Arrival time estimation method and device, storage medium, electronic equipment and product

Abstract

The application discloses a kind of arrival time estimation method, device, storage medium, electronic equipment and product, involve artificial intelligence technical field, the present application can be applied to blockchain, cloud technology, map and Internet of Vehicles etc. field, the method comprises: obtaining estimated arrival time, estimated arrival time is obtained using preset estimation model estimation;Obtain residual network parameters, residual network parameters are obtained by periodically training residual calculation network according to the travel data of user via candidate route;For scheduled departure time feature extraction processing, obtain scheduled departure time feature;Using residual calculation network, residual network parameters and scheduled departure time feature are based on residual calculation network parameter and scheduled departure time feature and carry out residual calculation processing, obtain arrival time residual;According to estimated arrival time and arrival time residual, obtain route arrival time.The present application can effectively improve the estimation effect of arrival time without adjusting preset estimation model, improve user experience.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a method, apparatus, storage medium, electronic device, and product for estimating arrival time. Background Technology

[0002] Estimated Time of Arrival (ETA) is the process of estimating the time required to complete a given route and departure time using a pre-defined estimation model, thus obtaining the arrival time.

[0003] Currently, the pre-defined models for time of arrival (TOA) estimation typically include rule-based, tree-based, and neural network-based models. Rule-based and tree-based models struggle with real-time updates, with the fastest updates occurring only daily. This delay in updates leads to poor TOA prediction accuracy. Neural network-based models can be updated in real-time, but this usually involves directly updating the parameters of the existing neural network. A drawback of this approach is the catastrophic forgetting effect; after training the model with the latest data, previously learned patterns are difficult to retain, further degrading the TOA prediction accuracy.

[0004] Therefore, the current arrival time prediction is ineffective, resulting in a poor user experience. Summary of the Invention

[0005] This application provides a method and related apparatus for estimating arrival time, which can effectively improve the estimation effect of arrival time and enhance the user experience without adjusting the preset estimation model.

[0006] To address the aforementioned technical problems, this application provides the following technical solutions:

[0007] According to one embodiment of this application, an arrival time estimation method is provided. The method includes: obtaining the estimated arrival time of a candidate route at a predetermined departure time, wherein the estimated arrival time is estimated using a preset estimation model; obtaining residual network parameters corresponding to the candidate route, wherein the residual network parameters are obtained by periodically training a residual calculation network based on the driving data of users traveling along the candidate route; performing feature extraction processing on the predetermined departure time to obtain predetermined departure time features; using the residual calculation network, performing residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain arrival time residuals; and obtaining the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residuals.

[0008] According to one embodiment of this application, an arrival time estimation device includes: a time acquisition module, configured to acquire the estimated arrival time of a candidate route at a predetermined departure time, wherein the estimated arrival time is estimated using a preset estimation model; a parameter acquisition module, configured to acquire residual network parameters corresponding to the candidate route, wherein the residual network parameters are obtained by periodically training a residual calculation network based on the driving data of users traveling along the candidate route; a time characterization module, configured to perform feature extraction processing on the predetermined departure time to obtain predetermined departure time features; a calculation processing module, configured to use the residual calculation network to perform residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain an arrival time residual; and an arrival time module, configured to obtain the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residual.

[0009] In some embodiments of this application, the apparatus further includes a parameter update module, configured to: periodically acquire driving data of users traveling via the candidate routes, the driving data including the current departure time and the road segment sequence corresponding to the candidate routes; extract current departure time features based on the current departure time, and extract initial road segment sequence features based on the road segment sequence; use the residual calculation network to perform residual calculation processing based on the current departure time features and the initial road segment sequence features to obtain a predicted residual; adjust the initial road segment sequence features according to the predicted residual until a predetermined condition is met to obtain a target road segment sequence feature, wherein the residual network parameters include the target road segment sequence features.

[0010] In some embodiments of this application, the driving data further includes a predicted arrival time and an actual arrival time; a parameter update module is used to: obtain a corrected arrival time based on the predicted residual and the predicted arrival time; adjust the initial road segment sequence features based on the error between the corrected arrival time and the actual arrival time until the error meets a predetermined error condition, thereby obtaining the target road segment sequence features.

[0011] In some embodiments of this application, the time characterization module is used to: calculate and process the predetermined departure time with time points within the predetermined time period to obtain calculated values ​​corresponding to the time points within the predetermined time period; and obtain the predetermined departure time feature corresponding to the predetermined departure time based on the calculated values ​​corresponding to the time points within the predetermined time period.

[0012] In some embodiments of this application, the time characterization module is used to implement one of the following methods: calculating the time difference between the predetermined departure time and a time point within the predetermined time period, and removing a predetermined time granularity from the time difference to obtain the calculated value corresponding to the time point within the predetermined time period; distributing the time points within the predetermined time period around a unit circle to obtain a time point circle, and calculating the coordinate value corresponding to the predetermined departure time on the time point circle to obtain the calculated value corresponding to the time point within the predetermined time period.

[0013] In some embodiments of this application, the calculation processing module is used to: fuse the residual network parameters with the predetermined departure time features to obtain residual features; and obtain the arrival time residual based on the residual features.

[0014] In some embodiments of this application, the residual network parameters include target road segment sequence features, which include road segment features corresponding to each road segment in the candidate route; the calculation processing module is used to: perform an inner product operation on the road segment features corresponding to each road segment and the predetermined departure time features to obtain the residual features corresponding to each road segment; process the residual features corresponding to each road segment into constant positive values, and average the constant positive values ​​corresponding to each road segment to obtain the arrival time residual.

[0015] In some embodiments of this application, the arrival time module is used to implement one of the following methods: summing the estimated arrival time and the arrival time residual to obtain the route arrival time corresponding to the candidate route; and weighting the estimated arrival time and the arrival time residual based on a predetermined weight to obtain the route arrival time corresponding to the candidate route.

[0016] According to another embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a computer's processor, causes the computer to perform the methods described in the embodiments of this application.

[0017] According to another embodiment of this application, an electronic device includes: a memory storing a computer program; and a processor reading the computer program stored in the memory to execute the methods described in the embodiments of this application.

[0018] According to another embodiment of this application, a computer program product or computer program includes 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 in the embodiments of this application.

[0019] In this embodiment, the estimated arrival time of a candidate route at a predetermined departure time is obtained, wherein the estimated arrival time is estimated using a preset estimation model; the residual network parameters corresponding to the candidate route are obtained, wherein the residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route; feature extraction processing is performed on the predetermined departure time to obtain predetermined departure time features; the residual calculation network is used to perform residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain arrival time residuals; and the route arrival time corresponding to the candidate route is obtained based on the estimated arrival time and the arrival time residuals.

[0020] In this way, for the estimated arrival time obtained by the preset prediction model, the arrival time residual is calculated by the residual calculation network based on the residual network parameters and the characteristics of the predetermined departure time. Based on the estimated arrival time and the arrival time residual, the calibrated route arrival time is obtained. This route arrival time is more accurate than the initial estimated arrival time. It can avoid the need to update the preset prediction model in real time and avoid the catastrophic forgetting caused by updating the preset prediction model. Therefore, the prediction effect of arrival time is effectively improved without adjusting the preset prediction model, thus improving the user experience. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A schematic diagram of a system to which embodiments of this application can be applied is shown.

[0023] Figure 2 A flowchart illustrating an arrival time estimation method according to an embodiment of this application is shown.

[0024] Figure 3 A framework diagram of an arrival time estimation service framework in one scenario according to this application is shown.

[0025] Figure 4 A schematic diagram of a parameter update process according to an example of this application is shown.

