Delivery order quantity adjustment method, device, storage medium and electronic device

Abstract

The present disclosure relates to a method and device for adjusting the amount of delivery orders, a storage medium and an electronic device. The method comprises: in response to receiving an order amount adjustment instruction, obtaining delivery characteristic information of a regional delivery personnel and incentive configuration information corresponding to the delivery personnel; inputting the delivery characteristic information and the incentive configuration information into a single quantity prediction model to obtain order quantity information of the delivery personnel under the corresponding incentive measures output by the single quantity prediction model; if the difference between the second order quantity and the first order quantity reaches a target threshold, the incentive configuration information is taken as target incentive configuration information; and adjusting the delivery order quantity of the region according to the target incentive configuration information. Thus, the target incentive configuration information can be determined based on the first order quantity and the second order quantity, and the delivery order quantity of the region can be adjusted according to the target incentive configuration information, thereby solving the problem of insufficient delivery capacity.

Description

Technical Field

[0001] The present disclosure relates to the field of information technology, and in particular to a method, device, storage medium, and electronic device for adjusting the delivery order quantity. Background Art

[0002] Takeout, as a convenient dining option, has experienced significant growth in recent years. Generally, this can be achieved by assigning a specific number of delivery drivers to a specific area to meet the order delivery needs. However, due to the supply and demand constraints of delivery platforms, the takeout delivery process is still prone to insufficient delivery resources, which affects delivery efficiency.

[0003] In relevant scenarios, incentives can be implemented to motivate limited delivery personnel to deliver more orders, thereby addressing the issue of insufficient delivery capacity. However, accurate data on order volume after implementing incentives is difficult to obtain using relevant technologies, making it difficult to implement incentives to address this issue, leading to continued low order delivery efficiency. Summary of the Invention

[0004] The purpose of the present disclosure is to provide a method, device, storage medium and electronic device for adjusting the delivery order quantity to solve the above-mentioned related technical problems.

[0005] To achieve the above objectives, a first aspect of an embodiment of the present disclosure provides a method for adjusting a delivery order quantity, the method comprising:

[0006] In response to receiving a delivery order quantity adjustment instruction, obtaining delivery characteristic information of a regional delivery person and incentive configuration information corresponding to the delivery person;

[0007] Inputting the delivery characteristic information and the incentive configuration information into an order quantity prediction model, and obtaining, as output by the order quantity prediction model, information on the order quantity of the delivery person under the corresponding incentive measures; wherein the order quantity prediction model includes a first neural network layer for predicting a first order quantity of the delivery person without the incentive measures based on the delivery characteristic information, and a second neural network layer for predicting a second order quantity of the delivery person under the corresponding incentive measures based on the first order quantity and the incentive configuration information;

[0008] If the difference between the second order quantity and the first order quantity reaches a target threshold, the incentive configuration information is used as target incentive configuration information;

[0009] The delivery order quantity of the area is adjusted according to the target incentive configuration information.

[0010] The above technical solution can determine the first order volume of a delivery driver without incentives through the first neural network layer in the order volume prediction model based on the delivery driver's delivery characteristic information. In this way, the second neural network can output the second order volume of the delivery driver with incentives based on the first order volume and the delivery driver's incentive information. In other words, using the above technical solution, the second order volume can be determined based on the incentive characteristics and the first order volume, thereby improving the accuracy of the second order volume. Furthermore, the above technical solution can also output the delivery driver's order volume with incentives and without incentives simultaneously through the order volume prediction model, thereby determining the change in order volume caused by the incentives. Furthermore, because the above technical solution outputs the delivery driver's order volume with incentives and without incentives through the order volume prediction model, it can avoid the cumulative error caused by separately predicting the order volume with incentives and without incentives when determining the change in order volume caused by incentives, thereby improving accuracy. Therefore, the target incentive configuration information can be determined based on the first order volume and the second order volume, and the delivery order volume of the area can be adjusted according to the target incentive configuration information, thereby solving the problem of insufficient delivery capacity.

[0011] Optionally, during the training process of the order quantity prediction model, constraints are set for the second neural network layer, and the constraints are used to ensure that the second order quantity output by the second neural network layer is not less than the first order quantity output by the first neural network layer.

[0012] In this way, by setting the constraint conditions, it can be ensured that the second order quantity output by the order quantity prediction model is not less than the first order quantity, thereby improving the accuracy of the output order quantity.

[0013] Optionally, the order quantity prediction model is obtained in the following manner:

[0014] Obtaining historical information of the delivery personnel, the historical information including historical delivery information and historical incentive information of the delivery personnel with incentive measures, and historical delivery information of the delivery personnel without incentive measures, the historical delivery information including the historical number of orders completed by the delivery personnel;

[0015] Constructing model training samples based on the historical information, the training samples include first-category samples, second-category samples, and third-category samples, the first-category samples being samples corresponding to delivery personnel without incentives, the second-category samples being samples corresponding to delivery personnel with incentives but not receiving rewards, and the third-category samples being samples corresponding to delivery personnel with incentives and receiving rewards;

[0016] The single quantity prediction model is obtained by training according to the training samples.

[0017] The above technical solution can train the order quantity prediction model based on the historical information of the delivery personnel, providing a basis for estimating the order quantity data of the delivery personnel.

[0018] Optionally, the loss function of the single quantity prediction model includes at least one of a first loss function, a second loss function, and a third loss function;

[0019] The first loss function is used to ensure that the absolute average error between the true value and the predicted value of the first type of samples is minimized;

[0020] The second loss function is used to ensure that the absolute average error between the true value and the predicted value of the second type of samples is minimized;

[0021] The third loss function is used to ensure that the absolute average error between the true value and the predicted value of the third type of samples is minimized.

