A method and system for optimizing the management of express order redistribution

Abstract

The invention belongs to the technical field of data processing, and discloses an express order reallocation optimization management method and system, comprising: collecting real-time delivery data of express orders; constructing an ARIMA model and a time-space joint model, processing the real-time delivery data of express orders, and obtaining a comprehensive feature data set of order delivery; training an abnormal order evaluation model according to the comprehensive feature data set of order delivery, and predicting the abnormal coefficient of the express order; judging whether it is necessary to reallocate the express order according to the abnormal coefficient of the express order; if it is necessary to reallocate the express order, using an intelligent reallocation algorithm to obtain an optimal reallocation path through a branch and bound method under multi-objective constraints; notifying a delivery person to execute the optimal reallocation path through an order reallocation management terminal, and completing the delivery of the reallocated express order; dynamically adjusting the delivery plan through intelligent reallocation, and improving the flexibility and anti-interference ability of logistics delivery.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to a method and system for optimizing the management of express order redistribution. Background Art

[0002] The patent application publication number CN115731007A discloses a reinforcement learning real-time order delivery method for regional collaboration, which constructs spatial geographic information based on regional fence processing; encodes spatial location value information to obtain the corresponding feature vector matrix; combines the order allocation matrix and the order acceptance rate factor to calculate the efficiency index in a single region; calculates the total efficiency index of the sub-region according to the traffic weight distribution, and determines the set of factor constraints; proposes a constrained optimization equation group based on the optimization objective function and constraints, and transforms the solution of the constrained optimization problem into an unconstrained optimization problem through Lagrange multipliers; uses the total efficiency index as the reward and punishment value of deep reinforcement learning; optimizes parameters through environmental interaction, and then obtains the trained real-time dispatch model. This method can significantly improve the real-time response speed of the express dispatch system, and help solve the problem of maximizing the efficiency of regional collaborative intelligent dispatch.

[0003] The existing express delivery order distribution management method and system have the following defects:

[0004] The inability to fully explore and capture the complex relationship between time and space dimensions leads to the neglect of important information such as the temporal and spatial distribution of orders, peak demand periods, and clustering patterns in specific geographic locations. The weak understanding of the correlation between orders leads to poor optimization of the model. The existing clustering methods cannot correctly identify the actual temporal and spatial distribution characteristics during the clustering of order data, leading to the neglect of some potential order clustering patterns.

[0005] The selection of delivery plans is prone to fall into local optimality, which cannot effectively reduce delivery costs, time and improve customer satisfaction. Ignoring multi-dimensional optimization may lead to waste of transportation resources or unbalanced workload, which will ultimately affect the overall scheduling efficiency. When faced with large-scale order and delivery personnel data, the computational complexity will increase significantly. When the order and delivery personnel's situation changes, the system cannot respond quickly.

[0006] Each search for a path optimization solution needs to be carried out in a wide solution space, resulting in a slow convergence process and a waste of a lot of computing resources; the search process lacks flexibility, which limits the system's ability to explore different paths, resulting in a decrease in the algorithm's global search capability, affecting the accuracy and diversity of path planning.

[0007] In view of this, the present invention proposes an express order reallocation optimization management method and system to solve the above-mentioned problems. Summary of the invention

[0008] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for optimizing the reallocation of express orders, comprising:

[0009] S1. Collect real-time delivery data of express orders;

[0010] S2, construct ARIMA model and time-space joint model, process the real-time delivery data of express orders, and obtain the comprehensive feature data set of order delivery;

[0011] S3. Train an abnormal order evaluation model based on the comprehensive feature data set of order delivery, and predict the abnormal coefficient of the express order; based on the abnormal coefficient of the express order, determine whether the express order needs to be reallocated;

[0012] S4. If express order redistribution is required, the intelligent redistribution algorithm is used to obtain the optimal redistribution path through the branch and bound method under multi-objective constraints;

[0013] S5. Through the order reallocation management terminal, notify the delivery personnel to execute the optimal reallocation path and complete the delivery of the reallocated express order.

[0014] Furthermore, the real-time delivery data of express orders includes order parameter data, delivery status data and traffic environment data; the order parameter data includes order quantity, order number, recipient information, order item weight, order item volume, delivery address and delivery time; the delivery status data includes delivery progress, estimated delivery time, delivery personnel number, delivery personnel location, workload, transport vehicle model and delivery route; the traffic environment data includes traffic flow, road conditions and weather conditions.

[0015] Furthermore, the method for processing the real-time delivery data of the express order includes:

[0016] The order parameter data, delivery status data and traffic environment data are cleaned, and outliers are identified and eliminated through the box plot method; the order parameter data, delivery status data and traffic environment data after outliers are eliminated are normalized by standard deviation, and converted into a standard state distribution with a mean of 0 and a standard deviation of 1, eliminating the dimensional influence between the data, and obtaining the normalized order parameter data, delivery status data and traffic environment data;

[0017] The ARIMA model is used to perform time series forecasting on the normalized delivery status data and traffic environment data to extract dynamic features, thereby obtaining a delivery status feature data set and a traffic environment feature data set. A time-space joint model is constructed to perform spatiotemporal correlation analysis on the normalized order parameter data to obtain a spatiotemporal density feature data set.

[0018] Furthermore, the method for acquiring the spatiotemporal density feature data set includes:

[0019] S41. Discretize the normalized order parameter data into data points in the order time-space sequence; each data point in the order time-space sequence is The feature vector of spatiotemporal attributes; the order spatiotemporal sequence is ;in, The index of the timestamp; for The spatiotemporal sequence of orders within time; is the total number of data point categories in the order spatiotemporal sequence; is the index of the data point category in the order spatiotemporal sequence, ; is the order space-time sequence The data points; is the order space-time sequence The data points;

[0020] S42. Define a spatiotemporal joint distance formula, and use the spatiotemporal joint distance formula to measure the spatiotemporal joint distance between any two data points in the order spatiotemporal sequence; the spatiotemporal joint distance formula is: ;in, For data points and data points The space-time joint distance between them; is the weight coefficient of spatial distance; For data points and data points The spatial distance between them is obtained by the Euclidean distance formula; is the weight coefficient of time distance; For data points and data points The time distance between ;in, is the index of the standardized time feature dimension; For data points No. A standardized time feature; For data points No. A standardized time feature; the time feature contained in the data point is mapped to the unit circle through the periodic function sin, and then the standardized time feature is obtained ;in, is the time period; is the ratio of pi; For data points The temporal features included;

[0021] S43, using K-Means clustering method, based on the time-space joint distance formula to cluster the order time-space sequence; randomly select 'Initial cluster center points ; For each data point in the order spatiotemporal sequence, calculate the spatiotemporal joint distance to all cluster center points, assign each data point to the nearest cluster center, and then obtain the initial cluster set;

[0022] S44, for each cluster in the initial cluster set, respectively calculate the mean of all data points in the cluster; use the obtained mean of all data points in the cluster as a new cluster center point, and respectively update the cluster center point of each cluster;

[0023] S45, repeating steps S43-S44 until the preset maximum number of iterations is reached, thereby obtaining a final cluster set; calculating the spatiotemporal density of each cluster in the final cluster set by a density extraction formula, thereby obtaining a spatiotemporal density feature data set; the density extraction formula is: ;in, For the The set of orders in the clusters; For the Weight of items in the order; For the Volume of items in an order; For the Delivery time of each order; For the The spatial area covered by the orders in each cluster; For the The length of the time window in a cluster.