[0026] Figure 5 A block diagram of an arrival time estimation device according to an embodiment of this application is shown.

[0027] Figure 6A block diagram of an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] It is understood that in the specific implementation of this application, data related to candidate routes, scheduled departure times, etc. are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0030] Figure 1 A schematic diagram of a system 100 to which embodiments of this application can be applied is shown. For example... Figure 1 As shown, system 100 may include server 101 and terminal 102.

[0031] Server 101 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0032] Terminal 102 can be any device, including but not limited to mobile phones, computers, smart voice interaction devices, smart home appliances, in-vehicle terminals, VR / AR devices, smartwatches, and computers, etc. In one embodiment, server 101 or terminal 102 can be a node device in a blockchain network, map, or vehicle networking platform.

[0033] In one embodiment of this example, server 101 or terminal 102 may: obtain the estimated arrival time corresponding to a candidate route at a predetermined departure time, wherein the estimated arrival time is estimated using a preset estimation model; obtain the residual network parameters corresponding to the candidate route, wherein the residual network parameters are obtained by periodically training a residual calculation network based on the driving data of users traveling along the candidate route; perform feature extraction processing on the predetermined departure time to obtain predetermined departure time features; use the residual calculation network to perform residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain arrival time residuals; and obtain the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residuals.

[0034] The arrival time estimation scheme implemented in this application can be applied to scenarios including but not limited to:

[0035] (1) When a user initiates navigation, the backend first provides several candidate routes, then uses the solution of this application to calculate the arrival time of each candidate route, and then selects the fastest candidate route from the several candidate routes and provides it to the user.

[0036] (2) After entering the navigation state, at regular intervals, the route arrival time of the remaining distance (i.e., the remaining route is used as the real-time candidate route) is calculated using the scheme of this application, so as to facilitate the user to arrange the trip.

[0037] (3) The solution of this application can be used to calculate an isochronous reachable circle (i.e., a reachable circle formed by candidate routes with the same arrival time), such as a half-hour reachable circle, a one-hour reachable circle, etc., to help users understand the living radius of a certain location;

[0038] (4) When delivering food, the solution of this application can be used to accurately calculate the arrival time of each candidate route, so as to better assign orders to deliverymen and improve delivery efficiency.

[0039] (5) In the context of mobile phone ride-hailing, the solution of this application can be used to accurately calculate the arrival time of each candidate route, thereby better arranging drivers to accept orders and improving passenger transport efficiency.

[0040] (6) The solution of this application can be used to provide the route arrival time of each candidate route for upstream services to evaluate the merits of each candidate route and then push the optimal route to the user.

[0041] (7) It can provide the influence weight of each segment of the candidate route on the route arrival time for upstream services, such as to avoid congestion and to interpret the estimated arrival time.

[0042] Figure 2A flowchart illustrating an arrival time estimation method according to an embodiment of this application is shown schematically. The entity performing this arrival time estimation method can be any device, such as… Figure 1 The server 101 or terminal 102 shown.

[0043] like Figure 2 As shown, the method for estimating arrival time may include steps S210 to S250.

[0044] Step S210: Obtain the estimated arrival time of the candidate route at the predetermined departure time. The estimated arrival time is obtained by using a preset prediction model. Step S220: Obtain the residual network parameters corresponding to the candidate route. The residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route. Step S230: Perform feature extraction processing on the predetermined departure time to obtain the predetermined departure time features. Step S240: Use the residual calculation network to perform residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain the arrival time residual. Step S250: Obtain the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residual.

[0045] Candidate routes can be routes between a starting point and an end point. In map applications, a route can be a complete line connecting the starting point and the end point. In real-world scenarios, the length of a route can typically range from one kilometer to tens of kilometers. Each candidate route can consist of at least one link. In map applications, routes are represented by a sequence of links. In map data, routes are divided into segments, the length of which can range from tens of meters to several kilometers. Each segment is called a road segment and can be assigned a globally unique identifier (id). Therefore, a route on a map is composed of all the road segments within that route. The scheduled departure time can be a user-specified time or real-time (e.g., the user's real-time time during navigation).

[0046] The preset prediction model can be rule-based, tree-based, or neural network-based. Before step S210, the preset prediction model can analyze and process the route features in the road segment sequence corresponding to the candidate route to obtain the time required to complete the candidate route. Then, the sum of the time taken and the scheduled departure time is the estimated arrival time. The rule-based model can estimate the passage time of each road segment based on its length, speed, traffic lights, and other route features in the road segment sequence using prediction rules. This time is then added to the passage time at each intersection to form the total time taken for the entire candidate route. In the tree-based model, the route features of the entire road segment sequence can be extracted first. These features include the total distance, the average speed at the scheduled departure time, the total number of traffic lights, and the percentage of congested mileage. These features are then input into the tree-based model to obtain the time taken for the entire candidate route output by the model. The tree-based model can be, for example, a model based on Gradient Boosting Decision Tree (GBDT). In a neural network-based model approach, route features can be extracted from the entire road segment sequence or from each segment within the sequence. These features are then input into the neural network-based model to obtain the time consumption of the entire candidate route output by the model. The neural network structure can be a convolutional neural network (CNN), a recurrent neural network (RNN), or similar structures.

[0047] After estimating the estimated arrival time of the candidate route at the predetermined departure time using a preset prediction model, the estimated arrival time of the candidate route at the predetermined departure time can be obtained in step S210. Further, the residual network parameters corresponding to the candidate route can be obtained. These residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route. The period can be set according to actual conditions; for example, the period could be once a week, once a day, once a minute, etc. The residual network parameters obtained from the periodic training can be stored at a predetermined location, and the residual network parameters can be obtained from this predetermined location.

[0048] Feature extraction of the scheduled departure time yields the corresponding scheduled departure time features, which can be feature vectors in vector form. A residual computation network is then used to perform residual calculations based on the residual network parameters and the scheduled departure time features, resulting in arrival time residuals. Furthermore, based on the estimated arrival time and the arrival time residuals, the calibrated arrival times for the candidate routes are obtained.

[0049] In this way, based on steps S210 to S250, for the estimated arrival time obtained by the preset prediction model, the arrival time residual is calculated by the residual calculation network based on the residual network parameters and the characteristics of the predetermined departure time. Based on the estimated arrival time and the arrival time residual, the calibrated route arrival time is obtained. This route arrival time is more accurate than the initial estimated arrival time. It is not necessary to update the preset prediction model in real time and avoids the catastrophic forgetting of the model caused by updating the preset prediction model. Therefore, the prediction effect of arrival time is effectively improved without adjusting the preset prediction model, thus improving the user experience.

[0050] For example, the preset arrival time prediction model can be a model deployed online on the navigation platform for estimating arrival times. Based on the solution proposed in this proposal, the estimated arrival time of the preset model can be corrected using the arrival time residual to obtain an accurate route arrival time. This eliminates the need for frequent real-time adjustments to the online preset model, achieving accurate route arrival times. Furthermore, it avoids the catastrophic forgetting of the model due to updates to the preset model. Based on the solution proposed in this proposal, the accuracy of arrival time prediction is effectively improved without adjusting the online preset model deployed on the navigation platform, thus enhancing the user experience of the navigation platform.

[0051] The following description Figure 2 Further specific alternative embodiments of the steps performed in the example of estimating arrival time.

