A distributed production material distribution scheduling method

Abstract

The application discloses a kind of distributed production material distribution scheduling method, it is applicable to multiple warehouse, multiple factory, multiple material and multiple vehicle collaborative distribution scene.This method minimizes each factory maximum material supply delay as target, constructs scheduling calculation model;Adopt three-layer joint coding structure including material allocation layer, batch loading layer and path optimization layer, through linkage reconstruction mechanism guarantee that scheduling scheme meets demand, capacity and time consistency constraint.Solving by using knowledge feedback intelligent optimization algorithm, establish material allocation, batch loading and path structure knowledge base;Then realize global search by layered crossover, carry out knowledge-driven local search to elite individual for delay bottleneck, replenishment waiting and path inefficiency, use high-quality fragment in knowledge base to feedback repair and dynamic update to poor individual.The application can collaboratively optimize material allocation, vehicle loading and distribution path, improve material supply timeliness and distributed production logistics scheduling efficiency.

Description

Technical Field

[0001] This invention relates to the field of production logistics scheduling and intelligent optimization technology in distributed manufacturing, and in particular to a distributed production material distribution scheduling method, belonging to the fields of intelligent manufacturing, smart logistics and operations optimization technology. Background Technology

[0002] To promote the high-end, intelligent, and green development of the manufacturing industry, combining information technology and artificial intelligence to achieve intelligent and green manufacturing has become an important means to improve the economic efficiency of the manufacturing industry, optimize resource allocation, and reduce production costs. With the continuous evolution of collaborative production methods and supply organization models among enterprises, distributed manufacturing is gradually becoming an important organizational form of modern manufacturing systems. This manufacturing model can improve the responsiveness of the manufacturing system to market demand fluctuations and reduce the management risks associated with centralized production and supply methods through collaboration among multiple warehouses and factories.

[0003] In a distributed manufacturing environment, materials, as a crucial foundational resource for production activities, directly impact the execution of factory production plans and order fulfillment capabilities through their timely and accurate supply. Especially in scenarios where multiple warehouses collaboratively supply various materials to multiple factories, a reasonable material distribution and scheduling scheme not only affects the effective utilization of warehouse inventory resources but also influences vehicle transportation efficiency, factory material shortage risks, and overall production delay levels. Furthermore, distributed production material distribution and scheduling not only needs to determine the material allocation relationship between warehouses and factories but also needs to consider various complex factors such as vehicle batch loading, vehicle capacity constraints, vehicle batch sequence, delivery route optimization, and replenishment waiting times. Material allocation, vehicle batch planning, and route optimization are interconnected: warehouse allocation results affect vehicle loading and route structure, vehicle batch arrangement affects factory arrival times, and route sequence affects material arrival times and delay levels. Therefore, developing efficient material allocation, vehicle scheduling, and route planning schemes has become one of the key issues in distributed manufacturing management decision-making. This problem is characterized by strong multidimensional decision coupling, complex constraints, large scale, and high difficulty in solving. Traditional mathematical programming methods struggle to obtain high-quality solutions to large-scale problems within a finite timeframe, while existing intelligent optimization methods generally suffer from insufficient knowledge utilization, limited search efficiency, and weak local improvement capabilities. Therefore, there is an urgent need for a distributed production material distribution scheduling method for collaborative scenarios involving multiple warehouses, factories, materials, and vehicle batches. This method aims to achieve collaborative optimization of material allocation, vehicle batch loading, and transportation routes, reduce the risk of material supply delays in factories, and improve the efficiency and quality of material distribution scheduling in a distributed manufacturing environment. Summary of the Invention

[0004] The technical objective of this invention is to propose a distributed production material distribution scheduling method based on a knowledge feedback intelligent optimization algorithm. This method addresses the challenges of highly coupled material allocation, vehicle batch loading, and route optimization under constraints involving multiple warehouses, factories, materials, and vehicles. Traditional mathematical programming methods struggle to solve large-scale problems within a limited timeframe, and existing intelligent optimization methods suffer from insufficient knowledge utilization and limited local development capabilities, lacking effective and efficient scheduling methods. This invention constructs a distributed production material distribution scheduling computational model considering warehouse inventory, replenishment waiting time, vehicle capacity, and delivery route constraints. It designs a three-layer joint encoding and linkage reconstruction mechanism and proposes a knowledge feedback intelligent optimization algorithm. This algorithm can quickly obtain high-quality scheduling schemes with minimal delays in maximum material supply to the factory, achieving collaborative optimization of material resource allocation and distribution processes in a distributed manufacturing environment.

[0005] To address the problems existing in the prior art, this invention provides a distributed production material distribution scheduling method. It constructs a distributed production material distribution scheduling calculation model, establishes scheduling scheme coding and reconstruction rules, uses hybrid heuristic rules to construct an initial population, and optimizes the quality of the population solution through hierarchical crossover, knowledge base feedback repair, and knowledge-driven local search, ultimately obtaining an optimized distributed production material distribution scheduling scheme.

