Method for training express volume prediction model, express transportation planning method and device

Abstract

The application provides a training method of a parcel volume prediction model, a parcel transportation planning method and device. After obtaining all historical shipping order information corresponding to historical shipping orders, the training method of the parcel volume prediction model first processes abnormal data in the historical shipping order information to obtain first shipping order information excluding the abnormal data; secondly, target parcel features affecting the parcel volume are determined from the first shipping order information to model the target parcel volume and obtain a target parcel volume prediction model, and the target parcel volume prediction model is used to predict the volume of the parcel. Through the shipping order volume explicitly in the historical shipping order and the cleaning of the abnormal data, the machine learning sample is obtained, the parcel features affecting the parcel volume are obtained, and then the detailed information of each shipping order is input, the volume of each shipping order is obtained through the prediction method, the prediction accuracy of the parcel volume can be effectively improved, and the efficiency of the parcel volume prediction is improved.

Description

Technical Field

[0001] This application relates to the field of express mail inspection, specifically to a training method for an express mail volume prediction model, an express mail transportation planning method, and an apparatus. Background Technology

[0002] Currently, there are two main methods for measuring the volume of express parcels: one is manual measurement based on length, width, and height, and the other is semi-automatic measurement based on visual algorithms and vision devices. However, both methods suffer from low measurement efficiency and high costs for both measurement equipment and labor. They can only be applied to express parcel volume measurement in small, specific scenarios and cannot be used for volume measurement of hundreds of millions of parcels per shipment. Summary of the Invention

[0003] This application aims to provide a training method for a parcel volume prediction model, a parcel transportation planning method, and an apparatus, in order to solve the problems of inaccurate parcel volume prediction and high computational cost and low efficiency in the prior art.

[0004] In a first aspect, embodiments of this application provide a training method for a package volume prediction model, the method comprising:

[0005] Retrieve historical waybill information corresponding to historical waybills within a preset historical time period;

[0006] The abnormal data in the historical waybill information is processed to obtain the first waybill information, which includes multiple basic characteristics of the express shipments corresponding to the express shipments in the historical waybill.

[0007] Based on multiple basic characteristics of express shipments in the first waybill information, the target express shipment characteristics that affect the shipment volume are determined;

[0008] The first parcel volume prediction model is trained based on the target parcel characteristics to obtain the target parcel volume prediction model for predicting parcel volume.

[0009] In one possible embodiment, processing the abnormal data in the historical waybill information to obtain the first waybill information includes:

[0010] Remove data from the historical waybill information that corresponds to features with a missing proportion greater than or equal to a preset missing proportion threshold.

[0011] Missing data in the historical waybill information is supplemented using a preset filling method;

[0012] Remove abnormal data from the historical waybill information;

[0013] The first waybill information includes multiple basic characteristics of the express shipments corresponding to the historical waybills.

[0014] In one possible embodiment, determining the target express shipment characteristics affecting the shipment volume based on multiple basic express shipment characteristics in the first waybill information includes:

[0015] Based on the multiple basic characteristics of express delivery mentioned above, the initial express delivery characteristics affecting the volume of express delivery are determined, resulting in multiple initial express delivery characteristics;

[0016] The multiple initial express delivery features are sorted to obtain multiple sorted initial express delivery features;

[0017] Among the sorted initial express features, the target express feature that affects the express volume is determined.

[0018] In one possible embodiment, training a first parcel volume prediction model based on the target parcel features to obtain the target parcel volume prediction model includes:

[0019] Obtain the volume prediction model for the first express shipment;

[0020] The first parcel volume prediction model is combined with the target parcel features to obtain the second parcel volume prediction model;

[0021] A training set including the volume of the parcel is obtained, and the second parcel volume prediction model is trained using the training set to obtain the target parcel volume prediction model.

[0022] In one possible embodiment, obtaining a training set including the parcel volume, and using the training set to train the second parcel volume prediction model to obtain the target parcel volume prediction model, includes:

[0023] The training set is divided into a first training set based on the delivery area and a second training set based on the item being delivered;

[0024] The second parcel volume prediction model is trained using the first training set to obtain a third parcel volume prediction model based on the delivery area.

[0025] The second express volume prediction model is trained using the second training set to obtain the fourth express volume prediction model based on the consigned item;

[0026] The target parcel volume prediction model is obtained by integrating the third parcel volume prediction model and the fourth parcel volume prediction model.

[0027] Secondly, embodiments of this application also provide a method for express delivery planning, the method further comprising:

[0028] Identify the set of parcels in the target vehicle, wherein the set of parcels includes multiple parcels to be transported;

[0029] The volume of the parcels in the parcel set is predicted using a target parcel volume prediction model, wherein the target parcel volume prediction model is any of the models described above;

[0030] Calculate the predicted vehicle volume loading rate for multiple express items in the express item collection;

[0031] Based on the predicted vehicle volume loading rate, adjust the parcels in the target vehicle.

[0032] In one possible embodiment, before determining the predicted vehicle volume loading rate for multiple express parcels in the parcel set, the method further includes:

[0033] The packages transported by different vehicles each time within the historical period are obtained to determine the historical vehicle volume loading rate for each transport by different vehicles.

[0034] Using the historical vehicle volume loading rate, a preset vehicle volume loading rate prediction model is trained to obtain a target vehicle volume loading rate prediction model.

[0035] Determining the predicted vehicle volume loading rate for multiple express parcels in the parcel set includes:

[0036] Using the target vehicle volume loading rate prediction model, the predicted vehicle volume loading rate values ​​for multiple express items in the express item set are determined;

[0037] In one possible embodiment, obtaining all express parcels transported by different vehicles each time within the historical period to determine the historical vehicle volume loading rate corresponding to each transport by different vehicles includes:

[0038] Obtain all express parcels transported by different vehicles for each time within the historical period;

[0039] Using a preset volume determination method, the sum of the visually estimated volumes of all express parcels transported by different vehicles each time is determined;

[0040] Based on the sum of the vehicle volume of each vehicle and the visually estimated volume, the historical vehicle volume loading rate for each transport trip of each vehicle is determined.

[0041] Thirdly, embodiments of this application also provide a training device for a parcel volume prediction model, the device comprising:

[0042] The acquisition module is used to acquire historical waybill information corresponding to all historical waybills within a preset historical time period;

[0043] The data processing module is used to process abnormal data in the historical waybill information to obtain first waybill information, which includes multiple basic characteristics of express shipments corresponding to the express shipments in the historical waybill.

[0044] The feature extraction module is used to determine the target express features that affect the express volume based on multiple basic express features in the first waybill information.

[0045] The training module is used to train a first parcel volume prediction model based on the characteristics of the target parcel to obtain a target parcel volume prediction model that predicts the parcel volume.

[0046] Fourthly, embodiments of this application also provide a piece transport planning device, the device comprising:

[0047] A determination module is used to determine the set of express parcels in the target vehicle, wherein the set of express parcels includes multiple express parcels to be transported;

[0048] The prediction module is used to predict the volume of the express items in the express item set using a target express item volume prediction model, wherein the target express item volume prediction model is any of the models described above.

[0049] The calculation module is used to calculate the predicted vehicle volume loading rate for multiple express items in the express item collection;

[0050] An adjustment module is used to adjust the parcels in the target vehicle based on the predicted vehicle volume loading rate.

[0051] Fifthly, embodiments of this application also provide a server, the server comprising:

[0052] One or more processors;

[0053] Memory; and

[0054] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the training method of the express volume prediction model as described in any of the preceding claims, or to implement the express transportation planning method as described in any of the preceding claims.

[0055] Sixthly, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute the steps in the training method of the express volume prediction model as described in any of the preceding claims, or to implement the steps in the express transportation planning method as described in any of the preceding claims.