[0022] Optionally, the loss function of the single quantity prediction model is:

[0023]

[0024] Among them, loss1 is the first loss function, loss2 is the second loss function, loss3 is the third loss function, MAE is the mean absolute error, label1 is the historical order volume completed by the deliveryman corresponding to the first category of samples, label2 is the historical order volume completed by the deliveryman corresponding to the second category of samples, label3 is the historical order volume completed by the deliveryman corresponding to the third category of samples, y is the order volume of the corresponding deliveryman output by the order volume prediction model, μ is the search parameter, reduce_mean() is the incentive loss, threshold1 is the first order volume threshold required to obtain the lowest reward value in the incentive configuration information, and threshold_act is the second order volume threshold required to obtain the highest reward value in the incentive configuration information.

[0025] By setting a variety of different loss functions, different loss functions can be used to calculate loss values ​​for samples of different categories during the training process of the single quantity prediction model, so that the weights of the single quantity prediction model can be adjusted according to the loss values, ultimately improving the prediction accuracy of the single quantity prediction model.

[0026] Optionally, constructing a model training sample based on the historical information includes:

[0027] Based on the historical delivery information and historical incentive information of the delivery personnel with incentive measures, new incentive information is constructed;

[0028] For the second type of samples, the winning threshold of the new incentive information is greater than the winning threshold of the historical incentive information and / or the reward amount of the new incentive information is less than the reward amount of the historical incentive information;

[0029] For the third type of samples, the award threshold of the new incentive information is lower than the award threshold of the historical incentive information and / or the reward amount of the new incentive information is higher than the reward amount of the historical incentive information;

[0030] The training sample is constructed according to the historical information and the new stimulus information.

[0031] That is to say, by adopting the above technical solution, the training samples can be expanded, thereby increasing the number of training samples and helping to improve the accuracy of the model.

[0032] Optionally, there are multiple pieces of incentive configuration information, and the step of inputting the delivery characteristic information and the incentive configuration information into an order quantity prediction model to obtain the order quantity information of the delivery person under the corresponding incentive measures output by the order quantity prediction model includes:

[0033] For each incentive configuration information, input the delivery characteristic information and the incentive configuration information into an order quantity prediction model to obtain the order quantity information of the delivery person under the incentive measure output by the order quantity prediction model;

[0034] If the difference between the second order quantity and the first order quantity reaches a target threshold, using the incentive configuration information as target incentive configuration information includes:

[0035] The incentive configuration information corresponding to the maximum difference between the second order quantity and the first order quantity is used as the target incentive configuration information.

[0036] In the above technical solution, the incentive configuration information corresponding to the maximum difference between the second order volume and the first order volume is used as the target incentive configuration information. Thus, adjusting the delivery order volume in the region based on this target incentive configuration information can achieve a greater increase in the delivery order volume in the region, thereby resolving the problem of insufficient regional delivery capacity.

[0037] According to a second aspect of the present disclosure, a device for adjusting a delivery order quantity is provided, the device comprising:

[0038] A first acquisition module is configured to acquire, in response to receiving a delivery order quantity adjustment instruction, delivery characteristic information of a regional delivery person and incentive configuration information corresponding to the delivery person;

[0039] An input module is configured to input the delivery characteristic information and the incentive configuration information into an order quantity prediction model, and obtain, as output by the order quantity prediction model, order quantity information of the delivery person under the corresponding incentive measures; wherein the order quantity prediction model includes a first neural network layer for predicting a first order quantity of the delivery person without the incentive measures based on the delivery characteristic information, and a second neural network layer for predicting a second order quantity of the delivery person under the corresponding incentive measures based on the first order quantity and the incentive configuration information;

[0040] a first execution module configured to use the incentive configuration information as target incentive configuration information if the difference between the second order quantity and the first order quantity reaches a target threshold;

[0041] The adjustment module is configured to adjust the delivery order quantity of the area according to the target incentive configuration information.

[0042] The above technical solution can determine the first order volume of a delivery driver without incentives through the first neural network layer in the order volume prediction model based on the delivery driver's delivery characteristic information. In this way, the second neural network can output the second order volume of the delivery driver with incentives based on the first order volume and the delivery driver's incentive information. In other words, using the above technical solution, the second order volume can be determined based on the incentive characteristics and the first order volume, thereby improving the accuracy of the second order volume. Furthermore, the above technical solution can also output the delivery driver's order volume with incentives and without incentives simultaneously through the order volume prediction model, thereby determining the change in order volume caused by the incentives. Furthermore, because the above technical solution outputs the delivery driver's order volume with incentives and without incentives through the order volume prediction model, it can avoid the cumulative error caused by separately predicting the order volume with incentives and without incentives when determining the change in order volume caused by incentives, thereby improving accuracy. Therefore, the target incentive configuration information can be determined based on the first order volume and the second order volume, and the delivery order volume of the area can be adjusted according to the target incentive configuration information, thereby solving the problem of insufficient delivery capacity.

[0043] Optionally, it also includes:

[0044] The second execution module is configured to set constraints for the second neural network layer during the training process of the order quantity prediction model, and the constraints are used to ensure that the second order quantity output by the second neural network layer is not less than the first order quantity output by the first neural network layer.

[0045] Optionally, the device further comprises:

[0046] The second acquisition module is configured to obtain the order quantity prediction model, and the second acquisition module includes:

[0047] an acquisition submodule configured to acquire historical information of delivery personnel, the historical information including historical delivery information and historical incentive information corresponding to delivery personnel with incentive measures, and historical delivery information corresponding to delivery personnel without incentive measures, the historical delivery information including the historical number of orders completed by the delivery personnel;

[0048] a creation submodule configured to construct model training samples based on the historical information, the training samples including first-category samples, second-category samples, and third-category samples, the first-category samples being samples corresponding to delivery personnel without incentives, the second-category samples being samples corresponding to delivery personnel with incentives but without rewards, and the third-category samples being samples corresponding to delivery personnel with incentives and rewards;

[0049] The training submodule is configured to obtain the single quantity prediction model based on the training samples.

[0050] Optionally, the loss function of the single quantity prediction model includes at least one of a first loss function, a second loss function, and a third loss function;

[0051] The first loss function is used to ensure that the absolute average error between the true value and the predicted value of the first type of samples is minimized;

[0052] The second loss function is used to ensure that the absolute average error between the true value and the predicted value of the second type of samples is minimized;

[0053] The third loss function is used to ensure that the absolute average error between the true value and the predicted value of the third type of samples is minimized.

