Method, device, storage medium and electronic device for predicting delivery status information

By training the delivery status information prediction model, using multi-task learning and DeepFM model to predict the delivery status information in extreme areas, the problem of poor prediction accuracy of the global regression model in extreme areas is solved, and more accurate delivery status information prediction and stress relief are achieved. .

CN110378522BActive Publication Date: 2024-11-12BEIJING SANKUAI ONLINE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201910590756.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2019-07-02
Publication Date
2024-11-12
Estimated Expiration
2039-07-02

AI Technical Summary

Technical Problem

The existing global regression model has poor accuracy when predicting delivery status information in extreme areas (such as high-load areas and severe weather areas) and cannot effectively alleviate delivery pressure.

Method used

By obtaining the delivery service data and area feature data of the target area, training the delivery status information prediction model, using multi-task learning and DeepFM model, predicting the average delivery time, order quantity and average order load of the target area within the preset time, improving the prediction accuracy.

Benefits of technology

It improves the accuracy of the delivery status information for extreme areas, and can promptly determine whether a burst order will occur, thereby taking measures to alleviate the delivery pressure and improve user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN110378522B_ABST
    Figure CN110378522B_ABST
Patent Text Reader

Abstract

The present disclosure relates to a method, device, storage medium and electronic device for predicting delivery status information, which can obtain delivery service data of a target area at a preset time and regional characteristic data corresponding to the target area, wherein the regional characteristic data includes time identification data, first weather data and regional identification data for characterizing the corresponding area; and obtain the delivery status information of the target area within a preset time period starting from the preset time through a delivery status information prediction model based on the delivery service data and the regional characteristic data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

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

[0002] Instant delivery business is a business model that has emerged and developed rapidly in recent years. If a delivery area experiences an explosion of orders, it means that the area is in a state of imbalance between supply and demand. At this time, corresponding measures should be taken to suppress users' willingness to place orders, thereby alleviating the delivery pressure in the area and improving user experience.

[0003] Normally, the delivery status information within a preset time in the future (such as the average delivery time, order volume or average load, etc.) can be used as the main basis for triggering regional order explosion. In the related technology, the global regression model is mainly used to predict the delivery status information within a preset time in the future. However, the global regression model is trained based on the delivery service data of all regions, which can only solve the problem of global optimization, that is, to ensure the overall accuracy of all regions. When the global regression model is used to estimate the delivery status information of extreme areas (such as high-load areas, areas with bad weather, etc.), the accuracy of the estimate is poor. Summary of the invention

[0004] The purpose of the present disclosure is to provide a method, device, storage medium and electronic device for predicting delivery status information.

[0005] In a first aspect, a method for predicting delivery status information is provided, the method comprising: obtaining delivery service data of a target area at a preset time and regional characteristic data corresponding to the target area, the regional characteristic data comprising time identification data, first weather data, and regional identification data for characterizing the corresponding area; obtaining the delivery status information of the target area within a preset time period starting from the preset time through a delivery status information prediction model based on the delivery service data and the regional characteristic data.

[0006] Optionally, the delivery status information includes average delivery time information, order quantity information, and average order load information; the average order load information is used to characterize the order quantity averagely distributed to each delivery person within the preset time period.

[0007] Optionally, the delivery service data includes one or more of the following data items: regional order volume data of the target area, regional order load data of the target area, regional delivery personnel data of the target area, order service time data of the target area within a preset historical time period with the preset moment as the end moment, and historical order statistical data of the target area within a preset historical time period.

[0008] Optionally, the delivery status information prediction model is trained in the following manner: obtaining historical delivery service data of a preset sample area at a preset historical moment, historical area feature data corresponding to the preset sample area, and historical delivery status information of the preset sample area within the preset time period from the preset historical moment; using the historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period from the preset historical moment as model training samples for training to obtain the delivery status information prediction model.

[0009] Optionally, the historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period from the preset historical moment are used as model training samples for training to obtain the delivery status information prediction model, including: using the historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period from the preset historical moment as model training samples, training the multi-task model to obtain the delivery status information prediction model, wherein the average delivery time information serves as the main task of the multi-task model, and the order quantity information and the average order load information serve as auxiliary tasks of the multi-task model.

