Recommended methods, devices, equipment, and media for split-order delivery mode.

Abstract

This invention relates to the field of big data processing technology, and provides a method, apparatus, device, and medium for recommending shipment delivery patterns. The method includes: acquiring cargo information data; and performing cluster analysis based on the cargo information data and previously acquired reference clusters to determine the shipment delivery pattern of the cargo information data. The reference clusters are obtained by first determining the corresponding shipment delivery patterns for historical cargo information based on the indicator states and influence weights of historical cargo information, and then clustering the historical cargo information based on the determined shipment delivery patterns. The indicator states characterize whether the cargo indicators of the historical cargo information exceed the corresponding thresholds, and the influence weights characterize the contribution of the cargo indicators to customs clearance speed. This invention determines the shipment delivery pattern by comparing cargo information data with reference clusters, ensuring a near real-time decision-making process and avoiding inappropriate or inconsistent decision results due to different agents' interpretations, thus guaranteeing the consistency of strategy execution.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to a method, apparatus, equipment and medium for recommending order delivery modes. Background Technology

[0002] With the advancement of globalization, the international express delivery and cross-border e-commerce industries have flourished, prompting the continuous upgrading of the cross-border air logistics market. The ever-growing international express delivery and cross-border e-commerce businesses have posed significant challenges to the cargo handling capacity and efficiency of airport cargo terminals. For outbound shipments, international express delivery and cross-border e-commerce goods need to be split into multiple shipments upon arrival at the cargo terminal to meet relevant regulatory requirements.

[0003] Currently, some airports offer various partial shipment arrival modes for shipping agents of outbound international express and cross-border e-commerce goods, allowing them to choose according to different partial shipment arrival points. These include gate arrival mode, terminal arrival mode, and warehouse arrival mode (gate arrival mode – partial shipment is sent when the vehicle passes through the terminal gate; terminal arrival mode – partial shipment is sent after the goods have passed inspection; warehouse arrival mode – partial shipment is sent after the goods have passed security check). This respects the shipping agents' wishes and facilitates air freight operations for international express and cross-border e-commerce goods. When entering arrival information into the system, shipping agents can select the partial shipment arrival mode, i.e., the partial shipment arrival point. The cargo terminal will then switch the corresponding subsequent cargo processing procedures based on the partial shipment arrival mode selected by the shipping agent.

[0004] However, the split shipment delivery mode is chosen independently by the shipping agent. Different split shipment delivery modes correspond to different subsequent processing procedures. It is affected by many factors such as the number of pieces, weight, category, name of goods, flight departure time, and customs receipt time. If an inappropriate split shipment delivery mode is used, it will lead to a decrease in the efficiency of subsequent processing of goods, thereby slowing down the customs clearance process. Summary of the Invention

[0005] This invention provides a method, apparatus, equipment, and medium for recommending split-order delivery modes, which addresses the shortcomings of existing technologies where manually selecting split-order delivery modes can easily lead to unsuitable modes and affect cargo handling efficiency. It ensures near real-time decision-making, avoids interference from human factors, guarantees consistency in strategy execution, and improves the accuracy of decision-making.

[0006] This invention provides a method for recommending split-shipment delivery patterns, comprising: acquiring cargo information data; and performing cluster analysis based on the cargo information data and previously acquired reference clusters to determine the split-shipment delivery patterns of the cargo information data. The reference clusters are obtained by first determining the split-shipment delivery patterns corresponding to historical cargo information based on the indicator states and influence weights of historical cargo information, and then clustering the historical cargo information based on the determined split-shipment delivery patterns. The indicator states characterize whether the cargo indicators of the historical cargo information exceed the corresponding thresholds, and the influence weights characterize the contribution of the cargo indicators to customs clearance speed.

[0007] According to the present invention, a method for recommending a split-order delivery mode is provided. Based on cargo information data and combined with previously acquired reference clusters, a cluster analysis is performed to determine the split-order delivery mode of the cargo information data. The method includes: SA1, determining the Euclidean distance from the cargo information data to the cluster centers of each reference cluster based on the cargo information data and the previously acquired reference clusters; SA2, updating the corresponding cargo information data to the reference cluster with the smallest distance based on the Euclidean distance, obtaining the corresponding updated cluster; SA3, determining the cluster center distance based on the cluster center of the updated cluster and the cluster center of the corresponding reference cluster; SA4, based on the cluster center distance being greater than a target threshold, re-determining the Euclidean distance from the cargo information data to the cluster centers of each cluster based on the cargo information data, combined with the unupdated reference clusters and the updated clusters, and iteratively executing steps SA2-SA4 until the corresponding cluster center distance is less than the target threshold or the preset number of iterations is reached, and determining the split-order delivery mode of the corresponding cargo information data based on the finally obtained updated clusters.

