Express time limit prediction method and device, computer device and storage medium

Abstract

The application provides an express time limit prediction method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining express attribute information of a target express; performing feature engineering processing on the express attribute information to obtain express features of the target express; inputting the express features into a trained time limit prediction model to output prediction probability values of the target express under each preset time limit type; wherein the trained time limit prediction model is composed of a full connection layer, a dropout layer, a fusion layer and a normalization layer; and analyzing the prediction probability values to determine the logistics time limit of the target express. The method can improve the prediction accuracy of the express time limit.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of artificial intelligence technology, in particular to a method and device for predicting the time limit of express delivery, a computer device and a storage medium. BACKGROUND

[0002] In the field of logistics, the promised time limit of express delivery is usually roughly estimated according to business experience, which is collectively referred to as a method for predicting the time limit of express delivery, including but not limited to the following ways: (1) taking the time limit that 85% of the express delivery of a certain product flow to a representative network point in a city can achieve as the promised time limit of the product flow to all network points in the city; (2) dividing the life cycle of express delivery into six and a half segments in the form of collection end-network point-transit site-transit site-network point-delivery end, and calculating the 85% achievable time limit according to the daily shift plan of each network / transit site and the historical 14-day data, and then accumulating.

[0003] However, the first way has the disadvantage of giving a fixed promised time limit to the same city network without distinction, which may cause certain deviation, and the time limit of the representative network point selected according to the quantity of pieces may not be representative; the second way has the disadvantage that the historical 14-day data can only cover 70% of the flow, resulting in missing data during prediction, and the independent calculation of each segment leads to error accumulation, and the dependence of the order of the five tables and the two times of refreshing per week on the refresh results in not only a large consumption of machine resources but also the risk of the need for re-running of the subsequent link due to the failure of the scheduling of a certain link.

[0004] Therefore, the traditional method for predicting the time limit of express delivery has the problem of low prediction accuracy. SUMMARY

[0005] The purpose of the present application is to provide a method and device for predicting the time limit of express delivery, a computer device and a storage medium, to improve the prediction accuracy of the time limit of express delivery.

[0006] In a first aspect, the present application provides a method for predicting the time limit of express delivery, comprising:

[0007] obtaining the express delivery attribute information of a target express delivery;

[0008] performing feature engineering processing on the express delivery attribute information to obtain the express delivery features of the target express delivery;

[0009] inputting the express delivery features into a trained time limit prediction model to output the prediction probability values of the target express delivery under each preset time limit type; wherein the trained time limit prediction model is composed of a full connection layer, a dropout layer, a fusion layer and a normalization layer;

[0010] analyzing each prediction probability value to determine the logistics time limit of the target express delivery.

[0011] In some embodiments of the present application, the express features are input into the trained time limit prediction model, and the prediction probability value of the target express under each preset time limit type is output, including: merging the express features to obtain the merged express features; inputting the express features into the trained time limit prediction model to output the first correlation information within the features; and inputting the merged express features into the trained time limit prediction model to output the second correlation information between the features; analyzing the first correlation information and the second correlation information to obtain the prediction probability value of the target express under each preset time limit type.

[0012] In some embodiments of the present application, the express features are input into the trained time limit prediction model, and the first correlation information within the features is output, including: inputting the express features into the trained time limit prediction model, and classifying the express features through a full connection layer and a dropout layer to obtain an initial feature vector and a differentiation coefficient of the initial feature vector; fusing the initial feature vector and the differentiation coefficient through a fusion layer to obtain a target feature vector carrying the differentiation coefficient; and normalizing the target feature vector through a normalization layer to obtain the first correlation information within the features.

[0013] In some embodiments of the present application, the first correlation information and the second correlation information are analyzed to obtain the prediction probability value of the target express under each preset time limit type, including: obtaining a first feature matrix corresponding to the first correlation information, and a second feature matrix corresponding to the second correlation information; multiplying the first feature matrix and the second feature matrix to obtain a comprehensive feature; classifying the comprehensive feature through a target full connection layer carrying a sigmoid function to obtain the prediction probability value of the target express under each preset time limit type.