[0024] Furthermore, the method for acquiring the order delivery comprehensive feature data set includes:

[0025] The delivery status feature dataset, traffic environment feature dataset and spatiotemporal density feature dataset are integrated through a weighted formula to obtain a comprehensive feature dataset of order delivery. The delivery status feature dataset is recorded as , the traffic environment feature dataset is recorded as , the spatiotemporal density feature dataset is recorded as ;

[0026] The weighted formula is: ;in, Comprehensive feature dataset for order delivery; is the weight coefficient of the delivery status feature data set; is the weight coefficient of the traffic environment characteristic dataset; is the weight coefficient of the spatiotemporal density feature dataset.

[0027] Furthermore, the training method of the abnormal order evaluation model includes:

[0028] The data set is divided into a training set, a validation set, and a test set, and an abnormal order evaluation model is constructed. The abnormal order evaluation model includes input data and output labels. The input data of the model is a historical order delivery comprehensive feature data set, and the output label of the model is the abnormal coefficient of the express order. The hyperparameters of the model are selected, and the hyperparameters include the number of trees and the maximum depth of sample segmentation. During the training process, each tree randomly selects sample features and sample subsets, generates a tree using random cutting, and finally calculates the abnormal coefficient. The abnormal order evaluation model is an isolation forest model.

[0029] The Isolation Forest Model does not require a loss function. The training process detects outliers by selecting split points and building trees. The model is trained on the data set, and the hyperparameters of the model are tuned using the validation set. The cross-validation method is used to find the optimal hyperparameter combination.

[0030] Use the test data set to evaluate the model, and evaluate the model performance by observing the performance indicators of the model on the prediction task. Stop testing when the model performance no longer improves, and obtain the trained abnormal order evaluation model; use the trained abnormal order evaluation model to predict the current order delivery comprehensive feature data set to obtain the abnormal coefficient of the express order.

[0031] Furthermore, the method for judging whether to reallocate the express order according to the abnormal coefficient of the express order includes:

[0032] Preset an abnormal coefficient threshold of the express order, and compare the abnormal coefficient of the predicted express order with the preset abnormal coefficient threshold of the express order;

[0033] If the predicted abnormal coefficient of the express order is less than the preset abnormal coefficient threshold of the express order, it is determined that there is no need to reallocate the express order;

[0034] If the predicted abnormal coefficient of the express order is greater than or equal to the preset abnormal coefficient threshold of the express order, it is determined that the express order needs to be reallocated.

[0035] Furthermore, the method for obtaining the optimal reallocation path includes:

[0036] S81. Define the objective function and set the total Express orders, including: For the Express orders; Index of the express order; is the total number of express orders; Delivery personnel, including For the A delivery person; is the index of the delivery person, ; is the total number of delivery personnel; Express orders are assigned to delivery personnel; optimize the objective function: ;in, is the total optimization target value; is the total delivery cost; is the total delivery time; For customer satisfaction; is the weight coefficient of the total delivery cost; is the weight coefficient of the total delivery time; is the weight coefficient of customer satisfaction;

[0037] S82. Use the branch and bound method to build a decision tree to enumerate all allocation schemes, use upper and lower bound optimization and pruning techniques to reduce the amount of calculation and find the global optimal solution; define the root node Indicates the initial state where all orders are unassigned, and calculates the initial lower bound and the upper bound of the current optimal solution ; Each node Indicates the state of the partial allocation plan. Generates child nodes from the current node. Each child node represents a courier order. Assign to a delivery person ;

[0038] S83. Calculate and update the current partial allocation scheme status using the lower bound calculation formula ; The lower bound calculation formula is: ;in, Assign a solution status to the current section The lower bound of is the total cost of currently allocated orders; For orders Assign to delivery person Costs; For unassigned orders;

[0039] S84. If the current partial allocation scheme status The lower bound Greater than or equal to the upper bound of the current optimal solution , then prune; if the total cost of the currently allocated orders Greater than the upper bound of the current optimal solution , then pruning is also performed; traverse the complete allocation plan, if the total cost of the currently allocated order Less than or equal to the upper bound of the current optimal solution , then no pruning is performed and the upper bound of the current optimal solution is updated ; When all nodes are processed, the upper bound of the current optimal solution This is the global optimal solution;

[0040] S85. After finding the global optimal solution through the branch and bound method, extract the express orders that each deliveryman is responsible for from the distribution plan corresponding to the optimal solution; each time an order is assigned to a deliveryman, calculate the cost and time of the current distribution state, and update the total cost, total time and customer satisfaction of the currently assigned orders; when traversing all distribution plans, the final optimal solution will be represented as a complete order distribution plan; the optimal reallocation path is the path corresponding to the complete order distribution plan.

[0041] Furthermore, the upper bound of the current optimal solution is The definition methods include:

[0042] S91. Use the ant colony algorithm to simulate the foraging behavior of ants and provide the upper bound of the optimal solution for the branch and bound method ; Define the number of ants as , initialize the pheromone matrix as ;

[0043] S92. Each ant starts with a random order, selects a path, and selects the delivery person for the next order by using the selection probability formula; the selection probability formula is: ;in, Choosing a path for the ants The probability of For path Pheromones on; is the weight coefficient of pheromone; for heuristic information; is the weight coefficient of heuristic information; For slave nodes The next node to choose from; For slave nodes The set of all next nodes that can be selected;

[0044] S93. Dynamically adjust the weight coefficient of the pheromone through the weight coefficient adjustment formula. The weight coefficient adjustment formula is: ;in, is the total number of nodes; For control Constant factor of maximum value; For control The constant factor of the rate of change with the number of nodes;

[0045] S94. After each ant completes the path construction, the path cost is calculated; the path cost is the total optimization target value. , taking the cost of each path as the quality index of the path, and updating the pheromone using the pheromone update formula according to the quality index of the path; the pheromone update formula is: ;in, For the updated pheromone; is the volatility coefficient; is the path gain;

[0046] S95, continuously update pheromones until the maximum number of iterations preset by the ant colony algorithm is reached. In each iteration, record and update the upper bound of the current optimal solution. .