[0008] According to the present invention, a method for recommending a split-order delivery pattern, before determining the split-order delivery pattern of cargo information data by performing cluster analysis based on cargo information data and previously acquired reference clusters, includes: traversing each target data in the cargo information data and checking the access status of the corresponding target data; if the access status is determined to be "accessed", skipping and processing the next target data; if the access status is determined to be "unaccessed", marking the corresponding target data as "accessed", and obtaining a neighborhood point set when the cargo information data is within the neighborhood of the target data; if the number of data in the neighborhood point set is determined to be greater than or equal to a preset minimum number of points, using the target data as the starting point of a new cluster, assigning a cluster identifier to the target data, and obtaining the density reachable points corresponding to the target data in the cargo information data and adding them to the new cluster to obtain an expanded cluster; if the number of data in the neighborhood point set is determined to be less than the preset minimum number of points, marking the corresponding target data as noise; and filtering out all target data marked as noise from the cargo information data.

[0009] According to the present invention, a method for recommending a split-order delivery pattern is provided. After determining the split-order delivery pattern of the cargo information data by performing cluster analysis based on cargo information data and previously acquired reference clusters, the method includes: acquiring all target data marked as noise to obtain noise data; for each noise data, determining the Euclidean distance between the noise data and the cluster centers of the finally obtained updated clusters and the unupdated reference clusters, and assigning the noise data to the cluster with the smallest distance.

[0010] According to the present invention, a method for recommending a split-shipment delivery pattern includes the following steps before determining the split-shipment delivery pattern of cargo information data by performing cluster analysis based on cargo information data and previously acquired reference clusters: acquiring historical cargo information, which includes the values ​​of at least two cargo information indicators, including at least two of the following: number of pieces, weight, category, cargo name, flight departure time, and customs receipt duration; comparing each cargo information indicator with its corresponding preset threshold based on the at least two cargo information indicators to determine the indicator status corresponding to each cargo information indicator; determining the corresponding influence weight of each cargo information indicator based on the at least two cargo information indicators and a structural causal model; wherein the structural causal model is constructed in advance based on the cargo information indicators of historical cargo information and the corresponding customs clearance speed of historical cargo information; determining the split-shipment delivery pattern of the corresponding historical cargo information based on the indicator status and influence weight of each cargo information indicator, and combining the correspondence between preset indicators and split-shipment delivery patterns; and clustering the historical cargo information using a preset clustering algorithm based on the split-shipment delivery pattern of the historical cargo information to obtain reference clusters.

[0011] According to the method for recommending a split-shipment delivery pattern provided by the present invention, after determining the split-shipment delivery pattern of the cargo information data by performing cluster analysis based on cargo information data and previously acquired reference clusters, the method further includes: obtaining the actual customs clearance speed; inputting each cargo information indicator and its corresponding influence weight into the customs clearance speed prediction model to extract the indicator features of each cargo information indicator, and determining the attention score of each indicator feature by combining an attention mechanism; determining the attention weight corresponding to each indicator feature based on the attention score of each indicator feature and its corresponding influence weight; obtaining a weighted fusion feature based on each indicator feature and its corresponding attention weight, and performing speed prediction based on the weighted fusion feature to obtain the predicted customs clearance speed; updating the customs clearance speed prediction model based on the actual customs clearance speed and the predicted customs clearance speed; obtaining the updated influence weight of the corresponding indicator feature based on the updated customs clearance speed prediction model, and updating the structural causal model based on the updated influence weight.

[0012] According to the present invention, a method for recommending a split-shipment delivery pattern, after determining the split-shipment delivery pattern of the cargo information data by performing cluster analysis based on the cargo information data and combining it with previously acquired reference clusters, further includes: obtaining the actual customs clearance rate, actual inspection rate, and actual operating cost, and obtaining an immediate reward by combining them with a preset reward function; updating the Q-value function of the reinforcement learning model based on the split-shipment delivery pattern, the updated cluster corresponding to the split-shipment delivery pattern, the immediate reward, and the next acquired cargo information data, to obtain the split-shipment delivery recommendation pattern output by the model; wherein, the Q-value function is constructed based on the expected returns obtained by each split-shipment delivery pattern and its corresponding updated cluster; and the method further includes: obtaining the split-shipment delivery recommendation pattern output by the model. In the recommendation mode, when different cargo information data within the same update cluster are assigned to at least two different order-based delivery recommendation modes, the information entropy of the distribution is determined based on all order-based delivery recommendation modes; or when different cargo information data within the same update cluster are assigned to the same order-based delivery mode and the Q values ​​differ, the variance of the Q value is determined. Based on the final updated cluster, the sum of squares within the corresponding updated cluster is determined. The information entropy or variance of the distribution is used to construct a loss function with the sum of squares within the corresponding updated cluster. Based on the convergence of the loss function, the preset clustering algorithm is updated, and the updated preset clustering algorithm is used to re-cluster the historical cargo information to obtain the updated reference cluster.

[0013] This invention also provides a device for recommending shipment delivery patterns, comprising: a data acquisition module for acquiring cargo information data; and a pattern recommendation module for performing cluster analysis based on the cargo information data and prior-acquired reference clusters to determine the shipment delivery patterns of the cargo information data; wherein, the reference clusters are obtained by prior-determining the shipment delivery patterns corresponding to historical cargo information based on the indicator states and influence weights of historical cargo information, and by clustering the historical cargo information based on the determined shipment delivery patterns; the indicator states are used to characterize whether the cargo indicators of historical cargo information exceed the corresponding thresholds, and the influence weights are used to characterize the contribution of cargo indicators to customs clearance speed.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the order splitting delivery mode recommendation method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the order delivery mode recommendation method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the order splitting and delivery mode recommendation method as described above.

