information for training, thereby improving the model effect and accuracy of the network model, thereby improving the prediction accuracy and reliability of the estimated arrival time. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1 is a schematic diagram of an application environment provided by an embodiment of the present application;

[0025] Figure 2 This is a flow chart of a method for determining a trip arrival time provided in an embodiment of the present application;

[0026] Figure 3 1 is a flow chart of another method for determining a trip arrival time provided in an embodiment of the present application;

[0027] Figure 4 This is a schematic diagram of another principle of network training provided by an embodiment of the present application;

[0028] Figure 5 1 is a flow chart of another method for determining a trip arrival time provided in an embodiment of the present application;

[0029] Figure 6 1 is a flow chart of another method for determining a trip arrival time provided in an embodiment of the present application;

[0030] Figure 7 1 is a flow chart of another method for determining a trip arrival time provided in an embodiment of the present application;

[0031] Figure 8 This is a schematic diagram of a framework of a device for determining a trip arrival time provided by an embodiment of the present application;

[0032] Figure 9 This is a hardware structure block diagram of an electronic device for a method for determining a trip arrival time provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server comprising a series of steps or submodules is not necessarily limited to those steps or submodules clearly listed, but may include other steps or submodules that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0036] Estimated Time of Arrival (ETA): This is a time estimation function in map software. Specifically, given a route and departure time on a map, it estimates the time required to complete the route.

[0037] Actual Time of Arrival (ATA): The actual time taken for a route in the historical data of the map service.

[0038] Route: In map applications, a route is a complete line connecting the starting point and the end point. In actual scenarios, the length of a route is usually in the range of one kilometer to several tens of kilometers.

[0039] Links: In map applications, routes are represented as sequences of links. In map data, roads are divided into multiple connected segments, ranging in length from tens of meters to several kilometers. Each segment is called a link and is assigned a globally unique ID. Therefore, a route on a map is a sequence of all the links in that route.

[0040] Neural Network: A neural network is a computational model with layers of connected nodes, similar to the network of neurons in the brain. Neural networks learn from data, training them to recognize patterns, classify data, and predict future events.

[0041] Generative Adversarial Networks (GANs): A new type of neural network designed to automatically generate highly realistic images. They consist of two components: a generator and a discriminator. The generator's goal is to produce an image that is indistinguishable from a real image, while the discriminator's goal is to determine whether an image is genuine by comparing it with real images. Through adversarial training, the images generated by the generator network gradually approach real images, achieving a level of fidelity.

[0042] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0043] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0044] Computer vision (CV) is the science of making machines "see." Specifically, it refers to machine vision, where cameras and computers replace the human eye in identifying and measuring objects, performing further image processing to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include image processing, image recognition, image semantic understanding, image retrieval, optical character recognition (OCR), video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0045] Natural language processing (NLP) is a key area of ​​research in computer science and artificial intelligence. It studies the theories and methods that enable effective communication between humans and computers using natural language. Natural language processing (NLP) integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language we use in everyday life—and is closely linked to the study of linguistics. Natural language processing technologies typically include text processing, semantic understanding, machine translation, robotic question answering, and knowledge graphs.

[0046] In recent years, with the research and progress of artificial intelligence technology, artificial intelligence technology has been widely used in many fields. The solutions provided in the embodiments of this application involve artificial intelligence machine learning / deep learning and natural language processing technologies, which are specifically illustrated by the following embodiments.

[0047] See also Figure 1 , Figure 1 This is a schematic diagram of an application environment provided by an embodiment of the present application. Figure 1 As shown, the application environment may include at least a terminal 01 and a server 02. In actual applications, the terminal 01 and the server 02 may be connected directly or indirectly via wired or wireless communication, which is not limited in this application.

[0048] The server 02 in the embodiment of the present application can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be 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, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0049] Specifically, cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to enable data computing, storage, processing, and sharing. Cloud technology can be applied in a variety of fields, such as healthcare cloud, cloud IoT, cloud security, cloud education, cloud conferencing, artificial intelligence cloud services, cloud applications, cloud calling, and cloud social networking. Based on the cloud computing business model, cloud technology distributes computing tasks across a resource pool consisting of a large number of computers, enabling various application systems to access computing power, storage space, and information services as needed. The network that provides resources is called the "cloud." To users, the resources in the cloud appear infinitely scalable and can be accessed at any time, used on demand, and expanded at any time, with a pay-per-use policy. Providers of cloud computing infrastructure establish a cloud computing resource pool (referred to as a cloud platform, commonly referred to as IaaS (Infrastructure as a Service)) and deploy various types of virtual resources within the resource pool for external clients to choose from. The cloud computing resource pool primarily includes computing devices (virtualized machines, including operating systems), storage devices, and network devices.

[0050] Based on logical functional divisions, the PaaS (Platform as a Service) layer can be deployed on top of the IaaS layer, and the SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. SaaS can also be deployed directly on top of IaaS. PaaS is a platform for software execution, such as databases and web containers. SaaS is a variety of business software, such as web portals and text messaging apps. Generally speaking, SaaS and PaaS are layers above IaaS.

[0051] Specifically, the server 02 mentioned above may include a physical device, which may specifically include a network communication submodule, a processor, a memory, etc., and may also include software running in the physical device, which may specifically include an application program, etc.

[0052] Specifically, terminal 01 may include physical devices such as smart phones, desktop computers, tablet computers, laptops, digital assistants, augmented reality (AR) / virtual reality (VR) devices, intelligent voice interaction devices, smart home appliances, smart wearable devices, and vehicle-mounted terminal devices, and may also include software running in physical devices, such as applications.

[0053] In an embodiment of the present application, terminal 01 can be used to send a trip arrival time estimation request to server 02, so that server 02 performs a corresponding trip arrival time determination operation. Server 02 can be used to provide a trip arrival time determination service, obtain multidimensional feature information of each route segment in the estimated route, generate a route segment feature sequence, and perform feature extraction and further regression prediction on the route segment feature sequence to obtain an estimated arrival time. It can be understood that terminal 01 can also be used to provide a trip arrival time determination service to obtain an estimated arrival time. Specifically, server 02 can also be used to provide network training services for a target feature generation network and a target time prediction network, and can also be used to store training data sets and multidimensional feature information of routes, etc.

