Logistics transportation time determination method and electronic equipment

Abstract

The invention discloses a logistics transportation time determination method and electronic equipment, and the method comprises the steps: generating a target hypergraph based on a transportation path and candidate risk events of a target logistics order, the target hypergraph being used for representing a geographic position of the target logistics order in the transportation path and a distribution diagram of the candidate risk events corresponding to the geographic position; processing the association relationship in the target hypergraph by using a diffusion model to obtain target influence information of a target risk event corresponding to the target logistics order; wherein the incidence relation comprises the geographic position in the target hypergraph and the relation between the corresponding candidate risk events; the target influence information represents quantitative influence information of the target risk event on the target logistics order in transportation time; and determining the transportation time of the target logistics order according to the target influence information of the target risk event.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and more particularly to a logistics transportation time determination method and an electronic device. BACKGROUND

[0002] In logistics transportation, predicting the estimated time of delivery (ETA) is a relatively important link. In logistics transportation, it is affected by some risk factors, and the states of these risk factors are also non-fixed, which will affect the logistics transportation state. If these risk factors are analyzed by human experience, there will be certain deviation and hysteresis, which reduces the accuracy and efficiency of logistics transportation time prediction. SUMMARY

[0003] Therefore, the present application provides the following technical solutions:

[0004] A logistics transportation time determination method comprises:

[0005] Based on the transportation path of the target logistics order and the candidate risk events, a target hypergraph is generated, which is used to represent the distribution of the candidate risk events at the corresponding geographical positions of the target logistics order in the transportation path;

[0006] The correlation relationship in the target hypergraph is processed using a diffusion model to obtain target influence information of a target risk event corresponding to the target logistics order; wherein the correlation relationship includes the relationship between the geographical positions and the corresponding candidate risk events in the target hypergraph; the target influence information represents the quantitative influence information of the target risk event on the transportation time of the target logistics order;

[0007] According to the target influence information of the target risk event, the transportation time of the target logistics order is determined.

[0008] Optionally, the target hypergraph is generated based on the transportation path of the target logistics order and the candidate risk events, comprising:

[0009] A first-level hypergraph is generated with the target logistics order as the node and the geographical positions in the transportation path of the target logistics order as the hyperedge;

[0010] A second-level hypergraph is generated with the geographical positions as the node and the candidate risk events as the hyperedge;

[0011] The target hypergraph is obtained based on the first-level hypergraph and the second-level hypergraph.

[0012] Optionally, the diffusion model represents a model for quantifying the influence of a risk event on a logistics order obtained by processing historical logistics order data through a graph structure diffusion process, and the diffusion model is obtained by:

[0013] determining a correlation matrix corresponding to a hypergraph based on historical logistics order data, wherein the hypergraph includes a first-level hypergraph and a second-level hypergraph, the nodes of the first-level hypergraph are logistics orders, and the hyperedges are geographic locations in the transportation paths of the logistics orders; the nodes of the second-level hypergraph are the geographic locations, and the hyperedges are risk events corresponding to the logistics orders; and the correlation matrix represents the correlation between logistics orders, geographic locations, and risk events;

[0014] performing diffusion processing on the correlation matrix of the hypergraph based on an analysis dimension for a risk event to obtain a diffusion result of the correlation matrix corresponding to the analysis dimension;

[0015] determining a diffusion model based on the diffusion result.

[0016] Optionally, the analysis dimension includes at least one of a geographic influence dimension of a risk event, an influence degree dimension of a risk event, and a time influence dimension of a risk event; and wherein the diffusion processing on the correlation matrix of the hypergraph based on the analysis dimension for a risk event to obtain a diffusion result of the correlation matrix corresponding to the analysis dimension includes at least one of:

[0017] performing first diffusion processing on the correlation matrix of the hypergraph based on the geographic influence dimension of the risk event to obtain a first diffusion result, wherein the first diffusion result represents an optimized correlation matrix after the geographic influence dimension of the risk event is optimized, and the optimized correlation matrix is used to associate the risk event to actual geographic locations affected by it; and the first diffusion processing is used to filter the range of geographic locations actually affected by the risk event, and to correct the association weights of logistics orders and geographic locations, and geographic locations and risk events;

[0018] performing second diffusion processing on the optimized correlation matrix based on the influence degree dimension of the risk event to obtain a second diffusion result, wherein the second diffusion processing represents the superimposed influence of quantifying the influence of at least one risk event on a logistics order, and the second diffusion result represents the comprehensive influence degree of a logistics order affected by the combined action of at least one risk event;

[0019] performing third diffusion processing on the matching relationship between the time characteristics of the risk event in the hypergraph and the transportation time of the logistics order based on the time influence dimension of the risk event to obtain a third diffusion result, wherein the third diffusion processing represents the dynamic calculation of the timeliness of the risk event on the transportation time of the logistics order, and the third diffusion result represents the remaining effective influence strength of the risk event at the arrival time of the logistics order.

[0020] Optionally, the determining the diffusion model based on the diffusion result comprises:

[0021] predicting a transport time of the historical logistics order based on the diffusion result of the correlation matrix of the corresponding analysis dimension, to obtain a predicted transport time;

[0022] adjusting a model parameter in an initial model structure based on a deviation between the predicted transport time and an actual transport time corresponding to the historical logistics order, to obtain the diffusion model.

[0023] Optionally, the first diffusion processing on the correlation matrix of the hypergraph to obtain a first diffusion result comprises:

[0024] adding controllable noise in the correlation matrix of the hypergraph, the controllable noise representing a false relationship and / or an error weight in the correlation matrix;

[0025] learning the controllable noise distribution through a neural network, removing the noise in the correlation matrix with the added controllable noise, to obtain a denoised correlation matrix;

[0026] performing a filtering operation on the denoised correlation matrix to obtain the first diffusion result;

[0027] Alternatively, the second diffusion processing on the optimized correlation matrix to obtain a second diffusion result comprises:

[0028] converting the optimized correlation matrix into a probability transition matrix, the probability transition matrix satisfying a propagation probability distribution of a risk event influence;

[0029] performing fusion processing on the probability transition matrix and a node feature matrix through hypergraph diffusion convolution to obtain the second diffusion result, the node feature matrix representing a state information matrix of a node in the hypergraph;

[0030] Alternatively, the third diffusion processing on a matching relationship between a time feature of the risk event in the hypergraph and a transport time of the logistics order to obtain a third diffusion result comprises:

[0031] performing matching calculation on the time feature of the risk event in the hypergraph and the transport time of the logistics order to obtain a calculation result;

[0032] determining the third diffusion result based on the calculation result and a time decay curve.