[0054] Optionally, the loss function of the single quantity prediction model is:

[0055]

[0056] Among them, loss1 is the first loss function, loss2 is the second loss function, loss3 is the third loss function, MAE is the mean absolute error, label1 is the historical order volume completed by the deliveryman corresponding to the first category of samples, label2 is the historical order volume completed by the deliveryman corresponding to the second category of samples, label3 is the historical order volume completed by the deliveryman corresponding to the third category of samples, y is the order volume of the corresponding deliveryman output by the order volume prediction model, μ is the search parameter, reduce_mean() is the incentive loss, threshold1 is the first order volume threshold required to obtain the lowest reward value in the incentive configuration information, and threshold_act is the second order volume threshold required to obtain the highest reward value in the incentive configuration information.

[0057] Optionally, the creating submodule includes:

[0058] A first creation subunit is configured to construct new incentive information based on historical delivery information and historical incentive information of the delivery personnel with incentive measures;

[0059] For the second type of samples, the winning threshold of the new incentive information is greater than the winning threshold of the historical incentive information and / or the reward amount of the new incentive information is less than the reward amount of the historical incentive information;

[0060] For the third type of samples, the award threshold of the new incentive information is lower than the award threshold of the historical incentive information and / or the reward amount of the new incentive information is higher than the reward amount of the historical incentive information;

[0061] The second creating subunit is configured to construct the training sample according to the historical information and the new stimulus information.

[0062] Optionally, the incentive configuration information is multiple, and the input module includes:

[0063] An input submodule is configured to input the delivery characteristic information and the incentive configuration information into an order quantity prediction model for each incentive configuration information, and obtain the order quantity information of the delivery person under the incentive measure output by the order quantity prediction model;

[0064] The first execution module includes:

[0065] The first execution submodule is configured to use the incentive configuration information corresponding to the maximum difference between the second order quantity and the first order quantity as the target incentive configuration information.

[0066] According to a third aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of the method described in any one of the first aspects are implemented.

[0067] According to a fourth aspect of the present disclosure, an electronic device is provided, including:

[0068] a memory having a computer program stored thereon;

[0069] A processor is used to execute the computer program in the memory to implement the steps of any one of the methods described in the first aspect above.

[0070] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following detailed description, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0072] Figure 1 It is a schematic diagram of a single quantity prediction shown in an exemplary embodiment of the present disclosure.

[0073] Figure 2 It is a flowchart of a method for adjusting the delivery order quantity shown in an exemplary embodiment of the present disclosure.

[0074] Figure 3 It is a schematic diagram of a single quantity prediction model shown in an exemplary embodiment of the present disclosure.

[0075] Figure 4 It is a training flowchart of a single quantity prediction model shown in an exemplary embodiment of the present disclosure.

[0076] Figure 5 It is a block diagram of a device for adjusting the delivery order quantity shown in an exemplary embodiment of the present disclosure.

[0077] Figure 6 It is a block diagram of an electronic device shown in an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0078] The following describes the specific embodiments of the present disclosure in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present disclosure and are not intended to limit the present disclosure.

[0079] Before introducing the delivery order quantity adjustment method, device, storage medium, and electronic device provided in this disclosure, we first introduce the application scenarios of the various embodiments provided in this disclosure. The various embodiments of this disclosure can be applied to various delivery order quantity adjustment scenarios. The order quantity can refer to the number of orders delivered by a delivery person within a certain time interval. The orders can include, for example, takeout orders, logistics orders, and so on.

[0080] Taking takeout delivery as an example, in relevant scenarios, delivery personnel can be assigned to corresponding delivery areas to meet the delivery needs of orders in those areas. However, orders often have distinct temporal and spatial distribution characteristics. Therefore, the delivery process of takeout orders in such scenarios is still prone to insufficient delivery resources, resulting in longer delivery times and affecting delivery efficiency.

[0081] Threshold 1 Reward value 1 Threshold 2 Reward value 2 Threshold 3 Reward value 3 Threshold 4 Reward value 4 5 6 7 12 12 20 18 30

[0082] Table 1

[0083] As shown in Table 1, in some implementation scenarios, multiple thresholds can be set for delivery personnel, and each threshold is set with a corresponding reward value. The threshold can be the delivery personnel's order volume, and the reward value can be, for example, the corresponding reward amount. That is, when the delivery personnel's delivery order volume reaches threshold 1 (i.e., 5 orders), the delivery personnel can be rewarded 6 yuan. When the delivery personnel's delivery order volume reaches threshold 2 (i.e., 7 orders), the delivery personnel can be rewarded 12 yuan, thereby incentivizing the delivery personnel to deliver more orders and thus improving the problem of insufficient delivery capacity.

[0084] However, it is still difficult to obtain accurate order volume data after the adoption of incentives in relevant scenarios. Figure 1 A schematic diagram of order quantity prediction is shown. In some implementation scenarios, the non-incentivized order quantity of the deliveryman under non-incentivized conditions and the incentivized order quantity under incentivized conditions can be predicted based on two models. In this way, the change in order quantity caused by the incentive measures can be determined through the non-incentivized order quantity and the incentivized order quantity, so that the effect of the incentive measures can be determined based on the change in order quantity, and then the incentive measures can be adjusted accordingly to solve the problem of insufficient delivery capacity. The applicant found that the above-mentioned implementation method still has the problem of large errors in the prediction results. Therefore, the order quantity obtained by the above-mentioned implementation method is easy to mislead the effectiveness evaluation process of the incentive measures, and it is difficult to improve the problem of insufficient delivery capacity.

[0085] To this end, the present disclosure provides a method for adjusting the delivery order quantity, referring to Figure 2 A flow chart of a method for adjusting a delivery order quantity is shown, the method comprising:

[0086] S21, in response to receiving a delivery order quantity adjustment instruction, obtaining delivery characteristic information of a regional delivery person and incentive configuration information corresponding to the delivery person.

