Cross-border e-commerce logistics order management system based on big data

Abstract

The invention relates to the technical field of logistics order delivery management, in particular to a cross-border e-commerce logistics order management system based on big data, which comprises a data acquisition processing unit, an order clustering delivery unit and a delivery execution feedback unit, the data acquisition and processing unit collects orders, geographic information and real-time traffic data from a cross-border e-commerce platform and performs cleaning preprocessing, repeated orders are removed by using a Hash algorithm, the order clustering and delivery unit clusters the orders by using a density peak clustering algorithm, and the clustering effect is optimized by combining the order emergency degree and the customer loyalty, so that the service quality of the cross-border e-commerce platform is improved. A genetic algorithm is adopted to optimize the delivery sequence, the crossover and mutation probability is dynamically adjusted, a fitness function is calculated by considering the delivery distance, time cost and regional limitation penalty factors, a delivery execution feedback unit sends an optimized route to a delivery vehicle, client feedback evaluation data is collected, a delivery strategy is dynamically adjusted by means of an upper confidence bound algorithm, and the delivery efficiency is improved. And the delivery efficiency and the customer satisfaction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics order delivery management, and specifically, to a cross-border e-commerce logistics order management system based on big data. Background Art

[0002] Logistics order delivery management is an important technology. At present, with the booming development of the e-commerce industry, the application prospect of logistics order delivery management is extremely broad. It is not only the key to ensuring the efficient delivery of goods and enhancing the customer shopping experience, but also helps e-commerce enterprises reduce costs and increase efficiency, enhance market competitiveness, and promote the sustainable development of the entire industry.

[0003] Cross-border e-commerce orders come from a wide range of sources, covering different markets around the world, and the order data formats vary greatly. This makes it difficult for traditional data processing algorithms to quickly and accurately classify and sort orders, and it is impossible to form effective clustering. Moreover, cross-border logistics involves different logistics nodes and transportation methods in multiple countries and regions, and the information flow in each link is not smooth, making it difficult to summarize and integrate to form a comprehensive data view. Therefore, it is impossible to optimize and plan the delivery route based on complete data. Delivery personnel can only blindly plan routes, resulting in repeated routes and wasted time. Due to the complexity of the cross-border logistics scenario, the traffic information collection and release standards in different countries and regions are not uniform, and it is difficult to obtain global traffic dynamic information in real time, resulting in the inability to incorporate real-time traffic conditions into the delivery plan. Regarding the urgency of orders, cross-border e-commerce lacks a unified order urgency identification system, and there are differences in the definitions and judgment criteria of urgency between merchants and logistics parties. It is difficult to quickly identify urgent orders during the order processing process, resulting in the mixed delivery of ordinary orders and urgent orders, affecting the logistics order delivery efficiency. To solve this technical problem, we provide a cross-border e-commerce logistics order management system based on big data. Summary of the Invention

[0004] The purpose of the present invention is to provide a cross-border e-commerce logistics order management system based on big data to solve the problems raised in the above background art.

[0005] To achieve the above purpose, a cross-border e-commerce logistics order management system based on big data is provided, including a data collection and processing unit, an order clustering and delivery unit, and a delivery execution and feedback unit;

[0006] The data collection and processing unit is used to collect cross-border e-commerce logistics order data from cross-border e-commerce platforms, and at the same time obtain the regional geographic information data and real-time traffic data, and clean and preprocess the collected data;

[0007] The order clustering and delivery unit uses the density peak clustering algorithm to cluster the pre - processed order data. By calculating the local density of each order data point, it evaluates the density of orders around this point. At the same time, it determines the relative distance between each point and the points with higher density than it, and identifies the order aggregation areas by drawing a decision graph. Finally, it assigns the order data points that are not cluster centers to the corresponding clusters. For each group of orders after clustering, a genetic algorithm is used to optimize the delivery order. In terms of the coding method, integer coding with added regional identification information is used to define the position of each order in the delivery sequence and its belonging area, and the crossover probability and mutation probability are dynamically adjusted according to population diversity and individual fitness. Then, using a fitness function that includes delivery distance, time cost, and regional restriction penalty factors, it evaluates the advantages and disadvantages of each delivery order plan. In the genetic operation link, the roulette wheel selection method is used to select high - quality individuals, and better delivery orders are generated through partially mapped crossover and swap mutation operations. After multiple iterations of optimization, the optimal delivery route is determined for each group of orders;

[0008] The delivery execution feedback unit sends the optimized delivery route to the delivery vehicle, collects customer feedback evaluation data after delivery, and adjusts the delivery strategy according to the feedback evaluation data.