[0056] This application provides a training method for a parcel volume prediction model, a parcel transportation planning method, and an apparatus. The training method for this parcel volume prediction model, after acquiring all historical waybill information corresponding to historical waybills, first processes abnormal data in the historical waybill information to obtain first waybill information excluding abnormal data; secondly, it identifies target parcel features affecting parcel volume from the first waybill information to model the target parcel volume, obtaining a target parcel volume prediction model, and then uses this model to predict the parcel volume. This application's embodiment obtains machine learning samples by using the clearly defined waybill volumes from historical waybills and cleaning abnormal data. By acquiring parcel features affecting parcel volume and inputting detailed information for each waybill, the volume of each waybill is obtained through prediction, effectively improving the accuracy and efficiency of parcel volume prediction. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a schematic diagram of a scenario for the express mail volume prediction system provided in an embodiment of this application;

[0059] Figure 2 A schematic flowchart of an embodiment of the training method for the express mail volume prediction model provided in this application;

[0060] Figure 3 A schematic diagram of an embodiment for determining the first waybill information provided in this application;

[0061] Figure 4 A schematic diagram of an embodiment for determining the characteristics of a target express shipment, provided in the application itself.

[0062] Figure 5 A schematic diagram of an embodiment of the cross features provided in this application;

[0063] Figure 6 A schematic diagram of another embodiment of the cross feature provided in this application;

[0064] Figure 7 A schematic diagram illustrating an embodiment of the target express mail features provided in this application.

[0065] Figure 8 This is a schematic flowchart of an embodiment of the express mail volume prediction model provided in this application.

[0066] Figure 9 A schematic flowchart of an embodiment of the target express shipment volume prediction model provided in this application;

[0067] Figure 10 A schematic flowchart of an embodiment of express mail volume prediction and vehicle planning provided in this application;

[0068] Figure 11 A schematic diagram of an embodiment of a training device for a parcel volume prediction model provided in this application;

[0069] Figure 12 A schematic diagram of an embodiment of the express delivery planning device provided in this application;

[0070] Figure 13 A schematic diagram of the structure of the electronic device involved in the embodiments of this application is shown. Detailed Implementation

[0071] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0073] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0074] It should be noted that since the method in this application embodiment is executed in an electronic device, the processing objects of each electronic device exist in the form of data or information, such as time, which is essentially time information. It can be understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data that exist so that the electronic device can process them. Specific details will not be elaborated here.

[0075] This application provides a training method for a parcel volume prediction model, a parcel transportation planning method, and an apparatus, which will be described in detail below.

[0076] Please see Figure 1 , Figure 1 This is a schematic diagram of a scenario for a parcel volume prediction system provided in an embodiment of this application. The parcel volume prediction system may include an electronic device 100, which integrates a parcel volume prediction device, such as... Figure 1 Electronic devices in the system.

[0077] In this embodiment, the electronic device 100 can be a standalone server, a server network, or a server cluster. For example, the electronic device 100 described in this embodiment includes, but is not limited to, a computer, a network host, a single network server, a set of multiple network servers, or a cloud server composed of multiple servers. The cloud server is composed of a large number of computers or network servers based on cloud computing.

[0078] Those skilled in the art will understand that Figure 1 The application environment shown is merely one application scenario of the present application solution and does not constitute a limitation on the application scenario of the present application solution. Other application environments may include more than one application scenario. Figure 1 The number of more or fewer electronic devices shown, for example Figure 1Only one electronic device is shown in the image. It is understood that the parcel volume prediction system may also include one or more other servers, which are not specified here.

[0079] In addition, such as Figure 1 As shown, the parcel volume prediction system may also include a memory 200 for storing data.

[0080] It should be noted that, Figure 1 The schematic diagram of the express volume prediction system shown is merely an example. The express volume prediction system and scenario described in this application are for the purpose of more clearly illustrating the technical solutions of this application and do not constitute a limitation on the technical solutions provided in this application. As those skilled in the art will know, with the evolution of express volume prediction systems and the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0081] First, this application provides a training method for a parcel volume prediction model. The execution subject of this training method is a parcel volume prediction device, which is applied to an electronic device. The training method includes: acquiring historical waybill information corresponding to historical waybills within a preset historical time period; processing abnormal data in the historical waybill information to obtain first waybill information, which includes multiple basic parcel features corresponding to the parcels in the historical waybills; determining target parcel features affecting parcel volume based on the multiple basic parcel features in the first waybill information; and training a first parcel volume prediction model based on the target parcel features to obtain a target parcel volume prediction model for predicting parcel volume.

[0082] like Figure 2 The diagram shown is a flowchart illustrating an embodiment of the training method for the express mail volume prediction model provided in this application, which may include:

[0083] 21. Obtain historical waybill information corresponding to historical waybills within a preset historical time period.

[0084] In the embodiments of this application, since it is necessary to determine the express characteristics of the express from the waybill information corresponding to the express, it is necessary not only to obtain the express volume corresponding to the express, but also to obtain other waybill information corresponding to the express, so as to determine the express characteristics corresponding to the express based on the complete waybill information.

[0085] Specifically, in the embodiments of this application, historical waybill information corresponding to express shipments within a historical time period is used to predict the volume of subsequent express shipments. The historical waybill information in this application includes not only the volume of express shipments within a historical time period, but also information such as the waybill number, time type, product, customer, origin, and destination. It also needs to include information such as the type of item being shipped, the number of pieces, the chargeable weight, the actual weight, and the chargeable revenue for each shipment.

[0086] In this context, each package corresponds to a specific product type, which is the goods offered by the logistics company to the sender. Different product types represent different delivery times and different pricing standards. For example, product types can include time-sensitive products and e-commerce products. Time-sensitive products offer strong delivery guarantees, and therefore command relatively higher prices. Conversely, e-commerce products are priced lower, and their delivery times are shorter than those of time-sensitive products.

[0087] The delivery time type indicates the corresponding delivery time requirement for the package. For example, a delivery time type of T4 means that the package needs to be delivered on the same day; a delivery time type of T6 means that the package needs to be delivered within three days. Different delivery time types correspond to different delivery time requirements.

[0088] "Consignment" refers to the way logistics companies categorize goods sent by users. It's typically divided into three levels: the first level includes 22 major categories, and the third level includes over two thousand subcategories. For example, if a user uses a standard express delivery service to send a bottle of alcohol, then "standard express" is the product type, and the alcohol is the consignment item. The alcohol would be classified as "alcoholic beverages" at the first level, "baijiu" (Chinese liquor) at the second level, and a specific brand of baijiu at the third level.

[0089] The actual weight refers to the actual weight of the package when it is weighed (usually including the weight of the outer packaging). Because express delivery charges use a tiered weight system, even if a package weighs less than one kilogram, it will still be charged as one kilogram; therefore, each package also has a chargeable weight. Generally, the chargeable weight of a package will be greater than its actual weight.

[0090] For example, the actual weight of a package including its outer packaging is 1.5 kg, but since 1.5 kg exceeds 1 kg, the chargeable weight is 2 kg.

[0091] Furthermore, in the embodiments of this application, the volume of the express delivery can also be divided into two categories: one is the actual measured volume of the express delivery; and the other is the volume of the outer packaging corresponding to the standard outer packaging used for the express delivery.

[0092] 22. Process the abnormal data in the historical waybill information to obtain the first waybill information.

[0093] The historical waybill information obtained includes not only the volume of the shipment, but also the corresponding waybill number, time type, product, customer, origin, and destination; as well as the corresponding consigned item, number of pieces, chargeable weight, actual weight, and chargeable revenue. This information may be recorded incorrectly or lost due to various reasons, resulting in abnormal data in the historical waybill information.

[0094] Therefore, this application embodiment requires cleaning the historical waybill information to remove abnormal data and obtain first waybill information that does not include abnormal data. Processing abnormal data in historical waybills includes not only removing some abnormal data but also filling in missing data.