[0010] Optionally, the loss function of the main task is a piecewise function; the piecewise function includes a first sub-function and a second sub-function; the first segmentation interval corresponding to the first sub-function is that the actual average delivery time of orders within the preset time period starting from the preset historical moment is less than a preset time threshold, and the first sub-function is obtained according to a preset first parameter adjustment coefficient and the actual average delivery time located in the first segmentation interval; the second segmentation interval corresponding to the second sub-function is that the actual average delivery time is greater than or equal to the preset time threshold, and the second sub-function is obtained according to a preset second parameter adjustment coefficient and the actual average delivery time located in the second segmentation interval; the preset time threshold represents the critical value of the average delivery time when a preset order abnormal event occurs in a preset area; the first parameter adjustment coefficient and the second parameter adjustment coefficient are different, and the first parameter adjustment coefficient is less than the second parameter adjustment coefficient.

[0011] Optionally, the regional characteristic data also includes regional type characteristic data, and the delivery status information prediction model includes multiple delivery status information prediction models, and different regional type characteristic data correspond to different delivery status information prediction models; obtaining the delivery status information of the target area within the preset time period starting from the preset moment through the delivery status information prediction model according to the delivery service data and the regional characteristic data includes: determining a target prediction model corresponding to the regional type characteristic data from multiple delivery status information prediction models; obtaining the delivery status information of the target area within the preset time period starting from the preset moment through the target prediction model according to the delivery service data and the regional characteristic data.

[0012] Optionally, the area type characteristic data includes any one of the following: second weather data used to characterize that the weather information of the corresponding area at the preset time meets the severe weather conditions; order load data used to characterize that the order load information of the corresponding area at the preset time meets the high load conditions, and the order load data is used to characterize the order quantity averagely distributed to each delivery person at the preset time; the second weather data and the order load data; abnormal area identification data used to characterize the corresponding area as an order abnormality area.

[0013] In a second aspect, a device for predicting delivery status information is provided, the device comprising: an acquisition module for acquiring delivery service data of a target area at a preset time and regional characteristic data corresponding to the target area, the regional characteristic data comprising time identification data, first weather data, and regional identification data for characterizing the corresponding area; a prediction module for obtaining the delivery status information of the target area within a preset time period starting from the preset time through a delivery status information prediction model based on the delivery service data and the regional characteristic data.

[0014] In a third aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in the first aspect of the present disclosure are implemented.

[0015] In a fourth aspect, an electronic device is provided, comprising: a memory on which a computer program is stored; and a processor for executing the computer program in the memory to implement the steps of the method described in the first aspect of the present disclosure.

[0016] Through the above technical scheme, the delivery service data of the target area at a preset time and the regional characteristic data corresponding to the target area are obtained, and the regional characteristic data include time identification data, first weather data and regional identification data for characterizing the corresponding area; according to the delivery service data and the regional characteristic data, the delivery status information of the target area within a preset time period starting from the preset time is obtained through a delivery status information prediction model. That is to say, the present disclosure predicts the delivery status information of the target area based on the delivery service data of the target area and the regional identification data of the target area, the first weather data and the time identification data and other regional characteristic data of the target area. In this way, the regional characteristic data of the target area can be learned, thereby improving the accuracy of predicting the delivery status information of the target area.

[0017] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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 specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0019] Figure 1 is a flow chart showing a method for predicting delivery status information according to an exemplary embodiment;

[0020] Figure 2 is a structural schematic diagram of a delivery status information prediction model according to an exemplary embodiment;

[0021] Figure 3 is a flow chart of a method for training a delivery status information prediction model according to an exemplary embodiment;

[0022] Figure 4 is a block diagram of a device for predicting delivery status information according to an exemplary embodiment;

[0023] Figure 5 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0024] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0025] The purpose of the present disclosure is to provide a method for estimating delivery status information such as average delivery time, order volume, average order load, etc. for a specific target area (the target area may be any of extreme areas such as high-load areas and areas with bad weather, or non-extreme areas) for a period of time in the future, and to serve as the main reference for measuring the capacity shortage in the target area. In the related art, a global regression model trained based on delivery service data of multiple areas is mainly used to predict delivery status information within a preset time period in the future, but this can only guarantee the overall accuracy of multiple areas. When the global regression model is used to estimate the delivery status information of extreme areas such as high-load areas and areas with bad weather, the accuracy of the estimate is poor.

