Emergency multi-target material scheduling method and device based on multi-task evolution framework

Abstract

The invention discloses an emergency multi-target material scheduling method and device based on a multi-task evolutionary framework. The method comprises the following steps: acquiring initial information required by emergency material scheduling; constructing a multi-task evolution framework comprising an original emergency material scheduling task and a simplified auxiliary task based on the initial information; in the multi-task evolution framework, population iteration optimization based on evolutionary computation is executed on the original emergency material scheduling task and the simplified auxiliary task, knowledge migration operation from the simplified auxiliary task to the original emergency material scheduling task is executed in the evolution process, and an optimized population of the original emergency material scheduling task is obtained; and according to a preset decision rule, determining a final emergency material scheduling scheme from the optimized population of the original emergency material scheduling task. A multi-task evolution framework is constructed, directed knowledge migration from an auxiliary task to an original task is executed, and the defects that single-task optimization is slow in convergence and direct multi-task parallel is prone to generating negative migration are overcome.

Description

Technical Field

[0001] This invention relates to the field of multi-objective optimization technology, and in particular to an emergency multi-objective material scheduling method and apparatus based on a multi-task evolutionary framework. Background Technology

[0002] In the field of emergency management, the efficiency and rationality of resource scheduling directly affect emergency response capabilities and rescue effectiveness, constituting an optimization problem involving multi-objective trade-offs and complex constraints. Existing technologies for addressing such resource scheduling problems can be broadly categorized into three types: traditional mathematical programming methods based on exact solutions, heuristic algorithms based on specific rules, and intelligent algorithms based on population iteration. Traditional mathematical programming methods often face limitations in handling large-scale, multi-constraint emergency scenarios, such as excessive computational complexity and difficulty in providing timely solutions. While heuristic algorithms can improve solution efficiency to some extent, the quality of their solutions heavily relies on design experience and is prone to getting trapped in local optima, making it difficult to guarantee the stability and reliability of the scheduling scheme. Intelligent methods such as evolutionary algorithms have gained widespread application in such problems due to their global search capabilities; however, most existing methods are designed based on a single-task framework, which suffers from slow convergence speed and difficulty escaping infeasible regions when dealing with complex scheduling problems like emergency situations. Furthermore, the entire optimization process lacks an effective knowledge-guided mechanism, making it difficult to simultaneously consider search efficiency and solution quality.

[0003] To improve optimization efficiency, a multi-task evolutionary framework has been proposed, which accelerates the search by simultaneously processing multiple related tasks and leveraging knowledge transfer between tasks. However, directly optimizing multiple complex emergency scheduling tasks in parallel can lead to "negative transfer" due to insufficient correlation between tasks; that is, ineffective or harmful knowledge transfer can actually reduce the optimization performance of the main task. Therefore, how to construct highly related and appropriately sized auxiliary tasks within a multi-task framework for the original highly complex emergency scheduling task, thereby effectively utilizing knowledge transfer to improve the optimization performance of the original task, has become a pressing technical challenge. Summary of the Invention

[0004] The purpose of this invention is to provide an emergency multi-objective material scheduling method and apparatus based on a multi-task evolution framework. By constructing a multi-task evolution framework that includes simplified auxiliary tasks and performing directed knowledge transfer from auxiliary tasks to the original tasks, it effectively overcomes the shortcomings of slow convergence in traditional single-task optimization and the tendency of negative transfer to occur in direct multi-task parallelism. Thus, in the complex multi-objective constraint problem of emergency material scheduling, it achieves faster convergence speed and higher quality scheduling scheme generation.

[0005] To address the aforementioned technical problems, a first aspect of this invention provides an emergency multi-objective resource scheduling method based on a multi-task evolution framework, comprising the following steps: Obtain the initial information required for emergency supplies dispatch; Based on the initial information, a multi-task evolution framework is constructed, which includes the original emergency material dispatching task and the simplified auxiliary task. In the multi-task evolutionary framework, population iterative optimization based on evolutionary computation is performed on the original emergency material dispatching task and the simplified auxiliary task respectively, and a knowledge transfer operation from the simplified auxiliary task to the original emergency material dispatching task is performed during the evolution process to obtain the optimized population of the original emergency material dispatching task. Based on preset decision-making rules, the final emergency material dispatching scheme is determined from the optimized population of the original emergency material dispatching tasks.

[0006] Furthermore, the construction of the multi-task evolutionary framework includes: For the original emergency material dispatching task, a multi-objective optimization model is constructed, which includes: a first objective function corresponding to the dispatching time, a second objective function corresponding to the total economic cost of transportation, and a third objective function corresponding to the dispatching unfairness. For the simplified auxiliary task, a single-objective optimization model is constructed, which includes a single objective function corresponding to scheduling time, total economic cost of transportation, and scheduling unfairness. A unified set of constraints is set for the multi-task evolution framework, which includes: vehicle number constraint, vehicle capacity constraint, supply point inventory constraint, and demand point demand constraint.

[0007] Furthermore, within the multi-task evolutionary framework, performing population iterative optimization based on evolutionary computation on both the original emergency material dispatching task and the simplified auxiliary task includes: A first initial population corresponding to the original emergency material dispatching task and a second initial population corresponding to the simplified auxiliary task are generated respectively. Each individual in the first initial population and the second initial population represents a complete dispatching scheme through an encoding structure. Based on the first initial population and the second initial population, the original emergency material dispatching task and the simplified auxiliary task are subjected to multiple rounds of co-evolution; wherein, each round of co-evolution includes: executing the population evolution and screening process of the original emergency material dispatching task and the simplified auxiliary task in parallel, and repairing infeasible solutions generated during the evolution of the original emergency material dispatching task based on the unified constraints.

[0008] Furthermore, the step of performing multiple rounds of co-evolution on the original emergency material dispatching task and the simplified auxiliary task based on the first initial population and the second initial population includes: Evolutionary operations are performed on the current population of the original emergency material dispatching task and the current population of the simplified auxiliary task, respectively, to generate corresponding offspring populations; An infeasibility solution repair operation is performed on the offspring population of the original emergency material dispatch task to ensure that each individual meets the unified constraint condition. Based on a multi-objective optimization strategy, individuals are selected from the current population and its repaired offspring population of the original emergency material dispatching task to form its next generation population. Based on a single-objective optimization strategy, individuals are selected from the current population and its offspring population of the simplified auxiliary task to form the next generation population.

[0009] Further, the step of performing evolutionary operations on the current population of the original emergency material dispatching task and the current population of the simplified auxiliary task, respectively, includes: For any given task, the current population is randomly paired with its individuals; For each pair of individuals, a single-point crossover operation is performed according to a preset crossover probability, and the encoding segments of the paired individuals after the randomly selected crossover point are swapped to generate intermediate offspring individuals. For all the intermediate offspring individuals, a mutation operation is performed according to a preset mutation probability. The mutation operation randomly selects the quantity of material transportation in the code of the intermediate offspring individual and increases or decreases it according to a preset adjustment range so that the adjusted quantity is a non-negative integer, thus obtaining the offspring population of the current task.

[0010] Furthermore, the infeasibility solution repair operation performed on the offspring population of the original emergency material dispatching task includes: For each individual in the offspring population, determine whether the individual violates the supply point inventory constraint. If it does, randomly select the transportation task corresponding to the individual's code and gradually reduce the amount of materials transported by the individual according to a first preset quantity until it does not violate the supply point inventory constraint. Check whether the individual violates the demand point constraint. If it does, identify the transportation task in the individual code where the quantity of materials transported exceeds the corresponding demand, and gradually reduce the quantity of materials transported in the identified transportation task by a second preset quantity until it does not violate the demand point constraint. Based on the adjusted quantity of materials transported for all transport tasks, the number of vehicles required to execute each transport task is recalculated, and the vehicle quantity field for all transport task sequences in the individual code is updated.

[0011] Furthermore, the selection of individuals from the current population and its repaired offspring population of the original emergency material dispatching task based on the multi-objective optimization strategy includes: The current population of the original emergency material dispatching task is merged with its repaired offspring population to obtain a merged population. Calculate the constraint violation degree value for each individual in the merged population; The individuals in the merged population are sorted by non-dominated order, and the non-dominated level of each individual is determined based on the function values ​​of the first objective function, the second objective function, and the third objective function. Calculate the crowding distance of the individuals with the same non-dominated hierarchy in the target space composed of the first objective function, the second objective function, and the third objective function; Based on the order of the non-dominated hierarchy from high to low, and within the same non-dominated hierarchy based on the order of the crowding distance from large to small, a predetermined number of individuals are selected from the merged population to form the next generation population for the original emergency material dispatching task.

[0012] Furthermore, the step of performing a knowledge transfer operation from the simplified auxiliary task to the original emergency material dispatching task during the evolutionary process to obtain an optimized population for the emergency material dispatching task includes: From the current population of the simplified auxiliary task, select all individuals that are not dominated by any individual in the current population of the original emergency material dispatch task, and form a set of transfer knowledge. For each individual in the set of transferred knowledge, determine the corresponding individual in the current population of the original emergency material dispatching task that has the minimum cosine angle distance with respect to the objective function value. The cosine angle distance is used to measure the directional similarity between two individuals in the target space composed of the first objective function, the second objective function, and the third objective function. For each pair of corresponding individuals, the information of the individuals in the knowledge transfer set is transferred to the corresponding individuals of the original emergency material dispatching task according to the preset learning rate, thereby generating new candidate individuals. If the new candidate individual dominates the corresponding individual of the original emergency material dispatching task, then the new candidate individual replaces the corresponding individual in the current population of the original emergency material dispatching task, resulting in an optimized population of the original emergency material dispatching task after knowledge transfer optimization.