[0052] In one embodiment, before obtaining the residual network parameters corresponding to the candidate route in step S220, the method may further include:

[0053] Periodically acquire driving data of users passing through candidate routes. The driving data includes the departure time and the road segment sequence corresponding to the candidate route. Extract the departure time feature based on the departure time and the initial road segment sequence feature based on the road segment sequence. Use a residual calculation network to perform residual calculation processing based on the departure time feature and the initial road segment sequence feature to obtain the prediction residual. Adjust the initial road segment sequence feature according to the prediction residual until it meets the predetermined conditions to obtain the target road segment sequence feature. The residual network parameters include the target road segment sequence feature.

[0054] Regularly acquiring user driving data via candidate routes can be achieved by periodically collecting trajectory data from relevant platforms (such as navigation platforms). This trajectory data can be a sequence of trajectory points (e.g., GPS point data corresponding to GPS points), each of which can include latitude, longitude, timestamp, speed, and direction angle. From this trajectory data, the user's departure time (i.e., the current departure time), the time taken to complete the entire candidate route (i.e., ATA, actual arrival time), and the predicted arrival time can be extracted. Furthermore, trajectory matching algorithms (such as Hidden Markov Models, Recurrent Neural Networks, etc.) can be used to match each trajectory point to a corresponding road segment, obtaining a road segment identifier sequence (link ID sequence). This sequence represents the road segments (links) traversed by the user, and can include a subsequence of trajectory point data corresponding to each road segment. Finally, the current departure time can be the time the user departs from the starting point of the candidate route. The sequence of road segments corresponding to the candidate route can be a sequence of trajectory point data of users in each road segment of the candidate route. The trajectory point data can include latitude and longitude, timestamp, speed, direction angle and other data.

[0055] Based on the departure time, the corresponding departure time feature can be extracted, which can be a feature vector in vector form. Based on the road segment sequence, the corresponding road segment sequence features can be extracted to obtain the initial road segment sequence features. The initial road segment sequence features can include the road segment features corresponding to each road segment in the candidate route, and the initial road segment sequence features can also be a feature vector in vector form. Specifically, common feature extraction (embedding) methods in neural networks can be used to extract the initial road segment sequence features, for example, convolutional networks or other feature extraction networks can be used.

[0056] A residual calculation network is employed, which calculates residuals based on the departure time characteristics and initial road segment sequence characteristics to obtain predicted residuals. The initial road segment sequence characteristics are then adjusted based on the predicted residuals until predetermined conditions are met (the predetermined conditions can be set according to actual conditions), resulting in adjusted target road segment sequence characteristics. The residual network parameters specifically refer to the target road segment sequence characteristics, which may include the road segment characteristics corresponding to each segment in the candidate route. In this implementation, the residual calculation network can perform residual calculations based on the target road segment sequence characteristics and predetermined departure time characteristics to obtain arrival time residuals.

[0057] In one embodiment, the travel data further includes the predicted arrival time and the actual arrival time; adjusting the initial road segment sequence features based on the prediction residual until predetermined conditions are met to obtain the target road segment sequence features may include:

[0058] Based on the predicted residual and the predicted arrival time, the corrected arrival time is obtained; based on the error between the corrected arrival time and the actual arrival time, the initial road segment sequence characteristics are adjusted until the error meets the predetermined error condition, and the target road segment sequence characteristics are obtained.

[0059] The predicted arrival time can be obtained from the preset prediction model in the relevant platform at the time the user travels along the candidate route. The actual arrival time can be the actual time the user takes to complete the candidate route. The corrected arrival time can be obtained by directly summing the prediction residual and the predicted arrival time, or by weighting the prediction residual and the predicted arrival time according to predetermined weights.

[0060] Based on the error between the corrected arrival time and the actual arrival time, the initial road segment sequence features are updated and adjusted. Specifically, the initial road segment sequence features can be updated using stochastic gradient descent based on a single sample (i.e., a single user) or a small batch of samples (i.e., a small batch of users). The residual calculation network can then calculate the predicted residual based on the adjusted initial road segment sequence features. Based on this predicted residual and the predicted arrival time, the corrected arrival time can be obtained, until the error between the corrected arrival time and the actual arrival time meets a predetermined error condition. The adjusted initial road segment sequence features are then the target road segment sequence features. The error can be calculated using methods such as MAPE and MSE, where MAPE = |ETA - ATA| / ATA, and MSE = (ETA - ATA). 2 ETA can represent the corrected arrival time, and ATA can represent the actual arrival time. A predetermined error condition can be that the error is less than a predetermined threshold.

[0061] In one embodiment, extracting the departure time feature based on the departure time may include: performing calculations on the departure time and time points within a predetermined time period to obtain calculated values ​​corresponding to the time points within the predetermined time period; and obtaining the departure time feature corresponding to the departure time based on the calculated values ​​corresponding to the time points within the predetermined time period.

[0062] The predetermined time period can be a pre-specified time period, such as one day, one week, or two weeks. In one embodiment of this example, the predetermined time period is one week, meaning it starts at 0:00 on Monday and ends at 24:00 on Sunday. The number of time points within the predetermined time period can be set according to actual circumstances; for example, the time points within the predetermined time period may include 0:00 on Monday and several time points after 0:00 on Monday.

[0063] The departure time can be calculated and processed against various time points within a predetermined time period to obtain the calculated values ​​corresponding to each time point within the predetermined time period. In one example, this calculation can be performed by multiplying the departure time by the values ​​for each time point within the predetermined time period. Based on the calculated values ​​for each time point within the predetermined time period, these values ​​can be concatenated to form a departure time feature corresponding to the current departure time. This departure time feature can then be a feature vector in vector form.

[0064] In one embodiment, the method of calculating the departure time and time points within a predetermined time period to obtain the calculated values ​​corresponding to the time points within the predetermined time period includes one of the following methods:

[0065] The first method calculates the time difference between the departure time and a time point within the predetermined time period, and then removes the predetermined time granularity from the time difference to obtain the calculated value corresponding to the time point within the predetermined time period. The second method distributes the time points within the predetermined time period around a unit circle to obtain a time point circle, and then calculates the coordinate value corresponding to the departure time on the time point circle to obtain the calculated value corresponding to the time point within the predetermined time period.

[0066] In the first method: If the predetermined time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and a time point within the predetermined time period is, for example, 0:00 on Monday, and the departure time is, for example, 8:00 AM on Tuesday, calculate the time difference between the departure time and the time point within the predetermined time period, in minutes (this can be understood as seconds or hours depending on the actual situation). The time difference is (24+8)*60 = 1920 minutes. Further, using 10-minute time granularities (this can be understood as setting the predetermined time granularity according to the actual situation), divide the time difference by the predetermined time granularity, and the calculated value corresponding to the time point "0:00 on Monday" within the predetermined time period is 1920 / 10 = 192.

[0067] In the second method, if the predetermined time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and further, if the smallest unit is the minute (which can be understood as 5 minutes, 10 minutes, etc. in actual use), then one week contains 24 * 7 * 60 = 10080 minutes. Assume these 10080 time points are evenly and orderly distributed around a unit circle, such that the last minute is adjacent to the first minute, resulting in a circle of time points around the unit circle (k = 1). This can be calculated using the formula... and Calculate the coordinates of the departure time t on the time circle. For example, if the departure time t was 8:00 AM on Tuesday, which is 1920 minutes different from 0:00 AM on Monday, then the coordinates of 8:00 AM on Tuesday (i.e., the two calculated values) would be: and In the above process, it is assumed that the number of minutes in a week revolves around the unit circle (k=1). Furthermore, we can let these 10080 minutes revolve around the unit circle any number of times, then according to the formula... and Calculate the coordinates of the departure time t on the time circle. For example, taking k=2, the coordinates of 8:00 AM on Tuesday (i.e., the two calculated values) are: and In the second approach, for a given departure time, its departure time feature can be expressed as an even number of features (i.e., calculated values), requiring only that k be chosen with more values. For cases where the departure time feature needs to be an odd number of features, the last feature can be set to 0, and the other features can be processed as described above. Therefore, the above processing can transform a moment into a departure time feature composed of any number of features. The applicant found that the second approach, compared to the first, can further improve the overall accuracy of arrival time prediction.

