A method and system for dynamic optimization scheduling of a round-corner container transportation path

Abstract

The application provides a rounded container transportation path dynamic optimization scheduling method and system, quantizes three types of constraints of rounded container size, barycenter and loading and unloading priority, converts into matching formula, path constraint threshold and weight rule, combines transportation cost, time, cargo loss risk to construct a weighted multi-objective optimization function; a hybrid framework of genetic algorithm and ant colony algorithm is built, a constraint adaptation layer is embedded to filter invalid solutions and improve iteration efficiency; two types of algorithm operators are improved, a constraint satisfaction degree, loading and unloading priority and dynamic parameter adjustment mechanism are integrated; a divide-and-conquer strategy is used to divide multiple regions into sub-regions for optimization, and then a large-scale dynamic scene is adapted through cross-region cooperation and local re-optimization; in iteration, a full-dimensional check scheduling scheme is performed, algorithm parameters are automatically adjusted based on constraint violation information, and iteration is terminated or constraint relaxation is started according to preset conditions; the improved algorithm is integrated into a dynamic scheduling system, real-time data is connected, parameters are optimized through a self-learning module, and a manual intervention interface is reserved.

Description

Technical Field

[0001] This invention relates to the field of dynamic optimization scheduling technology for transportation routes, and more specifically, to a method and system for dynamic optimization scheduling of rounded corner container transportation routes. Background Technology

[0002] Round-corner containers are increasingly used in cold chain transportation and high-value cargo transshipment due to their structural design advantages. However, their rounded corner structure leads to significant differences in size matching and center of gravity distribution compared to traditional right-angle containers, posing special requirements for vehicle type adaptation, route selection, and loading / unloading processes. Currently, traditional container transportation scheduling methods are mostly designed based on the characteristics of right-angle containers, failing to fully consider the core constraints of round-corner containers, such as their corner radius and center of gravity shift. They often simplify requirements such as size adaptation, route slope / turning radius limitations, and loading / unloading priorities into qualitative rules, lacking quantitative matching formulas, constraint thresholds, and weighting systems. This makes it difficult to accurately integrate them into the algorithm optimization process, resulting in scheduling schemes that are prone to problems such as wasted loading space, route violations, and increased risk of cargo damage.

[0003] Existing scheduling algorithms mostly employ single genetic algorithms or ant colony algorithms, failing to optimize operators for the multi-constraint characteristics of rounded-corner containers. This results in a high proportion of invalid solutions and slow convergence during iteration. When facing large-scale dynamic scheduling scenarios across multiple regions, they lack effective divide-and-conquer strategies and cross-regional collaboration mechanisms, making it difficult to adapt to dynamic factors such as container increases, capacity changes, and real-time road condition fluctuations. Furthermore, the balancing logic of cost, time, and cargo damage risk in multi-objective optimization is rigid, unable to dynamically adjust parameters based on constraint satisfaction. In addition, traditional scheduling systems rely heavily on human experience, with insufficient integration between algorithms and real-time data, and a lack of self-learning capabilities. This makes it difficult to balance compliance, efficiency, and economy in scheduling solutions, failing to meet the refined and dynamic scheduling needs of rounded-corner container transportation. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for dynamic optimization scheduling of rounded corner container transportation routes, the method comprising:

[0005] The three types of constraints—rounded corner container size, center of gravity, and loading / unloading priority—are quantified and transformed into matching formulas, path constraint thresholds, and weighting rules. A weighted multi-objective optimization function is then constructed by combining transportation costs, time, and cargo damage risk.

[0006] A hybrid framework combining genetic algorithm and ant colony algorithm is built, and a constraint adaptation layer is embedded to filter invalid solutions and improve iteration efficiency;

[0007] Two types of algorithm operators are improved by incorporating constraint satisfaction, loading / unloading priority, and dynamic parameter adjustment mechanisms.

[0008] The divide-and-conquer strategy is adopted to divide the region into multiple sub-regions for optimization, and then cross-regional collaboration and local re-optimization are used to adapt to large-scale dynamic scenarios.

[0009] The iterative full-dimensional verification scheduling scheme automatically adjusts algorithm parameters based on constraint violation information and terminates the iteration or initiates constraint relaxation according to preset conditions.

[0010] The improved algorithm is integrated into the dynamic scheduling system, connected to real-time data, and the parameters are optimized through a self-learning module, while retaining the interface for manual intervention.

[0011] Furthermore, embodiments of the present invention also provide a dynamic optimization scheduling system for the transportation path of rounded corner containers, characterized in that it includes:

[0012] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described dynamic optimization scheduling method for rounded corner container transportation routes by executing the machine-executable instructions.

[0013] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described dynamic optimization scheduling method for rounded corner container transportation paths.

[0014] Based on the above, by quantifying three core constraints—rounded corner container dimensions, center of gravity, and loading / unloading priority—a weighted multi-objective optimization function is constructed. This transforms qualitative requirements into computable quantitative rules, significantly improving the adaptability of the scheduling scheme to the special attributes of rounded corner containers. A genetic-ant colony hybrid framework embeds a constraint adaptation layer, filtering invalid solutions beforehand. Combined with improved algorithm operators and a dynamic parameter adjustment mechanism, this significantly reduces the proportion of invalid solutions during iteration, accelerates iteration convergence, and shortens the solution generation time, effectively addressing the pain points of low efficiency and poor adaptability of traditional algorithms.

[0015] By leveraging a divide-and-conquer strategy and a cross-regional collaborative mechanism, this invention efficiently adapts to large-scale dynamic scheduling scenarios. The combination of sub-regional optimization and local re-optimization enables rapid response to dynamic factors such as container volume increases and road condition changes, ensuring smooth cross-regional transportation connections. A comprehensive verification and constraint relaxation mechanism balances compliance and flexibility. The system integrates real-time data connectivity, a self-learning module, and a manual intervention interface, achieving automation, intelligence, and controllability in scheduling optimization. Ultimately, it achieves a comprehensive optimization of transportation costs, efficiency, and cargo damage risk, fully meeting the refined and dynamic scheduling needs of rounded-corner container transportation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the execution flow of the dynamic optimization scheduling method for rounded corner container transportation routes provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of exemplary hardware and software components of the rounded corner container transportation path dynamic optimization scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a dynamic optimization scheduling method for rounded corner container transportation routes provided in an embodiment of the present invention. The following is a detailed description of this dynamic optimization scheduling method for rounded corner container transportation routes.

[0019] Step S110: Quantify the three types of constraints of rounded corner container size, center of gravity, and loading and unloading priority, and transform them into matching formulas, path constraint thresholds and weight rules. Combine transportation costs, time and cargo damage risk to construct a weighted multi-objective optimization function.

[0020] For the scheduling scenario of 12m rounded corner refrigerated containers (for transporting Chilean cherries) from coastal ports to inland cities, three types of constraints are quantified: container size, center of gravity, and loading / unloading priority. A size matching formula is derived by combining the compatibility of the container with the Dongfeng Tianlong KL truck. The path slope / turning radius threshold is set according to vehicle dynamics. The loading / unloading priority weight rules are clarified through expert scoring. Finally, with the goals of lowest cost, shortest time, and lowest risk, a normalized weighted comprehensive optimization function is constructed to achieve the combination of constraint quantification and multi-objective optimization, providing a core model for scheduling scheme design.

[0021] Step S111: Obtain the key dimensions and rounded corner radius of the rounded corner container through 3D scanning, collect the cargo box dimensions, chamfer radius and maximum load of the transport vehicle, obtain the center of gravity related data of the container under different loading conditions through finite element mechanical simulation, and statistically analyze the cargo value, preservation requirement level, loading and unloading difficulty coefficient and emergency delivery level of the container.

[0022] The key dimensions of a 12m rounded corner refrigerated container (rounded corner radius 0.3m) and a Dongfeng Tianlong KL cargo box were obtained using a FARO 3D scanner. The coordinates of the center of gravity and the offset limit (lateral ≤200mm) of the container under full load / half load conditions were obtained through ANSYS finite element simulation. The core information of the container (value of goods 500,000 yuan, preservation grade A, loading and unloading difficulty 0.8, emergency delivery level 1) was collected. The data on dimensions, center of gravity, and priority were collected in a complete manner, providing accurate basic support for subsequent constraint quantification and function construction.