[0013] Furthermore, determining the final emergency material dispatching scheme from the optimized population of the original emergency material dispatching tasks according to preset decision rules includes: Determine whether an optimization instruction has been received for any of the first, second, and third objective functions; If the optimization instruction is not received, the first rule is executed: the function values ​​of the first objective function, the second objective function and the third objective function of each individual in the optimization population are normalized respectively, the comprehensive evaluation value of each individual is calculated, the comprehensive evaluation value is the sum of the normalized objective function values, and the individual with the smallest comprehensive evaluation value is selected as the final emergency material dispatch plan. If an optimization instruction for a given objective function is received, the second rule is executed: the individual with the smallest value of the given objective function is directly selected from the optimization population as the final emergency material dispatch plan.

[0014] Accordingly, a second aspect of the present invention provides an emergency multi-objective material scheduling device based on a multi-task evolutionary framework, wherein the scheduling scheme is determined based on the above-mentioned emergency multi-objective material scheduling method based on the multi-task evolutionary framework, including: The data acquisition module is used to acquire the initial information required for emergency material dispatch. The framework construction module is used to construct a multi-task evolution framework based on the initial information. The multi-task evolution framework includes the original emergency material dispatching task and the simplified auxiliary task. The population optimization module is used to perform population iterative optimization based on evolutionary computation on the original emergency material dispatching task and the simplified auxiliary task respectively in the multi-task evolutionary framework, and to perform knowledge transfer operation from the simplified auxiliary task to the original emergency material dispatching task during the evolution process, so as to obtain the optimized population of the original emergency material dispatching task. The scheme selection module is used to determine the final emergency material dispatch scheme from the optimized population of the original emergency material dispatch tasks according to preset decision rules.

[0015] Accordingly, a third aspect of the present invention provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the above-described emergency multi-objective material scheduling method based on a multi-task evolution framework.

[0016] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-described emergency multi-objective material scheduling method based on a multi-task evolution framework.

[0017] The above-described technical solutions of the embodiments of the present invention have the following beneficial technical effects: 1. By constructing a multi-task evolutionary framework consisting of an original multi-objective scheduling task and a simplified single-objective auxiliary task, this approach effectively addresses the inherent problems of slow convergence and susceptibility to local optima in traditional single-task optimization methods under complex emergency scenarios. The auxiliary task is simplified based on a key objective of the original task, ensuring a high degree of correlation between tasks and providing stable and efficient search guidance for the evolution of the original task. This significantly accelerates the overall optimization process, enabling the algorithm to quickly generate feasible, high-quality scheduling schemes within a short emergency response time. 2. By adopting a directional knowledge transfer strategy based on non-dominant relations and cosine angular distance, the "negative transfer" problem that may occur in multi-task optimization is fundamentally avoided. This strategy selects advantageous knowledge individuals from auxiliary tasks and achieves precise transfer by measuring their similarity to the original task individuals in the target space direction. This ensures that the transferred knowledge can effectively improve the quality of the original task population rather than interfere with its evolutionary direction, thereby improving search efficiency while ensuring the superiority and reliability of the final solution set. 3. The system integrates an infeasible solution repair mechanism for scheduling constraints with a multi-objective selection strategy based on non-dominated ordering and crowding distance, jointly ensuring the generation of diverse and balanced high-quality scheduling schemes under complex constraints. The repair mechanism can automatically and efficiently adjust scheduling schemes that violate realistic constraints into feasible solutions, ensuring the practicality of all evolutionary paths; while the multi-objective selection strategy accurately selects the Pareto optimal solution set from feasible solutions that achieves the best balance among multiple objectives such as time, cost, and fairness, providing decision-makers with rich and high-quality alternatives, greatly enhancing the scientific nature and flexibility of scheduling decisions. Attached Figure Description

[0018] Figure 1 This is a flowchart of an emergency multi-objective material scheduling method based on a multi-task evolution framework provided in an embodiment of the present invention; Figure 2 This is a logical schematic diagram of an emergency multi-objective material scheduling method based on a multi-task evolution framework provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of individual coding provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the Pareto non-dominated hierarchy provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a single-point intersection provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of congestion distance provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the knowledge transfer strategy provided in an embodiment of the present invention; Figure 8 This is a block diagram of an emergency multi-objective material dispatching device based on a multi-task evolution framework provided in an embodiment of the present invention.

[0019] Figure label: 1. Data acquisition module; 2. Framework construction module; 3. Population optimization module; 4. Scheme selection module. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0021] Please refer to Figure 1 The first aspect of this invention provides an emergency multi-objective material scheduling method based on a multi-task evolution framework, comprising the following steps: Step S100: Obtain the initial information required for emergency material dispatch.

[0022] Specifically, a complete information model containing supply points, demand points, materials, vehicles, and transportation routes needs to be read and constructed. Supply point information should include its geographical location, the inventory quantity of various emergency materials (such as personal protective equipment, radiation monitoring materials, and medical treatment materials), and the number of available transport vehicles. Demand point information should include its location and the urgent demand for various materials. Material information should define the physical attributes of each type of item, such as its volume or weight. Vehicle information should include its maximum load or volume, driving speed, and unit distance transportation cost. Transportation route information should be presented in matrix form, clearly defining the feasible paths and distances between each supply point and each demand point.

[0023] Step S200: Based on the initial information, construct a multi-task evolution framework, which includes the original emergency material dispatching task and simplified auxiliary tasks.

[0024] The structured initial information obtained in step S100 constitutes the sole data foundation and quantitative basis for constructing all mathematical models and optimization objectives of the multi-task evolutionary framework in step S200. The calculation of all objective functions for the original and auxiliary tasks within the framework directly depends on this initial information: the determination of the scheduling time objective requires calculation based on the distance between two locations in the transportation path information matrix and the driving speed in the vehicle information; the total economic cost objective is determined by the distance of each path and the unit transportation cost in the vehicle information; the scheduling unfairness objective relies on a comprehensive evaluation by comparing the demand and actual receipt quantities in the demand point information, and the inventory and actual outgoing quantities in the supply point information. Simultaneously, the specific boundary values ​​of all constraints also originate entirely from this: the upper limit of the vehicle number constraint is taken from the number of available vehicles at each supply point; the boundary of the vehicle capacity constraint directly corresponds to the rated cargo volume in the vehicle information; the upper limits of various materials in the supply point inventory constraint are clearly recorded in the supply point information; the lower limit of the demand point constraint is clearly defined by the demand point information; and the material volume conversion involved in the constraint calculation also requires consideration of the single-item volume parameters in the material information. Furthermore, this initial information collectively defines the decision variable space of the entire optimization problem, that is, the encoding structure of each individual (scheduling scheme) in the evolutionary algorithm. In essence, it is to allocate a specific number of vehicles and various types of materials from each known supply point to each known demand point. The set of all supply points, demand points, material types and vehicle types is completely determined by the initial information, thus defining the range of the solution space of the entire optimization search.

[0025] First, a multi-objective optimization model with three conflicting objectives is established for the original task: the first objective is to minimize the overall scheduling time, defined as the completion time of the longest transportation route; the second objective is to minimize the total economic cost of transportation, i.e., the sum of the products of the distance traveled by all vehicles and the unit cost; the third objective is to minimize scheduling unfairness, which comprehensively measures the balance of material satisfaction rates among different demand points and the balance of load rates among different supply points. Second, a simplified auxiliary task is constructed, which uses only the first objective of the original task—minimizing scheduling time—as its sole optimization objective. Finally, a set of common hard constraints is defined for both tasks, including that the total number of vehicles dispatched from each supply point does not exceed its inventory, the total volume of materials transported each time does not exceed the capacity of the vehicles used, the total amount of various materials transported from any supply point does not exceed its inventory, and the total amount of various materials delivered to any demand point is not less than its demand.

[0026] Step S300: In the multi-task evolutionary framework, population iterative optimization based on evolutionary computation is performed on the original emergency material dispatching task and the simplified auxiliary task respectively. During the evolution process, a knowledge transfer operation is performed from the simplified auxiliary task to the original emergency material dispatching task to obtain the optimized population of the original emergency material dispatching task.

[0027] At the start of optimization, initial populations are generated for both the original task and the auxiliary task. Each individual in the population uses a specific encoding structure to represent a complete scheduling scheme. This encoding consists of multiple subsequences, each precisely corresponding to a transportation task from a specific supply point to a specific demand point, and encoding the number of vehicles dispatched and the transportation volume of various materials. In each iteration, the two tasks independently perform population evolution operations, including single-point crossover based on random pairing and probability-based mutation. Specifically, for the offspring population generated by the original task, a dedicated infeasibility repair process is performed, sequentially correcting any potentially violated inventory and demand constraints, and recalculating the required number of vehicles to ensure that all individuals entering subsequent selections are feasible solutions. Subsequently, the two tasks update their populations according to their respective strategies: the original task uses environment selection based on fast non-dominated sorting and congestion distance to retain individuals that perform well across multiple objectives and are diverse in distribution; the auxiliary task simply sorts and selects based on its single objective function value.

[0028] Step S400: Based on preset decision rules, determine the final emergency material dispatch plan from the optimized population of the original emergency material dispatch tasks.

[0029] An optimized population typically contains a set of Pareto optimal solutions, i.e., high-quality alternatives that do not dominate each other. Pre-defined decision rules provide criteria for selecting the final solution from this set. If no priority instruction for a specific objective is received before the scheduling decision, the system employs a comprehensive decision rule: first, the values ​​of each individual in the population for scheduling time, economic cost, and unfairness are normalized to eliminate the influence of dimensions; then, the sum of the normalized objective values ​​for each individual is calculated as a comprehensive evaluation value; finally, the individual with the smallest comprehensive evaluation value is selected as the execution plan, achieving balanced optimization of multiple objectives. If a priority objective is explicitly specified before the decision (e.g., requiring the shortest scheduling time), the system employs a directional decision rule, directly selecting the individual with the smallest specified objective function value from the optimized population as the final plan to satisfy the urgent preferences in a specific scenario. Through these rules, this method can flexibly adapt to different decision-making needs, outputting deterministic scheduling instructions that conform to multi-objective optimization principles and meet actual command requirements.