[0033] Optionally, the processing on the correlation relationship in the target hypergraph by using the diffusion model to obtain target influence information of a target risk event corresponding to the target logistics order comprises:

[0034] determine an analysis dimension of the candidate risk event;

[0035] process, based on the diffusion model and the analysis dimension, an association relationship in the target hypergraph to obtain a diffusion result corresponding to each analysis dimension;

[0036] determine, based on the diffusion result corresponding to each analysis dimension, a target risk event in the candidate risk event and target influence information corresponding to the target risk event.

[0037] Optionally, the analysis dimension includes at least one of a geographical influence dimension of a risk event, an influence degree dimension of a risk event, and a time influence dimension of a risk event, and the diffusion model includes a first diffusion module corresponding to the geographical influence dimension of the risk event, a second diffusion module corresponding to the influence degree dimension of the risk event, and a third diffusion module corresponding to the time influence dimension of the risk event, wherein the processing, based on the diffusion model and the analysis dimension, of the association relationship in the target hypergraph to obtain the diffusion result corresponding to each analysis dimension includes at least one of:

[0038] filtering, based on the first diffusion module, a first association relationship between a geographical location in the target hypergraph and a risk event to obtain an actual influence geographical location corresponding to each candidate risk event;

[0039] analyzing, based on the second diffusion module, the actual influence geographical location corresponding to each candidate risk event to obtain a comprehensive influence degree of each candidate risk event on the target logistics order;

[0040] performing, based on the third diffusion module, time-effect reasoning processing on a time feature of the candidate risk event in the target hypergraph to obtain a change degree of each candidate risk event over time

[0041] Optionally, the determining, based on the diffusion result corresponding to each analysis dimension, of the target risk event in the candidate risk event and the target influence information corresponding to the target risk event includes:

[0042] determining, based on the actual influence geographical location corresponding to each candidate risk event, the change degree of each candidate risk event over time, and each geographical location in a transportation path of the target order, the target risk event in the candidate risk event;

[0043] determining, based on the comprehensive influence degree of each candidate risk event on the target logistics order, target influence information corresponding to the target risk event.

[0044] An electronic device includes:

[0045] a memory configured to store an application program and data generated by running of the application program;

[0046] a processor configured to execute the application program to implement:

[0047] generate a target hypergraph based on a transportation path of a target logistics order and candidate risk events, the target hypergraph being used to represent a distribution map of candidate risk events corresponding to geographical positions of the target logistics order in the transportation path;

[0048] process an association relationship in the target hypergraph by using a diffusion model to obtain target influence information of a target risk event corresponding to the target logistics order, wherein the association relationship comprises a relationship between the geographical positions and the corresponding candidate risk events in the target hypergraph, and the target influence information represents quantified influence information of the target risk event on a transportation time of the target logistics order;

[0049] determine the transportation time of the target logistics order according to the target influence information of the target risk event. BRIEF DESCRIPTION OF DRAWINGS

[0050] 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 embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0051] Figure 1 a flowchart of a logistics transportation time determination method provided by an embodiment of the present application;

[0052] Figure 2 a flowchart of a method for obtaining a diffusion model provided by an embodiment of the present application;

[0053] Figure 3 a hypergraph diagram provided by an embodiment of the present application;

[0054] Figure 4 a hypergraph diagram changing over time based on historical order data provided by an embodiment of the present application;

[0055] Figure 5 a structure diagram of a diffusion model provided by an embodiment of the present application;

[0056] Figure 6 a structure diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0057] With reference to the drawings, the technical solutions in the embodiments of the present application will be clearly and completely described in order to make apparent that the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0058] The terms "first" and "second" and the like in the present application are used to distinguish different objects, and are not used to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but can include steps or units not listed.

[0059] The embodiments of the present application provide a logistics transportation time determination method, which is applied to the scene of predicting the transportation time of a logistics order. The method can be used to predict the transportation time of a logistics order from a transportation starting point to a transportation ending point, or the transportation time of a logistics path in the middle of a logistics order, or the transportation time of a logistics order when it reaches a certain geographical location on the transportation path. In the embodiments of the present application, a target hypergraph can be generated according to a logistics order to be predicted and related risk events affecting the logistics order, and then a diffusion model is used to analyze the graph structure in the target hypergraph to determine the target risk event actually affecting the logistics order and the corresponding influence information of the target risk event. In this way, the influence information of the target risk event on the transportation time of the target logistics order can be quantified, the influence of multiple risk events on a single geographical location and logistics order is comprehensively considered, the risk influence on the logistics order can be more accurately analyzed, and the accuracy of the transportation time determination is improved.

[0060] Referring to Figure 1 A flowchart of a logistics transportation time determination method provided by the embodiments of the present application is shown in the figure. The method can include the following steps:

[0061] S101, generating a target hypergraph based on a transportation path of a target logistics order and candidate risk events.

[0062] The target hypergraph is used to represent the geographical positions of the target logistics order in the transportation path and the distribution of candidate risk events corresponding to the geographical positions. The target logistics order is a logistics order whose transportation time is to be predicted. The transportation time can be the transportation time of the target logistics order from the starting point to the end point, or the transportation time of a part of the transportation path. In order to clearly and accurately obtain the relationship between the geographical positions in the transportation path of the logistics order and the candidate risk events, the logistics order, the geographical positions in the transportation path, and the candidate risk events can be converted into a two-level hypergraph, that is, an association network of "logistics order-geographical position-candidate risk event" is established. The candidate risk event represents a risk event that can affect the transportation time of the target logistics order. For example, the transportation path of the target logistics order is from A to B for transfer and finally to C, and the current candidate risk events include a rainstorm warning in B and a typhoon warning in C. The first-level graph structure in the target hypergraph represents the information of the target order and the geographical positions A, B, and C, and the second-level graph structure represents the association information of B and the rainstorm warning and C and the typhoon warning. Through the establishment of the target hypergraph, more accurate basic information can be provided for the subsequent analysis process, thereby improving the accuracy of the subsequent determination of the transportation time.

[0063] In S102, a diffusion model is used to process the association relationship in the target hypergraph to obtain target influence information of a target risk event corresponding to the target logistics order.

[0064] The association relationship includes the relationship between the geographical positions and the corresponding candidate risk events in the target hypergraph. The target influence information represents the quantitative influence of the target risk event on the transportation time of the target logistics order. For example, the target influence information can be the influence degree of the target risk event on the transportation time, or the duration of the target risk event. The diffusion model is generated by using historical logistics order data, and can quantify the influence of the risk event on the logistics order. Through the diffusion model, the association relationship in the target hypergraph can be dynamically optimized and quantitatively calculated. The diffusion model can be analyzed by different dimensions to filter out the target risk events that can actually affect the transportation time of the target logistics order and obtain the target influence information. Through the graph structure diffusion processing mode, the learned association relationship and influence degree are explicitly output, and the explainability of the influence of the risk event on the logistics order is improved.

