A method and device for training a time of arrival estimation model, and a storage medium

Abstract

The application discloses a kind of training method, device and storage medium of time of arrival estimation model, can be applied to map and the field of Internet of Vehicles.The historical trajectory data is obtained;Then feature extraction is carried out to obtain a feature set;And the feature set is matched with time of arrival information to obtain training sample;Further, the scene estimation branch in time of arrival estimation model, future scene estimation branch and marked day scene estimation branch are uniformly trained based on training sample.Thereby, the uniform training process of multiple scene estimation branches is realized, since the corresponding type feature input is input according to the characteristics of different scenes, and the future scene estimation branch and the marked day scene estimation branch also utilize the real-time scene estimation branch for supervision during the training process, the accuracy of the training process of time of arrival estimation model is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a training method, apparatus, and storage medium for a time-of-arrival prediction model. Background Technology

[0002] With the rapid development of navigation technology, people have increasingly higher requirements for map navigation. Estimated time of arrival (ETA) is a function in map navigation that can estimate the time required to complete a given route on a map based on the departure time.

[0003] Generally, a rule-based, segment-by-segment accumulation method relies on human experience. Based on the length, speed, traffic light conditions, and other factors of each segment, the passage time of each segment is estimated. This time is then added to the passage time of each intersection, and the total time for the entire route is calculated.

[0004] However, for arrival time prediction in different scenarios, the model trained by manually input rules cannot accurately switch between arrival time prediction in different scenarios, which affects the accuracy of the arrival time prediction model training. Summary of the Invention

[0005] In view of this, this application provides a training method for an arrival time prediction model, which can effectively improve the accuracy of the arrival time prediction model training.

[0006] The first aspect of this application provides a training method for an arrival time estimation model, which can be applied to a system or program in a terminal device that includes a training function for an arrival time estimation model, specifically including:

[0007] Obtain historical trajectory data;

[0008] Feature extraction is performed on the historical trajectory data to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data;

[0009] The feature set is matched with the corresponding arrival time information to obtain training samples;

[0010] The arrival time prediction model is trained based on the training samples. The time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. The real-time scenario prediction branch is trained based on the static features, the historical features, and the real-time features, with the arrival time information as the training target. The future scenario prediction branch is trained based on the static features and the historical features. The marked-day scenario prediction branch is trained based on the static features and the real-time features. The future scenario prediction branch and the marked-day scenario prediction branch approximate the real-time scenario prediction branch during the training process.

[0011] Optionally, in some possible implementations of this application, the acquisition of historical trajectory data includes:

[0012] Acquire the collection trajectory;

[0013] Extract the location information sequence corresponding to the acquisition trajectory, and determine the time information corresponding to each location point in the location information sequence;

[0014] The location information sequence is matched with the time information to obtain the historical trajectory data.

[0015] Optionally, in some possible implementations of this application, the step of extracting features from the historical trajectory data to obtain a feature set includes:

[0016] Determine the road segment sequence in the historical trajectory data;

[0017] Feature extraction is performed on each path length and path type in the road segment sequence to obtain the static features;

[0018] The historical features are obtained by extracting the speed information of each path in the road segment sequence at different time dimensions.

[0019] Feature extraction is performed on the real-time speed information and congestion information corresponding to each path in the road segment sequence to obtain the real-time features;

[0020] The static features, historical features, and real-time features are collected to obtain the feature set.

[0021] Optionally, in some possible implementations of this application, training the arrival time prediction model based on the training samples includes:

[0022] The static features and historical features in the training samples are input into the fully connected layer in the future scene prediction branch for feature extraction, so as to train the future scene prediction branch based on the arrival time information. The fully connected layer in the future scene prediction branch only accepts the feature output of the future scene prediction branch.

[0023] The static features and real-time features in the training samples are input into the fully connected layer in the marked day scene prediction branch for feature extraction, so as to train the marked day scene prediction branch based on the arrival time information. The fully connected layer in the marked day scene prediction branch only accepts the feature output of the marked day scene prediction branch.

[0024] The static features, historical features, and real-time features in the training samples are input into the fully connected layer in the real-time scene prediction branch for feature extraction, so as to train the real-time scene prediction branch based on the arrival time information. The fully connected layer in the real-time scene prediction branch receives the feature outputs of the future scene prediction branch, the feature outputs of the marked day scene prediction branch, and the feature outputs of the real-time scene prediction branch.

[0025] Optionally, in some possible implementations of this application, the method further includes:

[0026] Obtain training metrics;

[0027] Based on the training metrics, the output of the future scene prediction branch is approximated to the arrival time information and the output of the real-time scene prediction branch to obtain the first loss information and the second loss information.

[0028] Based on the training metrics, the output of the marked day scene prediction branch is approximated to the arrival time information and the output of the real-time scene prediction branch to obtain the third loss information and the fourth loss information.

[0029] Based on the training metrics, the output of the real-time scene prediction branch is approximated to the arrival time information to obtain the fifth loss information.

[0030] Optionally, in some possible implementations of this application, the method further includes:

[0031] Obtain preset weighted information;

[0032] The first loss information, the second loss information, the third loss information, the fourth loss information, and the fifth loss information are weighted and calculated based on the preset weighted information to determine the target loss;

[0033] The arrival time prediction model is trained based on the target loss.

[0034] Optionally, in some possible implementations of this application, the method further includes:

[0035] Obtain the navigation request input from the target object;

[0036] Determine the path information and time information corresponding to the navigation request;

[0037] The path information is input into the arrival time prediction model to obtain the prediction information, which includes the real-time scene arrival time, the future scene arrival time, and the marked day scene arrival time.

[0038] Based on the time information, the matching items in the estimated information are determined, and the corresponding interface is displayed.

[0039] A second aspect of this application provides a training apparatus for an arrival time estimation model, comprising:

[0040] The acquisition unit is used to acquire historical trajectory data;

[0041] An extraction unit is used to extract features from the historical trajectory data to obtain a feature set, the feature set including static features, historical features and real-time features corresponding to the historical trajectory data;

[0042] The extraction unit is also used to match the feature set with the corresponding arrival time information to obtain training samples;

[0043] A training unit is used to train an arrival time prediction model based on the training samples. The time prediction model includes a real-time scene prediction branch, a future scene prediction branch, and a marked-day scene prediction branch. The real-time scene prediction branch is trained based on the static features, the historical features, and the real-time features, with the arrival time information as the training target. The future scene prediction branch is trained based on the static features and the historical features. The marked-day scene prediction branch is trained based on the static features and the real-time features. The future scene prediction branch and the marked-day scene prediction branch approximate the real-time scene prediction branch during the training process.