[0030] Furthermore, the construction of the multi-task evolutionary framework in step S200 includes: Step S210: For the original emergency material dispatching task, construct a multi-objective optimization model. The multi-objective optimization model includes: a first objective function corresponding to the dispatching time, a second objective function corresponding to the total economic cost of transportation, and a third objective function corresponding to the dispatching unfairness.

[0031] The first objective function, scheduling time, is defined as the total time required to complete the delivery of supplies from all supply points to all demand points along the longest or slowest route in the entire scheduling plan. This calculation is based on path distance and vehicle speed, and does not consider the time difference that may occur when multiple vehicles depart at different times on the same route. Its core is to evaluate the latest completion node of the emergency response. The second objective function, total economic cost of transportation, is calculated by summing the mileage of all vehicles performing transportation tasks and the preset unit mileage transportation cost. This objective is directly related to the economic burden and resource consumption of the rescue operation. The third objective function, scheduling unfairness, is a comprehensive measure aimed at evaluating the degree of spatial and resource balance in the allocation of supplies. It consists of two parts: first, the variance of the ratio of the actual amount of supplies received by each demand point to its demand, reflecting the fairness of rescue coverage; second, the variance of the ratio of the actual amount of supplies transported out by each supply point to its inventory capacity, reflecting the balance of resource allocation. The two are weighted and summed through a preset weighting factor.

[0032] Step S220: To simplify the auxiliary task, a single-objective optimization model is constructed. The single-objective optimization model includes a single objective function corresponding to the scheduling time, the total economic cost of transportation, and the scheduling unfairness.

[0033] Specifically, the auxiliary task selects only "scheduling time" from the three objectives of the original task as its sole optimization objective, completely abandoning explicit considerations of economic costs and unfairness. Choosing scheduling time as the simplified objective is inherently reasonable because time is often the primary and most critical factor in emergency response. An auxiliary task built around this core objective can guide the search to converge quickly to a high-performance region in the time dimension. The model of this auxiliary task shares all decision variables (i.e., resource allocation and vehicle scheduling schemes) and constraints of the original task, but its evaluation criteria become extremely singular and focused.

[0034] Step S230: Set unified constraints for the multi-task evolution framework. The unified constraints include: vehicle number constraints, vehicle capacity constraints, supply point inventory constraints, and demand point demand constraints.

[0035] This step defines a series of real-world physical and resource constraints that must be strictly adhered to in the entire multi-task optimization system, ensuring that any scheduling scheme generated by the evolutionary algorithm is theoretically executable. The vehicle number constraint stipulates that the total number of vehicles dispatched from any supply point (e.g., supply point A1 in the example) to perform transportation tasks cannot exceed the total number of vehicles actually owned and available at that supply point (e.g., A1 has 3 vehicles). The vehicle capacity constraint stipulates that the total volume of all types of supplies loaded by any vehicle in a specific transportation task (e.g., the sum of the volumes of personal protective equipment, radiation monitoring items, and medical supplies) cannot exceed the rated cargo capacity of that type of vehicle (e.g., in the example, the vehicle capacity is 50 cubic meters). The supply point inventory constraint stipulates that the total quantity of a specific type of supply (e.g., medical supplies) transported from any supply point cannot exceed the initial inventory of that type of supply at that supply point (e.g., supply point A1 has an inventory of 160 medical supplies). The demand constraint stipulates that the total quantity of a certain type of material delivered to any demand point must be at least equal to the urgent demand for that type of material declared by that demand point (e.g., demand point B1 needs 40 pieces of medical supplies).

[0036] Furthermore, in step S300, within the multi-task evolutionary framework, population iterative optimization based on evolutionary computation is performed on both the original emergency material dispatching task and the simplified auxiliary task, including: Step S310: Generate a first initial population corresponding to the original emergency material dispatching task and a second initial population corresponding to the simplified auxiliary task. Each individual in the first initial population and the second initial population represents a complete dispatching scheme through an encoding structure.

[0037] Each individual's coding structure directly corresponds to a complete material dispatching plan. This structure consists of multiple sequentially connected sub-sequences, the total number of which equals the product of the number of supply points and the number of demand points. In an embodiment containing five supply points (A1 to A5) and two demand points (B1, B2), each individual's coding consists of ten sub-sequences. Each sub-sequence precisely corresponds to a potential transportation task from a specific supply point to a specific demand point, and internally encodes four key decision variables: the number of vehicles assigned to the task, and the quantities of the three types of materials to be transported (e.g., personal protective equipment, radiation monitoring items, and medical supplies). For example, a sub-sequence "[2, 20, 10, 15]" indicates that from supply point A1 to demand point B1, two vehicles are dispatched to transport 20 pieces of personal protective equipment, 10 pieces of radiation monitoring items, and 15 pieces of medical supplies.

[0038] Step S320: Based on the first initial population and the second initial population, perform multiple rounds of co-evolution on the original emergency material dispatching task and the simplified auxiliary task. Each round of co-evolution includes: executing the population evolution and selection process of the original emergency material dispatching task and the simplified auxiliary task in parallel, and repairing infeasible solutions generated during the evolution of the original emergency material dispatching task based on unified constraints.

[0039] In each round of evolution, the original task and the auxiliary task first independently perform evolutionary operations on their current populations: based on a preset crossover probability, single-point crossover is performed on randomly paired parent individuals, exchanging some code fragments to generate new combinations; subsequently, based on a preset mutation probability, the quantity of goods transported in the codes of offspring individuals is randomly perturbed (e.g., increased or decreased by a percentage and rounded down from the current value), thereby introducing new changes. Since the evolutionary operators may compromise the feasibility of the solution, a repair process is specifically performed on the offspring population of the original task: first, violations of supply point inventory constraints are checked and corrected, randomly selecting excess transport tasks and gradually reducing their goods transport volume; then, violations of demand point demand constraints are checked and corrected, identifying insufficient supplies and gradually adjusting the transport volume of related tasks; finally, the number of vehicles required for each transport task is recalculated based on the corrected goods quantity, and the codes are updated. After the repair is completed, both tasks enter the population selection and update phase. The original task employs a multi-objective optimization strategy, merging its parent population with the repaired offspring population, calculating the constraint violation degree of each individual, and then performing a fast non-dominated sort to determine the Pareto level. Within the same level, the sparsity of the individual distribution is measured based on the crowding distance. Finally, individuals with high non-dominated levels and large crowding distances are preferentially selected to form the next generation population, thereby simultaneously approximating the Pareto front and maintaining solution diversity. The auxiliary task, on the other hand, adopts a single-objective optimization strategy. It simply sorts the individuals in its merged parent and offspring population based on its unique scheduling time objective function value, selecting the individual with the shortest time to form the next generation population, thus quickly converging towards minimizing the scheduling time.

[0040] Furthermore, in step S320, based on the first and second initial populations, multiple rounds of co-evolution are performed on the original emergency material dispatching task and the simplified auxiliary task, including: Step S321: Perform evolutionary operations on the current population of the original emergency material dispatching task and the current population of the simplified auxiliary task respectively to generate corresponding offspring populations.

[0041] For any task (original or auxiliary), the evolutionary process follows the same procedure. First, individuals in the current population are randomly paired to form multiple parent pairs. Then, based on a preset crossover probability, a single-point crossover operation is performed on each parent pair: a position in the individual's coding sequence is randomly selected as the crossover point, and all coding segments after that point are swapped between the two parent individuals, thus generating two new intermediate offspring individuals. For example, if one parent individual's partial coding is "2, 20, 10, 15, 1, 15, 8, 12..." and the other's is "1, 18, 12, 10, 2,22, 9, 15...", swapping them after the randomly selected crossover point may produce a new combination such as "2, 20, 10, 15, 2, 22, 9,15...". Next, based on a preset mutation probability, a mutation operation is performed on all intermediate offspring individuals: the gene locus representing the quantity of transported materials in the individual's code (e.g., the value representing the quantity of various types of materials in the subsequence) is randomly selected and modified according to a preset adjustment range, thereby introducing subtle random perturbations.

[0042] Step S322: Perform an infeasibility solution repair operation on the offspring population of the original emergency material dispatching task to ensure that each individual meets the uniform constraint conditions.

[0043] Because crossover and mutation operations may randomly alter the allocation of resources and vehicle assignments in the scheduling scheme, causing newly generated offspring individuals to violate actual physical or resource constraints, this step specifically performs a systematic repair on the offspring population of the original task to ensure that all individuals are feasible solutions. The repair operation is executed in a specific logical order. First, regarding the supply point inventory constraint, each individual in the offspring population is traversed to check whether the total amount of various resources transported from the same supply point by all its transportation tasks exceeds the actual inventory of that supply point; if an excess is found, some outgoing tasks corresponding to that supply point are randomly selected, and the quantity of excess resources transported is gradually reduced (each time by a preset unit, such as 1 piece), until the total outgoing volume of that supply point does not exceed its inventory limit. Secondly, regarding the demand constraints at each demand point, it is checked whether each individual meets the minimum demand for all types of materials at all demand points. If it is found that the total amount of a certain material received at some demand points is insufficient, then tasks for transporting this type of material to that demand point are searched in the coding, and the transport quantity of these tasks is gradually increased (increasing by a preset unit each time) until the minimum demand of that demand point is met. Conversely, if it is found that the total amount of a certain material received at a demand point exceeds its demand, then the transport quantity of the relevant transport tasks is gradually reduced. Finally, based on the material transport quantity of each transport task determined after the repair, combined with the unit volume of each type of material and the rated capacity of the vehicle, the actual number of vehicles required to execute each transport task is recalculated, and the vehicle number field of the corresponding subsequence in the individual coding is updated.