[0087] For example, the method can be applied to a server, which can be used to monitor and manage regional delivery orders. In some embodiments, the delivery order quantity adjustment instruction can be generated by the server in response to an operator's order quantity adjustment operation. In other embodiments, the delivery order quantity adjustment instruction can also be sent by other control terminals. For example, the server can receive the delivery order quantity adjustment instruction from a backend control terminal and then obtain the delivery characteristic information of the regional delivery personnel and the corresponding incentive configuration information of the delivery personnel.

[0088] With respect to the delivery characteristic information, the delivery characteristic information of the deliveryman may include, for example, the deliveryman's identity information, transportation information, years of service of the deliveryman, and the like. In some embodiments, the deliveryman's characteristic information may also include characteristic information of the geographical area where the deliveryman is located, such as weather information, road traffic information, and the like. Referring to Table 1, the deliveryman incentive configuration information may include multiple thresholds and reward values ​​corresponding to each of the thresholds. Among them, the threshold may be, for example, the number of orders of the deliveryman, and different thresholds may be obtained by dividing the number of orders of the deliveryman into steps. The reward value may be, for example, an amount corresponding to each of the thresholds. Of course, in some embodiments, the threshold and reward value may also be presented in other ways, for example, the reward value may also correspond to a physical reward, and the present disclosure does not limit the form of the threshold and the reward value.

[0089] S22, inputting the delivery characteristic information and the incentive configuration information into an order quantity prediction model, and obtaining the order quantity information of the delivery person under the corresponding incentive measures output by the order quantity prediction model.

[0090] Among them, the order volume prediction model can be implemented based on, for example, an MLP (Multilayer Perceptron). The order volume prediction model includes a first neural network layer for predicting a first order volume of the delivery person without incentive measures based on the delivery feature information, and a second neural network layer for predicting a second order volume of the delivery person under the corresponding incentive measures based on the first order volume and the incentive configuration information. For example, the first neural network layer can output the first order volume of the delivery person based on the input delivery person features (such as the number of delivery orders of the delivery person, delivery duration, etc.), weather features (such as general, medium, bad, etc.), supply-demand features (such as the order volume data compared with the same period in history), competition features (such as the regional market share ratio, etc.), as well as the delivery person cross features and delivery person ID features obtained based on the above features. Thus, the second neural network layer can determine the second order volume of the current delivery person based on the first order volume and the incentive configuration information of the current delivery person.

[0091] Referring to Figure 3 the schematic diagram of an order volume prediction model shown, in some embodiments, the first neural network layer can also generate an elastic hidden vector W based on the delivery feature information and historical incentive features i , (in the figure, 0 < i < 9 and i is an integer is示意). Thus, the second neural network layer can also determine the second order volume of the current delivery person based on the first order volume, the elastic hidden vector, and the incentive configuration information of the current delivery person. For example, the incentive information vector T can be determined according to the incentive configuration information i , taking Table 1 as an example, the incentive information vector T i can correspond to different thresholds and the reward values corresponding to the thresholds (i.e., 0 < i < 9 and i is an integer). For example, T1 can correspond to threshold 1, and T2 can correspond to the reward value 1 corresponding to threshold 1. Thus, the second order volume can be determined based on the first order volume, the incentive information vector T i and the elastic hidden vector W i .

[0092] Of course, for the order volume prediction model, those skilled in the art should also know that the order volume prediction model can also be implemented according to other neural network structures in specific implementations, and the present disclosure does not limit this.

[0093] In step S23, if the difference between the second order volume and the first order volume reaches a target threshold, the incentive configuration information can be used as the target incentive configuration information. The target threshold can be determined based on the historical delivery order statistics for the region. For example, the order delivery gap for the region can be determined based on the historical delivery order statistics for the region. The order delivery gap can be described as the difference between the number of orders generated in the region during a time interval (e.g., 10:00 to 14:00) and the number of orders delivered by the delivery personnel during that time interval.

[0094] It should be understood that when the difference between the second order volume and the first order volume is greater than the order volume gap, it can be understood that under this incentive measure, the number of orders delivered within the time interval using the same delivery capacity is greater than the number of delivery orders generated in the region within the time interval. In other words, the incentive measure can solve the problem of insufficient delivery capacity in the region.

[0095] Therefore, in step S24, the delivery order quantity of the region may be adjusted according to the target incentive configuration information, thereby solving the problem of insufficient delivery capacity in the region.

[0096] The above technical solution can determine the first order volume of a delivery driver without incentives through the first neural network layer in the order volume prediction model based on the delivery driver's delivery characteristic information. In this way, the second neural network can output the second order volume of the delivery driver with incentives based on the first order volume and the delivery driver's incentive information. In other words, using the above technical solution, the second order volume can be determined based on the incentive characteristics and the first order volume, thereby improving the accuracy of the second order volume. Furthermore, the above technical solution can also output the delivery driver's order volume with incentives and without incentives simultaneously through the order volume prediction model, thereby determining the change in order volume caused by the incentives. Furthermore, because the above technical solution outputs the delivery driver's order volume with incentives and without incentives through the order volume prediction model, it can avoid the cumulative error caused by separately predicting the order volume with incentives and without incentives when determining the change in order volume caused by incentives, thereby improving accuracy. Therefore, the target incentive configuration information can be determined based on the first order volume and the second order volume, and the delivery order volume of the area can be adjusted according to the target incentive configuration information, thereby solving the problem of insufficient delivery capacity.

[0097] It is worth noting that in some possible implementations, the incentive configuration information may also be multiple. Taking Table 1 as an example, by setting different thresholds and reward values, multiple incentive configuration information can be obtained. In this case, the delivery characteristic information and the incentive configuration information are input into the order volume prediction model to obtain the order volume information of the delivery person under the corresponding incentive measures output by the order volume prediction model, including:

[0098] For each incentive configuration information, the delivery characteristic information and the incentive configuration information are input into the order quantity prediction model to obtain the order quantity information of the delivery person under the incentive measure output by the order quantity prediction model.

