Logistics path planning method and system based on artificial intelligence

Abstract

The invention discloses a logistics path planning method and system based on artificial intelligence, and relates to the field of intelligent logistics, and the method comprises the steps: firstly generating a reference path efficiency factor based on the designated path efficiency direction data of a historical logistics order; taking a historical order as a reference, executing dynamic path evolution through a reinforcement learning framework, and screening transition routing strategy configuration of which the fitness score reaches the standard; updating a reference factor in combination with real-time order data, and continuously optimizing through online learning to obtain target routing strategy configuration; and generating a planning decision of the to-be-determined logistics path by using the target strategy, and pushing and displaying the planning decision. According to the method, intelligent optimization of multi-dimensional path efficiency is realized through dynamic evolution and online learning, and the adaptability and accuracy of path planning are improved.

Description

Technical Field

[0001] The present invention relates to the field of smart logistics, and specifically to a logistics route planning method and system based on artificial intelligence. Background Art

[0002] Traditional logistics routing methods primarily rely on manual experience or static algorithms, such as the Dijkstra algorithm and the A* algorithm. These methods can only optimize a single routing metric and struggle to dynamically adapt to complex scenarios such as real-time traffic congestion and fluctuating order demand. Furthermore, existing methods lack the ability to collaboratively optimize multi-dimensional routing performance, resulting in delayed strategy updates and inefficient resource allocation, making them unable to meet the comprehensive needs of modern logistics. Summary of the Invention

[0003] The purpose of the present invention is to provide a logistics path planning method and system based on artificial intelligence.

[0004] In a first aspect, an embodiment of the present invention provides a logistics path planning method based on artificial intelligence, comprising: Based on the performance prediction data of the designated path performance direction corresponding to each historical logistics order, a baseline path performance factor of the path planning model is obtained; wherein each historical logistics order is associated with multiple path performance directions; Each historical logistics order is used as a reference logistics order. Combined with the benchmark path efficiency factor, dynamic path evolution is performed through a reinforcement learning framework. The dynamically evolved transition routing strategy configuration is used as a candidate routing strategy configuration. Each round of evolution includes: For each basic routing policy configuration to be evolved, the following process is performed in sequence: Based on the performance integration data of multiple path performance directions of each reference logistics order obtained from a basic routing policy configuration, a historical path performance factor is obtained, and combined with the baseline path performance factor to obtain the corresponding path fitness score; The basic routing policy configuration whose path fitness score reaches the dynamic optimization threshold is used as the transition routing policy configuration, and each new basic routing policy configuration is selected; Based on the performance integration data of multiple path performance directions of each real-time logistics order itself obtained by the selected routing strategy configuration, a new benchmark path performance factor is obtained, and combined with each new reference logistics order composed of each real-time logistics order, the dynamic path evolution is continuously performed through the online learning mechanism to obtain the target routing strategy configuration; wherein, the target routing strategy configuration is used to obtain the path planning decision of the path planning model for the pending logistics path, so as to push and display the pending logistics path.

[0005] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is configured to execute the method described in at least one possible implementation of the first aspect.

[0006] Compared to existing technologies, the present invention offers the following advantages: It utilizes the disclosed AI-based logistics route planning method and system, which relates to the field of smart logistics, to generate a baseline route efficiency factor based on designated route efficiency direction data from historical logistics orders; uses historical orders as a reference, performs dynamic route evolution through a reinforcement learning framework, and screens transitional routing strategy configurations with satisfactory fitness scores; updates the baseline factor based on real-time order data, and continuously optimizes through online learning to obtain the target routing strategy configuration; and utilizes the target strategy to generate and display planning decisions for pending logistics routes. Through dynamic evolution and online learning, this method achieves intelligent optimization of multi-dimensional route efficiency, improving the adaptability and accuracy of route planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly describes the drawings required for use in the embodiments. It should be understood that the following drawings illustrate only certain embodiments of the present invention and should not be construed as limiting the scope of the present invention. Those skilled in the art can, without inventive effort, derive other relevant drawings from these drawings.

[0008] Figure 1 A schematic diagram of the steps of the artificial intelligence-based logistics path planning method provided in an embodiment of the present invention; Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be understood that the described embodiments are only a portion of the embodiments of the present invention, not all of them. Generally, the components of the embodiments of the present invention described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations.

[0010] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0011] In order to solve the technical problems in the above background technology, Figure 1 This is a flow chart of the artificial intelligence-based logistics path planning method provided in an embodiment of the present disclosure. The artificial intelligence-based logistics path planning method is introduced in detail below.

[0012] Step S201: Obtaining a baseline path efficiency factor for a path planning model based on the performance prediction data for a designated path efficiency direction corresponding to each historical logistics order; wherein each historical logistics order is associated with multiple path efficiency directions; In step S202, each historical logistics order is used as a reference logistics order, and the baseline path efficiency factor is combined to perform dynamic path evolution through a reinforcement learning framework, and the dynamically evolved transition routing strategy configuration is used as a candidate routing strategy configuration. Each round of evolution includes: For each basic routing policy configuration to be evolved, the following process is performed in sequence: Based on the performance integration data of multiple path performance directions of each reference logistics order obtained from a basic routing policy configuration, a historical path performance factor is obtained, and combined with the baseline path performance factor to obtain the corresponding path fitness score; The basic routing policy configuration whose path fitness score reaches the dynamic optimization threshold is used as the transition routing policy configuration, and each new basic routing policy configuration is selected; Step S203: Based on the performance integration data of multiple path performance directions of each real-time logistics order obtained from the candidate routing strategy configuration, a new baseline path performance factor is obtained. Combined with each new reference logistics order formed by each real-time logistics order, the dynamic path evolution is continuously performed through an online learning mechanism to obtain a target routing strategy configuration; Among them, the target routing strategy configuration is used to obtain the path planning decision of the path planning model for the pending logistics path, so as to push and display the pending logistics path.

[0013] In an embodiment of the present invention, the server first constructs a baseline path efficiency factor for the path planning model based on historical logistics order data, which serves as a reference for subsequent strategy optimization. For example, the server stores historical logistics order data from the past 18 months (e.g., 230,000 orders in total), with each order associated with multiple path efficiency metrics, including but not limited to transportation timeliness, transportation cost, cargo integrity rate, on-time delivery rate, and intra-regional path reuse rate. The server first selects the path efficiency metrics that require focus (e.g., balancing timeliness and cost, prioritizing high-value cargo integrity, etc., based on business needs. In this embodiment, the four metrics selected are transportation timeliness, transportation cost, cargo integrity rate, and on-time delivery rate). For each direction, the server extracts performance prediction data from historical orders: Shipping time: Calculates the actual shipping time for all historical orders, excluding delays due to extreme weather and other unusual factors (e.g., an order delayed 20 hours due to heavy rain), resulting in an average shipping time of 11.5 hours. Shipping cost: Actual costs are standardized by cargo weight / volume (to avoid inflated costs for bulky goods), resulting in an average cost of 780 yuan. Cargo integrity rate: Calculates the ratio of the value of damaged goods to the total value of the order, taking the median (to minimize the impact of individual high-damage orders) to obtain a value of 0.08%. On-time delivery rate: Calculates the proportion of orders with actual delivery times ≤ the promised time, taking the 95th percentile (to ensure that most orders meet the target) to obtain a value of 98.5%. The server combines the benchmark values for these five directions into a benchmark route efficiency factor, represented as a five-dimensional vector: Benchmark factor = [Time benchmark 11.5 hours, cost benchmark 780 yuan, integrity benchmark 0.08%, on-time rate benchmark 98.5%, same-region reuse rate benchmark 30%]. The server uses historical orders as reference logistics orders, combines them with the benchmark path efficiency factor, and performs dynamic path evolution through a reinforcement learning framework (such as the PPO algorithm) to generate a transition routing strategy configuration.

[0014] Each evolution cycle involves the following process: First, the server evaluates the performance of the underlying routing policy configuration to be evolved (initial policies such as "Prioritize Shortest Path," "Prioritize Lowest Cost," or "Balance Time and Cost") on reference orders. Taking Policy A ("Prioritize Shortest Path") as an example, the server simulates the routing results when this policy is applied to all reference orders: Transit Time: Based on theoretical map navigation time (e.g., 10 hours for Shanghai-Beijing), corrected by historical congestion data (the average historical congestion delay is 1 hour), the predicted transit time is 11 hours; Transit Cost: Calculated as route length (1,200 kilometers) multiplied by unit cost (0.6 yuan / km), the cost is 720 yuan; Cargo Integrity Rate: Based on the 90% highway portion of the route, the predicted damage rate is 0.05%; On-Time Delivery Rate: Because the predicted transit time of 11 hours is ≤ the promised time of 12 hours, the predicted on-time rate is 100%. Next, the server uses an attention mechanism to calculate the priority weight of each reference logistics order, using the following rule: Total weight = max(customer weight, cargo weight) + 0.1 × urgency (where VIP customer weight is 0.3, regular customer weight is 0.1; high-value cargo weight is 0.2, regular cargo weight is 0.05; urgency: time-limited delivery = 2, next-day delivery = 1, regular delivery = 0). For example, a VIP customer (0.3) with a high-value cargo (0.2) + a time-limited delivery (0.2) = a total weight of 0.3 + 0.2 = 0.5. The top 20% of orders (46,000 in total) are selected as the preferred logistics orders. Based on their performance prediction data, the historical path performance factor for this strategy is generated: Historical Factor (Strategy A) = [Timeliness 11 hours, Cost 720 yuan, Uptime 0.05%, On-time Rate 100%]. The server then calculates the path fitness score for this strategy. Specifically, for timeliness (time series data), an LSTM network is used to extract features (the input is a historical 30-day timeliness sequence, and the output is a 128-dimensional time series feature). For cost and intra-region reuse rate (static values), a fully connected network is used (the input is a standardized value, and it passes through two fully connected layers (64 to 128) to output a feature vector). For availability and punctuality (proportional data), an embedding layer + fully connected network is used (the input is a proportional value, which is embedded into a 32-dimensional vector and then passes through a fully connected layer (32 to 128) to output a feature vector). For example, the historical element 11h of the timeliness is extracted to obtain the vector V_historical timeliness, and the baseline element 11.5h is obtained to obtain the vector V_baseline timeliness. The cosine similarity between the two is calculated as the deviation (the smaller the deviation, the higher the fitness score). In this example, the timeliness deviation is 0.1 (fitness score 0.9), the cost deviation is 0.05 (fitness score 0.95), the availability deviation is 0.02 (fitness score 0.98), and the punctuality deviation is 0.01 (fitness score 0.99). After weighted averaging of the scores of each element (initial weight 0.25), the total score of Strategy A is 0.955.The server sets a dynamic optimization threshold (such as 0.9), configures strategies with a score ≥ 0.9 (such as Strategy A) as transitional routing strategies, and generates a new basic strategy based on it (such as an optimized version of Strategy A, "prioritizing the shortest path + dynamically avoiding congestion").