[0009] As a further improvement of this technical solution, when the data acquisition and processing unit cleans the cross - border e - commerce logistics order data collected, for duplicate order data, the following duplicate removal method based on the hash algorithm is adopted:

[0010] Extract key information from each order. The key information includes order number, customer name, commodity details, and order time, and form a string with the key information;

[0011] Use a hash function to calculate the combined string to obtain the hash value of each order, store the hash values of all orders in a hash table, traverse the hash table. If it is found that the hash values of different orders are the same, then compare the order amounts and delivery addresses of the orders to determine duplicate orders. For the determined duplicate orders, only keep one of them and delete the other duplicate orders.

[0012] As a further improvement of this technical solution, the order clustering and delivery unit includes a local density calculation module. When the local density calculation module calculates the local density of each order data point using the density peak clustering algorithm, the following method is adopted:

[0013] Extract the longitude and latitude data from the geographical location information of the orders, convert it into plane coordinates, then calculate the distance between orders using the weighted Euclidean distance formula, calculate the mean and standard deviation of the distances between all orders, draw a histogram of the distance data, observe the distribution pattern of the data, determine whether there is a clustering structure, and determine the cut-off distance according to the clustering structure;

[0014] Substitute the cut-off distance and the distance between orders into the Gaussian kernel function to calculate the local density of each order data point. The local density is used to reflect the density of orders in this area, and then introduce the order urgency and customer loyalty;

[0015] Set the order urgency of orders with a delivery time within 24 hours to 3, those within 24 - 48 hours to 2, and those over 48 hours to 1. Set the customer loyalty of customers with both the historical order quantity and consumption amount exceeding the preset threshold to 3, medium to 2, and low to 1. Adjust the local density value according to the order urgency and customer loyalty to obtain the adjusted local density.

[0016] As a further improvement of this technical solution, the order clustering and delivery unit includes a reference point selection module. When the reference point selection module determines the relative distance between each point and the points with higher density than it, if there are multiple points with higher density than the current point and equal distance, the following detailed method steps are used to select the reference point:

[0017] In addition to considering the order amount, introduce the order urgency, customer priority, and commodity type importance as evaluation attributes. The customer priority is determined according to the consumption amount, and the commodity type importance is determined according to the commodity value. Assign weights to each attribute according to the actual situation;

[0018] Obtain the actual amount value of the order from the data acquisition and processing unit, and convert the order urgency, customer priority, and commodity type importance into corresponding values according to the defined correspondence between grades and values. For each point with higher density than the current point and equal distance, calculate the comprehensive score;

[0019] Compare the comprehensive scores of all points with higher density than the current point and equal distance, select the point with the highest comprehensive score as the reference point to calculate the relative distance. If there are multiple points with the same and highest comprehensive score, compare the order amounts corresponding to the points, and select the point with the largest key attribute value as the reference point.

[0020] As a further improvement of this technical solution, when the order clustering and delivery unit uses the genetic algorithm to optimize the delivery order, the calculation of the fitness function adopts the following steps:

[0021] According to the geographical information data and the delivery addresses of the orders, use the path planning algorithm to calculate the total delivery distance of each delivery sequence. Combine the real-time traffic data, consider the traffic congestion conditions in different time periods, and calculate the total time cost of each delivery sequence.

[0022] According to the delivery area rules of the orders, check whether each delivery sequence violates the area restrictions. If it violates, determine the severity of the violation according to the violation threshold, then determine the value of the penalty factor according to the severity, and finally set the weight coefficient according to historical experience. Combine the total delivery distance, the total time cost, the value of the penalty factor, and the corresponding weight coefficient to calculate the fitness function.

[0023] As a further improvement of this technical solution, in the crossover operation of the genetic algorithm, the order clustering and delivery unit adopts a method combining two-point crossover and order crossover. The specific steps are as follows:

[0024] Before performing the crossover operation, for the gene encoding of each individual, calculate its Hamming distance from the gene encoding of other individuals, and then calculate the average value of the Hamming distances between all individuals. Measure the population diversity through the average value of the Hamming distances, and then calculate the fitness value of each individual according to the fitness function.

[0025] When the population diversity is lower than the population diversity threshold, it indicates that the individuals in the population have high similarity, so increase the crossover probability. When the individual fitness exceeds the threshold, in order to retain excellent individuals, reduce the crossover probability.

[0026] Select the crossover method according to the set probability. The crossover methods include two-point crossover and order crossover. Combine the offspring individuals obtained through the crossover operation with the individuals in the original population to form a new population for subsequent genetic operations.

