Logistics end distribution path dynamic optimization method and system

Abstract

The invention relates to the technical field of intelligent logistics, and discloses a dynamic optimization method and system for a logistics tail end distribution path. The method comprises the steps of collecting order distribution and transport capacity data in a delivery area, and dividing an order dense area and an order sparse area through a clustering algorithm; extracting order dense area fluctuation data according to a classification result, dynamically adjusting path planning granularity, establishing a prediction model in combination with historical orders and user consumption habits, and outputting future demand prediction; if the predicted value exceeds the current transport capacity, triggering resource configuration to generate an extension scheme, matching the transport capacity with the task in combination with the real-time scheduling request, and generating an optimized scheduling sequence; a response time index is extracted, and the prediction model is calibrated through comparison of an actual response and a predicted value; and on the basis of the calibrated model and performance indexes, integrating regional feature data to iteratively optimize path parameters, and forming a dynamic path optimization mechanism. According to the invention, the path planning accuracy and the resource utilization rate of logistics end distribution are improved.

Description

Technical Field

[0001] This application relates to the field of smart logistics technology, and in particular to a method and system for dynamic optimization of last-mile delivery routes. Background Technology

[0002] With the rapid development of e-commerce and online shopping, the demand for urban logistics and last-mile delivery services has increased dramatically, especially during peak periods and specific time slots. However, urban delivery faces a series of challenges, including large order volumes, high delivery time requirements, traffic congestion, and limited resources. Traditional delivery route planning methods typically rely on historical data and static rules, which cannot flexibly respond to real-time order changes and capacity fluctuations, easily leading to delivery delays, resource waste, and low delivery efficiency.

[0003] Existing route optimization methods generally focus on optimizing route planning and scheduling algorithms, but rarely incorporate dynamic adjustments based on real-time order distribution, user consumption habits, and fluctuations in transportation capacity. Traditional methods struggle to effectively address the challenge of flexibly adjusting delivery routes and capacity allocation according to order density, actual traffic conditions, and real-time demand forecasts in different regions. Even machine learning-based prediction models lack effective real-time feedback mechanisms, failing to provide timely corrections when prediction errors are significant.

[0004] Therefore, the market urgently needs an intelligent and dynamically adjustable route optimization method that can accurately predict demand based on real-time data, flexibly allocate resources, and calibrate according to actual delivery conditions, thereby improving overall delivery efficiency and reducing delays and resource waste. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and system for dynamic optimization of last-mile delivery routes, which solves the problems of inaccurate route planning and inefficient resource utilization caused by uneven order distribution and capacity fluctuations in last-mile delivery.

[0006] Firstly, this application provides a method for dynamic optimization of last-mile delivery routes, the method comprising: Step S1: Collect order distribution data and historical capacity usage records within the delivery area in real time, and use clustering algorithms to divide the delivery area and identify the classification results of order-dense and sparse areas; Step S2: Based on the classification results, extract the current order volume fluctuation data of the order-intensive area. If the fluctuation exceeds the preset threshold, dynamically adjust the path planning granularity. Step S3: Based on the adjusted path planning granularity and combined with historical order data and user consumption habit characteristics, establish a prediction model and output the demand forecast value for future periods; Step S4: If the demand forecast is higher than the current capacity, the resource allocation process is triggered to generate an expanded capacity allocation plan; Step S5: Obtain the expanded capacity allocation plan, combine it with real-time scheduling request data, and match capacity resources with delivery tasks using a matching algorithm to generate an optimized scheduling response sequence; Step S6: Extract the response time index from the scheduling response sequence, compare the response time index with the demand forecast value using a feedback loop mechanism, calculate the forecast deviation, and calibrate the forecast model; Step S7: Based on the calibrated prediction model and overall performance indicators, integrate regional feature data, dynamically adjust path planning parameters through iterative optimization algorithms, and generate the final dynamic path optimization mechanism.

[0007] Secondly, this application provides a dynamic optimization system for last-mile delivery routes, the system comprising: The data acquisition module is used to collect order distribution data and historical capacity usage records within the delivery area in real time. It uses a clustering algorithm to divide the delivery area and identify the classification results of order-intensive and sparse areas. The regional analysis module is used to extract the current order volume fluctuation data of the order-intensive area based on the classification results. If the fluctuation exceeds a preset threshold, the path planning granularity is dynamically adjusted. The model prediction module is used to build a prediction model based on the adjusted path planning granularity and combined with historical order data and user consumption habit characteristics, and output the demand forecast value for future periods. The resource allocation module is used to trigger the resource allocation process and generate an expanded capacity allocation plan if the demand forecast is higher than the current capacity. The scheduling and matching module is used to obtain the expanded capacity allocation plan, combine it with real-time scheduling request data, and match capacity resources with delivery tasks through a matching algorithm to generate an optimized scheduling response sequence. The feedback calibration module is used to extract the response time index from the scheduling response sequence, compare the response time index with the demand forecast value using a feedback loop mechanism, calculate the prediction deviation, and calibrate the prediction model. The optimization decision module is used to integrate regional feature data based on the calibrated prediction model and overall performance indicators, and dynamically adjust path planning parameters through iterative optimization algorithms to generate the final dynamic path optimization mechanism.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: This application provides a method and system for dynamic optimization of last-mile delivery routes. Through precise identification of order-intensive areas and dynamic adjustment of route planning granularity, it effectively improves the efficiency and resource utilization of logistics delivery. First, a clustering algorithm is used to divide the delivery area, identifying order-intensive and sparse areas. The route planning granularity is adjusted based on order fluctuations. When order demand fluctuations exceed a preset threshold, the route planning granularity is reduced according to the fluctuation ratio, reducing computational load, avoiding system overload, and improving response speed. Regarding demand forecasting, a prediction model is established using a long short-term memory network, combining historical order data and user consumption habits. This model accurately predicts future order demand, including peak and off-peak demand. This accurate demand forecasting provides a scientific basis for subsequent resource allocation and route planning. The backpropagation algorithm is used to optimize neural network parameters, continuously improving prediction accuracy and ensuring efficient allocation under different demand fluctuations.