[0023] Step S112: Based on the size data of the container and the cargo box, calculate the effective occupancy and capacity of both, determine a reasonable limit on the number of containers loaded, and clarify the placement rules to avoid rounded corners facing the corners of the cargo box by defining the placement method;

[0024] Calculate the effective carrying length of the cargo box (12000mm) to determine that only one target container can be loaded; establish placement constraints (the distance between the rounded corners of the container and the corners of the cargo box must be ≥100mm, and direct alignment is prohibited), and specify the specific placement plan (1000mm from the front of the cargo box, 600mm from the rear, and 31mm from each side, with the rounded corners facing the length direction); through quantitative calculations and rule definitions, solve the size compatibility problem between the container and the cargo box, and ensure the rational utilization of loading space and structural safety.

[0025] Step S113: Combining the container center of gravity offset data with the principles of vehicle driving dynamics, determine the maximum allowable slope and minimum allowable turning radius of the path, and clarify the slope and turning radius constraint rules that the path must meet;

[0026] Combining the container's center of gravity offset characteristics with vehicle driving dynamics, a simulation model was established using MATLAB / Simulink. The simulation showed that the vehicle was stable when the center of gravity offset of the fully loaded container was ≤200mm. Based on this, the maximum allowable slope of the route was set to 15° and the minimum allowable turning radius to be 18°. Route selection rules were formulated (excluding road sections with slope >15° and turning radius <18m, and prioritizing road sections with slope ≤10° and turning radius ≥20m), clarifying the route constraint thresholds and selection criteria to ensure vehicle stability during transportation.

[0027] Step S114: Using a weighted summation method, set the weight coefficients corresponding to cargo value, preservation requirements, loading and unloading difficulty, and emergency delivery, calculate the priority weights of container loading and unloading and normalize them, and clarify the loading and unloading order of containers loaded on the same vehicle in descending order of priority weight. When the priorities are the same, they are sorted in order of emergency delivery level and cargo value.

[0028] A scoring method using 5 logistics experts was adopted, and weight coefficients were set for cargo value (0.3), freshness requirements (0.4), loading and unloading difficulty (0.1), and urgent delivery (0.2). The loading and unloading priority weight of the target container was calculated to be 0.98 (after normalization) using a formula. A sorting rule was established (containers in the same vehicle are arranged in descending order of priority, and if the weights are the same, they are sorted by urgent delivery level and cargo value in turn), and the container was determined as the priority loading and unloading target in the same batch to ensure the timeliness of urgent fresh goods.

[0029] Step S115: Construct a total transportation cost function, which includes fuel consumption, labor and vehicle depreciation costs; construct a total transportation time function, which includes transportation travel time and the total loading and unloading time of all containers; and construct a cargo damage risk coefficient function, which comprehensively considers the cargo damage risks caused by center of gravity shift, excessive path slope, insufficient turning radius and preservation requirements. In the case of exceeding the standard, the corresponding risk item is calculated as extremely high risk.

[0030] The total transportation cost includes fuel consumption (245 yuan), labor (600 yuan), and depreciation (219.18 yuan), totaling 1064.18 yuan; the total time includes driving (1.67 hours) and loading and unloading (1 hour), totaling 2.67 hours; the risk of cargo damage is calculated based on factors such as center of gravity shift, slope, turning radius, and preservation factors, with a risk coefficient of 0.04 (the risk is considered extremely high when the slope exceeds the standard, and the coefficient rises to 0.34); through quantitative modeling, single-item objective functions of cost, time, and risk are formed, providing calculable basic indicators for comprehensive optimization.

[0031] Step S116: Construct a hierarchical structure model with comprehensive optimization as the objective and total transportation cost, total time, and cargo damage risk as criteria. Invite experts in relevant fields to conduct pairwise comparisons of the criteria layer objectives and construct a judgment matrix. After consistency verification, determine the weight of each objective and construct a normalized weighted comprehensive multi-objective optimization function.

[0032] A hierarchical model was constructed (objective layer: comprehensive optimization; criterion layer: cost, time, risk; solution layer: container / vehicle type / route / loading and unloading sequence), and a team of 7 experts (logistics + vehicle + supply chain fields) was assembled. A judgment matrix was constructed using the 1-9 scale method, and the criterion weights (cost 0.637, time 0.258, risk 0.105) were determined after consistency testing (CR=0.033<0.1). A weighted comprehensive optimization function was constructed, and the function value F=0.406 for path 1 was calculated (better than 0.32 for path 2). The optimal scheduling scheme was selected to achieve multi-objective balanced optimization.

[0033] Step S1161: Clarify the boundaries of the hierarchical structure, set comprehensive optimization as the target layer, focus on the balance of cost, efficiency and risk, define the three indicators of total transportation cost, total time and cargo damage risk and their respective connotations in the criteria layer, and cover the scheduling schemes related to containers, vehicle combinations, route planning and loading and unloading sequence in the solution layer to ensure that the hierarchical elements are clear and have no overlap.

[0034] The objective layer focuses on comprehensive optimization (balancing cost, efficiency, and risk), the criteria layer defines total transportation cost, total time, cargo damage risk, and their respective implications, and the solution layer covers scheduling schemes related to containers, vehicle combinations, route planning, and loading and unloading sequences, ensuring that the elements at each level are clear and non-overlapping, thus laying a structural foundation for subsequent multi-objective optimization.

[0035] Step S1162: Assemble an expert team covering multiple related fields and possessing corresponding experience, and explain to the experts in advance the constraints and influencing factors of rounded corner container transportation guidelines to ensure a consistent understanding of the assessment background;

[0036] A team of seven multidisciplinary experts (including logistics scheduling, vehicle engineering, and supply chain experts) was formed. The experts were informed in advance about the constraints and influencing factors of rounded corner container transportation, and a unified understanding of the background was established to ensure the professionalism and consistency of the subsequent assessment of the importance of the criteria.

[0037] Step S1163: Use the 1-9 scale method to formulate pairwise comparison rules, clarify the importance levels corresponding to different scales and the use of the intermediate values ​​of adjacent scales, and determine the reciprocity specification of the scales;

[0038] The 1-9 scale method is used to formulate pairwise comparison rules, clarify the importance levels corresponding to the 1-9 scale, specify the use scenarios of the intermediate values ​​of adjacent scales such as 2 and 4, and establish scale reciprocity norms (e.g., if A is 3 for B, then B is 1 / 3 for A), providing a unified standard for expert scoring.

[0039] Step S1164: Construct a judgment matrix based on the criteria layer indicators, organize experts to compare and score the importance of each criterion pairwise, and determine the final elements of the matrix after collecting the scores, calculating the mean and adjusting for outliers.

[0040] Using cost, time, and risk as criteria indicators, experts were organized to compare the importance of each pair and score them. After collecting the scores, extreme outliers were removed, the mean was calculated, and finally the elements of the judgment matrix were determined to form standardized criteria importance comparison data.

[0041] Step S1165: Perform a consistency check on the judgment matrix. Determine whether the consistency requirements are met by iteratively solving the eigenvectors, calculating the consistency index and proportion. If not, provide feedback to the experts to adjust the score until the standard is met.

[0042] A consistency check is performed on the judgment matrix: the largest eigenvalue is solved by iteratively solving the eigenvectors, and the consistency index CI and the ratio CR are calculated. After verification, CR = 0.033 < 0.1, which meets the consistency requirements, and no feedback is needed for expert adjustment of the score.

[0043] Step S1166: Extract the eigenvectors of the judgment matrix that have passed the test as initial weights, and obtain the weights of each criterion after normalization to ensure that the weight distribution is reasonable and the sum is 1;

[0044] The eigenvectors of the judgment matrix that passed the consistency test were extracted as initial weights. After normalization, the criterion weights (cost 0.637, time 0.258, risk 0.105) were obtained. The weight distribution was reasonable and the sum was 1, which provided a basis for constructing the comprehensive optimization function.