[0017] The method, apparatus, equipment, and medium for recommending split-shipment delivery patterns provided by this invention explicitly and quantitatively integrate the tacit knowledge and experience of business experts into the clustering process through indicator status and influence weights. This makes the final reference cluster no longer a cold set of data, but a pattern template with clear business meaning. By collecting information on new goods that need to be processed, it can quickly compare them with the previously acquired reference clusters to determine the split-shipment delivery patterns for these new goods. This makes the decision-making process nearly real-time, ensuring that all subsequent new goods are processed according to the reference clusters, completely avoiding interference from human factors, ensuring the consistency of strategy execution, and because the reference clusters are generated based on a large amount of historical data and optimized weights, their recommendation accuracy is far higher than ad-hoc decisions or simple rule judgments, thus significantly improving the accuracy of decision-making. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the split-order delivery method provided by the present invention;

[0020] Figure 2 This is a schematic diagram of the structure of the order splitting and delivery mode device provided by the present invention;

[0021] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is a flowchart illustrating the method for recommending split-order delivery modes provided by the present invention, as shown below. Figure 1 As shown, the method includes:

[0024] S11, Obtain cargo information data;

[0025] S12. Based on the cargo information data, cluster analysis is performed using previously acquired reference clusters to determine the cargo information data's shipment delivery pattern. The reference clusters are obtained by clustering historical cargo information based on the indicator status and influence weights of historical cargo information, and the historical cargo information is clustered based on the determined shipment delivery patterns. The indicator status is used to characterize whether the cargo indicators of historical cargo information exceed the corresponding threshold, and the influence weights are used to characterize the contribution of cargo indicators to customs clearance speed.

[0026] It should be noted that the step number "S1N" in this specification does not represent the order of the recommended method for split-order delivery mode. The split-order delivery mode recommendation method of the present invention is described in detail below.

[0027] Step S11: Obtain cargo information data.

[0028] It should be noted that the cargo information data includes at least one cargo information item, and each cargo information item includes at least two cargo information indicators. The cargo information indicators include at least two of the following indicators: number of pieces, weight, category, cargo name, flight departure time, and customs receipt time. For example, the cargo information data may include indicators such as the number of cargo pieces, category, and flight departure time. The specific selection can be made according to the actual design requirements, and no further restrictions are made here.

[0029] Step S12: Based on the cargo information data, cluster analysis is performed using previously acquired reference clusters to determine the cargo information data's shipment delivery pattern. The reference clusters are obtained by first determining the shipment delivery pattern corresponding to historical cargo information based on the indicator status and influence weights of historical cargo information, and then clustering the historical cargo information based on the determined shipment delivery pattern. The indicator status is used to characterize whether the cargo indicators of historical cargo information exceed the corresponding threshold, and the influence weights are used to characterize the contribution of cargo indicators to customs clearance speed.

[0030] In this embodiment, based on the cargo information data and combined with the previously acquired reference clusters, cluster analysis is performed to determine the order-based delivery mode of the cargo information data. This includes: SA1, determining the Euclidean distance from the cargo information data to the cluster centers of each reference cluster based on the cargo information data and the previously acquired reference clusters; SA2, updating the corresponding cargo information data to the reference cluster with the smallest distance based on the Euclidean distance, obtaining the corresponding updated cluster; SA3, determining the cluster center distance based on the cluster center of the updated cluster and the cluster center of the corresponding reference cluster; SA4, based on the cluster center distance being greater than a target threshold, re-determining the Euclidean distance from the cargo information data to the cluster centers of each cluster based on the cargo information data, combined with the unupdated reference clusters and the updated clusters, and iteratively executing steps SA2-SA4 until the corresponding cluster center distance is less than the target threshold or the preset number of iterations is reached. Based on the finally obtained updated clusters, the order-based delivery mode of the corresponding cargo information data is determined.

[0031] In an optional embodiment, since outliers can severely deflect the centroid of clusters, leading to inaccurate cluster division, before performing cluster analysis based on cargo information data and previously acquired reference clusters to determine the order delivery pattern of cargo information data, the process includes: traversing each target data in the cargo information data and checking the access status of the corresponding target data; if the access status is "accessed," skipping and processing the next target data; if the access status is "not accessed," marking the corresponding target data as "accessed," and obtaining a neighborhood point set when the cargo information data is within the neighborhood of the target data; if the number of data in the neighborhood point set is greater than or equal to a preset minimum number of points, using the target data as the starting point of a new cluster, assigning a cluster identifier to the target data, and obtaining the density reachable points corresponding to the target data in the cargo information data and adding them to the new cluster to obtain an extended cluster; if the number of data in the neighborhood point set is less than the preset minimum number of points, marking the corresponding target data as noise; and filtering out all target data marked as noise from the cargo information data.

[0032] It should be added that if the cargo information data includes only one piece of cargo information data, the target data can be understood as the cargo information indicator in that piece of cargo information data; if the cargo information data includes at least two pieces of cargo information data, the target data can be any piece of cargo information data. The target data can be set according to the actual cargo information data and design requirements, and no further restrictions are made here.