[0009] Furthermore, a matching algorithm is used to pair expanded transportation capacity resources with real-time scheduling request data to generate an optimized scheduling response sequence. When the matching degree does not meet the preset standard, the matching parameters can be adjusted and re-paired to ensure the accuracy of resource allocation and maximize the utilization of delivery resources. The feedback loop mechanism and the calibration of the prediction model enable real-time adaptation to changes in demand, further reducing resource waste and delays. Finally, through iterative optimization algorithms, the route planning parameters can be dynamically adjusted to generate the optimal route optimization mechanism, ensuring that the system can respond flexibly under high demand fluctuations. This method improves delivery efficiency, reduces resource waste, and provides a reliable intelligent delivery solution in uncertain demand environments. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an embodiment of a dynamic optimization method for last-mile delivery routes in this application. Figure 2 This is a schematic diagram showing the classification results of order-dense and sparse regions in an embodiment of this application. Figure 3 This is a schematic diagram comparing the granularity of path planning in the embodiments of this application; Figure 4 This is a schematic diagram of an embodiment of a dynamic optimization system for last-mile delivery routes in this application. Detailed Implementation

[0012] This application provides a method and system for dynamic optimization of last-mile delivery routes. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the dynamic optimization method for last-mile delivery routes in this application includes: Step S1: Collect order distribution data and historical capacity usage records within the delivery area in real time, and use clustering algorithms to divide the delivery area, identifying the classification results of order-dense and sparse areas.

[0014] The classification results of identifying order-dense and sparse areas in step S1 include: forming an initial dataset based on the collected current order distribution data and historical capacity usage records; dividing the delivery area using the K-means clustering algorithm, calculating cluster centers based on the order density index, and obtaining multiple cluster groups; calculating the average order density of each cluster group, and if the average order density is higher than a preset density threshold, the cluster group is marked as an order-dense area; otherwise, it is marked as a sparse area; extracting the boundary information between the order-dense and sparse areas from the marked cluster groups, generating a classification result map, and verifying the convergence of the clustering algorithm. If it does not converge, the K value is adjusted and the clustering is re-divided to form the final classification result.

[0015] Specifically, real-time order distribution data and historical capacity usage records are collected through delivery platforms or logistics management systems. Order distribution data includes information such as the delivery location, time requirements, and delivery volume for each order. Historical capacity usage records include data on capacity consumption in each region, including delivery personnel's working hours, vehicle usage, and historical peak transport data. This data forms the initial dataset. The initial dataset is then divided using a K-means clustering algorithm. Specifically, each sample point in the initial dataset represents a delivery area unit, and its feature vector includes geographical coordinates, order quantity, and capacity load value. The K-means clustering algorithm is initialized with a preset number of clusters K, and K sample points are randomly selected as initial cluster centers. Then, the clustering process is iterated. In each iteration, the Euclidean distance from all sample points to the current cluster centers is calculated. Each sample point is assigned to the corresponding cluster group according to the shortest distance principle. After all sample points are assigned, the new center point of each cluster group is recalculated. The new center is the mean of all sample points in the group in each feature dimension. This assignment and update process is repeated until the position change of all cluster centers is less than the preset convergence threshold. At this time, the algorithm is determined to have converged and outputs K stable cluster groups. After the algorithm converges, order density analysis and region labeling are performed on each cluster group. The average order density of each group is calculated, which is the sum of the number of orders of all sample points in the group divided by the total area of ​​the geographical area covered by the group. The calculated average order density is compared with a preset business density threshold, which is based on historical order density. If the average order density of a group is higher than the threshold, it is marked as an "order-dense area"; if it is lower than or equal to the threshold, it is marked as a "sparse area".

[0016] To visualize the classification results and use them for subsequent spatial analysis, boundary information needs to be extracted from the labeled clusters to generate a classification result map. Specifically, the convex hull algorithm from computational geometry is used to process all geographic coordinates within each labeled group, calculating the smallest convex polygon enclosing all points. The boundary of this polygon is the outline of the region. All these outlines are then superimposed onto the base map of the delivery area, ultimately generating a classification result map that visually displays the spatial distribution of order-intensive and sparse areas. Figure 2The diagram shows the classification results for dense and sparse order areas. To ensure the reliability of the clustering results, the convergence of this clustering needs to be verified. If the algorithm fails to converge due to poor initial values ​​or an unreasonable K-value setting (e.g., continuous oscillation of the center point, or a significant discrepancy between the generated map's regional division and the actual business situation), the K-value is automatically adjusted, for example, based on the silhouette coefficient, and the entire clustering process, starting from sample point allocation, is restarted until a final regional classification result that satisfies both the algorithm's convergence conditions and the business logic is understood is obtained. This method, by combining order distribution and capacity usage records, can reflect the dynamic changes in delivery areas in real time, providing strong support for optimizing last-mile delivery routes.

[0017] Step S2: Based on the classification results, extract the current order volume fluctuation data of the order-intensive area. If the fluctuation exceeds the preset threshold, dynamically adjust the path planning granularity.

[0018] The step S2, dynamically adjusting the path planning granularity, includes: extracting current order volume fluctuation data for order-intensive areas based on classification results, including fluctuation amplitude and frequency indicators; comparing the fluctuation amplitude with a preset computing resource threshold; if it exceeds the threshold, calculating the excess ratio of the excess portion relative to the computing resource threshold; and proportionally reducing the path planning granularity based on the excess ratio to reduce node resolution and optimize computing load.

[0019] Specifically, high-order-volume areas typically face the challenges of surging order volumes and resource constraints. Therefore, when planning routes in these areas, it's crucial to consider order volume fluctuations to ensure flexible adjustments based on actual demand, thereby optimizing delivery routes and effectively allocating transportation resources. The process of dynamically adjusting route planning granularity is achieved as follows: First, current order volume fluctuation data is acquired. This data includes fluctuation amplitude and frequency indicators. Fluctuation amplitude is calculated by comparing the number of orders in the current time period with the historical average order volume. For example, if the historical average order volume is 100 orders, and the current time period has 150 orders, then the fluctuation amplitude is 50 orders. The frequency indicator reflects the frequency of order fluctuations, typically measured by the number of times order fluctuations occur per unit of time. A higher frequency indicator indicates more frequent order volume fluctuations. This fluctuation data is vital for the system because it helps identify abnormal order volume fluctuations, thereby determining whether adjustments to the route planning granularity are necessary.