[0027] As a further improvement of this technical solution, in the mutation operation of the genetic algorithm, the order clustering and delivery unit adopts a combination of insertion mutation and inversion mutation. The specific steps are as follows:

[0028] Before starting the mutation operation, dynamically set the mutation probability according to the population diversity and individual fitness. For the gene encoding of the order delivery sequence, reflect the population diversity by measuring the distribution difference of different orders in the encoding sequence. Calculate the matching number of orders at the same position between each pair of individuals, and then calculate the average value as the diversity measurement value.

[0029] Combine the fitness function with the delivery distance, time cost, and area restriction penalty factor to calculate the fitness value of the order delivery plan corresponding to each individual. Dynamically adjust the mutation probability according to the above indicators. The formula is as follows:

[0030] ;

[0031] Among them is the adjusted mutation probability, is the initial mutation probability, is the adjustment coefficient, which is used to control the influence degree of diversity and fitness on the mutation probability, is the diversity metric value, and are respectively the minimum and maximum values of the population diversity, and are respectively the minimum and maximum values of the individual fitness, is the fitness function;

[0032] Based on the dynamically determined mutation probability , determine the mutation method according to the probability rule, put the individuals after the mutation operation back into the population, replace the original individuals, form a new population, and the new population carries the updated order delivery sequence information, and continuously optimize until the preset iteration termination condition is met, and finally output the optimal delivery route for each group of orders.

[0033] As a further improvement of the present technical solution, when the delivery execution feedback unit adjusts the delivery strategy according to the feedback evaluation data, the specific steps are as follows:

[0034] Input different delivery routes into the delivery execution feedback unit, calculate the reward value of each delivery route according to the satisfaction score feedback by customers, try different delivery routes in a random selection manner to obtain the preliminary reward information of each delivery route, then use the upper confidence bound algorithm to calculate the upper confidence bound value of each delivery route, and select the next execution delivery strategy according to the upper confidence bound value. As the system running time increases, continuously update the reward value and upper confidence bound value of each delivery route according to the new customer feedback satisfaction score, and dynamically adjust the delivery strategy.

[0035] Compared with the prior art, the beneficial effects of the present invention:

[0036] In a cross-border e-commerce logistics order management system based on big data, the order clustering and dispatching unit uses the density peak clustering algorithm to extract longitude and latitude data from the order geographical location information and convert it into plane coordinates, calculate the distance between orders, determine the cut-off distance through the mean, standard deviation and histogram of the distance, substitute it into the Gaussian kernel function to calculate the local density, and at the same time introduce the order urgency and customer loyalty to adjust the local density value, which can accurately identify the order aggregation area, reasonably allocate the non-clustering center orders, make the order classification more scientific, facilitate the subsequent dispatching plan, adopt the integer coding method of adding regional identification information, combine the geographical information data, real-time traffic data and dispatching area rules, use the fitness function containing the dispatching distance, time cost and regional restriction penalty factors to evaluate the advantages and disadvantages of the dispatching sequence plan. In the genetic operation, the roulette wheel selection method is used to select high-quality individuals, and through the partial mapping crossover and swap mutation operations, the optimal dispatching route is determined through multiple iterations of optimization, which improves the dispatching efficiency and reduces the transportation cost. Brief Description of the Drawings

[0037] Figure 1 It is the overall block diagram of the present invention.

[0038] The meanings of the various labels in the figure are as follows:

[0039] 1. Data acquisition and processing unit; 2. Order clustering and dispatching unit; 21. Local density calculation module; 22. Reference point selection module; 3. Dispatching execution and feedback unit. Detailed Embodiments

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0041] The present invention provides a cross-border e-commerce logistics order management system based on big data. Please refer to Figure 1 as shown, including a data acquisition and processing unit 1, an order clustering and dispatching unit 2, and a dispatching execution and feedback unit 3;

[0042] The data acquisition and processing unit 1 is used to collect cross-border e-commerce logistics order data from the cross-border e-commerce platform, and at the same time obtain the geographical information data and real-time traffic data of the region, and clean and preprocess the collected data.

[0043] When the data acquisition and processing unit 1 cleans the collected cross-border e-commerce logistics order data, for duplicate order data, the following duplicate removal method based on the hash algorithm is adopted:

[0044] In order to accurately and uniquely identify each order, key information is extracted from each order. The key information includes the order number, customer name, product details, and order placement time. A string is formed by combining the key information. By extracting and combining the key information into a string, these core data can be used to uniquely identify the order, avoiding incorrect judgments of duplicate orders caused by interference from other irrelevant information.