[0026] To solve the above-mentioned problems, the present disclosure provides a method, device, storage medium and electronic device for predicting delivery status information. In the process of predicting the delivery status information of a target area, the delivery service data of the target area at a preset time and the regional characteristic data corresponding to the target area can be first obtained, and then the delivery status information of the target area within a preset time period starting from the preset time can be obtained through a delivery status information prediction model based on the delivery service data and the regional characteristic data, wherein the regional characteristic data includes time identification data, first weather data and regional identification data for characterizing the corresponding area, and the delivery status information prediction model is trained based on the regional characteristic data and delivery service data of multiple preset sample areas. In this way, the delivery status information prediction model can perform personalized learning on the regional characteristic data of the target area, thereby improving the accuracy of predicting the delivery status information of the target area. In addition, the target area can be any area in extreme areas such as high-load areas and bad weather areas or non-extreme areas. Therefore, the prediction method provided by the present disclosure can also improve the accuracy of estimating the delivery status information of extreme areas.

[0027] Figure 1 is a flow chart of a method for predicting delivery status information according to an exemplary embodiment. Figure 1 As shown, the method comprises the following steps:

[0028] In step 101, the delivery service data of the target area at a preset time and the regional characteristic data corresponding to the target area are obtained, and the regional characteristic data includes time identification data, first weather data and regional identification data for characterizing the corresponding area.

[0029] Among them, the target area can be any area among multiple areas that need to predict the delivery status information, and the target area can be any area in extreme areas such as high-load areas and bad weather areas, or non-extreme areas. The preset moment can be the current moment. The delivery service data may include one or more of the following data items: regional order volume data of the target area, regional order load data of the target area, regional delivery personnel data of the target area, order service time data of the target area within a preset historical time period with the preset moment as the end moment, and historical order statistics data of the target area within a preset historical time period. Further, the regional order volume data can be data such as the number of orders generated by the target area within a period of time, the number of orders that have not been picked up at the current moment, the number of orders that have been picked up but not delivered at the current moment, and the cumulative number of orders completed on the day. The load data can be understood as the average number of orders distributed to each delivery person. The regional order load data may include data such as the regional load of the target area, the average load of the target area over a period of time, and the regional delivery personnel data may include data such as the current number of on-duty riders in the target area and the total number of riders in the target area.

[0030] The order service time data is explained below in the form of an example. It is assumed that the preset moment is the current moment, and the preset historical time is the past ten minutes with the current moment as the end moment. The order service time data may include the average actual arrival time at the store for orders that have been picked up in the target area in the past ten minutes (arrival time minus order time), the average actual waiting time for orders that have been picked up in the target area in the past ten minutes (pickup time minus arrival time), the average actual delivery time for orders that have been delivered in the target area in the past ten minutes (actual delivery time minus pick-up time), the average actual delivery time for orders that have been delivered in the target area in the past ten minutes (actual delivery time minus order time), the average actual delivery time for orders that have been delivered in the target area in the past ten minutes (actual delivery time minus order time), the average estimated delivery time for orders that have been delivered in the target area in the past ten minutes (estimated delivery time minus order time), and the average actual order receiving time for orders that have been picked up in the target area in the past ten minutes (order receiving time minus order time).

[0031] Assuming that the preset historical time period is the past 24 hours with the current moment as the end moment, the historical order statistics data may include the total order volume of the target area in the past 24 hours, the average load order volume per capita in the target area in the past 24 hours, the average pickup time and average delivery time of orders in the target area in the past 24 hours, and other data.

[0032] It should be noted that the above description of the delivery service data is only an example and the present disclosure does not limit this.

[0033] In addition, the time identification data is used to characterize the time period in which the preset moment is located in multiple preset time periods. In one possible implementation, the 24 hours of a day can be divided into multiple time periods. For example, a time period can be divided every 10 minutes. Specifically, 0:00-0:10 (indicating midnight to ten minutes after midnight) is the first time period, 0:10-0:20 is the second time period, and so on. The 24 hours of a day can be divided into 144 time periods. In this way, the time identification data can be the identification information of the time period in which the current moment is located. For example, the current moment is 0:15, and it can be determined that the time period in which the moment is located is the second time period 0:10-0:20. This is just an example and the present disclosure is not limited to this.

[0034] The first weather data is used to characterize the weather information of the target area at the preset time. For example, the first weather data may include weather data such as temperature, wind speed, rainfall, etc. of the target area at the preset time, and the area identification data may include identification information such as city ID and area ID.

[0035] In step 102, the delivery status information of the target area within a preset time period starting from the preset time is obtained through a delivery status information prediction model based on the delivery service data and the regional characteristic data.