[0099] That is, for each of the incentive configuration information, the second waybill quantity under the incentive configuration information can be determined by the order quantity prediction model. In this way, the target incentive configuration information can be determined based on each of the second waybill quantity and the first waybill quantity.

[0100] For example, in some embodiments, if the difference between the second order quantity and the first order quantity reaches a target threshold, using the incentive configuration information as target incentive configuration information includes:

[0101] The incentive configuration information corresponding to the maximum difference between the second order quantity and the first order quantity is used as the target incentive configuration information.

[0102] In the above technical solution, the incentive configuration information corresponding to the maximum difference between the second order volume and the first order volume is used as the target incentive configuration information. Thus, adjusting the delivery order volume in the region based on this target incentive configuration information can achieve a greater increase in the delivery order volume in the region, thereby further increasing the order delivery volume in the region and resolving the problem of insufficient regional delivery capacity.

[0103] Of course, in other embodiments, the target incentive configuration information may also be determined based on the difference between the second order quantity and the first order quantity, as well as other factors. For example, the target incentive configuration information may be determined comprehensively by considering the budget value corresponding to the incentive configuration information and assigning corresponding weights to the budget value and the difference between the second order quantity and the first order quantity. This disclosure is not limited to this.

[0104] In one possible embodiment, during the training process of the order quantity prediction model, constraints are set for the second neural network layer, and the constraints are used to ensure that the second order quantity output by the second neural network layer is not less than the first order quantity output by the first neural network layer.

[0105] For example, the constraint can be Where F2 is the second order quantity, F1 is the first order quantity, and ReLU is the function. N is the number of thresholds and reward values. Taking Table 1 as an example, Table 1 includes 4 thresholds and 4 reward values, then N can be 8. i The elastic latent vector generated by the first neural network is i Corresponding. T i is an excitation information vector generated according to the excitation information, wherein the excitation information vector T i It can correspond to different thresholds and the reward values ​​corresponding to the thresholds. Taking Table 1 as an example, T1 can correspond to threshold 1, and T2 can correspond to the reward value 1 corresponding to threshold 1. In this way, W i With t i The linear weighting is performed by multiplication and the output is obtained through the ReLU function. It is worth noting that the output value of the ReLU function is a non-negative number, so it can ensure that the second order amount is greater than or equal to the first order amount.

[0106] That is to say, by setting the constraint conditions, it is possible to ensure that the second order quantity output by the order quantity prediction model is not less than the first order quantity, thereby improving the accuracy of the output order quantity.

[0107] The above embodiment is based on the constraint condition However, those skilled in the art should know that other constraints can be used in specific implementations to ensure that the second order quantity output by the order quantity prediction model is not less than the first order quantity. For example, the constraint can also be Etc., the present disclosure does not limit this.

[0108] Optionally, refer to Figure 4 The training flow chart of a single quantity prediction model is shown. The single quantity prediction model can be obtained in the following manner:

[0109] In step S41, the historical information of the delivery person is obtained.

[0110] The delivery personnel may include delivery personnel with incentives and delivery personnel without incentives, and the historical information may include the delivery personnel's historical delivery information and historical incentive information. For example, the historical information may include the delivery personnel's historical delivery information and historical incentive information for delivery personnel with incentives, and the delivery personnel's historical delivery information for delivery personnel without incentives, and the historical delivery information may include the historical number of orders completed by the delivery personnel.

[0111] Of course, the delivery personnel with incentives may also include delivery personnel who have received rewards and delivery personnel who have not received rewards. Therefore, in step S42, a model training sample may be constructed based on the historical information. The training samples include first-category samples, second-category samples, and third-category samples. The first-category samples are samples corresponding to delivery personnel without incentives, the second-category samples are samples corresponding to delivery personnel with incentives but without rewards, and the third-category samples are samples corresponding to delivery personnel with incentives and rewards.

[0112] In this way, in step S43, the single quantity prediction model can be obtained by training according to the training samples.

[0113] For example, the delivery driver's historical information can be input into the order volume prediction model to obtain the order volume output by the model. In this way, a corresponding loss function can be used to describe the difference between the order volume in the historical information and the output order volume. The relevant weights in the order volume prediction model can be adjusted based on the function value of the loss function, thereby effectively training the order volume prediction model.

[0114] It is also worth noting that the training samples can be processed before training. For example, the historical information of the delivery person can be mined to obtain cross-features and ID-type features corresponding to the delivery person. Cross-features can include, for example, the delivery person's waybill type (such as high / low value of delivery orders, long / short delivery distance, etc.), attendance preferences (midweek / weekend, peak hours, weather, etc.), etc. ID-type features can include, for example, the delivery person's demographic attributes, delivery person level, income data, etc.

[0115] Furthermore, corresponding encoding can be performed on the cross-features and ID-based features. For example, one-hot encoding can be performed on features such as regional features and delivery driver levels. For features with multiple categories, such as city IDs, embedding can be performed to reduce the difficulty of feature processing.

[0116] The above technical solution can train the order quantity prediction model based on the historical information of the delivery personnel, providing a basis for estimating the order quantity data of the delivery personnel.

[0117] Regarding the loss function, in a possible implementation, the loss function of the single quantity prediction model includes a first loss function, and the first loss function is used to limit the absolute average error between the true value and the predicted value of the first category of samples to be minimum.

[0118] For example, the first loss function can be loss1 = MAE(label1 - y), where MAE is the mean absolute error, label1 is the historical number of orders completed by the delivery person corresponding to the first type of sample, and y is the order volume of the corresponding delivery person as output by the order volume prediction model. By setting the first loss function, the order volume prediction model can be trained, thereby improving the accuracy of the model's prediction of delivery orders for delivery persons without incentives.

[0119] In one embodiment, the loss function of the single quantity prediction model includes a second loss function, and the second loss function is used to limit the absolute average error between the true value and the predicted value of the second category of samples to be minimized.