[0045] A hash function can convert input data of any length into a hash value of a fixed length. By calculating the hash value of the order key information string, orders can be quickly compared and searched. The hash function is used to calculate the combined string, obtaining the hash value of each order. Storing the hash values of all orders in a hash table can quickly locate possible duplicate orders, reducing the workload of subsequent comparisons.

[0046] Although orders with the same hash value are very likely to be duplicate orders, due to different inputs generating the same hash value, it is necessary to further compare other information of the orders to confirm whether they are truly duplicate orders. Traverse the hash table. If it is found that the hash values of different orders are the same, then compare the order amounts and delivery addresses of the orders to determine duplicate orders. By further comparing the order amounts and delivery addresses, misjudgments caused by hash conflicts can be effectively avoided, improving the accuracy of duplicate order judgment. For the determined duplicate orders, only one of them is retained, and the other duplicate orders are deleted.

[0047] The order clustering and delivery unit 2 uses the density peak clustering algorithm to cluster the preprocessed order data. By calculating the local density of each order data point, the density of orders around this point is evaluated. At the same time, the relative distance between each point and the points with higher density than it is determined, and the order aggregation area is identified by drawing a decision graph. Finally, the order data points that are not cluster centers are assigned to the corresponding clusters. For each group of orders after clustering, a genetic algorithm is used to optimize the delivery order. In terms of the coding method, integer coding with added regional identification information is used to define the position of each order in the delivery sequence and the region to which it belongs, and the crossover probability and mutation probability are dynamically adjusted according to population diversity and individual fitness. Then, using a fitness function that includes delivery distance, time cost, and regional restriction penalty factors, the advantages and disadvantages of each delivery order plan are evaluated. In the genetic operation link, the roulette wheel selection method is used to select high-quality individuals, and better delivery orders are generated through partial mapping crossover and swap mutation operations. After multiple iterations of optimization, the optimal delivery route is determined for each group of orders.

[0048] The order clustering and delivery unit 2 includes a local density calculation module 21. When the local density calculation module 21 calculates the local density of each order data point using the density peak clustering algorithm, the following method is adopted:

[0049] When applying the density peak clustering algorithm, it is necessary to calculate the distance between orders, extract the longitude and latitude data from the geographical location information of the orders, convert it into plane coordinates, and then use the weighted Euclidean distance formula to calculate the distance between orders. , which provides the basic data for subsequent clustering analysis.

[0050] The cut-off distance is an important parameter in the density peak clustering algorithm. It determines the neighborhood range considered when calculating the local density. Calculate the mean and standard deviation of the distances between all orders, draw a histogram of the distance data, observe the distribution pattern of the data, judge whether there is a clustering structure, and determine the cut-off distance according to the clustering structure. , which provides a key parameter for accurately calculating the local density.

[0051] The Gaussian kernel function is a commonly used kernel function. It can calculate the local density of each order data point according to the distance between orders and the cut-off distance, reflecting the density of orders in this area. Substitute the cut-off distance and the distance between orders into the Gaussian kernel function, then the local density ; where is the distance between order and order , is the cut-off distance, is the total number of orders, and the initial local density value of each order data point is obtained, which provides an important basis for subsequent clustering analysis.

[0052] The order urgency and customer loyalty are important attributes of orders. To a certain extent, they reflect the importance and value of orders. Set the order urgency of orders with a delivery time within 24 hours to 3, those within 24 - 48 hours to 2, and those over 48 hours to 1. Set the customer loyalty of customers with both the historical order quantity and consumption amount exceeding the preset threshold to 3, medium to 2, and low to 1. Quantified values of order urgency and customer loyalty are assigned to each order, providing data support for the adjustment of local density.

[0053] By introducing the order urgency and customer loyalty to adjust the initial local density, the local density can better reflect the importance and aggregation degree of orders at the business level. The adjusted local density ; where is the initial local density, and are adjustment coefficients, is the urgency of order , is the The customer loyalty corresponding to each order is quantified for the business attributes of the order, enabling these attributes to be reflected in the clustering analysis and providing a more comprehensive and reasonable basis for more accurate subsequent clustering analysis.

[0054] The order clustering and delivery unit 2 includes a reference point selection module 22. When determining the relative distance between each point and the points with higher density than it, if there are multiple points with higher density than the current point and equal distance, the following detailed method steps are used to select the reference point:

[0055] When determining the reference point, only considering the order amount may not comprehensively reflect the importance and priority of the order. In addition to considering the order amount, the order urgency, customer priority, and importance of commodity type are introduced as evaluation attributes. The customer priority is determined according to the consumption amount, and the importance of commodity type is determined according to the commodity value. Weights are assigned to each attribute according to the actual situation. The weight of the order amount is The weight of the order urgency is The weight of the customer priority is The weight of the importance of commodity type is This provides a clear attribute and weight system for subsequent comprehensive evaluation, making the selection of reference points more scientific.