[0047] An express order reallocation optimization management system, comprising:

[0048] Data collection module, used to collect real-time delivery data of express orders;

[0049] The data processing module is used to build the ARIMA model and the time-space joint model, process the real-time delivery data of express orders, and obtain the comprehensive feature data set of order delivery;

[0050] The abnormality monitoring module is used to train an abnormal order evaluation model based on the comprehensive feature data set of order delivery, and predict the abnormal coefficient of the express order; according to the abnormal coefficient of the express order, it is determined whether the express order needs to be reallocated;

[0051] Redistribution optimization module: if express order redistribution is required, the intelligent redistribution algorithm is used to obtain the optimal redistribution path through the branch and bound method under multi-objective constraints;

[0052] The notification management module notifies the delivery personnel to execute the optimal reallocation path and complete the delivery of the reallocated express orders through the order reallocation management terminal.

[0053] Technical effects and advantages of the express order reallocation optimization management method and system of the present invention:

[0054] Converting order parameter data into a structured spatiotemporal sequence can conduct a more detailed analysis of data in the time and space dimensions, revealing the order distribution rules and spatiotemporal patterns; when involving geographic information and time dimensions, the spatiotemporal joint distance can effectively capture the spatiotemporal relationship between orders; using the K-Means clustering algorithm, clustering the order spatiotemporal sequence by the spatiotemporal joint distance can extract the distribution characteristics of orders in different regions and time periods, and provide clustering information for subsequent analysis and optimization;

[0055] Through the branch-and-bound method, a global search is performed to ensure that the optimal delivery person-order allocation plan is found, avoiding the problem of local optimal solutions, so that the balance between overall delivery cost, time and customer satisfaction is optimized; by defining a comprehensive objective function, multi-dimensional optimization of the objective function becomes possible, and the dispatch of delivery persons does not simply consider the optimization of one dimension, but comprehensively considers the optimal balance of cost, time and satisfaction; each time the optimal solution is updated, the decision on whether to continue in-depth exploration is made in combination with the upper limit of the currently known optimal solution, which not only ensures the discovery of the global optimal solution, but also avoids unnecessary calculations; each time an order is allocated, the allocated cost and time are dynamically calculated to ensure that the task burden of each delivery person is reasonable and avoid excessive load or waste of resources; the path and task of each delivery person are adjusted according to real-time order information, ensuring more flexible and efficient path planning;

[0056] By simulating the foraging behavior of ants, the upper bound of the global optimal solution is provided for the branch and bound method. By simulating the behavior of ants in the entire search space and accumulating pheromones, ants can find a better path, thereby providing an effective optimization guide for the branch and bound method, which helps to accelerate the convergence process of the algorithm. As the number of nodes increases, the influence of pheromones gradually weakens, and ants rely more on heuristic information, which can not only avoid premature convergence, but also enhance the diversity of the search process and avoid falling into the local optimal solution. By dynamically adjusting the weight coefficient of pheromones, ants can make full use of historical experience and appropriately rely on the superiority of the current path when selecting paths. By gradually updating the upper bound of the optimal solution in each round of iteration, the ant colony algorithm can limit unnecessary searches and avoid excessive exploration of non-optimal paths in the branch and bound method. Each time the ants build a path, they will evaluate the path cost and adjust the path quality through the pheromone update formula. By continuously adjusting the pheromone, the ants can optimize the path more accurately during the search process. In each iteration, the algorithm makes a selection based on the path quality, so that the optimal path is gradually found after multiple iterations, which improves the accuracy of path planning and avoids meaningless path search. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a flow chart of an express order reallocation optimization management method of the present invention;

[0058] Figure 2 It is a structural schematic diagram of an express order reallocation optimization management system of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Example 1

[0061] See also Figure 1 and Figure 2 As shown, this embodiment provides an express order reallocation optimization management method, including:

[0062] S1. Collect real-time delivery data of express orders;

[0063] S2, construct ARIMA model and time-space joint model, process the real-time delivery data of express orders, and obtain the comprehensive feature data set of order delivery;

[0064] S3. Train an abnormal order evaluation model based on the comprehensive feature data set of order delivery, and predict the abnormal coefficient of the express order; based on the abnormal coefficient of the express order, determine whether the express order needs to be reallocated;

[0065] S4. If express order redistribution is required, the intelligent redistribution algorithm is used to obtain the optimal redistribution path through the branch and bound method under multi-objective constraints;

[0066] S5. Through the order reallocation management terminal, notify the delivery personnel to execute the optimal reallocation path and complete the delivery of the reallocated express order.

[0067] Real-time delivery data of express orders include order parameter data, delivery status data and traffic environment data; order parameter data include order quantity, order number, recipient information, order product weight, order product volume, delivery address and delivery time; delivery status data include delivery progress, estimated delivery time, delivery personnel number, delivery personnel location, workload, transportation vehicle model and delivery route; traffic environment data include traffic flow, road conditions and weather conditions;

[0068] Collect order parameter data and delivery status data through intelligent business terminals; obtain real-time traffic environment data in each business area of ​​express delivery orders by integrating third-party traffic data sources (such as Amap, Baidu Map, etc.);

[0069] The order quantity refers to the number of parcels included in the same delivery task, which determines the complexity and resource requirements of delivery. For example, a customer may purchase multiple items, and each item may need to be delivered separately or in combination. Appropriate vehicles and personnel can be dispatched according to the order quantity to avoid resource waste or overload. The order number is the core identification mark of all delivery and logistics activities, which helps to track, manage and record the status of each order in the system to ensure the consistency and traceability of order information. The delivery time refers to the time requirement by which an order must be delivered, such as "next-day delivery" or "pre-scheduled delivery time", which is one of the most important performance indicators of express delivery companies. Collecting delivery time helps make scheduling decisions, and can determine which orders need to be delivered first based on the delivery time to ensure that the delivery time is met. Real-time monitoring of delivery progress can track the status of orders, promptly identify problems (such as the deliveryman not arriving on time) and make corresponding adjustments, while helping customers understand the delivery status of their orders in real time. Estimated delivery time The arrival time is the specific time when the order is expected to be delivered, calculated based on the current delivery progress and traffic environment data; the workload is the current workload undertaken by the delivery person, such as the number of orders delivered, the weight of the ordered goods, etc., which is used to determine whether the delivery person is overloaded; if a delivery person's workload is too heavy, some tasks need to be assigned to other couriers to ensure maximum work efficiency and avoid delivery fatigue; traffic flow refers to the number of vehicles passing through a certain road in a period of time, reflecting the degree of road congestion, and is used to avoid congested sections when planning the route for express order delivery, reduce delivery time and avoid delivery delays caused by traffic jams; road conditions refer to the road's capacity, including whether there are traffic accidents, road repairs, or closures; weather conditions are crucial for delivery, especially in severe weather conditions, which may lead to restricted road access and safety issues for delivery personnel; through weather data, delivery plans can be adjusted in a timely manner, giving priority to unaffected areas or adjusting delivery methods.