[0020] When the fluctuation range of a high-volume order area exceeds a preset computing resource threshold, it indicates that the current order demand has exceeded the system's processing capacity. The computing resource threshold is set based on historical capacity usage records and system computing load data, aiming to ensure that the system can handle surges in order demand without performance degradation during periods of significant order fluctuation. When order fluctuations exceed the computing resource threshold, the ratio between the excess fluctuation and the threshold, i.e., the excess ratio, is calculated using the following formula: For example, if the current fluctuation range is 30 orders, and the computing resource threshold is 20 orders, then the excess ratio is 50%. Based on this excess ratio, the path planning granularity is reduced proportionally, that is, the node resolution is reduced. Node resolution refers to the fineness of key nodes in path planning. Nodes represent key points on the path, such as intersections and turning points. In path planning, the higher the resolution, the more nodes there are, and the greater the computational complexity. Reducing the node resolution means reducing the number of key nodes in the planned path, thereby reducing the computational load. The process of reducing the path planning granularity is essentially reducing the computational complexity in path planning, thereby optimizing the computational load. When order fluctuations are large, by reducing the granularity and the number of nodes, the system can reduce the computational load and improve response speed and efficiency. For example, when the excess ratio is 50%, the path planning granularity is adjusted from every 100 meters to every 500 meters. Figure 3 The diagram illustrates a comparison of path planning granularity. It shows the distribution of path nodes at different granularities in order-dense and sparse areas. Specifically, the two parts of the diagram represent path planning at different granularities. Fine-grained planning results in denser path nodes (e.g., one node every 100 meters), while coarse-grained planning results in sparser path nodes (e.g., one node every 500 meters). This significantly reduces computational resource consumption and improves the system's responsiveness during order surges. By calculating the excess ratio in real-time based on order fluctuation data and adjusting the path planning granularity proportionally, the computational load is reduced. During peak periods of high order demand fluctuations, this ensures the system can process orders efficiently and stably, improving resource utilization.

[0021] Step S3: Based on the adjusted path planning granularity and combined with historical order data and user consumption habit characteristics, establish a prediction model and output the demand forecast value for future periods.

[0022] Step S3, which establishes a prediction model and outputs future demand forecasts, includes: acquiring historical order data and user consumption habit characteristics; extracting key path node information, including node location and connection weights, from the adjusted path planning granularity and historical order data; constructing a training dataset containing historical order data and user consumption habit characteristics; training the constructed neural network model based on the training dataset; calculating time series dependencies through the model's hidden layers to generate preliminary demand forecast results; weighting and fusing the key path node information as spatial features with the preliminary demand forecast results to obtain intermediate demand forecasts; calculating the deviation between the intermediate demand forecasts and historical order data; if the deviation exceeds a preset threshold, optimizing the weight parameters of the neural network; and determining the final future demand forecast based on the optimized neural network model.

[0023] Specifically, in order to further improve the accuracy of predictions and achieve refined resource allocation, it is necessary to extract the key path node locations from the adjusted path planning granularity, and combine historical order data with user consumption habits to establish a demand prediction model. Based on path node information and consumption habit characteristics, the model can output demand prediction values ​​for future periods, providing a more accurate basis for subsequent resource allocation and path optimization.

[0024] Specifically, the process begins by extracting the locations of critical path nodes from the adjusted path planning granularity. Path planning granularity refers to the level of detail used when planning delivery routes; finer granularity results in higher accuracy but also greater computational complexity. After adjusting the path planning granularity, the locations of critical path nodes are represented by geographic coordinates. Each node represents a key point on the route, typically an intersection, turning point, or other important location. Next, the connection weights between critical path nodes are obtained. These weights are calculated based on historical order flow data, which includes trends in order volume and the busyness of delivery routes. High-frequency order connections are assigned higher weights, while low-frequency connections are assigned lower weights. The locations of critical path nodes and the connection weights between them are defined as critical path node information. This approach ensures that the model pays more attention to node information in order-dense areas during subsequent predictions.

[0025] Historical order data is obtained from the enterprise order management database, including but not limited to the total order volume at multiple historical time points, the number of orders in each delivery area, and the average order value. Corresponding user consumption habit characteristics are obtained from the user profiling system, including but not limited to repeat order rate, preferred product types, and average spending. These characteristics help accurately reflect consumer behavior patterns. Combining historical order data and user consumption habit characteristics, a multi-dimensional training dataset is created and input into a neural network model. In particular, using a Long Short-Term Memory (LSTM) network can effectively process time-series data and capture long-term dependencies. The neural network model analyzes historical data and user consumption habits through methods such as backpropagation and gradient descent. How do habitual characteristics influence future order demand? For example, historical order data: Through the hidden layers of the Long Short-Term Memory (LSTM) network, the network can capture time dependencies. For instance, yesterday's order volume may affect today's order demand. User consumption habit characteristics help the model understand the behavioral patterns of specific users. For example, a user's high-frequency ordering may indicate that they will continue to place orders in the next few days. By learning the long-term dependencies of this data, the LTM network can predict future demand trends based on historical order volume and user behavior characteristics, outputting an estimate of order demand for a certain period in the future, i.e., preliminary demand forecast results. These preliminary demand forecast results will serve as the basis for subsequent capacity allocation, resource allocation, and route planning decisions.

[0026] Considering the demand fluctuations in certain regions over different time periods—for example, demand fluctuations in order-intensive areas may be significant—it is necessary to increase the demand forecast values ​​for these regions through weighted adjustments to ensure forecast accuracy. The specific process is as follows: Critical path node information is used as a spatial feature and weighted in conjunction with the preliminary demand forecast results. This weighted fusion operation involves multiplying the node weights by the preliminary demand forecast results to obtain intermediate demand forecast values. These intermediate demand forecast values ​​represent the weighted adjusted demand forecasts, taking into account regional characteristics such as the impact of order density on future demand. The intermediate demand forecast values ​​provide a more accurate prediction of future order volume and reflect the coordination between time-series demand and spatial features. For example, assuming the preliminary demand forecast predicts 100 orders for order-intensive area A in the next hour, and the connection weight of the node corresponding to order-intensive area A extracted from the critical path node information is 1.2, then the final intermediate demand forecast value is: 100 × 1.2 = 120 orders. Because the demand in this area is high, the adjusted forecast result is more consistent with reality.