[0015] The server applies the proposed routing strategy configuration to real-time logistics orders (e.g., the 5,000 new orders added that day) and continuously optimizes through online learning. First, the server calculates the performance of real-time orders under the proposed strategy (e.g., average delivery time of 11.2 hours, cost of 750 yuan, availability rate of 0.07%, and on-time rate of 99%) and updates the baseline path performance factor to [11.2 hours, 750 yuan, 0.07%, 99%]. Furthermore, the server introduces a pre-deployed comparison routing model (e.g., the enterprise's existing rule-based model) with a comparison path performance factor of [12 hours, 800 yuan, 0.15%, 97%]. Using Pareto frontier search, the server selects the core performance elements (e.g., timeliness and cost) that contribute to the top 20% of total fitness and performance constraint elements that are mandatory for business constraints (e.g., on-time rate ≥ 98%). In dynamic path evolution, the weighted parameters of the core performance elements are set to 0.4 (timeliness) and 0.4 (cost) in the first round. These weights are adjusted in subsequent rounds based on deviations: if the timeliness deviation decreases (from 0.1 to 0.05) and the cost deviation increases (from 0.05 to 0.1), the timeliness weight decreases to 0.35 and the cost weight increases to 0.45. The compensation parameter for the performance constraint element (punctuality) is adjusted based on the relationship between deviation and a threshold (e.g., ≤1%): if the on-timeness deviation is +0.5% (meeting the target), the compensation parameter is 0; if the deviation is -0.5% (not meeting the target), the compensation parameter is -0.1 (reducing the overall score). When the transition strategy configuration score fluctuation for five consecutive evolution rounds is ≤0.02 (reaching a stable state), the server switches the roles of the core and constraint elements (e.g., timeliness and cost, previously core, become constraints, and on-timeness, previously a constraint, becomes the core). The deviation threshold is reduced with each evolution round (0.1 in the first round, 0.08 in the second round, and 0.05 in the third round) to explore new optimization directions. Ultimately, through continuous evolution, the server arrives at a target routing policy configuration (e.g., "dynamically balancing timeliness, cost, and on-time performance, prioritizing the integrity of high-value goods for VIP customers"). For a pending logistics route (e.g., "Hangzhou-Guangzhou, VIP customers, high-value precision instruments, promised delivery time of 24 hours"), the server inputs the order information into the routing model and generates performance predictions for each dimension: timeliness of 22 hours (100% on-time performance), cost of 1,200 yuan, and integrity of 0.03%. Based on the target policy configuration, the server integrates this performance data to generate the optimal route (Hangzhou-Nanchang-Guangzhou, primarily on highways) and pushes the route information (route map, estimated timeliness, cost, and risk warnings) to the logistics scheduling system for dispatcher confirmation or automatic execution.

[0016] In summary, this method achieves intelligent and dynamic logistics route planning by building a benchmark based on historical data, strengthening learning dynamic optimization strategies, and continuously adapting to real-time needs through online learning. It also combines comparative models with dynamic parameter adjustment, significantly improving the efficiency and adaptability of route planning.

[0017] In an embodiment of the present invention, the reference path performance factor includes a plurality of reference path performance elements, the historical path performance factor includes a plurality of historical path performance elements, and each reference path performance element corresponds to one historical path performance element; In each round of dynamic path evolution, a corresponding path fitness score is obtained based on the obtained historical path effectiveness factor and combined with the benchmark path effectiveness factor, which can be implemented through the following example.

[0018] For the multiple historical path effectiveness elements in the historical path effectiveness factor, the following processes are performed in sequence: Extracting features of the historical path effectiveness elements to obtain a high-dimensional representation; determining a corresponding path fitness score based on a deviation of a high-dimensional representation of a historical path effectiveness element from a high-dimensional representation of a corresponding baseline path effectiveness element in the baseline path effectiveness factor; Based on the path fitness scores corresponding to the multiple historical path effectiveness elements, a corresponding path fitness score is obtained.

[0019] In this embodiment of the present invention, for example, the baseline path performance factor currently processed by the server includes four baseline path performance elements: baseline time = 11.5 hours, baseline cost = 780 yuan, baseline uptime = 0.08%, and baseline on-time rate = 98.5%. The historical path performance factor (generated by the "shortest path priority" strategy) includes the corresponding four historical path performance elements: historical time = 11 hours, historical cost = 720 yuan, historical uptime = 0.05%, and historical on-time rate = 100%. The server needs to convert each historical path performance element (such as time, cost, etc.) from raw values into a high-dimensional feature vector to capture its implicit business associations and potential patterns. The specific operations are as follows: Feature extraction model selection: The server calls a pre-trained deep neural network (such as a time series feature extractor based on LSTM, or a numerical feature extractor based on a fully connected layer). The model has been trained with historical logistics data and can map the original numerical value of a single performance element into a 128-dimensional high-dimensional representation (such as [v1, v2,..., v128]), retaining the business context information of the element (such as the correlation between timeliness and path congestion, and the correlation between cost and path length). Example of extracting specific elements: Historical time (11 hours): The server inputs "11 hours" into the time series feature extractor. The model combines implicit information such as the route type corresponding to this time (e.g., 90% expressway ratio) and the historical congestion probability (5%) to output a high-dimensional representation V_Time_History = [0.2, 0.5, 0.1, ..., 0.8] (128 dimensions). Benchmark time (11.5 hours): The server inputs "11.5 hours" into the same model. Combined with the average route conditions corresponding to the benchmark time (e.g., 85% expressway ratio, 8% historical congestion probability), the model outputs a high-dimensional representation V_Time_Baseline = [0.3, 0.4, 0.2, ... , 0.7] (128 dimensions); Historical cost (720 yuan): The server inputs "720 yuan" into the numerical feature extractor. The model combines the corresponding path length (1200 kilometers) and unit cost (0.6 yuan / km) with the cost, and outputs a high-dimensional representation V_CostHistory = [0.1, 0.6, 0.3, ..., 0.9]; Baseline cost (780 yuan): The server inputs "780 yuan" into the same model, combines the average path length (1250 kilometers) and unit cost (0.62 yuan / km) corresponding to the benchmark cost, and outputs a high-dimensional representation V_CostBaseline = [0.2, 0.5, 0.4, ..., 0.8]. The server compares the high-dimensional representation of each historical element with the corresponding benchmark element to quantify the strategy's performance in that direction. In this embodiment, the server uses cosine similarity to measure the deviation between two high-dimensional vectors (higher similarity means smaller deviation, and higher fitness score). Calculation of time-sensitive element deviation: The cosine similarity formula is: Sim(Va,Vb)=(Va·Vb) / (||Va||·||Vb||).Substituting V_Aging History and V_Aging Benchmark into the calculated similarity, the result is 0.92 (close to 1, with minimal deviation), resulting in a Time Adaptability Score of 0.92 (using the similarity directly as the score). Cost element deviation calculation: The cosine similarity between V_Cost History and V_Cost Benchmark is 0.88, resulting in a Cost Adaptability Score of 0.88. Similarly, for the Availability and On-Timeness elements: the high-dimensional representation similarity between the historical Availability (0.05%) and the benchmark (0.08%) is 0.95, resulting in a score of 0.95; the high-dimensional representation similarity between the historical On-Timeness (100%) and the benchmark (98.5%) is 0.99, resulting in a score of 0.99. The server then weights and averages the fitness scores of each element (with initial weights of 0.25) to obtain the strategy's overall fitness score: Total score = 0.92 × 0.25 + 0.88 × 0.25 + 0.95 × 0.25 + 0.99 × 0.25 = 0.935. Through this process, the server transforms raw values like timeliness and cost into high-dimensional features that incorporate business context, avoiding the bias of direct numerical comparison (e.g., comparing "11 hours vs. 11.5 hours" might overlook implicit factors like route congestion). Deviation calculation based on high-dimensional representations more accurately reflects strategy performance in real-world logistics scenarios. The final fitness score effectively selects optimal routing strategy configurations, providing a reliable basis for dynamic route evolution.

[0020] In the embodiments of the present invention, the following implementation modes are also provided.

[0021] Obtaining a comparative path effectiveness factor of a comparative path planning model; wherein the comparative path planning model is another path planning model that has been actually deployed, and the comparative path effectiveness factor includes a plurality of comparative path effectiveness elements, and each comparative path effectiveness element corresponds to a reference path effectiveness element in the new reference path effectiveness factor; Determining a core performance element and a performance constraint element among the multiple historical path performance elements included in the historical path performance factor based on deviations between the multiple baseline path performance elements in the new baseline path performance factor and the corresponding comparison path performance elements in the comparison path performance factor; In each round of the dynamic path evolution, a corresponding path fitness score is obtained based on the historical path effectiveness factor and the benchmark path effectiveness factor, including: Based on the deviations between the core performance elements and the performance constraint elements in the historical path performance factor and the corresponding benchmark path performance elements in the new benchmark path performance factor, a corresponding path fitness score is obtained.

[0022] In an embodiment of the present invention, after the server completes updating the baseline path efficiency factor based on historical and real-time orders, it further optimizes the strategy by integrating it with other deployed path planning models (i.e., comparative path planning models). For example, an enterprise may already have a fixed-weight, rule-driven path planning system (comparison model), whose performance data accumulated over long periods of operation can serve as a reference. The server first obtains the comparative path efficiency factor of the comparison model, which contains multiple comparative path efficiency elements corresponding to the new baseline path efficiency factor. For example, the new baseline path efficiency factor is derived from the performance data of 5,000 recent real-time orders and includes four baseline path efficiency elements: timeliness of 11.2 hours, cost of 750 yuan, uptime rate of 0.07%, and on-time rate of 99%. The comparative path efficiency factor is derived from one year of historical data from the comparison model and includes the corresponding elements: timeliness of 12 hours, cost of 800 yuan, uptime rate of 0.15%, and on-time rate of 97%. The server determines the core performance elements and performance constraint elements in the historical path performance factors based on the deviations between the corresponding elements in the new baseline path performance factors and the comparison path performance factors. Specifically, the server first calculates the deviations between each baseline element and its corresponding comparison element: the timeliness deviation is -0.8 hours (the new baseline is better), the cost deviation is -50 yuan (the new baseline is better), the uptime deviation is -0.08% (the new baseline is better), and the punctuality deviation is +2% (the new baseline is better). Then, combining business requirements with historical evolution data, the server uses the NSGA-II algorithm to generate a Pareto frontier solution for a multi-objective optimization (minimizing timeliness and cost while maximizing uptime, punctuality, and intra-regional reuse). From the Pareto frontier, the server selects elements with "deviations that do not meet the standard but contribute to the top 20% of total fitness" as core performance elements, and elements with "deviations that meet the standard and have mandatory business constraints" as performance constraint elements. For example, statistics show that timeliness and cost contribute 40% and 35% to the overall score, respectively (a total of 75%, representing the top 20%), thus making them core performance factors. The enterprise's CNN requirements specify "on-time delivery rate ≥ 98%" and "high-value cargo integrity rate ≤ 0.1%." Therefore, on-time delivery rate and integrity rate are used as performance constraints. In ongoing dynamic path evolution, the server no longer weights all elements equally, but instead designs separate scoring rules for core and constraint elements. For example, the basic routing policy configuration (Policy B: "Dynamically balancing timeliness and cost") from one round of evolution yields the following historical path performance factors: timeliness 11.1 hours, cost 740 yuan, integrity rate 0.06%, and on-time delivery rate 99.5%. The server then calculates the path fitness score for this policy: For the core performance elements (timeliness and cost), their weighting parameters are dynamically adjusted through meta-learning to reflect the current optimization priorities. In the first round of evolution, the weighted parameters of timeliness and cost were 0.4 and 0.4 respectively.The server calculated the cosine similarity of the high-dimensional representations of the historical timeliness (11.1 hours) and the new baseline timeliness (11.2 hours) to be 0.96 (small deviation). Combined with a weighting parameter of 0.4, the timeliness score is 0.96 × 0.4 = 0.384. The high-dimensional representation similarity of the historical cost (740 yuan) and the new baseline cost (750 yuan) is 0.93. Combined with a weighting parameter of 0.4, the cost score is 0.93 × 0.4 = 0.372. For performance constraints (punctuality and availability), the server adjusts the scores using compensation parameters to ensure business bottom lines. The deviation between the historical on-time rate (99.5%) and the new baseline on-time rate (99%) is +0.5% (meeting the target), the high-dimensional representation similarity is 0.98, and the deviation threshold is set to ±1% (the compensation parameter is 0 when meeting the target). Therefore, the on-time rate score is 0.98 × 0.1 (initial weight) + 0 = 0.098. The deviation between the historical uptime rate (0.06%) and the new baseline uptime rate (0.07%) is -0.01% (meeting the target), the high-dimensional representation similarity is 0.95, and the compensation parameter is 0, resulting in a score of 0.95 × 0.1 (initial weight) + 0 = 0.095. Finally, the total path fitness score of Strategy B is the sum of the core and constraint element scores: 0.384 + 0.372 + 0.098 + 0.095 = 0.949. This score indicates that Strategy B performs well in core indicators (timeliness and cost) and meets business constraints (punctuality and availability). Therefore, it is selected by the server as the transitional routing strategy configuration for subsequent evolution and optimization.

