Delivery duration determination method and device, computer program product and storage medium

Abstract

The embodiment of the invention provides a distribution duration determination method and device, a computer program product and a storage medium. When the delivery duration is adjusted, aiming at each adjustment value in a plurality of preset adjustment values, determining a corresponding order combining probability gain based on the waybill information of the target waybill and the initial delivery duration, the order combining probability gain can be a difference value between the order combining probability of the target waybill under the condition that the initial delivery duration is adjusted by using the adjustment value and the order combining probability of the target waybill under the condition that the initial delivery duration is not adjusted; and then an adjustment strategy for adjusting the initial delivery duration can be determined based on the order combining probability gains corresponding to the plurality of adjustment values, so that the target delivery duration of the target waybill can be obtained based on the adjustment strategy, and the target waybill can be scheduled and allocated. By quantifying the gain effects of different adjustment strategies, the optimal adjustment strategy can be selected, and the operation efficiency is improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present specification relate to the technical field of logistics distribution, and in particular to a delivery time length determination method, device, computer program product and storage medium. BACKGROUND

[0002] In the field of logistics distribution, the delivery time length of a waybill is a key indicator, which can be used to assist in decision-making for scheduling the waybill to improve delivery efficiency. For example, in order to improve the utilization rate of delivery resources and reduce the delivery cost of each waybill, a consolidation strategy can be adopted, i.e., multiple waybills with similar paths and matching time windows are consolidated for delivery. When consolidating the waybills, the delivery time length is usually taken into account to avoid the consolidation causing the delivery time length to be too long and bringing a bad experience to the user. Usually, after a new waybill is generated, the delivery time length of the waybill can be predicted first to obtain an initial delivery time length of the waybill, and then the initial delivery time length can be adjusted to expect that the delivery time length of the waybill is adjusted to improve the consolidation rate of the waybill. In related technologies, the adjustment of the delivery time length is not flexible enough, and the gain effect brought by the adjustment of the delivery time length cannot be evaluated, which cannot provide a basis for subsequent operation decision-making. SUMMARY

[0003] To overcome the problems in related technologies, embodiments of the present specification provide a delivery time length determination method, device, computer program product and storage medium.

[0004] According to a first aspect of embodiments of the present specification, a delivery time length determination method is provided, the method comprising: obtaining an initial delivery time length of a target waybill predicted in advance, and waybill information of the target waybill; for each adjustment value in a plurality of preset adjustment values, determining a consolidation probability gain corresponding to the adjustment value based on the initial delivery time length and the waybill information, wherein the consolidation probability gain is a difference between a consolidation probability of the target waybill in a case where the initial delivery time length is adjusted by using the adjustment value and a consolidation probability of the target waybill in a case where the initial delivery time length is not adjusted; determining an adjustment strategy for adjusting the initial delivery time length based on the consolidation probability gain corresponding to each adjustment value in the plurality of adjustment values; obtaining a target delivery time length of the target waybill based on the adjustment strategy and the initial delivery time length.

[0005] According to a second aspect of embodiments of the present specification, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the method mentioned in the first aspect above.

[0006] According to a third aspect of the embodiments of this specification, an electronic device is provided, the electronic device including a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein the computer program, when executed, implements the method mentioned in the first aspect above.

[0007] According to a fourth aspect of the embodiments of this specification, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, it implements the method mentioned in the first aspect above.

[0008] The beneficial effects of the embodiments in this specification are as follows: When adjusting the delivery time, multiple adjustment values ​​can be preset. For each adjustment value, the corresponding order merging probability gain can be determined based on the order information of the target order and the initial delivery time. The order merging probability gain can be the difference between the order merging probability of the target order when the initial delivery time is adjusted using the adjustment value and the order merging probability of the target order when the initial delivery time is not adjusted. Then, an adjustment strategy for adjusting the initial delivery time can be determined based on the order merging probability gains corresponding to the multiple adjustment values. The target delivery time of the target order can be obtained based on the adjustment strategy, so that the target order can be scheduled and allocated based on the target delivery time.

[0009] By calculating the order merging probability gain corresponding to each adjustment value, the most effective adjustment strategy for a specific order can be accurately identified. This "one-order-one-strategy" approach avoids the blindness of traditional fixed adjustment strategies, maximizes the merging potential of individual orders, and thus improves the overall order merging rate. Compared to traditional models that only focus on "changes in the overall order merging rate," the embodiments in this specification directly quantify the actual impact of each adjustment strategy on a single order through "order merging probability gain." By quantifying the incremental effect, data support is provided for operational decisions to achieve refined operations. Furthermore, this dynamic response mechanism makes delivery time adjustments more aligned with real-time needs, enhancing the flexibility and adaptability of the strategy. This allows for maximizing operational efficiency and reducing operational costs without significantly impacting user experience.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit the embodiments of this specification. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of the embodiments of this specification, illustrate embodiments consistent with those of this specification and, together with the specification, serve to explain the principles of the embodiments of this specification.

[0012] Figure 1 This is a schematic diagram illustrating an application scenario as an exemplary embodiment of this specification; Figure 2 A flowchart illustrating a delivery time determination method as an exemplary embodiment of this specification; Figure 3 This is a schematic diagram illustrating a waybill type as shown in an exemplary embodiment of this specification; Figure 4 This is a schematic diagram illustrating the training process of a first single probability prediction model as an exemplary embodiment of this specification. Figure 5 This is a schematic diagram illustrating the training process of the first gain prediction model and the second gain prediction model, which is an exemplary embodiment of this specification. Figure 6(a) is a schematic diagram of the training process of the second and third combined single probability prediction models shown in an exemplary embodiment of this specification. Figure 6(b) is a schematic diagram of the structure of the third gain prediction model shown in an exemplary embodiment of this specification; Figure 7 This is a logic block diagram of an electronic device illustrated in an exemplary embodiment of this specification. Detailed Implementation

[0013] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those described in this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the embodiments described in this specification as detailed in the appended claims.