[0056] In order to comprehensively compare different orders, it is necessary to convert attributes such as the order urgency, customer priority, and importance of commodity type into specific values. For each point with higher density than the current point and equal distance, calculate the comprehensive score. Let the order urgency be The customer priority be The importance of commodity type be The order amount be Then the comprehensive score This provides a specific comparison basis for selecting the reference point.

[0057] By comparing the comprehensive scores of all points with higher density than the current point and equal distance, the most advantageous order can be found as the reference point. Compare the comprehensive scores of all points with higher density than the current point and equal distance, select the point with the highest comprehensive score as the reference point to calculate the relative distance. If there are multiple points with the same and highest comprehensive score, compare the order amounts corresponding to the points, and select the point with the largest key attribute value as the reference point, determining the unique reference point, which provides an accurate basis for subsequent calculation of relative distance and clustering analysis.

[0058] When the order clustering and delivery unit 2 uses the genetic algorithm to optimize the delivery sequence, the calculation of the fitness function adopts the following steps:

[0059] The total delivery distance is one of the important indicators to measure the quality of a delivery plan. A shorter delivery distance usually means lower transportation costs and higher efficiency. Based on geographical information data and the delivery addresses of orders, the total delivery distance for each delivery sequence is calculated using a path planning algorithm , enabling objective comparison of different delivery plans in terms of distance, and helping to screen out plans with shorter distances.

[0060] It is incomplete to only consider the delivery distance and ignore the traffic congestion situation, because traffic conditions can significantly affect the delivery time and efficiency. Combining real-time traffic data and considering the traffic congestion situation in different time periods, the total time cost for each delivery sequence is calculated , obtaining the actual total time cost for each delivery sequence, which can more accurately evaluate the performance of the delivery plan in terms of time, and helping to select a plan with a shorter time consumption.

[0061] The delivery area rules for orders are to ensure the rationality and standardization of deliveries. Violating these rules may lead to additional costs or a decline in service quality. According to the delivery area rules of the orders, check whether each delivery sequence violates the area restrictions. If it does, determine the severity of the violation according to the violation threshold, and then determine the value of the penalty factor according to the severity , prompting the delivery plans generated by the genetic algorithm to be more in line with the delivery area rules, reducing potential risks and costs caused by violations, and improving the standardization and reliability of deliveries.

[0062] The fitness function is a key indicator in the genetic algorithm to evaluate the quality of an individual. It comprehensively considers the total delivery distance, the total time cost, and the penalty factor, and can comprehensively reflect the comprehensive performance of a delivery sequence. Finally, according to historical experience, set the weight coefficients, combine the values of the total delivery distance, the total time cost, and the penalty factor with the corresponding weight coefficients, and calculate the fitness function. Let the weight coefficient of the total delivery distance be , the weight coefficient of the total time cost be , and the weight coefficient of the penalty factor be , then the fitness function ; Calculate a comprehensive fitness value for each delivery sequence. The genetic algorithm can sort and select different delivery sequences according to this value, and finally find the delivery sequence with the optimal fitness value through continuous iterative evolution to achieve the optimization of the delivery plan.

[0063] In the crossover operation of the genetic algorithm for the order clustering delivery unit 2, a method combining two-point crossover and order crossover is adopted. The specific steps are as follows:

[0064] Before the crossover operation of the genetic algorithm, it is necessary to understand the diversity of the current population and the fitness of each individual. Before performing the crossover operation, for the gene encoding of each individual, calculate its Hamming distance from the gene encodings of other individuals, and then find the average value of the Hamming distances between all individuals. Measure the population diversity through the average value of the Hamming distances, and then calculate the fitness value of each individual according to the fitness function.

[0065] Let the individuals 、 have gene encodings and respectively. The Hamming distance is . Let there be individuals in the population. Then the population diversity is . Let the fitness function be . The fitness value of individual is . An accurate population diversity index and individual fitness values are obtained, laying a foundation for subsequent adjustment of the crossover probability and selection of the crossover method.

[0066] The crossover probability is an important parameter in the genetic algorithm, which affects the search ability and convergence speed of the algorithm. When the population diversity is lower than the population diversity threshold , it indicates that the individuals in the population have high similarity, so the crossover probability is increased; when the individual fitness exceeds the threshold , in order to retain excellent individuals, the crossover probability is reduced.