[0023] In summary, this implementation method clarifies the optimization direction of the new benchmark factor by introducing a comparative path planning model, combines business constraints to screen core and constraint elements, and adjusts the scoring rules in a targeted manner during dynamic evolution, so that the path planning strategy can not only efficiently optimize key indicators but also ensure basic service quality, significantly improving the practicality and robustness of the strategy.

[0024] 4. The method according to claim 3, wherein determining the core performance element and the performance constraint element of the plurality of historical path performance elements included in the historical path performance factor based on deviations between the plurality of baseline path performance elements in the new baseline path performance factor and the corresponding comparison path performance elements in the comparison path performance factor comprises: For the multiple reference path performance elements in the new reference path performance factor, the following processes are performed in sequence: Obtaining a reference path performance element and a deviation between the reference path performance element and the corresponding comparison path performance element in the comparison path performance factor; Based on the Pareto front search, the historical path performance element corresponding to the benchmark path performance element whose deviation does not reach the deviation condition and whose contribution to the total fitness is in the top K% is used as the core performance element, and the historical path performance element corresponding to the benchmark path performance element whose deviation reaches the deviation condition and has mandatory requirements for business constraints is used as the performance constraint element.

[0025] In an exemplary embodiment of the present invention, after obtaining a new baseline path performance factor (based on real-time logistics order updates) and a comparison path performance factor (based on historical data from a deployed rule-driven comparison model), the server then uses a Pareto front search to select core performance factors and performance constraint factors from multiple path performance elements. The following uses the four path performance elements of "timeliness, cost, availability, and on-time performance" as examples for detailed explanation. The server first calculates the deviation between each baseline path performance element in the new baseline path performance factor and the corresponding comparison path performance element in the comparison path performance factor. Specifically, the new baseline timeliness, derived from the average shipping time of 5,000 recent real-time orders, is 11.2 hours; the comparison model's timeliness, derived from a year's worth of historical order data, is 12 hours. The deviation between the two is 11.2 hours - 12 hours = -0.8 hours (a negative deviation indicates that the new baseline timeliness is superior). The new baseline cost is the standardized average cost of real-time orders, 750 yuan; the comparison cost is the standardized average cost of the comparison model, 800 yuan, with a deviation of 750 yuan - 800 yuan = -50 yuan (a negative deviation indicates a lower new baseline cost). The new baseline intactness rate is the median damage rate of real-time orders, 0.07%; the comparison intactness rate is the median damage rate of the comparison model, 0.15%, with a deviation of 0.07% - 0.15% = -0.08% (a negative deviation indicates a lower new baseline damage rate). The new baseline on-time rate is the 99th percentile of on-time delivery for real-time orders; the comparison on-time rate is the 97th percentile of on-time delivery for the comparison model, with a deviation of 99% - 97% = +2% (a positive deviation indicates a higher new baseline on-time rate).

[0026] The server must comprehensively evaluate the three dimensions of "deviation degree," "contribution to overall fitness," and "business constraints." Using Pareto frontier search, it identifies candidate elements that cannot be simultaneously surpassed by other elements in multiple dimensions. The specific steps are as follows: The server pre-sets "deviation conditions" based on business requirements, determining whether the deviation has reached a threshold requiring mandatory attention. For example: Timeliness deviation condition: an absolute value ≤ 0.5 hours (if the absolute value of the deviation is ≤ 0.5 hours, it is considered "deviation does not meet the standard" and requires optimization; > 0.5 hours is considered "deviation meets the standard" and meets the basic requirements); Cost deviation condition: an absolute value ≤ 30 yuan (≤ 30 yuan is considered "deviation does not meet the standard" and requires optimization; > 30 yuan is considered "deviation meets the standard"); Availability deviation condition: an absolute value ≤ 0.1% (≤ 0.1% is "deviation does not meet the standard"; > 0.1% is "deviation meets the standard"); Punctuality deviation condition: an absolute value ≤ 3% (≤ 3% is "deviation does not meet the standard"; > 3% is "deviation meets the standard"). At the same time, based on historical dynamic path evolution data, the server calculates the contribution of each path efficiency element to the overall fitness score (i.e., the percentage increase in the overall score for each element optimized by 1 unit). For example: for every hour shortened in time, the overall score increases by 0.1 (40% contribution); for every 50 yuan reduction in cost, the overall score increases by 0.08 (35% contribution); for every 0.1% reduction in availability, the overall score increases by 0.03 (15% contribution); for every 1% increase in punctuality, the overall score increases by 0.01 (10% contribution).

[0027] The server includes the four benchmark path efficiency elements and their corresponding deviation values and contributions into the candidate set, forming the following data: Timeliness: deviation -0.8h (absolute value 0.8h>0.5h, meeting the "deviation standard"), contribution 40%; Cost: deviation -50 yuan (absolute value 50 yuan>30 yuan, meeting the "deviation standard"), contribution 35%; Intact rate: deviation -0.08% (absolute value 0.08%≤0.1%, "deviation does not meet the standard"), contribution 15%; Punctuality: deviation +2% (absolute value 2%≤3%, "deviation does not meet the standard"), contribution 10%.

[0028] In this embodiment of the present invention, the specific judgment logic for the Pareto front search is as follows: If element A's deviation meets the standard and its contribution is higher than element B, or if element A's deviation does not meet the standard but its contribution is significantly higher than element B, then element A dominates element B and is retained on the Pareto front. Timeliness (deviation meets the standard, contribution 40%) and cost (deviation meets the standard, contribution 35%): Timeliness has a higher contribution and cannot be dominated by cost, so both are retained. Uptime (deviation does not meet the standard, contribution 15%) and punctuality (deviation does not meet the standard, contribution 10%): Both have lower contributions than timeliness and cost, are dominated by timeliness, and are not retained on the Pareto front.

[0029] Based on the Pareto frontier elements and actual business constraints (e.g., the company specifies that "the availability rate of high-value goods must be ≤0.1%" and "the on-time rate of ordinary orders must be ≥98%)," the server ultimately determines: Core Performance Elements: Select elements whose deviations do not meet the standard (no deviation conditions are triggered) and contribute to the top K% (K=30% in this example) of total fitness. While availability (contribution 15%) and on-time rate (contribution 10%) are not on the Pareto frontier, due to the mandatory business requirements for high-value goods and customer experience (e.g., a 30% proportion of high-value goods), these are included in the core performance elements after adjusting the K% to 30%, with their deviations being the focus of optimization. Performance Constraint Elements: Select elements whose deviations meet the standard (trigger the deviation conditions) and are mandatory business requirements. Timeliness (deviations meet the requirements, business requirements for timeliness ≤ 12 hours) and cost (deviations meet the requirements, business requirements for cost ≤ 800 yuan) already meet basic requirements, but they must be constrained to prevent regression to the previous levels of the comparison model (e.g., timeliness no more than 12 hours, cost no more than 800 yuan). Through Pareto frontier search, the server screens out core performance elements from a multi-dimensional evaluation that both have optimization potential (deviations do not meet the requirements) and significantly impact the overall score (high contribution), while retaining performance constraints required by the business (deviations meet the requirements but must be maintained). This process avoids the one-sidedness of relying solely on a single metric (such as contribution or deviation), ensuring that the path planning strategy focuses on key optimization areas (such as improving uptime and punctuality) while maintaining business bottom lines (such as controlling timeliness and cost), significantly improving the accuracy of strategy optimization and business adaptability.

[0030] In an embodiment of the present invention, the corresponding path fitness score is obtained based on the deviation between the core performance elements and the performance constraint elements in the historical path performance factor and the corresponding benchmark path performance elements in the new benchmark path performance factor, which can be implemented through the following examples.

[0031] For the core performance element, dynamically optimizing weighting parameters through meta-learning, and determining a path fitness score for the core performance element based on the weighting parameters of the core performance element and a deviation between the core performance element and a corresponding benchmark path performance element; For the performance constraint element, determining a path fitness score of the performance constraint element based on a weighting parameter and a compensation parameter of the performance constraint element, and a deviation between the performance constraint element and a corresponding benchmark path performance element; Based on the path fitness scores of the core performance elements and the path fitness scores of the performance constraint elements, a corresponding path fitness score is obtained.

[0032] In an embodiment of the present invention, after determining core performance elements (such as uptime and punctuality) and performance constraint elements (such as timeliness and cost) through Pareto front search, the server then designs scoring rules for each of these elements during each round of dynamic path evolution optimization, ultimately generating a comprehensive path fitness score. The following explanation uses the configuration of a basic routing policy (Policy C: "Prioritize the uptime of high-value goods") during a particular evolution round as an example. Optimizing core performance elements (such as uptime and punctuality) is the core objective of path planning, and their weighting parameters are dynamically adjusted through meta-learning to reflect changes in current optimization priorities. The specific steps are as follows: (1) Initial Weighting Parameter Setting: During the first evolution round, the server sets the initial weighting parameters for the core performance elements based on historical evolution data and business requirements. For example, the initial weighting parameters for uptime and punctuality are 0.4 and 0.3, respectively (the remaining 0.3 is allocated to the constraint elements). (2) Meta-learning dynamically optimizes weighted parameters: The server tracks the deviation changes of core elements in each round of evolution through meta-learning (such as gradient descent algorithm) and dynamically adjusts the weighted parameters. Take the integrity rate as an example: in the previous round of evolution, the historical integrity rate of strategy C was 0.06%, the new baseline integrity rate was 0.07%, and the deviation was -0.01% (better); in this round of evolution, the historical integrity rate of strategy C was optimized to 0.055%, and the new baseline integrity rate was adjusted to 0.068% due to real-time order updates, with a deviation of -0.013% (the deviation is further reduced); the server calculates the deviation change between this round and the previous round (-0.013%-(-0.0 1%) = -0.003%), combined with the total deviation change of all core elements (for example, the deviation change of on-time rate is +0.2%), adjust the weighting parameter of the uptime rate: this round's weighting parameter = the previous round's parameter × (1 + deviation change / total deviation change). Substituting this into the data yields: 0.4 × (1 + (-0.003%) / (-0.003% + 0.2%)) ≈ 0.4 × 1.015 = 0.406 (more significant deviation optimization, increased weight). (III) Calculation of Core Element Fitness Scores: The server calculates scores based on the dynamically adjusted weighting parameters and the deviations between the core elements and the baseline elements (quantified using high-dimensional representation similarity). For example: Intactness: The cosine similarity of the high-dimensional representation of 0.055% historical elements and 0.068% new benchmark elements is 0.97 (small deviation), the weighting parameter is 0.406, and the score is 0.97×0.406≈0.394; Punctuality: The cosine similarity of the high-dimensional representation of 99.6% historical elements and 99% new benchmark elements is 0.99 (small deviation), the weighting parameter is 0.3 (because the deviation of the punctuality rate changes little, the weight is not adjusted), and the score is 0.99×0.3=0.297.