[0044] Step S323: Based on a multi-objective optimization strategy, individuals are selected from the current population and its repaired offspring population of the original emergency material dispatching task to form the next generation population.

[0045] Among numerous feasible solutions, individuals exhibiting superior overall performance across multiple objectives and demonstrating diversity are selected to guide the original task population towards a high-quality Pareto front. First, the current population (parent generation) of the original task is merged with the offspring population repaired in step S322, forming a larger merged population. Next, each individual in the merged population is evaluated and compared. The evaluation process first calculates the constraint violation degree value for each individual, i.e., the quantitative sum of all constraints violated; for fully feasible individuals, this value is zero. Subsequently, a fast non-dominated sort is performed on the merged population: based on the individual's performance on the three objective functions of scheduling time, total transportation economic cost, and scheduling unfairness, pairwise comparisons are made to identify individuals not dominated by any other individual (i.e., superior in at least one objective and not inferior in others), defining them as the first non-dominated level. After removing these individuals, the next level of non-dominated individuals is identified from the remaining individuals, and so on, thus assigning all individuals a non-dominated level representing their performance level. For individuals within the same non-dominated hierarchy, to further differentiate their quality and maintain population diversity, the crowding distance of each individual in its target space is calculated. This distance reflects the density of other individuals around that individual, with individuals sparsely distributed in the target space (larger crowding distance) being more favored. Ultimately, the selection process proceeds according to the non-dominated hierarchy from high to low (i.e., the first hierarchy takes precedence over the second). When selection is required within the same hierarchy, individuals with larger crowding distances are prioritized, until the predetermined next generation population size is reached.

[0046] Step S324: Based on the single-objective optimization strategy, individuals are selected from the current population and its offspring population to form the next generation population, which simplifies the auxiliary task.

[0047] First, the current population (parent generation) of the auxiliary task is merged with its offspring population directly generated in step S321 (since the auxiliary task does not involve repairing infeasible solutions with complex constraints, its offspring population is directly used for selection). Then, since the auxiliary task's sole optimization objective is minimizing scheduling time, the function value of this objective function is directly used as the evaluation criterion. All merged individuals are sorted in ascending order of their scheduling times, from shortest to longest. Finally, starting from the top of the sorted list, a predetermined number of individuals are selected sequentially; these individuals represent candidate solutions that can complete resource scheduling fastest at the current evolutionary stage.

[0048] Furthermore, step S321 involves performing evolutionary operations on the current population of the original emergency supplies dispatching task and the current population of the simplified auxiliary task, respectively, including: Step S3211: For any task's current population, randomly pair its individuals.

[0049] All individuals in the current population (e.g., 100) are randomly paired to form several parent pairs. This random pairing strategy does not rely on individual fitness or any similarity measure, ensuring the fairness and randomness of the pairing process. For example, in a current population representing the original task, an individual encoded as transporting a large amount from supply point A1 to B1 may be paired with another individual encoded as transporting mainly from supply point A5 to B2.

[0050] Step S3212: For each pair of individuals, perform a single-point crossover operation according to the preset crossover probability, and exchange the encoding segments of the paired individuals after the randomly selected crossover point to generate intermediate offspring individuals.

[0051] By simulating chromosome crossover in biological genetics, a large-scale structured recombination of the parental scheduling scheme is performed. For each randomly paired parent individual, a predetermined crossover probability (e.g., 0.2) is used to determine whether to perform a crossover operation. If crossover is performed, a crossover point is randomly selected on the complete coding sequences of the two individuals; this point can fall after any gene locus in the coding sequence. Subsequently, all coding segments after the crossover point in the two parent individuals are exchanged. For example, the encoding of parent individual one is "[2,20,10,15], [1,15,8,12], [3,25,12,18]...", and the encoding of parent individual two is "[1,18,12,10], [2,22,9,15], [2,19,10,14]...". If crossover is randomly selected after the second subsequence, the generated intermediate child individual one is "[2,20,10,15], [1,15,8,12],[2,19,10,14]...", and individual two is "[1,18,12,10], [2,22,9,15], [3,25,12,18]...". Through this operation, transportation task combinations that originally belonged to different parent generations and targeted different supply-demand pairs were reintegrated into new individuals, thereby potentially creating entirely new scheduling schemes that combine the advantages of different parent generations (such as the time efficiency of one parent generation and the cost advantage of another), greatly expanding the search range of the algorithm.

[0052] Step S3213: For all intermediate offspring individuals, perform a mutation operation based on the preset mutation probability. The mutation operation randomly selects the quantity of material transportation in the code of the intermediate offspring individual and increases or decreases it according to the preset adjustment range so that the adjusted quantity is a non-negative integer, thus obtaining the offspring population of the current task.

[0053] The system performs local fine-tuning on intermediate offspring individuals generated after crossover to introduce new random variations and restore population diversity that may have been lost due to crossover, while also exploring better solutions at subtle levels. Specifically, for each intermediate offspring individual, the system iterates through all gene bits representing the quantity of goods transported in its encoding (i.e., the values ​​in each subsequence excluding the first vehicle number segment). For each such gene bit, a predetermined mutation probability (e.g., 0.8) is used to determine whether to perform mutation. If mutation is performed, the mutation direction (increase or decrease) is first randomly determined, and then the change is calculated according to a predetermined relative adjustment margin (e.g., 10%). The original value of the gene bit is multiplied by the adjustment margin, the result is rounded to the nearest integer, and then added to or subtracted from the original value according to the mutation direction, ensuring that the final result is a non-negative integer. For example, if the original value of a gene bit is "7" (representing the transport of 7 items of a certain type of goods), and the adjustment margin is 10%, the change is 0.7, rounded to 1. If the mutation direction is increase, the new value is 8; if it is decrease, the new value is 6. If the original value is zero, the reduction operation is ineffective, and the value remains zero.

[0054] Furthermore, step S322 involves performing an infeasibility solution repair operation on the offspring population of the original emergency supplies dispatch task, including: Step S3221: For each individual in the offspring population, determine whether the individual violates the supply point inventory constraint. If it does, randomly select the corresponding transportation task in the individual code and gradually reduce the amount of materials transported by the individual according to the first preset quantity until it does not violate the supply point inventory constraint.

[0055] Iterate through each individual in the offspring population and, for each supply point, calculate the total amount of various materials transported to each demand point from all transportation tasks originating from that point. Compare this total amount with the initial inventory of each type of material at that supply point. If the total outflow of a certain type of material exceeds the inventory, the individual is deemed to have violated the supply point's inventory constraint. Once a violation is detected, a repair mechanism is activated: from all transportation tasks originating from that supply point that transported excess of that type of material, randomly select one or more tasks as adjustment targets. Then, according to a preset, relatively large adjustment step size (i.e., a first preset quantity, for example, reducing by 10 units each time), gradually reduce the amount of that type of material transported in the selected tasks. This reduction process is repeated cyclically, with the total outflow recalculated after each reduction, until the total outflow of that type of material from that supply point decreases to equal to or below its inventory limit. For example, if the radiation monitoring items in supply point A2 are 20, and the total number of items transported from A2 in an individual plan reaches 30, the system may randomly select a transportation task from A2 to B1 to reduce the amount of radiation monitoring items transported by 10. If it still exceeds the limit, the system will continue to select tasks to reduce the amount of items transported until the total amount of items transported does not exceed 20.

[0056] Step S3222: Check whether the individual violates the demand point constraint. If it does, identify the transportation task in the individual code where the quantity of materials transported exceeds the corresponding demand, and gradually reduce the quantity of materials transported in the identified transportation task according to the second preset quantity until it does not violate the demand point constraint.

[0057] After completing the supply point inventory correction, further ensure that the scheduling plan can meet the minimum material requirements of all demand points and optimize and adjust for any potential oversupply. Continue to check the current individual, for each demand point (e.g., B1 and B2) and each type of material, accumulate the total amount of that type of material delivered to that point from all supply points, and compare it with the declared demand quantity of that demand point. There are two violations here: one is that the total amount is insufficient and does not meet the demand quantity; the other is that the total amount exceeds the demand quantity. This step mainly deals with the second situation, namely the oversupply problem, because the first situation (insufficiency) is usually indirectly caused or aggravated by reducing the amount shipped in step S3221, and more complex global adjustments may require the combination of other mechanisms. For the discovered oversupply, the system accurately locates all transportation tasks that transported the excess of that type of material to that demand point in the individual's code. Subsequently, using a preset, relatively fine adjustment step size (i.e., a second preset quantity, for example, decreasing by 1 unit each time), the transportation quantity of the corresponding material in these identified tasks is gradually reduced. The reduction process continues, with the total amount delivered recalculated after each adjustment, until the total amount of this type of material received by the demand point is precisely adjusted to meet or be no less than its demand. For example, if demand point B1 needs 50 radiation monitoring items, but the current plan delivers a total of 55 items, the system will find the tasks that have transported radiation monitoring items to B1 and reduce their quantity by 1 item one by one until the total amount is reduced to 50 items.

[0058] Step S3223: Based on the adjusted material transportation quantity of all transportation tasks, recalculate the number of vehicles required to execute each transportation task, and update the vehicle quantity field of all transportation task sequences in the individual code.