[0095] The processed first waybill information includes multiple basic characteristics of the shipments from historical waybills. Specifically, the first waybill information also includes information such as the shipment's volume, waybill number, time-sensitive type, product, customer, origin, and destination, as well as information such as the consigned item, number of pieces, chargeable weight, actual weight, and chargeable revenue. These are all basic characteristics of the shipment, and each of these basic characteristics could be a major factor affecting the shipment's volume.

[0096] 23. Based on the multiple basic characteristics of express items in the first waybill information, determine the characteristics of the target express item corresponding to the express item.

[0097] After obtaining the initial waybill information excluding abnormal data, the target package characteristics corresponding to different packages can be determined based on this information. This allows us to identify which package characteristics will affect the package's volume. Based on these target package characteristics that influence package volume, the package volume can then be predicted.

[0098] Typically, there are multiple target characteristics of a package that affect its volume.

[0099] 24. Train the first parcel volume prediction model based on the characteristics of the target parcel to obtain the target parcel volume prediction model for predicting parcel volume.

[0100] After identifying the target package features that will affect the package volume, the target package features can be modeled to obtain a target package volume prediction model. The relevant parameters in the package volume prediction model are all target package features, and the output of the package volume prediction module is the predicted package volume.

[0101] The method for training a parcel volume prediction model provided in this application involves, after acquiring all historical waybill information corresponding to historical waybills, first processing abnormal data in the historical waybill information to obtain first waybill information excluding abnormal data; secondly, determining target parcel features affecting parcel volume from the first waybill information to model the target parcel volume, obtaining a target parcel volume prediction model, and using the target parcel volume prediction model to predict the parcel volume. This application embodiment obtains machine learning samples by using the clearly defined waybill volumes from historical waybills and cleaning abnormal data. By acquiring parcel features affecting parcel volume and inputting detailed information for each waybill, the volume of each waybill is obtained through prediction, effectively improving the accuracy and efficiency of parcel volume prediction.

[0102] like Figure 3 The diagram shown is a flowchart illustrating an embodiment of determining first waybill information provided in this application, which may include:

[0103] 31. Remove data corresponding to features in historical waybill information where the missing percentage is greater than the preset missing percentage threshold.

[0104] Specifically, the historical waybill information obtained includes information such as time-of-delivery type, product, origin, and destination for all waybills within the historical period; for the time-of-delivery type data, it includes multiple different time-of-delivery type data.

[0105] In this embodiment, various reasons may cause not all time-sensor type data corresponding to all waybills to be recorded, resulting in missing data for some time-sensor types. In this embodiment, the missing proportion of missing time-sensor type data can be calculated, including: the number of missing time-sensor type data / the total number of time-sensor type data. After calculating the missing proportion, it can be determined whether the missing proportion is greater than or equal to a preset missing proportion threshold; if the missing proportion is greater than or equal to the preset missing proportion threshold, the data corresponding to the time-sensor type in the historical waybill information is directly removed.

[0106] In one specific embodiment, the historical waybill information includes various types of data corresponding to 1000 waybills, and theoretically, there should also be 10,000 time-delivery type data entries. However, in reality, there are only 3000 time-delivery type data entries. In this case, the missing time-delivery type data ratio is (10000-3000) / 10000 = 0.7, meaning that 70% of the time-delivery type data is missing. Therefore, the time-delivery type data in the historical waybill information can be directly removed. That is, the 3000 time-delivery type data entries in the historical waybill information are removed.

[0107] In one specific embodiment of this application, the preset missing ratio threshold can be 0.5, that is, if the missing ratio of a certain type of data is greater than or equal to 0.5, then this type of data can be directly removed.

[0108] In the embodiments of this application, all data corresponding to a certain type of information in historical waybill information can be removed, rather than removing a single missing data. Furthermore, different missing percentage thresholds can be set for different types of data according to actual circumstances.

[0109] 32. Use preset filling methods to supplement missing data in historical waybill information.

[0110] In the embodiments of this application, missing data in historical waybill information can also be supplemented. Specifically, for example, regarding the feature of "goods processing type", 1 represents parcel exchange and 2 represents parcel return. For most parcels, neither of these two processing types may exist. In this case, based on business experience, the parcels without values ​​in "goods processing type" can be grouped into one category, i.e., missing values ​​are filled in.

[0111] 33. Remove abnormal data from historical waybill information.

[0112] After removing data with a high proportion of missing information and supplementing some missing data, it is also necessary to process the abnormal data in the historical waybill information.

[0113] Abnormal data can include incorrectly recorded data or negative data that is impossible in reality. Specifically, negative numbers arising from financial logic can be removed; alternatively, abnormal data can be removed based on common sense, such as removing data that does not meet the density requirements of actual objects based on a reference density table.

[0114] In one specific embodiment, the chargeable weight of a single shipment can be set to be no more than 195kg and no less than 0.4kg; and the revenue per shipment can be no less than 0.1 yuan. If the chargeable weight of a certain shipment is less than 0.4kg, the chargeable weight of that shipment is removed; or if the chargeable revenue of a certain shipment is less than 0.1 yuan, the chargeable revenue data of that shipment is removed.

[0115] In another embodiment, for 3C product carriers, the common materials of 3C products are mostly plastic or metal. The density of plastic materials such as engineering plastics (ABS) and polycarbonate is below 1.5 g / cm3; while the density of metals such as magnesium-aluminum alloy and zinc alloy is less than 1.8 g / cm3 and 6.4 g / cm3, respectively. Therefore, about 1.5 times the maximum value, i.e., 10 g / cm3, is selected as the density threshold of such carriers.

[0116] A computer, a 3C product, was being shipped via a waybill. It weighed 15kg and had a volume of 640g / cm3. After calculation, its density was found to be 23.4g / cm3, which was far greater than the set threshold and even exceeded the density of platinum (21.45g / cm3). Based on common sense, it was concluded that the data was abnormal and therefore the item was rejected.

[0117] After processing historical waybill information to obtain the first waybill information excluding abnormal data, it is necessary to determine the target express shipment characteristics that affect the shipment volume based on the first waybill information. For example, generally speaking, the heavier the shipment, the larger its volume; this indicates that the weight of the shipment is a shipment characteristic affecting its volume. Of course, the volume of the shipment is not directly proportional to its weight, and other shipment characteristics can also affect its volume. Therefore, it is necessary to identify the most important shipment characteristics that affect the shipment volume from among the multiple shipment characteristics corresponding to the shipment, i.e., the target shipment characteristics.

[0118] like Figure 4 The diagram shown is a flowchart illustrating an embodiment of determining the characteristics of a target express shipment, as provided in this application. It may include:

[0119] 41. Based on multiple basic characteristics of express shipments, determine the initial characteristics of express shipments that affect their volume, and obtain multiple initial characteristics of express shipments.

[0120] In the embodiments of this application, multiple basic characteristics of the express shipment can be obtained based on the first waybill information, and multiple initial characteristics affecting the shipment volume can be further obtained based on these multiple characteristics. However, some characteristics have a significant impact on the shipment volume, while others have a smaller impact. This application mainly focuses on the characteristics that have a significant impact on the shipment volume, which can effectively reduce the computational load of subsequent shipment volume prediction.

[0121] The initial express delivery features obtained in this application embodiment can be divided into single express delivery features and cross-features. Single express delivery features can be further categorized by data type into continuous express delivery features, ordinal express delivery features, and category express delivery features. A single express delivery feature can be a basic express delivery feature; that is, the basic features of an express delivery are its single features. Cross-features can be express delivery features obtained by fusing multiple single express delivery features.

[0122] In some embodiments, the continuity characteristics of express shipments may include: calculating statistical indicators such as mean, median, mode, maximum, minimum, quantile, kurtosis, skewness, variance, and standard deviation from data such as chargeable weight, chargeable revenue, and discounts of express shipments.