[0065] In S103, the transportation time of the target logistics order is determined according to the target influence information of the target risk event.

[0066] After the target influence information is determined, the influence duration of the target risk event on the transportation time can be determined, and the transportation time of the target logistics order is further determined. For example, based on the transportation path determined for the target logistics order, the average transportation time corresponding to the transportation path can be obtained, and then the influence duration of the target risk event on the average transportation time is determined according to the target influence information. For example, the average transportation time is 8 hours, and the target risk event is a rainstorm at B in the transportation path. According to the duration of the rainstorm, the delay time is 2 hours, so the transportation time of the target logistics order is 8+2=10 hours.

[0067] In the embodiments of the present application, a target hypergraph can be generated according to a logistics order to be predicted and a related risk event affecting the logistics order, and then a diffusion model is used to analyze the graph structure in the target hypergraph to determine a target risk event actually affecting the logistics order and influence information corresponding to the target risk event. In this way, the influence of the target risk event on the transportation time of the target logistics order can be quantified, the influence of multiple risk events on a single geographic location and logistics order is comprehensively considered, the risk influence on the logistics order can be more accurately analyzed, and the accuracy of the transportation time determination is improved.

[0068] The logistics transportation time determination method provided in the embodiments of the present application will be described below in combination with specific application scenarios.

[0069] In order to accurately obtain the correlation between the geographic location in the logistics order and the candidate risk event that may affect the transportation time thereof, in the embodiments of the present application, the transmission relationship of “order-location-risk” is clearly represented in a hierarchical structure through a hypergraph, which facilitates subsequent analysis and application.

[0070] In one implementation, based on the transportation path of the target logistics order and the candidate risk event, the processing process of generating the target hypergraph can include: generating a first-level hypergraph with the target logistics order as a node and the geographic locations in the transportation path of the target logistics order as hyperedges; generating a second-level hypergraph with the geographic locations as nodes and the candidate risk events as hyperedges; and obtaining the target hypergraph based on the first-level hypergraph and the second-level hypergraph.

[0071] The first-level supergraph takes logistics orders as nodes and geographical positions as superedges, and can reflect the spatial path of the orders. The second-level supergraph takes geographical positions as nodes and candidate risk events as superedges, and can reflect the spatial distribution of the risk events. The two levels form a complete correlation network. For example, a logistics order is "Shanghai warehouse -> Hangzhou transfer station -> Ningbo distribution point". Taking the order as a node and the three geographical positions as superedges, a first-level supergraph is formed, in which the node features include the order weight and volume, and the superedge features include the geographical position coordinates. For the Hangzhou transfer station, candidate risk events (such as a high-temperature warning in Hangzhou on the same day) are obtained, taking Hangzhou as a node and the high-temperature warning as a superedge, to form a second-level supergraph, in which the node features include the transfer station processing capacity, and the superedge features include the high-temperature duration and impact level. Then, the first-level supergraph and the second-level supergraph are merged to form a correlation path including the order-Hangzhou-high-temperature warning. The explicit hierarchical relationship in the target supergraph facilitates the subsequent diffusion model to handle different dimensions of influence in a targeted manner, thereby improving the accuracy of the transportation time determination.

[0072] In the embodiments of the present application, the diffusion model represents the processing of historical logistics order data through a graph structure to obtain a model for quantifying the influence of a risk event on a logistics order. In one implementation, referring to Figure 2 , the obtaining process of the diffusion model can include the following steps:

[0073] S201, determining a correlation matrix corresponding to a supergraph based on historical logistics order data.

[0074] S202, performing diffusion processing on the correlation matrix of the supergraph based on an analysis dimension for a risk event, to obtain a diffusion result of the correlation matrix corresponding to the analysis dimension.

[0075] S203, determining a diffusion model based on the diffusion result.

[0076] The supergraph corresponding to the historical logistics order data includes a first-level supergraph and a second-level supergraph. The nodes of the first-level supergraph are logistics orders, and the superedges are geographical positions in the transportation path of the logistics orders. The nodes of the second-level supergraph are geographical positions, and the superedges are risk events corresponding to the logistics orders. Referring to the supergraph diagram shown in Figure 3 , only part of the historical logistics order data is shown in Figure 3 , such as order P1, order P2, and order P3 in the first-level supergraph. The superedges represent the geographical positions corresponding to the orders, such as the geographical positions corresponding to the orders at different time points (such as time t or time t+1). In Figure 3 , the geographical positions corresponding to the superedges include Beijing, Wuhan, and Guangzhou. Figure 3 The second-level supergraph is to associate the geographical positions in the first-level supergraph with risk events. The corresponding superedges can represent normal, heavy rain, or traffic control risk events.

[0077] For example, see Figure 4 A time-varying hypergraph diagram based on historical order data is provided for embodiments of the present application, in which only part of the historical order data is shown, in Figure 4 Different types of risk event sets (such as Figure 4 are represented in Risk Events) can be included in Figure 4 The corresponding risk events can be distinguished by different colors, such as in Figure 4 The second type of risk event can correspond to three, represented by blue, yellow and green, respectively, in 4. In 4, a set of logistics orders (represented by Logistic Items) is also included, such as package1-package5 representing the transport packages in the corresponding logistics orders. By analyzing the relevant historical logistics order data, the transport nodes in the logistics network, i.e. the geographic locations, are obtained in Figure 4 are represented by Geographic Locations.

[0078] In Figure 4 , the orders (such as Package 1-5) are the first layer nodes, and the geographic locations are the first layer hyperedges, which are used to represent the geographic location status of the orders at a certain time. For example, a geographic location hyperedge connects multiple order nodes, indicating that these orders are in the same geographic location at that time. The geographic locations are the second layer nodes, and the risk events (such as Event 1-2) are the second layer hyperedges, which are used to represent the risk events existing in each geographic location at that time. For example, a risk event hyperedge connects multiple geographic location nodes, indicating that the risk event affects these geographic locations.