[0044] Optionally, in some possible implementations of this application, the acquisition unit is specifically used to acquire the acquisition trajectory;

[0045] The acquisition unit is specifically used to extract the location information sequence corresponding to the acquisition trajectory and determine the time information corresponding to each location point in the location information sequence;

[0046] The acquisition unit is specifically used to match the location information sequence with the time information to obtain the historical trajectory data.

[0047] Optionally, in some possible implementations of this application, the extraction unit is specifically used to determine the road segment sequence in the historical trajectory data;

[0048] The extraction unit is specifically used to extract features from each path length and path type in the road segment sequence to obtain the static features;

[0049] The extraction unit is specifically used to extract features from the speed information of each path in the road segment sequence at different time dimensions to obtain the historical features;

[0050] The extraction unit is specifically used to extract features from the real-time speed information and congestion information corresponding to each path in the road segment sequence to obtain the real-time features;

[0051] The extraction unit is specifically used to collect the static features, the historical features, and the real-time features to obtain the feature set.

[0052] Optionally, in some possible implementations of this application, the training unit is specifically used to input the static features and historical features in the training samples into the fully connected layer in the future scene prediction branch for feature extraction, so as to train the future scene prediction branch based on the arrival time information, and the fully connected layer in the future scene prediction branch only accepts the feature output of the future scene prediction branch.

[0053] The training unit is specifically used to input the static features and real-time features in the training samples into the fully connected layer in the marked day scene prediction branch for feature extraction, so as to train the marked day scene prediction branch based on the arrival time information. The fully connected layer in the marked day scene prediction branch only accepts the feature output of the marked day scene prediction branch.

[0054] The training unit is specifically used to input the static features, historical features, and real-time features from the training samples into the fully connected layer in the real-time scene prediction branch for feature extraction, so as to train the real-time scene prediction branch based on the arrival time information. The fully connected layer in the real-time scene prediction branch receives the feature outputs of the future scene prediction branch, the feature outputs of the marked day scene prediction branch, and the feature outputs of the real-time scene prediction branch.

[0055] Optionally, in some possible implementations of this application, the training unit is specifically used to acquire training metrics;

[0056] The training unit is specifically used to approximate the output of the future scene prediction branch with the arrival time information and the output of the real-time scene prediction branch based on the training metrics, so as to obtain the first loss information and the second loss information.

[0057] The training unit is specifically used to approximate the output of the marked day scene prediction branch with the arrival time information and the output of the real-time scene prediction branch based on the training index, so as to obtain the third loss information and the fourth loss information.

[0058] The training unit is specifically used to approximate the output of the real-time scene prediction branch with the arrival time information based on the training metrics, so as to obtain the fifth loss information.

[0059] Optionally, in some possible implementations of this application, the training unit is specifically used to obtain preset weighting information;

[0060] The training unit is specifically used to perform weighted calculations on the first loss information, the second loss information, the third loss information, the fourth loss information, and the fifth loss information based on the preset weighted information, so as to determine the target loss;

[0061] The training unit is specifically used to train the arrival time prediction model based on the target loss.

[0062] Optionally, in some possible implementations of this application, the training unit is specifically used to obtain the navigation request input by the target object;

[0063] The training unit is specifically used to determine the path information and time information corresponding to the navigation request;

[0064] The training unit is specifically used to input the path information into the arrival time prediction model to obtain prediction information, which includes real-time scene arrival time, future scene arrival time, and marked day scene arrival time.

[0065] The training unit is specifically used to determine the matching items in the estimated information based on the time information and to display the corresponding interface.

[0066] A third aspect of this application provides a computer device, comprising: a memory, a processor, and a bus system; the memory is used to store program code; the processor is used to execute the training method of the arrival time estimation model described in the first aspect or any one of the first aspects according to the instructions in the program code.

[0067] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the training method for the arrival time estimation model described in the first aspect or any one of the first aspects.

[0068] According to one aspect of this application, a computer program product or computer program is provided, comprising 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 training method for the time-of-arrival estimation model provided in the first aspect or various alternative implementations thereof.

[0069] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0070] By acquiring historical trajectory data and then extracting features from it to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data, and matching the feature set with corresponding arrival time information to obtain training samples, the arrival time prediction model is trained based on these training samples. The time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. The real-time scenario prediction branch is trained based on static features, historical features, and real-time features, using arrival time information as the training objective. The future scenario prediction branch is trained based on static features and historical features, and the marked-day scenario prediction branch is trained based on static features and real-time features. During training, the future scenario prediction branch and the marked-day scenario prediction branch approximate the real-time scenario prediction branch. This achieves a unified training process for multiple scenario prediction branches. Because corresponding types of features are input according to the characteristics of different scenarios, and the future scenario prediction branch and the marked-day scenario prediction branch utilize the real-time scenario prediction branch for supervision during training, the accuracy of the arrival time prediction model training process is improved. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0072] Figure 1 A diagram showing the network architecture of the system used to train the time-of-arrival prediction model;

[0073] Figure 2 A flowchart illustrating the training process of an arrival time estimation model provided in this application embodiment;

[0074] Figure 3 A flowchart illustrating a training method for an arrival time estimation model provided in an embodiment of this application;

[0075] Figure 4 A schematic diagram illustrating a training method for an arrival time estimation model provided in an embodiment of this application;

[0076] Figure 5 A schematic diagram illustrating a training method for another arrival time estimation model provided in this application embodiment;

[0077] Figure 6 A schematic diagram illustrating a training method for another arrival time estimation model provided in this application embodiment;

[0078] Figure 7 A schematic diagram illustrating a training method for another arrival time estimation model provided in this application embodiment;

[0079] Figure 8 A schematic diagram of the structure of a training device for an arrival time estimation model provided in an embodiment of this application;

[0080] Figure 9 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application;

[0081] Figure 10 This is a schematic diagram of the structure of a server provided in an embodiment of this application. Detailed Implementation

[0082] This application provides a training method and related apparatus for an arrival time prediction model, which can be applied to a system or program in a terminal device that includes a training function for an arrival time prediction model. The method involves acquiring historical trajectory data; then extracting features from the historical trajectory data to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data; matching the feature set with corresponding arrival time information to obtain training samples; and then training the arrival time prediction model based on the training samples. The arrival time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. The real-time scenario prediction branch is trained based on static features, historical features, and real-time features, using arrival time information as the training target. The future scenario prediction branch is trained based on static features and historical features. The marked-day scenario prediction branch is trained based on static features and real-time features. During training, the future scenario branch and the marked-day scenario branch approximate the real-time scenario prediction branch. This enables a unified training process for multiple scenario prediction branches. By inputting corresponding types of features based on the characteristics of different scenarios, and by using the real-time scenario prediction branch for supervision during the training process for the future scenario prediction branch and the marked day scenario prediction branch, the accuracy of the arrival time prediction model training process is improved.