[0006] Its features include the following steps:

[0007] (1) Construct a distributed production material distribution scheduling calculation model, specifically as follows:

[0008] have Each factory needs Various materials, each factory For each material The demand is The materials are from One warehouse supplies, warehouse Medium material The inventory level is , warehouse To the factory Distributed materials Quantity is recorded as If warehouse If the inventory of materials is insufficient, replenishment is required, and the replenishment time is... ;storehouse The materials in the middle are made of Delivery by vehicles, each vehicle In each batch To the factory Transported materials Quantity is recorded as Each material The unit mass is The maximum load capacity of the vehicle is During transportation, the vehicle departs from the node. To the node The time is , warehouse The The car in Batch arrives at factory The time is recorded as ,factory The material supply deadline is The optimization objective is to minimize the material supply delays of all factories, as shown in (1);

[0009] (1)

[0010] in, For the factory The calculation method for the material supply delay is shown in (2):

[0011] (2)

[0012] in, For the factory The arrival time of all materials is calculated as shown in (3):

[0013] (3)

[0014] To meet the material needs of all factories, the factory The The allocation of each material should meet the requirements shown in (4):

[0015] (4)

[0016] To clarify the warehouse The The preparation time for each type of material is calculated, and the replenishment quantity is shown in (5):

[0017] (5)

[0018] To meet vehicle loading capacity limitations, calculate the warehouse vehicles In batch The total load mass is And satisfy the capacity constraints, as shown in (6) and (7):

[0019] (6) (7)

[0020] (2) Construct a three-layer joint coding structure and construction rules for each layer, and construct a material allocation layer. Batch loading layer and path optimization layer Three-layer joint coding structure Material distribution layer This indicates the supply quantity relationship of various materials between the warehouse and the factory, and the batch loading layer. This represents the loading plan for warehouse vehicles in each batch, and the route optimization layer. Indicate the access node order for each batch of warehouse vehicles; establish the construction rules for each layer, including given... structure and Given structure and Given structure When any layer of coding changes, the coding of the remaining layers is updated synchronously through the construction mechanism to ensure the feasibility of the scheduling scheme in terms of material demand, capacity constraints and time consistency.

[0021] (3) A knowledge feedback intelligent optimization algorithm is proposed, which includes five stages: population initialization method, knowledge base construction, global exploration, knowledge-driven local search, knowledge base feedback and update;

[0022] A hybrid initialization method based on random heuristics and urgency priority heuristics is used to construct a structure with a size of The initial population; the random heuristic generates a material allocation matrix by randomly allocating warehouse supply quantities. and call the constructor to generate and The urgency-first heuristic calculates the urgency of demand based on the factory's material deadline and transportation time, prioritizing factories with high urgency by selecting warehouses with short transportation times and sufficient inventory, and generating a material allocation matrix. and call the constructor to generate and ;

[0023] After sorting the initial population according to factory delay fitness, it was evenly divided into dominant and suboptimal individuals. High-quality local structural fragments from the three-layer coding were extracted from the dominant individuals to construct material allocation knowledge bases, batch loading knowledge bases, and path structure knowledge bases, respectively. The material allocation knowledge base is based on factory-material pairs. Using the index, extract all warehouse allocation quantities corresponding to the factory-material pair in the dominant individual. Store in the knowledge base; batch load the knowledge base using a warehouse-vehicle-batch combination. Using this as an index, extract the transport volume of this batch for all types of materials across all factories from the dominant individual. Stored in the knowledge base; the path structure of the knowledge base is a combination of warehouse-vehicle-batch. As an index, extract the factory access node sequence corresponding to this batch from the dominant individual as a path segment, and store it in the knowledge base along with the total path travel time and the maximum delay value of the source individual;

[0024] For each individual in the current population, it is paired with another individual randomly selected from the population with a certain probability, and a hierarchical crossover operation is performed targeting the material allocation layer, batch loading layer, and path optimization layer to generate offspring individuals. The hierarchical crossover operation includes: material allocation layer crossover, which uses the allocation submatrix corresponding to the factory or material subset as the exchange fragment, and exchanges the material allocation fragments of the two parents at the corresponding index positions to obtain a new allocation scheme. and call the constructor to generate and The batch loading layer uses the warehouse-vehicle-batch combination as the exchange unit, swapping the loading segments of the two parent generations in the selected batch to obtain a new batch loading scheme. and call the constructor to generate The path optimization layer uses either direct inheritance or sequential interleaving within a single path to reorganize the factory access sequence, resulting in a new path code. and call the constructor to generate and Combine the parent and offspring generations, rank them according to the factory's delay adaptability evaluation, and retain the top ones. Individuals form a new population;

[0025] ④ Perform a knowledge-driven local search on the top 20% of elite individuals in the current population, and perform fine-tuning on each individual by sequentially executing the following local search operators:

[0026] Material reallocation for factories facing delays: When an individual has a replenishment situation, identify the batch that is delayed due to replenishment waiting, attempt to transfer part of the demand for replenished materials to other warehouses with sufficient inventory, and reconstruct the relevant layer codes in conjunction with the process. If the replenishment waiting time is reduced and the maximum delay is reduced, the new solution is accepted.

[0027] Vehicle reassignment for replenishment waiting: Identify the routes with the longest total travel time and find the factories with the largest delays on those routes. Find the distance to the factory Other nearby warehouses Try to turn the factory Some or all of the demand will be from the original warehouse. Moved to a warehouse The batch loading layer and path optimization layer are reconstructed in conjunction; if the maximum delay is reduced, a new solution is accepted.

[0028] Path re-optimization for delays: Identify the bottleneck path that causes the factory's maximum delay. For the current bottleneck path, try to shorten the total travel time of the path by swapping nodes or using the 2-opt method. If the maximum delay is reduced, accept the new solution.

[0029] ⑤ High-quality coded knowledge is fed back to the current population through the knowledge base. Specifically, several poorly performing individuals with low evaluation function rankings are selected from the current population, and the current material supply delay for each factory is calculated. The factory with the largest delay value is recorded as the [factory name missing]. Factory-material pair As the target index in the material allocation knowledge base, it is the warehouse-vehicle-batch combination with the latest arrival time. As the target index in the batch loading knowledge base and path structure knowledge base, each layer of coding selects candidate fragments from the corresponding knowledge base that are valid with the current individual index, consistent with the requirements, and have better recorded delay values. The best candidate fragment is embedded into the current individual, and the constructor is called to reconstruct the coding of the remaining layers. After the individual is improved, it is re-evaluated. If the maximum delay is reduced, the improvement result is accepted; otherwise, the original individual is retained. After all the poor individuals complete the knowledge feedback, a new population is formed. The top-ranked individuals in the evaluation function are selected from the new population as the elite solution set. High-quality local structure fragments are extracted from their material allocation layer, batch loading layer, and path optimization layer, respectively, and the material allocation knowledge base, batch loading knowledge base, and path structure knowledge base are updated. The newly extracted fragments are deduplicated. If the knowledge base capacity exceeds the limit, the fragment with the lowest score or that has not been used for several consecutive generations is eliminated.