[0067] Let the initial crossover probability be , and the adjusted crossover probability be . When , , where is a positive adjustment amount. When , , where is a positive adjustment amount. is the fitness function of individual . The crossover probability can be reasonably adjusted according to the population state and individual quality, which helps the algorithm achieve a balance between global search and local optimization and improves the possibility of finding the optimal solution.

[0068] Select the crossover method according to the set probability. The crossover methods include two-point crossover and order crossover. Combine the offspring individuals obtained through the crossover operation with the individuals in the original population to form a new population for subsequent genetic operations. By reasonably selecting the crossover method and updating the population, the diversity of the population and the effectiveness of the search are increased, providing a better foundation for the subsequent iteration of the genetic algorithm and helping the algorithm converge to the optimal solution faster.

[0069] In the mutation operation of the genetic algorithm for the order clustering and delivery unit 2, the combination of insertion mutation and reverse order mutation is adopted, and the specific steps are as follows:

[0070] In the mutation operation of the genetic algorithm, the mutation probability plays a key role. Before starting the mutation operation, the mutation probability is dynamically set according to the population diversity and individual fitness. For the gene coding of the order delivery sequence, the population diversity is reflected by measuring the distribution difference of different orders in the coding sequence. Calculate the matching number of orders at the same position between each pair of individuals, and then take the average value as the diversity measurement value;

[0071] Combining the fitness function with the delivery distance, time cost, and regional restriction penalty factors, calculate the fitness value of the order delivery plan corresponding to each individual, and dynamically adjust the mutation probability according to the above indicators. The formula is as follows:

[0072] ;

[0073] Among them is the adjusted mutation probability, is the initial mutation probability, is the adjustment coefficient, which is used to control the influence degree of diversity and fitness on the mutation probability, is the diversity measurement value, and are respectively the minimum and maximum values of the population diversity, and are respectively the minimum and maximum values of the individual fitness, is the fitness function.

[0074] Based on the dynamically determined mutation probability , determine the mutation method according to the probability rule, put the individuals after the mutation operation back into the population, replace the original individuals, and form a new population. The new population carries the updated order delivery sequence information and continues to optimize until the preset iteration termination condition is met. Finally, the optimal delivery route for each group of orders is output, effectively improving the delivery efficiency and optimization effect.

[0075] The delivery execution feedback unit 3 sends the optimized delivery route to the delivery vehicle, collects the customer feedback evaluation data after the delivery is completed, and adjusts the delivery strategy according to the feedback evaluation data.

[0076] When the delivery execution feedback unit 3 adjusts the delivery strategy according to the feedback evaluation data, the specific steps are as follows:

[0077] In order to evaluate the advantages and disadvantages of different delivery routes, it is necessary to quantify their performance based on customer feedback. Input different delivery routes into the delivery execution feedback unit 3, and calculate the reward value of each delivery route according to the satisfaction score of customer feedback , which is convenient for comparison between different routes and initially screens out the delivery routes with better or worse performance.

[0078] In the initial stage of the system, due to limited understanding of each delivery route, randomly selecting different delivery routes can widely collect feedback under different circumstances, quickly accumulate performance data of each route in various scenarios, lay a foundation for subsequent precise decision-making, and avoid getting stuck in local optimal strategies prematurely. By randomly selecting different delivery routes, preliminary reward information for each delivery route is obtained, and preliminary reward information of each delivery route in different scenarios is accumulated, providing diverse data samples for subsequent more refined strategy adjustment.

[0079] The Upper Confidence Bound (UCB) algorithm can consider the existing reward information while taking into account the exploration of unknown situations. Use the UCB algorithm to calculate the upper confidence bound value of each delivery route. Let the average reward value of the th delivery route be ; where is the reward value obtained in the th execution of the th delivery route, the number of delivery times is , according to the UCB algorithm, the calculation formula of the upper confidence bound value is ; where is the total number of system runs, is the exploration coefficient, which is used to balance exploration and exploitation. Calculate the upper confidence bound value for each delivery route. When selecting a delivery strategy, this value can be used to more reasonably weigh whether to continue using the current better route or try a new route.

[0080] Select the delivery strategy for the next execution according to the upper confidence bound value. That is, for each delivery task, select the delivery route with the largest value for delivery. As the system running time increases, update the reward value and upper confidence bound value of each delivery route according to the new customer feedback satisfaction score. When receiving the satisfaction score corresponding to a delivery route, update the average reward value and recalculate the upper confidence bound value at the same time. The delivery strategy can be dynamically adjusted according to customer feedback, continuously optimize the delivery route selection, improve the delivery efficiency, enhance customer satisfaction with the delivery service, and ensure the high-quality operation of cross-border e-commerce logistics distribution.