[0083] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0084] First, some terms that may appear in the embodiments of this application will be explained.

[0085] Estimated Time of Arrival (ETA): A basic function in map software, it estimates the time required to complete a given route on a map, based on the given departure time.

[0086] Real-time ETA: The time required to complete the entire journey from the current moment.

[0087] Future ETA: The time required to complete the entire journey starting at a future point in time. The main difference between real-time ETA and future ETA lies in the road conditions. Real-time ETA uses the road conditions at the current moment; future ETA uses road conditions predicted for a future moment. Future ETA is sometimes referred to as future travel time.

[0088] Holiday ETA: During holidays, travel patterns differ significantly from weekdays, so directly using weekday real-time ETAs can lead to substantial errors. Therefore, special strategies are generally required for holidays. Furthermore, each holiday has its own characteristics; for example, travel patterns differ between the May Day and National Day holidays, necessitating different handling strategies.

[0089] Request time: The time when a user initiates a real-time or future ETA request on a platform such as an APP or mini-program is the request time; while the time when the user expects to depart is the departure time.

[0090] Departure Time: For real-time ETAs, the request time and departure time are basically the same, meaning that the user departs almost immediately after requesting the ETA. However, for future ETAs, a user can request the calculation of an ETA departing at 9:00 AM at 8:00 AM, meaning the request time is 8:00 AM and the departure time is 9:00 AM.

[0091] Route: In map applications, a route is simply a complete line connecting the start and end points. In real-world scenarios, the length of a route is usually between one kilometer and tens of kilometers.

[0092] Links: In map applications, routes are represented by sequences of links. In map data, roads are divided into line segments, ranging in length from tens of meters to several kilometers. Each line segment is called a link and is assigned a globally unique ID. Therefore, a route on a map is a sequence of all links within that route.

[0093] Actual Time of Arrival (ATA): Also known as arrival time information, this data can be extracted from the historical data of map services to identify a route. Therefore, this data can be used as the ground truth to train machine learning algorithms to estimate arrival time.

[0094] Historical Classic Speed: The speed mined for each link based on the historical trajectory of all users. Typically, the cycle is one week, with a granularity of 5 minutes, meaning each link yields 7*24*12=2016 speed values, representing the travel speed every 5 minutes from Monday to Sunday.

[0095] Trajectory matching: A user's GPS point sequence needs to be associated with the road network to obtain richer information. Matching each GPS point to its corresponding road segment is trajectory matching. Commonly used trajectory matching algorithms include Hidden Markov Models and Recurrent Neural Networks.

[0096] It should be understood that the training method for the arrival time estimation model provided in this application can be applied to systems or programs in terminal devices that include the training function of the arrival time estimation model, such as map applications. Specifically, the training system for the arrival time estimation model can run on, for example,... Figure 1 In the network architecture shown, such as Figure 1 The diagram shows the network architecture of the training system for the time-of-arrival (TOA) prediction model. As can be seen, the TOA training system can provide training processes for TOA models from multiple information sources. Specifically, navigation operations on the terminal side trigger the server to estimate the time for the corresponding route, and the estimation process uses a pre-trained TOA model. This means that… Figure 1 The diagram illustrates various terminal devices, which can be computer devices. In real-world scenarios, more or fewer types of terminal devices may participate in the training process of the arrival time prediction model. The specific number and types depend on the actual scenario and are not limited here. Figure 1 The example shows one server, but in real-world scenarios, multiple servers can be involved, especially in scenarios involving multi-model training and interaction. The specific number of servers depends on the actual scenario.

[0097] In this embodiment, the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, smart voice interaction device, smart home appliance, in-vehicle terminal, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, and the terminal and server can be connected to form a blockchain network; this application does not impose any restrictions.

[0098] It is understood that the training system for the aforementioned time of arrival prediction model can run on a personal mobile terminal, such as a map application, or on a server, or on a third-party device to provide training for the time of arrival prediction model in order to obtain the training processing results of the time of arrival prediction model for the information source. Specifically, the training system for the time of arrival prediction model can run as a program on the aforementioned device, or as a system component within the aforementioned device, or as a cloud service program. This embodiment can be applied to scenarios such as cloud technology and autonomous driving. The specific operating mode depends on the actual scenario and is not limited here.

[0099] With the rapid development of navigation technology, people have increasingly higher requirements for map navigation. Estimated time of arrival (ETA) is a function in map navigation that can estimate the time required to complete a given route on a map based on the departure time.

[0100] Generally, a rule-based, segment-by-segment accumulation method relies on human experience. Based on the length, speed, traffic light conditions, and other factors of each segment, the passage time of each segment is estimated. This time is then added to the passage time of each intersection, and the total time for the entire route is calculated.

[0101] However, for arrival time prediction in different scenarios, the model trained by manually input rules cannot accurately switch between arrival time prediction in different scenarios, which affects the accuracy of the arrival time prediction model training.

[0102] To address the aforementioned issues, this application proposes a training method for an arrival time estimation model, which is applied to... Figure 2 The training process framework for the arrival time estimation model is shown below, such as... Figure 2 The diagram shown is a flowchart of the training process for an arrival time prediction model provided in this application embodiment. Users interact with interactive videos through the interface layer, and trigger the target chart in the application layer through the interface operations involved in the interaction process, thereby switching media content between multiple interactive videos associated with the target chart.

[0103] It is understood that the method provided in this application can be a program written as processing logic in a hardware system, or it can be a training device for an arrival time prediction model, implementing the aforementioned processing logic in an integrated or external manner. As one implementation, the training device for the arrival time prediction model acquires historical trajectory data; then extracts features from the historical trajectory data to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data; and matches the feature set with corresponding arrival time information to obtain training samples; subsequently, it trains the arrival time prediction model based on the training samples. The time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. The real-time scenario prediction branch is trained based on static features, historical features, and real-time features, using arrival time information as the training target. The future scenario prediction branch is trained based on static features and historical features. The marked-day scenario prediction branch is trained based on static features and real-time features. During training, the future scenario branch and the marked-day scenario prediction branch approximate the real-time scenario prediction branch. This enables a unified training process for multiple scenario prediction branches. By inputting corresponding types of features based on the characteristics of different scenarios, and by using the real-time scenario prediction branch for supervision during the training process for the future scenario prediction branch and the marked day scenario prediction branch, the accuracy of the arrival time prediction model training process is improved.