[0033] If the target data is a cargo information indicator in a cargo information data set, then edge anomalies are filtered using the above method so that they can be assigned to the cluster with the smallest distance, thereby avoiding information loss that may be caused by direct discarding. If the target data is any one of at least two cargo information data sets, then edge anomalies are filtered using the above method so that they can be assigned to the cluster with the smallest distance, thereby determining the order delivery mode of the corresponding edge anomaly. The specific method of assigning edge anomalies to the cluster with the smallest distance can be found in the following text, and will not be repeated here.

[0034] It should be noted that before determining the shipment delivery pattern of the cargo information data by performing cluster analysis based on the cargo information data and the previously obtained reference clusters, the density clustering algorithm (DBSCAN) can be used to process outlier points at the neighborhood edges to temporarily remove noisy data marked by DBSCAN from the data.

[0035] Accordingly, after performing cluster analysis based on cargo information data and previously acquired reference clusters to determine the order delivery pattern of cargo information data, the process includes: acquiring all target data marked as noise to obtain noise data; for each noise data, determining the Euclidean distance between the noise data and the finally obtained updated cluster center and the unupdated reference cluster center, and assigning the noise data to the cluster with the smallest distance.

[0036] It should be added that if the target data is any one of at least two cargo information data sets, then after assigning the noisy data to the cluster with the smallest distance, the following steps are taken: based on the cluster assigned to the noisy data, determine the order delivery mode corresponding to the noisy data. Furthermore, by determining the cluster center distance between temporarily removed noise and the updated cluster, and assigning it to the updated cluster with the smallest distance, outliers are prevented from interfering with the core clustering, while the data information corresponding to the noise is preserved. This avoids the misclassification of outliers by traditional clustering methods, resulting in a higher quality and more business-interpretive clustering result.

[0037] In an optional embodiment, before determining the shipment delivery pattern of the cargo information data by performing cluster analysis based on the cargo information data and a previously acquired reference cluster, the process includes: acquiring historical cargo information, which includes the values ​​of at least two cargo information indicators, including at least two of the following: number of pieces, weight, category, cargo name, flight departure time, and customs receipt duration; comparing each cargo information indicator with a corresponding preset threshold based on the at least two cargo information indicators to determine the indicator status corresponding to each cargo information indicator; determining the influence weight corresponding to each cargo information indicator based on the at least two cargo information indicators and a structural causal model; wherein the structural causal model is constructed prior to the historical cargo information indicators and corresponding customs clearance speeds; determining the shipment delivery pattern of the corresponding historical cargo information based on the indicator status and influence weight of each cargo information indicator, and combining the correspondence between preset indicators and shipment delivery patterns; and clustering the historical cargo information using a preset clustering algorithm based on the shipment delivery pattern of the historical cargo information to obtain reference clusters.

[0038] It should be noted that the cluster center of the reference cluster can be selected from the data points within that cluster, i.e., by calculating the mean of all data points within that cluster. Furthermore, the preset clustering algorithm can be chosen based on actual design requirements or prior experience, such as K-means clustering or spectral clustering; no further limitations are made here. It is worth noting that when spectral clustering is used as the preset clustering algorithm, it can overcome the limitations of K-means clustering for non-convex data.

[0039] In addition, the preset thresholds for each cargo information indicator can be set according to actual design requirements or prior experience. For example, the threshold for the cargo number indicator can be set to 100 pieces. If the cargo number indicator is less than 100 pieces, the indicator status is less than the threshold. Another example is that the threshold for the flight departure time indicator can be set to 4 hours. If the flight departure time indicator exceeds 4 hours, the indicator status is greater than the threshold.

[0040] Furthermore, for time-related cargo information indicators, such as flight departure time indicators, time-sensitive thresholds can be used. This involves using a Long Short-Term Memory (LSTM) neural network to predict the probability of flight delays and then dynamically adjusting the threshold for flight departure time indicators based on the prediction results, thereby breaking the traditional fixed thresholds and static rules.

[0041] In addition, the correspondence between indicators and shipment arrival modes can be set based on prior experience. For example, shipment arrival modes include three types: checkpoint arrival, port arrival, and warehousing arrival. For shipment information where the number of goods exceeds the corresponding preset threshold, the customs declaration category is express shipment A, and the time before flight departure is less than 2 hours, the shipment arrival mode is marked as checkpoint arrival. For shipment information where the number of goods does not exceed the corresponding preset threshold, the customs declaration category is express shipment D, and the time before flight departure is more than 3 hours, the shipment arrival mode is marked as warehousing arrival. For other situations, the shipment information indicators and their influence weights are marked as port arrival.