[0123] At the same time, it is also necessary to determine whether the calculated statistical indicators meet the long-tail distribution, so as to perform logarithmic transformation on the data that meet the long-tail distribution, and further standardize the calculated statistical indicators.

[0124] For express delivery, the revenue generated tends to follow a long-tail distribution, meaning that most of the revenue from express deliveries is concentrated at relatively low prices—the prices users typically pay when sending a package. However, some express deliveries may have significantly higher revenue due to added value-added services, such as insurance (the premium for which is based on the value of the goods), or other reasons, resulting in a substantial difference in revenue compared to the majority of deliveries.

[0125] Such data is usually small in quantity and disproportionate to the larger amount of low-priced billing revenue. When these isolated extreme outliers exist in the data, taking the logarithm of this data protects the normal data. Without taking the logarithm, subsequent parcel volume prediction models may be hijacked by these isolated extreme outliers, resulting in inaccurate volume predictions.

[0126] In the embodiments of this application, the calculated statistical indicators are standardized to unify the data and facilitate subsequent processing. Standardization of the statistical indicators may specifically include subtracting the mean from the data and then dividing by the variance. This standardization method is applicable to normally distributed data. Furthermore, data obtained after logarithmic transformation of long-tailed distributions mostly exhibits a normal distribution; therefore, the aforementioned standardization method can be used to standardize the data.

[0127] Of course, in other embodiments of this application, other standardization methods can also be used to process the data, such as transformations based on maximum and minimum values, normalization transformations, etc. Specific standardization processes can be found in existing technologies, and are not limited in any way in this application.

[0128] The above embodiments describe obtaining the continuity characteristics of the express delivery, while the embodiments of this application also require obtaining the express delivery sequence number characteristics.

[0129] Among them, the ordinal feature of express delivery mainly refers to the distance area between the original sender and the destination based on information entropy sharing; that is, the distance of the express delivery is determined according to the sender's address and the destination, so as to classify the express delivery into different distance ranges.

[0130] Specifically, considering that the actual distance between the recipient address and the package address may be very close or very far, and that packages from farther away may affect packages from closer away, leading to the aforementioned long-tail distribution, this application embodiment needs to divide the packages into different distance ranges based on their distance to avoid the continuous variable values ​​being too sparse, thereby improving the fault tolerance and stability of the distance variable.

[0131] One approach is to classify parcels by distance using a boxing method. Common boxing methods can be divided into unsupervised boxing and supervised boxing. The most commonly used unsupervised boxing methods are equal-frequency boxing and equal-distance boxing. Equal-frequency boxing ensures that the number of samples in each box is equal, while equal-distance boxing ensures that the width of each box is equal. In the embodiments of this application, equal-frequency boxing divides all parcels into several categories with equal quantities; while equal-distance boxing groups parcels that are equidistant from each other into one category.

[0132] In a specific embodiment, based on real-world scenarios, express shipments with a distance of more than 800km are usually transported by air; therefore, based on business experience, 800km is divided into a distance segment.

[0133] In other embodiments of this application, information such as the delivery time type, product, origin, destination, region type, and mode of transport corresponding to the express shipment can also be processed. Specific processing methods may include various approaches such as label encoding, one-hot encoding, label binarization, and target encoding.

[0134] In some implementations, ordered, non-numerical discrete features, such as time-sensitivity types, are tagged with labels. Typically, in the logistics field, time-sensitivity types represent speed and are ordered rather than arbitrary. For example, time-sensitivity type T4 is faster than T6; therefore, such features are tagged with labels.

[0135] In one specific embodiment, assume there are three different timeframes: T4, T6, and T8; the three different timeframes represent different delivery speeds. The highest timeframe T4 is converted to 3, the next highest timeframe T6 is converted to 2, and the lowest timeframe T8 is converted to 1; in this way, the timeframe type of the express delivery is converted into data that the model can recognize.

[0136] For origin or destination data, since each address is independent and there is no distinction between size, this type of data is unordered; one-hot encoding can be used to process it.

[0137] In one specific embodiment, assuming there are four different regions—Beijing, Shanghai, Guangzhou, and Shenzhen—then the corresponding codes for these four regions would be [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], and [0, 0, 0, 1]. This transforms the different regions into data that the model can recognize. Furthermore, one-hot encoding handles unordered data, effectively addressing the problem of classifiers in the model struggling to process attribute data, and to some extent, also serves to expand the features.

[0138] Since one-hot encoding only uses 0 and 1 values, different types are stored in a vertical space. However, when the number of categories is large, the feature space becomes extremely large. When there are too many categories, we consider using target value encoding to process the data; specifically, this involves replacing a certain type of data. For example, consider products. There may be hundreds of products. This unordered data is not suitable for label encoding, and one-hot encoding would result in overly sparse data. Therefore, target value encoding can be used to process this type of product data. For example, the average, maximum, minimum, and median shipment volumes of hundreds of products can be used as features determined by the target value encoding.

[0139] The above embodiments define individual features of express shipments. However, considering actual business scenarios, express shipments may also have overlapping features. For example, the packaging box size for a certain type of goods shipped by an e-commerce customer is fixed. Therefore, we can consider the overlapping features of the customer and the type of goods shipped, that is, the waybill volume of the customer + the type of goods shipped can be used as an overlapping feature.

[0140] like Figure 5 The diagram shown is a schematic representation of an embodiment of the cross feature provided in this application. Figure 5 Specifically, this refers to a textile customer whose main products are clothing and footwear. Since the products shipped by this textile customer are of similar size, the volume of the express packages sent by this textile customer is also the same.

[0141] For example Figure 5 In this case, the textile customer mainly shipped clothing and footwear, and the waybill volume was consistently 50*50*40=10000cm². 3 The packaging box is the size of the item. Then the textile can be packaged with a 10000cm box. 3 The volume is bound together, which is the cross feature of the express delivery.

[0142] In the above embodiments, both the sender and the item being shipped are considered. For example, the textile company's main business is quilts, which are categorized as clothing and footwear items; while the textile company occasionally sends contract documents, which are categorized as documents and tickets. Based on the statistics of the textile company's items, the volume of express shipments in the subsequent training set can be statistically analyzed according to the dimensions of customer + item, such as average, median, mode, maximum and minimum values.

[0143] In actual business scenarios, it was found that the volume of express packages under this cross dimension is 10000 cm³. 3 Assuming the textile company's business structure remains largely unchanged, and if the goods shipped by the textile company are still clothing and footwear, the volume prediction model will likely determine the volume corresponding to the waybill to be 10,000 cm³.3 .

[0144] In the above embodiments, judging the clothing and footwear consignment alone, or judging the consignment corresponding to the textile company alone, cannot accurately predict the volume. However, if the textile company and the clothing and footwear consignment are combined as cross features, a more accurate volume prediction can be obtained.

[0145] The cross features in the embodiments of this application can achieve a 1+1>2 effect; through cross features, second-order or even higher-order combination information is obtained, thereby improving the fitting ability of complex relationships and improving the accuracy of the volume prediction model.

[0146] like Figure 6 The diagram shown illustrates another embodiment of the cross-features provided in this application. In actual logistics delivery scenarios, a parcel's receiving point may be located within an industrial park / commercial area, and most parcels dispatched from this industrial park / commercial area are packaged in similar sizes. Therefore, the binding of products and consigned items is considered to obtain the cross-features of the parcel.

[0147] specific Figure 6 As shown, all shipments dispatched from this distribution center are washing machines, a type of household appliance. Although there are different manufacturers and various brands of washing machines near the distribution center, their sizes are similar. Therefore, the logistics company places washing machines from different manufacturers into boxes of the same size, ensuring that all packages dispatched from this distribution center have the same volume. Thus, the distribution center can be compared with... Figure 6 120000cm 3 The volume is bound together, which is the cross feature of the express delivery.

[0148] By combining actual logistics and transportation scenarios, more cross-features of express shipments can be obtained; the above embodiments are merely examples. This application does not limit the cross-features of express shipments in any way.