[0014] In some embodiments of the present application, before the express features are input into the trained time limit prediction model, it further includes: constructing an initial time limit prediction model; based on a preset initial learning rate, preliminarily training the initial time limit prediction model until the training times reach a preset first threshold to obtain a preliminarily trained time limit prediction model; based on a preset peak learning rate, horizontally training the preliminarily trained time limit prediction model until the training times reach a preset second threshold to obtain a horizontally trained time limit prediction model; and according to a trigonometric function curve of the peak learning rate, debugging training the horizontally trained time limit prediction model to obtain the trained time limit prediction model.

[0015] In some embodiments of the present application, the express attribute information is subjected to feature engineering to obtain express features of the target express, including: extracting address information, time information and product information included in the express attribute information; subjecting the address information to feature engineering to obtain node information and distance information as express features; and subjecting the time information to feature engineering to obtain festival information and express sending time as express features; and subjecting the product information to feature engineering to obtain product type information as express features.

[0016] In some embodiments of the present application, each prediction probability value is analyzed to determine the logistics time limit of the target express, including: screening out a maximum value in each prediction probability value to obtain a target prediction probability value; and determining a time limit type corresponding to the target prediction probability value as the logistics time limit of the target express.

[0017] In a second aspect, the present application provides an express time limit prediction device, including:

[0018] An information acquisition module is configured to acquire express attribute information of a target express;

[0019] A feature engineering module is configured to subject the express attribute information to feature engineering to obtain express features of the target express;

[0020] A model analysis module is configured to input the express features into a trained time limit prediction model to output prediction probability values of the target express under each preset time limit type; wherein the trained time limit prediction model is composed of a full connection layer, a dropout layer, a fusion layer and a normalization layer.

[0021] A time limit determination module is configured to analyze each prediction probability value to determine the logistics time limit of the target express.

[0022] In a third aspect, the present application further provides a computer device, including:

[0023] One or more processors;

[0024] A memory; and one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the processor to implement the express time limit prediction method of the first aspect.

[0025] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute steps in the express time limit prediction method.

[0026] In a fifth aspect, an embodiment of the present application provides a computer program product or computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to perform the method provided in the first aspect.

[0027] The express time prediction method, device, computer device and storage medium described above, the server obtains the express attribute information of the target express, and performs feature engineering processing on the express attribute information, so as to obtain the express features of the target express, and then inputs the express features into the trained time prediction model, so as to output the prediction probability value of the target express under each preset time type, so as to analyze each prediction probability value and determine the logistics time of the target express. Therefore, the time prediction model provided in the present application is used to predict the time of the target express, without needing to pay attention to the time consumption of the express transfer link, so as to ensure the stability of the prediction result and finally improve the prediction accuracy of the express time. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0029] Figure 1 The scene schematic diagram of the express time prediction method provided in the embodiments of the present application;

[0030] Figure 2 The flowchart of the express time prediction method provided in the embodiments of the present application;

[0031] Figure 3 The architecture application flowchart of the time prediction model provided in the embodiments of the present application;

[0032] Figure 4 The structure schematic diagram of the express time prediction device provided in the embodiments of the present application;

[0033] Figure 5 The structure schematic diagram of the computer device provided in the embodiments of the present application. DETAILED DESCRIPTION

[0034] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort are within the scope of the present application.

[0035] In the description of the present application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0036] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can realize the present application without using these specific details. In other examples, well-known structures and processes will not be described in detail in order to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed in the present application.

[0037] In the embodiments of the present application, the express mail time limit prediction method provided by the embodiments of the present application can be applied to, for example Figure 1The express time limit prediction system shown in the figure. Among them, the express time limit prediction system includes terminal 102 and server 104. The terminal 102 can be a device that includes receiving and transmitting hardware, that is, a device with receiving and transmitting hardware capable of performing bidirectional communication on a bidirectional communication link. Such devices can include cellular or other communication devices with single-line displays or multi-line displays or cellular or other communication devices without multi-line displays. The terminal 102 can be a desktop terminal or a mobile terminal, and the terminal 102 can also be one of a mobile phone, a tablet computer, and a notebook computer. The server 104 can be a stand-alone server, or a server network or server cluster composed of servers, including but not limited to computers, network hosts, single network servers, multiple network server sets, or cloud servers composed of multiple servers. Among them, the cloud server is composed of a large number of computers or network servers based on cloud computing (Cloud Computing). In addition, the terminal 102 and the server 104 establish a communication connection through a network, and the network can be any one of a wide area network, a local area network, and a metropolitan area network.