[0054] Furthermore, it is understandable that Figure 1 What is shown is merely an application environment of a method for determining a trip arrival time. The application environment may include more or fewer nodes, and this application does not impose any limitation thereto.

[0055] The application environment involved in the embodiments of this application, or the terminal 01 and server 02 in the application environment, can be a distributed system formed by a client and multiple nodes (any type of computing device connected to the network, such as a server or user terminal) connected via network communication. The distributed system can be a blockchain system that can provide the above-mentioned trip arrival time determination service and data storage service, etc.

[0056] The following describes a method for determining the arrival time of a trip in this application based on the above application environment. The embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc. Please refer to Figure 2 , Figure 2It is a flowchart of a method for determining a trip arrival time provided in an embodiment of the present application. This specification provides method operation steps such as the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative work. The order of steps listed in the embodiment is only one way of executing the order of many steps and does not represent the only execution order. When the actual system or server product is executed, it can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment) according to the method shown in the embodiment or the accompanying drawings. Specifically, Figure 2 As shown, the method may include the following steps S201-S207.

[0057] S201: Obtain multi-dimensional feature information of the road segments in the estimated travel route.

[0058] In an embodiment of the present application, the estimated travel route can be obtained by a route calculation module in response to a travel route request, based on travel information. The travel information includes, but is not limited to, the travel starting point, the travel end point, the travel request time, and the travel departure time. The travel request time is the time when the travel route request is initiated, and the travel departure time is the time when the target object departs from the travel starting point. For example, in a navigation scenario, after initiating a navigation request, the target object departs immediately within a short period of time. In this case, the travel request time is substantially equal to the travel departure time. If the navigation request is for a future time, for example, a request for a future time of 10:00 AM on Monday morning at 8:00 AM on Monday morning, 8:00 AM is the request time, and 10:00 AM is the departure time. The target object can be a motor vehicle, a non-motor vehicle, or a pedestrian.

[0059] Specifically, the route calculation module can also divide the route into sections in response to the arrival time estimation request, and then provide the section identifiers for each section in the route, typically a sequence of section identifiers. Accordingly, a section identifier sequence and a target travel time are obtained for the estimated route. The section identifier sequence includes the section identifiers of each section in the estimated route, and each section identifier is a globally unique identifier. That is, the section identifier sequence is a sequence formed by arranging the section identifiers of each section in the estimated route based on bit order. The target travel time is the aforementioned travel request time or trip departure time.

[0060] Furthermore, based on the road segment identification sequence and the target travel time, multi-dimensional feature information for each routed road segment is obtained from the road network information and driving data. The road network information includes, but is not limited to, basic attribute feature information and road condition information for the routed road segments. Basic attribute feature information may include, but is not limited to, link mileage, road grade, width, and presence of traffic lights. Width refers to road width or driving width, and presence of traffic lights is represented by an identification code, e.g., a traffic light is 1 and no traffic light is 0. Road condition information is associated with time and includes, but is not limited to, the road condition, real-time speed, and real-time traffic volume of the routed road segments at each time. Road condition status may include, for example, the degree of congestion or weather conditions, and real-time speed may represent the real-time travel speed at that moment. Driving data is associated with time and includes speed feature information for the routed road segments. Speed ​​feature information may include, but is not limited to, historical classic speeds and the standard deviation of historical classic speeds.

[0061] Specifically, the aforementioned speed feature information can be derived based on historical trajectory information for historical travel routes. This historical trajectory information includes, but is not limited to, historical departure times, historical segment identifier sequences, travel speed information for each segment at each historical time, the target's historical arrival time at each segment, and the actual arrival time for the historical travel route. Accordingly, the historical classic speed is the speed mined for each link based on the historical trajectory information of multiple target objects, such as the historical trajectory information of all target objects in the backend. For example, using a weekly cycle and a 5-minute granularity, 7*24*12=2016 speed values ​​are mined for each link, representing the travel speed every 5 minutes from Monday to Sunday. This mining can, for example, involve calculating speed statistics for every 5 minutes. The transit time is the time when the target actually arrives at a particular segment. For example, if a trip departs at 10:00 and arrives at the 100th segment at 10:30, the transit time for the 100th segment is 10:30.

[0062] In some embodiments, the multidimensional feature information for each route segment may include the aforementioned various basic attribute feature information, road condition information within a first preset time period corresponding to the target travel time, and speed feature information within a second preset time period corresponding to the target travel time. In one example, the first preset time period is from -30 minutes to the travel departure time, and the second preset time period is ±30 minutes from the travel departure time. Accordingly, the multidimensional feature information includes basic attribute features such as link mileage, road grade, road width, and the presence of traffic lights; road condition information such as the road condition status, real-time speed, and real-time traffic volume at six moments in the route segment within the time period from t0-30 minutes to t0 departure time, with a granularity of 5 minutes; and driving data such as the historical classic speed and the standard deviation of the historical classic speed at 12 historical travel moments in the route segment within the time period from t0-30 minutes to t0+30 minutes from the travel departure time, with a granularity of 5 minutes. The historical transit time here can match the target travel time. It can be the time corresponding to the target travel time on the previous day, or the time corresponding to the target travel time last week. For example, if the target travel time is 9:00 AM on Tuesday, the historical transit time can be 9:00 AM on the previous day, or 9:00 AM on Tuesday last week. It should be noted that in the ETA prediction scenario, the transit time of each route segment in the future is unknown. Therefore, the target travel time is used to extract road condition information and speed characteristic information, rather than the transit time. The road condition information for the first preset time period corresponding to the target travel time is determined based on the real-time data of multiple other target objects in the first preset time period, or is published in real time by the road network platform during the time period.

[0063] It can be understood that the above-mentioned road network information and driving data may include more types of information related to the road network, driving and arrival time estimation. Correspondingly, the multi-dimensional feature information may also include more dimensions of information related to arrival time estimation. The first preset time period and the second preset time period can be set based on actual needs, such as based on computing power and timeliness requirements, which are not enumerated here in this application.