[0042] Accordingly, after determining the shipment delivery pattern based on cargo information data and prior reference clusters through cluster analysis, the process further includes: obtaining actual customs clearance rate, actual inspection rate, and actual operating cost, and combining these with a preset reward function to obtain an immediate reward; updating the Q-value function of the reinforcement learning model based on the shipment delivery pattern, the corresponding updated cluster, the immediate reward, and the next acquired cargo information data, to obtain the shipment delivery recommendation pattern output by the model; wherein, the Q-value function is constructed based on the expected returns obtained by each shipment delivery pattern and its corresponding updated cluster; and determining the same shipment delivery recommendation pattern based on the model output. When different cargo information data within an updated cluster are assigned to at least two different order-based delivery recommendation modes, the information entropy of the distribution is determined based on all order-based delivery recommendation modes; or, when different cargo information data within the same updated cluster are assigned to the same order-based delivery mode and the Q values ​​differ, the variance of the Q value is determined. Based on the final updated cluster, the sum of squares within the corresponding updated cluster is determined. The information entropy or variance of the distribution is used to construct a loss function with the sum of squares within the corresponding updated cluster. Based on the convergence of the loss function, the preset clustering algorithm is updated. The updated preset clustering algorithm is then used to re-cluster the historical cargo information to obtain the updated reference cluster.

[0043] It should be added that updating the reference cluster in the above manner proactively avoids areas that could lead to decision-making confusion, fundamentally ensuring the accuracy of the clustering results. Furthermore, based on the order-based delivery model, changes are made to any cargo information indicator, and the updated Q-value function is used to determine the change in Q-value. The importance of the corresponding cargo information indicator is then determined based on the change in Q-value, thereby updating the preset clustering algorithm. For example, if the preset clustering algorithm is K-means, the weights corresponding to the Euclidean distance can be updated; no further limitations are made here.

[0044] In an optional embodiment, after determining the shipment delivery pattern of the cargo information data by performing cluster analysis based on the cargo information data and the previously acquired reference clusters, the method further includes: obtaining the actual customs clearance speed; inputting each cargo information indicator and its corresponding influence weight into the customs clearance speed prediction model to extract the indicator features of each cargo information indicator, and determining the attention score of each indicator feature by combining an attention mechanism; determining the attention weight corresponding to each indicator feature based on the attention score of each indicator feature and its corresponding influence weight; obtaining a weighted fusion feature based on each indicator feature and its corresponding attention weight, and performing speed prediction based on the weighted fusion feature to obtain the predicted customs clearance speed; updating the customs clearance speed prediction model based on the actual customs clearance speed and the predicted customs clearance speed; obtaining the updated influence weight of the corresponding indicator feature based on the updated customs clearance speed prediction model, and updating the structural causal model based on the updated influence weight.

[0045] It should be added that when determining the attention weights corresponding to the indicator features based on the attention scores of each indicator feature and the corresponding influence weights of each cargo information indicator, the influence weights corresponding to the cargo information indicators can be used as bias terms or multiplicative factors of the attention score to directly intervene in the generation process of attention weights and improve the performance and interpretability of the model.

[0046] In summary, this invention, through indicator status and influence weights, explicitly and quantitatively integrates the tacit knowledge and experience of business experts into the clustering process. This transforms the final reference clusters from cold, impersonal data sets into pattern templates with clear business implications. By collecting information on new goods that need to be processed, and quickly comparing it with previously acquired reference clusters, the delivery patterns for these new goods can be determined. This makes the decision-making process nearly real-time, ensuring that all subsequent new goods are processed based on the reference clusters. This completely avoids interference from human factors, guarantees the consistency of strategy execution, and because the reference clusters are generated based on a large amount of historical data and optimized weights, their recommendation accuracy is far higher than ad-hoc decisions or simple rule judgments, thus significantly improving the accuracy of decision-making.

[0047] The following describes the split-order delivery mode recommendation device provided by the present invention. The split-order delivery mode recommendation device described below and the split-order delivery mode recommendation method described above can be referred to in correspondence.

[0048] Figure 2 A schematic diagram of a split-order delivery mode recommendation device is shown. The device includes:

[0049] Data acquisition module 21 acquires cargo information data;

[0050] The pattern recommendation module 22 performs cluster analysis based on cargo information data and previously acquired reference clusters to determine the cargo information data's order-based delivery patterns. The reference clusters are obtained by first determining the order-based delivery patterns corresponding to historical cargo information based on the indicator status and influence weights of historical cargo information, and then clustering the historical cargo information based on the determined order-based delivery patterns. The indicator status is used to characterize whether the cargo indicators of historical cargo information exceed the corresponding threshold, and the influence weight is used to characterize the contribution of cargo indicators to customs clearance speed.

[0051] In this embodiment, the pattern recommendation module 22 includes: a Euclidean distance determination unit, which determines the Euclidean distance from the cargo information data to the center of each reference cluster based on the cargo information data and the previously acquired reference clusters; an update unit, which updates the corresponding cargo information data to the reference cluster with the smallest distance based on the Euclidean distance, thereby obtaining the corresponding updated cluster; a cluster center distance determination unit, which determines the cluster center distance based on the cluster center of the updated cluster and the cluster center of the corresponding reference cluster; and an iteration unit, which, based on the cluster center distance being greater than a target threshold, re-determines the Euclidean distance from the cargo information data to the center of each cluster based on the cargo information data and the unupdated reference clusters and the updated clusters, and iteratively executes the steps from the update unit to the iteration unit until the corresponding cluster center distance is less than the target threshold or the preset number of iterations is reached, and determines the order delivery mode of the corresponding cargo information data based on the finally obtained updated clusters.