[0038] Those skilled in the art can understand that Figure 1 The application environment shown in the figure is only one application scenario applicable to the scheme of the present application, and does not constitute a limitation on the application scenarios of the scheme of the present application. Other application environments can include more or fewer devices than Figure 1 For example, Figure 1 Only 1 server is shown in the figure. It can be understood that the express time limit prediction system can also include one or more other servers, which are not specifically limited here. In addition, as shown in the figure, Figure 1 The express time limit prediction system can also include a memory for storing data, such as storing express attribute information of the target express.

[0039] It should be noted that Figure 1 The scenario diagram of the express time limit prediction system shown in the figure is only an example, and the express time limit prediction system and the scenario described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of the express time limit prediction system and the appearance of new business scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0040] Referring to Figure 2 The embodiments of the present application provide an express time limit prediction method, and the embodiments mainly take the server 104 in the above Figure 1 The method includes steps S201 to S204, and specifically as follows:

[0041] S201, obtain express attribute information of the target express.

[0042] The target express can be a to-be-shipped cargo selected for time prediction, and the to-be-shipped cargo is specifically a "express" with a relatively fast transport speed and a relatively high transport cost. It should be noted that the embodiments of the present application propose to use the target express for logistics time prediction, and the intention is not to distinguish the types of cargos in the logistics field, that is, not to distinguish "express" and "slow cargo" and the like, but to refer to the to-be-shipped cargos in the logistics field as "express", but if the "express" and "slow cargo" and the like are distinguished in other embodiments, the corresponding type of cargo information can be used for logistics time prediction.

[0043] The express attribute information can include address information, time information, and product information. The address information can include sending address information, receiving address information, and the like. The time information can include sending time, historical time consumption of the flow direction, and the like. The product information can include product type and the like.

[0044] In a specific implementation, in order to improve the prediction accuracy of the express time, the server 104 can first obtain the express attribute information of the target express to be predicted. The method for obtaining the express attribute information includes but is not limited to the following methods: 1. In a general network structure, the server 104 receives the express attribute information from the terminal 102 or other devices that have network connection; 2. In a pre-established blockchain network, the server 104 can synchronously obtain the express attribute information from other terminal nodes or server nodes. The blockchain network can be a public chain, a private chain, or the like; 3. In a pre-set tree structure, the server 104 can request the express attribute information from a superior server or poll the express attribute information from a subordinate server.

[0045] In addition, the server 104 can obtain the address information, time information, and product information of each express, and then integrate the express attribute information of the target express. At this time, the address information, time information, and product information of each express can be obtained by the same device or by multiple devices respectively.

[0046] S202, performing feature engineering processing on the express attribute information to obtain express features of the target express.

[0047] The feature engineering refers to a process of converting original data into training data of a model. The purpose is to obtain better training data features to make the machine learning model approach the upper limit. Feature engineering can improve the performance of the model, and sometimes even good results can be achieved on a simple model. Feature engineering plays a very important role in machine learning, and generally includes three parts of "feature construction, feature extraction, and feature selection".

[0048] In a specific implementation, the server 104 performs feature engineering processing on the express parcel attribute information, which can be regarded as feature design in practice. The feature design in the embodiment of the present application highlights the time information and static information, removes the influence of the date difference between the training set and the test set on the model, and improves the generalization ability of the model.

[0049] In one embodiment, the step includes: extracting address information, time information, and product information included in the express parcel attribute information; performing feature engineering processing on the address information to obtain network point information and distance information as express parcel features; and performing feature engineering processing on the time information to obtain holiday information and sending time as express parcel features; and performing feature engineering processing on the product information to obtain product type information as express parcel features.

[0050] In a specific implementation, the server 104 can perform feature engineering processing on the address information to obtain network point information of each of the receiving and sending ends, such as network point level, superior network point, network point type, district, large district, city representative network point, administrative district representative network point, and distance information such as city straight-line distance, city road distance, latitude and longitude, and distance type, which can all be used as express parcel features. The features obtained by the embodiment of the present application can include 11 numerical features and 27 category features, totaling 38 express parcel features.