[0030] (4) Repeat step (3) -⑤ The global exploration, knowledge-driven local search, knowledge base feedback and update are performed until the preset termination conditions are met, and the optimal scheduling scheme is output. The optimal scheduling scheme includes the allocation results of each warehouse to each factory for various types of materials, the loading scheme of each batch of each vehicle in the warehouse, the access path sequence of each batch of vehicles, and the actual arrival time and material supply delay of each factory.

[0031] The features and beneficial effects of this invention are as follows:

[0032] 1. This invention establishes a distributed production material distribution scheduling method that considers warehouse inventory, replenishment waiting time, vehicle capacity and route constraints, which can comprehensively describe the material supply and distribution process in scenarios involving multiple warehouses, multiple factories, multiple materials and multiple vehicles.

[0033] 2. This invention proposes a three-layer joint coding structure and a linkage reconstruction mechanism, which can uniformly represent three types of mutually coupled decision variables: material allocation, batch loading, and path optimization, and automatically restore the consistency and feasibility of the other two layers after any layer is modified.

[0034] 3. This invention proposes an initial population construction method based on hybrid heuristics, which can take into account both the quality of the initial solution and the diversity of the population, thus providing a better starting point for subsequent evolutionary search.

[0035] 4. This invention proposes a hierarchical fragment evaluation function and a hierarchical crossover mechanism, which can effectively preserve and propagate excellent local structures that have a positive effect on reducing delays during the evolution process, thereby improving the efficiency of global search.

[0036] 5. This invention proposes a hierarchical knowledge base construction and feedback repair mechanism, which can extract, store and reuse excellent structural knowledge generated during the evolution process, and realize feedback optimization that combines knowledge guidance and population evolution.

[0037] 6. This invention proposes multiple types of local search operators based on problem knowledge, which can make targeted improvements to key issues such as delay bottlenecks, replenishment waiting, allocation imbalance and path inefficiency, thereby further improving the quality of scheduling schemes. Attached Figure Description

[0038] Figure 1 A schematic diagram of a distributed production material distribution scheduling scenario and a three-layer joint coding structure provided for embodiments of the present invention.

[0039] Figure 2 A schematic diagram of the intelligent optimization algorithm for knowledge feedback provided in an embodiment of the present invention. Detailed Implementation

[0040] In a distributed manufacturing environment, multiple factories typically generate demand for various materials at different times based on production plans. Multiple warehouses then need to coordinate material supply and distribution tasks, taking into account factors such as inventory levels, replenishment status, vehicle resources, and transportation distances. Material distribution scheduling, as a crucial link connecting warehousing supply and factory production, directly impacts factory material arrival times, production continuity, and order fulfillment capabilities. Compared to single-warehouse or single-factory distribution scenarios, distributed production material distribution scheduling requires simultaneously determining the quantity of various materials supplied by each warehouse to each factory, the loading scheme of warehouse vehicles in different batches, and the factory access path sequence for each vehicle batch. These decisions are significantly coupled: material allocation results affect vehicle batch loading and warehouse replenishment waiting time; batch loading schemes affect vehicle capacity utilization and batch departure time; and the path access sequence further affects the actual arrival time at the factory and material supply delays. Due to the characteristics of this problem—multiple decision couplings, multiple constraints, and large-scale combinatorial optimization—traditional phased optimization methods struggle to fully coordinate the linkage between material allocation, vehicle loading, and path planning, easily leading to local optima, increased replenishment waiting times, or significant delays in material supply to critical factories. Therefore, designing a distributed production material distribution scheduling method that can unify modeling, link reconstruction, and enhance search capabilities through knowledge and feedback mechanisms is of great significance for improving the timeliness of material supply and the operating efficiency of distributed manufacturing systems.

[0041] This invention provides a distributed production material distribution scheduling method applicable to collaborative distribution scenarios involving multiple warehouses, factories, materials, and vehicles. The method aims to minimize factory material supply delays by collaboratively optimizing material allocation, batch loading, and transportation routes. It achieves high-quality scheduling solution search through global cross-referencing, knowledge extraction, knowledge-driven feedback repair, and local optimization. The specific implementation steps are as follows:

[0042] (1) Construct a distributed production material distribution scheduling calculation model, specifically as follows:

[0043] have Each factory needs Various materials, each factory For each material The demand is The materials are from One warehouse supplies, warehouse Medium material The inventory level is , warehouse To the factory Distributed materials Quantity is recorded as If warehouse If the inventory of materials is insufficient, replenishment is required, and the replenishment time is... ;storehouse The materials in the middle are made of Delivery by vehicles, each vehicle In each batch To the factory Transported materials Quantity is recorded as Each material The unit mass is The maximum load capacity of the vehicle is During transportation, the vehicle departs from the node. To the node The time is , warehouse The The car in Batch arrives at factory The time is recorded as ,factory The material supply deadline is The optimization objective is to minimize the material supply delays of all factories, as shown in (1);

[0044] (1)

[0045] in, For the factory The calculation method for the material supply delay is shown in (2):

[0046] (2)

[0047] in, For the factory The arrival time of all materials is calculated as shown in (3):

[0048] (3)

[0049] To meet the material needs of all factories, the factory The The allocation of each material should meet the requirements shown in (4):

[0050] (4)