[0033] Performance constraints (such as timeliness and cost) must meet business bottom lines (e.g., timeliness ≤ 12 hours, cost ≤ 800 yuan). The server adjusts the score using compensation parameters to ensure the strategy does not deviate from the constraints. The specific steps are as follows: (1) Setting weighted and compensation parameters: The weighted parameters of performance constraints are initially fixed (e.g., 0.2 for timeliness and 0.1 for cost). The compensation parameters are dynamically adjusted based on the relationship between deviation and threshold. For example, the deviation threshold for timeliness is ±0.5 hours (i.e., a timeliness ≤ 12 hours + 0.5 hours = 12.5 hours is considered met, and >12.5 hours is considered unmet), and the deviation threshold for cost is ±50 yuan (i.e., a cost ≤ 800 yuan + 50 yuan = 850 yuan is considered met, and >850 yuan is considered unmet). (2) Calculating the fitness score of the constraint element: The server first calculates the deviation between the constraint element and the baseline element, and then adjusts the score based on the compensation parameters. Taking timeliness as an example: the historical timeliness is 11.8 hours, the new benchmark timeliness is 11.2 hours, and the deviation is +0.6 hours (that is, the actual timeliness is 0.6 hours slower than the benchmark, but ≤ the compliance threshold of 12.5 hours). The high-dimensional representation similarity is 0.92 (small deviation), the weighting parameter is 0.2, and the base score is 0.92 × 0.2 = 0.184. Because the deviation does not exceed the threshold (assuming the compliance threshold is 12 hours, the new benchmark timeliness is 11.2 hours, and the business constraint is timeliness ≤ 12 hours, the deviation is 11.8 hours - 11.2 hours = +0.6 hours, but 11.8 hours ≤ 12 hours still meets the standard), the compensation parameter is 0 (no points are deducted if the standard is met), and the final timeliness score is 0.184 + 0 = 0.184. The same applies to cost: the historical cost is 760 yuan, the new baseline cost is 750 yuan, the deviation is +10 yuan (≤50 yuan for the compliance threshold), the high-dimensional representation similarity is 0.90, the weighting parameter is 0.1, the base score is 0.90 × 0.1 = 0.09, the compensation parameter is 0, and the final cost score is 0.09 + 0 = 0.09. The server adds the scores of the core performance elements and the performance constraint elements to obtain the total path fitness score of strategy C: 0.394 (uptime rate) + 0.297 (punctuality rate) + 0.184 (timeliness) + 0.09 (cost) = 0.965.

[0034] By dynamically adjusting the weighted parameters of core elements through meta-learning, servers can dynamically allocate optimization resources based on actual optimization results (e.g., prioritizing uptime, which has seen significant recent improvements). Compensation parameters are used to adjust the scores of constraint elements, ensuring that the strategy pursues core objectives while remaining within business boundaries (e.g., timeliness must not exceed 12 hours). This process ensures that path fitness scores are more aligned with actual business needs. Strategies selected through dynamic path evolution can effectively improve core metrics (e.g., uptime and punctuality) while ensuring basic service quality (e.g., timeliness and cost control), significantly enhancing the practicality and robustness of path planning.

[0035] In the embodiment of the present invention, the number of the core performance element is at least one, and in the dynamic path evolution, the weighted parameter of each core performance element in each round of evolution is a first value.

[0036] In this embodiment of the present invention, for example, during dynamic route evolution, the server sets a weighted parameter for each round of evolution for core performance elements (such as uptime and on-time performance) as a "first value." This value is determined based on historical evolution data, service priorities, and the contribution of the core elements to overall fitness, ensuring that optimization resources are prioritized in key areas. The following explanation uses the server's route planning strategy (Strategy D) that prioritizes high-value cargo uptime and on-time delivery experience as an example.

[0037] Through Pareto frontier search and business constraint analysis, the server determined that the core performance elements for this round of evolution are "uptime" and "punctuality" (two in total, meeting the "at least one" requirement). Based on historical evolution data, uptime contributes 40% to overall fitness (i.e., every 0.1% improvement in damage rate increases the overall score by 0.1), and punctuality contributes 30% (every 1% improvement in punctuality increases the overall score by 0.08). Considering the business priority requirements for high-value goods (40%) and VIP customers (30%), the server set the weighted parameters (first value) of the core performance elements in this round of evolution to: uptime 0.4, punctuality 0.3 (the remaining 0.3 is allocated to performance constraints). When calculating the path fitness score of strategy D, the server directly uses the first values set in this round (0.4 for intactness rate and 0.3 for on-time rate) to weight the deviation similarity of the core elements. The specific steps are as follows: After strategy D is applied to the reference logistics orders (including 2,000 high-value goods orders and 1,000 VIP customer orders), the core elements in the generated historical path efficiency factor are as follows: Intactness rate: The historical value is 0.05% (actual damage rate), and the new benchmark intactness rate is 0.07% (derived from real-time order statistics). The cosine similarity of their high-dimensional representations (including implicit information such as path bumpiness and cargo packaging strength) is 0.97 (small deviation, excellent strategy performance); On-time rate: The historical value is 99.8% (actual on-time delivery rate), and the new benchmark on-time rate is 99% (derived from real-time order statistics). The cosine similarity of their high-dimensional representations (including implicit information such as path congestion prediction and deliveryman efficiency) is 0.99 (extremely small deviation, excellent strategy performance). The server calculates the fitness scores of the core elements based on the similarity between the first value (weighting parameter) and the high-dimensional representation: The fitness score for the core elements is: The integrity score = high-dimensional representation similarity × weighting parameter = 0.97 × 0.4 = 0.388; the on-time score = high-dimensional representation similarity × weighting parameter = 0.99 × 0.3 = 0.297. The total fitness score for the core efficiency elements is the sum of these two: 0.388 + 0.297 = 0.685. By setting the weighting parameters for the core efficiency elements to the first value in each evolution round (e.g., 0.4 for integrity and 0.3 for on-time), the server ensures that optimization resources are allocated toward high-priority and high-contribution services. Strategy D's core element score (0.685) is significantly higher than that of non-core elements (e.g., the combined score of the timeliness and cost constraint elements is 0.275), demonstrating that this strategy excels in improving the integrity of high-value goods and the on-time delivery experience for VIP customers, meeting the server's optimization goals for dynamic path evolution. This process makes the path fitness score more in line with actual business needs through clear weighted parameter settings, effectively guiding the screening and optimization of routing strategies.

[0038] In one possible implementation, when the number of the core performance element is multiple, in the dynamic path evolution, the weighted parameter of each core performance element in the first round of evolution is a second value, and in each round of evolution after the first round of evolution, the weighted parameter of a core performance element is obtained by the following process, which can be implemented through the following example.

[0039] Based on the deviation between a core performance element in this round of evolution and the core performance element in the previous round of evolution, the sum of the deviations of all core performance elements in this round of evolution from the corresponding core performance element in the previous round of evolution, and the weighted parameter of the core performance element in the previous round of evolution, the weighted parameter of the core performance element in this round of evolution is obtained.

[0040] In an embodiment of the present invention, for example, during dynamic path evolution, if a server has multiple core performance elements (for example, if "uptime rate" and "on-time rate" are identified as two core elements through a Pareto front search), weighting parameters are set in stages: the first round of evolution uses a preset "second value" as the initial weight; in each subsequent round, the weight is dynamically adjusted based on the deviation of the core elements, ensuring that optimization resources are allocated to the elements with the most significant recent optimization effects. The following example illustrates a server's path planning strategy (Strategy E) that prioritizes improving the uptime rate of high-value goods and on-time delivery rates for VIP customers.

[0041] Based on historical evolution data and business needs, the server set a "second value" as the initial weighting parameter for each core performance element in the first round of evolution. For example, for Uptime Rate (Core Element 1), historical data shows that its contribution to overall fitness is 40% (every 0.1% improvement in damage rate increases the overall score by 0.1). The business prioritizes high-value goods at 40%, so the second value is set to 0.4. For On-Time Rate (Core Element 2), historical data shows that its contribution to overall fitness is 30% (every 1% improvement in on-time rate increases the overall score by 0.08). The business prioritizes VIP customers at 30%, so the second value is set to 0.3.

[0042] After the first round of evolution, the server adjusts the weighting parameters based on the deviation of each core element (i.e., the difference between the deviation of the current round and the previous round). The specific steps are as follows: The server first calculates the deviation of each core element in the current and previous rounds of evolution. Deviation is defined as the difference between the historical path performance element and the new baseline path performance element (for example, the uptime deviation is the difference between the historical damage rate and the baseline damage rate, and the on-time deviation is the difference between the historical on-time rate and the baseline on-time rate). Taking the second round of evolution as an example: For the previous (first) round of data: Uptime: The historical value was 0.06% (the damage rate after the previous round of strategy application), and the new baseline was 0.07% (the baseline after the first round of evolution). The deviation is 0.06% - 0.07% = -0.01% (a negative deviation indicates a better performance). For On-Time: The historical value was 99.2%, and the new baseline was 99%. The deviation is 99.2% - 99% = +0.2% (a positive deviation indicates a better performance). Data for this round (second round): Intact rate: The historical value has been optimized to 0.055% (the breakage rate after the application of this round of strategies). The new benchmark value has been adjusted to 0.068% (the benchmark after the second round of evolution) due to real-time order updates. The deviation is 0.055%-0.068%=-0.013% (the deviation has further decreased). On-time rate: The historical value has been increased to 99.5%. The new benchmark value has been adjusted to 99.1%. The deviation is 99.5%-99.1%=+0.4% (the deviation has increased). The server calculates the "current deviation and previous deviation" of each core element as the deviation change, and sums them up to get the total deviation change: Uptime rate deviation change: -0.013% - (-0.01%) = -0.003% (the deviation is better, the change is negative); On-time rate deviation change: +0.4% - (+0.2%) = +0.2% (the deviation is better, the change is positive); Total deviation change: -0.003% + 0.2% = 0.197% (the sum of the deviation changes of all core elements).

[0043] The server calculates the weighted parameter for this round (using the availability rate as an example) based on the following formula: Weighted parameter for this round = Weighted parameter for the previous round × (1 + Deviation change for this round / Total deviation change). Substituting the data: The weighted parameter for availability in the previous round was 0.4, the deviation change for this round was -0.003%, and the total deviation change was 0.197%, then: Weighted parameter for availability in this round = 0.4 × (1 + (-0.003%) / 0.197%) ≈ 0.4 × (10.015) ≈ 0.394 (due to the negative proportion of the availability deviation change to the total change, the weight is slightly reduced); The weighted parameter for punctuality in the previous round was 0.3, and the deviation change for this round was 0. The difference changes by +0.2%, and the total deviation changes by 0.197%. (Note: The total deviation change here is a positive value: 0.2% / 0.197%≈1.015). Therefore, the weighted parameter for on-time rate in this round = 0.3 × (1 + 0.2% / 0.197%) ≈ 0.3 × (1 + 1.015) ≈ 0.3 × 2.015 ≈ 0.605. (Because the positive change in on-time rate deviation accounts for a larger proportion of the total change, the weight is significantly increased.) The server uses this adjusted weighted parameter to calculate the path fitness score for strategy E. For example: Uptime: The cosine similarity between the high-dimensional representation of the historical value of 0.055% in this round and the new baseline value of 0.068% is 0.97 (small deviation), the weighting parameter is 0.394, and the score is 0.97×0.394≈0.382; Punctuality: The high-dimensional representation similarity between the historical value of 99.5% in this round and the new baseline value of 99.1% is 0.99 (extremely small deviation), the weighting parameter is 0.605, and the score is 0.99×0.605≈0.599; The total score of core elements is 0.382+0.599=0.981 (significantly higher than 0.685 in the first round, indicating that strategy E is more effective in optimizing punctuality).