[0014] The terminology used in the embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the embodiments of this specification. The singular forms “a,” “described,” and “the” as used in the embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0015] It should be understood that although the terms first, second, third, etc., may be used to describe various information in the embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of the embodiments of this specification, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0016] In the logistics and delivery sector, to improve the utilization rate of delivery resources and reduce the delivery cost per order, a consolidation strategy can be adopted, which involves merging multiple orders with similar routes and matching time windows for delivery. When consolidating orders, the relationship between the consolidation rate and delivery time must usually be considered. For example, the consolidation rate can often be improved by adjusting the delivery time of the orders. However, longer delivery times can also lead to a decline in user experience. Therefore, while adjusting the delivery time, the goal is usually to find an adjustment solution that can significantly improve the consolidation rate without causing significant loss of user experience.

[0017] Typically, after a new waybill is generated, its initial delivery time can be predicted. This initial delivery time can then be adjusted to improve the waybill consolidation rate. Related technologies usually employ a fixed adjustment strategy to change the initial delivery time before determining whether consolidation is feasible. For example, the initial delivery time might be uniformly extended by a fixed amount before subsequent consolidation and scheduling. This approach cannot dynamically adjust to different scenarios and needs, potentially leading to excessively long delivery times in some cases, negatively impacting user experience. Furthermore, because it doesn't consider other factors affecting waybill consolidation, the results are often less than ideal. Some technologies can predict the impact of delivery time adjustments on the overall order merging rate through pre-trained models. However, this approach mainly focuses on the overall effect of delivery time adjustments, such as the impact on the overall order merging rate, while ignoring the specific impact of a certain delivery time adjustment strategy on a single order. In this case, it is difficult to accurately assess the actual gains of each delivery time adjustment strategy on a specific order or rider, i.e., it is impossible to quantify the incremental effect of each adjustment strategy to provide a basis for decision-making in subsequent operations.

[0018] Based on this, embodiments of this specification provide a method for determining delivery time. After determining the initial delivery time of a target waybill, it is possible to predict the order merging probability gain brought to the target waybill by different delivery time adjustment strategies. This allows for the quantification of the incremental effect of each delivery time adjustment strategy on the target waybill, and further, the delivery time adjustment strategy for the target waybill can be decided based on this incremental effect. Specifically, when adjusting the delivery time, multiple adjustment values ​​can be preset. For each adjustment value, the corresponding order merging probability gain can be determined based on the waybill information and the initial delivery time of the target waybill. The order merging probability gain can be the difference between the order merging probability of the target waybill when the initial delivery time is adjusted using the adjustment value and the order merging probability of the target waybill when the initial delivery time is not adjusted. Then, based on the order merging probability gains corresponding to the multiple adjustment values, an adjustment strategy for adjusting the initial delivery time can be determined, and the target delivery time of the target waybill can be obtained based on the adjustment strategy, so that the target waybill can be scheduled and allocated based on the target delivery time.

[0019] By calculating the "order merging probability gain" (i.e., the difference in order merging probability between the adjusted and unadjusted values) corresponding to each adjustment value, the most effective adjustment strategy for a specific order can be accurately identified. This "one order, one strategy" approach avoids the blindness of traditional fixed adjustment strategies, maximizes the merging potential of individual orders, and thus improves the overall order merging rate. Compared to traditional models that only focus on "changes in the overall order merging rate," the embodiments in this specification directly quantify the actual impact of each adjustment strategy on a single order through "order merging probability gain." By quantifying the incremental effect, data support is provided for operational decisions to achieve refined operations. Furthermore, this dynamic response mechanism makes delivery time adjustments more aligned with real-time needs, enhancing the flexibility and adaptability of the strategy. This allows for maximizing operational efficiency and reducing operational costs without significantly impacting user experience.

[0020] The order merging rate mentioned in the embodiments of this specification is relative to multiple waybills. For example, if 2 out of 10 waybills are merged into one order within a certain period, the order merging rate is 20%. The order merging probability mentioned in the embodiments of this specification is relative to a single waybill. For example, if a waybill can be merged with other waybills, its order merging probability is 1; if it cannot be merged with other waybills, its order merging probability is 0. The embodiments of this specification optimize the order merging probability of a single waybill by quantifying the impact of different delivery time adjustment strategies on the order merging probability of a single waybill, thereby improving the overall order merging rate.

[0021] like Figure 1The diagram illustrates an application scenario of an embodiment of this specification. Delivery services are widely used in online shopping, food delivery, and errand-running services, involving multi-party interactions between the service provider, merchants, delivery capacity, and users. The service platform provides a server and a user client to users, allowing them to access services offered by the platform. In addition to the user client, the platform also provides a merchant client to merchants, enabling them to use the platform's services. Delivery capacity refers to entities with delivery capabilities, including but not limited to delivery personnel, such as delivery riders. Delivery capacity communicates with the server through a delivery capacity client. In other examples, delivery capacity may also include unmanned delivery equipment, such as drones and unmanned vehicles. Users can transact with merchants and initiate delivery orders through their user clients; the service provider can allocate delivery capacity for these immediate delivery orders.

[0022] For example, a user selects a target product on their client and places an order, generating a target waybill on their client. The client then sends this waybill to the server, which forwards it to the merchant's client for inventory preparation. Simultaneously, the server can predict the initial delivery time of the target waybill based on its information (e.g., delivery distance, delivery time, weather) and historical data. The server then determines the order-merging probability gain for each of several preset adjustment values ​​based on the waybill information. Based on the combined probability of these adjustments, the server determines a delivery time adjustment strategy for the target waybill. This strategy is used to adjust the initial delivery time, resulting in the target delivery time. Finally, the server can schedule the target waybill based on this target delivery time, such as performing order merging or allocation.

[0023] For example, the server can send the target order to the delivery capacity client. The delivery capacity can obtain the target order through the delivery capacity client. If the delivery capacity wants to deliver the target order, it can issue an order acceptance instruction. The user client will then detect the delivery capacity's order acceptance operation. At the same time, based on the delivery capacity's order acceptance operation, the server can send the target delivery time to the user client, so that the target user can obtain the delivery time of the target product they ordered on their user client.