[0081] In the present invention, the data acquisition and processing unit 1 collects order, geographical information and real-time traffic data from cross-border e-commerce platforms and performs cleaning and preprocessing. The duplicate orders are removed by using the hash algorithm. The order clustering and delivery unit 2 clusters the orders by using the density peak clustering algorithm, optimizes the clustering effect by combining the order urgency and customer loyalty, optimizes the delivery sequence by using the genetic algorithm, dynamically adjusts the crossover and mutation probabilities, calculates the fitness function considering the delivery distance, time cost and regional restriction penalty factors. The delivery execution and feedback unit 3 sends the optimized route to the delivery vehicle, collects the customer feedback evaluation data, and dynamically adjusts the delivery strategy by means of the upper confidence bound algorithm, so as to improve the delivery efficiency and customer satisfaction.

[0082] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the descriptions in the specification are only preferred examples of the present invention, which are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A cross-border e-commerce logistics order management system based on big data, characterized in that, It includes a data acquisition and processing unit (1), an order clustering and delivery unit (2), and a delivery execution and feedback unit (3); The data acquisition and processing unit (1) is used to collect cross-border e-commerce logistics order data from cross-border e-commerce platforms, and at the same time obtain the regional geographical information data and real-time traffic data, and clean and preprocess the collected data; The order clustering and delivery unit (2) uses the density peak clustering algorithm to cluster the preprocessed order data. By calculating the local density of each order data point, it evaluates the density of orders around this point, and at the same time determines the relative distance between each point and the points with higher density than it, and identifies the order aggregation area by drawing a decision graph. Finally, the order data points that are not cluster centers are assigned to the corresponding clusters. For each group of orders after clustering, a genetic algorithm is used to optimize the delivery order. In terms of coding method, integer coding with added regional identification information is used to define the position of each order in the delivery sequence and the region to which it belongs, and the crossover probability and mutation probability are dynamically adjusted according to population diversity and individual fitness. Then, using a fitness function that includes delivery distance, time cost, and regional restriction penalty factors, the advantages and disadvantages of each delivery order plan are evaluated. In the genetic operation link, the roulette wheel selection method is used to select high-quality individuals, and a better delivery order is generated through partial mapping crossover and swap mutation operations. After multiple iterations of optimization, the optimal delivery route is determined for each group of orders; The delivery execution and feedback unit (3) sends the optimized delivery route to the delivery vehicle, collects customer feedback evaluation data after delivery, and adjusts the delivery strategy according to the feedback evaluation data.

2. The cross-border e-commerce logistics order management system based on big data according to claim 1, wherein: When the data acquisition and processing unit (1) cleans the collected cross-border e-commerce logistics order data, for duplicate order data, the following deduplication method based on the hash algorithm is adopted: Extract the key information from each order. The key information includes the order number, customer name, commodity details, and order time, and form a string with the key information; Use a hash function to calculate the combined string to obtain the hash value of each order, store the hash values of all orders in a hash table, traverse the hash table, if it is found that the hash values of different orders are the same, then compare the order amounts and delivery addresses of the orders to determine duplicate orders. For the determined duplicate orders, only keep one of them and delete the other duplicate orders.

3. The cross-border e-commerce logistics order management system based on big data according to claim 2, characterized in that: The order clustering and delivery unit (2) includes a local density calculation module (21). When the local density calculation module (21) calculates the local density of each order data point using the density peak clustering algorithm, the following method is adopted: Extract the longitude and latitude data from the geographical location information of the order, convert it into a plane coordinate, and then use the weighted Euclidean distance formula to calculate the distance between orders. Calculate the mean and standard deviation of the distances between all orders, draw a histogram of the distance data, observe the distribution form of the data, judge whether there is a clustering structure, and determine the cut-off distance according to the clustering structure; Substitute the truncated distance and the distance between orders into the Gaussian kernel function to calculate the local density of each order data point, where the local density is used to reflect the density of orders in this area. Then, introduce the order urgency and customer loyalty; Set the order urgency of orders with a delivery time within 24 hours to 3, those within 24 - 48 hours to 2, and those over 48 hours to 1. Set the customer loyalty of customers with both the historical order quantity and consumption amount exceeding the preset threshold to 3, medium to 2, and low to 1. Adjust the local density value according to the order urgency and customer loyalty to obtain the adjusted local density.