[0070] The method for processing the real-time delivery data of express orders includes:

[0071] The order parameter data, delivery status data and traffic environment data are cleaned, and outliers are identified and eliminated through the box plot method; the order parameter data, delivery status data and traffic environment data after outliers are eliminated are normalized by standard deviation, and converted into a standard state distribution with a mean of 0 and a standard deviation of 1, eliminating the dimensional influence between the data, and obtaining the normalized order parameter data, delivery status data and traffic environment data;

[0072] The ARIMA model is used to perform time series forecasting on the normalized delivery status data and traffic environment data to extract dynamic features, thereby obtaining a delivery status feature data set and a traffic environment feature data set. A time-space joint model is constructed to perform spatiotemporal correlation analysis on the normalized order parameter data to obtain a spatiotemporal density feature data set. Spatiotemporal correlation analysis can capture the characteristics of both time and space dimensions and adapt to complex spatiotemporal dynamic scenarios. Time characteristics (such as the temporal changes in the number of orders) are combined with spatial characteristics (such as the regional distribution of order delivery addresses) to form a multi-dimensional data input, and a comprehensive exploration of spatiotemporal correlation is achieved by simultaneously optimizing the clustering objective functions of time and space.

[0073] Methods for obtaining spatiotemporal density feature datasets include:

[0074] S41. Discretize the normalized order parameter data into data points in the order time-space sequence; each data point in the order time-space sequence is The feature vector of spatiotemporal attributes; the order spatiotemporal sequence is ;in, The index of the timestamp; for The spatiotemporal sequence of orders within time; is the total number of data point categories in the order spatiotemporal sequence; is the index of the data point category in the order spatiotemporal sequence, ; is the order space-time sequence The data points; is the order space-time sequence The data points;

[0075] S42. Define a spatiotemporal joint distance formula, and use the spatiotemporal joint distance formula to measure the spatiotemporal joint distance between any two data points in the order spatiotemporal sequence for spatiotemporal analysis and clustering; the spatiotemporal joint distance formula is: ;in, For data points and data points The space-time joint distance between them; is the weight coefficient of spatial distance; For data points and data points The spatial distance between two points is used to measure the relative position difference between two points in geographic space and is obtained through the Euclidean distance formula; is the weight coefficient of time distance;

[0076] Based on historical experience, the weight coefficients of spatial distance and temporal distance are preset to 0.5 respectively. At this time, it is considered that spatial distance and temporal distance are equally important when calculating the spatiotemporal joint distance between two data points. The weight coefficients of spatial distance and temporal distance are regarded as a weight combination. The performance of different weight combinations on spatiotemporal joint distance is observed through grid search. The weight combination with the best performance is selected as the optimal weight combination. At the same time, it is also verified on an independent test data set to ensure the generalization ability of the optimal weight combination on new data.

[0077] For data points and data points The time distance between two points is used to measure the relative difference between two points in the time dimension. ;in, is the index of the standardized time feature dimension; For data points No. A standardized time feature; For data points No. A standardized time feature; the time feature contained in the data point is mapped to the unit circle through the periodic function sin, and then the standardized time feature is obtained ;in, is the time period; is the ratio of pi; For data points The temporal features included;

[0078] S43, using K-Means clustering method, based on the time-space joint distance formula to cluster the order time-space sequence; randomly select 'Initial cluster center points ; For each data point in the order spatiotemporal sequence, calculate the spatiotemporal joint distance to all cluster center points, assign each data point to the nearest cluster center, and then obtain the initial cluster set;

[0079] S44, for each cluster in the initial cluster set, respectively calculate the mean of all data points in the cluster; use the obtained mean of all data points in the cluster as a new cluster center point, and respectively update the cluster center point of each cluster;

[0080] S45, repeating steps S43-S44 until the preset maximum number of iterations is reached, thereby obtaining a final cluster set; calculating the spatiotemporal density of each cluster in the final cluster set by a density extraction formula, thereby obtaining a spatiotemporal density feature data set; the density extraction formula is: ;in, For the The set of orders in the clusters; For the Weight of items in the order; For the Volume of items in an order; For the Delivery time of each order; For the The spatial area covered by the orders in each cluster; For the The length of the time window in each cluster;

[0081] For example, 20 orders were received between 9:00 and 10:00 in the morning, and the spatial area covered by the orders was 1 square kilometer. The weight of the ordered goods was 2 kilograms, the volume of the ordered goods was 0.1 cubic meters, and the delivery time was one hour later. We can calculate the order density during this time period as follows: ; It can be concluded that the order density in the spatial area covered by orders from 9 am to 10 am is 4;

[0082] The density extraction formula comprehensively considers the weight of the order items, the volume of the order items, the delivery time and the spatial area covered by the order, quantifies the spatiotemporal density of each cluster, and provides an indicator to measure the compactness of clusters and the rationality of distribution. It can identify delivery peak periods and hot spots, improve the response speed of the logistics system, and accurately quantify the order distribution characteristics, combining the spatiotemporal characteristics and order characteristics, providing high-value data support for the optimization decision-making of logistics distribution.

[0083] The method for obtaining the comprehensive feature dataset of order delivery includes:

[0084] The delivery status feature dataset, traffic environment feature dataset and spatiotemporal density feature dataset are integrated through a weighted formula to obtain a comprehensive feature dataset of order delivery. The delivery status feature dataset is recorded as , the traffic environment feature dataset is recorded as , the spatiotemporal density feature dataset is recorded as ;

[0085] The weighted formula is: ;in, Comprehensive feature dataset for order delivery; is the weight coefficient of the delivery status feature data set; is the weight coefficient of the traffic environment characteristic dataset; is the weight coefficient of the spatiotemporal density feature dataset.