[0024] The delivery time determination method provided in the embodiments of this specification can be executed by the aforementioned server, which can be deployed on a server or a server cluster.

[0025] like Figure 2As shown, the method for determining delivery time may include the following steps: S202. Obtain the pre-predicted initial delivery time of the target waybill, and the waybill information of the target waybill; In step S202, after the target waybill is generated, the delivery time of the target waybill can be predicted. For example, the delivery time of the target waybill can be predicted based on factors such as delivery distance, delivery time (whether it is during morning or evening peak hours), and weather, to obtain the aforementioned initial delivery time. For example, a delivery time prediction model can be trained based on historical waybill data, and the initial delivery time of the target waybill can be predicted using this delivery time prediction model based on relevant information of the target waybill.

[0026] Simultaneously, the waybill information of the target waybill can be obtained. This waybill information can be various types of information related to the delivery of the target waybill. For example, the waybill information may include real-time features related to the target waybill, or historical features related to the target waybill. Real-time features may include the characteristics of the waybill, the merchant, the rider, features representing the interaction between the rider and the waybill, features of the business district, etc. Historical features include features representing the historical interaction between the rider and the waybill, features representing the historical interaction between waybills, etc. For example, the characteristics of the waybill may be the delivery distance and delivery time of the target waybill; the characteristics of the merchant may be related features such as the food preparation speed of the merchant corresponding to the target waybill; and the characteristics of the rider may be related features such as the number of riders in the business district. The specific content of the waybill information can be flexibly set based on actual needs, and this specification does not impose limitations on the embodiments.

[0027] S204. For each of the preset multiple adjustment values, based on the initial delivery time and the waybill information, determine the order merging probability gain corresponding to the adjustment value, wherein the order merging probability gain is: the difference between the order merging probability of the target waybill when the initial delivery time is adjusted using the adjustment value and the order merging probability of the target waybill when the initial delivery time is not adjusted. In step S204, to determine the optimal delivery time adjustment strategy for the current waybill, multiple adjustment values ​​can be pre-set. For example, these values ​​could be increasing the delivery time by 1 minute, 2 minutes, 3 minutes, 4 minutes, 5 minutes, etc., depending on actual needs. These adjustment values ​​can be the same or different for different waybills. For instance, the upper and lower limits of the adjustment values, as well as the step size, can differ for different waybills. For waybills with shorter initial delivery times, the upper limit of the adjustment value can be larger, while for waybills with longer initial delivery times, the upper limit can be smaller. Furthermore, the step size of the adjustment value can be flexibly adjusted based on actual needs. For example, in scenarios where users are sensitive to delivery time, the step size can be set smaller, while in scenarios where users are not sensitive to delivery time, the step size can be set larger. The order consolidation probability gain can be positive or negative.

[0028] For each of the preset adjustment values, based on the initial delivery time and the order information of the target order, the order merging probability gain corresponding to that adjustment value can be determined. This order merging probability gain is the difference between the order merging probability of the target order when the initial delivery time is adjusted using the adjustment value and the order merging probability of the target order when the initial delivery time is not adjusted. The order merging probability refers to the predicted order merging probability. For example, assuming the predicted order merging probability of the target order when the initial delivery time is not adjusted is P0, and the predicted order merging probability of the target order when the initial delivery time is adjusted using the adjustment value T1 is P1, then the order merging probability gain is P1 - P0.

[0029] The determination of the order merging probability gain can take several forms. For example, an order merging probability gain prediction model can be pre-trained, and this model can be used to predict the order merging probability gain based on the initial delivery time and the order information of the target order. Alternatively, an order merging probability prediction model can be pre-trained to predict the predicted order merging probability of the target order without adjusting the initial delivery time, and the predicted order merging probability of the target order with adjusted initial delivery time. The difference between the two is then calculated to obtain the order merging probability gain. Alternatively, the order information of the target order can be compared and analyzed with the order information of historical orders, and the order merging probability gain of similar historical orders can be used to determine the order merging probability gain of the target order.

[0030] S206. Determine an adjustment strategy for adjusting the initial delivery time based on the order combination probability gain corresponding to each of the multiple adjustment values. In step S206, after determining the order merging probability gain corresponding to each of the multiple adjustment values, an adjustment strategy for adjusting the initial delivery time can be determined based on this probability gain. This adjustment strategy can include not adjusting the initial delivery time, adjusting the initial delivery time, and specifying the adjustment value. The overall principle for determining the adjustment strategy is to increase the order merging probability while minimizing impact on the user experience.

[0031] S208. The target delivery time of the target waybill is obtained based on the initial delivery time and the adjustment strategy.

[0032] In step S208, after determining the adjustment strategy, it is possible to determine whether to adjust the initial delivery time and the specific adjustment method based on the adjustment strategy, so as to obtain the target delivery time for the target calculation. Then, the target waybill can be scheduled based on the target delivery time, for example, determining whether to combine the target waybill with other waybills, determining the allocation method of the target waybill, etc.

[0033] Considering that adjusting delivery time has a very small incremental effect on the probability of order merging, meaning the gain in the probability of order merging is very small, it means that even increasing delivery time will not significantly improve the probability of order merging. In this case, to avoid a decline in user experience due to increased delivery time, there is no need to adjust the delivery time. Alternatively, if the gain in the probability of order merging is negative, meaning that adjusting the delivery time reduces the probability of order merging (for example, causing orders that could previously be merged to become unmergeable), then there is also no need to adjust the delivery time.

[0034] Therefore, in some embodiments, when determining the adjustment strategy for adjusting the initial delivery time based on the order merging probability gain corresponding to each of the multiple adjustment values, if the order merging probability gain corresponding to each of the multiple adjustment values ​​is less than the preset gain, then the adjustment strategy is not to adjust the initial delivery time.