[0120] For example, the second loss function can be loss2 = MAE(label2-y)-μ*reduce_mean(threshold1-y), where MAE is the mean absolute error, label2 is the historical order volume completed by the delivery person corresponding to the second type of sample, y is the order volume of the corresponding delivery person output by the order volume prediction model, μ is a search parameter, for example, it can take values ​​of 0.1, 0.5, etc., and reduce_mean() is the incentive loss. threshold1 is the first order volume threshold required to obtain the minimum reward value in the incentive configuration information. Taking Table 1 as an example, the minimum reward value is the reward value 1, so the first order volume threshold is the first threshold. In this way, by setting the second loss function, when the predicted order quantity value for the delivery personnel in the second category of samples is high, for example, exceeding the threshold1, μ*reduce_mean(threshold1-y is a negative number, which increases the loss value in this case. In other words, through the above technical solution, the order quantity prediction model can be trained, thereby improving the prediction accuracy of the order quantity prediction model for delivery personnel who have incentives but have not received rewards.

[0121] In another embodiment, the single quantity prediction model may also include a third loss function, and the third loss function is used to limit the absolute average error between the true value and the predicted value of the third category of samples to be minimized.

[0122] For example, the third loss function is loss3 = MAE (label3-y) - μ * reduce_mean (y-threshold_act, where MAE is the absolute mean error, label3 is the historical order volume completed by the deliveryman corresponding to the third type of sample, y is the order volume of the corresponding deliveryman output by the order volume prediction model, μ is the search parameter, for example, it can take values ​​of 0.1, 0.5, etc., and reduce_mean() is the incentive loss. threshold_act is the second order volume threshold required to obtain the highest reward value in the incentive configuration information. Referring to Table 1, when the deliveryman has 6 delivery orders, the highest reward value he can obtain is reward value 1, and the reward value 1 corresponds to the threshold 1, so the threshold_act is threshold 1. Similarly, if the delivery person has 15 delivery orders, the highest reward value he can obtain is reward value 3, and the reward value 3 corresponds to threshold 3, so the threshold_act is threshold 3. In this way, by setting the third loss function, when the predicted order value for the delivery person in the third category of samples is low, for example, less than the threshold_act, then μ*reduce_mean(y-threshold_act) is a negative number, which increases the loss value in this case. In other words, through the above technical solution, the order quantity prediction model can be trained, and then the prediction accuracy of the order quantity prediction model for delivery persons who have incentives and receive rewards can be improved.

[0123] It is worth noting that the "minimum" mentioned in the above definitions of the loss functions can, in some scenarios, refer to the absolute average error between the true and predicted values ​​of the corresponding samples being less than a certain threshold. In some scenarios, it can also refer to the minimum value during multiple iterations of model training.

[0124] In addition, in the above embodiment, the loss function of the single quantity prediction model is described separately to facilitate understanding by those skilled in the art. However, those skilled in the art should be aware that this solution may include at least one of the first loss function, the second loss function, and the third loss function described above during specific implementation.

[0125] For example, in one possible implementation, the loss function of the single quantity prediction model is:

[0126]

[0127] Among them, loss1 is the first loss function, loss2 is the second loss function, loss3 is the third loss function, MAE is the mean absolute error, label1 is the historical order volume completed by the deliveryman corresponding to the first category of samples, label2 is the historical order volume completed by the deliveryman corresponding to the second category of samples, label3 is the historical order volume completed by the deliveryman corresponding to the third category of samples, y is the order volume of the corresponding deliveryman output by the order volume prediction model, μ is the search parameter, reduce_mean() is the incentive loss, threshold1 is the first order volume threshold required to obtain the lowest reward value in the incentive configuration information, and threshold_act is the second order volume threshold required to obtain the highest reward value in the incentive configuration information.

[0128] In this way, by setting a variety of different loss functions, during the training process of the single quantity prediction model, different loss functions can be used to calculate the loss values ​​for samples of different categories, so that the weights of the single quantity prediction model can be adjusted according to the loss values, ultimately improving the prediction accuracy of the single quantity prediction model.

[0129] The applicant discovered that, in the historical information of delivery drivers, the number of samples corresponding to active delivery drivers is smaller than the number of samples corresponding to inactive delivery drivers. Therefore, the samples of active delivery drivers can be amplified. Thus, in one possible implementation, when obtaining sample data of active delivery drivers, the sample data of active delivery drivers can be resampled based on the historical data, thereby amplifying the number of samples of active delivery drivers, thereby helping to improve the accuracy of the model.

[0130] Of course, in another possible implementation, the number of samples of active delivery personnel can be expanded by constructing new samples. Thus, the model training samples constructed based on the historical information include:

[0131] For the historical delivery information and historical incentive information of the delivery personnel with incentive measures, new incentive information is constructed.

[0132] For the second type of samples, the winning threshold of the new incentive information is greater than the winning threshold of the historical incentive information and / or the reward amount of the new incentive information is less than the reward amount of the historical incentive information;

[0133] For the third type of samples, the award threshold of the new incentive information is lower than the award threshold of the historical incentive information and / or the reward amount of the new incentive information is higher than the reward amount of the historical incentive information;

[0134] The training sample is constructed according to the historical information and the new stimulus information.

[0135] Still referring to Table 1, when the historical delivery volume of the target delivery person is 4 orders, that is, the target delivery person has not received any reward, it can be considered that when threshold 1 is raised, that is, threshold 1 is greater than 5, the order volume of the target delivery person still cannot meet the threshold 1, so a new sample can be constructed based on the adjusted threshold 1, the unadjusted reward value 1 and the historical data of the target delivery person (where the order volume is 4). Similarly, for the target delivery person, when the reward value 1 decreases, it can also be considered that the order volume of the target delivery person cannot meet the threshold 1, so a new sample can be constructed based on the threshold 1, the adjusted reward value 1 and the historical data of the target delivery person (where the order volume is 4). In addition, new sample data can also be constructed by simultaneously raising the threshold 1 and lowering the reward value 1.

[0136] Of course, in some embodiments, new samples can be constructed for the active delivery drivers based on the SMOTE method. For example, for a first target delivery driver, the sample distance between the first target delivery driver and other delivery drivers can be calculated, and samples of a second target delivery driver whose sample distance from the first target delivery driver is less than a threshold can be determined. In this way, new sample data can be randomly generated between the sample data of the first target delivery driver and the sample data of the second target delivery driver, thereby expanding the sample.