[0052] In an optional embodiment, the device further includes a discrete point filtering module, configured to: before performing cluster analysis based on the cargo information data and previously acquired reference clusters to determine the order delivery pattern of the cargo information data, traverse each target data in the cargo information data and check the access status of the corresponding target data; if the access status is determined to be accessed, skip and process the next target data; if the access status is determined to be unaccessed, mark the corresponding target data as accessed, and if the cargo information data is located within the neighborhood of the target data, obtain a neighborhood point set; if the number of data in the neighborhood point set is greater than or equal to a preset minimum number of points, use the target data as the starting point of a new cluster, assign a cluster identifier to the target data, and obtain the density reachable points corresponding to the target data in the cargo information data and add them to the new cluster to obtain an extended cluster; if the number of data in the neighborhood point set is less than the preset minimum number of points, mark the corresponding target data as noise; and filter out all target data marked as noise from the cargo information data.

[0053] Correspondingly, the pattern recommendation module 22 also includes: a noise data acquisition unit, which, after performing cluster analysis based on the cargo information data and the previously acquired reference clusters to determine the order delivery pattern of the cargo information data, acquires all target data marked as noise to obtain noise data; and a pattern recommendation unit, which, for each noise data, determines the Euclidean distance between the noise data and the finally obtained updated cluster center and the unupdated reference cluster center, and assigns the noise data to the cluster with the smallest distance.

[0054] It should be added that if the target data is any one of the at least two cargo information data, then the pattern recommendation module 22 also includes: a pattern determination unit, which, after assigning the noise data to the cluster with the smallest distance, determines the order delivery pattern corresponding to the noise data based on the cluster to which the noise data is assigned.

[0055] In an optional embodiment, the device further includes a reference cluster generation module, configured to: acquire historical cargo information before performing cluster analysis based on cargo information data and previously acquired reference clusters to determine the cargo information data's order-based delivery pattern; acquire historical cargo information including the values ​​of at least two cargo information indicators, which include at least two of the following: number of pieces, weight, category, cargo name, flight departure time, and customs receipt duration; compare each cargo information indicator with its corresponding preset threshold based on the at least two cargo information indicators to determine the indicator status corresponding to each cargo information indicator; determine the corresponding influence weight of each cargo information indicator based on the at least two cargo information indicators and a structural causal model; wherein the structural causal model is constructed prior to the historical cargo information's cargo information indicators and corresponding customs clearance speeds; determine the corresponding order-based delivery pattern for the historical cargo information based on the indicator status and influence weight of each cargo information indicator, combined with the correspondence between preset indicators and order-based delivery patterns; and cluster the historical cargo information using a preset clustering algorithm based on the historical cargo information's order-based delivery pattern to obtain reference clusters.

[0056] Accordingly, the device also includes a clustering update module, used for: after determining the order-based delivery mode of the cargo information data by performing clustering analysis based on the cargo information data and the previously acquired reference clusters, obtaining the actual customs clearance rate, actual inspection rate, and actual operating cost, and combining them with a preset reward function to obtain an immediate reward; updating the Q-value function of the reinforcement learning model based on the order-based delivery mode, the updated cluster corresponding to the order-based delivery mode, the immediate reward, and the next acquired cargo information data, to obtain the order-based delivery recommendation mode output by the model; wherein, the Q-value function is constructed based on the expected returns obtained by each order-based delivery mode and its corresponding updated cluster; and based on the order-based delivery recommendation mode output by the model. The model is as follows: when different cargo information data within the same update cluster are assigned to at least two different order-based delivery recommendation modes, the information entropy of the distribution is determined based on all order-based delivery recommendation modes; or when different cargo information data within the same update cluster are assigned to the same order-based delivery recommendation mode and the Q-values ​​differ, the variance of the Q-value is determined. Based on the final updated cluster, the sum of squares within the corresponding updated cluster is determined. The information entropy or variance of the distribution is used to construct a loss function with the sum of squares within the corresponding updated cluster. Based on the convergence of the loss function, the preset clustering algorithm is updated. The updated preset clustering algorithm is then used to re-cluster the historical cargo information to obtain the updated reference cluster.

[0057] In an optional embodiment, the device further includes a model update module, configured to: obtain the actual customs clearance speed after performing cluster analysis based on cargo information data and previously acquired reference clusters to determine the order delivery pattern of cargo information data; input each cargo information indicator and its corresponding influence weight from historical cargo information into the customs clearance speed prediction model to extract the indicator features of each cargo information indicator, and determine the attention score of each indicator feature by combining an attention mechanism; determine the attention weight corresponding to each indicator feature based on the attention score of each indicator feature and its corresponding influence weight; obtain a weighted fusion feature based on each indicator feature and its corresponding attention weight, and perform speed prediction based on the weighted fusion feature to obtain the predicted customs clearance speed; update the customs clearance speed prediction model based on the actual customs clearance speed and the predicted customs clearance speed; obtain the updated influence weight of the corresponding indicator feature based on the updated customs clearance speed prediction model, and update the structural causal model based on the updated influence weight.