[0035] For example, a preset gain can be set as a judgment threshold. This preset gain can be set based on actual needs, such as 5%. Therefore, after adjusting the initial delivery time using all adjustment values, the gain of the order merging probability is less than 5% (for example, a positive gain or a negative gain of less than 5%). This indicates that adjusting the initial delivery time has little impact on the order merging probability. Therefore, the initial delivery time can be kept unchanged to avoid the negative impact of ineffective adjustments on the user experience.

[0036] In some embodiments, if at least one of the adjustment values ​​corresponds to a merging probability gain greater than or equal to a preset gain, i.e., one or more adjustment values ​​significantly increase the merging probability, then the change in the predicted merging status of the target waybill can be determined by adjusting the initial delivery time using each of the at least one adjustment value compared to not adjusting the initial delivery time. An adjustment strategy is then determined based on this change in the predicted merging status, where the predicted merging status includes both merged and non-merged orders. For example, if at least one adjustment value achieves a gain (i.e., exceeds the preset gain), the difference between the predicted merging status (merged or non-merged) of the target waybill under these adjustment values ​​and the unadjusted state is further analyzed. For instance, whether adjusting the delivery time can change the waybill from non-merged to merged, or whether it can stabilize the merged status (e.g., the original merged status was merged but the probability of merged was low; after adjusting the delivery time, the status remains merged, but the probability of merged is greatly increased, making the merged status more stable). Finally, the actual value of the status change is used to determine whether to adjust the delivery time.

[0037] For example, if increasing the initial delivery time by 3 minutes or 5 minutes results in a greater gain in the order merging probability than a preset gain, then it can be further determined whether the order merging status of the target order changes after the initial delivery time increases by 3 minutes or 5 minutes. For instance, if the target order was originally not merged after the initial delivery time increases by 3 minutes or 5 minutes, and the merged status remains not merged after the delivery time adjustment (i.e., the merged status hasn't changed), then there's no need to perform invalid order merging to avoid impacting user experience. Alternatively, if the target order was originally merged, and the merged status remains merged after the delivery time adjustment, but the increase in merged status is not significant, then adjusting the delivery time is also unnecessary. Furthermore, if increasing the initial delivery time by 5 minutes results in the target order originally not merged, and the merged status becomes merged after the delivery time adjustment, then adjusting the initial delivery time can be considered to increase the merged status and improve delivery efficiency.

[0038] First, filtering out inefficient adjustments by setting a pre-defined gain threshold can reduce unnecessary delays and balance order merging efficiency with user experience. Then, combining qualitative analysis of order merging status (rather than relying solely on quantitative changes in probability gain) makes the adjustment strategy more aligned with the core objective (the actual outcome of order merging or not), avoiding resource waste caused by "a slight increase in the probability of order merging but no actual merging," ultimately achieving more accurate and valuable delivery time optimization.

[0039] In some embodiments, when determining an adjustment strategy based on changes in the predicted order merging status, if the predicted order merging status corresponding to each of the at least one adjustment value remains unchanged, the adjustment strategy is to not adjust the initial delivery time. If one of the at least one adjustment value corresponds to a target adjustment value whose order merging status changes from non-merging to merged, the adjustment strategy is to adjust the initial delivery time using the smallest adjustment value among the target adjustment values.

[0040] Specifically, if all adjusted predicted order merging statuses among the selected adjustment values ​​are consistent with the unadjusted status (e.g., always merged or never merged), then the initial delivery time is maintained to avoid meaningless time changes. If there exists at least one "target adjustment value" that can change the order merging status from non-merged to merged (i.e., achieving the key transition from "cannot be merged" to "can be merged"), then the smallest adjustment range among these valid adjustment values ​​is selected to adjust the initial delivery time. For example, if adding "3 min" or "5 min" to the initial delivery time can change the order merging status from non-merged to merged, then "adding 3 min" to the initial delivery time will yield the target delivery time.

[0041] By further filtering redundant adjustments based on whether the "state has undergone a qualitative change," we ensure that each extension of delivery time brings a real breakthrough in order merging results, rather than merely a probabilistic improvement, thus avoiding ineffective time increases. Prioritizing the minimum adjustment value minimizes the impact on user experience (such as avoiding excessively extended delivery times) while achieving order merging goals. This achieves an optimal balance between "improving order merging rates," "controlling the increase in delivery time," and "ensuring user experience," making the adjustment strategy both efficient and precise, and more aligned with core practical needs.

[0042] For example, such as Figure 3As shown, the horizontal and vertical axes represent the order merging and non-merging scenarios for a waybill under the conditions of adding 't' to the initial delivery time (i.e., extending the initial delivery time by 't') and not adding 't' (i.e., not adjusting the initial delivery time). Waybills can be divided into four types. The waybill in the upper left corner is one that merges due to adding 't'; this is the kind of waybill where adding 't' is truly necessary. For this type of waybill, the need to add 't' and by how much can be determined based on its merging probability gain. "Must-merge" means that the waybill will merge regardless of whether 't' is added. For example, if its merging status is merged before adding 't' and remains merged after adding 't', then there is no need to adjust its delivery time. "Adding 't' but not merging" means that its merging status is merged before adding 't' and changes to non-merging after adding 't' (i.e., the merging probability gain is negative), then there is no need to adjust its delivery time. For "non-combinable orders," which are orders that will not be combined regardless of whether a "t" is added (i.e., before adding a "t," their combined status is non-combined, and after adding a "t," their combined status changes to non-combined), then there is no need to adjust their delivery time.

[0043] In some embodiments, after obtaining the target delivery time for a target order based on the adjustment strategy and the initial delivery time, the target order can be scheduled based on this target delivery time. For example, after determining the target delivery time based on the adjustment strategy, the target delivery time can be used as the core constraint for scheduling the target order. For instance, the target delivery time can be incorporated into the order's basic attributes and matched with the target delivery times and route information of other orders to filter out orders that are compatible in terms of time windows and spatial routes for order merging. Then, by combining factors such as the rider's real-time location, load capacity, and historical delivery efficiency, the merged task package or individual order can be assigned to the optimal rider, and a specific delivery route can be planned based on the target delivery time. Throughout the process, the target delivery time serves as both the "time benchmark" for order merging and the "timeliness benchmark" for rider task execution, ensuring that the entire scheduling process revolves around the optimized target delivery time and improving overall operational efficiency.