[0068] In one embodiment, a residual calculation network is used to perform residual calculation processing based on the departure time characteristics and initial road segment sequence characteristics to obtain the predicted residual, which may include:

[0069] The initial road segment sequence features are fused with the departure time features to obtain residual features; the prediction residual is then obtained based on these residual features. The initial road segment sequence features and the departure time features can be fused using methods such as inner product operations or concatenation to obtain residual features; furthermore, the residual features can be averaged to obtain the prediction residual, or the residual features can be directly used as the prediction residual.

[0070] In one embodiment, the initial road segment sequence features include the road segment features corresponding to each road segment in the candidate route; fusing the initial road segment sequence features with the departure time features to obtain residual features includes: performing an inner product operation on the road segment features corresponding to each road segment in the initial road segment sequence features and the departure time features to obtain the residual features corresponding to each road segment; obtaining the prediction residual based on the residual features includes: processing the residual features corresponding to each road segment into constant positive values, and averaging the constant positive values ​​corresponding to each road segment to obtain the prediction residual.

[0071] The initial road segment sequence features include the road segment features corresponding to each road segment in the candidate route. For example, if the candidate route includes road segments 1 to N, then the initial road segment sequence features can include road segment feature 1 corresponding to road segment 1 to road segment feature N corresponding to road segment N. The road segment features can be M-dimensional feature vectors, and the departure time features can also be M-dimensional feature vectors. By performing an inner product operation between the road segment features corresponding to each road segment and the departure time features, the residual features corresponding to each road segment can be obtained. For example, the inner product of road segment feature 1 and the departure time features yields residual feature 1, and the inner product of road segment feature N and the departure time features yields residual feature N. Furthermore, the residual features corresponding to each road segment can be processed into constant positive values ​​using functions such as the softplus function, and then the constant positive values ​​corresponding to each road segment are averaged to obtain the overall average residual, i.e., the prediction residual.

[0072] In one embodiment, step S230 involves performing feature extraction processing on the predetermined departure time to obtain predetermined departure time features, including:

[0073] The scheduled departure time is calculated and processed with the time points within the scheduled time period to obtain the calculated values ​​corresponding to the time points within the scheduled time period; based on the calculated values ​​corresponding to the time points within the scheduled time period, the scheduled departure time characteristics corresponding to the scheduled departure time are obtained.

[0074] The predetermined time period can be a pre-specified time period, such as one day, one week, or two weeks. In one embodiment of this example, the predetermined time period is one week, meaning it starts at 0:00 on Monday and ends at 24:00 on Sunday. The number of time points within the predetermined time period can be set according to actual circumstances; for example, the time points within the predetermined time period may include 0:00 on Monday and several time points after 0:00 on Monday.

[0075] The scheduled departure time can be calculated and processed against each time point within a specified time period to obtain the calculated values ​​corresponding to each time point within the specified time period. In one example, this calculation can be performed by multiplying the scheduled departure time by each time point within the specified time period. Based on the calculated values ​​corresponding to each time point within the specified time period, these values ​​can be concatenated to form a scheduled departure time feature, which can then be a feature vector in vector form.

[0076] In one embodiment, the method of calculating the time points corresponding to the predetermined departure time and time points within a predetermined time period to obtain the calculated values ​​within the predetermined time period includes one of the following methods:

[0077] The first method calculates the time difference between the scheduled departure time and the time point within the scheduled time period, and then removes the time difference from the scheduled time granularity to obtain the calculated value corresponding to the time point within the scheduled time period. The second method distributes the time points within the scheduled time period around a unit circle to obtain a time point circle, and then calculates the coordinate value corresponding to the scheduled departure time on the time point circle to obtain the calculated value corresponding to the time point within the scheduled time period.

[0078] In the first method: If the scheduled time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and a scheduled departure time is, for example, 0:00 on Monday, then the time difference between the scheduled departure time and a specific point in the scheduled time period is calculated in minutes (this can be understood as seconds or hours depending on the actual situation). The time difference is (24+8)*60 = 1920 minutes. Further, using a 10-minute time granularity (this can be understood as setting the time granularity according to the actual situation), the time difference is divided by the time granularity, resulting in a calculated value of 1920 / 10 = 192 for the time point "0:00 on Monday" within the scheduled time period.

[0079] In the second method, if the predetermined time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and further, if the smallest unit is the minute (which can be understood as 5 minutes, 10 minutes, etc. in actual use), then one week contains 24 * 7 * 60 = 10080 minutes. Assume these 10080 time points are evenly and orderly distributed around a unit circle, such that the last minute is adjacent to the first minute, resulting in a circle of time points around the unit circle (k = 1). This can be calculated using the formula... and Calculate the coordinates of the scheduled departure time t on the time circle. For example, if the scheduled departure time t is 8:00 AM on Tuesday, which is 1920 minutes different from 0:00 AM on Monday, then the coordinates of 8:00 AM on Tuesday (i.e., the two calculated values) are: and In the above process, it is assumed that the number of minutes in a week revolves around the unit circle (k=1). Furthermore, we can let these 10080 minutes revolve around the unit circle any number of times, then according to the formula... and Calculate the coordinates of the scheduled departure time t on the time circle. For example, taking k=2, the coordinates of Tuesday morning 8:00 (i.e., the two calculated values) are: and In the second approach, for a given departure time, its characteristic can be expressed as an even number of features (i.e., calculated values), requiring only that k be chosen with more values. For cases where the desired departure time characteristic needs to be an odd number of features, the last feature can be set to 0, and the other features can be processed as described above. Therefore, the above processing can transform a moment into a desired departure time characteristic composed of any number of features. The applicant found that the second approach, compared to the first, can further improve the overall accuracy of arrival time prediction.

[0080] In one embodiment, step S240, which involves performing residual calculation processing based on the residual network parameters and the predetermined departure time characteristics to obtain the arrival time residual, includes: fusing the residual network parameters and the predetermined departure time characteristics to obtain residual characteristics; and obtaining the arrival time residual based on the residual characteristics. The residual network parameters and the predetermined departure time characteristics can be fused using methods such as inner product operations or concatenation to obtain residual characteristics; furthermore, the arrival time residual can be obtained by averaging the residual characteristics or by directly using the residual characteristics as the arrival time residual.

[0081] In one embodiment, the residual network parameters include target road segment sequence features, which include road segment features corresponding to each road segment in the candidate route; the residual network parameters are fused with the predetermined departure time features to obtain residual features, including: performing an inner product operation between the road segment features corresponding to each road segment and the predetermined departure time features to obtain the residual features corresponding to each road segment; the arrival time residual is obtained based on the residual features, including: processing the residual features corresponding to each road segment into constant positive values, and averaging the constant positive values ​​corresponding to each road segment to obtain the arrival time residual.

[0082] The target road segment sequence features include the road segment features corresponding to each road segment in the candidate route. For example, if the candidate route includes road segments 1 to N, then the target road segment sequence features can include road segment feature 1 corresponding to road segment 1 to road segment feature N corresponding to road segment N. The road segment features can be M-dimensional feature vectors, and the predetermined departure time features can also be M-dimensional feature vectors. By performing an inner product operation between the road segment features corresponding to each road segment in the target road segment sequence features and the predetermined departure time features, the residual features corresponding to each road segment can be obtained. For example, the inner product of road segment feature 1 and the predetermined departure time features yields residual feature 1, and the inner product of road segment feature N and the predetermined departure time features yields residual feature N. Furthermore, the residual features corresponding to each road segment can be processed into constant positive values ​​using functions such as the softplus function, and the average of the constant positive values ​​corresponding to each road segment is calculated to obtain the overall average residual, i.e., the arrival time residual.