[0137] The above technical solution can expand the training samples, thereby increasing the number of training samples, which in turn helps to improve the accuracy of the model.

[0138] The present disclosure also provides a device for adjusting the amount of delivery orders, referring to Figure 5 The block diagram of a device for adjusting the delivery order quantity is shown, wherein the device 500 includes:

[0139] A first acquisition module 501 is configured to acquire delivery characteristic information of a regional delivery person and incentive configuration information corresponding to the delivery person in response to receiving a delivery order quantity adjustment instruction;

[0140] Input module 502 is configured to input the delivery characteristic information and the incentive configuration information into an order quantity prediction model, and obtain, as output by the order quantity prediction model, information on the delivery driver's order quantity under corresponding incentive measures; wherein the order quantity prediction model includes a first neural network layer for predicting a first order quantity of the delivery driver without incentive measures based on the delivery characteristic information, and a second neural network layer for predicting a second order quantity of the delivery driver under corresponding incentive measures based on the first order quantity and the incentive configuration information;

[0141] A first execution module 503 is configured to use the incentive configuration information as target incentive configuration information if the difference between the second order quantity and the first order quantity reaches a target threshold;

[0142] The adjustment module 504 is configured to adjust the delivery order quantity of the area according to the target incentive configuration information.

[0143] The above technical solution can determine the first order volume of a delivery driver without incentives through the first neural network layer in the order volume prediction model based on the delivery driver's delivery characteristic information. In this way, the second neural network can output the second order volume of the delivery driver with incentives based on the first order volume and the delivery driver's incentive information. In other words, using the above technical solution, the second order volume can be determined based on the incentive characteristics and the first order volume, thereby improving the accuracy of the second order volume. Furthermore, the above technical solution can also output the delivery driver's order volume with incentives and without incentives simultaneously through the order volume prediction model, thereby determining the change in order volume caused by the incentives. Furthermore, because the above technical solution outputs the delivery driver's order volume with incentives and without incentives through the order volume prediction model, it can avoid the cumulative error caused by separately predicting the order volume with incentives and without incentives when determining the change in order volume caused by incentives, thereby improving accuracy. Therefore, the target incentive configuration information can be determined based on the first order volume and the second order volume, and the delivery order volume of the area can be adjusted according to the target incentive configuration information, thereby solving the problem of insufficient delivery capacity.

[0144] Optionally, it also includes: a second execution module, configured to set constraints for the second neural network layer during the training process of the order quantity prediction model, and the constraints are used to ensure that the second order quantity output by the second neural network layer is not less than the first order quantity output by the first neural network layer.

[0145] Optionally, the device also includes: a second acquisition module, configured to obtain the order quantity prediction model, the second acquisition module includes: an acquisition sub-module, configured to obtain historical information of the deliveryman, the historical information includes historical delivery information and historical incentive information corresponding to the deliveryman with incentives, and historical delivery information corresponding to the deliveryman without incentives, the historical delivery information includes the historical order quantity completed by the deliveryman; a creation sub-module, configured to construct model training samples based on the historical information, the training samples include first-class samples, second-class samples and third-class samples, the first-class samples are samples corresponding to deliverymen without incentives, the second-class samples are samples corresponding to deliverymen with incentives and no rewards, and the third-class samples are samples corresponding to deliverymen with incentives and rewards; a training sub-module, configured to obtain the order quantity prediction model through training based on the training samples.

[0146] Optionally, the loss function of the single-volume prediction model includes at least one of a first loss function, a second loss function, and a third loss function; the first loss function is used to limit the absolute average error between the true value and the predicted value of the first category of samples to be minimum; the second loss function is used to limit the absolute average error between the true value and the predicted value of the second category of samples to be minimum; the third loss function is used to limit the absolute average error between the true value and the predicted value of the third category of samples.

[0147] Optionally, the loss function of the single quantity prediction model is:

[0148]

[0149] Among them, loss1 is the first loss function, loss2 is the second loss function, loss3 is the third loss function, MAE is the mean absolute error, label1 is the historical order volume completed by the deliveryman corresponding to the first category of samples, label2 is the historical order volume completed by the deliveryman corresponding to the second category of samples, label3 is the historical order volume completed by the deliveryman corresponding to the third category of samples, y is the order volume of the corresponding deliveryman output by the order volume prediction model, μ is the search parameter, reduce_mean() is the incentive loss, threshold1 is the first order volume threshold required to obtain the lowest reward value in the incentive configuration information, and threshold_act is the second order volume threshold required to obtain the highest reward value in the incentive configuration information.

[0150] Optionally, the creation submodule includes: a first creation subunit, configured to construct new incentive information for the historical delivery information and historical incentive information of the delivery personnel with incentive measures; wherein, for the second type of samples, the winning threshold of the new incentive information is greater than the winning threshold of the historical incentive information and / or the reward amount of the new incentive information is less than the reward amount of the historical incentive information; for the third type of samples, the winning threshold of the new incentive information is less than the winning threshold of the historical incentive information and / or the reward amount of the new incentive information is greater than the reward amount of the historical incentive information; a second creation subunit, configured to construct the training sample based on the historical information and the new incentive information.

[0151] Optionally, the incentive configuration information is multiple, and the input module includes:

[0152] An input submodule is configured to input the delivery characteristic information and the incentive configuration information into an order quantity prediction model for each incentive configuration information, and obtain the order quantity information of the delivery person under the incentive measure output by the order quantity prediction model;

[0153] The first execution module includes:

[0154] The first execution submodule is configured to use the incentive configuration information corresponding to the maximum difference between the second order quantity and the first order quantity as the target incentive configuration information.

[0155] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0156] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the above embodiment when the program is executed by a processor.

[0157] The present disclosure also provides an electronic device, comprising:

[0158] a memory having a computer program stored thereon;

[0159] A processor is configured to execute the computer program in the memory to implement the steps of the method described in the above embodiment.