[0027] Next, it is necessary to calculate the deviation between the intermediate demand forecast and the actual historical order data. Historical order data represents the actual number of orders that have occurred, which is usually the order data that has already been executed. If the deviation exceeds a preset threshold, it means that the prediction accuracy of the model needs to be improved. Therefore, it is necessary to optimize the parameters of the neural network. The process of optimizing parameters is usually achieved through the backpropagation algorithm: the error between the prediction result and the actual result is calculated using a loss function, such as mean squared error. Based on the calculated error, the weights and biases of the neural network are updated through backpropagation using the gradient descent method. The neural network will better fit the data and reduce the deviation of future predictions. The optimized neural network model is retrained based on historical order data and user consumption habits to predict future order demand, generating future demand forecasts. From these forecasts, peak and off-peak demand are identified. Peak demand represents order volume during high-demand periods, such as holidays and weekday peak hours, when demand is typically high. The model identifies demand fluctuations during these peak periods based on historical data and predicts the maximum order volume. Off-peak demand represents order volume during low-demand periods (such as nighttime and weekends), when demand is typically low. Historical data is used to infer the minimum order volume during these low-demand periods. The final future demand forecasts, combining peak and off-peak demand, provide decision support for subsequent capacity allocation and route planning. These forecasts allow for the allocation of more resources during high-demand periods and reduced resource allocation during low-demand periods, ensuring optimal delivery efficiency and resource utilization. The optimized neural network model not only predicts future order demand but also dynamically adjusts based on these forecasts, helping the system make efficient and accurate decisions in the face of demand fluctuations, improving overall delivery efficiency and reducing resource waste.

[0028] Step S4: If the demand forecast is higher than the current capacity, the resource allocation process is triggered to generate an expanded capacity allocation plan.

[0029] The step S4, generating the expanded capacity allocation scheme, includes: obtaining the current capacity, comparing the future demand forecast with the current capacity, calculating the capacity gap when the future demand forecast is higher than the current capacity, quantifying the capacity gap into the required additional capacity resource units, triggering the resource allocation process, and calling the corresponding vehicle and personnel resources from the standby capacity resource pool; based on the capacity gap and available standby resources, formulating expanded resources, including the number of new delivery vehicles and the scheduling arrangement of delivery personnel, integrating the expanded resources with the current capacity, and generating the expanded capacity allocation scheme; verifying the coverage of the expanded capacity allocation scheme, and if the coverage is lower than the preset standard, re-triggering the resource allocation process and optimizing the allocation.

[0030] Specifically, the system compares the predicted demand in high-demand areas with the current transport capacity. If the predicted demand exceeds the current capacity, resources need to be expanded and adjusted. To ensure delivery requirements can still be met during peak periods, a resource allocation process is triggered, dynamically adjusting the system by calling upon the backup transport capacity pool. This process rapidly increases the number of delivery vehicles and personnel, optimizing resource allocation to ensure timely and efficient delivery services.

[0031] Specifically, the current delivery capacity data is obtained from the delivery management system, including the number of available delivery vehicles, the number of delivery personnel, and the number of orders these available resources can support. The number of orders refers to the maximum number of orders that currently available delivery resources, such as delivery vehicles and personnel, can handle within a specific time period. For example, with 10 vehicles and 15 delivery personnel, the maximum number of orders they can handle is calculated based on factors such as vehicle capacity and delivery personnel efficiency; this maximum value is the number of orders. Future demand forecasts are compared with the current delivery capacity. If the future demand forecast is greater than the current delivery capacity, it indicates that the current delivery capacity is insufficient. The difference between the future demand forecast and the current delivery capacity is calculated to obtain the excess capacity gap. After identifying the capacity gap, this gap is quantified into the required additional transportation resources, i.e., the number of delivery vehicles and personnel needed. For example, a gap of 50 orders requires 5 additional delivery vehicles and 10 delivery personnel. The allocation of these resources is based on historical data and experience. Based on the capacity gap and available resources in the backup transportation resource pool, a resource allocation process is triggered. The backup pool contains pre-reserved additional delivery vehicles and personnel, including but not limited to electric vehicles and heavy-duty trucks. These resources are typically configured in advance based on predicted demand fluctuations. An automatic resource allocation mechanism is activated based on the size of the gap to allocate the corresponding resources. After resource allocation is completed, an expanded resource plan is generated, including the number of additional delivery vehicles and the scheduling of delivery personnel. This plan not only needs to increase the required vehicles and personnel but also considers the optimal utilization of transportation resources. For example, in densely populated commercial areas, more electric vehicles may be needed to handle delivery demands on narrow streets, while in suburban areas, more heavy-duty trucks may be needed to cover a larger delivery area. By integrating the expanded resources with the current capacity, an expanded capacity allocation scheme is generated. The expanded capacity allocation scheme includes available delivery resources, including delivery vehicles and delivery personnel information. Delivery vehicles include electric tricycles, trucks, etc. Delivery personnel information includes the availability status, location, and order processing capabilities of delivery personnel. This scheme ensures that all newly added resources can be promptly dispatched and applied to the current delivery tasks.

[0032] To ensure the rationality of the capacity allocation plan and verify its coverage (i.e., whether the expanded resources can meet all future demand requirements), the coverage rate is calculated by the ratio between the currently allocated capacity resources and the predicted demand. If the coverage rate is lower than a preset standard, such as below 95%, the resource configuration process is retried to optimize resource allocation. This may include further increasing the use of backup resources or readjusting the priority of delivery tasks to ensure that all tasks can be completed on time. Finally, after optimization and adjustment, the final capacity allocation plan is output. This plan will be used to acquire subsequent real-time scheduling request data and guide capacity allocation during the delivery process, ensuring efficient and accurate delivery services. By dynamically adjusting capacity resources, the delivery system can quickly respond and optimize resource allocation even under conditions of large demand fluctuations, reducing delays caused by insufficient capacity, improving overall delivery efficiency, and reducing resource waste.

[0033] Step S5: Obtain the expanded capacity allocation plan, combine it with real-time scheduling request data, and match capacity resources with delivery tasks using a matching algorithm to generate an optimized scheduling response sequence.

[0034] Step S5, generating the optimized scheduling response sequence, includes: acquiring the expanded capacity allocation scheme and real-time scheduling request data, whereby the real-time scheduling request data includes request time and location information; using a matching algorithm, pairing the available capacity resources in the expanded capacity allocation scheme with the real-time scheduling request data and calculating the matching degree; prioritizing the application of the matching algorithm in order-intensive areas to determine whether the matching degree meets the preset standard; if not, adjusting the matching parameters and re-pairing; generating a preliminary scheduling response sequence based on the optimized matching results; processing the preliminary scheduling response sequence using a sorting algorithm to determine the response priority based on order urgency and location density, forming an optimized scheduling response sequence; verifying the response efficiency of the optimized scheduling response sequence; if the response efficiency is lower than a preset threshold, iterating the matching algorithm until the efficiency meets the standard.