[0083] In one embodiment, step S250, obtaining the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residual, includes one of the following methods: summing the estimated arrival time and the arrival time residual to obtain the route arrival time corresponding to the candidate route; or weighting the estimated arrival time and the arrival time residual based on a predetermined weight to obtain the route arrival time corresponding to the candidate route.

[0084] In one approach, the estimated arrival time and the arrival time residual are directly summed to obtain the arrival time of the candidate route. In another approach, the predetermined weights may include weights corresponding to the estimated arrival time and the arrival time residual, respectively. The estimated arrival time and the arrival time residual are then weighted and summed based on the predetermined weights to obtain the arrival time of the candidate route. The extent to which the residuals take effect can be controlled by configuring the predetermined weights.

[0085] The above embodiments are further described below with reference to a scenario in which the arrival time is estimated by applying the above embodiments of this application. Figure 3 This diagram illustrates the service framework for estimating arrival time in this scenario. Figure 4 The flowchart illustrates the residual network parameters during the training of the residual computation network in this scenario.

[0086] See Figure 3 In this scenario, the service framework for estimating arrival time can include four main modules: the original ETA service (i.e., online service) 310, the residual calculation service 320, the real-time trajectory feedback and processing service 330, and the parameter update service 340. It can also include the final arrival time calculation service (i.e., the final ETA) 350.

[0087] In this scenario, the arrival time estimation process may include steps (1) to (6).

[0088] (1) The original ETA service (i.e., online service) can use a preset prediction model to predict the estimated arrival time of the candidate route at the predetermined departure time (e.g., Figure 3 The original ETA is shown.

[0089] Specifically, the pre-defined prediction model can be rule-based, tree-based, or neural network-based. The pre-defined prediction model can analyze and process the route features in the segment sequence corresponding to the candidate route at the predetermined departure time to obtain the time required to complete the candidate route. Then, the sum of the time taken and the predetermined departure time is the estimated arrival time. Rule-based models can estimate the transit time of each segment based on its length, speed, traffic lights, and other route features in the segment sequence using prediction rules. This time is then added to the transit time at each intersection to form the total transit time for the entire candidate route. In tree-based models, route features such as the total distance, the average speed at the predetermined departure time, the total number of traffic lights, and the percentage of congested mileage are extracted first. These features are then input into the tree-based model to obtain the total transit time for the entire candidate route. Tree-based models can be, for example, models based on Gradient Boosting Decision Tree (GBDT). In a neural network-based model approach, route features can be extracted from the entire route sequence or from each segment within the sequence. These features are then input into the neural network-based model to obtain the time consumption of the entire candidate route output by the model. The neural network structure can be a convolutional neural network (CNN), a recurrent neural network (RNN), or similar structures.

[0090] (2) The final arrival time calculation service (i.e., the final ETA) can obtain the estimated arrival time (i.e., the estimated arrival time) for candidate routes at the scheduled departure time. Figure 3 The original ETA shown is estimated using a preset estimation model.

[0091] (3) Real-time trajectory feedback and processing service can periodically obtain the driving data of users passing through candidate routes. The driving data may include the departure time, the road segment sequence corresponding to the candidate route, the predicted arrival time and the actual arrival time.

[0092] Specifically, trajectory data of users traversing candidate routes can be collected periodically from relevant platforms (such as navigation platforms). This trajectory data can be a sequence of trajectory point data (e.g., GPS point data corresponding to GPS points), each of which can include information such as latitude and longitude, timestamp, speed, and direction angle. From this trajectory data, the user's departure time (i.e., the current departure time), the time it takes for the user to complete the candidate route (i.e., ATA, actual arrival time), and the predicted arrival time can be extracted. Furthermore, trajectory matching algorithms (such as Hidden Markov Models, Recurrent Neural Networks, etc.) can be used to match each trajectory point data to a corresponding road segment, obtaining a road segment identifier sequence (link ID sequence), representing the sequence of road segments (links) traversed by the user. This road segment sequence can include a subsequence of trajectory point data corresponding to each road segment. Thus, the current departure time can be the time the user departs from the starting point of the candidate route. The road segment sequence corresponding to the candidate route can be a sequence formed by the user's trajectory point data in each road segment of the candidate route, and the trajectory point data can include latitude and longitude, timestamp, speed, and direction angle.

[0093] (4) The parameter update service can update the residual network parameters, which include the target road segment sequence features. Specifically, the parameter update service can: extract the departure time features based on the current departure time, and extract the initial road segment sequence features based on the road segment sequence; use a residual calculation network to perform residual calculation processing based on the current departure time features and the initial road segment sequence features to obtain the predicted residual; adjust the initial road segment sequence features according to the predicted residual until they meet the predetermined conditions to obtain the target road segment sequence features. Specifically, adjusting the initial road segment sequence features according to the predicted residual until they meet the predetermined conditions to obtain the target road segment sequence features can include: obtaining the corrected arrival time based on the predicted residual and the predicted arrival time; adjusting the initial road segment sequence features according to the error between the corrected arrival time and the actual arrival time until the error meets the predetermined error conditions to obtain the target road segment sequence features.

[0094] Based on the departure time, the corresponding departure time feature can be extracted, which can be a feature vector in vector form. Based on the road segment sequence, the corresponding road segment sequence features can be extracted to obtain the initial road segment sequence features. The initial road segment sequence features can include the road segment features corresponding to each road segment in the candidate route, and the initial road segment sequence features can also be a feature vector in vector form. Specifically, common feature extraction (embedding) methods in neural networks can be used to extract the initial road segment sequence features, for example, convolutional networks or other feature extraction networks can be used.

[0095] A residual computation network is employed to perform residual computation based on the departure time characteristics and initial road segment sequence characteristics to obtain the prediction residual. Specifically, this process includes: fusing the initial road segment sequence characteristics with the departure time characteristics to obtain residual features; and obtaining the prediction residual based on these residual features. The initial road segment sequence characteristics include the road segment features corresponding to each road segment in the candidate route. The fusion of the initial road segment sequence characteristics with the departure time characteristics to obtain residual features specifically includes: performing an inner product operation between the road segment features corresponding to each road segment in the initial road segment sequence characteristics and the departure time characteristics to obtain the residual features corresponding to each road segment; and obtaining the prediction residual based on these residual features includes: processing the residual features corresponding to each road segment into constant positive values ​​and averaging the constant positive values ​​corresponding to each road segment to obtain the prediction residual.

[0096] See Figure 4 The initial road segment sequence features include the road segment features corresponding to each road segment in the candidate route. For example, if the candidate route includes road segments 1 to N, then the initial road segment sequence features may include road segment feature 1 corresponding to road segment 1 (e.g., Figure 4 The road segment features N corresponding to link1 emb) to road segment N (e.g., Figure 4 (linkN emb). Each road segment feature can be an M-dimensional feature vector, and the departure time feature can also be an M-dimensional feature vector. By performing an inner product operation between the road segment features (including link1 emb to linkN emb) and the departure time feature, the residual features corresponding to each road segment can be obtained (e.g., Figure 4 Including residuals 1 to N), for example, taking the inner product of road segment feature 1 and the departure time feature at that time to obtain residual feature 1 (e.g. Figure 4 The residual feature N is obtained by taking the inner product of the road segment feature N and the departure time feature (e.g., residual 1). Figure 4 The residuals (N) are further processed by the softplus function to convert the residual features of each road segment into constant positive values. The constant positive values ​​of each road segment are then averaged to obtain the "average residual" or prediction residual for the entire route.