4. The cross-border e-commerce logistics order management system based on big data according to claim 3, wherein: The order clustering and delivery unit (2) includes a reference point selection module (22). When the reference point selection module (22) determines the relative distance between each point and the points with higher density than it, if there are multiple points with higher density than the current point and equal distance, the following detailed method steps are used to select the reference point: In addition to considering the order amount, introduce the order urgency, customer priority, and importance of commodity type as evaluation attributes. The customer priority is determined according to the consumption amount, and the importance of commodity type is determined according to the commodity value. Assign weights to each attribute according to the actual situation; Obtain the actual amount value of the order from the data acquisition and processing unit (1). According to the defined correspondence between the level and the value, convert the order urgency, customer priority, and importance of commodity type into corresponding values. For each point with higher density than the current point and equal distance, calculate the comprehensive score; Compare the comprehensive scores of all points with higher density than the current point and equal distance, select the point with the highest comprehensive score as the reference point to calculate the relative distance. If there are multiple points with the same and highest comprehensive score, compare the order amounts corresponding to the points and select the point with the largest key attribute value as the reference point.

5. The cross-border e-commerce logistics order management system based on big data according to claim 4, characterized in that: When the order clustering and delivery unit (2) uses the genetic algorithm to optimize the delivery order, the calculation of the fitness function adopts the following steps: According to the geographical information data and the delivery addresses of the orders, use the path planning algorithm to calculate the total delivery distance of each delivery order. Combine the real-time traffic data and consider the traffic congestion situation in different time periods to calculate the total time cost of each delivery order; According to the delivery area rules of the orders, check whether each delivery order violates the area limit. If it violates, determine the severity of the violation according to the violation threshold, then determine the value of the penalty factor according to the severity, and finally set the weight coefficient according to historical experience. Combine the total delivery distance, total time cost, the value of the penalty factor, and the corresponding weight coefficient to calculate the fitness function.

6. The cross-border e-commerce logistics order management system based on big data according to claim 5, wherein: In the crossover operation of the genetic algorithm in the order clustering and delivery unit (2), a method combining two-point crossover and order crossover is adopted. The specific steps are as follows: Before the crossover operation, for the gene encoding of each individual, calculate its Hamming distance from the gene encoding of other individuals, then calculate the average value of the Hamming distances between all individuals. Measure the population diversity through the average value of the Hamming distances, and then calculate the fitness value of each individual according to the fitness function; When the population diversity is lower than the population diversity threshold, it indicates that the individuals in the population have high similarity, so the crossover probability is increased. When the individual fitness exceeds the threshold, in order to retain excellent individuals, the crossover probability is decreased; Select the crossover method according to the set probability. The crossover methods include two-point crossover and order crossover. The offspring individuals obtained through the crossover operation and the individuals in the original population are combined to form a new population for subsequent genetic operations.

7. The cross-border e-commerce logistics order management system based on big data according to claim 6, characterized in that: In the mutation operation of the genetic algorithm, the order clustering and delivery unit (2) adopts a combination of insertion mutation and inversion mutation. The specific steps are as follows: Before starting the mutation operation, the mutation probability is dynamically set based on the population diversity and individual fitness. For the gene coding of the order delivery sequence, the population diversity is reflected by measuring the distribution difference of different orders in the coding sequence. Calculate the matching number of orders at the same position between each pair of individuals, and then calculate the average value as the diversity metric value; Combining the fitness function with the delivery distance, time cost, and regional restriction penalty factors, calculate the fitness value of the order delivery plan corresponding to each individual, and dynamically adjust the mutation probability according to the above indicators. The formula is as follows: ； Among them is the adjusted mutation probability, is the initial mutation probability, is the adjustment coefficient, which is used to control the influence degree of diversity and fitness on the mutation probability, is the diversity metric value, and are respectively the minimum and maximum values of the population diversity, and are respectively the minimum and maximum values of the individual fitness, is the fitness function; Based on a dynamically determined mutation probability , determine the mutation method according to the probability rule, put the individual after the mutation operation back into the population, replace the original individual, form a new population, and the new population carries the updated order delivery sequence information, continuously optimize until the preset iteration termination condition is met, and finally output the optimal delivery route for each group of orders.

8. The cross-border e-commerce logistics order management system based on big data according to claim 7, characterized in that: When the delivery execution feedback unit (3) adjusts the delivery strategy according to the feedback evaluation data, the specific steps are as follows: Input different delivery routes into the delivery execution feedback unit (3). Calculate the reward value of each delivery route according to the satisfaction score feedback by the customers. Try different delivery routes in a random selection manner to obtain the preliminary reward information of each delivery route, and then use the upper confidence bound algorithm to calculate the upper confidence bound value of each delivery route. Select the next delivery strategy to be executed according to the upper confidence bound value. As the system running time increases, continuously update the reward value and upper confidence bound value of each delivery route according to the new customer feedback satisfaction score, and dynamically adjust the delivery strategy.

Images