[0149] It should be noted that the above-mentioned processing of data in historical waybill information is actually converting various data corresponding to the express shipment into data that the model can recognize and calculate, which is also to facilitate subsequent model training.

[0150] 42. Sort the multiple initial express delivery features to obtain sorted initial express delivery features.

[0151] In the above embodiments, multiple initial package volume features that may affect the package volume were identified, including basic features and cross features. However, not all package features affect the package volume, and some features have a relatively small impact. If all initial package features were considered, a large amount of computation would occur, affecting the efficiency of package volume prediction. Therefore, in the embodiments of this application, it is also necessary to screen multiple initial package features and then sort the screened initial package features.

[0152] Specifically, in the embodiments of this application, there may be hundreds of initial express delivery features initially acquired. However, not all express delivery features may be of significant value in predicting express delivery volume. Therefore, it is necessary to perform an initial screening of the initial express delivery volume to select "good" express delivery features. For a single initial express delivery feature, its linear and non-linear indices can be determined. The linear index represents the correlation of a single initial express delivery feature, while the non-linear index represents the correlation of a single initial express delivery feature.

[0153] In the embodiments of this application, correlation can be used to represent a linear relationship between a variable and volume, while association represents a non-linear relationship between a variable and volume. Predictive ability can be described to some extent by target correlation; changes in strongly correlated features will lead to changes in the predicted value, while changes in weakly correlated features may not affect the predicted value. Common correlation indicators include Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient. Common association indicators include mutual information (MI), maximum information coefficient (MIC), and information content (IV).

[0154] In the above embodiments, the calculated correlation coefficient and other indicators can be used to determine whether the initial express delivery characteristics need to be retained. In a specific embodiment, the Pearson correlation coefficient can be used, and a chi-square test can be performed to identify irrelevant express delivery characteristics.

[0155] Specifically, the first step is to calculate the relevant indicators. After obtaining the indicators, the correlation levels are referenced based on engineering experience ([0, 0.2] is very weak, [0.2, 0.4] is weak, [0.4, 0.6] is moderate, [0.6, 0.8] is strong, and [0.8, 1] is very strong). For example, the correlation between waybill revenue and waybill volume is approximately 0.6, which is a moderate or strong correlation, so it is retained. However, the correlation between value-added service revenue and waybill volume is approximately 0.1, which is a very weak correlation, so it is removed.

[0156] Of course, there is no "golden ratio" for deciding whether to remove all features. In practice, it requires continuous experimentation, combining model results with calculated feature importance to determine whether a feature should be removed. Similarly, the chi-square test works as follows: a larger chi-square value indicates a greater deviation between the actual and theoretical values; conversely, a smaller value indicates a smaller deviation. If both values ​​are exactly equal, the chi-square value is 0, indicating that the actual value perfectly matches the theoretical value. Feature selection is then based on the calculated values ​​combined with engineering experience.

[0157] In the above embodiment, multiple initial express delivery features were identified using indicators such as correlation coefficients for initial screening, removing some initial express delivery features that would not affect the express delivery volume. However, several initial express delivery features still remained that would affect the express delivery volume. Different express delivery features have different weights in their influence on the express delivery volume; therefore, it is necessary to sort the multiple initial express delivery features to obtain sorted initial express delivery features.

[0158] 43. Among the multiple initial express features after sorting, determine the target express feature that affects the express volume.

[0159] Since the number of initial express delivery features is usually large, but some express delivery features have little impact on the size of the express delivery, in this embodiment of the application, only the target express delivery features that have a greater impact on the size of the express delivery are determined from the sorted initial express delivery features.

[0160] The target package features are usually multiple; however, the number of target package features may vary in different embodiments.

[0161] like Figure 7 The diagram shown is an embodiment of the target express delivery feature provided in this application. Figure 7 The table selects the top ten express delivery features that significantly influence package volume; the horizontal axis represents different express delivery features, while the vertical axis represents the importance of each feature to package volume, i.e., the degree to which each feature affects package volume. Figure 7 In China, the chargeable weight of a parcel is one of the main factors affecting its volume, while other characteristics such as the type of item being shipped, the originating region, and additional shipping also affect the volume of the parcel.

[0162] Based on the above embodiments, after determining the main target express features that affect the volume of the express package, it is also necessary to model the target express package features to obtain a target express package volume prediction model, and use the target express package volume prediction model to predict the volume of the express package.

[0163] like Figure 8 The diagram shown is a flowchart illustrating an embodiment of the express mail volume prediction model provided in this application, which may include:

[0164] 81. Obtain the volume prediction model for the first express shipment.

[0165] 82. Combine the first express shipment volume prediction model with the target express shipment characteristics to obtain the second express shipment volume prediction model.

[0166] 83. Obtain a training set of parcel volume data to train a second parcel volume prediction model and obtain a target parcel volume prediction model.

[0167] Specifically, an initial, untrained first parcel volume prediction model can be obtained first. Then, the target parcel features that affect parcel volume are combined with the first parcel volume prediction model to obtain a second parcel volume prediction model. This allows the target parcel features to influence the model training process when training the second parcel volume prediction model.

[0168] In the embodiments of this application, the first parcel volume prediction model can be a tree model LightGBM. The tree model LightGBM includes multiple leaf nodes, and each leaf node can correspond to a target parcel feature. The first parcel volume prediction model is then combined with the target parcel feature to obtain a second parcel volume prediction model that includes the target parcel feature. In the second parcel volume prediction model, the weights corresponding to different leaf nodes are all different. The process of training the second parcel volume prediction model is actually the process of determining the weights corresponding to different leaf nodes (i.e., different target parcel features).

[0169] In the embodiments of this application, when training the second express delivery volume prediction model using a training set, different volume prediction models can be trained according to different training sets. Specifically, the training set can be divided into a first training set based on the delivery area and a second training set based on the item being shipped, so that the second express delivery volume prediction model can be trained using the first training set and the second training set respectively.

[0170] like Figure 9 The diagram shown is a flowchart illustrating an embodiment of the target parcel volume prediction model provided in this application, which may include:

[0171] 91. Divide the training set into a first training set based on the delivery area and a second training set based on the item being delivered.

[0172] In the embodiments of this application, the training set includes the data obtained after processing historical waybill information in the aforementioned embodiments. Furthermore, considering the large number of waybills, the high degree of autonomy in managing waybills in different regions, and the different management methods among regions, the training set can be divided based on different regions, and the second express shipment volume prediction model can be trained to obtain different volume prediction models corresponding to different regions.

[0173] It should be noted that when the training set is divided based on different regions, the regions can be East China, North China, South China, etc.; and this application embodiment does not consider express shipments to Hong Kong, Macao and Taiwan, as well as international express shipments.

[0174] Furthermore, considering the uneven distribution of consignment samples in actual logistics scenarios, the training set is divided based on the existing primary classification of consignments to obtain different volume prediction models corresponding to different consignments. In the embodiments of this application, the primary classification of consignments can be 22 types; the secondary classifications under the primary classification can be several hundred types. Considering the robustness of the model, the embodiments of this application do not use secondary classification for classification training to prevent the model from overfitting.

[0175] 92. Use the first training set to train the second express volume prediction model to obtain the third express volume prediction model based on the delivery area.

[0176] 93. The second express volume prediction model is trained using the second training set to obtain the fourth express volume prediction model based on the consigned item.

[0177] 94. By integrating the third and fourth express delivery volume prediction models, a target express delivery volume prediction model is obtained.

[0178] After training the third and fourth parcel volume prediction models, it is necessary to fuse the two parcel volume prediction models to avoid inaccurate prediction results from a single parcel volume prediction model.

[0179] It should be noted that, in the embodiments of this application, the loss functions corresponding to the third express volume prediction model and the fourth fast-forward volume prediction model can be the same, both being: Where n is the number of waybills and y is the actual volume of the express shipment. This is the volume of the package predicted by the model.