[0036] Among them, the preset time length can be the time length pre-set by the user according to his own needs in the actual prediction scenario. For example, the preset time length can be the next 20 minutes with the current moment as the starting moment, the next hour with the current moment as the starting moment, etc. The delivery status information may include average delivery time information, order quantity information, and average order load information. Usually, machine learning models (such as the delivery status information prediction model in the present disclosure) are all single-task learning, for example, only predicting one of the three data of average delivery time, order quantity and average order load, but considering that each task is also interrelated, the associated multi-task learning can achieve better generalization effect than single-task learning. In the present disclosure, since the average delivery time, order quantity and average order load in the same area are strongly correlated, based on this correlation, the present disclosure can simultaneously predict the average delivery time, order quantity and average order load of the target area within the preset time length through a delivery status information prediction model (which can be understood as multi-task learning), so as to improve the estimation accuracy of the model, wherein the average order load information is used to characterize the order quantity averagely distributed to each delivery person within the preset time length.

[0037] The delivery status information prediction model may include a DeepFM model. The following takes the delivery status information prediction model as the DeepFM model as an example to illustrate the specific implementation of this step:

[0038] For example, Figure 2 is a schematic diagram of the DeepFM model structure, such as Figure 2 As shown, the DeepFM model includes a feature data input layer, a word embedding layer, a DeepFM layer and an output layer. In this step, the time identification data of the target area, the first weather data and the area identification data can be input into the word embedding layer of the DeepFM model for embedding representation learning to convert the regional feature data of the target area into a word embedding vector, and then the word embedding vector is input into the DeepFM layer. In addition, the delivery service data belongs to continuous feature data and does not need to be processed by word embedding. The delivery service data can be directly input into the DeepFM layer for feature learning. In this way, the regional feature data and the delivery service data of the target area can be learned at the same time through the DeepFM layer, so that the delivery status information such as the average delivery time, order volume and average order load of the target area within a preset time can be output through the output layer. The above examples are only illustrative, and the present disclosure is not limited to this.

[0039] After executing step 102, the delivery status information prediction model can be used to predict the average delivery time, order volume, average order load and other delivery status information of the target area within a preset time period based on the above-mentioned delivery service data and regional characteristic data of the target area. In this way, it can be determined based on the delivery status information whether the target area will have a burst of orders within the preset time period, so that when it is predicted that a burst of orders will occur, effective measures can be taken in time to alleviate the delivery pressure of the target area.

[0040] In another possible implementation of the present disclosure, the regional characteristic data may further include regional type characteristic data, and the delivery status information prediction model may include multiple delivery status information prediction models, and different regional type characteristic data correspond to different delivery status information prediction models. In this way, in the process of obtaining the delivery status information of the target area within a preset time period starting from the preset moment through the delivery status information prediction model according to the delivery service data and the regional characteristic data, a target prediction model corresponding to the regional type characteristic data can be determined from the multiple delivery status information prediction models, and then the delivery status information of the target area within a preset time period starting from the preset moment can be obtained through the target prediction model according to the delivery service data and the regional characteristic data.

[0041] Among them, the area type characteristic data may include any one of the following: second weather data used to characterize that the weather information of the corresponding area at the preset moment meets the severe weather conditions; order load data used to characterize that the order load information of the corresponding area at the preset moment meets the high load conditions, wherein the order load data is used to characterize the order quantity averagely distributed to each delivery person at the preset moment; the second weather data and the order load data; abnormal area identification data used to characterize the corresponding area as an order abnormality area.

[0042] For example, the second weather data may be identification information of severe weather (for example, 1 may be used to represent severe weather, and 0 may be used to represent non-severe weather). In one possible implementation, multiple weather thresholds that meet severe weather conditions may be pre-set for different weather types. Specifically, a temperature threshold may be set for high temperature severe weather. If the temperature of the target area in the first weather data at the current moment is greater than or equal to the temperature threshold, the target area may be determined as a high temperature severe weather area. A rainfall threshold may be set for heavy rainfall severe weather. If the rainfall of the target area in the first weather data at the current moment is greater than or equal to the rainfall threshold, the target area may be determined as a high temperature severe weather area. The target area is determined as an area of ​​severe weather with heavy rainfall; a wind speed threshold can be set for severe weather with storms. If the wind speed of the target area in the first weather data at the current moment is greater than or equal to the wind speed threshold, the target area can be determined as an area of ​​severe weather with storms. It should be noted that when it is determined that at least one weather data in the first weather data (such as at least one of the various weather data such as temperature, rainfall, wind speed, etc.) meets the corresponding threshold conditions mentioned above, the current weather in the target area can be determined as severe weather. At this time, the area type of the target area can be determined as a severe weather area. The above examples are only for illustrative purposes and the present disclosure is not limited to this.