[0058] In summary, this invention, through indicator status and influence weights, explicitly and quantitatively integrates the tacit knowledge and experience of business experts into the clustering process. This transforms the final reference clusters from cold, impersonal data sets into pattern templates with clear business implications. By collecting information on new goods that need to be processed, and quickly comparing it with previously acquired reference clusters, the delivery patterns for these new goods can be determined. This makes the decision-making process nearly real-time, ensuring that all subsequent new goods are processed based on the reference clusters. This completely avoids interference from human factors, guarantees the consistency of strategy execution, and because the reference clusters are generated based on a large amount of historical data and optimized weights, their recommendation accuracy is far higher than ad-hoc decisions or simple rule judgments, thus significantly improving the accuracy of decision-making.

[0059] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a split-shipment delivery mode recommendation method. This method includes: acquiring cargo information data; performing cluster analysis based on the cargo information data and previously acquired reference clusters to determine the split-shipment delivery mode of the cargo information data; wherein the reference clusters are obtained by first determining the split-shipment delivery mode corresponding to historical cargo information based on the indicator status and influence weight of historical cargo information, and then clustering the historical cargo information based on the determined split-shipment delivery mode. The indicator status is used to characterize whether the cargo indicators of historical cargo information exceed the corresponding threshold, and the influence weight is used to characterize the contribution of the cargo indicators to the customs clearance speed.

[0060] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the order-based delivery mode recommendation method provided by the above methods. The method includes: acquiring cargo information data; performing cluster analysis based on the cargo information data and prior-acquired reference clusters to determine the order-based delivery mode of the cargo information data; wherein, the reference clusters are obtained by prior-determining the order-based delivery mode corresponding to historical cargo information based on the indicator status and influence weight of historical cargo information, and clustering the historical cargo information based on the determined order-based delivery mode. The indicator status is used to characterize whether the cargo indicators of historical cargo information exceed the corresponding threshold, and the influence weight is used to characterize the contribution of cargo indicators to customs clearance speed.

[0062] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the order-based delivery mode recommendation method provided by the above methods. The method includes: acquiring cargo information data; performing cluster analysis based on the cargo information data and prior-acquired reference clusters to determine the order-based delivery mode of the cargo information data; wherein, the reference clusters are obtained by prior-determining the order-based delivery mode corresponding to historical cargo information based on the indicator status and influence weight of historical cargo information, and clustering the historical cargo information based on the determined order-based delivery mode; the indicator status is used to characterize whether the cargo indicator of the historical cargo information exceeds the corresponding threshold, and the influence weight is used to characterize the contribution of the cargo indicator to the customs clearance speed.

[0063] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0064] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for recommending a separate delivery mode, characterized by, The method comprises the following steps: obtaining cargo information data; determining the single delivery mode of the cargo information data by clustering analysis according to the cargo information data and the reference cluster obtained in advance, wherein the reference cluster is obtained by determining the single delivery mode of historical cargo information based on the index state and influence weight of the historical cargo information in advance, and clustering the historical cargo information based on the determined single delivery mode, the index state is used to represent the state of whether the cargo index of the historical cargo information exceeds the corresponding threshold, and the influence weight is used to represent the contribution of the cargo index to the customs clearance speed; before determining the single delivery mode of the cargo information data by clustering analysis according to the cargo information data and the reference cluster obtained in advance, the method comprises the following steps: obtaining historical cargo information, the historical cargo information comprising values of at least two cargo information indexes, the at least two cargo information indexes comprising at least two of piece quantity, weight, category, cargo name, flight departure time and customs receipt time length; determining the index state corresponding to each cargo information index by comparing the size of each cargo information index with the corresponding preset threshold according to the at least two cargo information indexes; determining the influence weight corresponding to each cargo information index by combining a structural causal model according to the at least two cargo information indexes, wherein the structural causal model is constructed according to each cargo information index of historical cargo information and the customs clearance speed corresponding to the historical cargo information in advance; determining the single delivery mode corresponding to the historical cargo information according to the index state corresponding to each cargo information index and the influence weight corresponding to each cargo information index, and combining the corresponding relationship between the preset index and the single delivery mode; clustering the historical cargo information according to the single delivery mode of the historical cargo information by using a preset clustering algorithm to obtain a reference cluster.

2. The method according to claim 1, wherein determining the single delivery mode of the cargo information data by clustering analysis according to the cargo information data and the reference cluster obtained in advance, comprises the following steps: SA1, determining the Euclidean distance from the cargo information data to the cluster center of each reference cluster according to the cargo information data and the reference cluster obtained in advance; SA2, updating the corresponding cargo information data to the reference cluster with the smallest distance according to the Euclidean distance to obtain an updated cluster; SA3, determining the cluster center distance according to the cluster center of the updated cluster and the cluster center of the corresponding reference cluster of the updated cluster; SA4, based on the cluster center distance being greater than a target threshold, re-determining the Euclidean distance from the cargo information data to the cluster center of each cluster according to the cargo information data, the reference cluster not updated and the updated cluster, and iteratively executing steps SA2-SA4 until the corresponding cluster center distance is less than the target threshold or a preset iteration number is reached, and determining the single delivery mode of the corresponding cargo information data according to the updated cluster obtained finally.