[0051] To clarify the warehouse The The preparation time for each type of material is calculated, and the replenishment quantity is shown in (5):

[0052] (5)

[0053] To meet vehicle loading capacity limitations, calculate the warehouse vehicles In batch The total load mass is And satisfy the capacity constraints, as shown in (6) and (7):

[0054] (6) (7)

[0055] Figure 1 The matrix blocks provided in this embodiment of the invention represent, from top to bottom, the material allocation layer, the batch loading layer, and the path optimization layer. The material allocation layer matrix uses warehouses as rows and factories as columns, forming multiple allocation sub-matrices according to material type. Matrix elements represent the quantity of corresponding materials supplied by a warehouse to a corresponding factory. The batch loading layer matrix uses warehouses, vehicles, and batches as indices. Each matrix unit corresponds to a loading scheme for a vehicle batch, representing the quantity of various materials transported to each factory in that batch. The path optimization layer matrix uses warehouses, vehicles, and batches as indices. Each matrix unit corresponds to a vehicle delivery path, representing the node sequence where a vehicle departs from its warehouse, visits a factory, and returns to the warehouse. By combining these three matrices, a complete distributed production material distribution scheduling scheme can be represented.

[0056] Based on the above distributed production material distribution scheduling calculation model Figure 2 This is a schematic diagram of the scheduling method provided in an embodiment of the present invention. A knowledge-feedback intelligent optimization algorithm is proposed, comprising five main stages: hybrid heuristic generation of initial solutions, knowledge base construction, hierarchical global cross-referencing, knowledge-driven local search, knowledge base feedback and updating. After each iteration, it determines whether the termination condition is met. If met, the optimal scheduling scheme is output; otherwise, it returns to the hierarchical global cross-referencing stage to continue the next iteration. Specific steps include:

[0057] (2) Construct the three-layer joint encoding / decoding and linkage reconstruction rules for the scheduling scheme, specifically as follows:

[0058] The coding process involves constructing a three-layer joint coding structure for the scheduling scheme. ,in:

[0059] This represents the material allocation layer code, used to indicate the supply quantity relationship of various materials between the warehouse and the factory. (Element) Represents warehouse To the factory Supply materials Quantity;

[0060] This represents the batch loading layer code, used to indicate the loading scheme of warehouse vehicles in each batch. (Element) Represents warehouse The The car in Batch to factory Transport materials Quantity;

[0061] The path optimization layer encoding is used to represent the access node order of each batch of warehouse vehicles. Each path starts and ends at the warehouse to which it belongs, with the sequence of visited factory nodes in between.

[0062] The decoding process is based on three-layer joint coding. The following steps are taken to construct the decoding and delay calculation rules for the scheduling scheme: First, based on the batch loading layer encoding, extract the service factory set, material transportation details, and transportation route information for each batch of each vehicle in each warehouse; then, combine the vehicle availability time, warehouse inventory status, and material replenishment status to sequentially schedule all batches, calculate the actual start time of each batch and the actual arrival time of vehicles at each factory, and update the warehouse inventory and vehicle status during the delivery process; finally, extract the arrival time of all batches for each factory, take the last arrival time as the final material arrival time for that factory, and compare it with the planned arrival time to obtain the factory delay, thereby obtaining the total delay or maximum delay target value of the current scheduling scheme. Based on the three-layer encoding, a linkage reconstruction rule is constructed, specifically including:

[0063] ① Given Refactoring and Material allocation matrix The cargo volume is aggregated by warehouse, forming cargo packages to be loaded by factory, and then loaded into different vehicle batches in order of quality or urgency of demand. When the current batch of vehicles reaches its capacity limit, a new batch or a new vehicle is started to generate a batch loading plan. Then, based on the service factory set corresponding to each batch, a path construction and improvement method is used to generate the corresponding access path, resulting in the path code. .

[0064] ② Given Refactoring and : Path encoding Extract the set of factories that each warehouse, vehicle, and batch can serve, and combine this with the predicted arrival time of each batch. Based on the minimum incremental delay or earliest arrival priority rule, allocate the material requirements of the factories to the corresponding serviceable batches to generate a batch loading plan. Then, the loading plans for each batch are summarized according to the warehouse-factory-material dimension to obtain the material allocation plan. .

[0065] ③ Given Refactoring For each affected vehicle batch, an access path is constructed based on the non-zero destination set corresponding to its loading scheme, and the path is further locally improved to obtain the updated path code. .

[0066] Through the above-mentioned linkage reconstruction mechanism, when the coding of any layer changes, the other two layers can be reconstructed synchronously, thereby ensuring the feasibility of the scheduling scheme in terms of demand satisfaction, capacity constraints, and time consistency.

[0067] (4) A knowledge feedback intelligent optimization algorithm is proposed, which includes five stages: population initialization method, knowledge base construction, global exploration, knowledge-driven local search, knowledge base feedback and update;

[0068] ① The design is based on three-layer joint encoding and linkage reconstruction rules, and uses hybrid heuristic rules to construct the initial population, specifically:

[0069] Randomized heuristic initialization: for each factory and materials demand The warehouse order is randomly shuffled, and the supply quantity is allocated to each warehouse in a random order. For warehouses other than the last one, the allocation quantity is randomly generated within the current remaining demand range. For the last warehouse, all remaining demand is allocated to ensure demand conservation, resulting in a material allocation matrix. Then, the corresponding refactoring unit is invoked to generate a batch loading scheme. and path encoding This forms a complete initial solution.

[0070] Urgency-based heuristic initialization: Based on the time margin between the planned arrival time of the factory and the transportation time from the warehouse to the factory, calculate the urgency of demand for each factory and sort the factories from highest to lowest urgency. Process the demand for various types of materials from each factory in order of urgency, prioritizing warehouses with shorter transportation times and sufficient inventory to undertake the supply task. When the existing inventory of a warehouse is insufficient, allocate the remaining demand to the warehouse with the highest score and trigger replenishment, resulting in a material allocation matrix. Then the corresponding refactoring tool is called to generate... and This forms a complete initial solution.