[0043] In addition, a load threshold that meets the high load conditions can be pre-set. In this way, when the order load data of the target area at the current moment is greater than or equal to the load threshold, the area type of the target area can be determined as a high load area.

[0044] In a possible application scenario, if the target area is determined to be a severe weather area based on the first weather data of the target area, and is determined to be a high load area based on the order load data, then the area type of the target area can be determined as both a severe weather area and a high load area.

[0045] In another possible application scenario, there are usually some long-tail areas with abnormal orders due to unknown reasons. In a possible implementation method, in order to improve the accuracy of the delivery status information estimation in the long-tail area, a delivery status information prediction model for the long-tail area can be pre-established based on the delivery service data and regional characteristic data of multiple long-tail areas, and the regional identification information can be set in advance for this type of area. In this way, when the acquired regional characteristic data includes the abnormal area identification data, the area type of the target area can be determined as an order abnormality area (also referred to as a long-tail area). In this way, the delivery status information of the target area can be predicted by the delivery status information prediction model corresponding to the order abnormality area.

[0046] Four different types of regions are listed above, namely, severe weather regions, high-load regions, regions with both severe weather and high load, and order abnormality regions. In the actual application scenarios of the present disclosure, other types of regions other than the above four types of regions may also be included, and the present disclosure does not limit this. Thus, when predicting the delivery status information of the target region, the region type corresponding to the target region may be determined based on the region type characteristic data, and then the model corresponding to the region type of the target region is determined from a plurality of pre-set delivery status information prediction models as the target prediction model. Thus, the delivery status information of the target region within a preset time period from the preset moment may be obtained through the target prediction model based on the delivery service data and the region characteristic data. The specific prediction method has been described in. Figure 2 The example shown is described in detail and will not be repeated here.

[0047] It should also be noted that if it is determined based on the regional type characteristic data of the target area that the regional type of the target area does not belong to any of the above-mentioned regional types, the regional type of the target area can be determined as an ordinary regional type, and then the target prediction model corresponding to the ordinary regional type is obtained to predict the delivery status information of the target area through the target prediction model.

[0048] The above method can be used to predict the delivery status information of the target area based on the delivery service data of the target area and the regional characteristic data of the target area, such as the regional identification data, the first weather data and the time identification data. In this way, the regional characteristic data of the target area can be learned, thereby improving the accuracy of predicting the delivery status information of the target area. Moreover, since the target area can be any extreme area or non-extreme area such as a high-load area and a bad weather area, the prediction method provided by the present invention can also improve the accuracy of the estimation of the delivery status information of the extreme area.

[0049] In order to ensure the effectiveness and accuracy of the model prediction, the delivery status information prediction model needs to be trained and updated according to a preset period (such as every day). The training process of the delivery status information prediction model is described below. Figure 3 is a flowchart of a method for training the delivery status information prediction model according to an exemplary embodiment. Figure 3 As shown, the method comprises the following steps:

[0050] In step 301, historical delivery service data of a preset sample area at a preset historical moment, historical area feature data corresponding to the preset sample area, and historical delivery status information of the preset sample area within a preset time period from the preset historical moment are obtained.

[0051] The preset sample area may include multiple sample areas, and the preset historical moment may include multiple moments. For example, after dividing the 24 hours of a day into multiple time periods every ten minutes, the preset historical moment may be any moment in each of the time periods. In addition, when training the delivery status information prediction model, the preset time length should be consistent with the Figure 1 The preset time length in step 102 of the embodiment 1 shown is equal, so that the delivery status information prediction model obtained by training can be used to predict the delivery status information of the target area within the preset time length starting from the preset moment, and the historical delivery service data, the historical area characteristic data and the historical delivery status information can all be selected from the data in the preset historical time period before the model training date. For example, the preset historical time period can be from the day before the model training date to two weeks before the model training date.

[0052] In step 302, the historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period starting from the preset historical moment are used as model training samples to train the multi-task model to obtain the delivery status information prediction model.