[0035] Specifically, to ensure that delivery resources can respond quickly and optimize route planning when order volume increases, the expanded capacity allocation scheme is paired with real-time dispatch request data. The matching degree of each pairing is calculated, and a dispatch response sequence is generated based on the matching results. This allows for effective resource allocation during peak periods and in areas with high order density, reducing delivery delays. The following section will describe in detail how to generate optimized dispatch response sequences based on this data.

[0036] First, the expanded capacity allocation plans are obtained, including the added vehicles and delivery personnel. Real-time dispatch request data, containing the request time and location information for each order (e.g., specific street coordinates), is also acquired through a real-time interface. This real-time data ensures the system can respond and match orders immediately. Next, the capacity allocation plans are matched against the real-time dispatch request data using a matching algorithm. The matching degree is calculated using the Hungarian algorithm, which constructs a bipartite graph to match the capacity allocation plans with the dispatch request data. The matching degree depends on the order's location and time, and the formula is as follows: Here, distance is the Euclidean distance between the order location and the delivery resource location, usually calculated using map coordinates. Time difference is the difference between the order request time and the delivery resource availability time. The matching degree determines whether the pairing is reasonable; a high matching degree indicates that the delivery resource and order demand are highly compatible, while a low matching degree indicates an unsatisfactory pairing. Assume an order request has a delivery time of 10:00 AM, is located at point A, and a delivery person is riding an electric tricycle. The rider's current availability time is 9:50 AM, and the rider's current location is 200 meters from the order location. The matching degree between the two is: =0.012, the match between this order and the rider is relatively high because the distance is short and the time difference is suitable. Therefore, this rider will be assigned to this order, and the delivery response will be quick. Assume another order requests delivery at 10:00 AM from location B, and the delivery rider's electric tricycle is 5 kilometers away from the order location. The rider's available time is 9:30 AM, and the rider may have other tasks to perform. In this case, the match degree is: =0.001, the order and rider match is low because of the long distance and large time difference.

[0037] For order-intensive areas such as business centers or busy districts, order request data from these areas are prioritized for input into the matching algorithm for pairing. This is because order-intensive areas typically require more delivery resources, and ensuring timely processing of orders in these areas is paramount. The matching effect is evaluated by calculating the overall matching degree, such as the average matching degree. If the average matching degree is lower than a preset standard threshold, such as 0.001, it indicates that the current matching effect is poor, and delivery resources are not optimally allocated to orders with high demand. To improve the matching effect, the matching process is optimized by adjusting matching parameters. For example, the time difference tolerance may be increased, such as from 5 minutes to 10 minutes. This means that the system allows for a slightly wider time difference between the delivery person and the order. For example, if the original requirement was that the delivery person must be available within 5 minutes before or after the order request time, increasing the time difference tolerance to 10 minutes expands the range of available time for the delivery person, allowing the system to select more delivery persons for matching, thereby improving the matching success rate. Alternatively, the weight of distance can be reduced. For example, if the original distance weight is 0.6, it means that the system values ​​the distance between the delivery person and the order location. If the weight is adjusted to 0.4, it may prioritize the selection of available delivery persons and not consider distance restrictions as much. This allows for the faster allocation of more resources and makes it more adaptable to scenarios with a surge in order volume during peak periods. This adjustment can adapt to changes in demand in different regions and time periods, ensuring that resources are used to their maximum potential.

[0038] If the adjusted matching degree still does not meet the preset standard, parameters will be iteratively adjusted to generate a preliminary scheduling response sequence. This sequence lists the delivery resources for each order according to the matching results, ensuring that orders can be scheduled according to priority. After generating the preliminary scheduling response sequence, a quicksort algorithm will be used to sort the response priorities based on the urgency and location density of the orders. Urgent orders will be processed first, ensuring that high-priority orders are allocated to resources as early as possible, thereby improving overall delivery efficiency. Finally, the response efficiency of the optimized scheduling response sequence will be calculated, for example, by calculating the average response time for each order. If the response efficiency is lower than a set threshold, such as 15 minutes, the matching algorithm will be iteratively run to fine-tune the matching parameters, such as gradually reducing the weight of distance to improve efficiency, until the expected standard is met. Through this iterative process, resource matching can be continuously optimized, delivery delays can be reduced, and all tasks can be completed on time. At the same time, the iterative optimization process ensures that it can be flexibly adjusted in the context of different demand fluctuations to meet diverse delivery needs, improve overall delivery efficiency, and reduce capacity waste.

[0039] Step S6: Extract the response time index from the scheduling response sequence, use a feedback loop mechanism to compare the response time index with the demand forecast value, calculate the forecast deviation and calibrate the forecast model.

[0040] Step S6, which involves calculating the prediction deviation and calibrating the prediction model, includes: extracting response time indicators, including average response time and latency indicators, from the optimized scheduling response sequence; using a feedback loop mechanism to compare the response time indicators with future demand forecasts and calculate the prediction deviation; calibrating the model's prediction accuracy based on the prediction deviation and generating a calibrated accuracy evaluation score; when the accuracy evaluation score is lower than a preset threshold, collecting deviation data and updating the neural network parameters; retraining the prediction model using the updated neural network parameters and determining the refined prediction model based on historical order data; verifying the stability of the refined prediction model; if unstable, repeating the feedback loop and further updating the parameters; and fusing the verified stable prediction model with the response time indicators to output the final calibration result.

[0041] Specifically, in order to ensure that the delivery system can accurately predict future demand and adapt to actual delivery conditions, it is necessary to extract the response time index from the scheduling response sequence and compare it with the predicted future demand value to calculate the prediction deviation and calibrate the prediction model. This process is carried out through a feedback loop mechanism to continuously improve the accuracy of prediction and the responsiveness of the system.