[0180] In one specific embodiment, a shipment of mobile phones originating from Beijing will be processed using two different models for volume prediction: one for the North China region and the other for 3C products. Similarly, a shipment of automotive supplies from Guangzhou will be processed using both the South China region's volume prediction model and the primary automotive supplies category's volume prediction model.

[0181] In the above embodiments, the two types of parcel volume prediction models use the same loss function, but different training sets during training. For example, the parcel volume prediction model corresponding to a region selects all waybills for a certain region, regardless of the type of consignment; while the parcel volume prediction model corresponding to a specific type of consignment selects all waybills for a certain type of consignment, regardless of the sender's address. Each parcel volume prediction model focuses on a different aspect, and fusing different parcel volume prediction models can create complementary advantages.

[0182] The table below shows the volume prediction deviation rate of the tree model based on consigned items provided in the embodiments of this application. The table shows that consigned items are divided into several different categories, such as clothing and footwear, groceries and dried goods, industrial products, and household goods; and the predicted volume deviation rate for a single shipment corresponds to a different category.

[0183] First-level classification of consignment items Deviation rate of volume of single express shipment All items shipped are random. 0.308 Fresh food 0.105 Clothing and footwear 0.099 Dried food and other non-staple foods 0.243 Industrial production 0.348 Home furnishings 0.271 3C products 0.283 Culture, Sports and Entertainment 0.211 Medical and health care products 0.271 Personal care and cosmetics 0.277 Documents and receipts 0.378 Furniture 0.182 Bags 0.190 Home appliances 0.154 Home decoration and building materials 0.248 Agricultural production 0.220 Alcoholic beverages 0.246 Car accessories 0.296 Watches and jewelry 0.266 Maternity and baby products 0.140 Art category 0.264 Model the average value of each item entrusted for deposit. 0.235

[0184] Among these, documents and tickets are relatively small in both weight and volume, so even a slight deviation in the predicted volume can result in a significant deviation rate. For industrial production shipments, the density varies considerably; some shipments have a higher density, leading to a larger chargeable weight, even though their volume may not be large. Therefore, industrial production shipments also tend to have a higher volume deviation rate.

[0185] Furthermore, as can be directly seen from the table above, the volume deviation rate of most of the consignments obtained by predicting the volume of the consignments separately according to the primary classification is less than the volume deviation rate obtained by predicting the volume of all consignments together.

[0186] The table below shows the volume prediction deviation rate of the tree model based on the region provided in this application embodiment. As can be seen from the table, the packages are divided into different regions based on their sender addresses, including the South China Region, North China Region, East China Region, and Central and Western China Region; the volume deviation rate varies depending on the region.

[0187] Region Deviation rate of volume of single express shipment South China Region 0.299 North China Region 0.229 East China Region 0.275 Central and Western Region 0.234 Modeling averages separately for each major region 0.259

[0188] Furthermore, the table also shows that the volume deviation rate of most express parcels obtained by dividing them into different regions and predicting their volume separately is less than the volume deviation rate obtained by predicting the volume of all express parcels together.

[0189] Taking document and ticket parcels, which have the largest volume deviation rate among the primary categories of consigned items, as an example, the table below shows the final volume deviation rate of document and ticket parcels obtained by combining the volume deviation rate based on the consigned item and the volume deviation rate based on the region.

[0190] Region Deviation rate of volume for single express shipments of documents and tickets South China Region 0.360 North China Region 0.366 East China Region 0.365 Central and Western Region 0.360 Modeling averages separately for each major region 0.363

[0191] In the table above, by combining the parcel volume prediction model based on consignment items and the parcel volume prediction model based on regional distribution, the average deviation rate of the parcel volume for single parcels of documents and tickets is 0.363, while the average deviation rate for single parcels of documents and tickets considering only consignment items is 0.378. Since 0.378 is greater than 0.363, it can be seen that the deviation rate of the parcel volume after combining the models is less than the deviation rate of the parcel volume considering only consignment items, meaning that the parcel volume predicted by combining the models is more accurate.

[0192] Meanwhile, in the embodiments of this application, the volume of a package can be determined based on the volume deviation rate obtained from different package volume prediction models. In one specific embodiment, a shipment of mobile phones from Beijing will be divided into two package volume prediction models: one for the North China region and the other for 3C products. After obtaining two volumes, V1 and V2, using the two different package volume prediction models, the final predicted volume of the mobile phone is obtained by weighting the product of the inverse of the deviation rate.

[0193] For example, if V1 is 6000 and V2 is 4000, the table shows that the bias rate of the model for the North China region is 0.229, and the bias rate for the 3C product category is 0.283. At this point, the weighting rules will become... as well as The final weighted predicted volume of the phone is 5105.

[0194] In the above embodiments, the volume of the express shipment is predicted using a target shipment volume prediction model. In embodiments of this application, the predicted shipment volume can be further used to plan actual shipment transportation. Embodiments of this application also provide a shipment transportation planning method, which may include:

[0195] Identify the set of parcels in the target vehicle, which includes multiple parcels to be transported; predict the volume of the parcels in the parcel set using a target parcel volume prediction model; calculate the predicted vehicle volume loading rate for multiple parcels in the parcel set; and adjust the parcels in the target vehicle based on the predicted vehicle volume loading rate.

[0196] Meanwhile, before calculating the predicted vehicle volume loading rate for multiple express items in the express item set, the express item transportation method may further include: obtaining the express items transported by different vehicles each time within a historical period, so as to obtain the historical vehicle volume loading rate corresponding to each transport by different vehicles; using the historical vehicle volume loading rate, training a preset vehicle volume loading rate prediction model to obtain a target vehicle volume loading rate prediction model.

[0197] Once the target vehicle volume loading rate prediction model is obtained, it can be used to calculate the predicted vehicle volume loading rate values ​​for multiple express items in the express item collection.

[0198] In the embodiments of this application, different vehicles transport different packages, and each vehicle has a different volume. Furthermore, considering the different ways drivers place packages, and the principle of placing larger packages on top of smaller ones, heavier packages on top of lighter ones, and square packages on top of round ones, the actual vehicle volume loading rate varies for each vehicle.

[0199] Therefore, we can first obtain all the parcels transported by different vehicles for each trip within a historical period. It should be noted that this calculation is performed separately for each vehicle, obtaining all the parcels transported by each vehicle for each trip. After obtaining all the parcels transported by each vehicle for each trip, we can also use methods such as radar ranging or laser reflection to determine the sum of the visual volumes corresponding to all parcels. Because different parcels will have some gaps when placed, the sum of the visual volumes is not simply the sum of the volumes of all the parcels.

[0200] After determining the sum of the visually estimated volumes of all packages transported by a vehicle in a single trip, the vehicle's own volume can be determined; subsequently, the historical vehicle volume loading rate for each trip can be determined. Specifically, this can be expressed as: Historical vehicle volume loading rate = Sum of visually estimated volumes / Vehicle volume.

[0201] Since the sum of the visually estimated volumes is not the sum of the volumes of all packages, meaning there is a difference between the sum of the visually estimated volumes and the sum of the actual package volumes, this application defines a placement coefficient to determine the relationship between the sum of the visually estimated volumes and the sum of the actual package volumes. In this application's embodiments, the placement coefficient can be: Placement coefficient = Sum of actual package volumes / Sum of visually estimated volumes.

[0202] In actual parcel placement scenarios, different operators will use different placement methods, resulting in different placement coefficients for each vehicle and each transport. Therefore, this application embodiment needs to determine a relatively fixed placement coefficient to plan the parcels to be transported by different vehicles.

[0203] Specifically, a preset vehicle volume prediction loading rate prediction model can be obtained, and the model can be trained using multiple historical vehicle volume loading rates obtained in the aforementioned embodiments as a training set to obtain a target vehicle volume loading rate prediction model. The process of training the preset vehicle volume prediction loading rate prediction model is essentially the process of determining the loading coefficient.