[0053] In the actual training scenario, it is necessary to first construct a training set and a test set for model training. In a possible implementation, the data in the first preset historical sub-time period can be selected as the training sample, and the data in the second preset historical sub-time period can be selected as the test sample, and the preset historical time period includes the first preset historical sub-time period and the second preset historical sub-time period. For example, if the model training date is June 15, and the preset historical time period is from June 1 to June 14, the training sample can be the data obtained from June 1 to June 13, and the test sample can be the data obtained from June 1 to June 13. The data obtained on the 14th is only an example, and the present disclosure does not limit this. In addition, in the training sample or the test sample, each training sample includes two parts: features and labels, wherein the feature is the historical delivery service data and the historical area feature data, and the label is the historical delivery status information, such as the average delivery time of all orders within the preset time period from the preset historical moment, the order volume within the preset time period from the preset historical moment, and the average order volume distributed to each delivery person within the preset time period (that is, the average order load) are all the labels.

[0054] The average delivery time in each sample can be calculated by formula (1), and the average order load in each sample can be calculated by formula (2):

[0055]

[0056]

[0057] in, Indicates the average delivery time, Finish_duration n Indicates the delivery time of the nth order within the preset time. Indicates the average order load, load t It represents the order load at the t-th historical moment within the preset time (that is, the number of orders distributed to each delivery person on average at time t), N represents the total number of orders within the preset time, and T represents the preset time.

[0058] It should be noted that when selecting the training samples, samples whose historical delivery status information is abnormal data (for example, samples whose average delivery time in a region is greater than 2 hours are extremely abnormal cases) can be eliminated from the training samples to ensure the prediction accuracy of the trained model.

[0059] In addition, during the model training process, the network structure of the prediction model can be constructed first, such as Figure 2 As shown, assuming that the delivery status information prediction model is the DeepFM model, it can be Figure 2The structure shown in the figure is constructed. The specific method of constructing the model structure can refer to the description in relevant literature and will not be repeated here.

[0060] After building the model structure, the model parameters can be initialized according to the empirical values. For example, the word embedding vector length of the word embedding layer can be set to 64, the DNN part of the DeepFM model can be set to 3 fully connected layers, and dropout is set for the first-order and second-order terms of the FM part and each layer of the DNN. The dropout coefficient of the FM part can be set to 0.85, and the dropout coefficient of the DNN part can be set to 0.9. The initial learning rate of the model training can be set to 0.015, and the learning rate can be decayed every 2000 training steps, and the decay coefficient can be set to 0.95.

[0061] In addition, since the prediction method provided by the present invention can perform prediction learning on multiple tasks such as the average delivery time, order volume, average order load, etc. of the target area within a preset time from a preset moment, the multi-task model also needs to be trained in this step. Considering that in actual application scenarios, the average delivery time within a preset time is mainly used as the main basis for measuring whether an order explosion occurs in a region, therefore, in the present invention, the average delivery time information can be used as the main task of the multi-task model, and the order volume information and the average order load information can be used as auxiliary tasks of the multi-task model for model training. In addition, the loss function of the main task can use a piecewise function for model optimization to improve the problem of low prediction accuracy in extreme areas such as high-load areas and severe weather areas.

[0062] In which, the segmented function may include a first sub-function and a second sub-function, the first segmented interval corresponding to the first sub-function is that the actual average delivery time of orders within the preset time period starting from the preset historical moment is less than the preset time threshold, and the first sub-function is obtained according to the preset first parameter adjustment coefficient and the actual average delivery time located in the first segmented interval; the second segmented interval corresponding to the second sub-function is that the actual average delivery time is greater than or equal to the preset time threshold, and the second sub-function is obtained according to the preset second parameter adjustment coefficient and the actual average delivery time located in the second segmented interval; wherein, the preset time threshold represents the critical value of the average delivery time when a preset order abnormal event occurs in a preset area; the first parameter adjustment coefficient and the second parameter adjustment coefficient are different, and the first parameter adjustment coefficient is less than the second parameter adjustment coefficient.