[0045] Step S1167: Normalize each sub-objective, construct a comprehensive multi-objective optimization function by weighting according to the weights, and evaluate the quality of the scheduling scheme by the function value to achieve a comprehensive balance of cost, time and risk.

[0046] The sub-objectives of cost, time, and risk are normalized, and a comprehensive multi-objective optimization function is constructed by weighting according to the determined criteria. The quality of the scheduling scheme is judged by the magnitude of the function value, and finally a comprehensive balance of transportation cost, time, and risk is achieved.

[0047] Step S120: Build a hybrid framework of genetic algorithm and ant colony algorithm, embed a constraint adaptation layer to filter invalid solutions and improve iteration efficiency;

[0048] For the scheduling scenario of 100 rounded corner refrigerated containers from coastal ports to 5 inland cities, a genetic-ant colony hybrid framework was built: the genetic algorithm is responsible for the global allocation of multiple regions, multiple vehicle types, and multiple container combinations, while the ant colony algorithm is responsible for the path planning of a single combination. A constraint adaptation layer is embedded in the framework to connect the key stages of the algorithm, and to filter invalid solutions that violate size and center of gravity path constraints in real time, thereby improving iteration efficiency.

[0049] Step S121: The genetic algorithm is used as the global decision-making module, which is responsible for the global allocation of multi-region, multi-vehicle, and multi-rounded corner container combinations. The ant colony algorithm is used as the local decision-making module, which is responsible for the specific transportation route planning of a single container and vehicle combination. The global allocation scheme output by the genetic algorithm is used as the input premise of the ant colony algorithm. The path optimization results output by the ant colony algorithm are fed back to the genetic algorithm as the basis for its fitness calculation, forming a closed-loop interactive logic of global allocation, local optimization, and result feedback.

[0050] The genetic algorithm is defined as the global decision-making module, outputting the allocation schemes for containers, vehicle types, and regions. The ant colony algorithm is the local module, using the genetic output as input to plan specific paths. The ant colony feeds back the path cost and constraint satisfaction to the genetic algorithm as the basis for its fitness calculation, forming a closed-loop interactive logic of global allocation → local optimization → result feedback → global iteration.

[0051] Step S122: Build a two-layer architecture including a rule storage layer and a verification execution layer. The rule storage layer stores the size matching of rounded corner containers, the path slope / turning radius corresponding to the center of gravity, and the loading and unloading priority constraint rules according to the constraint type, verification logic, and violation judgment criteria. The verification execution layer designs a standardized verification interface, formulates the verification priority of verifying hard constraints first and then soft constraints, and filters invalid solutions that violate constraints.

[0052] The rule storage layer stores hard / soft constraint rules according to type, logic, and standard; the verification execution layer designs a standardized verification interface, sets the priority of hard verification first and soft verification later, and after receiving the algorithm solution, it first verifies the hard constraints and directly filters invalid solutions to ensure the compliance of the solution.

[0053] Step S123: Using a segmented chromosome encoding method of region segment, vehicle type segment, container combination segment, and loading / unloading node segment, after randomly generating the initial population, the initial solution is pre-filtered by calling hard constraint rules through the constraint adaptation layer to remove solutions that violate size matching and centroid path constraints. Genetic algorithm iteration is executed according to selection, crossover, and mutation. After generating a new solution in each iteration, it is passed to the constraint adaptation layer to complete soft constraint verification and record the constraint satisfaction. The fitness value of the solution is calculated based on the constraint satisfaction.

[0054] A segmented chromosome encoding method (region / vehicle type / container combination / loading / unloading node segment) is used to generate an initial population of 200. 32 invalid solutions are filtered out by a constraint adaptation layer. Through roulette wheel selection, crossover (probability 0.8), and mutation (probability 0.05) iterations, each new solution is verified by the constraint adaptation layer, and its satisfaction is recorded. The fitness value is calculated based on this index.

[0055] Step S124: Based on the global allocation scheme output by the genetic algorithm, initialize the parameters and search space of the ant colony algorithm, embed the path constraint rules in the constraint adaptation layer into the node connectivity judgment logic, mark the node connections that cannot be passed, optimize the pheromone initialization rules of the ant colony algorithm, increase the initial pheromone concentration of the path nodes that have passed the constraint adaptation layer verification, and call the verification interface of the constraint adaptation layer in real time during the state transition of the ant colony algorithm. If the path segment violates the constraint, terminate the path search of the corresponding ant.

[0056] Ant colony parameters are initialized based on a global genetic scheme, path constraints are embedded in node connectivity determination, and impassable road segments are marked. Pheromones are optimized to increase the initial concentration of compliant nodes. During ant colony state transitions, a verification interface is called in real time. If a path segment violates constraints, the search of the corresponding ant is immediately terminated to ensure path compliance.

[0057] Step S125: Establish a unified scheduling module to manage the algorithm iteration process. First, trigger the genetic algorithm to generate an initial global allocation solution. After filtering by the constraint adaptation layer, execute the iteration. When the genetic algorithm converges to the preset threshold, extract the high-quality global allocation scheme and pass it to the ant colony algorithm. After the ant colony algorithm completes the local path planning, it is verified by the constraint adaptation layer. The path cost and constraint satisfaction data are fed back to the genetic algorithm to update its fitness function. Repeat the iteration until the global solution and the local path both satisfy the constraints and the optimization objective reaches the preset requirements. Design a dynamic update mechanism for the constraint adaptation layer to automatically update the rules of the rule storage layer and the verification logic of the verification execution layer when the constraint rules are adjusted.

[0058] A unified scheduling module is built to manage iterations: after genetic convergence, the best solution is passed to the ant colony, and the ant colony updates the genetic fitness function with feedback data, repeating the iteration until the target is met. A dynamic update mechanism for the constraint adaptation layer is designed, which automatically updates the rule storage layer and the verification execution layer when a new constraint is added, without requiring a restart of the algorithm framework.

[0059] Step S126: Set a two-layer iteration termination condition where the constraint satisfaction reaches 100% and the comprehensive optimization function value is stable in the preset optimal range for multiple consecutive generations. Compare and test the hybrid algorithm connected to the constraint adaptation layer with the traditional hybrid algorithm, and count the number of iterations, the proportion of invalid solutions, and the time taken to generate the solution. If the efficiency improvement does not meet expectations, adjust the verification priority of the constraint adaptation layer or the algorithm iteration parameters.

[0060] Set a two-level iteration termination condition: constraint satisfaction is 100% and the comprehensive optimization function value remains stable in the optimal range for 20 consecutive iterations. Compare and test the hybrid algorithm with the traditional algorithm, and statistically analyze indicators such as the number of iterations and the proportion of invalid solutions; if the efficiency does not meet expectations, adjust the verification priority or algorithm parameters to ensure the optimization effect.

[0061] Step S130: Improve the two types of algorithm operators by incorporating constraint satisfaction, loading / unloading priority, and dynamic parameter adjustment mechanisms;

[0062] For the scheduling scenario of 100 round-corner refrigerated containers from coastal ports to 5 inland cities, the genetic and ant colony algorithm operators are improved: constraint satisfaction and loading / unloading priority are incorporated into the genetic operator, and path constraint satisfaction and priority matching are embedded into the ant colony operator; a dynamic parameter adjustment mechanism is built to associate operator parameters with constraint violation data, so as to realize the adaptive optimization of parameters according to the constraint satisfaction status, thereby improving the algorithm's adaptability to scheduling constraints.

[0063] Step S131: Comprehensively quantify the satisfaction of hard and soft constraints to form an index, adjust the fitness of the solution and eliminate substandard solutions, limit the crossover range according to segmented coding, lock the high-priority container coding segment, verify the constraints after crossover, set the mutation probability according to the loading and unloading priority, adjust the mutation direction in combination with the constraint satisfaction and verify and eliminate illegal solutions.

[0064] The quantification of hard / soft constraint satisfaction (overall satisfaction = hard constraint score × 0.7 + soft constraint score × 0.3) is incorporated into the genetic algorithm's fitness function to exclude solutions with an overall satisfaction score < 80. The crossover range is limited by segmented coding (only region / vehicle segment) to lock high-priority container coding segments. A lower mutation probability is set for high-priority containers, and after mutation, illegal solutions are checked and removed to enhance the genetic algorithm's adaptation to constraints and priorities.