[0044] By setting the initial weights for the secondary values (e.g., 0.4 for uptime rate and 0.3 for on-time rate) in the first round and dynamically adjusting them based on deviations in subsequent rounds (e.g., increasing the on-time rate weight to 0.605 in the second round), the server achieves dynamic allocation of optimization resources. This means that core elements with more significant recent optimization results (e.g., on-time rate) receive higher weights, guiding the strategy toward further optimization. This process ensures that path fitness scores more closely reflect actual optimization results. Strategies selected through dynamic path evolution can more accurately improve core metrics (e.g., on-time rate), significantly enhancing the adaptability and optimization efficiency of path planning strategies.

[0045] In the embodiment of the present invention, the number of the performance constraint element is at least one, and the compensation parameter of each performance constraint element is obtained by the following process, which can be implemented through the following example.

[0046] For each performance constraint element, a compensation parameter of the performance constraint element is obtained based on a deviation between the performance constraint element and a corresponding reference path performance element and a dynamic loading rule between the deviation threshold.

[0047] In an embodiment of the present invention, exemplarily, in the dynamic path evolution, the server sets compensation parameters for the performance constraint elements (such as timeliness, cost and other business mandatory requirements) to ensure that the strategy does not exceed the business bottom line. The acquisition of compensation parameters needs to be combined with the deviation between the performance constraint element and the benchmark element, as well as the preset deviation threshold, and adjusted through dynamic loading rules. The following is an example of a path planning strategy (strategy F) in which the server processes "timeliness" and "cost" as performance constraint elements (a total of 2, meeting the "at least one" requirement). The server first clarifies the deviation calculation method and deviation threshold of the performance constraint element. For example: Time constraint: The business requirement is "the transportation time of ordinary orders is ≤12 hours", and the new benchmark time is 11.5 hours (derived from real-time order statistics). Timeliness deviation is defined as "historical timeliness, new benchmark timeliness" (i.e., the difference between the actual timeliness after the policy is applied and the benchmark), with a deviation threshold set to "+0.5 hours" (i.e., actual timeliness ≤ 11.5 hours + 0.5 hours = 12 hours is considered met, >12 hours is considered unmet). Cost constraints: The business requirement is "normal order shipping cost ≤ 800 yuan," and the new benchmark cost is 780 yuan (derived from real-time order statistics). Cost deviation is defined as "historical cost, new benchmark cost" (i.e., the difference between the actual cost after the policy is applied and the benchmark), with a deviation threshold set to "+20 yuan" (i.e., actual cost ≤ 780 yuan + 20 yuan = 800 yuan is considered met, >800 yuan is considered unmet). The server sets dynamic loading rules based on the relationship between deviation and deviation threshold. The specific rules are as follows: If the deviation is ≤ the threshold (meeting the standard), the compensation parameter is 0 (no impact on the score); if the deviation is > the threshold (unmeeting the standard), the compensation parameter is -k × (deviation threshold) (k is the adjustment factor, k = 0.1 in this example, and the deviation unit is the same as the element). Take the evolution of Strategy F as an example: Time constraint: After Strategy F is applied, the historical timeliness is 11.8 hours (actual timeliness), while the new baseline timeliness is 11.5 hours. The deviation is 11.8 hours - 11.5 hours = +0.3 hours (≤ the 0.5 hour threshold, meeting the standard). Therefore, the compensation parameter is 0. Cost constraint: After Strategy F is applied, the historical cost is 810 yuan (actual cost), while the new baseline cost is 780 yuan. The deviation is 810 yuan - 780 yuan = +30 yuan (> the 20 yuan threshold, failing the standard). Therefore, the compensation parameter is -0.1 × (30 yuan 20 yuan) = -1 (i.e., a 1-point deduction).The server integrates the compensation parameter into the fitness score calculation of the performance constraint element. The specific steps are as follows: The basic score of the constraint element is obtained by multiplying the high-dimensional representation similarity between it and the benchmark element (reflecting the size of the deviation) by the weighting parameter (in this embodiment, the time weighting is 0.2 and the cost weighting is 0.1): Time: The cosine similarity of the high-dimensional representation of the historical time of 11.8 hours and the new benchmark time of 11.5 hours is 0.93 (small deviation), and the basic score is 0.93×0.2=0.186; Cost: The high-dimensional representation similarity of the historical cost of 810 yuan and the new benchmark cost of 780 yuan is 0.85 (large deviation), and the basic score is 0.85×0.1=0.085. The server adds the compensation parameter to the base score to obtain the final fitness score for the constraint element: Timeliness score: 0.186 + 0 (compensation for meeting the target) = 0.186; Cost score: 0.085 + (-1) (compensation for failing to meet the target) = -0.915 (the score is significantly lower due to cost failure). Strategy F's overall path fitness score is determined by both core elements (such as uptime and punctuality) and the constraint element scores. Because the cost constraint fails to meet the target (a score of -0.915), even with a high core element score (e.g., 0.8), the overall score may still fall below the dynamic optimization threshold (e.g., 0.9), causing the server to eliminate the strategy, ensuring that the strategy does not exceed the business bottom line. By calculating the compensation parameter through dynamic load allocation rules, the server implements "bottom-line control" for the efficiency constraint element. This means that if the target is met, the score remains unchanged; if it is not, the overall score is reduced by a negative compensation parameter, forcing the strategy to optimize the constraint metric. This process ensures that the path planning strategy must meet the business requirements for basic indicators such as timeliness and cost while pursuing core goals (such as improving uptime), significantly enhancing the robustness and business adaptability of the strategy.

[0048] In an embodiment of the present invention, during the continuous dynamic path evolution, when the transition routing policy configuration of the current round of evolution and the transition routing policy configuration of the previous round of evolution reach a stable state, starting from the next round of evolution, each round performs the following process until a stable state is reached again: For each basic routing policy configuration to be evolved, perform the following steps in sequence: Based on the performance integration data of multiple path performance directions of each reference logistics order obtained by a basic routing strategy configuration, a historical path performance factor is obtained; The core efficiency elements in the previous round of evolution are used as efficiency constraint elements in the historical path efficiency factors of this round of evolution, and the efficiency constraint elements in the previous round of evolution are used as the core efficiency elements in the historical path efficiency factors of this round of evolution; Based on the deviations between the core performance elements and performance constraint elements in this round of evolution and the corresponding benchmark path performance elements in the new benchmark path performance factors, a corresponding path fitness score is obtained; The basic routing policy configuration whose path fitness score reaches the dynamic optimization threshold is used as the transition routing policy configuration, and each new basic routing policy configuration is selected.

[0049] In an embodiment of the present invention, for example, while continuously performing dynamic path evolution, the server must first determine whether a "stable state" has been triggered. A stable state is defined as: over three consecutive rounds of evolution, the path fitness score of the transition routing policy configuration fluctuates by ≤ 0.02 (for example, the score is 0.95 in round n, 0.96 in round n+1, and 0.95 in round n+2, with the fluctuation range within 0.01). At this point, the server deems the optimization of the current core and constraint elements to have reached saturation, and needs to explore new optimization directions through role swapping. For example, a server processes a "high-value cargo routing" scenario. In the initial phase (the first five evolution rounds), the core performance elements are "uptime" (weighted by a parameter of 0.4) and "punctuality" (0.3), while the performance constraints are "timeliness" (≤12 hours) and "cost" (≤800 yuan). In the sixth, eighth, and eighth evolution rounds, the transition policy configuration scores reach 0.94, 0.95, and 0.94, respectively (with fluctuations ≤0.01), triggering a stable state. Starting with the ninth evolution round, the server switches roles between the core and constraint elements. After a stable state is reached, the server begins the next evolution round (the ninth round) and performs the following steps: The server selects a base routing policy configuration to be evolved (e.g., Policy G: "Prioritize Optimization of High-Value Cargo Routing Resilience") and applies this policy to all reference logistics orders (including historical and real-time orders, for example, 30,000 in total). It then integrates performance data for each route performance dimension (e.g., timeliness, cost, uptime, and punctuality) to generate a historical route performance factor. For example, after applying Strategy G, the average delivery time for each reference order is 11.3 hours, the average cost is 790 yuan, the average availability rate is 0.06%, and the average on-time rate is 99.2%. Therefore, the historical path performance factor is [delivery time 11.3 hours, cost 790 yuan, availability rate 0.06%, on-time rate 99.2%]. The server marks the core performance elements (availability rate, on-time rate) from the previous round (Round 8) as performance constraints for the current round (Round 9), and marks the performance constraints from the previous round (delivery time, cost) as the core performance elements for this round. The specific adjustments are as follows: Core performance elements for this round: delivery time (original constraint element) and cost (original constraint element); Performance constraints for this round: availability rate (original core element) and on-time rate (original core element). The server calculates the deviation between the core and constraint elements of this round and the new benchmark path efficiency factor (updated by real-time orders; the new benchmark factor for this round is [timeliness 11.2 hours, cost 780 yuan, uptime rate 0.07%, punctuality rate 99%]) and calculates the fitness score based on the weighted and compensation parameters.

[0050] Scoring calculation for core performance elements (timeliness and cost): The weighting parameters of the core elements are dynamically adjusted through meta-learning (the first round uses the second value, and subsequent rounds adjust based on deviation changes). The initial weighting parameters for timeliness and cost in this round are 0.4 and 0.3, respectively (due to the historical contributions of the original constraint elements of 35% and 30%). Timeliness deviation: 11.3 hours (historical value) - 11.2 hours (new baseline) = +0.1 hours, high-dimensional representation cosine similarity is 0.96 (small deviation), and the score is 0.96 × 0.4 = 0.384; cost deviation: 790 yuan (historical value) - 780 yuan (new baseline) = +10 yuan, high-dimensional representation similarity is 0.93 (small deviation), and the score is 0.93 × 0.3 = 0.279. The total score of the core elements is 0.384 + 0.279 = 0.663. Scoring calculation for performance constraint elements (uptime rate, on-time rate): Compensation parameters for these constraint elements are determined based on a dynamic loading rule based on deviations and thresholds (e.g., the uptime threshold is ≤0.1% and the on-time threshold is ≥98%). The uptime deviation is: 0.06% (historical value) - 0.07% (new baseline) = -0.01% (meeting target, compensation parameter 0), with a high-dimensional representation similarity of 0.95 and a weighting parameter of 0.2 (the original core element's initial weight), resulting in a score of 0.95 × 0.2 + 0 = 0.19. The on-time deviation is: 99.2% (historical value) - 99% (new baseline) = +0.2% (meeting target, compensation parameter 0), with a high-dimensional representation similarity of 0.98 and a weighting parameter of 0.1, resulting in a score of 0.98 × 0.1 + 0 = 0.098. The total score for the constraint element is: 0.19 + 0.098 = 0.288.