[0042] Specifically, response time metrics are extracted from the optimized scheduling response sequence. These metrics include average response time and latency metrics. By traversing the scheduling response sequence, the start and end times of each order are identified, and the response time for each order is calculated. Then, the response times of all orders are summed to calculate the average response time, which reflects the overall delivery efficiency. The latency metric refers to the proportion of orders whose response time exceeds a preset standard. For example, if the set response time standard is 15 minutes, and 5% of the orders have a response time exceeding this standard, then the latency metric is 5%. Next, through a feedback loop mechanism, the actual response time is compared with the predicted future demand to calculate the prediction deviation. Based on the calculated prediction deviation, the model's prediction accuracy is calibrated. Specifically, the neural network parameters of the model are updated according to the deviation value, generating a calibrated accuracy evaluation score. The calibrated accuracy evaluation score is obtained by calculating the weighted deviation value. When the accuracy evaluation score is lower than a threshold, all deviation data is collected to form a new training set, and the neural network parameters are updated using the backpropagation algorithm. This process uses historical order data and new deviation data to retrain the neural network, optimizing the model's weights and biases to ensure that the model can better adapt to the actual situation and improve prediction accuracy. Specific operations include adjusting the weights of the neural network using the gradient descent method until the ideal prediction effect is achieved. The retrained neural network will process new historical order data to obtain a refined prediction model. It will then calculate new demand predictions through forward propagation. Finally, the stability of the refined prediction model will be verified. Stability verification is achieved by calculating the variance of the predictions over multiple test rounds. If the variance of the model's predictions is too large, it indicates that the model is unstable and the feedback loop needs to be re-executed and the neural network parameters updated. Only when the model's prediction variance is lower than a preset threshold can the model's stability be confirmed and final optimization be performed.

[0043] Once the stability of the refined prediction model has been verified, the response time metric will be combined with the updated prediction model output to generate the final calibration result. This calibration result will be used for subsequent integration of overall performance metrics, ensuring that the model continues to improve efficiency and accuracy in future scheduling tasks. By combining actual response time and demand forecast results, and utilizing a feedback loop mechanism to continuously optimize the prediction model, and by calibrating the accuracy of the prediction model, it can more accurately adapt to different scenarios and order fluctuations, reduce delivery delays, improve the efficiency of capacity planning, increase delivery efficiency, reduce resource waste, and ensure that delivery tasks can be completed on time in a changing demand environment.

[0044] Step S7: Based on the calibrated prediction model and overall performance indicators, integrate regional feature data, dynamically adjust path planning parameters through iterative optimization algorithms, and generate the final dynamic path optimization mechanism.

[0045] The final dynamic route optimization mechanism generated in step S7 includes: constructing a multi-objective optimization function based on future demand forecasts output by a refined prediction model, real-time delivery efficiency indicators and resource utilization data, and integrated regional feature data including delivery area classification results and fluctuation data; using an iterative optimization algorithm with the multi-objective optimization function as the objective, calculating the optimization adjustment amount of the route planning parameters; updating the route planning parameters based on the optimization adjustment amount in each iteration cycle to generate corresponding intermediate optimization schemes; verifying the performance improvement effect of each intermediate optimization scheme; determining optimization convergence when the performance improvement of consecutive iterations is less than a preset threshold; extracting the optimal parameter configuration from the convergence result to form a dynamic optimization mechanism that includes route adjustment strategies and scheduling rules; and verifying the dynamic optimization mechanism in a test environment. If the key performance indicators do not meet the preset standards, the optimization process is re-executed by updating the regional feature data and performance data.

[0046] Specifically, in order to further improve delivery efficiency and resource utilization, based on a refined prediction model and overall performance indicators, combined with regional characteristic data, the path planning parameters are dynamically adjusted through an iterative optimization algorithm to generate the final dynamic path optimization mechanism.

[0047] Specifically, a multi-objective optimization function is constructed by integrating refined predictive model outputs of future demand forecasts, real-time acquired delivery efficiency indicators and resource utilization data, and regional characteristic data. This function includes multiple objectives, each corresponding to different delivery optimization needs, such as improving efficiency, reducing resource waste, and adapting to regional demand fluctuations. This data establishes an optimization objective system, enabling delivery route planning to be optimized in multiple aspects. Next, iterative optimization algorithms, such as gradient descent, are used to calculate the optimization adjustment of route planning parameters, including route planning granularity and capacity allocation strategies. Gradient descent adjusts the route planning parameters to optimize the objective function by calculating the gradient of the loss function. In each iteration, the performance of the current route planning is evaluated, and the gradient is calculated, representing the direction of change of the objective function relative to the route planning parameters. Parameters are updated based on the direction and step size of the gradient, gradually optimizing the route planning until the optimal solution is reached. Each updated parameter further adjusts the route planning until the preset convergence condition is met, ultimately achieving delivery route optimization. The route planning parameters are updated based on the optimization adjustments, generating intermediate optimization schemes. These schemes include adjustments to the granularity of route planning, optimization of capacity resource allocation strategies, and priority ranking. Each iteration generates an intermediate scheme that helps the system evaluate the adjustment effect and provides a basis for the next optimization step. Finally, through multiple iterations and verifications, the optimal route planning and resource allocation scheme is obtained, forming the final dynamic route optimization mechanism. For example, in each cycle, the granularity of route planning may be adjusted from coarser planning (e.g., every 500 meters) to finer planning (e.g., every 100 meters) to optimize the accuracy of resource allocation and route calculation. The performance improvement effect of each intermediate optimization scheme is then verified, i.e., checking whether each iteration effectively improves delivery efficiency and resource utilization. These performance improvements are measured by the overall improvement in delivery efficiency and resource utilization. If, in continuous iterations, the performance improvement is less than a preset threshold (e.g., less than 0.5%), the optimization is considered converged, and iteration stops. The optimal parameter configuration is extracted from the convergence results to form a dynamic optimization mechanism that includes route adjustment strategies and scheduling rules. For example, the optimal parameter configuration might include adjusted route planning granularity, specific capacity allocation schemes, and scheduling rules for specific regions. Next, the applicability of this dynamic optimization mechanism is verified in a test environment. If the verification results show that key performance indicators do not meet preset standards, such as actual delivery efficiency being below 95%, the optimization process is re-executed by updating regional characteristic data and performance data. For example, if insufficient resource allocation is found in certain regions during testing, parameters are readjusted, more regional fluctuation data or historical order records are added, and optimization is performed again.

[0048] By combining refined predictive models, real-time performance metrics, and regional characteristic data, the dynamic adjustment of delivery routes and resources ensures that delivery efficiency and resource utilization are maximized. Through continuous iterative optimization, precise scheduling can be carried out for changes in demand in different regions and time periods, reducing resource waste and delivery delays. Ultimately, the optimized dynamic route adjustment mechanism can efficiently respond to different delivery needs, continuously improving overall delivery efficiency and system flexibility, especially during periods of large demand fluctuations or peak periods, ensuring the timely completion of delivery tasks.

[0049] The construction of the multi-objective optimization function includes: acquiring delivery efficiency indicators and resource utilization data from the delivery management system in real time, aligning the data with the future demand forecasts output by the refined prediction model, overlaying the classification results of order-intensive and sparse areas with historical order fluctuation data to form a regional feature matrix with spatial characteristics, establishing a multi-objective optimization function containing efficiency, resource, and spatial constraint terms based on future demand forecasts, delivery efficiency indicators, resource utilization data, and the regional feature matrix, assigning dynamic weight coefficients to each optimization term according to the characteristics of the operating period, and constructing an optimization objective system.