[0071] Multiple candidate individuals are generated by a hybrid random heuristic and a urgency-first heuristic, forming a group of [size missing]. The initial population, thus taking into account both the quality and diversity of the initial population.

[0072] ② Sort the initial population according to the factory delay fitness and divide it evenly into dominant and inferior individuals. Extract high-quality local structural fragments from the three-layer coding from the dominant individuals to construct material allocation knowledge base, batch loading knowledge base, and path structure knowledge base, respectively.

[0073] Among them, the material allocation knowledge base is based on plant-material pairs. For the index, extract all warehouse allocation quantities corresponding to the factory-material pair in the dominant individual: And store it in the material allocation knowledge base.

[0074] Batch loading knowledge base in warehouse-vehicle-batch combination Using this as an index, extract the transport volume of this batch for all types of materials across all factories within the dominant individual: And store it in the batch loading knowledge base.

[0075] The path structure knowledge base is composed of warehouse-vehicle-batch combinations. For indexing, extract the factory access node sequence corresponding to the batch in the dominant individual as a path segment, and store the total path travel time and the maximum delay value of the source individual together in the knowledge base.

[0076] The three-layer knowledge base can record the high-quality local structures in high-quality scheduling individuals, providing knowledge guidance for subsequent improvement of inferior individuals and local search.

[0077] ③ Select parent individuals from the current population and perform a crossover operation to generate offspring individuals. Combine the parent individuals with all offspring individuals to form a joint candidate set. Evaluate the fitness of each individual in the joint candidate set for factory delays, using the minimum delay for maximum material supply to the factory as the optimization objective. Based on the evaluation results, select the individual with the best fitness. Individuals constitute an elite solution set; then, high-quality local structure fragments are extracted from the three-layer joint encoding in the elite solution set, and the material allocation fragment library, batch loading fragment library, and path structure fragment library are updated respectively.

[0078] Material allocation layer fragment evaluation and crossover: Using the allocation submatrix corresponding to a factory subset or material subset as the evaluation fragment, and comprehensively considering factory delays, approximate transportation time and cost, and replenishment pressure caused by warehouse overstocking, a material allocation layer fragment evaluation function is constructed. During the crossover process, a (factory, material) subset is randomly selected, and the material allocation fragments of the two parent generations at the corresponding index positions are exchanged. Local structures with higher fragment evaluation values ​​are retained first to obtain a new allocation scheme. Then call the refactorer to generate the corresponding and .

[0079] Batch loading layer segment evaluation and crossover: Using the warehouse-vehicle-batch combination as the evaluation segment, and considering the impact of batches on delays, a batch loading layer segment evaluation function is constructed. During the crossover process, several batch combinations are randomly selected as exchange units, and the loading segments of the two parent generations on the selected batches are exchanged. Loading structures with higher segment evaluation values ​​are retained first, resulting in a new batch loading scheme. Then call the refactorer to generate a new one. .

[0080] Path optimization layer segment evaluation and crossover: Using factory visit sequence segments within a path set or a single path as the evaluation object, and considering the impact of batches on delays, a path optimization layer segment evaluation function is constructed. During the crossover process, some batch paths are reorganized using direct inheritance, or factory visit sequences within a single path are recombined using sequential crossover. Path segments with shorter travel times and smaller delays are prioritized for retention, resulting in new path codes. Then call the refactorer to generate compatible ones. and .

[0081] Through the above-mentioned hierarchical cross-operations, a collaborative global search is achieved across the three decision-making levels of material allocation, batch loading, and path optimization.

[0082] ④ Based on a knowledge-driven local search strategy, the current population's top search points are analyzed. The elite individuals perform knowledge-driven local search, and for each individual, the following local search operators are executed sequentially for fine-tuning:

[0083] Material reallocation for factories facing delays: When an individual has replenishment needs, identify batches delayed due to replenishment waiting, attempt to transfer some of the required replenishment materials to other warehouses with sufficient inventory, and refactor the relevant layer codes accordingly. If the replenishment waiting time decreases and the maximum delay is reduced, accept the new solution; otherwise, retain the original solution.

[0084] Vehicle reassignment for replenishment waiting: Identify the routes with the longest total travel time and find the factories with the largest delays on those routes. Find the distance to the factory Other nearby warehouses Try to turn the factory Some or all of the demand will be from the original warehouse. Moved to a warehouse This process involves reconstructing the batch loading layer and the path optimization layer. If the maximum delay is reduced, the new solution is accepted; otherwise, the original solution is retained.

[0085] Path re-optimization for delays: Identify bottleneck paths causing maximum factory delays, and for the current bottleneck path, optimize through node swapping or... The approach attempts to shorten the total travel time of the route. If the maximum delay is reduced, the new solution is accepted; otherwise, the original solution is retained.

[0086] The specific execution strategy is as follows: Rank the top-performing individuals in the current population based on their evaluation function. Elite individuals perform knowledge-driven local searches. To balance the ability to refine local optimization with the ability to escape local optima, this invention designs a progressive local search strategy that combines small-step search and large-step search. Small-step search is used to make minor perturbations in the neighborhood of the current solution, while large-step search is used to make larger-scale structural adjustments when small-step search fails to improve the plant's delay fitness. After each search operation, the constructor is invoked to update the material allocation layer, batch loading layer, and path optimization layer, and the fitness is recalculated through the scheduling calculation model.