[0059] Based on the final material allocation plan determined in the first two steps, the vehicle resource requirements are recalculated to ensure the integrity and consistency of the coding information. Since the quantity of materials to be transported has been adjusted, the number of vehicles specified for each transport task in the original coding may no longer be accurate. The system needs to perform independent calculations for each transport task (i.e., each subsequence in the coding): First, multiply the quantity of each type of material required for the task by its corresponding unit volume (e.g., personal protective equipment 1 cubic meter / piece, radiation monitoring items 2 cubic meters / piece, medical supplies 1 cubic meter / piece) to obtain the total volume of each type of material. Then, sum these volumes to obtain the total volume of materials required for the transport task. Next, divide this total volume of materials by the rated cargo capacity of a single transport vehicle (e.g., 50 cubic meters). Since the number of vehicles must be an integer and sufficient to load all materials, the result of the division is rounded up to obtain the theoretical minimum number of vehicles required to execute the task. Finally, replace the original vehicle number field in the coding of the transport task subsequence with this newly calculated number of vehicles. For example, a transportation task originally coded as "[2, 30, 25, 15]" indicates that two vehicles will be dispatched to transport 30 pieces of personal protective equipment, 25 pieces of radiation monitoring equipment, and 15 pieces of medical supplies. The total volume of the supplies is calculated to be 30 × 1 + 25 × 2 + 15 × 1 = 95 cubic meters. The vehicle capacity is 50 cubic meters, so 95 / 50 = 1.9, rounded up to 2. Therefore, the vehicle number "2" may not need to be changed. However, if the amount of supplies decreases, the calculated number of vehicles may become 1, and the code needs to be updated to "[1, …]". Through this step, for each corrected individual, all decision variables (number of vehicles and amount of supplies) in its code achieve a logically consistent match based on real-world physical constraints, representing a fully feasible and resource-matched scheduling scheme.

[0060] Furthermore, step S323, based on a multi-objective optimization strategy, selects individuals from the current population and its repaired offspring population of the original emergency supplies dispatching task, including: Step S3231: Merge the current population of the original emergency material dispatching task with its repaired offspring population to obtain a merged population.

[0061] Specifically, after the evolutionary operation in step S321 and the repair operation in step S322, the original task yields a new set of feasible offspring individuals. At this point, the system merges this repaired offspring population with the current population of the original task before this round of evolution (i.e., the parent population). For example, if the initial population size is set to one hundred, the parent population has one hundred individuals, and the offspring population, after crossover and repair, also has one hundred individuals; the temporary population size formed after merging is two hundred. This merging strategy ensures that when selecting the next generation, not only are the new areas explored by the newly generated offspring considered, but also the proven superior genes (elite individuals) of the previous generation are preserved, thus avoiding the loss of excellent solutions.

[0062] Step S3232: Calculate the constraint violation value for each individual in the merged population.

[0063] A feasibility quality assessment is performed on all individuals in the merged population, distinguishing between fully feasible individuals and those that, although repaired, may still have minor constraint violations (which may occur under certain complex constraints or repair logic). The constraint violation severity value is a non-negative scalar value, calculated by quantifying and summing the degree to which an individual violates various uniform constraints. Specifically, each individual is checked for violations of vehicle number constraints, vehicle capacity constraints, supply point inventory constraints, and demand point demand constraints. For each constraint, the violation amount is calculated: for example, if a supply point is assigned more vehicles than it has, the excess amount is the violation amount; if the total volume of goods transported exceeds the total capacity of the assigned vehicles, the excess portion is the violation amount. All these violation amounts (zero for satisfied constraints) are summed to obtain the individual's total constraint violation severity value.

[0064] Step S3233: Perform non-dominated sorting on the individuals in the merged population, and determine the non-dominated level of each individual based on the function values ​​of the first objective function, the second objective function, and the third objective function.

[0065] Without requiring manually assigned weights, all feasible and some infeasible individuals in the merged population are stratified and categorized according to the natural superiority-inferiority relationships of solutions across multiple objectives. The ranking process is based on the Pareto dominance concept: for any two individuals, if one individual's value is no worse than the other individual's (i.e., less than or equal to) in all three objective functions (scheduling time, economic cost, unfairness), and is strictly better (i.e., strictly less than) in at least one of the objectives, then the former is said to dominate the latter. The system first identifies all individuals in the merged population that are not dominated by any other individual and marks them as the first non-dominated level (or frontier one). These individuals constitute the currently available set of optimal solutions. Then, these individuals are temporarily removed, and from the remaining individuals, all individuals that are not dominated by the other remaining individuals are identified again and marked as the second non-dominated level. This process is repeated until all individuals are assigned a non-dominated level.

[0066] Step S3234: Calculate the crowding distance between individuals with the same non-dominated level in the target space composed of the first objective function, the second objective function, and the third objective function.

[0067] Within the same non-dominated hierarchy, the distribution characteristics of individuals are further evaluated to promote the diversity of the next generation population and prevent the algorithm from converging to a narrow region on the Pareto front. Individuals classified into the same hierarchy are first independently ranked according to their values ​​on each objective function. For a given individual, its crowding distance is defined as the sum of the normalized distance differences on each objective function between its two nearest neighbors ("left" (individuals with slightly smaller objective values) and "right" (individuals with slightly larger objective values) after ranking, measuring the density of other individuals around that individual. If an individual is an extreme value (i.e., maximum or minimum) on a certain objective, it is usually assigned an infinitely large crowding distance (or a very large value) to maintain the extensibility of the front. A larger crowding distance indicates a more "open" environment around the individual, and retaining that individual helps maintain the population's distribution breadth in the objective space.

[0068] Step S3235: Based on the order of non-dominated hierarchy from high to low, and within the same non-dominated hierarchy based on the order of crowding distance from large to small, a predetermined number of individuals are selected from the merged population to form the next generation population for the original emergency material dispatching task.

[0069] This study integrates the non-dominated hierarchy and crowding distance information obtained in previous steps to implement a selection strategy that combines elite preservation with diversity maintenance. The selection process begins with the first non-dominated hierarchy. First, all individuals at this hierarchy are attempted to be included in the next generation. If the number of individuals at this hierarchy has reached or exceeded the predetermined next generation population size (e.g., 100), then only within this hierarchy are the required number of individuals selected, in descending order of crowding distance. If the number of individuals in the first hierarchy is less than 100, all of them are selected, and the process continues to the second non-dominated hierarchy, similarly selecting individuals based on descending crowding distance. This process continues until 100 individuals are selected. In comparisons, the constraint violation value is used as the primary criterion: fully feasible individuals (violation degree of zero) are prioritized over any infeasible individuals (violation degree of greater than zero); only when feasibility is equal are the non-dominated hierarchies compared.

[0070] Furthermore, in step S300, a knowledge transfer operation is performed during the evolutionary process from the simplified auxiliary task to the original emergency material dispatching task, resulting in an optimized population for the original emergency material dispatching task, including: Step S351: From the current population of the simplified auxiliary task, select all individuals that are not dominated by any individual in the current population of the original emergency material dispatch task, and form a set of transfer knowledge.

[0071] The auxiliary task population has undergone multiple rounds of single-objective (scheduling time) optimization, containing a large number of individuals that excel in reducing scheduling time. However, not all of these individuals are suitable for migration because the auxiliary task completely disregards economic costs and scheduling fairness. Some individuals may have very short time requirements, but at the cost of extremely high costs or highly unfair allocation. Directly introducing them could negatively impact the multi-objective balance of the original task. Therefore, a screening criterion needs to be established: comparing the multi-objective dominance of an auxiliary task individual with all individuals in the current population of the original task. If an auxiliary task individual is not dominated by any individual in the original task in any of the three objective functions (scheduling time, economic cost, and unfairness) (i.e., there is no original task individual that is no worse than the auxiliary individual in all three objectives, and is strictly better than the auxiliary individual in at least one objective), it means that the auxiliary individual provides unique value in at least one objective dimension (usually its focused scheduling time) that the original population has not yet reached or surpassed.

[0072] Step S352: For each individual in the knowledge transfer set, determine the corresponding individual in the current population of the original emergency material dispatching task that has the minimum cosine angle distance with respect to the objective function value. The cosine angle distance is used to measure the directional similarity between the two individuals in the target space composed of the first objective function, the second objective function, and the third objective function.

[0073] To find the most suitable "receiver" or "learning object" in the original task population for each high-quality "knowledge source" in the knowledge transfer set, the matching principle is based on the directional similarity between the two in the optimization target space, rather than simply the distance. Specifically, for an individual k in the knowledge transfer set, its three objective function values ​​are represented as a vector in a three-dimensional space. Similarly, each individual p in the current population of the original task is also represented as a vector in this space. Then, the cosine angular distance between individual k and each individual p is calculated. This distance is defined by the cosine of the angle between the two vectors, ranging from -1 to 1; the smaller the value (closer to -1), the more opposite the directions of the two vectors; the larger the value (closer to 1), the more consistent the directions; zero indicates orthogonality. This step selects the original task individual p with the largest cosine value (i.e., the smallest angle and the most similar direction) as the corresponding individual k. For example, the vector of the auxiliary individual k might point in the direction of "extremely short time and high cost". In that case, the individual in the original task population that points in the direction of "shorter time and slightly higher cost" (even if the absolute values ​​are different) will be selected because it is closest to the direction of improvement in the target space.

[0074] Step S353: For each pair of identified corresponding individuals, according to the preset learning rate, the information of the individuals in the transfer knowledge set is transferred to the corresponding individuals of the original emergency material dispatching task to generate new candidate individuals.

[0075] Here, "information" at the algorithm level typically refers to the encoding (genotype) representing the scheduling scheme. For a matched pair of individuals—the knowledge source individual k (from the auxiliary task) and the receiver individual p (from the original task)—the specific transfer operation is implemented through a linear interpolation process: the encoding of the new candidate individual k' is calculated using the formula k' = p + α × (kp), where α is a preset learning rate parameter randomly selected between 0 and 1. For example, if a certain gene position of p (such as the transport volume of protective items from A1 to B1) is 20, the corresponding gene of k is 30, and α is 0.5, then the corresponding gene of k' becomes 20 + 0.5 × (30 - 20) = 25. This transfer method is a gentle knowledge injection; it is not simply replacing p with k, but rather absorbing some features of k based on p, thus potentially producing a compromise new individual that absorbs the temporal optimization characteristics of k (reflected in the gene encoding) while retaining the advantages of p in other aspects (such as cost- and fairness-related gene patterns).