[0051] The server set a dynamic optimization threshold of 0.9 (the sum of the core and constraint element scores ≥ 0.9). Policy G's total score of 0.663 + 0.288 = 0.951 (meets the target), and therefore was selected as the transitional routing policy configuration. Simultaneously, the server generated a new basic routing policy configuration based on Policy G (e.g., Policy G+: "Prioritize timeliness while controlling cost fluctuations") for the next round of evolution. By swapping the roles of the core and constraint elements, the server proactively switched optimization directions (e.g., from "improving uptime" to "optimizing timeliness and cost") during a stable state, preventing the policy from falling into a local optimum. Policy G's total score (0.951) indicates that the new core elements (timeliness and cost) were effectively optimized after the role swap, while the original core elements (uptime and on-time performance) remained met as constraints (uptime 0.06% ≤ 0.1%, on-time 99.2% ≥ 98%). This process enables the path planning strategy to continuously adapt to dynamic changes in business needs, significantly improving the system's long-term optimization capabilities and flexibility.

[0052] In an embodiment of the present invention, during the continuous dynamic path evolution, when the transition routing policy configuration of the current round of evolution and the transition routing policy configuration of the previous round of evolution reach a stable state, starting from the next round of evolution, the deviation threshold in each round of evolution is obtained by the following process, which can be implemented through the following example.

[0053] The deviation threshold is obtained based on the tuning frequency of the core performance element and the performance constraint element in multiple rounds of evolution; wherein, in multiple rounds of evolution, the deviation threshold decreases with the evolution rounds.

[0054] In an embodiment of the present invention, for example, during continuous dynamic path evolution, a server determines that a stable state has been reached when the fitness score fluctuation of the transition routing policy configuration is ≤ 0.02 for three consecutive rounds (e.g., 0.95 in the 10th round, 0.96 in the 11th round, and 0.95 in the 12th round). At this point, the server must adjust the deviation threshold based on the "tuning frequency" of the core and constraint elements (i.e., the number of times the element has been actively optimized in the past five rounds), starting with the 13th round of evolution, and reduce the threshold with each round. Take the scenario where "timeliness" and "cost" are the current core performance elements (original constraint elements), and "uptime" and "punctuality" are the current performance constraint elements (original core elements) as an example: in the past five rounds (rounds 8-12) of evolution, the core element "timeliness" was tuned (i.e., the strategy optimization direction was targeted at timeliness) four times (tuning frequency 4), and "cost" was tuned three times (tuning frequency 3); the constraint element "uptime" was tuned once (because it had already met the standard, the optimization demand was low), and "punctuality" was tuned twice. The server sets the deviation threshold calculation formula as follows: Current round deviation threshold = Initial threshold × (1 tuning frequency / Total tuning times) × (1 round attenuation coefficient); where the initial threshold is set according to the element type (for example, the initial threshold for timeliness is 0.5 hours, and the initial threshold for cost is 20 yuan), the total number of tuning times is the sum of the number of tuning times for all core and constraint elements in the past five rounds (total number of tuning times in this embodiment = 4 + 3 + 1 + 2 = 10), and the round attenuation coefficient is 0.1 (the threshold decreases by 10% each round). Taking the 13th round of evolution as an example: The time deviation threshold is: Initial threshold 0.5h, tuning frequency 4, calculated as: 0.5h × (14 / 10) × (10.1) = 0.5h × 0.6 × 0.9 = 0.27h (a 46% decrease from the initial threshold); Cost deviation threshold: Initial threshold 20 yuan, tuning frequency 3, calculated as: 20 yuan × (13 / 10) × 0.9 = 20 yuan × 0.7 × 0.9 = 12.6 yuan (a 37% decrease from the initial threshold). Uptime deviation threshold (constraint): Initial threshold 0.1%, tuning frequency 1, calculated as: 0.1% × (11 / 10) × 0.9 = 0.1% × 0.9 × 0.9 = 0.081% (a 19% decrease from the initial threshold); On-time deviation threshold (constraint): Initial threshold 3%, tuning frequency 2, calculated as: 3% × (12 / 10) × 0.9 = 3% × 0.8 × 0.9 = 2.16% (a 28% decrease from the initial threshold). The server applied the adjusted deviation threshold in the 13th round of evolution, significantly improving optimization accuracy.Taking the basic routing policy configuration (Policy H: "Prioritize shortening time while controlling cost fluctuations") as an example: Time constraint: After applying Policy H, the historical timeliness is 11.3 hours, the new baseline timeliness is 11.2 hours, the deviation is +0.1 hours (≤ the new threshold of 0.27 hours, meeting the standard), and the compensation parameter is 0. Cost constraint: The historical cost is 785 yuan, the new baseline cost is 780 yuan, the deviation is +5 yuan (≤ the new threshold of 12.6 yuan, meeting the standard), and the compensation parameter is 0. Uptime constraint: The historical uptime is 0.065%, the new baseline uptime is 0.07%, the deviation is -0.005% (≤ the new threshold of 0.081%, meeting the standard), and the compensation parameter is 0. On-time constraint: The historical on-time is 99.1%, the new baseline is 99%, the deviation is +0.1% (≤ the new threshold of 2.16%, meeting the standard), and the compensation parameter is 0. By dynamically reducing the deviation threshold based on the frequency of tuning, the server gradually tightened the optimization criteria in a stable state (for example, reducing the time threshold from 0.5h to 0.27h), forcing the strategy to optimize in a more refined direction (for example, shortening the time threshold from 11.3h to 11.25h). The fitness score of strategy H (0.97) further improved compared to the pre-stabilization score (0.95), indicating that the dynamic adjustment of the deviation threshold effectively promoted the continuous optimization of the path planning strategy, avoided optimization stagnation caused by excessively large thresholds, and significantly enhanced the system's long-term optimization capabilities.

[0055] In an embodiment of the present invention, the historical path efficiency factor is obtained based on the efficiency integration data of multiple path efficiency directions of each reference logistics order itself obtained by a basic routing strategy configuration, which can be implemented through the following examples.

[0056] Based on the performance integration data of multiple path performance directions of each reference logistics order obtained by a basic routing strategy configuration, the priority weight of each reference logistics order is calculated through an attention mechanism, and each preferred logistics order is selected from each reference logistics order based on the priority weight; The historical path efficiency factor is obtained based on the efficiency prediction data and characteristic identification of multiple path efficiency directions of each preferred logistics order itself.

[0057] In an embodiment of the present invention, for example, when evaluating a basic routing policy configuration (e.g., "prioritize high-value cargo transportation time"), the server first applies the policy to a reference set of logistics orders (including historical orders and real-time orders during a shopping festival, e.g., 50,000 in total), obtaining performance integration data and feature identifiers for each order. Specifically, performance integration data (path performance direction): Transportation Time (T): actual transportation time (unit: hours, e.g., 10.2 hours for Shanghai-Beijing); Transportation Cost (C): standardized cost (excluding weight / volume effects, unit: RMB, e.g., RMB 820); Goods Integrity Rate (D): percentage of damaged goods by value (unit: %, e.g., 0.3%); On-Time Delivery Rate (P): percentage of orders delivered on time (unit: %, e.g., 99%). Feature identification (key attributes affecting priority): Customer type (CT): VIP (1), ordinary (0); Cargo type (GT): high-value electronic equipment (2), fresh food (1), ordinary daily necessities (0); Urgency (UR): time-limited delivery (≤24 hours, 2), next-day delivery (≤48 hours, 1), ordinary delivery (>48 hours, 0); Cargo weight (W): light (≤50kg, 0), medium (50-200kg, 1), heavy (>200kg, 2); Transportation distance (D): short distance (≤500 kilometers, 0), medium distance (500-1500 kilometers, 1), long distance (>1500 kilometers, 2).

[0058] In order for the attention mechanism to effectively capture the priority of orders, the server needs to discretize and embed the features, converting the original features into numerical vectors that the model can process.

[0059] (1) Discretization: Binning of Continuous Features: For continuous features (such as cargo weight and transportation distance), the server discretizes them based on the quantiles of historical data to ensure a balanced distribution of features. For example: Cargo weight (weight distribution of 50,000 orders): Light (≤50kg): 30%, labeled 0; Medium (50-200kg): 50%, labeled 1; Heavy (>200kg): 20%, labeled 2. Transportation distance (distance distribution of 50,000 orders): Short distance (≤500km): 25%, labeled 0; Medium distance (500-1500km): 60%, labeled 1; Long distance (>1500km): 15%, labeled 2.

[0060] (II) Embedding Layer Processing: Discrete Feature Vectorization: The server inputs the discretized features into the embedding layer, mapping each discrete value to a low-dimensional dense vector, preserving the semantic association between features. For example: Customer Type (CT): VIP (1) is mapped to the vector [0.8, 0.2], and Ordinary (0) is mapped to [0.1, 0.9]; Goods Type (GT): High-value electronic equipment (2) is mapped to [1.2, -0.3], fresh food (1) is mapped to [0.5, 0.7], and ordinary daily necessities (0) is mapped to [-0.1, 0.4]; Similarly, for urgency (UR), cargo weight (W), and transportation distance (D), each discrete value corresponds to a 2-dimensional embedding vector. Finally, the five features of each order are embedded and concatenated into a 10-dimensional feature vector (5 features × 2 dimensions / feature). For example, the embedding vector of order A (VIP, high-value electronic equipment, limited delivery time, medium weight, medium distance) is: [0.8, 0.2, 1.2, -0.3, 1.0, 0.5, 0.6, 0.4, 0.7, -0.1].

[0061] The server uses a multi-head attention network to calculate the priority weight of each order. The core is to evaluate the impact of the order on the strategy effect through feature similarity.

[0062] (1) Generation of query (Q), key (K), and value (V) matrices: The server takes the 10-dimensional embedding vectors of 50,000 orders as input and generates Q, K, and V matrices (all with dimensions of 50,000×10). Among them: Q (Query): represents the features that need to be "queried" for the current order; K (Key): represents the "key" features of other orders; V (Value): represents the "value" features of other orders (in this embodiment, V=K, that is, the importance of the feature itself). (2) Attention score calculation and weight normalization: The server calculates the attention score (similarity) between orders using the following formula: Attention score =; Where is the feature dimension (10), which is used for scaling to prevent gradient disappearance. Taking order A as an example, its attention score with order B is: =(0.8×0.7+0.2×(-0.1)+...+0.7×0.4) / 3.16≈2.5 / 3.16≈0.79.

[0063] The attention scores of all order pairs form a 50,000 x 50,000 matrix. After softmax normalization, each order is assigned a priority weight (in the range [0, 1]), representing its relative importance relative to other orders. To enhance business objectives (such as ensuring timely delivery of high-value goods), the server adds an additional weight offset to orders with key characteristics (goods type = 2, urgency = 2). For example, an order with goods type = 2 (high-value electronic equipment) receives a weight of +0.3; an order with urgency = 2 (time-sensitive delivery) receives a weight of +0.2. After this summation, the final weight of order A (goods type = 2, urgency = 2) is 0.79 (attention score) + 0.3 + 0.2 = 1.29 (orders outside the [0, 1] range are truncated to 1.0).