[0087] For each elite individual, a small-step search is first performed on the material reallocation operator for the delayed bottleneck plant to obtain a new individual and calculate its delay fitness. If the fitness of the new individual is improved, the new individual is accepted, and the small-step search on the vehicle reallocation operator for replenishment waiting continues; if the fitness of the new individual is not improved, a large-step search on the material reallocation operator for the delayed bottleneck plant is performed, and the fitness is recalculated. If the fitness is improved after the large-step search, the new individual is accepted, and the small-step search on the vehicle reallocation operator for replenishment waiting continues; otherwise, the original individual is retained.

[0088] When executing the vehicle reallocation operator for replenishment waiting, the operator's small-step search is first performed, and the delay fitness is calculated. If the fitness improves, a new individual is accepted, and the small-step search for the delay-oriented path re-optimization operator continues. If the fitness does not improve, the operator's large-step search is performed, and the fitness is recalculated. If the fitness improves after the large-step search, a new individual is accepted, and the small-step search for the delay-oriented path re-optimization operator continues; otherwise, the current individual is retained.

[0089] When executing the delay-oriented path re-optimization operator, the bottleneck path causing the maximum material supply delay is first identified, and then addressed through node swapping or... The method involves small-step path adjustments. If the fitness improves after the adjustment, the new individual is accepted; if the fitness does not improve, a large-step path re-optimization can be performed, through multi-node re-insertion and multi-segment... Alternatively, a multi-path joint adjustment method can be used to re-optimize the vehicle access order. If the fitness improves after the large-step path re-optimization, the new individual is accepted; otherwise, the current individual is retained.

[0090] ⑤ High-quality coded knowledge is fed back to the current population through the knowledge base. Specifically, several poorly performing individuals with low evaluation function rankings are selected from the current population, and the current material supply delay for each factory is calculated. The factory with the largest delay value is recorded as the [factory name missing]. Factory-material pair As the target index in the material allocation knowledge base, it is the warehouse-vehicle-batch combination with the latest arrival time. As the target index in the batch loading knowledge base and path structure knowledge base, each layer of coding selects candidate fragments from the corresponding knowledge base that are valid with the current individual index, consistent with the requirements, and have better recorded delay values. The best candidate fragment is embedded into the current individual, and the constructor is called to reconstruct the coding of the remaining layers. After the individual is improved, it is re-evaluated. If the maximum delay is reduced, the improvement result is accepted; otherwise, the original individual is retained. After all the poor individuals complete the knowledge feedback, a new population is formed. The top-ranked individuals in the evaluation function are selected from the new population as the elite solution set. High-quality local structure fragments are extracted from their material allocation layer, batch loading layer, and path optimization layer, respectively, and the material allocation knowledge base, batch loading knowledge base, and path structure knowledge base are updated. The newly extracted fragments are deduplicated. If the knowledge base capacity exceeds the limit, the fragment with the lowest score or that has not been used for several consecutive generations is eliminated.

[0091] (4) Repeat step (3) -⑤ The global exploration, knowledge-driven local search, knowledge base feedback and update are performed until the preset termination conditions are met, and the optimal scheduling scheme is output. The optimal scheduling scheme includes the allocation results of each warehouse to each factory for various types of materials, the loading scheme of each batch of each vehicle in the warehouse, the access path sequence of each batch of vehicles, and the actual arrival time and material supply delay of each factory.

[0092] The distributed production material distribution scheduling method proposed in this invention constructs an initial population with good quality and diversity through hybrid heuristic rules. It utilizes a three-layer joint encoding and linkage reconstruction mechanism to uniformly represent the coupling relationship between material allocation, batch loading, and path optimization, and achieves collaborative global search across different decision-making layers through hierarchical cross-operations. Simultaneously, this invention extracts, stores, and reuses high-quality local structures generated during the evolution process by constructing material allocation knowledge bases, vehicle loading knowledge bases, and path knowledge bases. Furthermore, it combines a knowledge-driven local search strategy that integrates small-step and large-step searches to address key issues such as delay bottlenecks, replenishment waiting times, and path inefficiencies. This enables the generation of high-quality distribution scheduling schemes with minimal material supply delays, satisfying material demand constraints, vehicle capacity constraints, and time consistency constraints in a relatively short time, thereby effectively improving the timeliness of material supply and the efficiency of distribution scheduling in a distributed manufacturing environment.

Claims

1. A distributed production material distribution and scheduling method, characterized in that, The steps include the following: (1) Construct a distributed production material distribution scheduling calculation model. The model aims to minimize the material supply delay of all factories and includes material demand, warehouse inventory, material allocation, replenishment time, vehicle loading capacity, vehicle load constraints, and vehicle transportation time constraints. (2) Construct a three-layer joint coding structure including a material allocation layer, a batch loading layer and a path optimization layer, and establish construction rules between each layer. When the coding of any layer changes, the coding of the remaining layers is updated synchronously through the construction mechanism to ensure the feasibility of the scheduling scheme in terms of material demand, capacity constraints and time consistency. (3) The solution is obtained by using a knowledge feedback intelligent optimization algorithm. The algorithm includes five stages: population initialization, knowledge base construction, global exploration, knowledge-driven local search, knowledge base feedback and update. The five stages are executed in sequence and form a closed-loop iterative optimization process. (4) Repeat the global exploration, knowledge-driven local search, knowledge base feedback and update in step (3) until the preset termination condition is met, and output the optimal scheduling scheme. The optimal scheduling scheme fully covers all decision variables and evaluation indicators in the model.