[0079] The information reflected in the hypergraph at different times (such as Figure 4 , t, t+1, t+H) is different, for example, at the t-1 time shown in Figure 4 , the orders are distributed in different geographic locations, and the hyperedge connection represents the location of the orders at the current t-1 time. For example, Package 1 is connected to a geographic location hyperedge, indicating that it is in that location at t-1. Some geographic locations are connected to risk event hyperedges. For example, Event 1 can affect a certain geographic location (such as Figure 5The orange range in the t-1 moment represents the geographical position affected by Event 1, at this time, the coverage of the risk event Event 1 is small, and the potential impact on the order has not yet significantly appeared. At the t moment, the orders distributed in various geographical positions have changed, and correspondingly, the geographical position affected by the risk event changes over time, for example, the coverage of the risk event Event 1 at the t moment is larger than that at the t-1 moment, that is, the affected geographical position has changed. Correspondingly, at the t+1 moment, it can be seen that the influence range of the yellow Event2 corresponding to the risk event has disappeared. In this way, through the changes of nodes and hyperedges in the hypergraph over time, the correlation between the logistics order, the geographical position, and the risk event, and the difference over time can be more clearly and accurately displayed. It provides more accurate graph structure information for subsequent diffusion processing, so that the risk events and related information that actually affect the transportation time can be more accurately obtained, and the accuracy of the transportation time prediction is improved.

[0080] The correlation matrix corresponding to the hypergraph represents the correlation between the logistics order, the geographical position, and the risk event. The correlation matrix can be processed by diffusion based on the analysis dimension of the risk event to obtain the corresponding diffusion result. The correlation matrix can be optimized by corresponding diffusion processing in different analysis dimensions, and then the model parameters are adjusted to the prediction error as the target. Among them, the diffusion processing is mainly processed by geographical influence, degree, time decay, etc. For example, the diffusion processing is realized by hierarchical dynamic optimization of the hypergraph correlation matrix to achieve accurate quantitative evaluation of risk influence. For example, noise injection and noise removal learning can be performed on the order-location-risk correlation matrix in the structure dimension to filter out the real and effective risk path (such as eliminating the false "light rain affects the warehouse" correlation); the purified correlation matrix can be converted into a probability transition matrix in the transfer dimension, and the nonlinear superposition effect of multiple risk events (such as strike + heavy rain combined delay 4.2 hours) can be calculated by hypergraph convolution; finally, the influence weight is dynamically adjusted based on the life cycle of the risk event in the time sequence dimension (such as 60% delay attenuation before the strike ends 2 hours), and finally the transportation time prediction is output after time and space calibration.

[0081] Based on the diffusion result of the corresponding analysis dimension, the diffusion result representing the risk event after eliminating the error in the risk event and the corresponding influence information is obtained, and then an initial model is trained based on this to obtain a diffusion model. For example, in one embodiment, the transportation time of the historical logistics order is predicted based on the diffusion result of the correlation matrix of the corresponding analysis dimension to obtain the predicted transportation time; the model parameters in the initial model structure are adjusted based on the deviation between the predicted transportation time and the actual transportation time corresponding to the historical logistics order to obtain the diffusion model.

[0082] In the embodiments of the present application, the analysis dimension of the risk event can be determined according to the logistics characteristics of the target logistics order. If the target logistics order is long-distance transportation, the corresponding analysis dimension can include the time dimension of the risk event, so that the influence degree of the risk event over time can be analyzed, and it can be determined whether the current risk event will have an actual impact on the transportation of the target logistics order. The corresponding analysis dimension can also be determined according to the transportation path of the target logistics order, so that accurate, objective and multi-angle analysis of the risk event based on different analysis dimensions can be achieved, and accurate reference basis for determining the transportation time can be obtained.

[0083] In an implementation of the embodiments of the present application, the analysis dimension includes at least one of a geographical influence dimension of the risk event, an influence degree dimension of the risk event, and a time influence dimension of the risk event. Correspondingly, based on the analysis dimension of the risk event, the association matrix of the hypergraph is diffused to obtain a diffusion result of the association matrix of the corresponding analysis dimension, including at least one of the following:

[0084] (1) Based on the geographical influence dimension of the risk event, the association matrix of the hypergraph is subjected to first diffusion processing to obtain a first diffusion result, wherein the first diffusion result represents the association matrix after optimization of the geographical influence dimension of the risk event, and the optimized association matrix is used to associate the risk event to the actual affected geographical location. The first diffusion processing is used to filter the geographical location range actually affected by the risk event, and to correct the association weight between the logistics order and the geographical location and between the geographical location and the risk event.

[0085] The diffusion processing of the geographical influence dimension of the risk event is to filter out the geographical location actually affected by the risk event through denoising and filtering. In an implementation, the association matrix of the hypergraph is subjected to first diffusion processing to obtain a first diffusion result, including:

[0086] Controllable noise is added to the association matrix of the hypergraph, and the controllable noise represents false relationships and / or incorrect weights in the association matrix. The controllable noise distribution is learned through a neural network, the noise in the association matrix with the controllable noise is removed, and a denoised association matrix is obtained.

[0087] The first diffusion processing performs spatial range filtering and weight correction processing on the original association matrix, so that the risk event that actually affects the current geographical location in the historical logistics order data can be obtained. For example, the first-level hypergraph takes the order as the first-level node and the geographical location as the first-level hyperedge, and represents the geographical location state of the order at a certain time, denoted as wherein represents the set of orders existing in the logistics network at time t, and the feature matrix is represents the matching relationship between all orders and geographical positions at time t, that is, the incidence matrix (also called the adjacency matrix) in the hypergraph; the second-level hypergraph, with geographical positions as the second layer nodes and risk events as the second layer hyperedges, represents the risk events existing in each geographical position at this moment, denoted as wherein represents the set of geographical positions existing in the logistics network at time t, and the characteristic matrix thereof is represents the matching relationship between all geographical positions and risk events at time t, that is, the adjacency matrix in the hypergraph, and the characteristic matrix of the risk event is denoted as

[0088] The first diffusion processing of the geographical influence dimension of the risk event can be realized by structure dimension diffusion graph convolution, that is, adding and removing noise on the incidence matrix (also called the adjacency matrix) of the two-level hypergraph, so as to filter out the associated relationship consistent with the target and predict the associated relationship, that is, to associate the risk event to the actual geographical position affected thereby. For example, adding noise on the adjacency matrix of each level of hypergraph, the corresponding transition kernel can be represented as:

[0089]

[0090] wherein is a hyperparameter.

[0091] The reverse process (i.e., the process of learning to remove the above-mentioned added noise): diffuse the noise in the reverse direction on the above-mentioned adjacency matrix, and the transition kernel is:

[0092]

[0093] wherein, is the mean and variance of the first layer reverse process, is the mean and variance of the second layer reverse process.

[0094] Then, irrelevant influence information is removed by filtering operation:

[0095]

[0096] wherein, σ is an activation function, which filters out irrelevant edges weakened by the diffusion module and new edges enhanced.

[0097] For example, the initial incidence matrix shows that the typhoon affects Shenzhen, but after diffusion processing, it is found that the actual influence range of the typhoon extends to Guangzhou and Changsha; the filtering operation can be used to enhance the associated edges of Shenzhen-Guangzhou-Changsha and weaken the false association.