[0065] Step S132: The pheromone update incorporates path constraint satisfaction and loading / unloading priority, penalizes the decay of illegal paths, sets and dynamically fine-tunes the evaporation coefficient in stages, reconstructs the heuristic function to incorporate constraint satisfaction and priority matching degree, and adjusts the weights in stages.

[0066] The ant colony algorithm incorporates path constraint satisfaction and loading / unloading priority into its pheromone update. It sets a 1.2-fold penalty decay coefficient for illegal paths and adjusts the evaporation coefficient in stages (ρ=0.1 for the first 50 generations, then adjusted to 0.08). It reconstructs the heuristic function (including path length, constraint satisfaction, and priority matching) and adjusts the weights in stages (focusing on path length for the first 50 generations, then strengthening constraints) to improve the constraint compliance of path optimization.

[0067] Step S133: Build a module to collect key data for iteration, formulate rules to adaptively adjust the parameters and constraint weights of the two types of algorithms, and record the adjustment process and effects;

[0068] A Python+MySQL iterative data acquisition module was built to collect data such as iteration count, constraint satisfaction, and algorithm parameters; parameter adjustment rules were formulated (e.g., if the hard constraint violation rate is >10%, the genetic crossover range is narrowed); the time, value, and constraint satisfaction changes of each parameter adjustment were recorded to form a parameter adjustment ledger to support the adaptive optimization of algorithm parameters.

[0069] Step S134: Connect the improved operator to the hybrid algorithm framework, connect the adaptation layer and the monitoring end, compare the indicators before and after the improvement, and backtrack to adjust the parameters until the target is met.

[0070] The improved operator is integrated into the hybrid algorithm framework, connecting the constraint adaptation layer and the monitoring end; the indicators before and after the improvement are compared (the number of iterations is reduced from 200 generations to 150 generations, and the proportion of invalid solutions is reduced from 35% to 10%); if the indicators are not met (such as the proportion of invalid solutions > 8%), parameters such as genetic mutation probability are adjusted backtracking until the scheduling optimization requirements are met.

[0071] Step S140: Use a divide-and-conquer strategy to divide the region into sub-regions for optimization, and then adapt to large-scale dynamic scenarios through cross-region collaboration and local re-optimization.

[0072] For the large-scale rounded container scheduling scenario radiating from a coastal port to five inland cities, a divide-and-conquer strategy is adopted: first, sub-regions are divided and locally optimized based on geographical, resource, and task balance; then, resources are allocated and the main routes are optimized through cross-regional collaborative allocation; finally, local re-optimization is initiated in response to dynamic changes to adapt to large-scale dynamic scheduling needs.

[0073] Step S141: Based on geographical correlation, resource matching degree and task balance, the cluster analysis method is used to divide the multiple regions into several sub-regions, clarify the boundary range, core scheduling nodes and resource endowment of each sub-region, and ensure that the scheduling tasks within the sub-regions are relatively independent and retain cross-regional collaborative interfaces.

[0074] K-means clustering (k=3) was used to divide the five cities into three sub-regions based on geographical relevance, resource matching degree, and task balance. The boundaries of each sub-region and core scheduling nodes (such as Shanghai Port) were clearly defined, and cross-regional data sharing API interfaces were retained to ensure that tasks within the sub-regions are independent and collaborative.

[0075] Step S142: For each sub-region, the improved genetic and ant colony hybrid algorithm is called to perform local optimization. The genetic algorithm is responsible for the global allocation of vehicle type and container combination within the sub-region, while the ant colony algorithm is responsible for the local path planning of a single container and vehicle type combination. During the optimization process, the constraint adaptation layer is used to filter invalid solutions. Based on the preset iteration termination condition that the constraint satisfaction reaches 100% and the optimization target is stable, a sub-region scheduling sub-scheme is generated.

[0076] For sub-region 1 (Shanghai and Suzhou), an improved hybrid algorithm is used for optimization: a genetic algorithm is used to allocate 20 cold chain vehicles and 50 containers, and an ant colony algorithm is used to plan the path; an access constraint adaptation layer is used to filter invalid solutions, and the iteration is terminated according to the condition that "constraint satisfaction is 100% + the fluctuation of the comprehensive optimization function value is less than 1% for 15 consecutive generations", generating a sub-region scheduling sub-scheme.

[0077] Step S143: Summarize all sub-regional scheduling sub-plans, analyze the vehicle capacity gap between sub-regions, cross-regional route congestion points, and container transfer connection bottlenecks through a unified data platform, and establish a classified ledger that marks the bottleneck type, scope of impact, and related sub-regions;

[0078] The three sub-regional solutions were summarized, and cross-regional bottlenecks (capacity gap in sub-region 2, congestion on G15 expressway, and transfer interruption at Wuxi hub) were analyzed through the Alibaba Cloud ECS unified data platform. A bottleneck classification ledger was established, marking the bottleneck type, scope of impact, and related sub-regions, providing a basis for cross-regional collaboration.

[0079] Step S144: Based on the bottleneck ledger, establish resource allocation rules between sub-regions, dispatch vehicle models to support the capacity gap in areas with surplus capacity, match the size and load constraints of rounded corner containers in the target area during allocation, formulate resource allocation priorities according to emergency delivery level and cargo damage risk, and record allocation path and occupation time.

[0080] Based on the bottleneck log, five Dongfeng Tianlong KL cold chain trucks were allocated from sub-region 1 to support sub-region 2. The trucks were matched with the container size and load constraints. Allocation priorities were set according to the emergency delivery level and cargo value, with priority given to supporting level 1 cherry containers. Allocation routes and capacity occupation time were recorded to ensure cross-regional capacity adaptation.

[0081] Step S145: Construct a cross-regional road network model with inter-regional transportation hubs as key nodes, embed the path constraints of rounded corner containers into the connectivity determination of road network nodes, call the ant colony algorithm to uniformly optimize the cross-regional trunk path, combine the time nodes of the sub-regional scheduling sub-schemes to plan the path connection sequence, and consider the real-time traffic conditions to dynamically adjust the path.

[0082] A cross-regional road network model is constructed with ports and hubs as key nodes, embedding container route constraints (slope ≤ 15°); the ant colony algorithm is used to optimize the main routes, and the connection sequence is planned in combination with the sub-regional scheduling time nodes; and the real-time traffic API is connected to avoid congested sections and dynamically adjust the routes to ensure smooth cross-regional transportation.

[0083] Step S146: Define two types of local re-optimization trigger conditions: dynamic changes within a sub-region and cross-regional collaborative feedback problems. Establish a trigger condition monitoring module to collect sub-regional scheduling data, traffic data, and order data in real time, and mark the affected sub-regions and scheduling tasks that meet the trigger conditions.

[0084] Define two types of local re-optimization trigger conditions (≥10 new containers added in a sub-region, and cross-regional capacity allocation delay ≥1 hour); build an MQTT monitoring module to collect order, road condition, and capacity data in real time, and mark the affected sub-regions and tasks (e.g., 10 new containers added in sub-region 3).

[0085] Step S147: Initiate a local re-optimization process for the marked affected sub-regions, call the hybrid algorithm to iteratively update the scheduling sub-scheme, reuse the original effective decisions, focus on adjusting the parts affected by dynamic changes, and quickly verify the new scheme through the constraint adaptation layer;

[0086] For the tasks affected in sub-region 3, a local re-optimization was initiated, reusing 80% of the original effective decisions (such as vehicle type-cargo container combination), and only adjusting the allocation and path of newly added cargo containers; the constraint adaptation layer completed the verification within 5 minutes, eliminated 2 illegal solutions, and generated an updated sub-solution.

[0087] Step S148: Synchronize the locally re-optimized sub-regional scheduling sub-schemes to the unified data platform, verify the rationality of cross-regional resource allocation and the smoothness of path connection. If the cross-regional bottleneck changes, dynamically adjust the coordination strategy, re-execute resource allocation and backbone path optimization to form a closed loop.