[0051] For example, the network point of each of the receiving and sending ends can be a school, a building, a community office, etc.; the network point level can be divided into levels 1, 2, and 3 according to business rules, such as the size of the mail volume, the coverage area of the network point, and the number of staff; the network point type can include hospitals and schools; the district can include city districts such as Nanshan District and Futian District in Shenzhen; the large district can include Central China, South China, West China, and North China; and the city straight-line distance can be the straight-line distance between the receiving and sending ends on a map, and the distance type can include intra-city, intra-province, inter-province, and economic circle.

[0052] Further, the server 104 can perform feature engineering processing on the time information to obtain information such as whether it is a holiday, city road time consumption, and 75th, 80th, and 85th percentiles of historical time consumption of the flow direction, and convert the sending time in the format of “yyyy-MM-dd hh:mm:ss” into a float value in the format of “hh*3600+mm*60+ss”, which can all be used as express parcel features. The “yyyy-MM-dd hh:mm:ss” represents the conversion of time into 12-hour format, for example, “2018-06-27 03:24:21”.

[0053] Further, the server 104 can perform feature engineering on the product information to obtain product type information, such as express product types like express and special express, as express features. It should be noted that in the embodiment, all express features are "static features" describing the sending and receiving locations, and the sending and receiving location information is generally complete, and other features related to distance and historical time consumption can be directly obtained by associating with existing known information, so the data coverage rate can reach 100%, and there is no data missing problem, that is, the accuracy of the express time prediction will not be reduced.

[0054] In S203, the express features are input into the trained time prediction model to output predicted probability values of the target express under each preset time type.

[0055] The preset time type can be a time type composed of a time range, for example, the time types include "2d12", "2d18", "2d22", etc. "2d12" represents a time range of "today to tomorrow 12 o'clock", "2d18" represents a time range of "today to tomorrow 18 o'clock", and "2d22" represents a time range of "today to tomorrow 22 o'clock", and so on. It can be understood that the specific time range can be set according to actual business needs, and the present application does not make specific limitations.

[0056] In a specific implementation, after the server 104 analyzes and obtains the express features, the trained time prediction model can be called, and then the express features are input into the trained time prediction model for feature analysis, so that the trained time prediction model outputs the predicted probability values of the target express under each preset time type. In addition, before the server 104 calls the trained time prediction model, the time prediction model can be trained, and the present application proposes to use the learning rate automatic training of "straight line rise-> keep-> curve decline" to optimize the model performance. In this way, the model prediction target is only based on the information of the sending and receiving ends, and the various intermediate transfer links are blurred, the "overall time consumption" of the end-to-end inference of the express life cycle is obtained, that is, there is no need to calculate and superimpose the time consumption of each stage, avoiding the defect of mutual dependence of multi-link scheduling, and improving the prediction accuracy of the express time. The model training steps and model analysis steps involved in the embodiment will be described in detail below.

[0057] In an embodiment, the step includes: performing feature merging on the express features to obtain merged express features; inputting the express features into the trained time limit prediction model to output first correlation information within the features; and inputting the merged express features into the trained time limit prediction model to output second correlation information between the features; and analyzing the first correlation information and the second correlation information to obtain the predicted probability value of the target express under each preset time limit type.

[0058] In a specific implementation, reference can be made to Figure 3 The architecture application flowchart of the time limit prediction model provided in the embodiments of the present application adopts a unique network structure, specifically composed of a full connection layer, a dropout layer, a fusion layer, and a normalization layer, which can not only realize the abstraction of express features, but also automatically learn the difference coefficients of different dimensions of express features.

[0059] Specifically, after the server 104 inputs the 38 express features into the trained time limit prediction model, each express feature will sequentially pass through the full connection layer, the dropout layer, the fusion layer, and the normalization layer to obtain the difference information and the correlation information within the features as the first correlation information; at the same time, the time limit prediction model can be applied not only to the 38 independent express features, but also to the total express features after merging. That is to say, the server 104 can perform feature merging on the express features to obtain merged express features as total express features, and then input the total express features into the trained time limit prediction model for analysis, which will sequentially pass through the full connection layer, the dropout layer, the fusion layer, and the normalization layer to obtain the difference information and the correlation information between the features as the second correlation information. Finally, by analyzing the first correlation information and the second correlation information, the predicted probability value of the target express under each preset time limit type can be obtained.