[0098] (2) based on the influence degree dimension of the risk event, the second diffusion processing is performed on the optimized correlation matrix to obtain a second diffusion result; wherein the second diffusion processing represents the superimposed influence of quantifying the influence of at least one risk event on the logistics order, and the second diffusion result represents the comprehensive influence degree of the logistics order under the joint action of at least one risk event.

[0099] In the diffusion processing process of the influence degree dimension of the risk event, the optimized correlation matrix in (1) can be used for processing, or the original correlation matrix in the hypergraph can be used for processing, so as to quantify the superimposed influence of multiple risk events.

[0100] In one embodiment, the second diffusion processing is performed on the optimized correlation matrix to obtain a second diffusion result, including: converting the optimized correlation matrix into a probability transition matrix, the probability transition matrix satisfying the propagation probability distribution of the risk event influence; and performing fusion processing on the probability transition matrix and a node feature matrix through hypergraph diffusion convolution to obtain the second diffusion result, the node feature matrix representing a matrix of state information of nodes in the hypergraph.

[0101] In the process of the second diffusion processing, the optimized correlation matrix is diffused by the processing mode of probability propagation and nonlinear aggregation to obtain the second diffusion result. For example, the second diffusion processing can be realized by the way of passing dimension diffusion graph convolution, noise is added and removed in the process of information passing from the hyperedge to the node, and the information aggregation process is transformed on each node, so as to depict the transmission process of the risk event and the superimposed influence of multiple risk events.

[0102] The optimized correlation matrix is subjected to row random change to obtain a probability transition matrix:

[0103]

[0104] wherein, and is the optimized correlation matrix, and is a hyperparameter.

[0105] Then, the hypergraph diffusion convolution processing is performed on the above-mentioned probability transition matrix, that is, noise is added in the graph convolution operation on each level of hypergraph, and the transition kernel can be represented as:

[0106]

[0107] For example, the correlation matrix of typhoon (influence degree 0.8) and traffic control (influence degree 0.5) is converted into a probability transition matrix, and the joint influence is calculated by hypergraph convolution: 0.8x0.5=0.4 (superimposed attenuation), and the final comprehensive influence degree on the order is 0.6.

[0108] (3) Based on the time impact dimension of risk events, the third diffusion processing is performed on the matching relationship between the time characteristics of risk events in the hypergraph and the transportation time of logistics orders to obtain the third diffusion result; the third diffusion processing represents the timeliness of the dynamic calculation of risk events on the transportation time of logistics orders; the third diffusion result represents the residual effective impact intensity of the risk event at the time of arrival of the logistics order.

[0109] By performing diffusion processing on the time impact dimension of risk events, the timeliness of the impact of risk events can be dynamically calculated, and dynamic attenuation calculation can be performed on the time-aligned risk events (aligning the occurrence time of the risk event with the time of the geographical location through which the logistics order passes) to obtain the timeliness correction result of the risk event.

[0110] In one embodiment, a third diffusion process is performed on the matching relationship between the time characteristics of the risk events in the hypergraph and the transportation time of the logistics order to obtain a third diffusion result, including: performing a matching calculation on the time characteristics of the risk events in the hypergraph and the transportation time of the logistics order to obtain a calculation result; and determining the third diffusion result based on the calculation result and the time attenuation curve.

[0111] The third diffusion process can be performed through time-series diffusion graph convolution. In the time dimension, noise is added and removed during the representation transfer of node objects, thereby depicting the process of risk event impact defense over time. For example, by using the time-series random walk processing method to perform diffusion enhancement in each random walk, the transfer kernel can be expressed as:

[0112]

[0113] in, and They are and The similarity association matrix calculated based on attention is shown above. τ1 and τ2 are hyperparameters, which are the probabilities of restarting in logistics orders and wandering over time during the random walk process, respectively. τ1 and τ2 are hyperparameters, which are the probabilities of restarting in risk events and wandering over time during the random walk process, respectively.

[0114] For example, the order is expected to arrive in Changsha at t+12 hours, and the typhoon ends at t+8 hours. The remaining impact intensity is calculated to be 0.54 hours through the time decay curve (such as exponential decay).

[0115] After the above diffusion processing of each dimension of each historical order data, the influence characteristics of risk events of different dimensions can be obtained, so that the influence characteristics can be used for model training to obtain a diffusion model, and the transport time can be predicted based on the diffusion model. For example, the geographical location range affected by each risk event (i.e., the superedge in the two-level hypergraph) is obtained, the influence degree of multiple risk events on the order (i.e., the influence of superedge information on node information), and the change process of the influence over time (i.e., the trend of the change of the two hypergraphs over time).

[0116] In the above learning process of historical order data, the related parameters of each diffusion process or the overall model can be adjusted based on the deviation between the predicted transport time and the actual transport time to obtain a final diffusion model for subsequent prediction of the transport time of the order. Correspondingly, the diffusion model can include different diffusion modules, each diffusion module being used to perform diffusion processing of the corresponding analysis dimension. In this way, when predicting the transport time of a logistics order, the diffusion module can be used to perform data diffusion processing of the corresponding hypergraph of the logistics order in the corresponding analysis dimension to obtain the diffusion result of the analysis dimension, and then output the diffusion result to the prediction layer of the diffusion model for prediction of the transport time.

[0117] For example, the output information of the prediction output layer is y,

[0118] i.e., the predicted value of the transport time of each order to its next logistics transport node. Wherein, W o represents a weight matrix, i.e., the contribution weight of different features to the transport time, which can include a geographical coverage intensity weight, a risk superposition value weight, and a failure coefficient weight, b o represents the adjustment amount of the baseline transport time, reflecting the inherent deviation in the risk-free scenario (such as the inherent time consumption of the route). may represent the influence characteristics obtained after diffusion processing, and the influence time parameter corresponding to the corresponding influence characteristics.

[0119] The above describes the obtaining process of the diffusion model. Correspondingly, when the diffusion model is applied, the data of the target logistics order is processed by the diffusion model, which is similar to the training process of the diffusion model, which will not be described in detail here.

[0120] In an embodiment of the present application, the correlation in the target hypergraph is processed by using a diffusion model to obtain the target influence information of the target risk event corresponding to the target logistics order, including: determining the analysis dimension of the candidate risk event; processing the correlation in the target hypergraph based on the diffusion model and the analysis dimension to obtain the diffusion result corresponding to each analysis dimension; determining the target risk event and the target influence information corresponding to the target risk event in the candidate risk event based on the diffusion result corresponding to each analysis dimension.