[0064] The server sorted 50,000 orders by final weight from highest to lowest, selecting the top 20% (10,000 orders total) as preferred logistics orders. The order characteristics of these orders were: Cargo Type = 2 (High-value Electronic Equipment), accounting for 85%; Urgency = 2 (Time-Limited Delivery), accounting for 70%; Customer Type = 1 (VIP), accounting for 65%; Average Transport Distance = 1,200 km (Medium Distance); Average Cargo Weight = 150 kg (Medium Distance). The server extracted performance data (transportation timeliness, cost, availability, and on-time performance) from these 10,000 preferred orders and, combining these data with their characteristic identifiers, generated a historical route performance factor. The specific integration rules are as follows: Shipping Time (T): calculates the average actual shipping time for preferred orders (10.5 hours); Shipping Cost (C): calculates the average cost normalized by cargo weight (820 yuan); Cargo Integrity Rate (D): calculates the median damage rate (0.3%, due to stricter packaging for high-value cargo); On-Time Delivery Rate (P): calculates the proportion of on-time deliveries (99%, due to priority scheduling for time-sensitive orders); Feature Identification: identifies key features (e.g., "85% of high-value cargo," "70% of time-sensitive orders"). Ultimately, the historical path performance factor is expressed as: [T = 10.5 hours, C = 820 yuan, D = 0.3%, P = 99%] (features: 85% high-value cargo, 70% time-sensitive orders, 65% VIP customers). By using a multi-head attention mechanism and adjusting weights based on business objectives, the server accurately selects the preferred logistics orders (such as high-value, time-sensitive VIP orders) that are most critical to evaluating strategy effectiveness, preventing the influence of common orders (such as low-value, time-sensitive orders) on the evaluation results. The generated historical route efficiency factors focus on core business scenarios, more accurately reflecting the performance of the underlying routing strategy configuration on key metrics (such as the timeliness of high-value shipments). This provides a reliable basis for calculating fitness scores for subsequent dynamic route evolution, significantly improving the targetedness and accuracy of strategy optimization. Furthermore, combined with feature processing at the embedding layer, the attention mechanism effectively captures semantic correlations between order features (such as the strong correlation between high-value shipments and time-sensitive delivery), further enhancing the rationality of weight calculations.

[0065] In an embodiment of the present invention, the historical path performance factor includes one of the following historical path performance elements: The total transportation time efficiency index, transportation time efficiency index of the same regional route, total transportation cost index, cargo integrity rate index and on-time delivery rate index of each preferred logistics order.

[0066] In an embodiment of the present invention, for example, the server uses an attention mechanism to select 10,000 high-priority preferred logistics orders (accounting for 20%) from 50,000 reference logistics orders. The core features of these orders are: 85% of high-value electronic equipment, 70% of time-limited delivery orders, and 65% of VIP customers, which reflect the needs of key business scenarios during the "Shopping Festival". Based on the performance integration data (transportation time, cost, integrity rate, punctuality rate, etc.) of the 10,000 preferred orders, the server calculates five types of historical path performance elements, as follows: (1) Total transportation time index: The total transportation time index reflects the overall transportation efficiency of the preferred orders. The server calculates the average actual transportation time of the 10,000 preferred orders. For example, the actual transportation time of 10,000 orders is 9.8 hours, 10.2 hours, 10.5 hours, etc. (covering multiple routes such as Shanghai-Beijing and Guangzhou-Chengdu); the server takes the arithmetic average of all times and obtains a total transportation time index of 10.5 hours. (2) Same-region route transit time index: This same-region route transit time index is used to evaluate the strategy's performance across different geographic regions. The server calculates average transit time by order's shipping and receiving region (e.g., North China: Beijing-Tianjin, East China: Shanghai-Hangzhou). For example: For 2,000 premium orders in North China, the average actual transit time is 8.2 hours; for 3,000 premium orders in East China, the average actual transit time is 9.5 hours; and for 5,000 premium orders in Central China, the average actual transit time is 11.8 hours. The final transit time index for the same-region route is: 8.2 hours for North China, 9.5 hours for East China, and 11.8 hours for Central China. (3) Total transportation cost index: This total transportation cost index reflects the overall cost control effectiveness of premium orders. The server calculates the average standardized transportation cost (after adjusting for weight and volume) for 10,000 orders. For example, the standardized costs for each order are 800, 820, 850, and so on (adjusted for weight, e.g., the cost factor for 150kg of cargo is 1.2). The server takes the arithmetic average of all costs, resulting in a total transportation cost of 825. (IV) Cargo Integrity Rate: This index reflects the level of control over cargo damage in preferred orders. The server calculates the median percentage of damaged cargo value across 10,000 orders (to avoid the influence of extreme values). For example, the damage rates for orders are 0.1%, 0.3%, 0.5%, and 0.8%, respectively (high-value electronic equipment generally has a lower damage rate due to its rigorous packaging). After sorting, the server takes the median damage rate for the 5,000th order, resulting in a cargo integrity rate of 0.3% (i.e., 99.7% of the cargo is intact). (V) On-Time Delivery Rate: This index reflects the ability of preferred orders to be delivered on time. The server calculates the percentage of orders out of 10,000 with an "actual delivery time ≤ promised time."For example, out of 10,000 orders, 9,850 were delivered on time (e.g., promised within 24 hours, but actually took 23 hours), while 150 were delayed (e.g., took 25 hours). The on-time delivery rate indicator is 9,850 / 10,000 × 100% = 98.5%. The server integrates these five indicators into a historical route performance factor, ultimately expressed as: [total delivery time 10.5 hours, same-region delivery time (North China 8.2 hours / East China 9.5 hours / Central China 11.8 hours), total cost 825 yuan, uptime rate 0.3%, on-time rate 98.5%]. This factor is directly used to calculate fitness scores in subsequent dynamic route evolution. For example, when evaluating the "prioritize high-value cargo delivery" strategy, if the total delivery time of 10.5 hours is better than the baseline value (e.g., 11 hours), the strategy scores high in terms of delivery time. If the same-region delivery time (e.g., 9.5 hours in East China) is significantly lower than the historical average (e.g., 10 hours), the strategy demonstrates outstanding optimization results in that region. By defining five specific elements of historical route efficiency factors, the server implements a multi-dimensional quantitative assessment of basic routing strategy configurations: The combination of total time efficiency and same-region time efficiency reflects both overall efficiency and regional differences; and the integration of total cost with uptime and punctuality rates balances cost control and service quality. This process enables more targeted strategy optimization (for example, adjusting routes to address high time efficiency in the central China region), significantly improving the refinement of logistics route planning.

[0067] In the embodiments of the present invention, the following implementation modes are also provided.

[0068] Inputting the undetermined logistics path into the path planning model to obtain performance prediction data of multiple path performance directions corresponding to the undetermined logistics path; Based on the target routing strategy configuration, integrating the performance prediction data of multiple path performance directions corresponding to the undetermined logistics path to obtain performance integration data corresponding to the undetermined logistics path; Based on the efficiency integration data corresponding to the pending logistics path, the pending logistics path is pushed and displayed.

[0069] In an embodiment of the present invention, exemplarily, before executing the push display of the pending logistics path, the server has completed the following key technical steps: generation of baseline path efficiency factor: based on the efficiency prediction data (transportation time, cost, intactness rate, punctuality rate) of historical logistics orders (230,000), the initial baseline factor is calculated as [timeliness 11.5h, cost 780 yuan, intactness rate 0.08%, punctuality rate 98.5%]; dynamic path evolution optimization: through the reinforcement learning framework, with historical orders as a reference, multiple rounds of dynamic evolution are performed in combination with the baseline factor to screen the transition routing strategy configuration. In each round of evolution, the attention mechanism is used to screen the preferred orders (accounting for 20%, focusing on high-value goods), generate the historical path efficiency factor (such as [timeliness 11h, cost 720 yuan, intactness rate 0.05%, punctuality rate 100%]), and through feature extraction (high-dimensional representation), Pareto frontier search (determining the core elements as "intactness rate" and "punctuality rate", and the constraint elements as "timeliness" and "cost"), meta-learning to dynamically adjust the weighted parameters of the core elements (the second value in the first round is 0.4 and 0.3, and subsequent rounds are adjusted based on deviation changes), compensation parameter constraints (if the cost deviation exceeds the threshold, points will be deducted), and other steps. , optimization strategy; continuous iteration of online learning: the selected strategy is applied to real-time orders (5,000 new orders during the "Shopping Festival"), and the benchmark factors are updated to [timeliness 11.2 hours, cost 750 yuan, completeness rate 0.07%, on-time rate 99%]. Through mechanisms such as core / constraint element role swapping (after the stable state is triggered, the original core "completeness rate" and "on-time rate" become constraints, and the original constraints "timeliness" and "cost" become core), and dynamic reduction of deviation thresholds (for example, the timeliness threshold is reduced from 0.5 hours to 0.27 hours), the target routing strategy configuration (the optimal strategy that balances core and constraint elements) is finally obtained.

[0070] After receiving a new logistics order (pending logistics route), the server inputs it into the optimized path planning model (based on the target routing policy configuration) to generate performance predictions for each route. For example, consider the order "Shanghai-Beijing, VIP customer, high-value electronic equipment (150kg, promised 24-hour delivery)": (I) Input Data: Fusion of Historical and Real-Time Information: The model inputs include: basic order information: shipping location (Shanghai), delivery location (Beijing), cargo type (high-value electronic equipment), weight (150kg), promised delivery time (24 hours), and customer type (VIP); historical contextual data: the route's performance for historically prioritized orders (high-priority orders selected through the attention mechanism) (e.g., historical average delivery time of 11 hours, cost of 720 yuan, availability rate of 0.05%, and on-time rate of 100%); and real-time environmental data: the day's congestion forecast for the Jinan section of the Beijing-Shanghai Expressway (possible delay of 0.5 hours between 10:00 and 12:00), weather (sunny), and the surge in truck traffic during the "Shopping Festival" (dispatching priority adjustment). (2) Model prediction: Multi-dimensional output based on target routing strategy: The model is based on the target routing strategy configuration (the core elements are "timeliness" and "cost", and the constraint elements are "intactness" and "punctuality", because the roles are swapped after the stable state), combined with the previously optimized weighted parameters (timeliness 0.4, cost 0.3) and compensation parameters (intactness threshold 0.07%, punctuality threshold 99%), and outputs the following predicted data: Transportation timeliness: The model combines the theoretical time of map navigation (10 hours), historical congestion correction (0.5-hour delay in the Jinan section), and VIP order priority scheduling (fast lane allocation, shortened by 0.3 hours), and predicts the actual timeliness to be 10h+0.5h-0.3h=10.2 hours (≤ the promised 24 hours). Transportation Cost: Based on the route length (1,200 km), unit cost (0.6 yuan / km), weight factor (1.2 for 150 kg), and high-value insurance premium (30 yuan), the standardized cost is calculated as 1,200 × 0.6 × 1.2 + 30 = 894 yuan. (The new baseline cost configured for the target routing policy is 750 yuan. The actual predicted cost of 894 yuan may exceed the baseline and needs to be considered in conjunction with compensation parameters.) Cargo Integrity: Analyzing the proportion of high-speed routes (90%, indicating minimal bumps), packaging protection (shockproof box, protection factor 1.5), and historical damage rate for similar orders (0.05%), the predicted damage rate is 0.05% / 1.5 = 0.033%, indicating a 99.967% intact rate, which exceeds the constraint threshold of 0.07%. On-time Delivery Rate: Given a predicted delivery time of 10.2 hours ≤ 24 hours and a stable historical on-time rate of 100% for similar orders, the predicted on-time rate is 99.8% (exceeding the constraint threshold of 99%).