[0086] According to the importance of the delivery status feature dataset, traffic environment feature dataset, and spatiotemporal density feature dataset in the order delivery comprehensive feature dataset, the weight coefficient is set. and weight coefficient The initial value range is , the initial default value is ; Set weight coefficient The initial value range is , the initial default value is ; Through the historical order delivery comprehensive feature data set, analyze the contribution of the delivery status feature data set, traffic environment feature data set and spatiotemporal density feature data set to the abnormal coefficient evaluation results of express orders; if the contribution of any one of the delivery status feature data set, traffic environment feature data set and spatiotemporal density feature data set to the abnormal coefficient evaluation results of express orders is significantly higher than that of other data sets (such as the contribution of the abnormal coefficient evaluation results affecting express orders exceeds 50%), then increase the value of the weight coefficient of the corresponding data set, otherwise reduce it, and at the same time gradually optimize the setting of the weight coefficient according to the feedback of the real-time order delivery comprehensive feature data set.

[0087] The training method of the abnormal order evaluation model includes:

[0088] The dataset is divided into training set, validation set and test set, and an abnormal order evaluation model is constructed. The abnormal order evaluation model includes input data and output labels. The input data of the model is the historical order delivery comprehensive feature dataset, and the output label of the model is the abnormal coefficient of the express order. The hyperparameters of the model are selected, including the number of trees and the maximum depth of sample segmentation. During the training process, each tree will randomly select sample features and sample subsets, use random cutting to generate trees, and finally calculate the abnormal coefficient. The abnormal order evaluation model is an isolation forest model.

[0089] The Isolation Forest Model does not require a loss function. The training process detects outliers by selecting split points and building trees. The model is trained on the data set, and the hyperparameters of the model are tuned using the validation set. The cross-validation method is used to find the optimal hyperparameter combination.

[0090] Use the test data set to evaluate the model, and evaluate the model performance by observing the performance indicators of the model on the prediction task. Stop testing when the model performance no longer improves, and obtain the trained abnormal order evaluation model; use the trained abnormal order evaluation model to predict the current order delivery comprehensive feature data set to obtain the abnormal coefficient of the express order.

[0091] According to the abnormal coefficient of the express order, the method for judging whether the express order needs to be reallocated includes:

[0092] Preset an abnormal coefficient threshold of the express order, and compare the abnormal coefficient of the predicted express order with the preset abnormal coefficient threshold of the express order;

[0093] If the predicted abnormal coefficient of the express order is less than the preset abnormal coefficient threshold of the express order, it is determined that there is no need to reallocate the express order;

[0094] If the predicted abnormal coefficient of the express order is greater than or equal to the preset abnormal coefficient threshold of the express order, it is determined that the express order needs to be reallocated.

[0095] The method for obtaining the optimal redistribution path includes:

[0096] S81. Define the objective function and set the total Express orders, including: For the Express orders; Index of the express order; is the total number of express orders; Delivery personnel, including For the A delivery person; is the index of the delivery person, ; is the total number of delivery personnel; Express orders are assigned to delivery personnel; optimize the objective function: ;in, is the total optimization target value; is the total delivery cost; is the total delivery time; For customer satisfaction; is the weight coefficient of the total delivery cost; is the weight coefficient of the total delivery time; is the weight coefficient of customer satisfaction;

[0097] Weight coefficient of total delivery cost , considering that cost is usually one of the important factors of logistics, it can be set to [0.1, 0.5], and the initial default value is 0.3; the weight coefficient of the total delivery time , since timeliness is also crucial in modern logistics, it can be set to [0.1,0.5], with an initial default value of 0.4; the weight coefficient of customer satisfaction , customer satisfaction is the key to long-term business success, so it can be set to [0.1, 0.5], with an initial default value of 0.3; use historical order delivery data to analyze the impact of different factors on the abnormal coefficient evaluation results, and adjust the corresponding weight coefficient according to their contribution:

[0098] Analysis process:

[0099] Collect and analyze historical order delivery data, and calculate the contribution of total delivery cost, total delivery time, and customer satisfaction to the abnormal coefficient. If the contribution of a factor is significantly higher than other factors (for example, more than 50%), increase the weight coefficient of the factor.

[0100] Adjustment mechanism:

[0101] If the contribution of the total delivery cost to the abnormal coefficient is high, increase The value of

[0102] If the contribution of total delivery time to the abnormal coefficient is high, increase The value of

[0103] If customer satisfaction contributes more to the abnormal coefficient, then increase The value of

[0104] S82. Use the branch and bound method to build a decision tree to enumerate all allocation schemes, use upper and lower bound optimization and pruning techniques to reduce the amount of calculation and find the global optimal solution; define the root node Indicates the initial state where all orders are unassigned, and calculates the initial lower bound and the upper bound of the current optimal solution ; Each node Indicates the state of the partial allocation plan. Generates child nodes from the current node. Each child node represents a courier order. Assign to a delivery person ; The upper bound here represents the target value of the currently known optimal feasible solution, that is, the total optimization target value of the currently found best distribution solution , the lower bound is the partial allocation scheme state corresponding to the current node The minimum possible target value, that is, the most optimistic estimate based on the current state.

[0105] S83. Calculate and update the current partial allocation scheme status using the lower bound calculation formula ; The lower bound calculation formula is: ;in, Assign a solution status to the current section The lower bound of is the total cost of currently allocated orders; For orders Assign to delivery person Costs; For unassigned orders;

[0106] S84. If the current partial allocation scheme status The lower bound Greater than or equal to the upper bound of the current optimal solution , then prune; that is, skip further branches of this node to avoid wasting computing resources to explore paths that may not be better than the optimal solution; if the lower bound of a node is already greater than the known optimal solution, or the current cost has exceeded the upper bound of the optimal solution, then the branch of this node cannot bring a better solution, so it can be skipped to avoid wasting computing resources. This strategy greatly improves search efficiency and reduces unnecessary calculations; if the total cost of the currently assigned orders is Greater than the upper bound of the current optimal solution , then pruning is also performed; traverse the complete allocation plan, if the total cost of the currently allocated order Less than or equal to the upper bound of the current optimal solution , then no pruning is performed and the upper bound of the current optimal solution is updated ; When all nodes are processed, the upper bound of the current optimal solution This is the global optimal solution;

[0107] S85. After finding the global optimal solution through the branch and bound method, extract the express orders that each deliveryman is responsible for from the distribution plan corresponding to the optimal solution; each time an order is assigned to a deliveryman, calculate the cost and time of the current distribution state, and update the total cost, total time and customer satisfaction of the currently assigned orders; when traversing all distribution plans, the final optimal solution will be represented as a complete order distribution plan, that is, the optimal deliveryman-order distribution relationship; the optimal reallocation path is the path corresponding to the complete order distribution plan.