[0121] In this embodiment, the transportation time of the target logistics order needs to be determined, the target risk event is screened from the candidate risk event by using the diffusion model, and the influence is quantified. First, the analysis dimension (such as geography, degree, and time) is determined, the corresponding diffusion module is applied to process the target hypergraph, and the key risk event is screened based on the diffusion result and the influence is quantified.

[0122] For example, the target logistics order is transported from A to B, the candidate risk events include: C sandstorm, B high temperature, and D traffic control, and C and D are the passing places of the target logistics order. Through the diffusion processing of the geographical influence dimension, the actual influence range of C sandstorm and D traffic control is obtained, and then the superimposed influence degree of each candidate risk event can be calculated, and the remaining influence intensity of B high temperature when the order arrives can also be obtained by diffusion processing: such as 0.5 hours (the high temperature lasts after the order arrives). Based on the transportation path, the risk events C sandstorm and D traffic control affecting the geographical position in the path are screened as the target risk events. Then the two target influence information, such as C sandstorm causing a transportation delay of 1 hour and D traffic control causing a delay of 0.5 hours, and the comprehensive delay of the two target events is 1.5 hours. In the present application, the key influencing factors are accurately screened from multiple candidate risk events, the interference of irrelevant risks is avoided, and the path information in the transportation scheme can also be adjusted according to the quantified results.

[0123] Referring to Figure 6 It shows a structure schematic diagram of a diffusion model provided by an embodiment of the present application, which includes an input module for obtaining the feature information of a target logistics order to be predicted, a first diffusion module, a second diffusion module, and a third diffusion module, and a prediction module and an output module.

[0124] In the case that the analysis dimension comprises at least one of the geographical influence dimension of the risk event, the influence degree dimension of the risk event, and the time influence dimension of the risk event, the diffusion model comprises a first diffusion module corresponding to the geographical influence dimension of the risk event, a second diffusion module corresponding to the influence degree dimension of the risk event, and a third diffusion module corresponding to the time influence dimension of the risk event, wherein, based on the diffusion model and the analysis dimension, the association relationship in the target hypergraph is processed to obtain a diffusion result corresponding to each analysis dimension, comprising at least one of the following:

[0125] Based on the first diffusion module, the first association relationship between the geographical position in the target hypergraph and the risk event is filtered to obtain an actual influence geographical position corresponding to each candidate risk event;

[0126] Based on the second diffusion module, the actual influence geographical position corresponding to each candidate risk event is analyzed to obtain a comprehensive influence degree of each candidate risk event on the target logistics order;

[0127] Based on the third diffusion module, the time characteristics of the candidate risk event in the target hypergraph are processed to obtain a change degree of each candidate risk event over time.

[0128] The processing process of each diffusion module can refer to the corresponding diffusion processing process in the foregoing embodiments, which will not be described in detail here. In the learning process based on historical logistics order data, each diffusion module has module parameters that can accurately output the corresponding diffusion result through repeated parameter adjustment, facilitating the subsequent prediction module to predict the transportation time based on the output information of each diffusion module, and then output the transportation time through the output module of the diffusion module. Through the cooperative processing of each diffusion module in the diffusion module, the prediction accuracy of the transportation time corresponding to the complex risk scenario is improved.

[0129] After obtaining the diffusion results of each diffusion module, the risk event can be screened and the influence can be quantified based on the diffusion results. In an implementation manner, based on the diffusion result corresponding to each analysis dimension, a target risk event is determined from the candidate risk event, and target influence information corresponding to the target risk event comprises:

[0130] Based on the actual influence geographical position corresponding to each candidate risk event, the change degree of each candidate risk event over time, and each geographical position in the transportation path of the target order, the target risk event is determined from the candidate risk event; based on the comprehensive influence degree of each candidate risk event on the target logistics order, the target influence information corresponding to the target risk event is determined.

[0131] In this embodiment, whether the geographical position of the risk impact is in the transportation path is determined through the geographical dimension, the overlapping of the risk and the arrival time of the order is determined through the time dimension, the impact intensity is determined through the degree dimension, and finally the key risks are screened and the total impact is quantified. Path-related risk events can be accurately screened, irrelevant risks (such as risks not on the path) can be excluded, calculation redundancy can be reduced, and the accuracy of predicting the transportation time can be improved.

[0132] In another embodiment of the present application, an electronic device is also provided, referring to ​ The electronic device comprises:

[0133] The memory 601 is configured to store an application program and data generated by running of the application program.

[0134] The processor 602 is configured to execute the application program to implement:

[0135] generating a target hypergraph based on a transportation path of a target logistics order and candidate risk events, the target hypergraph being used to represent a distribution diagram of geographical positions of the target logistics order in the transportation path and the candidate risk events corresponding to the geographical positions;

[0136] processing, by using a diffusion model, an association relationship in the target hypergraph to obtain target impact information of a target risk event corresponding to the target logistics order, wherein the association relationship comprises a relationship between the geographical positions and the corresponding candidate risk events in the target hypergraph, and the target impact information represents quantified impact information of the target risk event on a transportation time of the target logistics order;

[0137] determining the transportation time of the target logistics order according to the target impact information of the target risk event.

[0138] Optionally, the generating of the target hypergraph based on the transportation path of the target logistics order and the candidate risk events comprises:

[0139] generating a first-level hypergraph by taking the target logistics order as a node and a geographical position in the transportation path of the target logistics order as a hyperedge;

[0140] generating a second-level hypergraph by taking the geographical position as a node and a candidate risk event as a hyperedge;

[0141] obtaining the target hypergraph based on the first-level hypergraph and the second-level hypergraph.

[0142] Optionally, the diffusion model represents a model for quantifying an impact of a risk event on a logistics order, the model being obtained by processing historical logistics order data through a graph structure diffusion process.

[0143] determine a correlation matrix corresponding to the hypergraph based on historical logistics order data, wherein the hypergraph comprises a first-level hypergraph and a second-level hypergraph, nodes of the first-level hypergraph are logistics orders, and hyperedges are geographical positions in a transportation path of the logistics orders; nodes of the second-level hypergraph are the geographical positions, and hyperedges are risk events corresponding to the logistics orders; and the correlation matrix represents an association relationship between the logistics orders, the geographical positions, and the risk events;

[0144] perform diffusion processing on the correlation matrix of the hypergraph based on an analysis dimension of the risk event, to obtain a diffusion result of the correlation matrix corresponding to the analysis dimension;

[0145] determine a diffusion model based on the diffusion result.