[0050] Specifically, by integrating multiple key data sources, including delivery efficiency, resource utilization, demand forecasting, and regional characteristics, into a multi-objective optimization function, dynamic path adjustment and resource allocation can be achieved during iterative optimization. First, delivery efficiency indicators and resource utilization data are acquired in real time from the delivery management system. These indicators reflect the execution efficiency of current delivery tasks and the actual use of transportation resources. Efficiency indicators are obtained by calculating the ratio of the number of orders completed per unit time to the total number of orders, while resource utilization is calculated by the ratio of current transportation capacity usage records to total transportation capacity. By acquiring this data in real time, a comprehensive understanding of the current operational status of the delivery system can be achieved.

[0051] Next, the refined prediction model's output of future demand forecasts is aligned with real-time delivery efficiency indicators and resource utilization data. This alignment process ensures that future demand forecasts match actual delivery efficiency and resource utilization, facilitating subsequent optimization. The classification results of order-dense and sparse areas are combined with current order volume fluctuation data to form a regional feature matrix incorporating spatial characteristics. This matrix integrates delivery demand characteristics of different regions, such as order density and fluctuation amplitude, providing foundational data for further optimization. Based on this data, a multi-objective optimization function is established, including efficiency, resource, and spatial constraint terms. By inputting various data into the optimization function, delivery efficiency, resource utilization, and regional characteristics can be considered simultaneously. In the multi-objective optimization function, the efficiency term reflects the overall efficiency of the delivery task, the resource term reflects the actual utilization of transportation capacity, and the spatial constraint term considers the delivery demand characteristics of different regions. To make the optimization more accurate, dynamic weight coefficients are assigned to each optimization term according to the characteristics of the operating period.

[0052] For example, during peak periods, the weights of efficiency and resource items may increase, while during off-peak periods, the weight of spatial constraints may increase to adapt to different delivery needs. In this way, the system can flexibly adjust the weights of optimization objectives, enabling optimal resource allocation and route planning under various operational scenarios. Ultimately, by integrating these factors, a multi-objective optimization function is constructed, providing a scientific basis for subsequent route planning and resource allocation, ensuring efficient delivery services under different environments and demand fluctuations.

[0053] The above describes a method for dynamic optimization of last-mile delivery routes in this application. The following describes a system for dynamic optimization of last-mile delivery routes in this application. Please refer to [link to relevant documentation]. Figure 4 One embodiment of the dynamic optimization system for last-mile delivery routes in this application includes: The data acquisition module is used to collect order distribution data and historical capacity usage records within the delivery area in real time. It uses a clustering algorithm to divide the delivery area and identify the classification results of order-intensive and sparse areas.

[0054] The regional analysis module is used to extract the current order volume fluctuation data of the order-intensive area based on the classification results. If the fluctuation exceeds the preset threshold, the path planning granularity will be dynamically adjusted.

[0055] The model prediction module is used to build a prediction model based on the adjusted path planning granularity and combined with historical order data and user consumption habit characteristics, and output the demand forecast value for future periods.

[0056] The resource allocation module is used to trigger the resource configuration process and generate an expanded capacity allocation plan if the demand forecast is higher than the current capacity.

[0057] The scheduling and matching module is used to obtain the expanded capacity allocation plan, combine it with real-time scheduling request data, and match capacity resources with delivery tasks through a matching algorithm to generate an optimized scheduling response sequence.

[0058] The feedback calibration module is used to extract response time indicators from the scheduling response sequence, and uses a feedback loop mechanism to compare the response time indicators with the demand forecast values, calculate the forecast deviation, and calibrate the forecast model.

[0059] The optimization decision module is used to integrate regional feature data based on the calibrated prediction model and overall performance indicators, and dynamically adjust path planning parameters through iterative optimization algorithms to generate the final dynamic path optimization mechanism.

[0060] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0061] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for dynamic optimization of last-mile delivery routes, characterized in that, The method includes: Step S1: Collect order distribution data and historical capacity usage records within the delivery area in real time, and use clustering algorithms to divide the delivery area and identify the classification results of order-dense and sparse areas; Step S2: Based on the classification results, extract the current order volume fluctuation data of the order-intensive area. If the fluctuation exceeds the preset threshold, dynamically adjust the path planning granularity. Step S3: Based on the adjusted path planning granularity and combined with historical order data and user consumption habit characteristics, establish a prediction model and output the demand forecast value for future periods; Step S4: If the demand forecast is higher than the current capacity, the resource allocation process is triggered to generate an expanded capacity allocation plan; Step S5: Obtain the expanded capacity allocation plan, combine it with real-time scheduling request data, and match capacity resources with delivery tasks using a matching algorithm to generate an optimized scheduling response sequence; Step S6: Extract the response time index from the scheduling response sequence, compare the response time index with the demand forecast value using a feedback loop mechanism, calculate the forecast deviation, and calibrate the forecast model; Step S7: Based on the calibrated prediction model and overall performance indicators, integrate regional feature data, dynamically adjust path planning parameters through iterative optimization algorithms, and generate the final dynamic path optimization mechanism.

2. The method according to claim 1, characterized in that, The classification results of identifying order-dense and sparse areas in step S1 include: An initial dataset is formed based on the collected current order distribution data and historical capacity usage records; the K-means clustering algorithm is used to divide the delivery area, and the cluster centers are calculated based on the order density index to obtain multiple cluster groups; Calculate the average order density of each cluster group. If the average order density is higher than the preset density threshold, the cluster group is marked as an order-dense region; otherwise, it is marked as a sparse region. Extract the boundary information between the order-dense and sparse regions from the marked cluster groups to generate a classification result map and verify the convergence of the clustering algorithm. If it does not converge, adjust the K value and re-cluster to form the final classification result.

3. The method according to claim 1, characterized in that, The dynamic adjustment of path planning granularity in step S2 includes: Based on the classification results, extract the current order volume fluctuation data of the order-intensive area, including fluctuation amplitude and frequency indicators; compare the fluctuation amplitude with a preset computing resource threshold, and if it exceeds the threshold, calculate the excess ratio of the excess part relative to the computing resource threshold; according to the excess ratio, proportionally reduce the path planning granularity, reduce node resolution, and optimize the computing load. Extract adjustment parameters from the reduced path planning granularity to determine whether to use coarse or fine granularity; verify the effect of the adjusted path planning granularity on saving computing resources. If the saving effect is lower than expected, further reduce the granularity level and redetermine the adjusted path planning granularity.