[0064] S203: Generate a road segment feature sequence of the to-be-estimated travel route based on the multi-dimensional feature information.

[0065] In an embodiment of the present application, feature encoding processing is performed on the multi-dimensional feature information of each of the passed road sections to obtain the section vector of each passed road section, and feature splicing processing is performed on each section vector based on the section identification sequence to obtain a section feature sequence.

[0066] Specifically, for each route segment, a K-dimensional route segment vector can be derived based on the multidimensional feature information. K represents the dimensionality of the multidimensional feature information, that is, the amount of information included in the multidimensional feature information. In the above example, K can be 46 dimensions. The K-dimensional route segment vectors for each route segment in the estimated route are then concatenated in the other direction to form a route segment feature sequence of length M, where M is the total number of routes in the estimated route. This means that the route segment feature sequence is an M*K feature matrix.

[0067] S205: Extracting features from the road segment feature sequence based on the target feature generation network to obtain target route features of the route to be estimated.

[0068] In an embodiment of the present application, the road section feature sequence is input into a target feature generation network to perform feature extraction to obtain target route features. The target feature generation network can be constructed based on a convolutional neural network or a recurrent neural network.

[0069] S207: Performing regression prediction processing on the target route features based on the target time prediction network to obtain an estimated arrival time of the route to be estimated.

[0070] In this embodiment of the present application, the target route features output by the target feature generation network are used as input to the target time prediction network for regression prediction to obtain an estimated arrival time. The target time prediction network can be constructed based on a fully connected neural network, a factorization machine network, or an encoder (transformer). In this way, a segment feature sequence for the entire route is generated based on the multidimensional feature information of each segment along the route. The target feature generation network and the target time prediction network are combined to perform time prediction on this segment feature sequence, significantly improving the prediction accuracy.

[0071] In actual applications, after obtaining the estimated arrival time, the historical travel information of the request object of the arrival time estimation request can be obtained, where the request object can be a terminal identifier or an account identifier, etc.; based on the information such as the starting point, end point and travel time of the route to be estimated, at least one reference historical trip that matches the route to be estimated is screened out, and the reference historical trip is a historical trip that has the same starting point and end point as the route to be estimated, and the time difference between the travel times is within a preset range; the actual arrival time of at least one reference historical trip is obtained, and the estimated arrival time is corrected according to the actual arrival time of at least one reference historical trip to obtain a corrected estimated arrival time.

[0072] In actual applications, the estimated arrival time can be used as basic information in a variety of scenarios for use by upstream services to evaluate route quality, including but not limited to: in the navigation scenario, after the starting point and end point of the trip are determined, the background provides several candidate routes, and then uses the above-mentioned estimated time determination method to obtain the estimated arrival time of each candidate route, and then selects the optimal route as the target route; during the navigation process, at regular intervals, the above-mentioned method is used to calculate the time remaining in the route based on the real-time starting point, so that the remaining time can be included in the itinerary arrangement; in the food delivery or commercial vehicle scenario, the estimated arrival time of each route can be calculated to optimize the order quality, reduce empty driving or waiting time, and improve delivery or passenger transport efficiency.

[0073] Furthermore, in an area search scenario, an isochronous area corresponding to a location can be determined based on the estimated arrival time. Accordingly, prior to step S201 , the method further includes: obtaining a target location and multiple candidate locations surrounding the target location; and determining estimated travel routes between the target location and the multiple candidate locations.

[0074] Specifically, the candidate location is within a preset area around the target location. The preset area can be an area with a preset radius centered on the target location, such as a preset radius of 20 km. The preset area can be set as needed based on a preset duration. The candidate location can be a pre-marked point of interest (POI) or a key road section. With the target location as the starting point of the trip and the candidate location as the end point of the trip, at least one estimated route between the target location and each candidate location can be determined.

[0075] Furthermore, after S207, the method may also include: filtering out a target candidate location that meets a preset time condition from multiple candidate locations based on the estimated arrival time; determining a target area corresponding to the target location based on the target candidate location; and displaying an area mark corresponding to the target area on a preset map interface.

[0076] Specifically, satisfying the preset time condition may be that the estimated arrival time is less than or equal to a preset time threshold, or that the estimated arrival time is within a preset time range; exemplarily, the preset time threshold may be 120min, 60min or 30min, etc., or the preset time range may be 120±10min, 60±5min, or 30±5min, etc.

[0077] Specifically, the target area can be an isochronous reachable area with the target location as the starting point of the journey, such as a half-hour reachable circle or a one-hour reachable circle, and the target candidate location is located on the regional boundary of the target area. Based on the multiple target candidate locations with the same or similar estimated arrival times, an isochronous reachable area is delineated, and the target area is marked on the preset map interface through area marking. The area marking may include displaying the target area in a preset color, and may also include a location tag of the target candidate location. In some cases, it may also include location tags of other candidate locations within the target area. A query control may be set at the location tag of the preset map interface, and in response to the triggering operation of the query control, the travel reference information of the corresponding candidate location is displayed. The travel reference information includes but is not limited to the location identifier and the estimated arrival time. In this way, the isochronous reachable circle of the search location is determined by the estimated arrival time, so as to display the radiation area of ​​the search location and optimize the user experience.

[0078] In this embodiment, the target feature generation network and target time prediction network are derived by iteratively training the initial generation network, initial discriminant network, and initial prediction network through adversarial generation and regression prediction based on the historical route segment feature sequence, multi-dimensional route statistics, and actual arrival times. This adversarial training method assists in optimizing the feature extraction process. Furthermore, the actual arrival times of historical itineraries and the statistical information generated by the target object's journey through each route segment serve as supervisory information for training, improving the network model's performance and accuracy, and thereby enhancing the accuracy and reliability of estimated arrival times.

[0079] In actual applications, similar to the aforementioned road segment feature sequence, the historical road segment feature sequence is obtained based on the multi-dimensional historical feature information of each road segment in the historical travel route, including: obtaining the historical road segment identification sequence and historical travel time corresponding to the historical travel route, and based on the historical road segment identification sequence and historical travel time, obtaining the multi-dimensional historical feature information of each road segment from the road network information and driving data.