[0146] Optionally, the analysis dimension comprises at least one of a geographical influence dimension of the risk event, an influence degree dimension of the risk event, and a time influence dimension of the risk event; and wherein the diffusion processing on the correlation matrix of the hypergraph based on the analysis dimension of the risk event, to obtain a diffusion result of the correlation matrix corresponding to the analysis dimension, comprises at least one of:

[0147] perform first diffusion processing on the correlation matrix of the hypergraph based on the geographical influence dimension of the risk event, to obtain a first diffusion result, wherein the first diffusion result represents an optimized correlation matrix after the geographical influence dimension of the risk event is optimized, and the optimized correlation matrix is used to associate the risk event to actual geographical positions affected thereby; and the first diffusion processing is used to screen a range of geographical positions actually affected by the risk event, and correct association weights of the logistics orders and the geographical positions, and the geographical positions and the risk event;

[0148] perform second diffusion processing on the optimized correlation matrix based on the influence degree dimension of the risk event, to obtain a second diffusion result, wherein the second diffusion processing represents superimposed influences of quantifying influences of at least one risk event on logistics orders, and the second diffusion result represents a comprehensive influence degree of logistics orders affected by combined influences of at least one risk event;

[0149] perform third diffusion processing on a matching relationship between a time feature of the risk event in the hypergraph and a transportation time of the logistics order based on the time influence dimension of the risk event, to obtain a third diffusion result, wherein the third diffusion processing represents dynamic calculation of timeliness of the risk event on the transportation time of the logistics order, and the third diffusion result represents a remaining effective influence strength of the risk event at an arrival time of the logistics order.

[0150] Optionally, the determining of the diffusion model based on the diffusion result comprises:

[0151] The transport time of the historical logistics order is predicted based on a diffusion result of the correlation matrix of the correspondence analysis dimension, to obtain a predicted transport time;

[0152] Based on the deviation between the predicted transport time and the actual transport time corresponding to the historical logistics order, the model parameters in the initial model structure are adjusted to obtain a diffusion model.

[0153] Optionally, the first diffusion processing on the correlation matrix of the hypergraph is performed to obtain a first diffusion result, including:

[0154] Controllable noise is added to the correlation matrix of the hypergraph, and the controllable noise represents false relationships and / or incorrect weights in the correlation matrix;

[0155] The controllable noise distribution is learned by a neural network, and the noise in the correlation matrix with added controllable noise is removed to obtain a denoised correlation matrix;

[0156] The denoised correlation matrix is filtered to obtain the first diffusion result;

[0157] Alternatively, the second diffusion processing on the optimized correlation matrix is performed to obtain a second diffusion result, including:

[0158] The optimized correlation matrix is converted into a probability transition matrix, and the probability transition matrix satisfies a propagation probability distribution of a risk event influence;

[0159] The probability transition matrix and a node feature matrix are fused by hypergraph diffusion convolution to obtain the second diffusion result, and the node feature matrix represents a matrix of state information of nodes in the hypergraph;

[0160] Alternatively, the third diffusion processing on the matching relationship between the time feature of the risk event in the hypergraph and the transport time of the logistics order is performed to obtain a third diffusion result, including:

[0161] The time feature of the risk event in the hypergraph and the transport time of the logistics order are matched to obtain a calculation result;

[0162] Based on the calculation result and a time decay curve, the third diffusion result is determined.

[0163] Optionally, the correlation relationship in the target hypergraph is processed by using the diffusion model to obtain target influence information of a target risk event corresponding to the target logistics order, including:

[0164] An analysis dimension of the candidate risk event is determined;

[0165] process the association relationship in the target hypergraph based on the diffusion model and the analysis dimension, to obtain a diffusion result corresponding to each analysis dimension;

[0166] determine a target risk event and target impact information corresponding to the target risk event in the candidate risk events based on the diffusion result corresponding to each analysis dimension.

[0167] Optionally, the analysis dimension includes at least one of a geographical impact dimension of a risk event, an impact degree dimension of a risk event, and a time impact dimension of a risk event, and the diffusion model includes a first diffusion module corresponding to the geographical impact dimension of the risk event, a second diffusion module corresponding to the impact degree dimension of the risk event, and a third diffusion module corresponding to the time impact dimension of the risk event, wherein the processing of the association relationship in the target hypergraph based on the diffusion model and the analysis dimension, to obtain a diffusion result corresponding to each analysis dimension, includes at least one of:

[0168] filtering a first association relationship between a geographical location and a risk event in the target hypergraph based on the first diffusion module, to obtain an actual impact geographical location corresponding to each candidate risk event;

[0169] analyzing the actual impact geographical location corresponding to each candidate risk event based on the second diffusion module, to obtain a comprehensive impact degree of each candidate risk event on the target logistics order;

[0170] processing a time feature of the candidate risk event in the target hypergraph based on the third diffusion module, to obtain a change degree of each candidate risk event over time

[0171] Optionally, the determining of the target risk event and the target impact information corresponding to the target risk event in the candidate risk events based on the diffusion result corresponding to each analysis dimension includes:

[0172] determining the target risk event in the candidate risk events based on the actual impact geographical location corresponding to each candidate risk event, the change degree of each candidate risk event over time, and each geographical location in a transportation path of the target order;

[0173] determining the target impact information corresponding to the target risk event based on the comprehensive impact degree of each candidate risk event on the target logistics order.

[0174] In another embodiment of the present application, a readable storage medium is also provided, which has a computer program stored thereon, and the computer program is executed by a processor to implement the logistics transportation time determination method as described above.

[0175] It should be noted that the specific implementation of the processor in this embodiment can refer to the corresponding content in the foregoing, which will not be described in detail here.

[0176] The various embodiments are described in a progressive manner in the specification, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0177] The skilled person can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0178] The steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, a software module executed by a processor, or a combination of both. The software module can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0179] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining logistics transportation time, comprising: Based on the transportation path of the target logistics order and the candidate risk events, a target hypergraph is generated, wherein the target hypergraph is used to represent the geographical location of the target logistics order in the transportation path and the distribution map of the candidate risk events corresponding to the geographical location; The association relationships in the target hypergraph are processed using a diffusion model to obtain target impact information of the target risk event corresponding to the target logistics order; wherein the association relationships include the relationship between the geographic locations in the target hypergraph and the corresponding candidate risk events; the target impact information represents the quantitative impact of the target risk event on the transportation time of the target logistics order; The transportation time of the target logistics order is determined according to the target impact information of the target risk event.

2. The method according to claim 1, wherein generating a target hypergraph based on the transportation path of the target logistics order and the candidate risk events comprises: Generate a first-level hypergraph with target logistics orders as nodes and geographic locations in the transportation path of the target logistics orders as hyperedges; Generate a second-level hypergraph with the geographical location as a node and the candidate risk events as a hyperedge; The target hypergraph is obtained based on the first-level hypergraph and the second-level hypergraph.