[0088] The sub-regional triple optimization plan was synchronized to a unified data platform to verify the rationality of cross-regional capacity coordination; after discovering congestion on the S25 expressway, the coordination strategy was adjusted (by changing to the G60 expressway), and the cross-regional trunk route was re-optimized to ensure the timing of container transfers.

[0089] Step S149: Establish a dynamic adjustment mechanism for sub-regions, which can redefine boundaries or adjust resource allocation based on task increments and resource changes, establish information sharing channels between sub-regions, synchronize capacity status, container turnover progress and constraint rule updates in real time, and regularly evaluate and optimize the rationality of sub-region division and the effectiveness of collaborative strategies.

[0090] Establish a dynamic adjustment mechanism for sub-regions (if the increase in containers exceeds 20 units, a new region will be created), establish an information sharing channel through MQTT to synchronize transportation capacity and container progress in real time; evaluate the rationality of sub-region division every month, and adjust the clustering k value if the iteration convergence is slow to optimize task balance.

[0091] Step S1410: Compare the scheduling scheme generation time, iteration convergence speed, cross-regional transportation connection efficiency, constraint satisfaction rate and other indicators before and after the implementation of the divide-and-conquer strategy. If the sub-region optimization effect is not good or there are breakpoints in cross-regional collaboration, adjust the sub-region division granularity, hybrid algorithm iteration parameters or collaborative allocation rules until the scheduling needs of large-scale dynamic scenarios are met.

[0092] Compare the metrics before and after implementing the divide-and-conquer strategy (time reduced from 40 min to 20 min, constraint satisfaction rate increased from 90% to 98%); if the constraint satisfaction rate of sub-region 2 does not reach 95%, adjust the sub-region partitioning granularity or the crossover probability of the genetic algorithm until it adapts to the needs of large-scale dynamic scheduling.

[0093] Step S150: Verify the scheduling scheme in all dimensions during iteration, automatically adjust algorithm parameters based on constraint violation information, and terminate the iteration or start constraint relaxation according to preset conditions;

[0094] For the scheduling iteration process of 100 rounded corner containers, a closed-loop framework of "full-dimensional verification - constraint violation analysis - parameter adjustment - termination / relaxation" is built: during the iteration, hard / soft / cooperative constraints are verified, algorithm parameters are adjusted based on violation information, the iteration is terminated when the conditions are met, and constraint relaxation is initiated in extreme cases.

[0095] Step S151: Verify the core dimensions, which cover hard constraints, soft constraints, and cross-regional collaborative constraints. Hard constraints include size matching, compliance of path slope / turning radius corresponding to the center of gravity, and load adaptability. Soft constraints are loading and unloading priority ranking. Cross-regional collaborative constraints include capacity allocation adaptability, path connection sequence, and hub transfer matching. Set quantitative standards for each dimension. Hard constraints must be 100% met, soft constraint satisfaction rate ≥ preset threshold, and collaborative constraint connection deviation ≤ limited range. Classify the degree of constraint violation, including mild violation (minor deviation of a single type of soft constraint), moderate violation (partial violation of a single type of hard constraint or deviation of multiple types of soft constraints), and severe violation (violation of multiple types of hard constraints or breakage of core collaborative constraints).

[0096] Three types of verification dimensions and quantitative standards are defined: hard constraints (size, slope, load) must be 100% met, soft constraints (loading and unloading priority) must be met at a rate of ≥95%, and coordination constraints (allocation delay ≤1h) must be met. Violations are classified into three levels: minor (small deviation of soft constraints), medium (local violation of a single type of hard constraint), and severe (violation of multiple types of hard constraints).

[0097] Step S152: During the iteration process, the verification is performed in the order of sub-regional verification, cross-regional collaborative verification, and global comprehensive verification. Sub-regional verification focuses on the compliance of local constraints, cross-regional collaborative verification focuses on the adaptability of capacity allocation, the compliance of main route constraints, and the rationality of hub connection timing, and global comprehensive verification integrates all dimensions to form a complete verification report, which clarifies the type, number, degree and scope of impact of constraint violations.

[0098] The iteration proceeds in the following order: "sub-regional verification (local constraint compliance) → cross-regional collaborative verification (capacity / route / connection) → global comprehensive verification". This generates a comprehensive report that includes the type, number, degree, and scope of constraint violations, allowing for precise identification of the problem.

[0099] Step S153: Build an information collection module, collect data according to constraint type, degree of violation, affected container batch, associated sub-region, and corresponding algorithm parameter fields, establish a root cause analysis model, and explore the correlation between constraint violation and algorithm parameters through statistical analysis;

[0100] A Python data collection module was built to collect data according to dimensions such as "constraint type - violation degree - algorithm parameters"; a root cause analysis model was constructed to explore the correlation between parameters and violation rate (e.g., the path violation rate is 12% when ant colony ρ=0.1, and drops to 5% when ρ=0.08), providing a basis for parameter adjustment.

[0101] Step S154: For the genetic algorithm, if the hard constraint violation rate is high, the crossover range is narrowed and the mutation probability of high-priority container coding segments is reduced. If the soft constraint violation rate is high, the weight of soft constraints in the comprehensive constraint satisfaction is adjusted. For the ant colony algorithm, if there are many path constraint violations, the constraint satisfaction weight in the heuristic function is increased and the pheromone increment of the violation segment is reduced. If the connection time sequence constraint is violated, the pheromone evaporation coefficient is adjusted. The parameters are adjusted according to the preset amplitude to avoid iterative oscillation.

[0102] Adjust parameters based on the root cause of constraint violation: if the ant colony path violation rate is >10%, increase the weight of the heuristic function constraint; if the genetic soft constraint violation rate is >5%, increase the weight of the soft constraint in the overall satisfaction. The parameter adjustment range should be ≤0.1 to avoid iterative oscillation. Restart the iteration after adjustment.

[0103] Step S155: The first layer is the constraint satisfaction condition, namely, hard constraints are 100% satisfied, soft constraint satisfaction rate is ≥ preset threshold, and there is no violation of heavy collaborative constraints. The second layer is the optimization objective stability condition, namely, the fluctuation amplitude of the comprehensive optimization function value for multiple consecutive generations is ≤ preset threshold. If both conditions are satisfied at the same time, the iteration is terminated and the optimal solution is output.

[0104] Set dual iteration termination conditions: first, constraints are met (hard constraints 100%, soft constraints ≥95%, no severe collaborative violations); second, the optimization objective is stable (the comprehensive function value fluctuates by ≤1% for 20 consecutive generations). When both conditions are met, the iteration terminates and the optimal solution is output.

[0105] Step S156: The prerequisite for starting is that there are still severe constraint violations after the number of iterations reaches the preset upper limit, but the core hard constraints have been satisfied. The remaining violated constraints are non-core. The relaxation range is limited to non-core soft constraints or local collaborative constraints. A relaxation threshold is set. The execution process is to evaluate the relaxation effect, adjust the constraint verification standard, and re-verify in all dimensions until the termination condition is met.

[0106] When the iteration reaches the upper limit of 150 generations and the core hard constraints are satisfied, with only moderate violations remaining, constraint relaxation is initiated: the relaxation range is limited (non-core soft constraints / local collaborative constraints, such as adjusting the slope threshold to 16°); the impact of relaxation is assessed (the risk of cargo damage increases slightly), the verification standard is adjusted and re-verified, until the termination condition is met.

[0107] Step S157: After parameter adjustment, restart the iteration, repeat the verification, analysis, and parameter tuning process, record the verification results, constraint violation information, parameter adjustment content and effects of each iteration, and form an optimization log.

[0108] After adjusting the parameters, restart the iteration and repeat the "verification-analysis-parameter tuning" process; build an optimization log module to record the verification results, root causes of violations, parameter tuning content and effects of each iteration (such as a 7% reduction in path violation rate after ρ is adjusted to 0.08), forming a traceable iteration ledger.

[0109] Step S160: Integrate the improved algorithm into the dynamic scheduling system, connect to real-time data, optimize parameters through the self-learning module, and retain the manual intervention interface.

[0110] The improved algorithm is integrated into the SpringBoot dynamic scheduling system, connecting with real-time data from container AIS, road conditions, and orders; an LSTM self-learning module is embedded to optimize algorithm parameters based on historical logs; and a web-based manual intervention interface is retained to support adjusting the scheme and modifying constraint thresholds, balancing automation and controllability.