[0076] Step S354: If a new candidate individual dominates the corresponding individual of the original emergency material dispatching task, then replace the corresponding individual in the current population of the original emergency material dispatching task with the new candidate individual to obtain the optimized population of the original emergency material dispatching task after knowledge transfer optimization.

[0077] For each new candidate individual k' generated after matching and migration, and its corresponding original individual p, the system needs to evaluate the effect of this knowledge transfer. The evaluation criterion is Pareto dominance: if the new individual k' dominates the original individual p in a multi-objective sense (i.e., k' is no worse than p in terms of scheduling time, economic cost, and scheduling unfairness, and is strictly better than p in at least one objective), then it proves that the knowledge introduced from the auxiliary task has successfully produced a better solution. At this time, the system replaces the old individual p in the current population of the original task with this better new individual k'. If k' does not dominate p, or is even dominated by p, it means that this migration has failed to produce improvement or has degenerated. In this case, the migration result will be discarded, and the original individual p will remain in the population.

[0078] Simultaneously, step S400, based on preset decision rules, determines the final emergency material dispatching plan from the optimized population of the original emergency material dispatching tasks, including: Step S410: Determine whether an optimization instruction has been received for any of the first objective function, the second objective function, and the third objective function.

[0079] The "optimization instruction" here is a clear, structured technical parameter input that specifies the particular objective dimension that needs to be prioritized and optimized at the current decision-making moment. This instruction may originate from real-time commands from the superior emergency command system, or it may be preset by the decision-maker through a human-computer interface based on the phased priorities of the accident situation (such as prioritizing time control in the early stages of the accident, and focusing on cost or fairness in the later stages). The content of the instruction explicitly specifies the objective function number or name, such as "prioritize minimizing scheduling time (first objective function)" or "prioritize minimizing transportation costs (second objective function)." When executing step S400, the system will first check whether such a valid instruction exists. For example, in an emergency phase where the accident situation escalates and every second counts, the command center may issue a "time-optimal" instruction; while in the subsequent optimization of material allocation or the stage of strict budget control, a "cost-optimal" instruction may be issued.

[0080] Step S420: If no optimization instruction is received, execute the first rule: normalize the function values ​​of the first objective function, the second objective function, and the third objective function of each individual in the optimization population, calculate the comprehensive evaluation value of each individual, the comprehensive evaluation value is the sum of the normalized objective function values, and select the individual with the smallest comprehensive evaluation value as the final emergency material dispatch plan.

[0081] First, since the three objective functions (scheduling time, economic cost, and unfairness) have different dimensions and orders of magnitude (e.g., time is measured in hours, cost in yuan, and unfairness is dimensionless but has a very small value), direct summation and comparison are meaningless. Therefore, it is necessary to normalize the values ​​of all individuals in the optimization population for each objective. A common method is to find the maximum and minimum values ​​of each objective function in the population, and then map each individual's value for that objective to the range of 0 to 1, specifically (individual value - minimum value) / (maximum value - minimum value). After normalizing the three objectives, each individual obtains three normalized objective values ​​within the same numerical range. Subsequently, the three normalized objective values ​​of each individual are simply summed to obtain its "overall evaluation value." This value reflects the degree to which the individual deviates from the ideal point in terms of the three objectives on the normalized scale. For example, a small overall evaluation value for an individual means that its overall performance in terms of time, cost, and fairness is better after standardization, even if it is not the best in any single objective, but it is the best in terms of global trade-offs. The system ultimately selects the individual with the lowest overall evaluation value from the optimized population as the final execution plan.

[0082] Step S430: If an optimization instruction for a set objective function is received, then execute the second rule: directly select the individual with the smallest set objective function value from the optimization population as the final emergency material dispatch plan.

[0083] Once the system confirms receipt of an optimization instruction targeting a specific objective function (e.g., "minimize scheduling time"), it skips the complex comprehensive evaluation process and directly enters the targeted selection phase. The system iterates through each individual in the optimization population (i.e., the high-quality Pareto solution set obtained through multi-task co-evolution and knowledge transfer), reading and comparing its original function value (without normalization) on the objective function specified by the instruction. For example, if the instruction is "prioritize ensuring the shortest time," the system directly finds the individual with the smallest "scheduling time (first objective function)" value. If the instruction is "prioritize controlling the lowest cost," it finds the individual with the smallest "total transportation economic cost (second objective function)." This rule is logically direct and responsive, fully utilizing the characteristic that the optimization population itself is already a high-quality Pareto front solution—even when pursuing only a single objective optimum, the selected individual comes from a solution set that has been sufficiently optimized across multiple objectives. Therefore, the individual's performance on non-priority objectives is usually not too bad, avoiding extreme solutions. For example, the fastest solution selected under the "time-optimal" instruction is likely to have better cost and fairness than an unoptimized random solution. This demonstrates the advantages of targeted selection within high-quality frontier solution sets.

[0084] The implementation process of this invention will be described in detail below with a specific embodiment. In this example, the emergency material dispatching task involves 5 supply points and 2 demand points, and covers 3 types of transported goods (personal protective equipment, radiation detection and testing items, and medical supplies). The multi-objective emergency material dispatching method based on a multi-task evolutionary framework proposed in this invention is used to solve the dispatching scheme of this embodiment. The main logic diagram is as follows: Figure 2 As shown, the specific steps are as follows: S1. Obtain initial input information. Tables 1 and 2 are supply point and demand point information tables, respectively, where each row represents information about a supply point or demand point. In this example, the vehicle types are the same. Tables 3 and 4 are material and vehicle information, respectively, and Table 5 is the transportation route information.

[0085] Table 1 Supply Point Information Table 2 Demand Point Information Table 3 Material Information Table 4 Vehicle Information Table 5 Transportation Route Information S2. Build a multi-task framework. The steps are as follows: Construct multiple objective functions for the original emergency supplies dispatching task. There are three objective functions, including minimizing the dispatching time. Minimize scheduling economic costs Minimize scheduling unfairness .

[0086] ; The maximum transport time for each transport route in the scheduling plan is calculated using the following formula, without considering the departure time difference of multiple vehicles cooperating to transport the same batch of goods on the same route: ; in, Indicates the number of supply points; Indicates the number of demand points ； V represents the distance between supply point i and demand point j; V represents the vehicle's speed.

[0087] The sum of the products of the travel distance of all transport vehicles and the unit transportation cost is calculated using the following formula: ； Where C represents the unit distance transportation cost of the vehicle.

[0088] The sum of the demand imbalance and the supply imbalance is calculated using the following formula: ； in, and This represents the variance of the demand fulfillment rate and the supply rate. The formula is: ； ； in, For the first Supply rate at each supply point The average supply rate at the supply points. This is a weighting factor, and in this example, it is set to 0.8 to prioritize ensuring that the demand points meet the balance. by As the objective function in the assignment task.

[0089] Four constraints are established in the construction of the original emergency material dispatch task and the simplified auxiliary task: vehicle number constraint, vehicle capacity constraint, supply point constraint, and demand point constraint.

[0090] The vehicle number constraint means that the total number of vehicles assigned to a supply point to transport goods cannot exceed the total number of vehicles already in the supply point.

[0091] , ; in, This represents the number of vehicles at supply point i. This represents the number of vehicles involved in transportation from supply point i to demand point j.

[0092] Vehicle capacity constraints refer to the requirement that the total volume of transported goods cannot exceed the vehicle's maximum capacity.

[0093] , ; in, Indicates the first k The individual volume of a type of item; This represents the quantity of the k-th type of goods transported from supply point i to demand point j; Indicates from the supply point i To the point of demand j The rated capacity of the vehicle used.

[0094] Supply point inventory constraints mean that the quantity of goods supplied by a supply point cannot exceed the supply point's inventory of that type of goods.

[0095] ， ； in, This represents the quantity of the k-th type of goods transported from supply point i to demand point j. Indicates supply point i middle k Inventory levels of this type of product; This indicates the number of types of goods at the supply point.

[0096] The demand point constraint means that the total quantity of all goods accepted by a demand point must not be less than the quantity of goods it requires.

[0097] ， ； in, Indicate demand points j For the first k The demand for such materials.

[0098] S3, Multi-task collaborative evolution.

[0099] The material scheduling scheme in this embodiment is solved using the multi-objective optimization algorithm based on a multi-task evolutionary framework proposed in this invention. The main logic diagram is as follows: Figure 2 As shown, the specific steps are as follows: 1. The initial population of individuals that generates the original task. and the initial individual population for auxiliary tasks And generate the external population from the original task. .

[0100] In the initial population, each individual represents a scheduling scheme, such as Figure 3 As shown, an individual is encoded as a combination of multiple sequences. Each sequence represents a transportation task from a supply point to a demand point. For example, the first sequence [2, 20, 10, 15] in an individual represents a transportation task from supply point A1 to demand point B1, assigning 2 vehicles to transport 20 personal protective equipment items, 10 radiation detection and testing items, and 10 medical supplies. The number of sequences contained in an individual is equal to the number of demand points N. Number of supply points: M. Encoding length for a single individual: N. M The number of elements in a single sequence.

[0101] N individuals are generated for the original task and the auxiliary task respectively to form the initial population. and , where N is the initial population size. The number of transported goods and the number of vehicles in the N individuals are randomly generated integers within the supply point's inventory.