[0104] The solutions provided in this application relate to deep learning technology in artificial intelligence, and are specifically illustrated through the following embodiments:

[0105] Based on the above process architecture, the training method of the arrival time prediction model in this application will be introduced below. Please refer to [link / reference]. Figure 3 , Figure 3 The flowchart illustrates a training method for an arrival time estimation model provided in this application embodiment. This management method can be executed by a server or a terminal, and this application embodiment includes at least the following steps:

[0106] 301. Obtain historical trajectory data.

[0107] In this embodiment, the historical trajectory data can be the collected trajectory data or the trajectory data stored on the server, or it can be the trajectory data downloaded from the cloud through the Internet of Vehicles. The specific data source is not limited.

[0108] Specifically, the historical trajectory data collection process begins by acquiring the trajectory itself, which is a sequence of GPS points. The location information sequence corresponding to the trajectory is then extracted, and the time information corresponding to each location point is determined. The location information sequence is then matched with the time information to obtain the historical trajectory data, as each GPS point also contains information such as latitude and longitude, timestamp, speed, and direction angle. From this information, the user's departure time t0 and the time ATA to complete a route can be extracted.

[0109] Additionally, links along a user's route can be obtained from GPS point sequences; this process is also known as trajectory matching. Commonly used algorithms include Hidden Markov Models (HMMs) and Recurrent Neural Networks (RNNs). The obtained departure time t0, ATA, and link sequence constitute a data set; all samples collected over a period of time constitute the entire dataset.

[0110] In one possible scenario, the historical trajectory data in this embodiment is used for time of arrival (ETA) estimation in multiple scenarios, specifically real-time ETA, future ETA, and marked-date ETA; where marked-date ETA is an ETA for a specific date, and can also be called holiday ETA.

[0111] Specifically, real-time ETA represents the time required to complete the entire journey from the current moment. Future ETA represents the time required to complete the entire journey from a future moment; the main difference between real-time and future ETA lies in road conditions. Real-time ETA uses current road conditions, while future ETA uses predicted road conditions for future moments. Future ETA is sometimes referred to as future travel time. Furthermore, for holidays, travel patterns differ significantly from weekdays, and directly using weekday real-time ETAs will result in substantial errors. Therefore, special strategies are generally required for holidays. Moreover, each holiday has its own characteristics; for example, travel patterns differ between May Day and National Day holidays, leading to different processing strategies. This embodiment considers these differences and employs a unified multi-branch model training process.

[0112] 302. Extract features from historical trajectory data to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data.

[0113] In this embodiment, static features, historical features, and real-time features are descriptions of feature types. The specific acquisition process can be as follows: First, determine the link sequence in the historical trajectory data; then, extract features from the length and type of each path in the link sequence to obtain static features, such as the total mileage and the proportion of expressways; then, extract features from the speed information of each path in the link sequence at different time dimensions to obtain historical features, such as the average historical speed of each link in the entire route; and then extract features from the real-time speed and congestion information corresponding to each path in the link sequence to obtain real-time features, such as the average real-time speed of each link in the entire route and the total congestion mileage; finally, collect static features, historical features, and real-time features to obtain a feature set.

[0114] Specifically, the determination of the aforementioned static, historical, and real-time features can be achieved through statistical analysis of the basic attributes of the link sequence. Since a route is a link sequence, the following types of features need to be extracted for each link: basic attribute features: such as link mileage, road grade, width, and presence of traffic lights; historical classic speed features at departure time t0: for example, within the time interval from t0-30min to t0+30min, using a 5-minute granularity, the 12 historical classic speeds and their standard deviations for this link; and real-time traffic condition features at departure time t0: for example, within the time interval from t0-30min to t0, using a 5-minute granularity, the 6 traffic condition states, 6 real-time speeds, and 6 real-time traffic flows for this link.

[0115] After the above feature statistics, each link will obtain a K-dimensional vector. The K-dimensional vectors of all links form a sequence of length M. Therefore, the feature extracted from the link sequence is an M*K matrix. Furthermore, route-level features can be extracted based on human experience, which is equivalent to manual feature engineering. For example: total mileage, the proportion of highways along the entire route, the average historical classic speed of each link along the entire route (since each link has 12 historical classic speeds, there are 12 average historical classic speeds), the average real-time speed of each link along the entire route (which can be an arithmetic average or a harmonic average), the total congestion mileage, etc. The specific feature composition depends on the actual scenario and is not limited here.

[0116] 303. Match the feature set with the corresponding arrival time information to obtain training samples.

[0117] In this embodiment, through feature statistics in step 302, each route will obtain a set of features, denoted as L-dimensional; the above features, together with ATA, constitute a training sample; furthermore, the training samples extracted from all samples constitute the entire training set.

[0118] 304. The arrival time prediction model is trained based on training samples. The time prediction model includes a real-time scene prediction branch, a future scene prediction branch, and a marked-day scene prediction branch. The real-time scene prediction branch is trained based on static features, historical features, and real-time features with arrival time information as the training target. The future scene prediction branch is trained based on static features and historical features. The marked-day scene prediction branch is trained based on static features and real-time features. The future scene prediction branch and the marked-day scene prediction branch approach the real-time scene prediction branch during the training process.

[0119] Understandably, the approximation process indicates that when the input data is the same, the outputs of the model branches are close, that is, the output differences of the model branches are as small as possible. This output difference can be indicated by the difference in absolute value or by the approximation index, such as making the approximation index (mean absolute percentage error, mean square error, etc.) as close as possible to a certain specific value.

[0120] Furthermore, since this embodiment includes multiple approximation processes, namely, the output of the future scene prediction branch approximating the output of the real-time scene prediction branch, the output of the marked-day scene prediction branch approximating the output of the real-time scene prediction branch, the output of the future scene prediction branch approximating the arrival time information, the output of the real-time scene prediction branch approximating the arrival time information, and the output of the marked-day scene prediction branch approximating the arrival time information; by using approximation metrics to guide the training process, the overall effect of the training process of multiple branches can be improved.