[0044] In some embodiments, to determine the combined order probability gain corresponding to each adjustment value, a combined order probability prediction model can be pre-trained to predict the combined order probability, hereinafter referred to as the first combined order probability prediction model. For example, such as Figure 4As shown, a large amount of historical waybill information can be collected. These historical waybills include two categories: Category 1 (with tags) and Category 2 (with tags). Category 1 historical waybills are those whose initial delivery time has not been adjusted. The tags on Category 1 historical waybills indicate whether the waybill has been merged. Category 2 historical waybills are those whose initial delivery time has been adjusted. The tags on Category 2 historical waybills also indicate whether the waybill has been merged. These Category 1 and Category 2 historical waybills can then be used as training samples to train a pre-defined initial model. For example, the initial model can be used to predict the probability of merging Category 1 and Category 2 historical waybills. The model parameters of the initial model are adjusted based on the difference between the predicted merging probability and the tag (actual merging probability) until a pre-defined condition is met, thus training the first merging probability prediction model.

[0045] When determining the order merging probability gain corresponding to the adjustment value based on the initial delivery time and the order information of the target order, the order information of the target order and the initial delivery time can be input into the pre-trained first order merging probability prediction model. The first order merging probability prediction model predicts the non-intervention order merging probability of the target order without adjusting the initial delivery time. Then, the order information of the target order and the delivery time obtained by adjusting the initial delivery time using the adjustment value can be input into the first order merging probability prediction model. The first gain prediction model predicts the intervention order merging probability of the target order with the initial delivery time adjusted using the adjustment value. The difference between the intervention order merging probability and the non-intervention order merging probability can then be used as the order merging probability gain.

[0046] In some embodiments, the first order merging probability prediction model can be a deep learning model, such as a DNN. Considering that deep learning models possess strong nonlinear fitting capabilities and can automatically uncover the complex interactions between waybill information, initial delivery time, and adjustment values, a single model can be used to uniformly process both types of historical waybill data—unadjusted and adjusted initial delivery times—to obtain a prediction model that can accurately predict the merging probability of waybills that have not been intervened and those that have been intervened. This single-model architecture reduces model maintenance costs, eliminates the need for separate optimization for different waybill types, and lowers the complexity of engineering implementation.

[0047] In some embodiments, a gain prediction model can also be directly constructed and trained to directly predict the probability gain of order consolidation. For example, a gain prediction model can be trained separately for waybills with unadjusted initial delivery time (i.e., uninterrupted waybills) and waybills with adjusted initial delivery time (i.e., intervened waybills) to improve the accuracy of prediction results by using different gain prediction models to predict the two types of waybills respectively.

[0048] For example, such as Figure 5As shown, a large number of historical waybills of type I without adjusted initial delivery times can be collected. Counterfactual prediction can be performed on these historical waybills, that is, predicting the probability of merging these historical waybills when their initial delivery times are adjusted. This is referred to as the predicted intervention merging probability. Then, the difference between the predicted intervention merging probability and the actual merging probability of the historical waybills can be calculated and used as the merging probability gain label for the historical waybills. The labeled historical waybills are then used to train a preset initial model. For example, the initial model can be used to predict the merging probability gain of the historical waybills. The model parameters of the initial model are adjusted based on the difference between the predicted merging probability gain and the label until preset conditions are met, thus training a first gain prediction model. This first gain prediction model can be used to predict the merging probability gain corresponding to waybills without adjusted initial delivery times.

[0049] Simultaneously, a large number of second-type historical waybills with adjusted initial delivery times can be collected. Counterfactual predictions can be performed on these second-type historical waybills, i.e., predicting the probability of merging these waybills without adjusting their initial delivery times. This is referred to as the predicted probability of merging without intervention. Then, the difference between the actual probability of merging and the predicted probability of merging without intervention can be calculated and used as the merging probability gain label for these second-type historical waybills. These labeled second-type historical waybills are then used to train a pre-set initial model. For example, the initial model can be used to predict the merging probability gain of second-type historical waybills. The model parameters of the initial model are adjusted based on the difference between the predicted merging probability gain and the label until preset conditions are met, thus training a second-gain prediction model. This second-gain prediction model can be used to predict the merging probability gain corresponding to waybills with adjusted initial delivery times.

[0050] When determining the order merging probability gain corresponding to the adjustment value based on the initial delivery time and the order information of the target order, the initial delivery time and the order information of the target order can be input into the first gain prediction model, and the first gain prediction model outputs the first order merging probability gain of the target order. Here, the first order merging probability gain represents the order merging probability gain corresponding to the target order when it is calculated as an order that does not require adjustment of the initial delivery time.

[0051] Simultaneously, the waybill information of the target delivery time and the adjustment value can be input into the second gain prediction model, which then outputs the second merging probability gain of the target waybill. The second merging probability gain represents the merging probability gain corresponding to the target waybill when its initial delivery time needs adjustment. Then, a weighted average of the first and second merging probability gains can be performed to obtain the merging probability gain of the target waybill.

[0052] In some embodiments, the first gain prediction model and the second gain prediction model can employ machine learning models, such as XGBoost. Machine learning models offer faster training and inference speeds when the amount of data is limited or the feature relationships are relatively simple. The separate model design can reduce noise interference caused by the mixing of two types of waybill data. In scenarios where the features of the two types of waybills differ significantly, the prediction accuracy is more controllable, and the model is more interpretable, making it easier for technicians to understand the logic of gain prediction and providing a more transparent basis for adjusting strategies.