3. The method according to claim 1, wherein the diffusion model represents a model for quantifying the impact of risk events on logistics orders obtained by processing historical logistics order data through graph structure diffusion. The process of obtaining the diffusion model includes: Based on a hypergraph corresponding to historical logistics order data, determining an association matrix corresponding to the hypergraph, wherein the hypergraph includes a first-level hypergraph and a second-level hypergraph, wherein the nodes of the first-level hypergraph are logistics orders, and the hyperedges are geographic locations in the transportation path of the logistics orders; the nodes of the second-level hypergraph are the geographic locations, and the hyperedges are risk events corresponding to the logistics orders; and the association matrix represents the association relationship between logistics orders, geographic locations, and risk events; Based on the analysis dimension of the risk event, the association matrix of the hypergraph is diffused to obtain the diffusion result of the association matrix of the corresponding analysis dimension; Based on the diffusion results, a diffusion model is determined.

4. The method according to claim 3, wherein the analysis dimension comprises at least one of the geographical impact dimension of the risk event, the impact degree dimension of the risk event, and the temporal impact dimension of the risk event; wherein, The diffusion processing is performed on the association matrix of the hypergraph based on the analysis dimension for the risk event to obtain a diffusion result of the association matrix corresponding to the analysis dimension, including at least one of the following: Based on the geographical impact dimension of the risk event, a first diffusion process is performed on the association matrix of the hypergraph to obtain a first diffusion result, where the first diffusion result represents the association matrix after optimization of the geographical impact dimension of the risk event, and the optimized association matrix is ​​used to associate the risk event with the geographical location actually affected by the risk event; The first diffusion process is used to screen the geographical location range actually affected by the risk event, and to modify the association weights between logistics orders and geographical locations, and between geographical locations and risk events; Based on the impact dimension of the risk event, a second diffusion process is performed on the optimized correlation matrix to obtain a second diffusion result; wherein the second diffusion process quantifies the cumulative impact of at least one risk event on the logistics order, and the second diffusion result represents the comprehensive impact of the at least one risk event on the logistics order; Based on the time impact dimension of the risk event, a third diffusion processing is performed on the matching relationship between the time characteristics of the risk event in the hypergraph and the transportation time of the logistics order to obtain a third diffusion result; the third diffusion processing represents the timeliness of the dynamically calculated risk event on the transportation time of the logistics order; the third diffusion result represents the remaining effective impact intensity of the risk event at the arrival time of the logistics order.

5. The method according to claim 3, wherein determining a diffusion model based on the diffusion result comprises: Based on the diffusion results of the correlation matrix of the corresponding analysis dimension, the transportation time of historical logistics orders is predicted to obtain the predicted transportation time; Based on the deviation between the predicted transportation time and the actual transportation time corresponding to the historical logistics order, the model parameters in the initial model structure are adjusted to obtain a diffusion model.

6. The method according to claim 4, wherein performing a first diffusion process on the incidence matrix of the hypergraph to obtain a first diffusion result comprises: adding controllable noise to an incidence matrix of the hypergraph, wherein the controllable noise represents spurious relationships and / or erroneous weights in the incidence matrix; Learning the controllable noise distribution through a neural network, removing noise from the correlation matrix to which the controllable noise is added, and obtaining a denoised correlation matrix; Performing a filtering operation on the denoised correlation matrix to obtain a first diffusion result; Alternatively, performing a second diffusion process on the optimized correlation matrix to obtain a second diffusion result includes: Converting the optimized association matrix into a probability transfer matrix that satisfies the propagation probability distribution of the risk event impact; fusing the probability transfer matrix and the node feature matrix through a hypergraph diffusion convolution to obtain a second diffusion result, wherein the node feature matrix represents a matrix of state information of nodes in the hypergraph; Alternatively, performing a third diffusion process on the matching relationship between the time characteristics of the risk events in the hypergraph and the transportation time of the logistics orders to obtain a third diffusion result includes: Matching and calculating the time characteristics of the risk events in the hypergraph with the transportation time of the logistics orders to obtain calculation results; A third diffusion result is determined based on the calculation result and the time decay curve.

7. The method according to claim 1, wherein the process of processing the association relationships in the target hypergraph using a diffusion model to obtain target impact information of the target risk event corresponding to the target logistics order comprises: Determining analysis dimensions of the candidate risk events; Based on the diffusion model and the analysis dimension, the association relationship in the target hypergraph is processed to obtain the diffusion result corresponding to each analysis dimension; Based on the diffusion results corresponding to each analysis dimension, a target risk event and target impact information corresponding to the target risk event are determined from the candidate risk events.

8. The method according to claim 7, wherein the analysis dimension includes at least one of a geographical impact dimension of the risk event, a risk event impact degree dimension, and a risk event time impact dimension; the diffusion model includes a first diffusion module corresponding to the geographical impact dimension of the risk event, a second diffusion module corresponding to the risk event impact degree dimension, and a third diffusion module corresponding to the risk event time impact dimension, wherein: The processing of the association relationships in the target hypergraph based on the diffusion model and the analysis dimension to obtain a diffusion result corresponding to each analysis dimension includes at least one of the following: Based on the first diffusion module, the first association relationship between the geographical location and the risk event in the target hypergraph is screened to obtain the actual impact geographical location corresponding to each candidate risk event; Analyzing the actual impact geographical location corresponding to each candidate risk event based on the second diffusion module to obtain the comprehensive impact degree of each candidate risk event on the target logistics order; Based on the third diffusion module, temporal reasoning processing is performed on the temporal features of the candidate risk events in the target hypergraph to obtain the degree of change of each candidate risk event over time.

9. The method according to claim 8, wherein determining a target risk event from the candidate risk events based on the diffusion results corresponding to each analysis dimension and target impact information corresponding to the target risk event comprises: Determine a target risk event from the candidate risk events based on the actual impact geographic location corresponding to each candidate risk event, the degree of change of each candidate risk event over time, and each geographic location in the transportation route of the target order; Based on the comprehensive impact of each of the candidate risk events on the target logistics order, target impact information corresponding to the target risk event is determined.

10. An electronic device comprising: A memory, used to store applications and data generated by the execution of the applications; A processor, configured to execute the application program to implement: Based on the transportation path of the target logistics order and the candidate risk events, a target hypergraph is generated, wherein the target hypergraph is used to represent the geographical location of the target logistics order in the transportation path and the distribution map of the candidate risk events corresponding to the geographical location; The association relationships in the target hypergraph are processed using a diffusion model to obtain target impact information of the target risk event corresponding to the target logistics order; wherein the association relationships include the relationship between the geographic locations in the target hypergraph and the corresponding candidate risk events; the target impact information represents the quantitative impact of the target risk event on the transportation time of the target logistics order; The transportation time of the target logistics order is determined according to the target impact information of the target risk event.

Images