[0160] Figure 6 6 is a block diagram of an electronic device 600 according to an exemplary embodiment. For example, the electronic device 600 may be provided as a server, and the server may be an order management server. Figure 6The electronic device 600 includes a processor 622, which may be one or more, and a memory 632 for storing a computer program executable by the processor 622. The computer program stored in the memory 632 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor 622 may be configured to execute the computer program to perform the aforementioned method for adjusting the delivery order quantity.

[0161] In addition, the electronic device 600 may further include a power supply component 626 and a communication component 650. The power supply component 626 may be configured to perform power management of the electronic device 600, and the communication component 650 may be configured to implement communication, such as wired or wireless communication, of the electronic device 600. In addition, the electronic device 600 may further include an input / output (I / O) interface 658. The electronic device 600 may operate based on an operating system stored in the memory 632, such as Windows Server™, Mac OS X™, Unix™, Linux™, etc.

[0162] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided. When executed by a processor, the program instructions implement the steps of the aforementioned method for adjusting the delivery order quantity. For example, the computer-readable storage medium may be the aforementioned memory 632 including the program instructions. The program instructions may be executed by the processor 622 of the electronic device 600 to implement the aforementioned method for adjusting the delivery order quantity.

[0163] In another exemplary embodiment, a computer program product is further provided. The computer program product includes a computer program that can be executed by a programmable device, and the computer program has a code portion for executing the above-mentioned method for adjusting the delivery order quantity when executed by the programmable device.

[0164] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the technical concept of the present disclosure, various simple modifications can be made to the technical solutions of the present disclosure, and these simple modifications all fall within the scope of protection of the present disclosure.

[0165] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0166] In addition, the various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A method for adjusting the quantity of delivery orders, characterized in that: The method comprises: In response to receiving a delivery order quantity adjustment instruction, obtaining delivery characteristic information of a regional delivery person and incentive configuration information corresponding to the delivery person; Inputting the delivery characteristic information and the incentive configuration information into an order quantity prediction model, and obtaining, as output by the order quantity prediction model, information on the order quantity of the delivery person under the corresponding incentive measures; wherein the order quantity prediction model includes a first neural network layer for predicting a first order quantity of the delivery person without the incentive measures based on the delivery characteristic information, and a second neural network layer for predicting a second order quantity of the delivery person under the corresponding incentive measures based on the first order quantity and the incentive configuration information; If the difference between the second order quantity and the first order quantity reaches a target threshold, the incentive configuration information is used as target incentive configuration information; Adjusting the delivery order volume of the region according to the target incentive configuration information; The deliveryman incentive configuration information includes multiple thresholds and reward values ​​corresponding to each threshold, wherein the threshold is the deliveryman's order volume. Different thresholds are obtained by dividing the deliveryman's order volume into steps. The reward value is an amount corresponding to each threshold. The effect of the incentive measures is determined according to the change in the order volume, and the incentive measures are adjusted accordingly. The order quantity prediction model is obtained by: Obtaining historical information of the delivery personnel, the historical information including historical delivery information and historical incentive information of the delivery personnel with incentive measures, and historical delivery information of the delivery personnel without incentive measures, the historical delivery information including the historical number of orders completed by the delivery personnel; Constructing model training samples based on the historical information, the training samples include first-category samples, second-category samples, and third-category samples, the first-category samples being samples corresponding to delivery personnel without incentives, the second-category samples being samples corresponding to delivery personnel with incentives but not receiving rewards, and the third-category samples being samples corresponding to delivery personnel with incentives and receiving rewards; Based on the historical delivery information and historical incentive information of the delivery personnel with incentive measures, new incentive information is constructed; For the second type of samples, the winning threshold of the new incentive information is greater than the winning threshold of the historical incentive information and / or the reward amount of the new incentive information is less than the reward amount of the historical incentive information; For the third type of samples, the award threshold of the new incentive information is lower than the award threshold of the historical incentive information and / or the reward amount of the new incentive information is higher than the reward amount of the historical incentive information; Constructing the training sample according to the historical information and the new incentive information; The single quantity prediction model is obtained by training according to the training samples.

2. The method according to claim 1, characterized in that During the training process of the order quantity prediction model, constraints are set for the second neural network layer, and the constraints are used to ensure that the second order quantity output by the second neural network layer is not less than the first order quantity output by the first neural network layer.

3. The method according to claim 1, characterized in that The loss function of the single quantity prediction model includes at least one of a first loss function, a second loss function, and a third loss function; The first loss function is used to ensure that the absolute average error between the true value and the predicted value of the first type of samples is minimized; The second loss function is used to ensure that the absolute average error between the true value and the predicted value of the second type of samples is minimized; The third loss function is used to ensure that the absolute average error between the true value and the predicted value of the third type of samples is minimized.

4. The method according to claim 3, characterized in that The loss function of the single quantity prediction model is: ； in, is the first loss function, is the second loss function, is the third loss function, MAE is the mean absolute error, is the historical order volume completed by the delivery personnel corresponding to the first type of samples, is the historical order volume completed by the delivery personnel corresponding to the second type of samples, is the historical order volume completed by the deliveryman corresponding to the third type of sample, y is the order volume of the corresponding deliveryman output by the order volume prediction model, is the search parameter, To motivate loss, The first order quantity threshold required to obtain the minimum reward value in the incentive configuration information, The second order quantity threshold required to obtain the highest reward value in the incentive configuration information.

5. The method according to claim 1, wherein The incentive configuration information includes multiple items, and the delivery characteristic information and the incentive configuration information are input into the order quantity prediction model to obtain the order quantity information of the delivery person under the corresponding incentive measures output by the order quantity prediction model, including: For each incentive configuration information, input the delivery characteristic information and the incentive configuration information into an order quantity prediction model to obtain the order quantity information of the delivery person under the incentive measure output by the order quantity prediction model; If the difference between the second order quantity and the first order quantity reaches a target threshold, using the incentive configuration information as target incentive configuration information includes: The incentive configuration information corresponding to the maximum difference between the second order quantity and the first order quantity is used as the target incentive configuration information.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

7. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 5.

Images