[0063] For example, in a possible implementation, the piecewise function can be expressed as the following formula (3):

[0064]

[0065] Among them, y label represents the actual average delivery time, T represents the preset time threshold, which is the critical value of the average delivery time when a burst of orders occurs in the preset area (for example, T can be equal to 2400 seconds), y label <T represents the first segment interval, y label ≥T represents the second segment interval, L main represents the loss function value of the main task, α low represents the first parameter adjustment coefficient, α high represents the second parameter adjustment coefficient, and when the actual average delivery time is less than the preset time threshold (i.e., y label <T), there will usually be no explosion of orders. On the contrary, when the actual average delivery time is greater than or equal to the preset time threshold (i.e., y label ≥T), it is usually easy to have a burst of orders. Therefore, in order to ensure the accuracy of the average delivery time prediction for the burst of orders scenario, α is usually set low Less than α high , Ω(W) represents the regularization coefficient of the main task loss function.

[0066] In addition, the loss function for the auxiliary task can be the square loss function shown in formula (4):

[0067] L aux =||y pred -y label ||2 (4)

[0068] Among them, L aux Represents the loss function value of the auxiliary task. When the auxiliary task is to predict the order quantity information, the y label represents the actual order quantity, y pred Represents the order volume predicted by the model output; when the auxiliary task is to predict the average order load information, the y label represents the actual average order load, y pred Represents the average order load output by the model prediction.

[0069] After constructing the loss function, the multi-task model can be trained based on the historical delivery service data, the historical regional feature data and the historical delivery status information. In one possible implementation, the model training can be performed using an alternating training method.

[0070] For example, first, a random number in the range of 0-1 can be generated, and then the random number can be used as the basis for the training task of the current training round. The training task selection rules are shown in Table 1. It can be seen from the table that different training probabilities can be set for the main task and the auxiliary task, respectively. For example, the main task can be set to 0.6 and the auxiliary task can be set to 0.2. This can ensure that the main task can be fully trained with a limited number of training times, so that an order explosion warning can be issued more accurately. For example, the random number interval corresponding to the average delivery time predicted by the main task can be set to [0,0.6], which is consistent with the auxiliary task prediction. The random number interval corresponding to the average order load can be set to (0.6, 0.8], and the random number interval corresponding to the auxiliary task predicting the order volume can be set to (0.8, 1]. In this way, when the generated random number is in the interval [0, 0.6], the training task of the current training round is to predict the average delivery time; when the generated random number is in the interval (0.6, 0.8], the training task of the current training round is to predict the average order load; when the generated random number is in the interval (0.8, 1], the training task of the current training round is to predict the order volume. The above examples are only for illustration and the present disclosure is not limited to this.

[0071] Training Mission Training probability Estimated average delivery time 0.6 Forecast average order load 0.2 Forecast order volume 0.2

[0072] Table 1

[0073] By adopting the model training method provided above, a prediction model capable of multi-task learning is trained, and the loss function of the main task of the model is customized as a piecewise function, which can improve the problem of low prediction accuracy in extreme areas such as high-load areas and severe weather areas.

[0074] Figure 4 is a block diagram of a device for predicting delivery status information according to an exemplary embodiment. Figure 4 As shown, the device comprises:

[0075] An acquisition module 401 is used to acquire the delivery service data of a target area at a preset time and the regional characteristic data corresponding to the target area, wherein the regional characteristic data includes time identification data, first weather data, and regional identification data for characterizing the corresponding area;

[0076] The prediction module 402 is used to obtain the delivery status information of the target area within a preset time period starting from the preset time through a delivery status information prediction model according to the delivery service data and the regional characteristic data.

[0077] By using the above-mentioned device, the delivery status information of the target area can be predicted based on the delivery service data of the target area and the regional characteristic data such as the regional identification data, the first weather data and the time identification data of the target area. In this way, the regional characteristic data of the target area can be learned, so as to improve the accuracy of predicting the delivery status information of the target area. Moreover, since the target area can be any extreme area or non-extreme area such as a high-load area and a bad weather area, the prediction device provided by the present invention can also improve the accuracy of the estimation of the delivery status information of the extreme area.

[0078] Figure 5 is a block diagram of an electronic device 500 according to an exemplary embodiment. For example, the electronic device 500 may be provided as a server. Figure 5 The electronic device 500 includes a processor 522, which may be one or more, and a memory 532 for storing a computer program executable by the processor 522. The computer program stored in the memory 532 may include one or more modules, each corresponding to a set of instructions. In addition, the processor 522 may be configured to execute the computer program to perform the above-mentioned method for predicting delivery status information.