2. The distributed production material distribution and scheduling method according to claim 1, characterized in that, The specific construction method of the distributed production material distribution scheduling calculation model in step (1) is as follows: have Each factory needs Various materials, each factory For each material The demand is The materials are from One warehouse supplies, warehouse Medium material The inventory level is , warehouse To the factory Distributed materials Quantity is recorded as If warehouse If the inventory of materials is insufficient, replenishment is required, and the replenishment time is... ;storehouse The materials in the middle are made of Delivery by vehicles, each vehicle In each batch To the factory Transported materials Quantity is recorded as Each material The unit mass is The maximum load capacity of the vehicle is During transportation, the vehicle departs from the node. To the node The time is , warehouse The The car in Batch arrives at factory The time is recorded as ,factory The material supply deadline is The optimization objective is to minimize material supply delays for all factories. The optimization objective is to minimize the maximum delay in material supply across all factories; where, factories The material supply delay is defined as the positive portion of the difference between the final arrival time and the planned deadline for all materials. The final arrival time is defined as the maximum value among the arrival times of materials delivered to the plant by all warehouses, vehicles, and batches. The model satisfies the following constraints: the demand for each material at each plant is equal to the sum of the allocations of that material to the plant by all warehouses; when a warehouse allocation exceeds its existing inventory, the stockout needs to be replenished, and the replenishment amount is the difference between the allocation and the inventory; the total mass of all materials loaded in each vehicle batch must not exceed the maximum load capacity of the vehicle.

3. The distributed production material distribution and scheduling method according to claim 1, characterized in that, The specific implementation of the three-layer joint coding structure and construction rules in step (2) is as follows: Construct a three-layer joint coding structure comprising a material allocation layer X, a batch loading layer B, and a path optimization layer O. The material allocation layer X is represented in matrix form, with warehouses as rows and factories as columns, and multiple allocation sub-matrices are formed according to material type. Matrix elements... This represents the quantity of material k supplied by warehouse i to factory j, used to characterize the macroscopic supply quantity relationship of various materials between warehouses and factories; the batch loading layer B is represented in multidimensional matrix form, with warehouses, vehicles, and batches as indices. Each matrix unit corresponds to a loading scheme for a vehicle batch, and the matrix elements... The quantity of material k transported by the l-th vehicle of warehouse i to factory j in the r-th batch is represented by the l-th vehicle of warehouse i. This is used to characterize the specific loading details of warehouse vehicles in each batch. The path optimization layer O is represented in the form of a sequence set, with warehouse, vehicle and batch as indices. Each matrix unit corresponds to a vehicle delivery path, with the warehouse as the starting point and the end point, and the middle part as the sequence of visited factory nodes. This is used to characterize the factory access order of each batch of warehouse vehicles. Based on the above three-layer coding, the following linkage reconstruction rules are established: Rule ①—Given the material allocation layer X, reconstruct the batch loading layer B and the path optimization layer O. Specifically, the quantities in the material allocation matrix X are summarized by warehouse, forming a batch of goods to be loaded by factory, and then loaded into different vehicle batches in order of material quality or urgency. When the current batch of vehicles reaches its capacity limit, a new batch or a new vehicle is started, generating a batch loading scheme B. Then, based on the service factory set corresponding to each batch, a path construction and improvement method is used to generate the corresponding access path, resulting in path code O; Rule ②—Given the path optimization layer O, reconstruct the material allocation layer X and the batch loading layer B. Layer B specifically involves extracting the set of factories that each warehouse, vehicle, and batch can serve from the path code O, and combining the predicted arrival time of each batch. The material requirements of the factories are then allocated to the corresponding serviceable batches according to the rule of minimum incremental delay priority or earliest arrival priority, generating batch loading scheme B. Then, the loading schemes of each batch are summarized according to the warehouse-factory-material dimension to obtain the material allocation scheme X. Rule ③ - Given the batch loading layer B, the path optimization layer O is reconstructed. Specifically, for each affected vehicle batch, an access path is constructed based on the non-zero destination set corresponding to the loading scheme, and the path is further improved locally to obtain the updated path code O. Through the above-mentioned linkage and reconstruction mechanism, the collaborative representation of the three-level decision-making of material allocation, batch loading and path optimization is realized. When the coding of any layer changes, the corresponding construction rule is called to automatically and synchronously reconstruct the coding of the other two layers, thereby ensuring the feasibility of the scheduling scheme in terms of demand satisfaction, capacity limitation and time consistency.

4. The distributed production material distribution and scheduling method according to claim 3, characterized in that, The specific implementation method of the population initialization phase in step (3) is as follows: The initialization phase employs a hybrid initialization method based on random heuristics and urgency priority heuristic to construct an initial population of size PS. This phase provides a high-quality and diverse starting point for subsequent evolutionary search, and the initial population also serves as the data source for subsequent knowledge base construction. The specific operation of random heuristic initialization is as follows: for each factory j and material k, the demand Q jk The warehouse order is randomly shuffled, and the supply quantity is allocated to each warehouse in a random order. For warehouses other than the last one, the allocation quantity is randomly generated within the current remaining demand range. For the last warehouse, all remaining demand is allocated to ensure demand conservation, resulting in the material allocation matrix X. The specific operation of the urgency-first heuristic initialization is as follows: calculate the urgency of each factory's demand based on the time margin between the planned arrival time of the factory and the transportation time from the warehouse to the factory, sort the factories from high to low urgency, process the demand of each factory for various types of materials in the sorting order, prioritize the warehouse with shorter transportation time and sufficient inventory to undertake the supply task, and when the warehouse's existing inventory is insufficient, allocate the remaining demand to the warehouse with the highest score and trigger replenishment to obtain the material allocation matrix X; Material allocation matrices generated by random heuristic initialization and urgency-first heuristic initialization B and O are generated according to the linkage reconstruction rule described in claim 3 to form a complete initial solution; multiple candidate individuals are generated by a combination of random heuristic and urgency priority heuristic to form an initial population of size PS, thereby taking into account both the quality and diversity of the initial population.