[0108] The upper bound of the current optimal solution The definition methods include:

[0109] S91. Use the ant colony algorithm to simulate the foraging behavior of ants and provide the upper bound of the optimal solution for the branch and bound method ; Define the number of ants as , initialize the pheromone matrix as ;

[0110] S92. Each ant starts with a random order, selects a path, and selects the delivery person for the next order by using the selection probability formula; the selection probability formula is: ;in, Choosing a path for the ants The probability of To Node The probability of selection; For path Pheromones on a path reflect the superiority of the path. The more pheromones there are on a path, the better the path may have performed in previous iterations, and the more ants are inclined to choose this path. is the weight coefficient of the pheromone, which controls the influence of the pheromone on the path selection. This means that pheromones have a greater impact on ants’ path selection, and ants are more likely to choose paths with stronger pheromones; is the heuristic information, which is related to the distance or time of the path, indicating that To Node The higher the "superiority" of the heuristic information, the more the ants tend to choose this path; is the weight coefficient of heuristic information, which controls the influence of heuristic information in path selection. It means that heuristic information has a greater impact on path selection, and ants will be more inclined to choose paths with higher heuristic information; For slave nodes The next node to choose from; For slave nodes The set of all next nodes that can be selected;

[0111] S93. Dynamically adjust the weight coefficient of the pheromone through the weight coefficient adjustment formula. The weight coefficient adjustment formula is: ;in, is the total number of nodes; For control Constant factor of maximum value; For control The constant factor of the rate of change with the number of nodes;

[0112] When the total number of nodes is small, since the space for path selection is small, pheromones have a greater guiding effect on the search, so a higher To enhance the effect of pheromones. At this time, pheromones will help ants quickly gather on a shorter path and improve search efficiency; when the total number of nodes is large, the selection space between nodes increases, and ants may rely too much on pheromones, causing the search to converge prematurely or fall into a local optimal solution. Therefore, It should gradually decrease with the increase of the number of nodes, so that ants can rely more on heuristic information, thus avoiding over-reliance on pheromones and enhancing the exploratory nature of the search; ants use pheromones to guide path selection, but the strength of pheromones is directly related to the historical experience of path selection. When the path selection space increases, the influence of pheromones will gradually weaken, and ants will rely more on other heuristic information in the environment; by With the increase of the number of nodes, the influence of pheromone gradually weakens, which helps to improve the algorithm's exploration and avoid falling into the local optimal solution, thereby improving the algorithm's global search ability;

[0113] For example, the total number of nodes 5 o'clock; control Constant factor of maximum value is 1; control Constant factor for the rate of change with the number of nodes is 0.1; the weight coefficient of pheromone ; When the total number of nodes When it increases to 50, the weight coefficient of pheromone ;

[0114] S94. After each ant completes the path construction, calculate its path cost; the path cost is the total optimization target value , taking the cost of each path as the quality index of the path, and updating the pheromone using the pheromone update formula according to the quality index of the path; the pheromone update formula is: ;in, For the updated pheromone; is the volatility coefficient; is the path gain;

[0115] S95, continuously update pheromones until the maximum number of iterations preset by the ant colony algorithm is reached. In each iteration, record and update the upper bound of the current optimal solution. .

[0116] The abnormal coefficient threshold of the preset express order is set by the staff. The abnormal coefficients of different express orders are collected through the order reallocation management terminal, and the average value of the abnormal coefficients of multiple express orders is taken as the abnormal coefficient threshold of the preset express order.

[0117] In this embodiment, by converting the order parameter data into a structured spatiotemporal sequence, a more detailed analysis of the data in the time and space dimensions can be performed to reveal the order distribution law and spatiotemporal pattern; when involving geographic information and time dimensions, the spatiotemporal joint distance can effectively capture the spatiotemporal relationship between orders; the K-Means clustering algorithm is used to cluster the order spatiotemporal sequence by the spatiotemporal joint distance, which can extract the distribution characteristics of orders in different regions and time periods, and provide clustering information for subsequent analysis and optimization;

[0118] Through the branch-and-bound method, a global search is performed to ensure that the optimal delivery person-order allocation plan is found, avoiding the problem of local optimal solutions, so that the balance between overall delivery cost, time and customer satisfaction is optimized; by defining a comprehensive objective function, multi-dimensional optimization of the objective function becomes possible, and the dispatch of delivery persons does not simply consider the optimization of one dimension, but comprehensively considers the optimal balance of cost, time and satisfaction; each time the optimal solution is updated, the decision on whether to continue in-depth exploration is made in combination with the upper limit of the currently known optimal solution, which not only ensures the discovery of the global optimal solution, but also avoids unnecessary calculations; each time an order is allocated, the allocated cost and time are dynamically calculated to ensure that the task burden of each delivery person is reasonable and avoid excessive load or waste of resources; the path and task of each delivery person are adjusted according to real-time order information, ensuring more flexible and efficient path planning;

[0119] By simulating the foraging behavior of ants, the upper bound of the global optimal solution is provided for the branch and bound method. By simulating the behavior of ants in the entire search space and accumulating pheromones, ants can find a better path, thereby providing an effective optimization guide for the branch and bound method, which helps to accelerate the convergence process of the algorithm. As the number of nodes increases, the influence of pheromones gradually weakens, and ants rely more on heuristic information, which can not only avoid premature convergence, but also enhance the diversity of the search process and avoid falling into the local optimal solution. By dynamically adjusting the weight coefficient of pheromones, ants can make full use of historical experience and appropriately rely on the superiority of the current path when selecting paths. By gradually updating the upper bound of the optimal solution in each round of iteration, the ant colony algorithm can limit unnecessary searches and avoid excessive exploration of non-optimal paths in the branch and bound method. Each time the ants build a path, they will evaluate the path cost and adjust the path quality through the pheromone update formula. By continuously adjusting the pheromone, the ants can optimize the path more accurately during the search process. In each iteration, the algorithm makes a selection based on the path quality, so that the optimal path is gradually found after multiple iterations, which improves the accuracy of path planning and avoids meaningless path search.

[0120] Example 2

[0121] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an express order reallocation optimization management system, including:

[0122] Data collection module, used to collect real-time delivery data of express orders;

[0123] The data processing module is used to build the ARIMA model and the time-space joint model, process the real-time delivery data of express orders, and obtain the comprehensive feature data set of order delivery;

[0124] The abnormality monitoring module is used to train an abnormal order evaluation model based on the comprehensive feature data set of order delivery, and predict the abnormal coefficient of the express order; according to the abnormal coefficient of the express order, it is determined whether the express order needs to be reallocated;

[0125] Redistribution optimization module: if express order redistribution is required, the intelligent redistribution algorithm is used to obtain the optimal redistribution path through the branch and bound method under multi-objective constraints;

[0126] The notification management module notifies the delivery personnel to execute the optimal reallocation path and complete the delivery of the reallocated express orders through the order reallocation management terminal.