[0111] Based on the same inventive concept, please refer to Figure 2 This diagram illustrates a schematic block diagram of a rounded-corner container transportation route dynamic optimization scheduling system 100, provided in an embodiment of this application, for executing the aforementioned dynamic optimization scheduling method for rounded-corner container transportation routes. The system 100 may include a communication unit 110, a machine-readable storage medium 120, and a processor 130. Alternatively, the machine-readable storage medium 120 may be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the rounded-corner container transportation route dynamic optimization scheduling method provided in the aforementioned method embodiment.

[0112] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for dynamic optimization scheduling of a rounded corner container transportation path, characterized in that, The method comprises: Quantifying three types of constraints of the rounded corner container size, barycenter and loading and unloading priority, converting them into matching formulas, path constraint thresholds and weight rules, and combining transportation cost, time and cargo loss risk to build a weighted multi-objective optimization function; A hybrid framework of genetic algorithm and ant colony algorithm is built, and a constraint adaptation layer is embedded to filter invalid solutions, including genetic algorithm responsible for global distribution of multi-region, multi-vehicle type and multi-container combination, ant colony algorithm responsible for path planning of single combination, constraint adaptation layer embedded in the framework to connect algorithm stages, and real-time filtering of invalid solutions violating size, barycenter and path constraints, the constraint adaptation layer including rule storage layer and verification execution layer; Two types of algorithm operators are improved, and constraint satisfaction, loading and unloading priority and dynamic parameter adjustment mechanism are integrated, including constraint satisfaction and loading and unloading priority in genetic operator, and path constraint satisfaction and priority matching degree in ant colony operator; a dynamic parameter adjustment mechanism is built to associate operator parameters and constraint violation data; A divide-and-conquer strategy is used to divide multiple regions into sub-regions for optimization, and then large-scale dynamic scenarios are adapted through cross-region collaboration and local re-optimization; Full-dimensional verification of scheduling scheme is iterated, algorithm parameters are automatically adjusted based on constraint violation information, and iteration is terminated or constraint relaxation is started according to preset conditions, including building a full-dimensional verification, constraint violation analysis, parameter adjustment, termination / relaxation closed loop framework: verifying hard / soft / collaborative constraints during iteration, adjusting algorithm parameters based on violation information, terminating iteration if conditions are met, and starting constraint relaxation in extreme cases; The improved algorithm is integrated into a dynamic scheduling system, connected to real-time data, optimized parameters through a self-learning module, and a manual intervention interface is retained.

2. The rounded corner container transport path dynamic optimization scheduling method of claim 1, wherein, The three types of constraints of the rounded corner container size, barycenter and loading and unloading priority are quantified, converted into matching formulas, path constraint thresholds and weight rules, and combined with transportation cost, time and cargo loss risk to build a weighted multi-objective optimization function, including: The size and corner radius of the rounded corner container are obtained through three-dimensional scanning, the cargo compartment size, corner chamfer radius and maximum load of the transport vehicle type are collected, and the barycenter related data of the container under different loading states are obtained through finite element mechanics simulation, and the cargo value, freshness requirement level, loading and unloading difficulty coefficient and emergency delivery level of the container are counted; Based on the size data of the container and the cargo compartment, the effective occupation and containing size of the two are calculated, the reasonable container loading quantity limit is determined, and the placement rules for avoiding the direct placement of the corner to the cargo compartment corner are defined; Combined with the barycenter offset data of the container and the vehicle driving dynamics principle, the maximum allowed slope and minimum allowed turning radius of the path are determined, and the slope and turning radius constraint rules that the path needs to meet are defined; The weighted summation method is used to set the weight coefficients corresponding to the cargo value, freshness requirement, loading and unloading difficulty, and emergency delivery, calculate the loading and unloading priority weight of the container and normalize the processing, and the loading and unloading order of the containers loaded on the same vehicle is arranged in descending order of priority weight, and the containers with the same priority are sorted according to the emergency delivery level and cargo value in turn; A total transportation cost function is constructed, covering fuel consumption, labor and vehicle depreciation costs, a total transportation time function is constructed, including transportation driving time and total loading and unloading time of all containers, a cargo loss risk coefficient function is constructed, comprehensively considering the cargo loss risks caused by gravity center deviation, path slope exceeding, insufficient turning radius and preservation requirements, and corresponding risk items are counted as extremely high risks in case of exceeding; A hierarchical structure model is constructed, taking comprehensive optimization as the target, total transportation cost, total time and cargo loss risk as the criteria, experts in the relevant fields are invited to compare the target of the criteria layer two by two and build a judgment matrix, the weights of each target are determined after consistency test, and a weighted comprehensive multi-objective optimization function after normalization processing is constructed.

3. The rounded corner container transport path dynamic optimization scheduling method of claim 2, wherein, The hierarchical structure model is constructed, taking comprehensive optimization as the target, total transportation cost, total time and cargo loss risk as the criteria, experts in the relevant fields are invited to compare the target of the criteria layer two by two and build a judgment matrix, the weights of each target are determined after consistency test, and a weighted comprehensive multi-objective optimization function after normalization processing is constructed, including: The boundaries of the hierarchical structure are defined, taking comprehensive optimization as the target layer, focusing on the balance of cost, efficiency and risk, the criteria layer defines three indexes of total transportation cost, total time and cargo loss risk and their respective connotations, the scheme layer covers related scheduling schemes of container, vehicle type combination, path planning and loading and unloading sequence, and ensures that the hierarchical elements are clear and do not overlap; An expert team covering multiple relevant fields and having corresponding experience is formed, the experts are informed in advance about the constraints of round-corner container transportation and the influencing factors of the criteria to ensure consistent understanding of the evaluation background; The 1-9 scale method is used to develop two-by-two comparison rules, the importance levels corresponding to different scales and the use of intermediate values between adjacent scales are determined, and the reciprocal nature of the scale is also determined; The judgment matrix is constructed based on the indexes of the criteria layer, the experts are organized to compare and score the importance of each criterion pair by pair, and the final elements of the matrix are determined after mean calculation and abnormal value adjustment after the scores are recovered; The consistency of the judgment matrix is checked, whether the consistency requirement is met is judged through the iteration solution of the eigenvector, the consistency index and the proportion calculation, if not, the experts are fed back to adjust the scores until the requirement is met; The eigenvector of the judgment matrix that passes the test is extracted as the initial weight, the weights of each criterion are obtained after normalization processing, and the weight distribution is reasonable and the sum is 1; Each sub-target is normalized, a comprehensive multi-objective optimization function is constructed by weighting, the advantages and disadvantages of the scheduling scheme are judged by the function value, and the comprehensive trade-off of cost, time and risk is realized.