[0060] In an embodiment, inputting the express features into the trained time limit prediction model to output first correlation information within the features includes: inputting the express features into the trained time limit prediction model, performing feature classification on the express features through the full connection layer and the dropout layer to obtain an initial feature vector and a difference coefficient of the initial feature vector; performing fusion on the initial feature vector and the difference coefficient through the fusion layer to obtain a target feature vector carrying the difference coefficient; and performing normalization processing on the target feature vector through the normalization layer to obtain the first correlation information within the features.

[0061] In a specific implementation, the server 104 can input the express features into the trained time limit prediction model, so that each express feature sequentially passes through a dense fully connected layer with a ReLu activation function, a dense fully connected layer without an activation function, a dropout layer, and a 16-dimensional dense fully connected layer, to obtain a 16-dimensional feature vector as an initial feature vector. At this time, if the initial feature vector further passes through a fully connected layer with a sigmoid function, the difference coefficient of each initial feature vector will be calculated.

[0062] Further, after the initial feature vector calculated by "dense-16" and the difference coefficient calculated by "Dense(Sigmoid)-16" are multiplied, the target feature vector with the difference coefficient is obtained. The mean and variance of the elements of the 38 target feature vectors in the same dimension are calculated using layer normalization technology, and normalization processing is performed, so that the distribution with a mean of "0" and a variance of "1" is obtained, and the "difference" and "correlation" information between the features are learned as the first association information.

[0063] In one embodiment, the first association information and the second association information are analyzed to obtain the prediction probability value of the target express under each preset time limit type, including: obtaining a first feature matrix corresponding to the first association information, and a second feature matrix corresponding to the second association information; multiplying the first feature matrix and the second feature matrix to obtain a comprehensive feature; and classifying the comprehensive feature through a target fully connected layer with a sigmoid function to obtain the prediction probability value of the target express under each preset time limit type.

[0064] In a specific implementation, the first association information is represented as internal abstract information of 38 express features, and the first feature matrix of "16*38" is actually output by the time limit prediction model; at the same time, the second association information is represented as external abstract information of 38 express features, and the first feature matrix of "16*1" is actually output by the time limit prediction model. Thereafter, the server 104 can transpose the first feature matrix so that the matrix dimension is "38*16", and then perform matrix multiplication with the second feature matrix through "matrix multiplication" to realize cross fusion of each express feature. Finally, the comprehensive feature obtained by matrix multiplication is subjected to a fully connected layer with a sigmoid function, and the time limit prediction model outputs the prediction probability value of the target express under each preset time limit type.

[0065] It can be understood that the activation function set in the embodiment of the present application is specifically used to add a nonlinear factor, improve the expression ability of the neural network to the model, solve the problems that cannot be solved by the linear model, use the ReLu function to improve the calculation speed and convergence speed, and use the Sigmoid function to transform the input continuous real value into the output between 0 and 1.

[0066] In an embodiment, before this step, it further comprises: constructing an initial time limit prediction model; based on a preset initial learning rate, preliminarily training the initial time limit prediction model until the number of training reaches a preset first threshold, obtaining a preliminarily trained time limit prediction model; based on a preset peak learning rate, horizontally training the preliminarily trained time limit prediction model until the number of training reaches a preset second threshold, obtaining a horizontally trained time limit prediction model; and debugging training the horizontally trained time limit prediction model according to a trigonometric function curve of the peak learning rate, obtaining the trained time limit prediction model.

[0067] In a specific implementation, the learning rate is usually kept unchanged in different iteration rounds of model training, for example, the learning rate is often set to 0.1, 0.01, 0.005, etc. A larger learning rate can easily make the model fall into a local optimal solution, and the loss fluctuates greatly. A smaller learning rate can make the model converge too slowly, and eventually cannot fully converge, still cannot find the global optimal solution. Therefore, the present application proposes a learning rate that automatically changes with the number of iterations, that is, the learning rate is automatically adjusted according to the "linear rise -> keep -> curve decline" rule, which can avoid large fluctuations, and when the loss is stable, the learning rate is accelerated, and when the model approaches the global optimal solution, the learning rate is reduced, and the optimal solution is approached steadily, avoiding the learning rate being too large to pass the optimal solution and causing the loss to deteriorate.