[0080] Specifically, the historical segment identification sequence and historical travel time are similar to the aforementioned route segments, segment identification sequence, and travel time, respectively, and will not be further described here. Historical travel time includes historical request times and historical departure times. The multi-dimensional historical feature information for each historical route segment can include the various basic attribute feature information described above, historical road condition information within a first preset time period corresponding to the historical travel time, and speed feature information within a second preset time period corresponding to the historical travel time.

[0081] In actual application scenarios, the traffic information corresponding to the current estimated route is the real-time traffic information obtained by the backend at the target travel time, and the real-time traffic information at the target object's future passing time cannot be known; the traffic information corresponding to the historical route is the historical real-time traffic information at the historical passing time obtained based on the historical trajectory information. Accordingly, in an example, the multi-dimensional historical feature information includes basic attribute features such as link mileage, road grade, width, and whether there are traffic lights, and the historical departure time t x -30min to t x During the time period, with a granularity of 5 minutes, the historical road condition information of the 6 moments of the route, including historical real-time speed and historical real-time traffic flow, and the historical departure time t x -30min to t x Within the +30 minute period, with a 5-minute granularity, the historical typical speed and standard deviation of the historical typical speed for the 12 historical travel times of the route segment are collected. The historical travel time here refers to the actual time when the target object corresponding to the historical travel route passed through the route segment.

[0082] Furthermore, feature encoding is performed on the multidimensional feature information of each route segment in the historical itinerary to obtain a historical route vector for each route segment. Based on the historical route segment identifier sequence, feature concatenation is performed on each historical route segment vector to obtain a historical route segment feature sequence. The historical route segment feature sequence and the actual arrival time of the historical route constitute a first training sample binary.

[0083] Please refer to Figure 3 and Figure 4 ,The target feature generation network and target time prediction network are obtained in the following way.

[0084] S301: Generate route statistical features of historical travel routes based on multi-dimensional route statistical information.

[0085] In practice, multi-dimensional route statistics include both overall and segment-by-segment statistics for historical routes. Overall statistics may include, but are not limited to, total mileage, percentage of expressways, number of traffic lights, and total congestion time. Segment statistics for each segment include, but are not limited to, the average historical real-time speed and average historical speed for that segment.

[0086] In one case, the average historical real-time speed is the overall average speed of a single route segment, which can be obtained by averaging the historical real-time speeds of the route segment at each historical moment within the first preset time period corresponding to the historical passage time, or by sampling the route segment at intervals within the first preset time period corresponding to the historical passage time to obtain multiple historical real-time speeds, and then averaging the multiple historical real-time speeds. In other cases, the average historical real-time speed includes the historical average speed of multiple moments in a single route segment, such as the historical passage time t link -30min to t link Within the time period, the historical real-time speed is averaged every 5 minutes at a granularity of 5 minutes to obtain the historical average speed of the 6 historical moments of the route section. The above averaging can be arithmetic or harmonic average.

[0087] Similarly, in one case, the average historical classic speed is the overall historical classic speed of a single route segment, which can be obtained by averaging the classic speeds of each historical moment in the second preset time period corresponding to the historical passing moment of the route segment, or the average historical classic speed can be obtained by sampling the route segment at intervals in the second preset time period corresponding to the historical passing moment, obtaining multiple historical classic speeds, and then averaging the multiple historical classic speeds; in other cases, the average historical classic speed includes the average classic speed of multiple moments in a single route segment, such as the average classic speed of the historical passing moment t link -30min to t link During the +30 minute period, the historical classic speeds are averaged every 5 minutes at a granularity of 5 minutes to obtain the average classic speed for the 12 historical moments of the route. This averaging can be either arithmetic or harmonic.

[0088] It should be noted that the above-mentioned road section statistics are all based on the historical passing time t of the target object passing through the road section. link , the corresponding data is generated, rather than the historical departure time t x Corresponding data.

[0089] It is understandable that the above multi-dimensional route statistical information may include more information related to arrival time estimation, not limited to extraction from link-level features, such as the entire congestion time and the entire high-speed time, etc., which are not enumerated here in this application.

[0090] In some embodiments, feature encoding processing may be performed on the multi-dimensional route statistical information to obtain a route statistical feature, where the route statistical feature is an L-dimensional vector, where L is the information dimension in the multi-dimensional route statistical information.

[0091] In other embodiments, generating route statistical features may include: performing feature encoding processing on multi-dimensional route statistical information to obtain initial statistical features of historical travel routes. Feature mapping processing is then performed on the initial statistical features to obtain route statistical features. Specifically, the initial statistical features are L-dimensional vectors. After feature mapping processing, the dimensionality of the initial statistical features is transformed to N dimensions. In some cases, N can be less than or equal to L, or greater than or equal to L; in other cases, N can be the same as the dimensionality of the historical route features, i.e., the feature space of the route statistical features and the historical route features is the same. In this way, the dimensionality of the route statistical features and the historical route features is unified, thereby optimizing network performance.

[0092] Specifically, feature mapping processing can be performed on the initial statistical features based on a feature mapping network, which can be a fully connected neural network. By performing feature mapping on the initial statistical features, even if the types and dimensions of the multi-dimensional route statistical information are increased, the dimensionality of the route statistical features will not change. This ensures that the input to the subsequent initial discriminant network is fixed-length features, facilitating data expansion and network update training, and improving system scalability. In some embodiments, route statistical features can be extracted using a combination of historical data and manual experience, or provided by a road network platform and a driving data platform. Route statistical features can be truth vectors.

[0093] S303: Extracting features from the historical road segment feature sequence based on the initial generation network to obtain historical route features.

[0094] In practical applications, the initial generation network extracts variable-length features into fixed-length features. It can be constructed based on convolutional neural networks, recurrent neural networks, or encoders, such as BERT. The historical road section feature sequence is input into the initial generation network for feature extraction to obtain historical route features.

[0095] S305: Using historical route features, route statistical features, and actual arrival time as inputs to the initial discriminant network, a first discriminant probability and a second discriminant probability are obtained. The first discriminant probability represents the degree of match between the actual arrival time and the historical route features, and the second discriminant probability represents the degree of match between the actual arrival time and the route statistical features.