[0053] In some embodiments, when performing a weighted average of the first and second order merging probability gains, the weights corresponding to the two order merging probabilities can be dynamically adjusted based on the probability that the target order is an order requiring adjustment of its initial delivery time. Specifically, the weight corresponding to the first order merging probability gain is positively correlated with the probability that the target order is an order that does not require adjustment of its initial delivery time; that is, the higher the probability that the target order is an order that does not require adjustment of its initial delivery time, the larger the weight corresponding to the first order merging probability gain. Conversely, the higher the probability that the target order is an order that requires adjustment of its initial delivery time, the larger the weight corresponding to the second order merging probability gain.

[0054] In some embodiments, the probability that a target waybill requires an adjustment to its initial delivery time can be determined based on the waybill information and the initial delivery time. For example, if the target waybill's delivery destination is relatively remote, it is unlikely to be combined with other waybills, meaning the probability of needing to adjust its initial delivery time is relatively low. Alternatively, if the target waybill's initial delivery time is already relatively long, it is generally preferred not to adjust it to avoid impacting user experience; therefore, the probability of needing to adjust its initial delivery time is also relatively low.

[0055] In some embodiments, to obtain the counterfactual predicted merging probabilities of the first and second types of historical waybills, a merging probability prediction model can be pre-trained to predict the counterfactual merging probabilities of both. Two merging probability prediction models can be trained, one for waybills with uninterrupted initial delivery times and the other for waybills with intervened initial delivery times. For example, as shown in Figure 6(a), a large amount of waybill information for the first type of historical waybills whose initial delivery times have not been adjusted can be collected, and a tag can be determined for each waybill. This tag indicates whether the first type of historical waybill is merged. Then, based on the waybill information of the first type of historical waybills with tags, a pre-set initial model is trained to obtain a second merging probability prediction model. This second merging probability prediction model can be used to predict the merging probability of uninterrupted waybills (i.e., waybills whose initial predicted delivery times have not been adjusted). For the second type of historical waybills, the corresponding predicted uninterrupted merging probability can be predicted using this second merging probability prediction model. For example, the waybill information of the second historical waybill can be input into the second combined waybill probability prediction model to predict the above-mentioned probability of combined waybill without intervention.

[0056] Similarly, a large amount of waybill information for a second type of historical waybill with adjusted initial delivery times can be collected, and a tag can be determined for this second type of historical waybill. This tag indicates whether the second type of historical waybill has been merged. Then, based on the waybill information of the second type of historical waybill with the tag, a preset initial model is trained to obtain a third merging probability prediction model. This third merging probability prediction model can be used to predict the merging probability of intervened waybills (i.e., waybills with adjusted initial predicted delivery times). For the first type of historical waybill, its corresponding predicted intervention merging probability can be predicted by this third merging probability prediction model. For example, the waybill information of the first historical waybill can be input into this third merging probability prediction model to predict the aforementioned predicted intervention merging probability.

[0057] In some embodiments, the merging of historical waybills can also be analyzed. Waybills with similar or identical waybill information and initial delivery time adjustment strategies (i.e., initial delivery time and adjustment value) typically exhibit similar merging patterns. Therefore, historical waybills can be analyzed to determine the merging patterns of waybills with similar or identical waybill information and initial delivery times, distinguishing between those with adjusted initial delivery times and those without. Based on the differences in merging patterns, the merging probability gain corresponding to different adjustment values ​​under specific waybill information and initial delivery times can be determined. This allows for the construction of a mapping relationship between "waybill information + initial delivery time + adjustment value" and the merging probability gain. For example, a large amount of waybill information and merging probabilities for the first type of historical waybill, as well as a large amount of information and merging probabilities for the second type of historical waybill, can be obtained. From this, second-type and first-type historical waybills with identical coding information can be identified. Based on the difference in their merging probabilities, the merging probability gain corresponding to different adjustment values ​​under specific coding information can be obtained. This allows us to construct a mapping relationship between the coded information corresponding to "waybill information + initial delivery time adjustment strategy" and the order consolidation probability gain.

[0058] When determining the order merging probability gain corresponding to the adjustment value based on the initial delivery time and waybill information, the waybill information, initial delivery time, and adjustment value can be encoded to obtain target encoding information. Based on the pre-determined mapping relationship between the encoding information and the order merging probability gain, the order merging probability gain corresponding to the target encoding information is determined as the order merging probability gain corresponding to the adjustment value.

[0059] In some embodiments, the operation of encoding waybill information and the adjustment value to obtain target encoded information, and determining the order consolidation probability gain corresponding to the target encoded information based on a pre-determined mapping relationship between the encoded information of the waybill information and the order consolidation probability gain, can be performed by a pre-trained model. This mapping relationship can be learned by the model from waybill information of multiple historical waybills. For example, a large amount of historical waybill data can be collected, such as various features of historical waybills, whether the initial delivery time of historical waybills has been adjusted, and the adjustment value, etc. This information is input into the model, which can then encode the waybill information and adjustment value of historical waybills to obtain encoded information, and then learn the mapping relationship between different encoded information and the order consolidation probability gain. When the waybill information and adjustment value of the target waybill are input, the model can encode the waybill information and adjustment value of the target waybill, and determine the order consolidation probability gain of the target waybill under that adjustment value based on the mapping relationship between the encoded information and the order consolidation probability gain. By leveraging the principle that "similar or identical waybill information leads to similar order merging situations," cluster analysis is performed on historical waybills to filter out those with similar information (such as merchant location, business district characteristics, and time period). The merging situation (such as merging rate and number of merging) is then statistically analyzed for those with "adjusted initial delivery time" and those without "unadjusted initial delivery time." The difference between these two is used to calculate the merging probability gain corresponding to different adjustment values ​​under specific waybill information, forming a basic mapping of "waybill information - adjustment strategy - merging probability gain."

[0060] Then, the waybill information and adjustment values ​​are encoded (e.g., through hashing, feature embedding, or clustering labels) so that similar waybill information corresponds to the same or similar codes, transforming the basic mapping into a simplified mapping of "coded information - consolidated probability gain". In practical applications, this process can be completed by a pre-trained model, which automatically completes the encoding and learns the mapping relationship by learning the features, adjustment values, and corresponding gains of a large number of historical waybills.