[0127] Since the electronic device introduced in this embodiment is an electronic device used to implement an express order reallocation optimization management method and system in the embodiment of the present application, based on the express order reallocation optimization management method and system introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as a person skilled in the art implements the electronic device used in the express order reallocation optimization management method and system in the embodiment of the present application, it belongs to the scope of protection of this application.

[0128] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0129] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for optimizing the reallocation of express orders, characterized in that: include: S1. Collect real-time delivery data of express orders; S2, construct ARIMA model and time-space joint model, process the real-time delivery data of express orders, and obtain the comprehensive feature data set of order delivery; The method for processing the real-time delivery data of the express order comprises: The order parameter data, delivery status data and traffic environment data are cleaned, and outliers are identified and eliminated through the box plot method; the order parameter data, delivery status data and traffic environment data after outliers are eliminated are normalized by standard deviation, and converted into a standard state distribution with a mean of 0 and a standard deviation of 1, eliminating the dimensional influence between the data, and obtaining the normalized order parameter data, delivery status data and traffic environment data; Use the ARIMA model to perform time series prediction on the normalized delivery status data and traffic environment data to extract dynamic features, and then obtain the delivery status feature data set and the traffic environment feature data set; build a time-space joint model to perform spatiotemporal correlation analysis on the normalized order parameter data, and then obtain the spatiotemporal density feature data set; The method for acquiring the spatiotemporal density feature data set includes: S41, discretizing the normalized order parameter data into an order time-space sequence containing N data points; each data point in the order time-space sequence is a feature vector having n time-space attributes; S42, defining a spatiotemporal joint distance formula, and measuring the spatiotemporal joint distance between any two data points in the order spatiotemporal sequence by the spatiotemporal joint distance formula; S43, using K-Means clustering method, cluster analysis of order time and space series is performed based on the time and space joint distance formula; S44, for each cluster in the initial cluster set, respectively calculate the mean of all data points in the cluster; use the obtained mean of all data points in the cluster as a new cluster center point, and respectively update the cluster center point of each cluster; S45, repeating steps S43-S44 until the preset maximum number of iterations is reached, thereby obtaining a final cluster set; calculating the spatiotemporal density of each cluster in the final cluster set by a density extraction formula, thereby obtaining a spatiotemporal density feature data set; S3. Train an abnormal order evaluation model based on the comprehensive feature data set of order delivery, and predict the abnormal coefficient of the express order; based on the abnormal coefficient of the express order, determine whether the express order needs to be reallocated; S4. If express order redistribution is required, the intelligent redistribution algorithm is used to obtain the optimal redistribution path through the branch and bound method under multi-objective constraints; S5. Through the order reallocation management terminal, notify the delivery personnel to execute the optimal reallocation path and complete the delivery of the reallocated express order.

2. The express order reallocation optimization management method according to claim 1, characterized in that: The real-time delivery data of express orders includes order parameter data, delivery status data and traffic environment data; order parameter data includes order quantity, order number, recipient information, order item weight, order item volume, delivery address and delivery time; delivery status data includes delivery progress, estimated delivery time, delivery personnel number, delivery personnel location, workload, transport vehicle model and delivery route; traffic environment data includes traffic flow, road conditions and weather conditions.

3. The express order reallocation optimization management method according to claim 2 is characterized in that: The method for acquiring the spatiotemporal density feature data set also includes: The order space-time sequence is {X t }={x1,...,x i ,...,x N }; where t is the index of the timestamp; X t is the order spatiotemporal sequence within time t; N is the total number of data point categories in the order spatiotemporal sequence; i is the index of the data point category in the order spatiotemporal sequence, i = 1, 2, ..., N; x i is the order space-time sequence X t The i-th data point in N is the order space-time sequence X t The Nth data point in ; The space-time joint distance formula is d(x i ,x j )=ω s ·d s (x i ,x j )+ω t ·d t (x i ,x j ), where d(x i ,x j ) is the data point x i and data point x j The space-time joint distance between s is the weight coefficient of spatial distance; d s (x i ,x j ) is the data point x i and data point x j The spatial distance between them is obtained by the Euclidean distance formula; ω t is the weight coefficient of time distance; s is the index of the spatial feature category; t is the index of the temporal feature category; d t (x i ,x j ) is the data point x i and data point x j The time distance between Among them, k is the index of the standardized time feature dimension; ti i,k For data point x i The kth standardized time feature of ti j,k For data point x j The kth standardized time feature of the data point is mapped to the unit circle through the periodic function sin, and then the standardized time feature is obtained. Where T' is the time period; π is the circumference of a circle; For data point x i The temporal features included; Randomly select K' initial cluster centers C1, C2, ..., C K′ ; For each data point in the order spatiotemporal sequence, calculate the spatiotemporal joint distance to all cluster center points, assign each data point to the nearest cluster center, and then obtain the initial cluster set; The density extraction formula is: Among them, E K′ is the order set in the K′th cluster; W I is the weight of the item in the first order; V I is the volume of the item in the first order; t I is the delivery time of the first order; SQ K′ is the spatial area covered by the orders in the K′th cluster; Δt K′ is the time window length in the K′th cluster.

4. The express order reallocation optimization management method according to claim 3 is characterized in that: The method for obtaining the order delivery comprehensive feature data set includes: The delivery status feature dataset, traffic environment feature dataset and spatiotemporal density feature dataset are integrated through a weighted formula to obtain a comprehensive feature dataset of order delivery; the delivery status feature dataset is recorded as M1, the traffic environment feature dataset is recorded as M2, and the spatiotemporal density feature dataset is recorded as M3; The weighted formula is: Among them, SE is the comprehensive feature dataset of order delivery; is the weight coefficient of the delivery status feature data set; is the weight coefficient of the traffic environment characteristic dataset; is the weight coefficient of the spatiotemporal density feature dataset.