[0068] Specifically, the initial learning rate can be set to "0.001", which is linearly increased to the peak learning rate "0.004" for 2 rounds, and then gradually decreased according to the trigonometric function curve rule after 2 rounds of horizontal training, and the decrease speed can be slow at first and then fast, so that the trained time limit prediction model is obtained. In this way, the dynamic adaptive learning rate can be configured for different stages of model optimization, the steady convergence of the model to the global optimal solution is realized, and the situation that the model does not converge due to the fixed learning rate in the traditional scheme is avoided.

[0069] S204, analyzing each prediction probability value to determine the logistics time limit of the target express.

[0070] The logistics time limit can include the logistics commitment time, also known as the "estimated delivery time", for example, the logistics time limit of the target express "A" is "today to tomorrow 12:00", and the logistics time limit of the target express "B" is "today to tomorrow 18:00".

[0071] In a specific implementation, after the server 104 analyzes the prediction probability value of the target express under each preset time limit type based on the above steps, it can further analyze each prediction probability value to determine the logistics time limit of the target express, and the analysis method can be to compare the size of each prediction probability value.

[0072] In one embodiment, this step includes: screening out the maximum value in each prediction probability value to obtain a target prediction probability value; determining the time limit type corresponding to the target prediction probability value as the logistics time limit of the target express.

[0073] In a specific implementation, reference can be made to Figure 3 If a 13-dimensional fully connected layer with a sigmoid function is provided in the time limit prediction model, 13 prediction probability values are output, and the maximum value in the 13 prediction probability values can be selected as the logistics time limit of the target express. However, it can be understood that if an N-dimensional fully connected layer with a sigmoid function is provided in the time limit prediction model, N prediction probability values will be output, and N≥2.

[0074] In the express time limit prediction method in the above embodiments, the server can obtain express attribute information of a target express, perform feature engineering processing on the express attribute information, obtain express features of the target express, input the express features into a trained time limit prediction model, and output prediction probability values of the target express under each preset time limit type. By analyzing each prediction probability value, the logistics time limit of the target express can be determined. Therefore, the time limit prediction model provided in the present application can be used to predict the time limit of the target express, which can greatly increase the data coverage rate and avoid the influence of mutual dependence of multiple links, thereby improving the prediction accuracy of the express time limit in the best state of the model performance.

[0075] It should be understood that, although Figure 2 the steps in the flowchart are shown in order according to the arrows, these steps are not necessarily executed in order according to the arrows. Unless otherwise specified in this document, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, Figure 2 At least part of the steps in may include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0076] Figure 4 In order to better implement the express time limit prediction method provided in the embodiments of the present application, on the basis of the express time limit prediction method provided in the embodiments of the present application, an express time limit prediction device is further provided in the embodiments of the present application, as shown in

[0077] The information acquisition module 410 is configured to acquire express attribute information of a target express.

[0078] The feature engineering module 420 is configured to perform feature engineering on the express attribute information to obtain express features of the target express;

[0079] The model analysis module 430 is configured to input the express features into the trained time limit prediction model to output prediction probability values of the target express under each preset time limit type; the trained time limit prediction model is composed of a full connection layer, a dropout layer, a fusion layer, and a normalization layer;

[0080] The time limit determination module 440 is configured to analyze the prediction probability values to determine the logistics time limit of the target express.

[0081] In an embodiment, the model analysis module 430 is further configured to perform feature merging on the express features to obtain merged express features; input the express features into the trained time limit prediction model to output first correlation information within the features; input the merged express features into the trained time limit prediction model to output second correlation information between the features; and analyze the first correlation information and the second correlation information to obtain the prediction probability values of the target express under each preset time limit type.

[0082] In an embodiment, the model analysis module 430 is further configured to input the express features into the trained time limit prediction model, perform feature classification on the express features through the full connection layer and the dropout layer to obtain an initial feature vector and a differentiation coefficient of the initial feature vector; perform fusion on the initial feature vector and the differentiation coefficient through the fusion layer to obtain a target feature vector carrying the differentiation coefficient; and perform normalization processing on the target feature vector through the normalization layer to obtain the first correlation information within the features.

[0083] In an embodiment, the model analysis module 430 is further configured to obtain a first feature matrix corresponding to the first correlation information and a second feature matrix corresponding to the second correlation information; perform matrix multiplication on the first feature matrix and the second feature matrix to obtain a comprehensive feature; perform feature classification on the comprehensive feature through a target full connection layer carrying a sigmoid function to obtain the prediction probability values of the target express under each preset time limit type.