[0096] In practical applications, historical route features and actual arrival time form the first sample pair, and route statistical features and actual arrival time form the second sample pair. The first sample pair and the second sample pair are used as the input of the initial discriminant network to obtain the first discriminant probability of the first sample pair and the second discriminant probability of the second sample pair, and then conduct adversarial generative training.

[0097] S307: Perform regression prediction processing on historical route features based on the initial prediction network to obtain a reference arrival time.

[0098] In practical applications, historical route features are input into the initial prediction network for regression prediction. The output result is the reference arrival time of the historical travel route, and the reference arrival time is the estimated arrival time predicted by the initial prediction network.

[0099] S309: Calculate the loss based on the first discrimination probability, the second discrimination probability, the reference arrival time, and the actual arrival time to obtain the generated loss, the discriminant loss, and the predicted loss.

[0100] In actual application, please refer to Figure 5 , S309 may include the following steps S3091-S3093.

[0101] S3091: Calculate the loss based on the first discriminant probability and the second discriminant probability to obtain the discriminant loss.

[0102] S3092: Calculate the loss based on the deviation between the reference arrival time and the actual arrival time to obtain the predicted loss.

[0103] S3093: Calculate the loss based on the first discriminant probability and the predicted loss to obtain the generated loss.

[0104] Specifically, the discrimination loss represents the loss of the initial discrimination network, the prediction loss represents the loss of the initial prediction network, and the generation loss represents the loss of the initial generation network.

[0105] In one embodiment, the loss functions for training the initial generation network G and the initial prediction network C are respectively represented by the following formulas. Wherein, L(D) is the loss function of the initial discriminant network D, L(G) is the loss function of the initial generation network G, L(C) is the loss function of the initial prediction network C, F represents the number of historical travel routes in the training sample set, the training sample set includes historical segment feature sequences, actual arrival times, and segment statistical features corresponding to multiple historical travel routes, real represents route statistical features, ATA represents the actual arrival time, x represents the historical route feature sequence, D(G(x), ATA) represents the first discriminant probability, D(real, ATA) represents the second discriminant probability, and (G(x)) represents the reference arrival time.

[0106]

[0107]

[0108]

[0109] Specifically, loss(C(G(x)),ATA) represents the reference arrival time ETA rThe deviation from ATA can be calculated based on square loss or absolute value loss, as shown in the following formula.

[0110] Loss=-(ETA r -ATA) 2

[0111] Loss=-|ETA r -ATA|

[0112] It should be noted that ETA r The deviation between the ATA and the ATA is not limited to the above calculation method, and other methods can be used to calculate the loss, which are not enumerated in this application.

[0113] S311: Based on the generation loss, discrimination loss and prediction loss, the initial generation network, the initial discrimination network and the initial prediction network are iteratively trained until the preset convergence conditions are met to obtain the target feature generation network and the target time prediction network.

[0114] For some examples, please refer to Figure 6 , S311 may include the following steps S3111-S3113.

[0115] S3111: Fix the model parameters of the initial generation network and the initial prediction network, and perform a first training on the initial discriminant network based on the discriminant loss to update the model parameters of the initial discriminant network.

[0116] S3112: Fix the updated model parameters of the discriminant network, and perform a second training on the initial generation network and the initial prediction network based on the generation loss and the prediction loss to update the model parameters of the initial generation network and the initial prediction network.

[0117] S3113: Alternately perform the above-mentioned first training and second training steps until the preset convergence condition is met, and obtain the target feature generation network and the target time prediction network.

[0118] In practical applications, the training process may specifically include: initializing the D, G, and C networks, fixing the G and C networks, and training the D network using the above formula L(D), that is, updating the model parameters of the D network until a first preset number of iterations is reached or the discrimination loss of the D network meets the preset loss condition; then, fixing the D network, and training the G and C networks using formulas L(G) and L(C) until a second preset number of iterations is reached, or the generation loss and prediction loss of the G and C networks meet the preset loss condition, and alternately performing the above steps of training the D network (first training) and training the G and C networks (second training) until the number of alternating executions reaches a preset number of alternating executions, or the prediction loss, generation loss, and discrimination loss all meet the corresponding preset convergence conditions.

[0119] For other embodiments, please refer to Figure 7 , S311 may include the following steps S3114-S3115.

[0120] S3114: Determine the total loss based on the generation loss, the discrimination loss, and the prediction loss.

[0121] S3115: Based on the total loss, the initial generation network, the initial discriminant network and the initial prediction network are jointly trained for adversarial generation and regression prediction to obtain the target feature generation network and the target time prediction network.

[0122] Specifically, the generation loss, discrimination loss and prediction loss can be weighted and summed to obtain the total loss, and then the model parameters of the initial generation network, initial discrimination network and initial prediction network can be adjusted based on the total loss to obtain updated initial generation network, initial discrimination network and initial prediction network. The above process is repeated until the preset convergence conditions are met, such as the total loss meets the preset total loss condition or the number of iterations reaches the preset number.

[0123] In summary, the technical solution of the present application makes full use of the multi-dimensional statistical information of historical travel routes and the variable-length multi-dimensional statistical information of historical passing times in each route segment as supervision information for training, and uses adversarial training to assist in optimizing the feature extraction process, thereby improving the accuracy of feature extraction, and thus significantly improving the model effect and prediction accuracy, avoiding the training difficulty and accuracy improvement bottleneck caused by using only actual arrival time as supervision information, and combining human experience and artificial intelligence algorithms to enhance the interpretability and accuracy of the model.

[0124] The embodiment of the present application also provides a device 600 for determining a trip arrival time, such as Figure 8 , which shows a structural schematic diagram of a device for determining a trip arrival time provided in an embodiment of the present application. The device may include the following modules.