[0071] The server must integrate the predicted data from these four directions into comprehensive performance data to quantify the quality of the path. This integration rule is based on the weighted parameters of the core / constraint elements configured in the target routing policy (0.4 for timeliness and 0.3 for cost), the compensation parameters (0.07% for uptime threshold and 99% for on-time threshold), and the weights adjusted through meta-learning (due to the role swap, the core now becomes timeliness and cost, while the constraints become uptime and on-time). (I) Calculation of fitness scores for core elements (timeliness and cost): Timeliness score: The deviation between the predicted value of 10.2 h and the baseline value of 11.2 h is -1.0 h (better), the high-dimensional representation similarity is 0.98 (small deviation), the weighting parameter is 0.4, and the score is 0.98 × 0.4 = 0.392; Cost score: The deviation between the predicted value of 894 yuan and the baseline value of 750 yuan is +144 yuan (exceeding the baseline), the high-dimensional representation similarity is 0.75 (large deviation), the weighting parameter is 0.3, and the score is 0.75 × 0.3 = 0.225 (due to exceeding the baseline, correction with compensation parameters is required). (2) Calculation of fitness scores for constraint elements (uptime rate, punctuality rate): Uptime rate score: The deviation between the predicted value of 0.033% and the constraint threshold of 0.07% is -0.037% (meets the standard), the compensation parameter is 0, the high-dimensional representation similarity is 0.99, the weighting parameter is 0.2, and the score is 0.99×0.2+0=0.198; Punctuality score: The deviation between the predicted value of 99.8% and the constraint threshold of 99% is +0.8% (meets the standard), the compensation parameter is 0, the high-dimensional representation similarity is 0.99, the weighting parameter is 0.1, and the score is 0.99×0.1+0=0.099. (3) Comprehensive performance integration data: The total fitness score is the sum of the core and constraint scores: 0.392 (timeliness) + 0.225 (cost) + 0.198 (uptime rate) + 0.099 (punctuality rate) = 0.914 (≥ dynamic optimization threshold 0.9, met the standard).

[0072] The server combines the integrated performance data with the path planning results to generate visual push content and display it to the logistics scheduling system and customers: (1) Basic path information and core advantages: Recommended path: Map visualization (Shanghai-G2 Beijing-Shanghai Expressway-Jinan (avoiding congested sections)-Beijing), with key nodes marked (Jinan Service Area as an alternative transit point); Core advantages: High timeliness: predicted 10.2 hours (better than the benchmark 11.2 hours, the result of core element optimization); High integrity rate: 99.967% (better than the constraint threshold of 0.07%, due to 90% of the path being expressway + shockproof packaging); On-time rate guarantee: 99.8% (better than the constraint threshold of 99%, VIP orders are prioritized for scheduling). (II) Constraint indicators and risk warnings: Cost explanation: The predicted cost is RMB 894 (slightly exceeding the benchmark of RMB 750). This is due to the high volume of truck traffic during the "Shopping Festival", which resulted in an increase in tolls (RMB 0.6 / km to RMB 0.65 / km). However, the additional costs have been controlled by optimizing the route length (shortening it by 50 km); Risk warning: The Jinan section may be congested from 10:00 to 12:00 (delayed by 0.5 hours), but buffer time is reserved (predicted timeliness 10.2 hours ≤ 24 hours), and delivery can still be made on time (technical basis: the same-region timeliness indicator in the historical route efficiency factor shows that the impact of congestion in the Jinan section on the total timeliness is ≤ 0.5 hours). (3) Basis for strategy optimization: Dynamic evolution support: The target routing strategy configuration is optimized through multiple rounds of reinforcement learning. The weighted parameters of the core elements (timeliness and cost) are dynamically adjusted through meta-learning (0.4 and 0.3 in the first round, and subsequently significantly increased to 0.45 due to timeliness optimization), ensuring that resources are tilted towards key indicators; Attention mechanism screening: The historical path efficiency factor is generated based on preferred orders (high-value goods account for 85%), and the predicted data is more in line with the core business needs of the "Shopping Festival"; Role swap verification: After the stable state is triggered, the original core elements (intact rate, punctuality rate) are converted into constraints to ensure that the strategy does not reduce service quality while optimizing the new core (timeliness and cost) (intact rate ≥ 0.07%, punctuality rate ≥ 99%).

[0073] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned artificial intelligence-based logistics path planning method and system. Figure 2 As shown, Figure 2 This is a block diagram of the structure of a computer device 100 provided in an embodiment of the present invention. Computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or exchange, memory 111, processor 112, and communication unit 113 are electrically connected to each other, directly or indirectly. For example, these components can be electrically connected via one or more communication buses or signal lines.

[0074] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. These embodiments have been selected and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the present disclosure and to utilize various embodiments with various modifications as appropriate for the specific application contemplated.

Claims

1. The logistics path planning method based on artificial intelligence is characterized by: include: Based on the performance prediction data of the designated path performance direction corresponding to each historical logistics order, a baseline path performance factor of the path planning model is obtained; wherein each historical logistics order is associated with multiple path performance directions; Each historical logistics order is used as a reference logistics order. Combined with the benchmark path efficiency factor, dynamic path evolution is performed through a reinforcement learning framework. The dynamically evolved transition routing strategy configuration is used as a candidate routing strategy configuration. Each round of evolution includes: For each basic routing policy configuration to be evolved, the following process is performed in sequence: Based on the performance integration data of multiple path performance directions of each reference logistics order obtained from a basic routing policy configuration, a historical path performance factor is obtained, and combined with the baseline path performance factor to obtain the corresponding path fitness score; The basic routing policy configuration whose path fitness score reaches the dynamic optimization threshold is used as the transition routing policy configuration, and each new basic routing policy configuration is selected; Based on the performance integration data of multiple path performance directions of each real-time logistics order itself obtained by the selected routing strategy configuration, a new benchmark path performance factor is obtained, and combined with each new reference logistics order composed of each real-time logistics order, the dynamic path evolution is continuously performed through the online learning mechanism to obtain the target routing strategy configuration; wherein, the target routing strategy configuration is used to obtain the path planning decision of the path planning model for the pending logistics path, so as to push and display the pending logistics path.

2. The method according to claim 1, characterized in that The reference path performance factor includes a plurality of reference path performance elements, the historical path performance factor includes a plurality of historical path performance elements, and each reference path performance element corresponds to one historical path performance element; In each round of dynamic path evolution, the corresponding path fitness score is obtained based on the obtained historical path effectiveness factor and the benchmark path effectiveness factor, including: For the multiple historical path effectiveness elements in the historical path effectiveness factor, the following processes are performed in sequence: Extracting features of the historical path effectiveness elements to obtain a high-dimensional representation; determining a corresponding path fitness score based on a deviation of a high-dimensional representation of a historical path effectiveness element from a high-dimensional representation of a corresponding baseline path effectiveness element in the baseline path effectiveness factor; Based on the path fitness scores respectively corresponding to the multiple historical path effectiveness elements, a corresponding path fitness score is obtained.

3. The method according to claim 1, characterized in that The method further comprises: Obtaining a comparative path effectiveness factor of a comparative path planning model; wherein the comparative path planning model is another path planning model that has been actually deployed, and the comparative path effectiveness factor includes a plurality of comparative path effectiveness elements, and each comparative path effectiveness element corresponds to a reference path effectiveness element in the new reference path effectiveness factor; For the multiple reference path performance elements in the new reference path performance factor, the following processes are performed in sequence: Obtaining a reference path performance element and a deviation between the reference path performance element and the corresponding comparison path performance element in the comparison path performance factor; Based on the Pareto front search, the historical path performance elements corresponding to the benchmark path performance elements whose deviations do not meet the deviation conditions and whose contributions to the total fitness are in the top K% are used as core performance elements, and the historical path performance elements corresponding to the benchmark path performance elements whose deviations meet the deviation conditions and have mandatory requirements for business constraints are used as performance constraint elements. In each round of the dynamic path evolution, a corresponding path fitness score is obtained based on the historical path effectiveness factor and the benchmark path effectiveness factor, including: For the core performance element, dynamically optimizing weighting parameters through meta-learning, and determining a path fitness score for the core performance element based on the weighting parameters of the core performance element and a deviation between the core performance element and a corresponding benchmark path performance element; For the performance constraint element, determining a path fitness score of the performance constraint element based on a weighting parameter and a compensation parameter of the performance constraint element, and a deviation between the performance constraint element and a corresponding benchmark path performance element; Based on the path fitness scores of the core performance elements and the path fitness scores of the performance constraint elements, a corresponding path fitness score is obtained.

4. The method according to claim 3, characterized in that When there are multiple core performance elements, in the dynamic path evolution, the weighted parameter of each core performance element in the first round of evolution is the second value. In each round of evolution after the first round of evolution, the weighted parameter of a core performance element is obtained by the following process, including: Based on the deviation between a core performance element in this round of evolution and the core performance element in the previous round of evolution, the sum of the deviations of all core performance elements in this round of evolution from the corresponding core performance element in the previous round of evolution, and the weighted parameter of the core performance element in the previous round of evolution, the weighted parameter of the core performance element in this round of evolution is obtained.

5. The method according to claim 3, characterized in that The number of the performance constraint elements is at least one, and the compensation parameter of each performance constraint element is obtained by the following process, including: For each performance constraint element, a compensation parameter of the performance constraint element is obtained based on a deviation between the performance constraint element and a corresponding reference path performance element and a dynamic loading rule between the deviation threshold.

6. The method according to claim 5, characterized in that During the continuous dynamic path evolution, when the transition routing policy configuration of the current evolution round and the transition routing policy configuration of the previous evolution round reach a stable state, the following process is performed in each round starting from the next evolution round until a stable state is reached again: For each basic routing policy configuration to be evolved, perform the following steps in sequence: Based on the performance integration data of multiple path performance directions of each reference logistics order obtained by a basic routing strategy configuration, a historical path performance factor is obtained; The core efficiency elements in the previous round of evolution are used as efficiency constraint elements in the historical path efficiency factors of this round of evolution, and the efficiency constraint elements in the previous round of evolution are used as the core efficiency elements in the historical path efficiency factors of this round of evolution; Based on the deviations between the core performance elements and performance constraint elements in this round of evolution and the corresponding benchmark path performance elements in the new benchmark path performance factors, a corresponding path fitness score is obtained; The basic routing policy configuration whose path fitness score reaches the dynamic optimization threshold is used as the transition routing policy configuration, and each new basic routing policy configuration is selected.

7. The method according to claim 6, characterized in that During the continuous dynamic path evolution, when the transition routing policy configuration of the current evolution round and the transition routing policy configuration of the previous evolution round reach a stable state, starting from the next evolution round, the deviation threshold in each evolution round is obtained by the following process, including: The deviation threshold is obtained based on the tuning frequency of the core performance element and the performance constraint element in multiple rounds of evolution; wherein, in multiple rounds of evolution, the deviation threshold decreases with the evolution rounds.

8. The method according to claim 1, characterized in that The historical path efficiency factor is obtained by integrating the efficiency data of multiple path efficiency directions of each reference logistics order itself based on a basic routing strategy configuration, including: Based on the performance integration data of multiple path performance directions of each reference logistics order obtained by a basic routing strategy configuration, the priority weight of each reference logistics order is calculated through an attention mechanism, and each preferred logistics order is selected from each reference logistics order based on the priority weight; Based on the performance prediction data and feature identifiers of multiple path performance directions of each preferred logistics order itself, the historical path performance factor is obtained, wherein the historical path performance factor includes one of the following historical path performance elements: the total transportation time efficiency index of each preferred logistics order, the transportation time efficiency index of the same-region path, the total transportation cost index, the cargo integrity rate index and the on-time delivery rate index.

9. The method according to claim 1, characterized in that The method further comprises: Inputting the undetermined logistics path into the path planning model to obtain performance prediction data of multiple path performance directions corresponding to the undetermined logistics path; Based on the target routing strategy configuration, integrating the performance prediction data of multiple path performance directions corresponding to the undetermined logistics path to obtain performance integration data corresponding to the undetermined logistics path; Based on the efficiency integration data corresponding to the pending logistics path, the pending logistics path is pushed and displayed.

10. A server system, characterized in that: The method comprises a server, wherein the server is configured to execute the method according to any one of claims 1 to 9.

Images