[0084] In an embodiment, the express time limit prediction device 400 further includes a model training module configured to construct an initial time limit prediction model; perform preliminary training on the initial time limit prediction model based on a preset initial learning rate until a training frequency reaches a preset first threshold to obtain a preliminary trained time limit prediction model; perform horizontal training on the preliminary trained time limit prediction model based on a preset peak learning rate until the training frequency reaches a preset second threshold to obtain a horizontally trained time limit prediction model; and perform debugging training on the horizontally trained time limit prediction model according to a trigonometric function curve of the peak learning rate to obtain the trained time limit prediction model.

[0085] In one embodiment, the feature engineering module 420 is further configured to extract address information, time information and product information included in the express parcel attribute information; perform feature engineering processing on the address information to obtain network point information and distance information as express parcel features; perform feature engineering processing on the time information to obtain holiday information and parcel sending time as express parcel features; and perform feature engineering processing on the product information to obtain product type information as express parcel features.

[0086] In one embodiment, the time limit determination module 440 is further configured to screen out a maximum value in each prediction probability value to obtain a target prediction probability value; and determine a time limit type corresponding to the target prediction probability value as a logistics time limit of the target express parcel.

[0087] In the above embodiments, the time limit prediction model proposed in the present application is used to predict the time limit of the target express parcel, which can greatly increase the data coverage rate and avoid the influence of mutual dependence of multiple links, and finally improve the prediction accuracy of the express parcel time limit in the best state of the model performance.

[0088] It should be noted that the specific limitations of the express parcel time limit prediction device can be referred to the limitations of the express parcel time limit prediction method in the above, which will not be repeated here. Each module in the above express parcel time limit prediction device can be realized by software, hardware and their combinations in whole or in part. The above modules can be embedded in or independent of the processor in the electronic device in hardware form, or can be stored in the memory in the electronic device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above modules.

[0089] In some embodiments of the present application, the express parcel time limit prediction device 400 can be realized in the form of a computer program, which can run on a computer device as shown in Figure 5 The memory of the computer device can store various program modules constituting the express parcel time limit prediction device 400, such as the information acquisition module 410, the feature engineering module 420, the model analysis module 430 and the time limit determination module 440 as shown in Figure 4 The computer program constituted by each program module makes the processor execute the steps in the express parcel time limit prediction method of each embodiment of the present application described in the specification. For example, Figure 5 The computer device as shown in Figure 4The information obtaining module 410 in the express time limit prediction device 400 shown performs step S201. The computer device can perform step S202 through the feature engineering module 420. The computer device can perform step S203 through the model analysis module 430. The computer device can perform step S204 through the time limit determination module 440. The computer device includes a processor, a memory and a network interface connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is configured to communicate with external computer devices through network connection. The computer program is executed by the processor to implement an express time limit prediction method.

[0090] Those skilled in the art can understand that, Figure 5 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0091] In some embodiments of the present application, a computer device is provided, including one or more processors, a memory, and one or more application programs, wherein the one or more application programs are stored in the memory and configured to execute the steps of the express time limit prediction method by the processor. The steps of the express time limit prediction method can be the steps in the express time limit prediction method of each of the above embodiments.

[0092] In some embodiments of the present application, a computer readable storage medium is provided, which stores a computer program. The computer program is loaded by the processor, so that the processor executes the steps of the express time limit prediction method. The steps of the express time limit prediction method can be the steps in the express time limit prediction method of each of the above embodiments.

[0093] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0094] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0095] The above has carried out the detailed introduction to the express time prediction method, device, computer equipment and storage medium provided by the embodiment of the present application, the principle and implementation mode of the present application have been described in this paper, the above embodiment is only used to help understanding the method and core idea of the present application; At the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed; In view of the above, the content of the specification should not be understood as the limitation of the present application.