[0125] Multi-dimensional feature acquisition module 10: used to obtain multi-dimensional feature information of the road segments in the estimated travel route;

[0126] The road segment feature generation module 20 is used to generate a road segment feature sequence of the to-be-estimated travel route based on the multi-dimensional feature information;

[0127] Route feature extraction module 30: used to extract features from the road segment feature sequence based on the target feature generation network to obtain target route features of the route to be estimated;

[0128] Regression prediction module 40: used to perform regression prediction processing on the target route characteristics based on the target time prediction network to obtain the estimated arrival time of the route to be estimated;

[0129] Among them, the target feature generation network and the target time prediction network are obtained by iteratively training the initial generation network, the initial discriminant network and the initial prediction network through adversarial generation and regression prediction based on the historical section feature sequence of the historical travel route, multi-dimensional route statistical information and the actual arrival time.

[0130] In some embodiments, the apparatus may further include:

[0131] Statistical feature generation module: used to generate route statistical features of historical travel routes based on multi-dimensional route statistical information;

[0132] Historical feature extraction module: used to extract features from historical road segment feature sequences based on the initial generation network to obtain historical route features;

[0133] Discriminant network module: This module uses historical route features, route statistical features, and actual arrival time as inputs to the initial discriminant network to obtain a first discriminant probability and a second discriminant probability. The first discriminant probability represents the degree of match between the actual arrival time and the historical route features, while the second discriminant probability represents the degree of match between the actual arrival time and the route statistical features.

[0134] Prediction network module: used to perform regression prediction processing on historical route characteristics based on the initial prediction network to obtain the reference arrival time;

[0135] Loss calculation module: used to calculate the loss based on the first discriminant probability, the second discriminant probability, the reference arrival time and the actual arrival time to obtain the generated loss, the discriminant loss and the predicted loss;

[0136] Iterative training module: used to iteratively train the initial generation network, initial discriminant network and initial prediction network based on generation loss, discrimination loss and prediction loss until the preset convergence conditions are met to obtain the target feature generation network and target time prediction network.

[0137] In some embodiments, the iterative training module may include:

[0138] The first training submodule is used to fix the model parameters of the initial generation network and the initial prediction network, and perform the first training on the initial discriminant network based on the discriminant loss to update the model parameters of the initial discriminant network;

[0139] The second training submodule is used to fix the updated model parameters of the discriminant network, and perform a second training on the initial generation network and the initial prediction network based on the generation loss and the prediction loss to update the model parameters of the initial generation network and the initial prediction network;

[0140] Alternating execution submodule: used to alternately execute the above-mentioned first training and second training steps until the preset convergence conditions are met to obtain the target feature generation network and the target time prediction network.

[0141] In some other embodiments, the iterative training module may include:

[0142] Total loss determination submodule: used to determine the total loss based on generation loss, discrimination loss and prediction loss;

[0143] Joint training submodule: used to perform joint training of adversarial generation and regression prediction on the initial generation network, initial discriminant network and initial prediction network based on the total loss to obtain the target feature generation network and target time prediction network.

[0144] In some embodiments, the loss calculation module may include:

[0145] Discrimination loss calculation submodule: used to calculate the loss based on the first discrimination probability and the second discrimination probability to obtain the discrimination loss;

[0146] Prediction loss calculation submodule: used to calculate the loss based on the deviation between the reference arrival time and the actual arrival time to obtain the predicted loss;

[0147] Generation loss calculation submodule: used to calculate the loss based on the first discriminant probability and the prediction loss to obtain the generation loss.

[0148] In some embodiments, the multi-dimensional feature acquisition module 10 may include:

[0149] The first acquisition submodule is used to obtain a road segment identification sequence and a target travel time corresponding to the estimated travel route, wherein the road segment identification sequence includes the road segment identifications of each road segment in the estimated travel route;

[0150] The second acquisition submodule is used to obtain the multi-dimensional feature information of each route segment from the road network information and driving data based on the route segment identification sequence and departure time.

[0151] In some embodiments, the road segment feature generation module 20 may include:

[0152] Feature coding submodule: used to perform feature coding processing on the multi-dimensional feature information of each of the passed road sections to obtain the road section vector of each passed road section;

[0153] Feature splicing submodule: used to perform feature splicing processing on each road section vector based on the road section identification sequence to obtain a road section feature sequence.

[0154] In some embodiments, the statistical feature generation module may include:

[0155] Historical feature encoding submodule: used to perform feature encoding processing on multi-dimensional route statistical information to obtain the initial statistical features of the historical travel route;

[0156] Feature mapping submodule: used to perform feature mapping processing on the initial statistical features to obtain route statistical features.

[0157] It should be noted that the above device embodiments and method embodiments are based on the same implementation method.

[0158] An embodiment of the present application provides a device for determining a trip arrival time. The scheduling device can be a terminal or a server, including a processor and a memory. The memory stores at least one instruction or at least one program. The at least one instruction or the at least one program is loaded and executed by the processor to implement the trip arrival time determination method provided in the above-mentioned method embodiment.

[0159] The memory can be used to store software programs and modules. The processor executes the software programs and modules stored in the memory to perform various functional applications and determine the trip arrival time. The memory may primarily include a program storage area and a data storage area. The program storage area can store the operating system, applications required for functions, etc.; the data storage area can store data generated based on device usage. Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide processor access to the memory.

[0160] The method embodiments provided in the embodiments of the present application can be executed in electronic devices such as mobile terminals, computer terminals, servers or similar computing devices. Figure 9 This is a hardware structure block diagram of an electronic device for determining a trip arrival time provided by an embodiment of the present application. Figure 9As shown, the electronic device 900 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 910 (the processor 910 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 930 for storing data, and one or more storage media 920 (such as one or more mass storage devices) for storing application programs 923 or data 922. Among them, the memory 930 and the storage medium 920 can be temporary storage or permanent storage. The program stored in the storage medium 920 may include one or more modules, each module may include a series of instruction operations in the electronic device. Furthermore, the central processing unit 910 can be configured to communicate with the storage medium 920 and execute a series of instruction operations in the storage medium 920 on the electronic device 900. The electronic device 900 may also include one or more power supplies 960, one or more wired or wireless network interfaces 950, one or more input and output interfaces 940, and / or, one or more operating systems 921, such as Windows Server TM , Mac OS X TM , Unix TM , LinuxTM, FreeBSDTM, etc.