[0061] When it is necessary to determine the probability gain of merging a target waybill, simply input its waybill information and adjustment value into the model. After encoding the input, the model directly calls the pre-stored mapping relationship and outputs the corresponding probability gain of merging. This solution, through the approach of "similarity clustering + encoding mapping," achieves efficient calculation of the probability gain of merging while ensuring a certain level of prediction accuracy, balancing practicality and cost, and providing lightweight and reliable support for delivery time adjustment strategies.

[0062] In some embodiments, the operation of determining the order merging probability gain corresponding to the adjustment value based on the initial delivery time and waybill information can be implemented by a pre-trained third gain prediction model. As shown in Figure 6(b), this third gain prediction model includes an encoding module, a self-interaction module, and an intervention-aware interaction module. The encoding module encodes the waybill information to obtain a first feature, and encodes the initial delivery time and the adjustment value to obtain a second feature. The self-interaction module predicts the non-intervention order merging probability of the target waybill without adjusting the initial delivery time based on the first feature. The intervention-aware interaction module predicts the intervention-merging probability of the target waybill when the initial delivery time is adjusted using the adjustment value, based on the first and second features. Then, it can determine the order merging probability gain based on the difference between the intervention-merging probability and the non-intervention-merging probability. The encoding module can use various networks that can be used for feature extraction from the data. The self-interaction module can learn the intrinsic relationship between waybill information and the order merging probability gain, and the intervention-aware interaction module can learn the intrinsic relationship between different intervention strategies and the order merging probability gain. The third gain prediction model can be trained based on a large amount of first-class historical waybill information and order merging probability, as well as a large amount of second-class historical waybill information and order merging probability.

[0063] In some embodiments, the third gain prediction model is the EFIN (EfficientNet) model.

[0064] In some embodiments, when determining the order merging probability gain corresponding to the adjustment value based on the initial delivery time and waybill information, the waybill features of the target waybill can be extracted from the waybill information first. Then, based on the initial delivery time and waybill features, the order merging probability gain corresponding to the adjustment value is determined. The features of a waybill include one or more of the following: The characteristics of the merchant corresponding to the target order, the characteristics of the rider, the characteristics of the business district where the target order is located, the characteristics representing the relationship between riders and orders within the business district where the target order is located, the characteristics representing the spatiotemporal relationship between the target order and other orders, and the characteristics representing the relationship between riders and business districts.

[0065] By systematically extracting multi-dimensional order features from order information, including merchant features (such as merchant location and food preparation speed), rider features (such as rider real-time location and delivery proficiency), business district features (such as order density and traffic conditions), the correlation between riders and orders within the business district (such as riders' ability to accept orders within the business district), the spatiotemporal correlation between the target order and other orders (such as the overlap of time windows and path similarity between orders in the same area), and the correlation between riders and the business district (such as riders' familiarity with the business district routes), and then combining these with the initial delivery time, the model calculates the order merging probability gain under different adjustment values. These multi-dimensional order features comprehensively capture the complex factors affecting order merging, making the calculation of order merging probability gain more aligned with real-world scenarios and avoiding gain evaluation bias caused by single features. Furthermore, accurately quantifying the impact of adjustment values ​​on the order merging probability based on these features provides a more scientific basis for subsequent adjustment strategy formulation, thereby improving the accuracy of order merging rate prediction and the effectiveness of adjustment strategies, ultimately achieving optimal allocation of delivery resources.

[0066] For example, in some embodiments, in order to more comprehensively capture the factors affecting the consolidated shipment, the shipment feature may include one or more features listed in Table 1 below.

[0067] Table 1 The delivery time determination method described in this specification, by constructing an adjustment value-order consolidation probability gain prediction model in an instant logistics scenario, can more accurately assess the impact of different adjustment strategies on the order consolidation rate. The model can predict the changes in individual effects of a certain intervention (such as adjusting the T value), thereby improving overall scheduling efficiency and reducing costs while ensuring service quality.

[0068] The embodiments in this specification can link scheduling, delivery time, and pricing efficiency deduction capabilities. By extending the predicted delivery time of high-efficiency orders, utilizing time through dynamic scheduling timelines, order consolidation and tracking can be achieved, and cost recovery can be realized through efficiency deduction.

[0069] The experiment was conducted in four cities: one large city with tight supply and demand, one large city with sufficient supply and demand, one small and medium-sized city with tight supply and demand, and one small and medium-sized city with sufficient supply and demand. The results showed significant gains, including a 1.33pt increase in the order completion rate, a 0.13pt increase in the accuracy of prediction T, and a 0.84-point decrease in the average efficiency per order.

[0070] Corresponding to the memory allocation method embodiments provided in this specification, this specification also provides a computer program product, including a computer program that, when executed by a processor, implements the method mentioned in any of the above embodiments.

[0071] This description also provides an electronic device, such as... Figure 7 The diagram shown is a structural schematic of an electronic device according to an embodiment of this specification, except... Figure 7 In addition to the processor 72 and memory 74 shown, the device may also include other hardware, such as a forwarding chip responsible for processing messages; from a hardware structure perspective, the device may also be a distributed device, possibly including multiple interface cards to extend message processing at the hardware level. The memory 74 stores computer instructions, and when the processor 72 executes the computer instructions, it implements the methods mentioned in any of the above embodiments.

[0072] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0073] Since the parts of the embodiments in this specification that contribute to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, the computer software product is stored in a storage medium and includes several instructions to cause a terminal device to execute all or part of the steps of the methods in the various embodiments of this specification. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above description is merely a preferred embodiment of the embodiments of this specification and is not intended to limit the embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this specification should be included within the scope of protection of the embodiments of this specification.