[0102] Suppose that the objective function values ​​of individuals a and b are [0.299, 10134.9, 0.006] and [0.299, 1240.7, 0.0013], respectively. If a is no worse than b in all three objective function values ​​and is better than b in at least one objective function, then individual a is said to dominate individual b. Figure 4 A schematic diagram illustrating the non-dominated ranking results of individuals in a three-objective minimization problem is provided. For example... Figure 4 As shown, an individual with a non-dominant level of 1 is not dominated by any individual, and is therefore also called a non-dominant individual. Non-dominated individuals are stored in middle.

[0103] 2. Analyze the parent populations in the original task and the auxiliary task respectively. and Crossover and mutation operators are used to generate offspring populations. and The specific steps are as follows: (1) To Individuals are randomly paired up.

[0104] (2) Each pair of parent individuals follows a certain crossover probability New offspring individuals are generated by single-point crossover, and these individuals are aggregated into a temporary offspring set. Figure 5 This is a diagram illustrating a single-point intersection.

[0105] (3) Apply the mutation probability to individuals in the temporary offspring set. Randomly modify the quantity of transported goods to obtain the final offspring population. Assume the individuals to be mutated are [1,2,10,10| 2,5,7,9| 1,20,10,10| 2,20,20,20|3,10,10,10| 2,10,10,10| 3,10,10,10| 1,10,5,10| 2,10,0,10| 1,10,0,5]. The specific mutation method is: for the quantity of items in the 2nd to 4th positions of each sequence (excluding the number of vehicles in the 1st position), mutate according to the mutation probability. Randomly select whether to mutate; if mutated, increase or decrease the item quantity value by 10% (randomly selected in the ± direction, the result is rounded to the nearest integer and not less than 0). For example, in the sequence [2,5,7,9], item 1 has a quantity of 5, which could become 5 + 0.5 after mutation. 6 or 5-0.5 5 (remains unchanged after rounding); Item 2 quantity 7, after mutation, may become 7 + 0.7. 8 or 7-0.7 6; After the item's quantity 3 is 9, it may become 9 + 0.9. 10 or 9-0.9 8; In the sequence [2,10,0,10], the quantity of item 2 is 0, because it cannot be negative, it can only remain 0. After processing the quantity of all items according to this rule, the final offspring population is formed. .

[0106] (4) Repeating the same crossover and mutation process, the offspring population for the auxiliary task is eventually obtained. .

[0107] 3. To The individual components undergo infeasible solution repair operations. The specific steps are as follows: (1) First, the supply point inventory constraint is corrected, the amount of inventory violated by the individual is counted, and the transportation task is randomly selected to reduce the minimum (10, the amount of inventory violated) until the inventory constraint is not violated.

[0108] (2) Secondly, the demand point constraint is modified. First, the individuals are traversed to find transportation tasks that exceed the acceptance limit of 5. For each transportation task, the transportation volume is gradually reduced by 1 unit until the individual does not violate the demand point constraint. (3) Recalculate the vehicles required for the current individual transportation task and modify the vehicles required for the transportation task in the individual's genes.

[0109] 4. From and Select the N best individuals as the population after iteration. This completes the population update operation of the original task. The specific steps are as follows: (1) The parent population With offspring population They merge to form a temporary population of size 2N.

[0110] (2) Calculate the constraint violation value for each individual in the temporary population. Assume the individuals are [1,2,10,10 |2,25,7,9 |1,20,10,10 |2,20,20,20 |3,15,10,10 |2,15,10,10 |3,20,10,10 |1,30,5,10 |2,20,0,10 |1,30,0,5]. Regarding the vehicle number constraint: Supply point A2 exceeds 1 vehicle, A3 exceeds 3 vehicles, A4 exceeds 1 vehicle, and A5 exceeds 2 vehicles, with a total violation of 7. Regarding the vehicle capacity constraint: The total volume of all sequences is ≤ number of vehicles × 50, with a total violation of 0. Regarding the supply point inventory constraint: Supply point A2 exceeds 10 radiation monitoring items, A4 exceeds 10 medical treatment items, and A5 exceeds 20 personal protective equipment, with a total violation of 40. Regarding the demand point constraint: Demand point 1 is short 10 radiation monitoring items, with a total violation of 10. Therefore, the violation value of this individual constraint is 7 + 0 + 40 + 10 = 57.

[0111] (3) Perform non-dominated sorting on the temporary population to obtain the non-dominated level of each individual. The smaller the non-dominated level, the closer the individual is to the ideal Pareto front.

[0112] (4) For individuals in the same non-dominated level, the individual is ranked according to the crowding distance in the target space.

[0113] This invention uses crowding distance to measure the neighborhood density of an individual; a larger crowding distance indicates better individual performance. For example... Figure 6 As shown, individual The crowding distance is the sum of the distances between the two individuals to its left and right on each target. The crowding distance at the boundary point is set to infinity.

[0114] (5) Select N feasible solutions from the temporary population. If the number of feasible solutions in the temporary population exceeds N, prioritize individuals with smaller non-dominated levels. If the number of individuals in the current frontier exceeds the population size N, select from largest to smallest based on crowding distance until N individuals are selected. If the number of feasible solutions in the temporary population is less than N, continue selecting from infeasible solutions based on non-dominated level and crowding distance until N individuals are selected.

[0115] 5. Merge and And select feasible individuals with a non-dominant level of 1 as new ARCo.

[0116] 6. From and Select the N best individuals as the population after iteration. This completes the population update operation for the auxiliary task. The specific steps are as follows: (1) The parent population With offspring population They merge to form a temporary population of size 2N.

[0117] (2) Using transportation time as the objective function Using this as the sole evaluation criterion, all individuals are sorted in ascending order of their function values.

[0118] (3) Select the N individuals with smaller function values.

[0119] 7. A knowledge transfer strategy is employed to extract effective information from the auxiliary task to guide the evolutionary process of the original task. A schematic diagram of the knowledge transfer strategy used in this invention is shown below. Figure 7 As shown.

[0120] The specific steps for knowledge transfer from auxiliary tasks to the original tasks are as follows: (1) From Select the ones that are not Individuals controlled by all individuals in the group are used for knowledge transfer, and the selected individuals are stored in K.

[0121] (2) For each individual in K ,from Select with The individual with the closest cosine angle distance The formula for calculating the cosine angle distance is as follows: ; in, for and The inner product, and They are respectively and The modulus of C. and The cosine angle value of the angular distance; the smaller the C value, the stronger the angular distance. and The smaller the angular distance between them.

[0122] (3) Calculation .in, The learning rate is randomly selected from [0,1]; if Dominate Then use replace .

[0123] 8. Based on the decision-maker's needs, output a scheduling plan for the original task. The specific steps are as follows: (1) If the decision-maker has no preference for the objective function, then first... The individual's objective function value is normalized, and then the individual with the smallest sum of the three objective values ​​is selected.

[0124] (2) If the decision-maker has a clear preference, then from The output is the individual with the smallest objective function value.

[0125] In the embodiments of the present invention, the parameters of the multi-objective evolutionary algorithm are set as follows: the population size N is 100; the crossover and mutation probabilities are 0.2 and 0.8, respectively; and the maximum number of function evaluations is 20000.

[0126] The effects of this invention can be further illustrated by the following simulation experiments: 1. Experimental conditions Simulation was performed using MATLAB R2024a on an Apple M2 Pro CPU, 16GB of RAM, and macOS system.

[0127] 2. Experiment Content This invention provides a solution for the scheduling scheme in the above-mentioned emergency material scheduling implementation example with 5 supply points, 2 demand points, 3 types of materials, 4 constraints, and 3 optimization objectives.

[0128] 3. Experimental Results Based on the above experimental conditions and parameter settings, 30 repeated experiments were conducted to test the performance of the present invention and four mainstream constrained multi-objective algorithms, namely NSGAIIARSBX, MTCMO, MCCMO and CMOSMA, in the emergency material scheduling problem. The results are shown in Table 6.

[0129] As shown in Table 6, the average running time of this invention is 0.588s, which is better than other algorithms. Furthermore, this invention addresses the three core optimization objectives of emergency material dispatching (…). Scheduling time, Dispatch economic costs, The optimization effect on scheduling unfairness is outstanding. In summary, the multi-objective evolutionary algorithm of this invention has significantly better overall performance than existing mainstream constrained multi-objective algorithms in emergency material scheduling problems. It has high efficiency, stability and practicality, and can be widely applied to similar multi-constraint, multi-objective emergency resource scheduling scenarios.

[0130] Table 6 Experimental Results Accordingly, please refer to Figure 8 A second aspect of this invention provides an emergency multi-objective material scheduling device based on a multi-task evolutionary framework, which determines a scheduling scheme based on the aforementioned emergency multi-objective material scheduling method based on a multi-task evolutionary framework, including: Data acquisition module 1 is used to acquire the initial information required for emergency material dispatch; Framework building module 2 is used to build a multi-task evolution framework that includes an original emergency material dispatching task and a simplified auxiliary task. The original emergency material dispatching task is a material dispatching model that contains at least two optimization objectives, and the simplified auxiliary task is a material dispatching model that is related to the original emergency material dispatching task but with simplified objectives. Population optimization module 3 is used to perform population iterative optimization based on evolutionary computation for the original emergency material dispatching task and the simplified auxiliary task in the multi-task evolutionary framework, and to perform knowledge transfer operation from the simplified auxiliary task to the original emergency material dispatching task during the evolution process, so as to obtain the optimized population of the original emergency material dispatching task. The scheme selection module 4 is used to determine a final emergency material dispatch scheme from multiple alternative dispatch schemes included in the optimized population of the original emergency material dispatch task according to preset decision rules.

[0131] Accordingly, a third aspect of the present invention provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the above-described emergency multi-objective material scheduling method based on a multi-task evolution framework.

[0132] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-described emergency multi-objective material scheduling method based on a multi-task evolution framework.