[0204] After determining the loading coefficient, the set of parcels corresponding to the target vehicle can be obtained. The volume of each parcel can then be predicted using the target parcel volume prediction model, resulting in the sum of the predicted parcel volumes for that vehicle. Furthermore, based on the loading coefficient, the predicted volume loading rate for that vehicle can be determined.

[0205] The set of parcels corresponding to different vehicles can be determined based on information such as the parcel's delivery address and delivery time type. Existing technologies can be referenced to determine the set of parcels corresponding to different vehicles; no specific limitations are specified here.

[0206] The predicted parcel volume loading rate can be: in, This is the sum of the predicted parcel volumes corresponding to all parcels in the actual vehicle.

[0207] The above embodiments only obtain the predicted volume loading rate, which is a predicted value. In the actual express delivery loading process, there are many interfering factors that cause the actual express delivery volume loading rate to change. Therefore, this application also defines a vehicle total volume deviation rate to determine whether the predicted express delivery volume is accurate and to reallocate the express delivery to be transported by each vehicle.

[0208] Wherein, the vehicle total volume deviation rate = (σ t *Predicted parcel volume loading rate - Actual parcel volume loading rate) / Number of parcels loaded per truck * 100%.

[0209] σ t This refers to the placement coefficient described in the previous embodiments. The predicted parcel volume loading rate is actually also affected by the placement coefficient, so it needs to be multiplied by the placement coefficient to obtain the final predicted parcel volume loading rate.

[0210] The method described in this application predicts the volume of each package and the overall vehicle volume loading rate. Both predictions need to be combined to plan vehicle transportation. Furthermore, in predicting the overall vehicle volume loading rate, this application utilizes the aforementioned target package volume prediction model to predict the volume of packages loaded onto the vehicle.

[0211] like Figure 10 The diagram shown is a flowchart illustrating an embodiment of the express mail volume prediction and vehicle planning provided in this application. Figure 10 First, we can obtain the historical waybill information for all express shipments within the past 45 days and determine the volume of each shipment. At the same time, we need to clean up and remove abnormal data from the historical waybill information.

[0212] After obtaining the initial waybill information excluding abnormal data, multiple initial express shipment features can be extracted and classified from the initial waybill information. Then, target express shipment features with a significant impact on shipment volume are identified from these initial features. Machine learning validation is then performed on the target express shipment features to obtain a shipment volume prediction model capable of predicting shipment volume.

[0213] After obtaining the parcel volume prediction model, the volume of a single parcel can be predicted by inputting the corresponding waybill information. Furthermore, by combining the parcel sets corresponding to each vehicle, the sum of the predicted parcel volumes for all parcels transported by each vehicle can be determined.

[0214] Due to variations in placement methods, the sum of the actual vehicle volume and the package volume may not perfectly match. Therefore, it's necessary to verify whether the package loading rate meets the requirements. Specifically, this can be achieved by first obtaining the actual volume loading rate for each vehicle over the past 45 days. The actual loading rate can be determined by summing the vehicle volume and the visually estimated volume of the packages obtained using ranging radar or laser reflection. Multiplying the visually estimated volume by a placement coefficient and then dividing by the actual vehicle volume yields the vehicle's actual loading rate.

[0215] Since the placement coefficient varies for each vehicle during transport, it is necessary to train a pre-defined vehicle volume loading rate prediction model to obtain a target vehicle volume loading rate prediction model, thus achieving a relatively stable placement coefficient. At this point, the predicted volume loading rate can be determined based on the sum of the vehicle's actual volume, the placement coefficient, and the predicted volume of the parcels. Furthermore, the number of parcels transported by each vehicle can be adjusted according to the predicted volume loading rate.

[0216] This application also provides a device for predicting the volume of express mail, such as... Figure 11 The diagram shown is a schematic representation of an embodiment of a training device for a parcel volume prediction model provided in this application, which may include:

[0217] The acquisition module 1101 is used to acquire historical waybill information corresponding to all historical waybills within a preset historical time period.

[0218] The data processing module 1102 is used to process abnormal data in historical waybill information to obtain first waybill information; the first waybill information includes multiple basic characteristics of express shipments corresponding to the express shipments in the historical waybill.

[0219] The feature extraction module 1103 is used to determine the target express features that affect the express volume based on multiple basic express features in the first waybill information.

[0220] Training module 1104 is used to train the first express shipment volume prediction model based on the characteristics of the target express shipment, so as to obtain the target express shipment volume prediction model.

[0221] The express shipment volume prediction device provided in this application, after acquiring all historical waybill information corresponding to historical waybills, first processes abnormal data in the historical waybill information to obtain first waybill information excluding abnormal data; secondly, it determines the target express shipment features affecting the shipment volume from the first waybill information to model the target express shipment volume, obtaining a target express shipment volume prediction model, and then uses the target express shipment volume prediction model to predict the shipment volume. This application embodiment obtains machine learning samples by using the clearly defined waybill volumes in historical waybills and cleaning abnormal data. By acquiring the express shipment features affecting the shipment volume and inputting detailed information for each waybill, the volume of each waybill is obtained through prediction, which can effectively improve the prediction accuracy and efficiency of express shipment volume.

[0222] In some embodiments of this application, the data processing module 1102 can be specifically used to: remove data corresponding to features in historical waybill information where the missing proportion is greater than or equal to a preset missing proportion threshold; supplement missing data in historical waybill information using a preset filling method; and remove abnormal data in historical waybill information. The first waybill information includes multiple basic express delivery features corresponding to the express delivery in the historical waybill.

[0223] In some embodiments of this application, the feature extraction module 1103 may be specifically used to: determine the initial express features that affect the express volume based on multiple basic express features, and obtain multiple initial express features; sort the multiple initial express features to obtain sorted multiple initial express features; and determine the target express feature that affects the express volume among the sorted multiple initial express features.

[0224] In some embodiments of this application, the training module 1104 may be specifically used to: obtain a first express delivery volume prediction model; combine the first express delivery volume prediction model with the target express delivery features to obtain a second express delivery volume prediction model; obtain a training set including the express delivery volume, so as to train the second express delivery volume prediction model using the training set to obtain a target express delivery volume prediction model.

[0225] In some embodiments of this application, the training module 1104 may further be used to: divide the training set into a first training set based on the delivery area and a second training set based on the consigned item; train the second express volume prediction model using the first training set to obtain a third express volume prediction model based on the delivery area; train the second express volume prediction model using the second training set to obtain a fourth express volume prediction model based on the consigned item; and fuse the third express volume prediction model and the fourth express volume prediction model to obtain a target express volume prediction model.

[0226] This application also provides a fast mail transportation planning device, such as... Figure 12 The diagram shown is a schematic representation of an embodiment of the express delivery planning device provided in this application, which may include:

[0227] The determination module 1201 is used to determine the set of express parcels in the target vehicle, which includes multiple express parcels to be transported.

[0228] The prediction module 1202 is used to predict the volume of express items in the express item set using the target express item volume prediction model.

[0229] The calculation module 1203 is used to calculate the predicted vehicle volume loading rate for multiple express items in the express item collection.

[0230] Adjustment module 1204 is used to adjust the express parcels in the target vehicle based on the predicted value of the vehicle volume loading rate.

[0231] This application also provides an electronic device that integrates any of the express mail volume prediction devices or express mail transportation planning devices provided in this application. For example... Figure 13 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0232] The electronic device may include components such as a processor 1301 with one or more processing cores, a memory 1302 with one or more computer-readable storage media, a power supply 1303, and an input unit 1304. Those skilled in the art will understand that the electronic device structure shown in the figures does not constitute a limitation on the electronic device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0233] The processor 1301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 1302, and by calling data stored in the memory 1302, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 1301 may include one or more processing cores; preferably, the processor 1301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1301.