5. The express order reallocation optimization management method according to claim 4 is characterized in that: The training method of the abnormal order evaluation model includes: The data set is divided into a training set, a validation set, and a test set, and an abnormal order evaluation model is constructed. The abnormal order evaluation model includes input data and output labels. The input data of the model is a historical order delivery comprehensive feature data set, and the output label of the model is the abnormal coefficient of the express order. The hyperparameters of the model are selected, and the hyperparameters include the number of trees and the maximum depth of sample segmentation. During the training process, each tree randomly selects sample features and sample subsets, generates a tree using random cutting, and finally calculates the abnormal coefficient. The abnormal order evaluation model is an isolation forest model. The Isolation Forest Model does not require a loss function. The training process detects outliers by selecting split points and building trees. The model is trained on the data set, and the hyperparameters of the model are tuned using the validation set. The cross-validation method is used to find the optimal hyperparameter combination. Use the test data set to evaluate the model, and evaluate the model performance by observing the performance indicators of the model on the prediction task. Stop testing when the model performance no longer improves, and obtain the trained abnormal order evaluation model; use the trained abnormal order evaluation model to predict the current order delivery comprehensive feature data set to obtain the abnormal coefficient of the express order.

6. The express order reallocation optimization management method according to claim 5 is characterized in that: The method for determining whether express order reallocation is required according to the abnormal coefficient of the express order includes: Preset an abnormal coefficient threshold of the express order, and compare the abnormal coefficient of the predicted express order with the preset abnormal coefficient threshold of the express order; If the predicted abnormal coefficient of the express order is less than the preset abnormal coefficient threshold of the express order, it is determined that there is no need to reallocate the express order; If the predicted abnormal coefficient of the express order is greater than or equal to the preset abnormal coefficient threshold of the express order, it is determined that the express order needs to be reallocated.

7. The express order reallocation optimization management method according to claim 6, characterized in that: The method for obtaining the optimal redistribution path includes: S81, define the objective function, presuppose that there are o={o1, o2, ..., o p ,...,o a } express orders, of which o p is the pth express order; p is the index of the express order; a is the total number of express orders; there are v = {v1, v2, ..., v q ,...,v b } delivery personnel, among which, v q is the qth deliveryman; q is the deliveryman's index, q = 1, 2, ..., b; b is the total number of deliverymen; o express orders are assigned to v deliverymen; optimization objective function: Z = δ1G + δ2T - δ3H; where Z is the total optimization target value; G is the total delivery cost; T is the total delivery time; H is customer satisfaction; δ1 is the weight coefficient of the total delivery cost; δ2 is the weight coefficient of the total delivery time; δ3 is the weight coefficient of customer satisfaction; S82. Build a decision tree through the branch and bound method to enumerate all allocation schemes, use upper and lower bound optimization and pruning techniques to reduce the amount of calculation and find the global optimal solution; define the root node D0 to represent the initial state of all unallocated orders, calculate the initial lower bound LB(D0) and the current optimal solution upper bound UB; each node D represents the state of part of the allocation scheme, and generates child nodes from the current node. Each child node represents the allocation of a courier order o p Assign to a delivery person v q ; S83, calculate and update the current partial allocation scheme state D′ by using the lower bound calculation formula; the lower bound calculation formula is: Among them, LB(S′) is the lower bound of the current partial allocation scheme state D′; G(D′) is the total cost of the currently allocated orders; g(o p ,v q ) is the order o p Assigned to the delivery personv q The cost of; F is the unallocated order; S84. If the lower bound LB(S′) of the current partial allocation scheme state D′ is greater than or equal to the upper bound UB of the current optimal solution, pruning is performed; if the total cost G(D′) of the currently allocated orders is greater than the upper bound UB of the current optimal solution, pruning is also performed; traverse the complete allocation scheme, if the total cost G(D′) of the currently allocated orders is less than or equal to the upper bound UB of the current optimal solution, no pruning is performed, and the upper bound UB of the current optimal solution is updated; when all nodes are processed, the upper bound UB of the current optimal solution is the global optimal solution; S85. After finding the global optimal solution through the branch and bound method, extract the express orders that each deliveryman is responsible for from the distribution plan corresponding to the optimal solution; each time an order is assigned to a deliveryman, calculate the cost and time of the current distribution state, and update the total cost, total time and customer satisfaction of the currently assigned orders; when traversing all distribution plans, the final optimal solution will be represented as a complete order distribution plan; the optimal reallocation path is the path corresponding to the complete order distribution plan.

8. The express order reallocation optimization management method according to claim 7, characterized in that: The definition method of the upper bound UB of the current optimal solution includes: S91, simulate the foraging behavior of ants through the ant colony algorithm to provide the upper bound UB of the optimal solution for the branch and bound method; define the number of ants as m, and initialize the pheromone matrix as τ0; S92. Each ant starts with a random order, selects a path, and selects the delivery person for the next order by using the selection probability formula; the selection probability formula is: Among them, P ef The probability of the ant choosing the path (e,f); τ ef is the pheromone on the path (e,f); α is the weight coefficient of the pheromone; η ef is the heuristic information; β is the weight coefficient of the heuristic information; r is the next node that can be selected from node e; Q is the set of all next nodes that can be selected from node e; S93. Dynamically adjust the weight coefficient of the pheromone through the weight coefficient adjustment formula. The weight coefficient adjustment formula is: Where L is the total number of nodes; γ is the constant factor that controls the maximum value of α; θ is the constant factor that controls the rate at which α changes with the number of nodes; S94, after each ant completes the path construction, the path cost is calculated; the path cost is the total optimization target value Z, and each path cost is used as the quality index of the path. According to the quality index of the path, the pheromone is updated using the pheromone update formula; the pheromone update formula is: τ′ ef =(1-φ)τ ef +Δτ ef ; where τ′ ef is the updated pheromone; φ is the volatility coefficient; Δτ ef is the path gain; S95. Continuously update the pheromone until the maximum number of iterations preset by the ant colony algorithm is reached. In each iteration, the upper bound UB of the current optimal solution is recorded and updated.

9. An express order reallocation optimization management system, which is used to implement the express order reallocation optimization management method according to any one of claims 1 to 8, characterized in that: include: Data collection module, used to collect real-time delivery data of express orders; The data processing module is used to build the ARIMA model and the time-space joint model, process the real-time delivery data of express orders, and obtain the comprehensive feature data set of order delivery; The abnormality monitoring module is used to train an abnormal order evaluation model based on the comprehensive feature data set of order delivery, and predict the abnormal coefficient of the express order; according to the abnormal coefficient of the express order, it is determined whether the express order needs to be reallocated; Redistribution optimization module: if express order redistribution is required, the intelligent redistribution algorithm is used to obtain the optimal redistribution path through the branch and bound method under multi-objective constraints; The notification management module notifies the delivery personnel to execute the optimal reallocation path and complete the delivery of the reallocated express orders through the order reallocation management terminal.

Images