4. The rounded corner container transport path dynamic optimization scheduling method of claim 1, wherein, The genetic algorithm and ant colony algorithm hybrid framework is built, the constraint adaptation layer is embedded to filter invalid solutions and improve iteration efficiency, including: The genetic algorithm is used as a global decision module, responsible for global distribution of multiple regions, multiple vehicle types and multiple round-corner container combinations, the ant colony algorithm is used as a local decision module, responsible for specific transportation path planning of a single container group and vehicle type combination, the global distribution scheme output by the genetic algorithm is used as the input premise of the ant colony algorithm, the path optimization result output by the ant colony algorithm is fed back to the genetic algorithm as the basis for its fitness calculation, forming a closed-loop interactive logic of global distribution, local optimization and result feedback; A double-layer architecture including a rule storage layer and a verification execution layer is built. The rule storage layer stores size matching, barycenter path slope / turning radius, and loading / unloading priority constraint rules of the round-corner container according to constraint types, verification logic, and violation judgment standards. The verification execution layer designs a standardized verification interface, formulates a verification priority of first verifying hard constraints and then verifying soft constraints, and filters invalid solutions violating constraints. A segmented chromosome coding mode of area segment, vehicle type segment, container combination segment, and loading / unloading node segment is adopted. After an initial population is randomly generated, the initial solution is pre-filtered by calling hard constraint rules of the constraint adaptation layer to eliminate solutions violating size matching and barycenter path constraints. The genetic algorithm is executed according to selection, crossover, and mutation. After a new solution is generated in each iteration, it is transmitted to the constraint adaptation layer to complete soft constraint verification and record constraint satisfaction degree. The fitness value of the solution is calculated in combination with the constraint satisfaction degree. Based on the global allocation scheme output by the genetic algorithm, the parameters and search space of the ant colony algorithm are initialized. The path constraint rules in the constraint adaptation layer are embedded into the node connectivity judgment logic to mark the node connection lines that cannot be passed. The initialization rule of the ant colony algorithm pheromone is optimized to improve the initial pheromone concentration of the path nodes that pass the verification of the constraint adaptation layer. During the state transition process of the ant colony algorithm, the verification interface of the constraint adaptation layer is called in real time. If a path segment violates the constraint, the path search of the corresponding ant is terminated. A unified scheduling module is built to control the algorithm iteration process. The genetic algorithm is first triggered to generate an initial solution of global allocation. After being filtered by the constraint adaptation layer, the iteration is executed. When the genetic algorithm converges to a preset threshold, the high-quality global allocation scheme is extracted and transmitted to the ant colony algorithm. After the ant colony algorithm completes local path planning, the path cost and constraint satisfaction degree data are fed back to the genetic algorithm to update its fitness function. The iteration is repeated until the global scheme and local path both satisfy the constraints and the optimization target meets the preset requirements. A dynamic updating mechanism of the constraint adaptation layer is designed to automatically update the rules in the rule storage layer and the verification logic in the verification execution layer when the constraint rules are adjusted. The double-layer iteration termination conditions of the constraint satisfaction degree reaching 100% and the comprehensive optimization function value being continuously stable in the preset optimal interval for multiple generations are set. The hybrid algorithm with the constraint adaptation layer is compared with the traditional hybrid algorithm for test. The iteration number, invalid solution ratio, and solution generation time are counted. If the efficiency improvement does not meet the expectation, the verification priority or algorithm iteration parameter of the constraint adaptation layer is adjusted.

5. The rounded corner container transport path dynamic optimization scheduling method of claim 1, wherein, The two types of algorithm operators incorporate the constraint satisfaction degree, loading / unloading priority, and dynamic parameter adjustment mechanism, including: An index is formed by quantitatively integrating the hard and soft constraint satisfaction degrees to adjust the fitness of the solution and exclude substandard solutions. The crossover range is limited according to the segmented coding. The high-priority container coding segment is locked. The constraints are verified after crossover. The mutation probability is set differently according to the loading / unloading priority. The mutation direction is adjusted in combination with the constraint satisfaction degree, and the solutions violating the rules are filtered out. In the formula, the quantitative hard / soft constraint satisfaction degree is integrated into the genetic algorithm fitness function, and solutions with a comprehensive satisfaction degree less than 80 points are excluded; the cross range is limited according to the segmented coding, and the coding segment of high-priority containers is locked; a lower mutation probability is set for high-priority containers, and the solutions that violate the rules are checked and removed after mutation, thereby strengthening the adaptation of the genetic algorithm to the constraints and priorities; The pheromone update is integrated with the path constraint satisfaction degree and the loading and unloading priority, the illegal path is punished with a decay coefficient of 1.2, the evaporation coefficient is adjusted in stages, the heuristic function is reconstructed to integrate the constraint satisfaction degree and the priority matching degree, and the weight is adjusted in stages; In the formula, the pheromone update of the ant colony algorithm integrates the path constraint satisfaction degree and the loading and unloading priority, and sets a decay coefficient of 1.2 for illegal paths; the heuristic function is reconstructed, and the weight is adjusted in stages to improve the constraint compliance of path optimization; The building module collects iterative data, formulates rules to adaptively adjust the parameters of the two types of algorithms and the constraint weight, and records the adjustment process and effect; The improved operator is connected to the hybrid algorithm framework, and the adaptation layer and the monitoring end are connected, the indicators before and after improvement are compared, and the adjustment parameters are traced back until the standard is met.

6. The rounded corner container transport path dynamic optimization scheduling method of claim 5, wherein, The strategy divides multiple regions into sub-regions for optimization, and then adapts to large-scale dynamic scenarios through cross-regional collaboration and local re-optimization, including: Based on geographical relevance, resource matching degree, and task balance, a clustering analysis method is used to divide multiple regions into several sub-regions, and the boundary range, scheduling node, and resource endowment of each sub-region are determined to ensure that the scheduling tasks within the sub-region are relatively independent and retain cross-regional collaboration interfaces; For each sub-region, the improved genetic and ant colony hybrid algorithm is called to perform local optimization, where the genetic algorithm is responsible for the global allocation of vehicle types and container combinations within the sub-region, and the ant colony algorithm is responsible for the local path planning of single container and vehicle type combinations. The constraint adaptation layer is connected during the optimization process to filter invalid solutions, and the preset iteration termination condition is met when the constraint satisfaction degree reaches 100% and the optimization target is stable, generating a sub-region scheduling sub-solution; All sub-region scheduling sub-solutions are summarized, and the unified data platform is used to analyze the vehicle capacity gap between sub-regions, cross-regional path congestion points, and container transfer connection breakpoints, establish a classification account that labels the bottleneck type, influence range, and associated sub-regions; Based on the bottleneck account, resource allocation rules are established between sub-regions to support the capacity gap in the scheduling region with surplus vehicles, and the size and load constraints of the target region's rounded containers are matched during allocation. The resource allocation priority is determined according to the emergency delivery level and the risk of cargo damage, and the allocation path and occupancy duration are recorded; The inter-sub-region transportation hub is used as a node to build a cross-regional road network model, and the path constraints of rounded containers are embedded in the road network node connectivity judgment. The ant colony algorithm is used to optimize the cross-regional main path, and the path connection sequence is planned according to the time nodes of the sub-region scheduling sub-solution. The path is dynamically adjusted considering real-time traffic conditions; Two types of local re-optimization trigger conditions are defined, including dynamic changes within sub-regions and cross-regional collaboration feedback problems. A trigger condition monitoring module is established to collect sub-region scheduling data, road condition data, and order data in real time, and mark the affected sub-regions and scheduling tasks that meet the trigger conditions. Start a local re-optimization process for the marked affected sub-region, call the hybrid algorithm to iteratively update the scheduling sub-scheme, reuse the original effective decisions, focus on adjusting the part affected by dynamic changes, and quickly check the new scheme through the constraint adaptation layer; Synchronize the local re-optimized sub-region scheduling sub-scheme to the unified data platform, check the rationality of cross-regional resource allocation and the smoothness of path connection, dynamically adjust the coordination strategy if the cross-regional bottleneck changes, and re-execute resource allocation and main path optimization to form a closed loop; Build a dynamic adjustment mechanism for sub-regions, which can re-divide boundaries or adjust resource allocation according to task increments and resource changes, establish information sharing channels between sub-regions, synchronize real-time transportation capacity, container flow progress, and constraint rule update information, and periodically evaluate the rationality of sub-region division and the effectiveness of coordination strategy and optimize them; Compare the scheduling scheme generation time, iteration convergence speed, cross-regional transportation connection efficiency, and constraint satisfaction rate before and after the implementation of the divide-and-conquer strategy. If the sub-region optimization effect is not good or there is a breakpoint in cross-regional coordination, adjust the sub-region division granularity, hybrid algorithm iteration parameters, or coordination allocation rules until the scheduling demand of large-scale dynamic scenarios is met.