4. The method according to claim 1, characterized in that, The demand forecast values ​​for future time periods output in step S3 include: Historical order data and user consumption habit characteristics are acquired. Key path node information, including node location and connection weight, is extracted from the adjusted path planning granularity and the historical order data. A training dataset containing the historical order data and the user consumption habit characteristics is constructed. The constructed neural network model is trained based on the training dataset. Time series dependencies are calculated through the hidden layers of the model to generate preliminary demand prediction results. The key path node information is used as a spatial feature and weighted and fused with the preliminary demand prediction results to obtain intermediate demand prediction values. Calculate the deviation between the intermediate demand forecast and the historical order data. If the deviation exceeds a preset threshold, optimize the weight parameters of the neural network. Based on the optimized neural network model, determine the final future demand forecast.

5. The method according to claim 1, characterized in that, The step S4 of generating the expanded capacity allocation scheme includes: Obtain the current capacity, compare the future demand forecast with the current capacity, and when the future demand forecast is higher than the current capacity, calculate the capacity gap of the excess, quantify the capacity gap into the required additional capacity resource units, trigger the resource allocation process, and call the corresponding vehicle and personnel resources from the standby capacity resource pool. Based on the capacity gap and available backup resources, expand resources are formulated, including the number of new delivery vehicles and the scheduling of delivery personnel. The expanded resources are integrated with the current capacity to generate an expanded capacity allocation scheme. The coverage of the expanded capacity allocation scheme is verified. If the coverage is lower than the preset standard, the resource configuration process is re-triggered and the allocation is optimized.

6. The method according to claim 1, characterized in that, The optimized scheduling response sequence generated in step S5 includes: The expanded capacity allocation scheme and real-time scheduling request data are obtained, wherein the real-time scheduling request data includes request time and location information; the available capacity resources in the expanded capacity allocation scheme are paired with the real-time scheduling request data using a matching algorithm, and the matching degree is calculated. In the order-intensive area, a matching algorithm is applied first to determine whether the matching degree meets the preset standard. If not, the matching parameters are adjusted and the matching is re-paired. A preliminary scheduling response sequence is generated based on the optimized matching results. The preliminary scheduling response sequence is processed using a sorting algorithm to determine the response priority based on the urgency of the orders and the density of the locations, forming an optimized scheduling response sequence. The response efficiency of the optimized scheduling response sequence is verified. If the response efficiency is lower than the preset threshold, the matching algorithm is iterated until the efficiency meets the standard.

7. The method according to claim 1, characterized in that, Step S6, which involves calculating the prediction bias and calibrating the prediction model, includes: Response time metrics, including average response time and latency metrics, are extracted from the optimized scheduling response sequence. A feedback loop mechanism is used to compare the response time metrics with future demand forecasts to calculate the prediction deviation. The model's prediction accuracy is calibrated based on the prediction deviation, generating a calibrated accuracy evaluation score. When the accuracy evaluation score is below a preset threshold, deviation data is collected and neural network parameters are updated. The prediction model is retrained using the updated neural network parameters, and a refined prediction model is determined based on historical order data. The stability of the refined prediction model is verified; if unstable, the feedback loop is repeated and parameters are further updated. The verified stable prediction model is fused with the response time metrics to output the final calibration result.

8. The method according to claim 1, characterized in that, Step S7 generates the final dynamic path optimization mechanism, which includes: Based on the future demand forecasts output by the refined prediction model, the real-time delivery efficiency indicators and resource utilization data, and the integrated regional feature data including delivery area classification results and fluctuation data, a multi-objective optimization function is constructed. An iterative optimization algorithm is adopted, with a multi-objective optimization function as the objective, to calculate the optimization adjustment amount of the path planning parameters. In each iteration cycle, the path planning parameters are updated based on the optimization adjustment amount to generate the corresponding intermediate optimization scheme. Verify the performance improvement effect of each intermediate optimization scheme. When the performance improvement of continuous iterations is less than a preset threshold, the optimization is determined to be converged. Extract the optimal parameter configuration from the convergence result to form a dynamic optimization mechanism that includes path adjustment strategy and scheduling rules. Verify the dynamic optimization mechanism in a test environment. If the key performance indicators do not meet the preset standards, the optimization process is re-executed by updating the regional feature data and performance data.

9. The method according to claim 8, characterized in that, The construction of the multi-objective optimization function includes: The system acquires delivery efficiency indicators and resource utilization data in real time from the delivery management system and aligns them with the future demand forecasts output by the refined prediction model. It then overlays the classification results of order-intensive and sparse areas with historical order fluctuation data to form a regional feature matrix with spatial characteristics. Based on the future demand forecasts, delivery efficiency indicators, resource utilization data, and regional feature matrix, a multi-objective optimization function is established that includes efficiency, resource, and spatial constraints. Dynamic weight coefficients are assigned to each optimization item according to the characteristics of the operating period to construct an optimization objective system.

10. A dynamic optimization system for last-mile delivery routes, used to implement the dynamic optimization method for last-mile delivery routes as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is used to collect order distribution data and historical capacity usage records within the delivery area in real time. It uses a clustering algorithm to divide the delivery area and identify the classification results of order-intensive and sparse areas. The regional analysis module is used to extract the current order volume fluctuation data of the order-intensive area based on the classification results. If the fluctuation exceeds a preset threshold, the path planning granularity is dynamically adjusted. The model prediction module is used to build a prediction model based on the adjusted path planning granularity and combined with historical order data and user consumption habit characteristics, and output the demand forecast value for future periods. The resource allocation module is used to trigger the resource allocation process and generate an expanded capacity allocation plan if the demand forecast is higher than the current capacity. The scheduling and matching module is used to obtain the expanded capacity allocation plan, combine it with real-time scheduling request data, and match capacity resources with delivery tasks through a matching algorithm to generate an optimized scheduling response sequence. The feedback calibration module is used to extract the response time index from the scheduling response sequence, compare the response time index with the demand forecast value using a feedback loop mechanism, calculate the prediction deviation, and calibrate the prediction model. The optimization decision module is used to integrate regional feature data based on the calibrated prediction model and overall performance indicators, and dynamically adjust path planning parameters through iterative optimization algorithms to generate the final dynamic path optimization mechanism.

Images