Claims

1. A method for predicting the delivery time of express mail, characterized in that, include: Obtain the package attribute information of the target package, including address information, time information, and product information; The express delivery attribute information is subjected to feature engineering processing to obtain the express delivery features of the target express delivery, including: performing feature engineering processing on the address information to obtain network point information and distance information as the express delivery features; performing feature engineering processing on the time information to obtain holiday information and dispatch time as the express delivery features; and performing feature engineering processing on the product information to obtain product type information as the express delivery features. The process involves inputting the express delivery features into a trained timeliness prediction model and outputting the predicted probability value of the target express delivery under various preset timeliness types. This includes: merging the express delivery features to obtain merged express delivery features; inputting the express delivery features into the trained timeliness prediction model and outputting first correlation information within the features; inputting the merged express delivery features into the trained timeliness prediction model and outputting second correlation information between the features; and analyzing the first correlation information and the second correlation information to obtain the predicted probability value of the target express delivery under various preset timeliness types. The trained timeliness prediction model consists of a fully connected layer, a dropout layer, a fusion layer, and a normalization layer. Analyze the predicted probability values ​​to determine the delivery time of the target package.

2. The method as described in claim 1, characterized in that, The step of inputting the express delivery features into the trained timeliness prediction model and outputting the first correlation information within the features includes: The express delivery features are input into a trained time-delivery prediction model. The express delivery features are classified through the fully connected layer and the dropout layer to obtain an initial feature vector and the difference coefficient of the initial feature vector. The initial feature vector and the differentiation coefficient are fused through the fusion layer to obtain a target feature vector carrying the differentiation coefficient; The target feature vector is normalized by the normalization layer to obtain the first correlation information within the feature.

3. The method as described in claim 1, characterized in that, The analysis of the first and second association information to obtain the predicted probability value of the target express shipment under various preset time-sensitive types includes: Obtain the first feature matrix corresponding to the first association information, and the second feature matrix corresponding to the second association information; Multiply the first feature matrix by the second feature matrix to obtain the comprehensive features; By using a target fully connected layer carrying a sigmoid function, the comprehensive features are classified to obtain the predicted probability value of the target express delivery under each preset time-sensitive type.

4. The method according to any one of claims 1 to 3, characterized in that, Before inputting the express delivery features into the trained time-delivery prediction model, the method further includes: Construct an initial timeliness prediction model; Based on a preset initial learning rate, the initial timeliness prediction model is initially trained until the number of training iterations reaches a preset first threshold, thus obtaining the initially trained timeliness prediction model. Based on the preset peak learning rate, the time-efficiency prediction model after initial training is horizontally trained until the number of training times reaches the preset second threshold, thus obtaining the horizontally trained time-efficiency prediction model. The time-efficiency prediction model after horizontal training is debugged and trained according to the trigonometric function curve of the peak learning rate to obtain the trained time-efficiency prediction model.

5. The method as described in claim 1, characterized in that, The analysis of each predicted probability value to determine the delivery time of the target express shipment includes: The maximum value among the predicted probability values ​​is selected to obtain the target predicted probability value; The timeliness type corresponding to the predicted probability value of the target is determined as the logistics timeliness of the target express shipment.

6. A device for predicting the delivery time of express mail, characterized in that, include: The information acquisition module is used to acquire the package attribute information of the target package, including address information, time information and product information; The feature engineering module is used to perform feature engineering processing on the express delivery attribute information to obtain the express delivery features of the target express delivery, including: performing feature engineering processing on the address information to obtain network point information and distance information as the express delivery features; performing feature engineering processing on the time information to obtain holiday information and dispatch time as the express delivery features; and performing feature engineering processing on the product information to obtain product type information as the express delivery features. The model analysis module is used to input the express delivery features into a trained timeliness prediction model and output the predicted probability value of the target express delivery under various preset timeliness types. This includes: merging the express delivery features to obtain merged express delivery features; inputting the express delivery features into the trained timeliness prediction model and outputting first correlation information within the features; inputting the merged express delivery features into the trained timeliness prediction model and outputting second correlation information between the features; and analyzing the first correlation information and the second correlation information to obtain the predicted probability value of the target express delivery under various preset timeliness types. The trained timeliness prediction model consists of a fully connected layer, a dropout layer, a fusion layer, and a normalization layer. The timeliness determination module is used to analyze each of the predicted probability values ​​to determine the logistics timeliness of the target express shipment.

7. A computer device, characterized in that, The computer device includes: One or more processors; The memory; and 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 express delivery time prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the steps in the express delivery time prediction method according to any one of claims 1 to 5.

Images