[0097] The predicted arrival time can be obtained from a pre-set prediction model on the relevant platform at the time the user traverses the candidate route. The actual arrival time can be the actual time it takes the user to complete the candidate route. See also Figure 4 The predicted residuals (i.e. Figure 4 (mean residual) and predicted arrival time (i.e. Figure 4 The corrected arrival time (ETA) can be obtained by directly summing the original ETA, or by weighting the predicted residuals and predicted arrival times based on predetermined weights. Figure 4 China-Singapore ETA).

[0098] Based on the corrected arrival time (i.e.) Figure 4 China-Singapore ETA) and actual arrival time (i.e. Figure 4 The initial road segment sequence features are updated and adjusted based on the error of the ATA (Advanced Time of Arrival). Specifically, the initial road segment sequence features can be updated using stochastic gradient descent based on a single sample (i.e., a single user) or a small batch of samples (i.e., a small batch of users). The residual calculation network can continue to calculate the predicted residual based on the adjusted initial road segment sequence features. Based on this predicted residual and the predicted arrival time, the corrected arrival time can be obtained until the error between the corrected arrival time and the actual arrival time meets a predetermined error condition. The adjusted initial road segment sequence features are then the target road segment sequence features. The error can be calculated using methods such as MAPE (Mean Error Probability) and MSE (Mean Error Probability), where MAPE = |ETA - ATA| / ATA, and MSE = (ETA - ATA). 2 ETA can represent the corrected arrival time, and ATA can represent the actual arrival time. A predetermined error condition can be that the error is less than a predetermined threshold.

[0099] Furthermore, extracting the departure time feature based on the departure time can specifically include: calculating and processing the departure time and time points within a predetermined time period to obtain the calculated values ​​corresponding to the time points within the predetermined time period; and obtaining the departure time feature corresponding to the departure time based on the calculated values ​​corresponding to the time points within the predetermined time period.

[0100] The predetermined time period can be a pre-specified time period, such as one day, one week, or two weeks. In one example, the predetermined time period is one week, meaning it starts at 0:00 on Monday and ends at 24:00 on Sunday. The number of time points within the predetermined time period can be set according to actual circumstances; for example, the time points within the predetermined time period can include 0:00 on Monday and several time points after 0:00 on Monday. The current departure time can be calculated and processed with each time point within the predetermined time period to obtain the calculated value corresponding to each time point within the predetermined time period. In one example, the current departure time can be processed by multiplying the current departure time with each time point within the predetermined time period. Based on the calculated values ​​corresponding to each time point within the predetermined time period, the calculated values ​​corresponding to each time point can be concatenated to form the current departure time feature, which can then be a feature vector in vector form.

[0101] The method of calculating the value corresponding to the time point within the predetermined time period by comparing the current departure time with the time point within the predetermined time period can specifically include one of the following methods: The first method is to calculate the time difference between the current departure time and the time point within the predetermined time period, and then remove the predetermined time granularity from the time difference to obtain the calculated value corresponding to the time point within the predetermined time period; The second method is to distribute the time points within the predetermined time period around a unit circle to obtain a time point circle, and then calculate the coordinate value corresponding to the current departure time on the time point circle to obtain the calculated value corresponding to the time point within the predetermined time period.

[0102] In the first method: If the predetermined time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and a time point within the predetermined time period is, for example, 0:00 on Monday, and the departure time is, for example, 8:00 AM on Tuesday, calculate the time difference between the departure time and the time point within the predetermined time period, in minutes (this can be understood as seconds or hours depending on the actual situation). The time difference is (24+8)*60 = 1920 minutes. Further, using 10-minute time granularities (this can be understood as setting the predetermined time granularity according to the actual situation), divide the time difference by the predetermined time granularity, and the calculated value corresponding to the time point "0:00 on Monday" within the predetermined time period is 1920 / 10 = 192.

[0103] In the second method, if the predetermined time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and further, if the smallest unit is the minute (which can be understood as 5 minutes, 10 minutes, etc. in actual use), then one week contains 24 * 7 * 60 = 10080 minutes. Assume these 10080 time points are evenly and orderly distributed around a unit circle, such that the last minute is adjacent to the first minute, resulting in a circle of time points around the unit circle (k = 1). This can be calculated using the formula... and Calculate the coordinates of the departure time t on the time circle. For example, if the departure time t was 8:00 AM on Tuesday, which is 1920 minutes different from 0:00 AM on Monday, then the coordinates of 8:00 AM on Tuesday (i.e., the two calculated values) would be: and In the above process, it is assumed that the number of minutes in a week revolves around the unit circle (k=1). Furthermore, we can let these 10080 minutes revolve around the unit circle any number of times, then according to the formula... and Calculate the coordinates of the departure time t on the time circle. For example, taking k=2, the coordinates of 8:00 AM on Tuesday (i.e., the two calculated values) are: and In the second approach, for a given departure time, its departure time feature can be expressed as an even number of features (i.e., calculated values), requiring only that k be chosen with more values. For cases where the departure time feature needs to be an odd number of features, the last feature can be set to 0, and the other features can be processed as described above. Therefore, the above processing can transform a moment into a departure time feature composed of any number of features.

[0104] (5) The residual calculation service can calculate arrival time residuals based on residual network parameters (e.g., Figure 3 (The residuals shown). Specifically, the residual calculation service can: perform feature extraction processing on the scheduled departure time to obtain the scheduled departure time features; and use a residual calculation network to perform residual calculation processing based on the residual network parameters and the scheduled departure time features to obtain the arrival time residuals.

[0105] Specifically, the process of extracting features from the scheduled departure time to obtain the scheduled departure time features may include: calculating the scheduled departure time and time points within a scheduled time period to obtain the calculated values ​​corresponding to the time points within the scheduled time period; and obtaining the scheduled departure time features corresponding to the scheduled departure time based on the calculated values ​​corresponding to the time points within the scheduled time period.

[0106] The predetermined time period can be a pre-specified time period, such as one day, one week, or two weeks. In one embodiment of this example, the predetermined time period is one week, meaning it starts at 0:00 on Monday and ends at 24:00 on Sunday. The number of time points within the predetermined time period can be set according to actual circumstances; for example, the time points within the predetermined time period can include 0:00 on Monday and several time points after 0:00 on Monday. The predetermined departure time can be calculated and processed with each time point within the predetermined time period to obtain the calculated values ​​corresponding to each time point within the predetermined time period. Based on the calculated values ​​corresponding to each time point within the predetermined time period, the calculated values ​​corresponding to each time point can be concatenated to form a predetermined departure time feature, which can then be a feature vector in vector form.

[0107] The method of calculating the time points within a predetermined time period by comparing the predetermined departure time with the time points within that predetermined time period can specifically include one of the following methods: The first method calculates the time difference between the predetermined departure time and the time points within the predetermined time period, and then removes the predetermined time granularity from the time difference to obtain the calculated value corresponding to the time points within the predetermined time period; the second method distributes the time points within the predetermined time period around a unit circle to obtain a time point circle, and then calculates the coordinate value of the predetermined departure time on the time point circle to obtain the calculated value corresponding to the time points within the predetermined time period.