[0133] The embodiments of the present invention aim to protect an emergency multi-objective material scheduling method and device based on a multi-task evolution framework, which has the following effects: 1. By constructing a multi-task evolutionary framework consisting of an original multi-objective scheduling task and a simplified single-objective auxiliary task, this approach effectively addresses the inherent problems of slow convergence and susceptibility to local optima in traditional single-task optimization methods under complex emergency scenarios. The auxiliary task is simplified based on a key objective of the original task, ensuring a high degree of correlation between tasks and providing stable and efficient search guidance for the evolution of the original task. This significantly accelerates the overall optimization process, enabling the algorithm to quickly generate feasible, high-quality scheduling schemes within a short emergency response time. 2. By adopting a directional knowledge transfer strategy based on non-dominant relations and cosine angular distance, the "negative transfer" problem that may occur in multi-task optimization is fundamentally avoided. This strategy selects advantageous knowledge individuals from auxiliary tasks and achieves precise transfer by measuring their similarity to the original task individuals in the target space direction. This ensures that the transferred knowledge can effectively improve the quality of the original task population rather than interfere with its evolutionary direction, thereby improving search efficiency while ensuring the superiority and reliability of the final solution set. 3. The system integrates an infeasible solution repair mechanism for scheduling constraints with a multi-objective selection strategy based on non-dominated ordering and crowding distance, jointly ensuring the generation of diverse and balanced high-quality scheduling schemes under complex constraints. The repair mechanism can automatically and efficiently adjust scheduling schemes that violate realistic constraints into feasible solutions, ensuring the practicality of all evolutionary paths; while the multi-objective selection strategy accurately selects the Pareto optimal solution set from feasible solutions that achieves the best balance among multiple objectives such as time, cost, and fairness, providing decision-makers with rich and high-quality alternatives, greatly enhancing the scientific nature and flexibility of scheduling decisions.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An emergency multi-objective material scheduling method based on a multi-task evolutionary framework, characterized in that, Includes the following steps: Obtain the initial information required for emergency supplies dispatch; Based on the initial information, a multi-task evolution framework is constructed, which includes the original emergency material dispatching task and the simplified auxiliary task. In the multi-task evolutionary framework, population iterative optimization based on evolutionary computation is performed on the original emergency material dispatching task and the simplified auxiliary task respectively, and a knowledge transfer operation from the simplified auxiliary task to the original emergency material dispatching task is performed during the evolution process to obtain the optimized population of the original emergency material dispatching task. Based on preset decision-making rules, the final emergency material dispatching scheme is determined from the optimized population of the original emergency material dispatching tasks.

2. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to claim 1, characterized in that, The construction of the multi-task evolutionary framework includes: For the original emergency material dispatching task, a multi-objective optimization model is constructed, which includes: a first objective function corresponding to the dispatching time, a second objective function corresponding to the total economic cost of transportation, and a third objective function corresponding to the dispatching unfairness. For the simplified auxiliary task, a single-objective optimization model is constructed, which includes a single objective function corresponding to scheduling time, total economic cost of transportation, and scheduling unfairness. A unified set of constraints is set for the multi-task evolution framework, which includes: vehicle number constraint, vehicle capacity constraint, supply point inventory constraint, and demand point demand constraint.

3. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to claim 2, characterized in that, Within the multi-task evolutionary framework, population-based iterative optimization based on evolutionary computation is performed on both the original emergency material dispatching task and the simplified auxiliary task, including: A first initial population corresponding to the original emergency material dispatching task and a second initial population corresponding to the simplified auxiliary task are generated respectively. Each individual in the first initial population and the second initial population represents a complete dispatching scheme through an encoding structure. Based on the first initial population and the second initial population, the original emergency material dispatching task and the simplified auxiliary task are subjected to multiple rounds of co-evolution; wherein, each round of co-evolution includes: executing the population evolution and screening process of the original emergency material dispatching task and the simplified auxiliary task in parallel, and repairing infeasible solutions generated during the evolution of the original emergency material dispatching task based on the unified constraints.

4. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to claim 3, characterized in that, The process of performing multiple rounds of collaborative evolution on the original emergency material dispatching task and the simplified auxiliary task based on the first initial population and the second initial population includes: Evolutionary operations are performed on the current population of the original emergency material dispatching task and the current population of the simplified auxiliary task, respectively, to generate corresponding offspring populations; An infeasibility solution repair operation is performed on the offspring population of the original emergency material dispatch task to ensure that each individual meets the unified constraint condition. Based on a multi-objective optimization strategy, individuals are selected from the current population and its repaired offspring population of the original emergency material dispatching task to form its next generation population. Based on a single-objective optimization strategy, individuals are selected from the current population and its offspring population of the simplified auxiliary task to form the next generation population.

5. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to claim 4, characterized in that, The step of performing evolutionary operations on the current population of the original emergency material dispatching task and the current population of the simplified auxiliary task, respectively, includes: For any given task, the current population is randomly paired with its individuals; For each pair of individuals, a single-point crossover operation is performed according to a preset crossover probability, and the encoding segments of the paired individuals after the randomly selected crossover point are swapped to generate intermediate offspring individuals. For all the intermediate offspring individuals, a mutation operation is performed according to a preset mutation probability. The mutation operation randomly selects the quantity of material transportation in the code of the intermediate offspring individual and increases or decreases it according to a preset adjustment range so that the adjusted quantity is a non-negative integer, thus obtaining the offspring population of the current task.

6. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to claim 4, characterized in that, The infeasibility solution repair operation performed on the offspring population of the original emergency material dispatch task includes: For each individual in the offspring population, determine whether the individual violates the supply point inventory constraint. If it does, randomly select the transportation task corresponding to the individual's code and gradually reduce the amount of materials transported by the individual according to a first preset quantity until it does not violate the supply point inventory constraint. Check whether the individual violates the demand point constraint. If it does, identify the transportation task in the individual code where the quantity of materials transported exceeds the corresponding demand, and gradually reduce the quantity of materials transported in the identified transportation task by a second preset quantity until it does not violate the demand point constraint. Based on the adjusted quantity of materials transported for all transport tasks, the number of vehicles required to execute each transport task is recalculated, and the vehicle quantity field for all transport task sequences in the individual code is updated.

7. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to claim 4, characterized in that, The multi-objective optimization strategy for selecting individuals from the current population and its repaired offspring population of the original emergency material dispatching task includes: The current population of the original emergency material dispatching task is merged with its repaired offspring population to obtain a merged population. Calculate the constraint violation degree value for each individual in the merged population; The individuals in the merged population are sorted by non-dominated order, and the non-dominated level of each individual is determined based on the function values ​​of the first objective function, the second objective function, and the third objective function. Calculate the crowding distance of the individuals with the same non-dominated hierarchy in the target space composed of the first objective function, the second objective function, and the third objective function; Based on the order of the non-dominated hierarchy from high to low, and within the same non-dominated hierarchy based on the order of the crowding distance from large to small, a predetermined number of individuals are selected from the merged population to form the next generation population for the original emergency material dispatching task.

8. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to any one of claims 1-7, characterized in that, The process of performing knowledge transfer operations from the simplified auxiliary task to the original emergency material dispatching task during evolution, to obtain an optimized population for the emergency material dispatching task, includes: From the current population of the simplified auxiliary task, select all individuals that are not dominated by any individual in the current population of the original emergency material dispatch task, and form a set of transfer knowledge. For each individual in the set of transferred knowledge, determine the corresponding individual in the current population of the original emergency material dispatching task that has the minimum cosine angle distance with respect to the objective function value. The cosine angle distance is used to measure the directional similarity between two individuals in the target space composed of the first objective function, the second objective function, and the third objective function. For each pair of corresponding individuals, the information of the individuals in the knowledge transfer set is transferred to the corresponding individuals of the original emergency material dispatching task according to the preset learning rate, thereby generating new candidate individuals. If the new candidate individual dominates the corresponding individual of the original emergency material dispatching task, then the new candidate individual replaces the corresponding individual in the current population of the original emergency material dispatching task, resulting in an optimized population of the original emergency material dispatching task after knowledge transfer optimization.

9. The emergency multi-objective material scheduling method based on a multi-task evolutionary framework according to any one of claims 1-7, characterized in that, The step of determining the final emergency material dispatching scheme from the optimized population of the original emergency material dispatching tasks according to preset decision rules includes: Determine whether an optimization instruction has been received for any of the first, second, and third objective functions; If the optimization instruction is not received, the first rule is executed: the function values ​​of the first objective function, the second objective function and the third objective function of each individual in the optimization population are normalized respectively, the comprehensive evaluation value of each individual is calculated, the comprehensive evaluation value is the sum of the normalized objective function values, and the individual with the smallest comprehensive evaluation value is selected as the final emergency material dispatch plan. If an optimization instruction for a given objective function is received, the second rule is executed: the individual with the smallest value of the given objective function is directly selected from the optimization population as the final emergency material dispatch plan.

10. An emergency multi-objective material dispatching device based on a multi-task evolutionary framework, characterized in that, The emergency multi-objective material scheduling method based on the multi-task evolutionary framework as described in any one of claims 1-9 determines the scheduling scheme, including: The data acquisition module is used to acquire the initial information required for emergency material dispatch. The framework construction module is used to construct a multi-task evolution framework based on the initial information. The multi-task evolution framework includes the original emergency material dispatching task and the simplified auxiliary task. The population optimization module is used to perform population iterative optimization based on evolutionary computation on the original emergency material dispatching task and the simplified auxiliary task respectively in the multi-task evolutionary framework, and to perform knowledge transfer operation from the simplified auxiliary task to the original emergency material dispatching task during the evolution process, so as to obtain the optimized population of the original emergency material dispatching task. The scheme selection module is used to determine the final emergency material dispatch scheme from the optimized population of the original emergency material dispatch tasks according to preset decision rules.

Images