[0121] In this embodiment, the arrival time estimation model is a model with multiple branches, specifically as follows: Figure 4 As shown, Figure 4 This is a schematic diagram of a training method for an arrival time prediction model provided in an embodiment of this application. The diagram shows that the time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. Different feature groups are used for training the holiday, future, and real-time scenarios. Specifically, the holiday scenario uses only static and real-time features; the future scenario uses only static and historical features; and the real-time scenario uses all three types of features.

[0122] Specifically, the training process for the time prediction model involves inputting the static and historical features from the training samples into the fully connected layer of the future scene prediction branch for feature extraction. This allows the future scene prediction branch to be trained based on arrival time information, and the fully connected layer in this branch only accepts the feature outputs of the future scene prediction branch. Similarly, the static and real-time features from the training samples are input into the fully connected layer of the marked-day scene prediction branch for feature extraction. This allows the marked-day scene prediction branch to be trained based on arrival time information, and the fully connected layer in this branch only accepts the feature outputs of the marked-day scene prediction branch. Simultaneously, the static, historical, and real-time features from the training samples are input into the fully connected layer of the real-time scene prediction branch for feature extraction. This allows the real-time scene prediction branch to be trained based on arrival time information, and the fully connected layer in this branch accepts the feature outputs of the future scene prediction branch, the marked-day scene prediction branch, and the real-time scene prediction branch. Each scenario contains several layers of fully connected networks; the real-time ETA scenario accepts all three types of features as input, and at each layer, it accepts the outputs of the other two scenarios as input; while the holiday ETA only uses real-time features and static features, and the future ETA only uses historical features and static features; in addition, the intermediate layers of the holiday ETA scenario and the future ETA scenario only accept the output of the layer before itself as input.

[0123] For the loss during training, training metrics can be obtained. Then, based on the training metrics, the output of the future scenario prediction branch is approximated to the output of the arrival time information and the real-time scenario prediction branch to obtain the first and second loss information. Similarly, based on the training metrics, the output of the marked day scenario prediction branch is approximated to the output of the arrival time information and the real-time scenario prediction branch to obtain the third and fourth loss information. Simultaneously, based on the training metrics, the output of the real-time scenario prediction branch is approximated to the arrival time information to obtain the fifth loss information. In other words, the ETA for all three scenarios needs to approximate the ATA; and the holiday ETA and future ETA also need to approximate the real-time ETA.

[0124] In one possible scenario, the training metric can be called an "approximation" metric, which can be MAPE, MSE, etc., and is defined as follows:

[0125]

[0126] MSE=(ETA-ATA) 2

[0127] ETA represents the estimated arrival time for each scenario, while ATA represents the actual arrival time. Other metrics can be used for approximation, and no restrictions are imposed here.

[0128] Furthermore, the five approximation targets mentioned above can be weighted, that is, firstly, pre-defined weighted information is obtained; then, based on the pre-defined weighted information, the first loss information, the second loss information, the third loss information, the fourth loss information, and the fifth loss information are weighted and calculated to determine the target loss; and then the arrival time prediction model is trained based on the target loss.

[0129] Specifically, the weighted calculation can be performed according to the following formula:

[0130]

[0131] Where, ω i The weights for each approximation target. Generally, it's sufficient to simply set ω. i =1 is sufficient; if necessary, it can also be set based on experience.

[0132] It's important to note that when calculating the loss function above, the approximation is made from the future and holiday ETAs towards the real-time ETA, not the other way around. Therefore, during backpropagation, it's crucial to ensure that gradients only propagate along the paths of future and holiday ETAs, avoiding backpropagation along the real-time ETA. In TensorFlow, this can be achieved using `tf.stop_gradient(eta_realtime)`, and similar approaches can be used for other deep learning frameworks.

[0133] By training a unified model as described above, there will be no disconnect or jump when switching between ETA before and after holidays, in real time, and in the future. Furthermore, the collaborative training of future, holiday, and real-time ETA improves the synergy among the three scenarios. In addition, all three use ATA as the target, and the future and holiday ETA also use real-time ETA as the target. By utilizing historical trajectory data, supervision information is added, which improves the accuracy of time prediction.

[0134] The following describes the process of using the above arrival time prediction model: First, the navigation request input by the target object is obtained; then, the path information and time information corresponding to the navigation request are determined, namely the destination and departure time of the navigation; the path information is input into the arrival time prediction model to obtain the prediction information, which includes the real-time scene arrival time, the future scene arrival time, and the marked day scene arrival time; then, based on the time information, the matching items in the prediction information are determined and the corresponding interface is displayed.

[0135] For the matching process, if the current time is a weekday and it is a real-time request, the real-time ETA is returned; if the current time is a weekday and it is a future request, the future ETA is returned; if the current time is a holiday and it is a real-time request, the holiday ETA is returned; if the current time is a holiday and it is a future request, the future ETA is returned.

[0136] In one possible scenario, the interface display of this embodiment is as follows: Figure 5 As shown, Figure 5 This diagram illustrates a scenario for training another time-of-arrival (ETA) prediction model provided in this application embodiment. The diagram shows a possible display scenario for real-time ETA and future ETA. The upper part displays the real-time ETA for three candidate routes. Selecting one route will display its future travel time below. The future travel time is calculated every 15 minutes; actual products may have different display formats, which are not limited here.

[0137] Additionally, regarding the setup process for navigation requests, such as... Figure 6 As shown, Figure 6 This is a schematic diagram of a training method for another arrival time prediction model provided in an embodiment of this application; the diagram shows that clicking "Set Departure Time" will take you to another page where you can select a specific departure time, thus supporting infinitely far-reaching future times.

[0138] In one possible scenario, for the selected scenario at the departure time, Figure 7 This is a schematic diagram of a training method for another arrival time estimation model provided in this application embodiment; the diagram shows the route and several ETA bar charts before and after the selected departure time, thus intuitively displaying the future ETA.

[0139] Specifically, the arrival time prediction model trained in this embodiment can be applied when initiating navigation. The background first provides several candidate routes, then uses this embodiment to calculate the estimated arrival time of each candidate route, and then selects the fastest route to provide to the user. After entering the navigation state, this embodiment can also be used to calculate the remaining travel time at regular intervals to facilitate the user's trip planning.

[0140] In addition, this embodiment can be used to calculate an isochronous reachable circle, such as a half-hour reachable circle or a one-hour reachable circle, to help users understand the living radius of a certain location. Specifically, in food delivery, this embodiment can be used to accurately calculate the time taken for each route, thereby better assigning orders to delivery personnel and improving delivery efficiency. In the context of mobile ride-hailing, this embodiment can be used to accurately calculate the time taken for each route, thereby better arranging drivers to accept orders and improving passenger transport efficiency.