[0108] In the first method: If the scheduled time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and a scheduled departure time is, for example, 0:00 on Monday, then the time difference between the scheduled departure time and a specific point in the scheduled time period is calculated in minutes (this can be understood as seconds or hours depending on the actual situation). The time difference is (24+8)*60 = 1920 minutes. Further, using a 10-minute time granularity (this can be understood as setting the time granularity according to the actual situation), the time difference is divided by the time granularity, resulting in a calculated value of 1920 / 10 = 192 for the time point "0:00 on Monday" within the scheduled time period.

[0109] In the second method, if the predetermined time period is one week, that is, a time period starting at 0:00 on Monday and ending at 24:00 on Sunday, and further, if the smallest unit is the minute (which can be understood as 5 minutes, 10 minutes, etc. in actual use), then one week contains 24 * 7 * 60 = 10080 minutes. Assume these 10080 time points are evenly and orderly distributed around a unit circle, such that the last minute is adjacent to the first minute, resulting in a circle of time points around the unit circle (k = 1). This can be calculated using the formula... and Calculate the coordinates of the scheduled departure time t on the time circle. For example, if the scheduled departure time t is 8:00 AM on Tuesday, which is 1920 minutes different from 0:00 AM on Monday, then the coordinates of 8:00 AM on Tuesday (i.e., the two calculated values) are: and In the above process, it is assumed that the number of minutes in a week revolves around the unit circle (k=1). Furthermore, we can let these 10080 minutes revolve around the unit circle any number of times, then according to the formula... and Calculate the coordinates of the scheduled departure time t on the time circle. For example, taking k=2, the coordinates of Tuesday morning 8:00 (i.e., the two calculated values) are: and In the second approach, for a given departure time, its departure time feature can be expressed as an even number of features (i.e., calculated values), requiring only that k be chosen with more values. For cases where the departure time feature needs to be an odd number of features, the last feature can be set to 0, and the other features can be processed as described above.

[0110] Furthermore, residual calculation processing is performed based on the residual network parameters and the predetermined departure time characteristics to obtain the arrival time residual. Specifically, this may include: fusing the residual network parameters and the predetermined departure time characteristics to obtain residual features; and obtaining the arrival time residual based on the residual features. The residual network parameters include target road segment sequence features, which include the road segment features corresponding to each road segment in the candidate route. Fusing the residual network parameters and the predetermined departure time characteristics to obtain residual features includes: performing an inner product operation between the road segment features corresponding to each road segment and the predetermined departure time features to obtain the residual features corresponding to each road segment; and obtaining the arrival time residual based on the residual features includes: processing the residual features corresponding to each road segment into constant positive values ​​and averaging the constant positive values ​​corresponding to each road segment to obtain the arrival time residual.

[0111] The target road segment sequence features include the road segment features corresponding to each road segment in the candidate route. For example, if the candidate route includes road segments 1 to N, then the target road segment sequence features can include road segment feature 1 corresponding to road segment 1 to road segment feature N corresponding to road segment N. The road segment features can be M-dimensional feature vectors, and the predetermined departure time features can also be M-dimensional feature vectors. By performing an inner product operation between the road segment features corresponding to each road segment in the target road segment sequence features and the predetermined departure time features, the residual features corresponding to each road segment can be obtained. For example, the inner product of road segment feature 1 and the predetermined departure time features yields residual feature 1, and the inner product of road segment feature N and the predetermined departure time features yields residual feature N. Furthermore, the residual features corresponding to each road segment can be processed into constant positive values ​​using functions such as the softplus function, and the average of the constant positive values ​​corresponding to each road segment is calculated to obtain the overall average residual, i.e., the arrival time residual.

[0112] (6) The final arrival time calculation service (i.e., the final ETA) can obtain the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residual.

[0113] The process involves obtaining the arrival time of a candidate route based on the estimated arrival time and the arrival time residual. This can be achieved through one of the following methods: 1) summing the estimated arrival time and the arrival time residual; or 2) weighted summing the estimated arrival time and the arrival time residual based on predetermined weights. In the first method, the estimated arrival time and the arrival time residual are directly summed. In the second method, the predetermined weights can include the weights corresponding to the estimated arrival time and the arrival time residual. The estimated arrival time and the arrival time residual are then weighted and summed based on these predetermined weights to obtain the arrival time of the candidate route. The extent to which the residuals take effect can be controlled by configuring these predetermined weights.

[0114] In this manner, in this scenario, by applying the foregoing embodiments of this application, at least the following beneficial effects are achieved: It can improve the response speed of the arrival time estimation service, improve the accuracy of the arrival time estimation service, optimize and correct the estimated arrival time based on the arrival time residual, and decouple it from the original preset estimation model. The solution is applicable to the optimization of preset estimation models using various technical approaches such as rule-based, tree-based, and neural network-based methods, and can avoid catastrophic forgetting problems. It can effectively improve the arrival time estimation effect without adjusting the preset estimation model, thereby enhancing the user experience.

[0115] To facilitate better implementation of the arrival time estimation method provided in the embodiments of this application, the embodiments of this application also provide an arrival time estimation device based on the above-described arrival time estimation method. The meanings of the terms used are the same as in the above-described arrival time estimation method, and specific implementation details can be found in the descriptions in the method embodiments. Figure 5 A block diagram of an arrival time estimation device according to an embodiment of this application is shown.

[0116] like Figure 5 As shown, the arrival time estimation device 400 may include a time acquisition module 410, a parameter acquisition module 420, a time characterization module 430, a calculation and processing module 440, and an arrival time module 450.

[0117] The time acquisition module 410 can be used to acquire the estimated arrival time corresponding to the candidate route at a predetermined departure time, wherein the estimated arrival time is estimated using a preset prediction model; the parameter acquisition module 420 can be used to acquire the residual network parameters corresponding to the candidate route, wherein the residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route; the time representation module 430 can be used to perform feature extraction processing on the predetermined departure time to obtain the predetermined departure time features; the calculation processing module 440 can be used to use the residual calculation network to perform residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain the arrival time residual; the arrival time module 450 can be used to obtain the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residual.

[0118] In some embodiments of this application, the apparatus further includes a parameter update module, configured to: periodically acquire driving data of users traveling via the candidate routes, the driving data including the current departure time and the road segment sequence corresponding to the candidate routes; extract current departure time features based on the current departure time, and extract initial road segment sequence features based on the road segment sequence; use the residual calculation network to perform residual calculation processing based on the current departure time features and the initial road segment sequence features to obtain a predicted residual; adjust the initial road segment sequence features according to the predicted residual until a predetermined condition is met to obtain a target road segment sequence feature, wherein the residual network parameters include the target road segment sequence features.

[0119] In some embodiments of this application, the driving data further includes a predicted arrival time and an actual arrival time; a parameter update module is used to: obtain a corrected arrival time based on the predicted residual and the predicted arrival time; adjust the initial road segment sequence features based on the error between the corrected arrival time and the actual arrival time until the error meets a predetermined error condition, thereby obtaining the target road segment sequence features.

[0120] In some embodiments of this application, the time characterization module is used to: calculate and process the predetermined departure time with time points within the predetermined time period to obtain calculated values ​​corresponding to the time points within the predetermined time period; and obtain the predetermined departure time feature corresponding to the predetermined departure time based on the calculated values ​​corresponding to the time points within the predetermined time period.

[0121] In some embodiments of this application, the time characterization module is used to implement one of the following methods: calculating the time difference between the predetermined departure time and a time point within the predetermined time period, and removing a predetermined time granularity from the time difference to obtain the calculated value corresponding to the time point within the predetermined time period; distributing the time points within the predetermined time period around a unit circle to obtain a time point circle, and calculating the coordinate value corresponding to the predetermined departure time on the time point circle to obtain the calculated value corresponding to the time point within the predetermined time period.