Claims

1. A method for determining delivery time, the method comprising: Obtain the pre-predicted initial delivery time of the target waybill, as well as the waybill information of the target waybill; For each of the preset adjustment values, based on the initial delivery time and the waybill information, the order merging probability gain corresponding to the adjustment value is determined, wherein the order merging probability gain is: the difference between the order merging probability of the target waybill when the initial delivery time is adjusted using the adjustment value and the order merging probability of the target waybill when the initial delivery time is not adjusted. An adjustment strategy for adjusting the initial delivery time is determined based on the order combination probability gain corresponding to each of the multiple adjustment values. The target delivery time for the target waybill is obtained based on the adjustment strategy and the initial delivery time.

2. The method according to claim 1, wherein determining the adjustment strategy for adjusting the initial delivery time based on the order combination probability gain corresponding to each of the plurality of adjustment values ​​includes: If the order combination probability gain corresponding to each of the multiple adjustment values ​​is less than the preset gain, then the adjustment strategy is not to adjust the initial delivery time.

3. The method according to claim 1, wherein determining the adjustment strategy for adjusting the initial delivery time based on the order combination probability gain corresponding to each of the plurality of adjustment values ​​includes: If at least one of the multiple adjustment values ​​corresponds to a merging probability gain greater than or equal to the preset gain, then the change in the predicted merging status of the target waybill is determined by comparing adjusting the initial delivery time using each of the at least one adjustment value with or without adjusting the initial delivery time; the adjustment strategy is determined based on the change in the predicted merging status; wherein, the predicted merging status includes merging and not merging.

4. The method according to claim 3, wherein determining the adjustment strategy based on the changes in the predicted consolidated order status includes: If the predicted order status corresponding to each of the at least one adjustment value remains unchanged, the adjustment strategy is not to adjust the initial delivery time. If any of the at least one adjustment value corresponds to a target adjustment value whose order status changes from non-order to order-merged, then the adjustment strategy is to adjust the initial delivery time using the minimum adjustment value among the target adjustment values.

5. The method according to claim 1, wherein after obtaining the target delivery time of the target waybill based on the adjustment strategy, the method further includes: The target waybill is scheduled based on the target delivery time.

6. The method according to claim 1, wherein determining the order consolidation probability gain corresponding to the adjustment value based on the initial delivery time and the waybill information includes: The waybill information and the initial delivery time are input into a pre-trained first order merging probability prediction model, and the non-intervention order merging probability of the target waybill is predicted through the first order merging probability prediction model. The waybill information and the delivery time after adjusting the initial delivery time using the adjustment value are input into the first order merging probability prediction model, and the intervention merging probability of the target waybill is predicted through the first order merging probability prediction model. The difference between the probability of order merging with intervention and the probability of order merging without intervention is taken as the order merging probability gain; The first order merging probability prediction model is trained based on the order information of the first type of historical waybills with tags and the second type of historical waybills with tags. The first type of historical waybills are historical waybills whose initial delivery time has not been adjusted. The tags of the first type of historical waybills are used to indicate whether the first type of historical waybills are merged. The second type of historical waybills are historical waybills whose initial delivery time has been adjusted. The tags of the second type of historical waybills are used to indicate whether the second type of historical waybills are merged.

7. The method according to claim 1, wherein determining the order consolidation probability gain corresponding to the adjustment value based on the initial delivery time and the waybill information includes: The waybill information and the initial delivery time are input into a pre-trained first gain prediction model, and the first order merging probability gain is predicted by the first gain prediction model. The first gain prediction model is trained based on the waybill information of a first type of historical waybill with a tag. The first type of historical waybill is a historical waybill whose initial delivery time has not been adjusted. The tag of the first type of historical waybill is the difference between the predicted intervention merging probability and the actual merging probability of the first type of historical waybill. The predicted intervention merging probability is the predicted merging probability of the first type of historical waybill when the initial delivery time of the first type of historical waybill is adjusted. The waybill information, the initial delivery time, and the adjustment value are input into a pre-trained second gain prediction model, and the second gain prediction model is used to predict the second order merging probability gain. The second gain prediction model is trained based on waybill information of a second type of historical waybill with a tag. The second type of historical waybill is a historical waybill with an adjusted initial delivery time. The tag of the second type of historical waybill is the difference between the actual order merging probability and the predicted order merging probability without intervention. The predicted order merging probability without intervention is the predicted order merging probability of the second type of historical waybill without adjusting the initial delivery time. The first and second combined probability gains are weighted and averaged to obtain the combined probability gain.

8. The method according to claim 7, wherein the predicted intervention probability of order merging is predicted by a pre-trained second order merging probability prediction model, the second order merging probability prediction model being trained based on the order information of the first type of historical waybills carrying a tag, the tag being used to indicate whether the historical waybill is merged; The predicted probability of uninterrupted order merging is obtained by a pre-trained third order merging probability prediction model, which is trained based on the order information of the second type of historical waybills with tags. These tags are used to indicate whether the second type of historical waybills have been merged.

9. The method according to claim 7, wherein the weight corresponding to the first order probability gain is positively correlated with the probability that the target waybill is a waybill that does not require adjustment of the initial delivery time; The weight corresponding to the second order probability gain is positively correlated with the probability that the target order is an order that needs to adjust the initial delivery time.

10. The method according to claim 1, wherein determining the order consolidation probability gain corresponding to the adjustment value based on the initial delivery time and the waybill information includes: The waybill information, the initial delivery time, and the adjustment value are encoded to obtain target encoding information; Based on the predetermined mapping relationship between the encoded information and the order combination probability gain, the order combination probability gain corresponding to the target encoded information is determined as the order combination probability gain corresponding to the adjustment value. The mapping relationship is constructed based on the difference between the merging probability of the second type of historical waybills and the merging probability of the first type of historical waybills. The waybill information and the coding information corresponding to the adjustment strategy of the initial delivery time of the second type of historical waybills and the first type of historical waybills are the same. The first type of historical waybills are historical waybills whose initial delivery time has not been adjusted, and the second type of historical waybills are historical waybills whose initial delivery time has been adjusted.

Images