[0161] The input / output interface 940 can be used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by a communications provider of the electronic device 900. In one embodiment, the input / output interface 940 includes a network adapter (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one embodiment, the input / output interface 940 can be a radio frequency (RF) module for wirelessly communicating with the Internet.

[0162] It can be understood by those skilled in the art that Figure 9 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 9 More or fewer components than shown, or with Figure 9 Different configurations shown.

[0163] An embodiment of the present application also provides a computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to implementing a method for determining a trip arrival time in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the method for determining the trip arrival time provided in the above method embodiment.

[0164] Optionally, in this embodiment, the storage medium may be located in at least one of a plurality of network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0165] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described above.

[0166] It can be seen from the embodiments of the method, device, equipment, server, terminal, storage medium and program product for determining the travel arrival time provided by the above-mentioned application that the application obtains multidimensional feature information of the sections passed through in the travel route to be estimated, and generates a section feature sequence of the travel route to be estimated based on the multidimensional feature information; then, based on the target feature generation network, feature extraction is performed on the section feature sequence to obtain the target route features of the travel route to be estimated; then, based on the target time prediction network, regression prediction processing is performed on the target route features to obtain the estimated arrival time of the travel route to be estimated; based on the multidimensional feature information of each section passed through in the route, a section feature sequence of the entire route is generated, and the target feature generation network and the target time prediction network are combined to perform time prediction on the section feature sequence, thereby significantly improving the prediction accuracy. In addition, the target feature generation network and the target time prediction network are obtained by iteratively training the initial generation network, the initial discriminant network and the initial prediction network through adversarial generation and regression prediction based on the historical segment feature sequence of the historical travel route, multi-dimensional route statistical information and the actual arrival time. In this way, the adversarial training method is used to assist in optimizing the feature extraction process, and the actual arrival time of the historical travel and the statistical information generated by the target object passing through each route segment are combined as the supervision information for training, thereby improving the model effect and accuracy of the network model, thereby improving the prediction accuracy and reliability of the estimated arrival time.

[0167] It should be noted that the order of the embodiments of the present application described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. The above description is of specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0168] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device, equipment, and storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0169] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk, or an optical disk, etc.

[0170] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for determining a trip arrival time, characterized in that: The method comprises: Generate route statistical features of historical travel routes based on multi-dimensional route statistical information; Extracting features from a historical segment feature sequence of the historical route based on an initial generation network to obtain historical route features; Using the historical route features, the route statistical features, and the actual arrival time as inputs to an initial discriminant network, a first discriminant probability and a second discriminant probability are obtained, wherein the first discriminant probability represents the degree of match between the actual arrival time and the historical route features, and the second discriminant probability represents the degree of match between the actual arrival time and the route statistical features; Performing regression prediction processing on the historical route features based on the initial prediction network to obtain a reference arrival time; Calculate the loss based on the first discrimination probability, the second discrimination probability, the reference arrival time, and the actual arrival time to obtain a generated loss, a discriminant loss, and a predicted loss; Based on the generation loss, the discrimination loss and the prediction loss, the initial generation network, the initial discrimination network and the initial prediction network are iteratively trained until a preset convergence condition is met, thereby obtaining a target feature generation network and a target time prediction network; Obtain multi-dimensional feature information of the road segments in the estimated travel route; Generating a road segment feature sequence of the to-be-estimated travel route according to the multi-dimensional feature information; Extracting features from the road segment feature sequence based on a target feature generation network to obtain target route features of the route to be estimated; The target route features are subjected to regression prediction processing based on the target time prediction network to obtain the estimated arrival time of the route to be estimated.

2. The method according to claim 1, characterized in that The iterative training of the initial generation network, the initial discrimination network, and the initial prediction network based on the generation loss, the discrimination loss, and the prediction loss until a preset convergence condition is satisfied to obtain the target feature generation network and the target time prediction network includes: Fixing the model parameters of the initial generation network and the initial prediction network, and performing a first training on the initial discriminant network based on the discriminant loss to update the model parameters of the initial discriminant network; Fixing the updated model parameters of the discriminant network, and performing a second training on the initial generation network and the initial prediction network based on the generation loss and the prediction loss to update the model parameters of the initial generation network and the initial prediction network; The first training and second training steps are performed alternately until the preset convergence condition is met, thereby obtaining the target feature generation network and the target time prediction network.

3. The method according to claim 1, characterized in that The iterative training of the initial generation network, the initial discrimination network, and the initial prediction network based on the generation loss, the discrimination loss, and the prediction loss until a preset convergence condition is satisfied to obtain the target feature generation network and the target time prediction network includes: determining a total loss based on the generation loss, the discrimination loss, and the prediction loss; Based on the total loss, the initial generation network, the initial discriminant network and the initial prediction network are jointly trained for adversarial generation and regression prediction to obtain the target feature generation network and the target time prediction network.

4. The method according to claim 1, wherein The loss calculation is performed based on the first discrimination probability, the second discrimination probability, the reference arrival time, and the actual arrival time to obtain the generated loss, the discrimination loss, and the predicted loss, including: Calculating the loss based on the first discrimination probability and the second discrimination probability to obtain the discrimination loss; Calculating the loss based on the deviation between the reference arrival time and the actual arrival time to obtain the predicted loss; Loss calculation is performed based on the first discriminant probability and the predicted loss to obtain the generated loss.

5. The method according to any one of claims 1 to 4, characterized in that The method of obtaining multi-dimensional feature information of the road segments passed through in the estimated travel route includes: Obtaining a road segment identification sequence and a target travel time corresponding to the estimated travel route, wherein the road segment identification sequence includes road segment identifications of each road segment in the estimated travel route; Based on the road segment identification sequence and the departure time, the multi-dimensional feature information of each of the passed road segments is obtained from the road network information and the driving data.

6. The method according to claim 5, characterized in that The step of generating the segment feature sequence of the to-be-estimated travel route according to the multi-dimensional feature information includes: Performing feature coding processing on the multi-dimensional feature information of each of the passed road sections to obtain a road section vector for each of the passed road sections; Feature splicing processing is performed on each road section vector based on the road section identification sequence to obtain the road section feature sequence.