[0122] In some embodiments of this application, the calculation processing module is used to: fuse the residual network parameters with the predetermined departure time features to obtain residual features; and obtain the arrival time residual based on the residual features.

[0123] In some embodiments of this application, the residual network parameters include target road segment sequence features, which include road segment features corresponding to each road segment in the candidate route; the calculation processing module is used to: perform an inner product operation on the road segment features corresponding to each road segment and the predetermined departure time features to obtain the residual features corresponding to each road segment; process the residual features corresponding to each road segment into constant positive values, and average the constant positive values ​​corresponding to each road segment to obtain the arrival time residual.

[0124] In some embodiments of this application, the arrival time module is used to implement one of the following methods: summing the estimated arrival time and the arrival time residual to obtain the route arrival time corresponding to the candidate route; and weighting the estimated arrival time and the arrival time residual based on a predetermined weight to obtain the route arrival time corresponding to the candidate route.

[0125] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0126] Furthermore, embodiments of this application also provide an electronic device, which can be a terminal or a server, such as... Figure 6 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0127] The electronic device may include components such as a processor 501 with one or more processing cores, a memory 502 with one or more computer-readable storage media, a power supply 503, and an input unit 504. Those skilled in the art will understand that... Figure 6 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0128] The processor 501 is the control center of the electronic device. It connects to various parts of the computer device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 502, and by calling data stored in the memory 502, it performs various functions of the computer device and processes data, thereby detecting the electronic device. Optionally, the processor 501 may include one or more processing cores; preferably, the processor 501 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user page, and application programs, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 501.

[0129] The memory 502 can be used to store software programs and modules. The processor 501 executes various functional applications and data processing by running the software programs and modules stored in the memory 502. The memory 502 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one 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 computer device, etc. In addition, the memory 502 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. Accordingly, the memory 502 may also include a memory controller to provide the processor 501 with access to the memory 502.

[0130] The electronic device also includes a power supply 503 that supplies power to various components. Preferably, the power supply 503 can be logically connected to the processor 501 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 503 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0131] The electronic device may also include an input unit 504, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0132] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 501 in the electronic device loads the executable files corresponding to the processes of one or more computer programs into the memory 502 according to the following instructions, and the processor 501 runs the computer programs stored in the memory 502, thereby realizing the various functions of the foregoing embodiments of this application.

[0133] For example, processor 501 can execute the following: obtaining the estimated arrival time corresponding to the candidate route at a predetermined departure time, wherein the estimated arrival time is estimated using a preset estimation model; obtaining the residual network parameters corresponding to the candidate route, wherein the residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route; performing feature extraction processing on the predetermined departure time to obtain predetermined departure time features; using the residual calculation network, performing residual calculation processing based on the residual network parameters and the predetermined departure time features to obtain arrival time residuals; and obtaining the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residuals.

[0134] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by a computer program, or by a computer program controlling related hardware. The computer program can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0135] Therefore, embodiments of this application also provide a computer-readable storage medium storing a computer program that can be loaded by a processor to perform the steps in any of the methods provided in embodiments of this application.

[0136] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0137] Since the computer program stored in the computer-readable storage medium can execute the steps of any of the methods provided in the embodiments of this application, the beneficial effects that the methods provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0138] 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 methods provided in the various optional implementations of the above embodiments of this application.

[0139] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0140] It should be understood that this application is not limited to the embodiments described above and shown in the accompanying drawings, but various modifications and changes can be made without departing from its scope.

Claims

1. A method for estimating arrival time, characterized in that, include: Obtain the estimated arrival time of the candidate route at the predetermined departure time, wherein the estimated arrival time is estimated using a preset estimation model; Periodically acquire the driving data of users who travel through the candidate routes, including the departure time and the road segment sequence corresponding to the candidate routes; Based on the departure time at that time, extract the departure time feature at that time; and based on the road segment sequence, extract the initial road segment sequence feature. A residual calculation network is used to perform residual calculation processing based on the departure time characteristics and the initial road segment sequence characteristics to obtain the prediction residual; the initial road segment sequence characteristics are adjusted according to the prediction residual until they meet the predetermined conditions to obtain the target road segment sequence characteristics; Obtain the residual network parameters corresponding to the candidate route. The residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route. The residual network parameters include the target road segment sequence features, which include the road segment features corresponding to each road segment in the candidate route. The predetermined departure time is subjected to feature extraction processing to obtain predetermined departure time features, including: calculating the predetermined departure time and time points within a predetermined time period to obtain calculated values ​​corresponding to the time points within the predetermined time period; and obtaining the predetermined departure time features corresponding to the predetermined departure time based on the calculated values ​​corresponding to the time points within the predetermined time period. Using the residual calculation network, residual calculation processing is performed based on the residual network parameters and the predetermined departure time characteristics to obtain the arrival time residual. This includes: performing an inner product operation between the road segment characteristics corresponding to each road segment and the predetermined departure time characteristics to obtain the residual characteristics corresponding to each road segment; processing the residual characteristics corresponding to each road segment into constant positive values, and averaging the constant positive values ​​corresponding to each road segment to obtain the arrival time residual. The route arrival time corresponding to the candidate route is obtained based on the estimated arrival time and the arrival time residual, including: summing the estimated arrival time and the arrival time residual to obtain the route arrival time corresponding to the candidate route; or, weighting and summing the estimated arrival time and the arrival time residual based on a predetermined weight to obtain the route arrival time corresponding to the candidate route.

2. The method according to claim 1, characterized in that, The driving data also includes the predicted arrival time and the actual arrival time; adjusting the initial road segment sequence features based on the prediction residual until a predetermined condition is met to obtain the target road segment sequence features includes: The corrected arrival time is obtained based on the predicted residual and the predicted arrival time. Based on the error between the corrected arrival time and the actual arrival time, the initial road segment sequence features are adjusted until the error meets the predetermined error condition, thereby obtaining the target road segment sequence features.

3. The method according to claim 1, characterized in that, The step of calculating the predetermined departure time and the time points within the predetermined time period to obtain the calculated values ​​corresponding to the time points within the predetermined time period includes one of the following methods: Calculate the time difference between the predetermined departure time and the time point within the predetermined time period, and remove the predetermined time granularity from the time difference to obtain the calculated value corresponding to the time point within the predetermined time period. The time points within the predetermined time period are distributed around a unit circle to obtain a time point circle. The coordinate value corresponding to the predetermined departure time on the time point circle is calculated to obtain the calculated value corresponding to the time point within the predetermined time period.

4. An arrival time estimation device employing the method described in any one of claims 1 to 3, characterized in that, include: The time acquisition module is used to acquire the estimated arrival time of the candidate route at the predetermined departure time. The estimated arrival time is estimated using a preset estimation model. The parameter acquisition module is used to acquire the residual network parameters corresponding to the candidate route. The residual network parameters are obtained by periodically training the residual calculation network based on the driving data of users passing through the candidate route. The time representation module is used to perform feature extraction processing on the predetermined departure time to obtain the predetermined departure time features; The calculation and processing module is used to perform residual calculation processing based on the residual network parameters and the predetermined departure time characteristics using the residual calculation network to obtain the arrival time residual; The arrival time module is used to obtain the route arrival time corresponding to the candidate route based on the estimated arrival time and the arrival time residual.

5. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the method described in any one of claims 1 to 3.

6. An electronic device, characterized in that, include: Memory, which stores computer programs; A processor reads a computer program stored in memory to execute the method according to any one of claims 1 to 3.

7. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 3.

Images