7. The rounded corner container transport path dynamic optimization scheduling method of claim 6, wherein, The scheduling sub-scheme of all sub-regions is summarized, and the unified data platform is used to analyze the vehicle capacity gap between sub-regions, cross-regional path congestion points, and container transfer connection breakpoints. A classification account is established to mark the bottleneck type, impact range, and associated sub-regions, including: Clearly define the sub-region identifier, available vehicle types and parameters, container scheduling task details, transportation capacity supply and demand data, cross-regional path planning details, and container transfer hub connection time window data fields. Standardize data formats, coding rules, and data integrity verification standards; Build a distributed storage architecture to receive sub-region scheduling sub-scheme data, interface real-time traffic, transportation capacity status, and container flow tracking systems to obtain dynamic data. Set up a data synchronization mechanism to store and index data by sub-region, data type, and timestamp; Clean and aggregate data, remove duplicates, errors, and outliers, verify the compliance of container, vehicle matching, and path planning based on the rounded container constraint rules, mark abnormal data and associated sub-regions, and integrate them into a unified analysis dataset; Build a transportation capacity supply and demand balance calculation model to count the supply and demand of each sub-region by vehicle type, calculate the capacity gap, match the surplus and gap vehicles across sub-regions, count the gap size, affected container tasks, and emergency level, and identify the associated sub-regions; Combine path planning data and real-time traffic data to calculate path travel time deviation rate, filter congestion path segments and locate congestion nodes according to the preset threshold, analyze the impact range and constraint violation risk points; Focus on the container transfer hub node, extract the container connection time data, build a connection time window matching model, check the connection rationality, filter abnormal connection situations, and record the breakpoint-related information and expected impact; Design fields including bottleneck unique ID, type, specific description, impact range, associated sub-region list, severity classification, discovery timestamp, and data source, and store them in a structured table to support multi-dimensional filtering and querying; Based on dynamic data, recheck the bottleneck status regularly, update the account and mark the solved bottleneck, input the new bottleneck, trigger the automatic warning for high severity bottleneck and push the related personnel; Show the bottleneck distribution in the form of chart, support drilling query, set access rights, open online viewing and data export functions, and establish a version management mechanism to record update information.

8. The rounded corner container transport path dynamic optimization scheduling method of claim 6, wherein, The model is constructed with inter-sub-regional transportation hubs as nodes, the path constraints of the round-corner container are embedded in the connectivity judgment of the nodes, the ant colony algorithm is called to optimize the inter-regional main path, the time node of the sub-regional scheduling sub-scheme is combined to plan the path connection order, the real-time road condition is considered to dynamically adjust the path, including: The location, functional orientation and connected sub-regional range of the inter-sub-regional transportation hub are determined, the loading points and unloading points of each sub-region are integrated as nodes, the basic attributes of the connected road sections between nodes such as mileage, slope, turning radius and design speed are collected, the maximum allowable slope and minimum allowable turning radius of the round-corner container are associated as path constraint parameters to form a road section constraint attribute library; The node is taken as the vertex and the connected road section between nodes as the edge to construct an undirected road network graph model, the node connectivity judgment rule is embedded in the road network graph, when the road section slope meets the maximum allowable slope of the round-corner container and the turning radius meets the minimum allowable turning radius, the node connectivity is determined to be valid, otherwise it is marked as invalid, the mileage, expected travel time calculated based on the design speed and real-time road condition update interface are marked for each valid edge to form a road network model that can dynamically adapt to constraints; The state transition rule of the ant colony algorithm is adjusted, the road section constraint satisfaction degree is included in the transition probability calculation to improve the selection probability of compliant road sections, the road sections connecting the transportation hubs are given higher initial pheromone concentration, the target functions of the total path mileage, constraint satisfaction degree and sub-regional time node matching degree are set, and the optimization goal is to minimize the comprehensive cost; The earliest and latest time of the container arriving at the transportation hub, the latest time of the container leaving the hub and the expected arrival time of the unloading node in each sub-regional scheduling sub-scheme are extracted to establish a time window matching matrix with the transportation hub as the center, the time interval of receiving and transferring containers at the hub node is determined, and when the ant colony algorithm optimizes the inter-regional main path, the path travel time is ensured to match the time interval to plan the ordered connection path of the sub-region, hub and sub-region; The real-time road condition monitoring system is connected, the actual travel time, congestion state and temporary traffic control information of each road section in the road network are obtained at a preset period, the deviation rate of the actual travel time and the expected time is calculated, the deviation rate threshold is set, when the road section deviation rate exceeds the threshold or the road section is invalid due to congestion or control, the path dynamic adjustment mechanism is triggered, and the constraint compliance of the road section after the road condition changes is verified synchronously; After triggering the adjustment, the ant colony algorithm reiterates based on the updated road section travel time and the road network model of the invalid road section, the original compliant and good road sections are preferentially reused, after the new path is generated, the matching degree with the sub-regional time node is verified, if there is a time window conflict, the path node connection order is adjusted or an alternative compliant road section is selected. The optimized cross-regional main path is synchronized to the unified data platform, pushed to the relevant sub-regional scheduling system, and the time node of the sub-regional scheduling sub-scheme is confirmed for adaptation. If there is a mismatch, feedback is given to the ant colony algorithm for secondary re-optimization until the cross-regional path and the sub-regional scheme are completely adapted, and the cross-regional main path scheme containing constraint compliance description, time connection plan and real-time adjustment interface is output.

9. The rounded corner container transport path dynamic optimization scheduling method of claim 1, wherein, The iteration of the full-dimensional verification scheduling scheme automatically adjusts the algorithm parameters based on the constraint violation information, and terminates the iteration or starts the constraint relaxation under the preset conditions, including: The verification dimensions cover hard constraints, soft constraints and cross-regional collaborative constraints. The hard constraints include size matching, path slope / turning radius compliance corresponding to the center of gravity, and load adaptation. The soft constraints are loading and unloading priority sorting. The cross-regional collaborative constraints include transport capacity allocation adaptability, path connection timing, and hub transfer matching. Quantitative standards are set for each dimension. The hard constraints must be 100% satisfied. The soft constraint satisfaction rate is greater than or equal to a preset threshold. The collaborative constraint connection deviation is less than or equal to a limited range. The constraint violation degree level is divided, including single soft constraint slight deviation, mild violation; single hard constraint local violation or multiple soft constraint deviations, moderate violation; multiple hard constraint violations or collaborative constraint rupture, severe violation. The iteration process is executed in the order of sub-regional verification, cross-regional collaborative verification, and global comprehensive verification. The sub-regional verification focuses on local constraint compliance. The cross-regional collaborative verification focuses on transport capacity allocation adaptability, main path constraint compliance, and hub connection timing rationality. The global comprehensive verification integrates all dimensions to form a complete verification report, clearly indicating the type, number, degree and influence range of constraint violations. An information collection module is built to collect data according to constraint type, violation degree, affected container batch, associated sub-region, and corresponding algorithm parameter field. A root cause analysis model is established to statistically analyze and mine the correlation between constraint violations and algorithm parameters. For the genetic algorithm, if the hard constraint violation rate is high, the crossover range is reduced and the high-priority container code segment mutation probability is reduced. If the soft constraint violation rate is high, the weight of the soft constraint in the comprehensive constraint satisfaction degree is adjusted. For the ant colony algorithm, if the path constraint violation is high, the constraint satisfaction degree weight in the heuristic function is increased and the illegal path segment pheromone increment is reduced. If the connection timing constraint violation is high, the pheromone evaporation coefficient is adjusted. The parameters are adjusted by a preset amplitude to avoid iteration oscillation. The first level is the constraint satisfaction condition, i.e. the hard constraint is 100% satisfied, the soft constraint satisfaction rate is greater than or equal to a preset threshold, and there is no severe collaborative constraint violation. The second level is the optimization target stability condition, i.e. the comprehensive optimization function value fluctuation amplitude is less than or equal to a preset threshold for consecutive generations. When both conditions are met, the iteration is terminated and the optimal scheme is output. The starting premise is that the iteration number reaches a preset upper limit and there is still a severe constraint violation, but the hard constraint has been satisfied. The remaining violated constraints are non-core types. The relaxation range is limited to non-core soft constraints or local collaborative constraints. A relaxation threshold is set. The execution process is to evaluate the relaxation impact, adjust the constraint verification standard, and re-verify all dimensions until the termination condition is met. After the parameter adjustment, restart the iteration, repeat the verification, analysis, and parameter adjustment process, record the verification results, constraint violation information, parameter adjustment content and effect of each iteration, and form an optimization log.

10. A rounded corner container transport path dynamic optimization scheduling system, characterized in that, Comprise: a processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the machine-executable instructions to perform the rounded corner container transportation path dynamic optimization scheduling method of any one of claims 1 to 9.

Images