3. The method according to claim 2, wherein before determining the single delivery mode of the cargo information data by clustering analysis according to the cargo information data and the reference cluster obtained in advance, the method comprises the following steps: Traverse each target data in the cargo information data, and check the access state of the corresponding target data. When the access state is determined as accessed, the next target data is skipped and processed; When the access state is determined as not accessed, the corresponding target data is marked as accessed, and when the cargo information data is determined to be located in the neighborhood range of the target data, a neighborhood point set is obtained; When the number of data in the neighborhood point set is greater than or equal to a preset minimum point number, the target data is taken as a starting point of a new cluster, the target data is assigned a cluster identifier, and density reachable points corresponding to the target data in the cargo information data are obtained and added to the new cluster to obtain an extended cluster; When the number of data in the neighborhood point set is less than the preset minimum point number, the corresponding target data is marked as noise; All target data marked as noise is filtered out from the cargo information data.

4. The method according to claim 3, wherein, After clustering analysis is performed on the cargo information data in combination with the previously obtained reference cluster to determine the single delivery mode of the cargo information data, the method further includes: Obtaining all target data marked as noise to obtain noise data; For each noise data, the Euclidean distance between the noise data and the cluster center of the finally obtained updated cluster and the cluster center of the un-updated reference cluster is determined, and the noise data is assigned to the cluster with the smallest distance.

5. The method of claim 1, wherein, After clustering analysis is performed on the cargo information data in combination with the previously obtained reference cluster to determine the single delivery mode of the cargo information data, the method further includes: Obtaining an actual customs clearance speed; Each cargo information index of the historical cargo information and the corresponding influence weight of each cargo information index are input into a customs clearance speed prediction model to extract index features of each cargo information index, and the attention mechanism is combined to determine the attention score of each index feature; According to the attention score of each index feature, the corresponding influence weight of each cargo information index is determined to determine the attention weight corresponding to the index feature; According to each index feature and the attention weight corresponding to the index feature, a weighted fusion feature is obtained, and speed prediction is performed based on the weighted fusion feature to obtain a customs clearance prediction speed; According to the actual customs clearance speed and the customs clearance prediction speed, the customs clearance speed prediction model is updated; According to the updated influence weight of the corresponding index feature, the structural causal model is updated.

6. The method of claim 1, wherein, After clustering analysis is performed on the cargo information data in combination with the previously obtained reference cluster to determine the single delivery mode of the cargo information data, the method further includes: Obtaining an actual customs clearance speed, an actual inspection rate and an actual operation cost, and combining a preset reward function to obtain an immediate reward; According to the single delivery mode, the updated cluster corresponding to the single delivery mode, the immediate reward and the next obtained cargo information data, the Q value function of the reinforcement learning model is updated to obtain a single delivery recommendation mode output by the model; wherein the Q value function is constructed according to the expected income of each single delivery mode and the updated cluster corresponding thereto; When the different goods information data in the same update cluster are assigned at least two different single-shipment delivery recommendation modes according to the single-shipment delivery recommendation mode output by the model, the distribution information entropy is determined according to all the single-shipment delivery recommendation modes, or the Q value variance is determined when the different goods information data in the same update cluster are assigned the same single-shipment delivery recommendation mode and the Q values are different; The intra-cluster sum of squares corresponding to the update cluster is determined according to the finally obtained update cluster; The distribution information entropy or the Q value variance is constructed into a loss function with the intra-cluster sum of squares corresponding to the update cluster, and the preset clustering algorithm is updated based on the convergence of the loss function, and the historical goods information is re-clustered by using the updated preset clustering algorithm to obtain an updated reference cluster.

7. A single delivery mode recommendation device, characterized by comprising: Comprise: A data acquisition module that acquires goods information data; A mode recommendation module that determines a single-shipment delivery mode of the goods information data by performing clustering analysis on the goods information data in combination with a previously acquired reference cluster; wherein the reference cluster is obtained by clustering historical goods information based on index states and influence weights of the historical goods information, the index states being used to represent states of whether goods indexes of the historical goods information exceed corresponding thresholds, and the influence weights being used to represent contribution degrees of the goods indexes to customs clearance speed; The device further comprises a reference cluster generation module that is configured to: acquire historical goods information before determining the single-shipment delivery mode of the goods information data by performing clustering analysis on the goods information data in combination with the previously acquired reference cluster, the historical goods information comprising values of at least two goods information indexes, the at least two goods information indexes comprising at least two of piece count, weight, category, goods name, flight departure time, and customs receipt duration; determine corresponding index states of the goods information indexes according to the at least two goods information indexes by comparing sizes of the goods information indexes and corresponding preset thresholds; determine corresponding influence weights of the goods information indexes by combining a structural causal model according to the at least two goods information indexes; wherein the structural causal model is constructed according to the goods information indexes of historical goods information and corresponding customs clearance speeds of the historical goods information; determine a single-shipment delivery mode corresponding to the historical goods information by combining a preset index and a corresponding relationship between the single-shipment delivery mode according to the index states of the goods information indexes and the influence weights of the goods information indexes; cluster the historical goods information by using a preset clustering algorithm to obtain the reference cluster according to the single-shipment delivery mode of the historical goods information.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor implements the single-shipment delivery mode recommendation method according to any one of claims 1 to 6 when executing the computer program. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the single-shipment delivery mode recommendation method according to any one of claims 1 to 6 when executed by the processor.

Images