7. The method according to claim 1, characterized in that The generating of the route statistical features of the historical travel route according to the multi-dimensional route statistical information includes: Performing feature coding processing on the multi-dimensional route statistical information to obtain initial statistical features of the historical travel route; Perform feature mapping processing on the initial statistical features to obtain the route statistical features.

8. The method according to claim 1, characterized in that Before obtaining the multi-dimensional feature information of the road segments in the estimated travel route, the method further includes: Acquire a target location and multiple candidate locations around the target location; Determine respectively the estimated travel routes between the target location and the plurality of candidate locations.

9. The method according to claim 8, characterized in that After performing regression prediction processing on the target route features based on the target time prediction network to obtain the estimated arrival time of the to-be-estimated travel route, the method further includes: Filtering a target candidate location that meets a preset time condition from the plurality of candidate locations according to the estimated arrival time; determining a target area corresponding to the target position based on the candidate target position; An area mark corresponding to the target area is displayed on a preset map interface.

10. A device for determining a trip arrival time, characterized in that: The device comprises: Statistical feature generation module: used to generate route statistical features of historical travel routes based on multi-dimensional route statistical information; A historical feature extraction module is used to extract the feature sequence of the historical sections of the historical route based on the initial generation network to obtain the historical route features; A discriminant network module is configured to use the historical route features, the route statistical features, and the actual arrival time as inputs to an initial discriminant network to obtain a first discriminant probability and a second discriminant probability, wherein the first discriminant probability represents the degree of match between the actual arrival time and the historical route features, and the second discriminant probability represents the degree of match between the actual arrival time and the route statistical features; Prediction network module: used to perform regression prediction processing on the historical route characteristics based on the initial prediction network to obtain a reference arrival time; Loss calculation module: used to calculate the loss according to the first discrimination probability, the second discrimination probability, the reference arrival time and the actual arrival time to obtain the generated loss, the discrimination loss and the predicted loss; Iterative training module: used for iteratively training the initial generation network, the initial discrimination network and the initial prediction network based on the generation loss, the discrimination loss and the prediction loss until the preset convergence conditions are met, thereby obtaining the target feature generation network and the target time prediction network; Multi-dimensional feature acquisition module: used to obtain multi-dimensional feature information of the road sections in the estimated travel route; A road segment feature generation module is configured to generate a road segment feature sequence of the to-be-estimated travel route based on the multi-dimensional feature information; Route feature extraction module: used for extracting features from the road segment feature sequence based on the target feature generation network to obtain target route features of the route to be estimated; Regression prediction module: used to perform regression prediction processing on the target route characteristics based on the target time prediction network to obtain the estimated arrival time of the route to be estimated.

11. The device according to claim 10, characterized in that The iterative training modules include: A first training submodule: used to fix the model parameters of the initial generation network and the initial prediction network, and perform a first training on the initial discriminant network based on the discriminant loss to update the model parameters of the initial discriminant network; A second training submodule: configured to fix the model parameters of the updated discriminant network, and perform a second training on the initial generation network and the initial prediction network based on the generation loss and the prediction loss to update the model parameters of the initial generation network and the initial prediction network; Alternating execution submodule: used to alternately execute the above-mentioned first training and second training steps until the preset convergence condition is met, thereby obtaining the target feature generation network and the target time prediction network.

12. The device according to claim 10, characterized in that The iterative training module includes: A total loss determination submodule: configured to determine a total loss based on the generation loss, the discrimination loss, and the prediction loss; Joint training submodule: used to perform joint training of adversarial generation and regression prediction on the initial generation network, the initial discriminant network and the initial prediction network based on the total loss to obtain the target feature generation network and the target time prediction network.

13. The device according to claim 10, characterized in that The loss calculation module includes: A discrimination loss calculation submodule: configured to perform loss calculation based on the first discrimination probability and the second discrimination probability to obtain the discrimination loss; Predicted loss calculation submodule: configured to calculate the loss according to the deviation between the reference arrival time and the actual arrival time to obtain the predicted loss; Generation loss calculation submodule: used to perform loss calculation based on the first discrimination probability and the predicted loss to obtain the generation loss.

14. The device according to any one of claims 10 to 13, characterized in that The multi-dimensional feature acquisition module includes: A first acquisition submodule is used to acquire a road segment identification sequence and a target travel time corresponding to the estimated travel route, wherein the road segment identification sequence includes the road segment identifications of each road segment in the estimated travel route; The second acquisition submodule is used to acquire the multi-dimensional feature information of each of the passed road sections from the road network information and driving data based on the road section identification sequence and the departure time.

15. The device according to claim 14, characterized in that The road segment feature generation module includes: A feature coding submodule is used to perform feature coding processing on the multi-dimensional feature information of each of the passed road sections to obtain a road section vector of each of the passed road sections; Feature splicing submodule: used for performing feature splicing processing on each road section vector based on the road section identification sequence to obtain the road section feature sequence.

16. The device according to claim 10, characterized in that The statistical feature generation module includes: Historical feature coding submodule: used to perform feature coding processing on the multi-dimensional route statistical information to obtain the initial statistical features of the historical travel route; Feature mapping submodule: used for performing feature mapping processing on the initial statistical features to obtain the route statistical features.

17. The device according to claim 10, characterized in that The device is also used for: Before obtaining the multi-dimensional feature information of the road segments in the estimated travel route, obtaining a target location and a plurality of candidate locations around the target location; Determine respectively the estimated travel routes between the target location and the plurality of candidate locations.

18. The device according to claim 17, characterized in that The device is also used for: After performing regression prediction processing on the target route features based on the target time prediction network to obtain an estimated arrival time of the to-be-estimated travel route, a target candidate location that meets a preset duration condition is screened from the plurality of candidate locations according to the estimated arrival time; determining a target area corresponding to the target position based on the candidate target position; An area mark corresponding to the target area is displayed on a preset map interface.

19. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the trip arrival time determination method according to any one of claims 1 to 9.

20. A computer device, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the trip arrival time determination method according to any one of claims 1 to 9.

21. A computer program product, characterized in that The computer program product comprises computer instructions, and when the computer instructions are executed by a processor, the method for determining the trip arrival time according to any one of claims 1 to 9 is implemented.

Images