[0141] Optionally, the ETA for each route can be provided for upstream services to evaluate the advantages and disadvantages of each route and then push the optimal route to the user; and the influence weight of each road segment on the ETA can be provided for upstream services to use, for example, to avoid congestion and interpret the estimated time.

[0142] Based on the above application scenario description, after the user selects the starting point and destination, the map software provides the shortest route. Finding this route requires estimating the time for each candidate route. Alternatively, a future time can be specified, and the future travel time for each of the multiple candidate routes provided by the backend can be calculated using this embodiment. Then, the fastest route is selected and provided to the user. After the user initiates navigation, the remaining travel time needs to be continuously reported to the user during navigation. Furthermore, given a route, a graph showing its future travel time can be provided to assist the user's travel decision.

[0143] In practical scenarios, food delivery applications need to allocate orders reasonably to delivery drivers. This allocation requires calculating the total time from order pickup to delivery based on the customer's location, the store's location, and the delivery driver's location. In other words, during logistics delivery, specifying a future time allows for the calculation of future travel times for each route, thus better planning delivery times and improving efficiency. Similarly, in ride-hailing applications, it's crucial to match users with taxis to minimize empty taxi runs. Route planning requires accurate time estimates for each possible route. This also allows for the accurate calculation of future travel times for each route by specifying a future time, enabling better driver booking and improved passenger transport efficiency.

[0144] As described in the above embodiments, historical trajectory data is acquired; then features are extracted from the historical trajectory data to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data; the feature set is then matched with the corresponding arrival time information to obtain training samples; and finally, the arrival time prediction model is trained based on the training samples. The time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. The real-time scenario prediction branch is trained based on static features, historical features, and real-time features, using arrival time information as the training target. The future scenario prediction branch is trained based on static features and historical features, and the marked-day scenario prediction branch is trained based on static features and real-time features. During training, the future scenario prediction branch and the marked-day scenario prediction branch approximate the real-time scenario prediction branch. This achieves a unified training process for multiple scenario prediction branches. Because corresponding types of features are input according to the characteristics of different scenarios, and the future scenario prediction branch and the marked-day scenario prediction branch also utilize the real-time scenario prediction branch for supervision during training, the accuracy of the arrival time prediction model training process is improved.

[0145] To better implement the above-described solutions of the embodiments of this application, related apparatus for implementing the above solutions is also provided below. Please refer to... Figure 8 , Figure 8 This application provides a schematic diagram of the structure of a training device for an arrival time estimation model, the training device 800 including:

[0146] Acquisition unit 801 is used to acquire historical trajectory data;

[0147] The extraction unit 802 is used to extract features from the historical trajectory data to obtain a feature set, the feature set including static features, historical features and real-time features corresponding to the historical trajectory data;

[0148] The extraction unit 802 is also used to match the feature set with the corresponding arrival time information to obtain training samples;

[0149] Training unit 803 is used to train the arrival time prediction model based on the training samples. The time prediction model includes a real-time scene prediction branch, a future scene prediction branch, and a marked-day scene prediction branch. The real-time scene prediction branch is trained based on the static features, the historical features, and the real-time features, with the arrival time information as the training target. The future scene prediction branch is trained based on the static features and the historical features. The marked-day scene prediction branch is trained based on the static features and the real-time features. The future scene prediction branch and the marked-day scene prediction branch approximate the real-time scene prediction branch during the training process.

[0150] Optionally, in some possible implementations of this application, the acquisition unit 801 is specifically used to acquire the acquisition trajectory;

[0151] The acquisition unit 801 is specifically used to extract the location information sequence corresponding to the acquisition trajectory and determine the time information corresponding to each location point in the location information sequence;

[0152] The acquisition unit 801 is specifically used to match the location information sequence with the time information to obtain the historical trajectory data.

[0153] Optionally, in some possible implementations of this application, the extraction unit 802 is specifically used to determine the road segment sequence in the historical trajectory data;

[0154] The extraction unit 802 is specifically used to extract features from each path length and path type in the road segment sequence to obtain the static features;

[0155] The extraction unit 802 is specifically used to extract features from the speed information of each path in the road segment sequence at different time dimensions to obtain the historical features;

[0156] The extraction unit 802 is specifically used to extract features from the real-time speed information and congestion information corresponding to each path in the road segment sequence to obtain the real-time features;

[0157] The extraction unit 802 is specifically used to collect the static features, the historical features, and the real-time features to obtain the feature set.

[0158] Optionally, in some possible implementations of this application, the training unit 803 is specifically used to input the static features and historical features in the training samples into the fully connected layer in the future scene prediction branch for feature extraction, so as to train the future scene prediction branch based on the arrival time information, and the fully connected layer in the future scene prediction branch only accepts the feature output of the future scene prediction branch.

[0159] The training unit 803 is specifically used to input the static features and real-time features in the training samples into the fully connected layer in the marked day scene prediction branch for feature extraction, so as to train the marked day scene prediction branch based on the arrival time information. The fully connected layer in the marked day scene prediction branch only accepts the feature output of the marked day scene prediction branch.

[0160] The training unit 803 is specifically used to input the static features, historical features and real-time features in the training samples into the fully connected layer in the real-time scene prediction branch for feature extraction, so as to train the real-time scene prediction branch based on the arrival time information. The fully connected layer in the real-time scene prediction branch receives the feature outputs of the future scene prediction branch, the feature outputs of the marked day scene prediction branch and the feature outputs of the real-time scene prediction branch.

[0161] Optionally, in some possible implementations of this application, the training unit 803 is specifically used to acquire training metrics;

[0162] The training unit 803 is specifically used to approximate the output of the future scene prediction branch with the arrival time information and the output of the real-time scene prediction branch based on the training index, so as to obtain the first loss information and the second loss information.

[0163] The training unit 803 is specifically used to approximate the output of the marked day scene prediction branch to the arrival time information and the output of the real-time scene prediction branch based on the training index, so as to obtain the third loss information and the fourth loss information.

[0164] The training unit 803 is specifically used to approximate the output of the real-time scene prediction branch with the arrival time information based on the training index, so as to obtain the fifth loss information.