[0079] In addition, the electronic device 500 may further include a power supply component 526 and a communication component 550, wherein the power supply component 526 may be configured to perform power management of the electronic device 500, and the communication component 550 may be configured to implement communication, such as wired or wireless communication, of the electronic device 500. In addition, the electronic device 500 may further include an input / output (I / O) interface 558. The electronic device 500 may operate based on an operating system stored in the memory 532, such as Windows Server™, Mac OSX™, Unix™, Linux™, etc.

[0080] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the method for predicting delivery status information are implemented. For example, the computer-readable storage medium may be the memory 532 including the program instructions, and the program instructions may be executed by the processor 522 of the electronic device 500 to complete the method for predicting delivery status information.

[0081] In another exemplary embodiment, a computer program product is also 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 of predicting delivery status information when executed by the programmable device.

[0082] 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 in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0083] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0084] In addition, 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 predicting delivery status information, characterized in that: The method comprises: Acquire the delivery service data of the target area at a preset time and the regional characteristic data corresponding to the target area, wherein the regional characteristic data includes time identification data, first weather data, and regional identification data for characterizing the corresponding area; Obtaining the delivery status information of the target area within a preset time period starting from the preset time through a delivery status information prediction model according to the delivery service data and the regional characteristic data; The regional characteristic data also includes regional type characteristic data, and the regional type characteristic data includes: bad weather area, high load area, area with both bad weather and high load, and abnormal order area; Wherein, the delivery status information prediction model includes a plurality of delivery status information prediction models, and different area type feature data correspond to different delivery status information prediction models; The step of obtaining the delivery status information of the target area within a preset time period from the preset time by using a delivery status information prediction model according to the delivery service data and the regional characteristic data includes: Determining a target prediction model corresponding to the area type feature data from a plurality of the delivery status information prediction models; Obtaining, by using the target prediction model, the delivery status information of the target area within a preset time period starting from the preset time according to the delivery service data and the regional characteristic data; The region type characteristic data includes any of the following: Second weather data used to indicate that the weather information of the corresponding area at the preset time meets the severe weather conditions; Order load data used to indicate that the order load information of the corresponding area at the preset time meets the high load condition, and the order load data is used to indicate the order quantity averagely distributed to each delivery person at the preset time; the second weather data and the order load data; Abnormal area identification data used to indicate that the corresponding area is an abnormal order area; The delivery status information prediction model is trained in the following way: Obtaining historical delivery service data of a preset sample area at a preset historical moment, historical area feature data corresponding to the preset sample area, and historical delivery status information of the preset sample area within the preset time period from the preset historical moment; The historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period from the preset historical moment are used as model training samples for training to obtain the delivery status information prediction model; The training is performed using the historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period from the preset historical moment as model training samples to obtain the delivery status information prediction model, including: The historical delivery service data corresponding to the preset sample area at the preset historical moment, the historical area feature data corresponding to the preset sample area, and the historical delivery status information of the preset sample area within the preset time period from the preset historical moment are used as model training samples to train the multi-task model to obtain the delivery status information prediction model, wherein the average delivery time information is used as the main task of the multi-task model, and the order quantity information and the average order load information are used as auxiliary tasks of the multi-task model; The loss function of the main task is a piecewise function; The piecewise function includes a first sub-function and a second sub-function; The first segmentation interval corresponding to the first sub-function is that the actual average delivery time of the orders within the preset time period from the preset historical moment is less than the preset time threshold, and the first sub-function is obtained according to the preset first parameter adjustment coefficient and the actual average delivery time in the first segmentation interval; The second segmentation interval corresponding to the second sub-function is that the actual average delivery time is greater than or equal to the preset time threshold, and the second sub-function is obtained according to the preset second parameter adjustment coefficient and the actual average delivery time in the second segmentation interval; The preset time threshold represents the critical value of the average delivery time when a preset order abnormal event occurs in a preset area; the first parameter adjustment coefficient is different from the second parameter adjustment coefficient, and the first parameter adjustment coefficient is less than the second parameter adjustment coefficient.

2. The method according to claim 1, characterized in that The delivery status information includes average delivery time information, order quantity information, and average order load information; the average order load information is used to characterize the order quantity averagely distributed to each delivery person within the preset time.

3. The method according to claim 1, characterized in that The delivery service data includes one or more of the following data items: The target area's regional order volume data, the target area's regional order load data, the target area's regional delivery staff data, the target area's order service duration data within a preset historical duration with the preset moment as the end moment, and the target area's historical order statistics data within a preset historical time period.

4. 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 described in any one of claims 1 to 3 are implemented.

5. 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 3.