[0234] The memory 1302 can be used to store software programs and modules. The processor 1301 executes various functional applications and data processing by running the software programs and modules stored in the memory 1302. The memory 1302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 1302 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 1302 may also include a memory controller to provide the processor 1301 with access to the memory 1302.

[0235] The electronic device also includes a power supply 1303 that supplies power to various components. Preferably, the power supply 1303 can be logically connected to the processor 1301 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 1303 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0236] The electronic device may also include an input unit 1304, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0237] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 1301 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 1302 according to the following instructions, and the processor 1301 runs the applications stored in the memory 1302 to realize various functions, as follows:

[0238] Obtain historical waybill information corresponding to historical waybills within a preset historical time period; process abnormal data in the historical waybill information to obtain the first waybill information; determine the target express features affecting the express volume based on multiple basic express features; train the first express volume prediction model based on the target express features to obtain the target express volume prediction model for predicting the express volume.

[0239] Alternatively, determine the set of parcels in the target vehicle, which includes multiple parcels to be transported; predict the volume of the parcels in the parcel set using a target parcel volume prediction model, which can be any of the models mentioned above; calculate the predicted vehicle volume loading rate for multiple parcels in the parcel set; and adjust the parcels in the target vehicle based on the predicted vehicle volume loading rate.

[0240] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0241] Therefore, embodiments of this application provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in the training method of any of the express delivery volume prediction models provided in embodiments of this application. For example, the computer program loaded by the processor can execute the following steps:

[0242] Obtain historical waybill information corresponding to historical waybills within a preset historical time period; process abnormal data in the historical waybill information to obtain the first waybill information; determine the target express features affecting the express volume based on multiple basic express features; train the first express volume prediction model based on the target express features to obtain the target express volume prediction model for predicting the express volume.

[0243] Alternatively, determine the set of parcels in the target vehicle, which includes multiple parcels to be transported; predict the volume of the parcels in the parcel set using a target parcel volume prediction model, which can be any of the models mentioned above; calculate the predicted vehicle volume loading rate for multiple parcels in the parcel set; and adjust the parcels in the target vehicle based on the predicted vehicle volume loading rate.

[0244] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.

[0245] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.

[0246] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0247] The training method, transportation planning method, and apparatus for a parcel volume prediction model provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A training method for a package volume prediction model, characterized in that, The method includes: Retrieve historical waybill information corresponding to historical waybills within a preset historical time period; The abnormal data in the historical waybill information is processed to obtain the first waybill information, which includes multiple basic characteristics of the express shipments corresponding to the express shipments in the historical waybill. Based on multiple basic characteristics of express shipments in the first waybill information, the target express shipment characteristics that affect the shipment volume are determined; Train a first parcel volume prediction model based on the target parcel characteristics to obtain a target parcel volume prediction model that predicts the parcel volume. The step of training a first parcel volume prediction model based on the target parcel features to obtain a target parcel volume prediction model includes: acquiring a first parcel volume prediction model; combining the first parcel volume prediction model with the target parcel features to obtain a second parcel volume prediction model; acquiring a training set including parcel volume, dividing the training set into a first training set based on the delivery area and a second training set based on the parcel item; training the second parcel volume prediction model using the first training set to obtain a third parcel volume prediction model based on the delivery area; training the second parcel volume prediction model using the second training set to obtain a fourth parcel volume prediction model based on the parcel item; and fusing the third parcel volume prediction model and the fourth parcel volume prediction model to obtain the target parcel volume prediction model.

2. The training method for the express mail volume prediction model according to claim 1, characterized in that, The process of processing abnormal data in the historical waybill information to obtain first waybill information includes: Remove data from the historical waybill information that corresponds to features with a missing proportion greater than or equal to a preset missing proportion threshold. Missing data in the historical waybill information is supplemented using a preset filling method; Remove abnormal data from the historical waybill information; The first waybill information includes multiple basic characteristics of the express shipments corresponding to the historical waybills.

3. The training method for the express mail volume prediction model according to claim 2, characterized in that, The step of determining the target express shipment characteristics affecting the shipment volume based on multiple basic express shipment characteristics in the first waybill information includes: Based on the aforementioned multiple basic characteristics of express shipments, initial express shipment characteristics affecting the volume of express shipments are determined, resulting in multiple initial express shipment characteristics; The multiple initial express delivery features are sorted to obtain multiple sorted initial express delivery features; Among the sorted initial express features, the target express feature that affects the express volume is determined.

4. A method for express delivery planning, characterized in that, The method further includes: Identify the set of parcels in the target vehicle, wherein the set of parcels includes multiple parcels to be transported; The volume of the parcels in the parcel set is predicted using a target parcel volume prediction model, wherein the target parcel volume prediction model is a model trained by the training method of any one of claims 1-3. Calculate the predicted vehicle volume loading rate for multiple express items in the express item collection; Based on the predicted vehicle volume loading rate, adjust the parcels in the target vehicle.

5. The express delivery planning method according to claim 4, characterized in that, Before calculating the predicted vehicle volume loading rate for multiple express parcels in the express parcel set, the method further includes: The packages transported by different vehicles each time within the historical period are obtained to determine the historical vehicle volume loading rate for each transport by different vehicles. Using the historical vehicle volume loading rate, a preset vehicle volume loading rate prediction model is trained to obtain a target vehicle volume loading rate prediction model. The calculation of the predicted vehicle volume loading rate for multiple express parcels in the express parcel set includes: Using the target vehicle volume loading rate prediction model, the predicted vehicle volume loading rate values ​​for multiple express items in the express item set are calculated.

6. The express delivery planning method according to claim 5, characterized in that, The step of obtaining all express parcels transported by different vehicles each time within the historical period to determine the historical vehicle volume loading rate corresponding to each transport by different vehicles includes: Obtain the parcels transported by different vehicles each time within the historical period; Using a preset volume determination method, the sum of the visually estimated volumes of all express parcels transported by different vehicles each time is determined; Based on the sum of the vehicle volume of each vehicle and the visually estimated volume, the historical vehicle volume loading rate for each transport trip of each vehicle is determined.

7. A training device for a package volume prediction model, characterized in that, The device includes: The acquisition module is used to acquire historical waybill information corresponding to historical waybills within a preset historical time period; The data processing module is used to process abnormal data in the historical waybill information to obtain first waybill information, which includes multiple basic characteristics of express shipments corresponding to the express shipments in the historical waybill. The feature extraction module is used to determine the target express features that affect the express volume based on multiple basic express features in the first waybill information. The training module is used to train a first parcel volume prediction model based on the target parcel features to obtain a target parcel volume prediction model that predicts the parcel volume. The training module is also used to: acquire the first parcel volume prediction model; combine the first parcel volume prediction model with the target parcel features to obtain a second parcel volume prediction model; acquire a training set including parcel volume, and divide the training set into a first training set based on the delivery area and a second training set based on the parcel item; train the second parcel volume prediction model using the first training set to obtain a third parcel volume prediction model based on the delivery area; train the second parcel volume prediction model using the second training set to obtain a fourth parcel volume prediction model based on the parcel item; and fuse the third parcel volume prediction model and the fourth parcel volume prediction model to obtain the target parcel volume prediction model.

8. A parcel delivery planning device, characterized in that, The device includes: A determination module is used to determine the set of express parcels in the target vehicle, wherein the set of express parcels includes multiple express parcels to be transported; The prediction module is used to predict the volume of the express items in the express item set using a target express item volume prediction model, wherein the target express item volume prediction model is a model trained by the training method of the express item volume prediction model according to any one of claims 1-3. The calculation module is used to calculate the predicted vehicle volume loading rate for multiple express items in the express item collection; An adjustment module is used to adjust the parcels in the target vehicle based on the predicted vehicle volume loading rate.

Images