[0165] Optionally, in some possible implementations of this application, the training unit 803 is specifically used to obtain preset weighting information;

[0166] The training unit 803 is specifically used to perform weighted calculations on the first loss information, the second loss information, the third loss information, the fourth loss information, and the fifth loss information based on the preset weighted information, so as to determine the target loss;

[0167] The training unit 803 is specifically used to train the arrival time prediction model based on the target loss.

[0168] Optionally, in some possible implementations of this application, the training unit 803 is specifically used to obtain the navigation request input by the target object;

[0169] The training unit 803 is specifically used to determine the path information and time information corresponding to the navigation request;

[0170] The training unit 803 is specifically used to input the path information into the arrival time prediction model to obtain prediction information, which includes real-time scene arrival time, future scene arrival time and marked day scene arrival time.

[0171] The training unit 803 is specifically used to determine the matching items in the estimated information based on the time information and to display the corresponding interface.

[0172] By acquiring historical trajectory data and then extracting features from it to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data, and matching the feature set with corresponding arrival time information to obtain training samples, the arrival time prediction model is trained based on these training samples. The time prediction model includes a real-time scenario prediction branch, a future scenario prediction branch, and a marked-day scenario prediction branch. The real-time scenario prediction branch is trained based on static features, historical features, and real-time features, using arrival time information as the training objective. The future scenario prediction branch is trained based on static features and historical features, and the marked-day scenario prediction branch is trained based on static features and real-time features. During training, the future scenario prediction branch and the marked-day scenario prediction branch approximate the real-time scenario prediction branch. This achieves a unified training process for multiple scenario prediction branches. Because corresponding types of features are input according to the characteristics of different scenarios, and the future scenario prediction branch and the marked-day scenario prediction branch utilize the real-time scenario prediction branch for supervision during training, the accuracy of the arrival time prediction model training process is improved.

[0173] This application also provides a terminal device, such as... Figure 9 The diagram shown is a structural schematic of another terminal device provided in an embodiment of this application. For ease of explanation, only the parts related to the embodiment of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiment of this application. The terminal can be any terminal device including mobile phones, tablet computers, personal digital assistants (PDAs), point-of-sale (POS) terminals, in-vehicle computers, etc. Taking a mobile phone as an example:

[0174] Figure 9 This is a block diagram illustrating a portion of the structure of a mobile phone related to the terminal provided in the embodiments of this application. (Reference) Figure 9The mobile phone includes components such as a radio frequency (RF) circuit 910, a memory 920, an input unit 930, a display unit 940, a sensor 950, an audio circuit 960, a wireless fidelity (WiFi) module 970, a processor 980, and a power supply 990. Those skilled in the art will understand that... Figure 9 The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0175] The following is combined Figure 9 A detailed introduction to each component of a mobile phone:

[0176] RF circuit 910 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with processor 980; additionally, it transmits uplink data to the base station. Typically, RF circuit 910 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), and a duplexer. Furthermore, RF circuit 910 can also communicate wirelessly with networks and other devices. The aforementioned wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, and Short Message Service (SMS).

[0177] The memory 920 can be used to store software programs and modules. The processor 980 executes various functions and data processing of the mobile phone by running the software programs and modules stored in the memory 920. The memory 920 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 920 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0178] The input unit 930 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the mobile phone. Specifically, the input unit 930 may include a touch panel 931 and other input devices 932. The touch panel 931, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 931, and air touch operations within a certain range on the touch panel 931), and drive the corresponding connection devices according to a pre-set program. Optionally, the touch panel 931 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, sends it to the processor 980, and can receive and execute commands sent by the processor 980. In addition, the touch panel 931 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 931, the input unit 930 may also include other input devices 932. Specifically, other input devices 932 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0179] The display unit 940 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 940 may include a display panel 941, which may optionally be configured as a liquid crystal display (LCD), organic light-emitting diode (OLED), or similar form. Further, a touch panel 931 may cover the display panel 941. When the touch panel 931 detects a touch operation on or near it, it transmits the information to the processor 980 to determine the type of touch event. Subsequently, the processor 980 provides corresponding visual output on the display panel 941 based on the type of touch event. Although in Figure 9 In this embodiment, the touch panel 931 and the display panel 941 are two separate components to realize the input and output functions of the mobile phone. However, in some embodiments, the touch panel 931 and the display panel 941 can be integrated to realize the input and output functions of the mobile phone.

[0180] The mobile phone may also include at least one sensor 950, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 941 according to the ambient light level, and the proximity sensor can turn off the display panel 941 and / or backlight when the phone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity, which can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, taps), etc. Other sensors that may be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0181] Audio circuit 960, speaker 961, and microphone 962 provide an audio interface between the user and the mobile phone. Audio circuit 960 converts received audio data into electrical signals and transmits them to speaker 961, where speaker 961 converts them into sound signals for output. On the other hand, microphone 962 converts collected sound signals into electrical signals, which are received by audio circuit 960, converted into audio data, and then processed by processor 980 before being transmitted via RF circuit 910 to, for example, another mobile phone, or the audio data can be output to memory 920 for further processing.

[0182] WiFi is a short-range wireless transmission technology. Mobile phones, through the WiFi module 970, can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access. Although Figure 9The WiFi module 970 is shown, but it is understood that it is not an essential component of a mobile phone and can be omitted as needed without changing the essence of the invention.

[0183] The processor 980 is the control center of the mobile phone, connecting various parts of the phone through various interfaces and lines. It executes software programs and / or modules stored in the memory 920, and calls data stored in the memory 920 to perform various functions and process data, thereby providing overall monitoring of the phone. Optionally, the processor 980 may include one or more processing units; optionally, the processor 980 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the aforementioned modem processor may also not be integrated into the processor 980.

[0184] The phone also includes a power supply 990 (such as a battery) that supplies power to various components. Optionally, the power supply can be logically connected to the processor 980 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0185] Although not shown, mobile phones may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0186] In this embodiment of the application, the processor 980 included in the terminal also has the function of performing the various steps of the page processing method described above.

[0187] This application also provides a server; please refer to [link / reference]. Figure 10 , Figure 10 This is a schematic diagram of a server structure provided in an embodiment of this application. The server 1000 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 1022 (e.g., one or more processors) and memory 1032, and one or more storage media 1030 (e.g., one or more mass storage devices) for storing application programs 1042 or data 1044. The memory 1032 and storage media 1030 can be temporary or persistent storage. The program stored in the storage media 1030 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the server. Furthermore, the CPU 1022 may be configured to communicate with the storage media 1030 and execute the series of instruction operations in the storage media 1030 on the server 1000.

[0188] Server 1000 may also include one or more power supplies 1026, one or more wired or wireless network interfaces 1050, one or more input / output interfaces 1058, and / or one or more operating systems 1041, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0189] The steps performed by the management device in the above embodiments can be based on this Figure 10 The server structure shown.

[0190] This application embodiment also provides a computer-readable storage medium storing training instructions for an arrival time estimation model, which, when run on a computer, causes the computer to perform the aforementioned actions. Figures 3 to 7 The steps performed by the training apparatus for the arrival time prediction model in the method described in the illustrated embodiment.

[0191] This application also provides a computer program product including training instructions for an arrival time estimation model, which, when run on a computer, causes the computer to perform the aforementioned... Figures 3 to 7 The steps performed by the training apparatus for the arrival time prediction model in the method described in the illustrated embodiment.

[0192] This application also provides a training system for an arrival time estimation model, which may include... Figure 8 The training device for the arrival time estimation model in the described embodiments, or Figure 9 The terminal device in the described embodiments, or Figure 10 The server described.

[0193] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0195] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0197] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a training device for a time-of-arrival prediction model, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0198] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A training method for an arrival time prediction model, characterized in that, include: Obtain historical trajectory data; Feature extraction is performed on the historical trajectory data to obtain a feature set, which includes static features, historical features, and real-time features corresponding to the historical trajectory data; The feature set is matched with the corresponding arrival time information to obtain training samples; The arrival time prediction model is trained based on the training samples. The time prediction model includes a real-time scene prediction branch, a future scene prediction branch, and a marked-day scene prediction branch. The real-time scene prediction branch is trained based on the static features, the historical features, and the real-time features, using the arrival time information as the training target. The future scene prediction branch is trained based on the static features and the historical features. The marked-day scene prediction branch is trained based on the static features and the real-time features. During training, the future scene prediction branch and the marked-day scene prediction branch approximate the real-time scene prediction branch. During training, training metrics are obtained. Based on the training metrics, the output of the future scene prediction branch is approximated to the arrival time information and the output of the real-time scene prediction branch to obtain first and second loss information. Based on the training metrics, the output of the marked-day scene prediction branch is approximated to the arrival time information and the output of the real-time scene prediction branch to obtain third and fourth loss information. Based on the training metrics, the output of the real-time scene prediction branch is approximated to the arrival time information to obtain fifth loss information.

2. The method according to claim 1, characterized in that, The acquisition of historical trajectory data includes: Acquire the collection trajectory; Extract the location information sequence corresponding to the acquisition trajectory, and determine the time information corresponding to each location point in the location information sequence; The location information sequence is matched with the time information to obtain the historical trajectory data.

3. The method according to claim 1, characterized in that, The step of extracting features from the historical trajectory data to obtain a feature set includes: Determine the road segment sequence in the historical trajectory data; Feature extraction is performed on each path length and path type in the road segment sequence to obtain the static features; The historical features are obtained by extracting the speed information of each path in the road segment sequence at different time dimensions. Feature extraction is performed on the real-time speed information and congestion information corresponding to each path in the road segment sequence to obtain the real-time features; The static features, historical features, and real-time features are collected to obtain the feature set.

4. The method according to claim 1, characterized in that, The training of the arrival time prediction model based on the training samples includes: The static features and historical features in the training samples are input into the fully connected layer in the future scene prediction branch for feature extraction, so as to train the future scene prediction branch based on the arrival time information. The fully connected layer in the future scene prediction branch only accepts the feature output of the future scene prediction branch. The static features and real-time features in the training samples are input into the fully connected layer in the marked day scene prediction branch for feature extraction, so as to train the marked day scene prediction branch based on the arrival time information. The fully connected layer in the marked day scene prediction branch only accepts the feature output of the marked day scene prediction branch. The static features, historical features, and real-time features in the training samples are input into the fully connected layer in the real-time scene prediction branch for feature extraction, so as to train the real-time scene prediction branch based on the arrival time information. The fully connected layer in the real-time scene prediction branch receives the feature outputs of the future scene prediction branch, the feature outputs of the marked day scene prediction branch, and the feature outputs of the real-time scene prediction branch.

5. The method according to claim 4, characterized in that, The method further includes: Obtain preset weighted information; The first loss information, the second loss information, the third loss information, the fourth loss information, and the fifth loss information are weighted and calculated based on the preset weighted information to determine the target loss; The arrival time prediction model is trained based on the target loss.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Obtain the navigation request input from the target object; Determine the path information and time information corresponding to the navigation request; The path information is input into the arrival time prediction model to obtain the prediction information, which includes the real-time scene arrival time, the future scene arrival time, and the marked day scene arrival time. Based on the time information, the matching items in the estimated information are determined, and the corresponding interface is displayed.

7. A training device for an arrival time prediction model, characterized in that, include: The acquisition unit is used to acquire historical trajectory data; An extraction unit is used to extract features from the historical trajectory data to obtain a feature set, the feature set including static features, historical features and real-time features corresponding to the historical trajectory data; The extraction unit is also used to match the feature set with the corresponding arrival time information to obtain training samples; A training unit is used to train an arrival time prediction model based on the training samples. The time prediction model includes a real-time scene prediction branch, a future scene prediction branch, and a marked-day scene prediction branch. The real-time scene prediction branch is trained based on the static features, the historical features, and the real-time features, using the arrival time information as the training target. The future scene prediction branch is trained based on the static features and the historical features. The marked-day scene prediction branch is trained based on the static features and the real-time features. During training, the future scene prediction branch and the marked-day scene prediction branch approximate the real-time scene prediction branch. During training, training metrics are obtained. Based on the training metrics, the output of the future scene prediction branch is approximated to the arrival time information and the output of the real-time scene prediction branch to obtain first and second loss information. Based on the training metrics, the output of the marked-day scene prediction branch is approximated to the arrival time information and the output of the real-time scene prediction branch to obtain third and fourth loss information. Based on the training metrics, the output of the real-time scene prediction branch is approximated to the arrival time information to obtain fifth loss information.

8. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store program code; the processor is used to execute the training method of the arrival time estimation model according to any one of claims 1 to 6 according to the instructions in the program code.

9. A computer program product comprising a computer program / instructions stored in a computer-readable storage medium, characterized in that, When the computer program / instructions in the computer-readable storage medium are executed by a processor, they implement the steps of the training method for the arrival time estimation model according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the training method for the time-of-arrival prediction model according to any one of claims 1 to 6.

Images