5. The distributed production material distribution and scheduling method according to claim 1, characterized in that, The specific implementation method of the knowledge base construction stage in step (3) is as follows: In the knowledge base construction phase, the initial population is sorted by factory delay fitness and divided into dominant and inferior individuals. High-quality local structural fragments are extracted from the three-layer coding of the dominant individuals to construct material allocation knowledge base, batch loading knowledge base and path structure knowledge base respectively. The core function of this phase is to extract and store the excellent structural knowledge generated during the evolution process, and provide guidance information for subsequent knowledge feedback and local search. The material allocation knowledge base is constructed as follows: using the factory-material pair (j,k) as the index, all warehouse allocation quantities corresponding to that factory-material pair are extracted from each dominant individual. The allocation vector is stored as a knowledge fragment in the material allocation knowledge base to record the supply allocation ratio of each warehouse for specific material needs of a specific factory. The batch loading knowledge base is constructed using a combination of warehouse, vehicle, and batch. As an index, extract the shipment volume of this batch for all types of materials across all factories from each dominant individual. The transportation volume matrix is ​​stored as a knowledge fragment in the batch loading knowledge base to record the loading structure of a specific vehicle in a specific batch in a specific warehouse. The path structure knowledge base is constructed using a combination of warehouse, vehicle, and batch. For indexing, the sequence of factory access nodes corresponding to the batch is extracted from each dominant individual as a path segment, and the total travel time of the path and the maximum delay value of the source individual are stored together in the path structure knowledge base to record the high-quality vehicle access sequence and its performance indicators.

6. The distributed production material distribution and scheduling method according to claim 1, characterized in that, The specific implementation method of the global exploration phase described in step (3) is as follows: The global exploration phase performs a hierarchical crossover operation oriented towards three-level coding on individuals in the current population to generate offspring individuals. After that, a new generation of population is formed through factory delay fitness evaluation and selection. This phase is responsible for conducting a broad search in the solution space to explore the combination possibilities between different scheduling schemes and to propagate the best coding fragments in the population. The specific operation is as follows: For each individual in the current population, it is paired with another randomly selected individual in the population with a certain probability, and a hierarchical crossover operation oriented towards the three-layer encoding is performed to generate offspring individuals; among them, the material allocation layer crossover randomly selects a factory-material subset, and swaps the allocation submatrices of the two parent individuals at the corresponding index positions to obtain a new allocation scheme. The batch loading layer randomly selects several warehouse-vehicle-batch combinations and swaps the loading segments of two parent individuals in the selected batches to obtain a new batch loading scheme. The path optimization layer uses direct inheritance to exchange the entire path code for some batch paths, or uses sequential interleaving within a single path to reorganize the factory access sequence according to the order of appearance in another parent, resulting in a new path code. After the above-mentioned layers are crossed, the coding of the other two layers is updated synchronously in accordance with the linkage reconstruction rules described in claim 3 to ensure the consistency of the three-layer coding. The constructor is invoked to generate the corresponding X' and B' according to the linkage reconstruction rules described in claim 3; The parent individuals are merged with all offspring individuals, and the evaluation function value of each individual is calculated. The optimization objective is to minimize the maximum material supply delay of the factory. The top PS individuals with the best factory delay fitness are retained to form a new population.

7. The distributed production material distribution and scheduling method according to claim 3, characterized in that, The specific implementation method of the knowledge-driven local search stage in step (3) is as follows: The knowledge-driven local search phase involves executing three local search operators sequentially to finely optimize the top 10% of elite individuals in the current population based on their evaluation function. The core function of this phase is to conduct a deep and fine search in the vicinity of the elite individuals and make targeted improvements to the key bottlenecks that cause delays. If the maximum delay is reduced after each operator is executed, the new solution is accepted; otherwise, the original solution is retained. Operator 1 is for vehicle reassignment for replenishment waiting, used to identify delivery routes with long total travel time and include delayed factories, and adjust some or all of the material requirements of the delayed factories to nearby alternative warehouses or vehicle batches. Operator 2 is for vehicle reassignment for replenishment waiting, used to identify delivery routes with long total travel time and include delayed factories, and adjust some or all of the material requirements of the delayed factories to nearby alternative warehouses or vehicle batches. Operator 3 is a path re-optimization oriented towards delays, used to identify the bottleneck path that causes the maximum delay and perform node swapping or path subsequence reversal operations on that path. The three local search operators mentioned above adopt a progressive strategy that combines small-step search and large-step search: first, small-step search is performed, i.e., single-point adjustment. If the factory delay fitness does not improve after small-step search, large-step search is performed, i.e., multi-point joint adjustment or multi-round iterative adjustment. After each search operation is completed, the constructor is called to update the three-layer code and recalculate the fitness.

8. The distributed production material distribution and scheduling method according to claim 1, characterized in that, The specific implementation method of the knowledge base feedback and update stage in step (3) is as follows: The knowledge base feedback and update stage includes two sub-stages: knowledge feedback and knowledge base update. This stage realizes the two-way coupling of knowledge guidance and population evolution: the knowledge base provides guidance for the improvement of inferior individuals, and the new elite solutions generated by evolution, in turn, enrich and update the knowledge base. The specific operation of the knowledge feedback sub-step is as follows: Select several poor individuals with low evaluation function rankings in the current population, and identify the bottleneck factory and bottleneck batch in each poor individual; use the combination of the bottleneck factory and material as the retrieval index of the material allocation knowledge base, filter out matching allocation fragments with better delay values ​​from the source individuals, and embed them into the material allocation layer; use the warehouse-vehicle-batch combination corresponding to the bottleneck batch as the retrieval index of the batch loading knowledge base and path structure knowledge base, filter out matching loading fragments and path fragments with better delay values ​​from the source individuals, and embed them into the corresponding encoding layer; reconstruct and re-evaluate the individuals after embedding the knowledge fragments according to the linkage reconstruction rules described in claim 3. If the maximum delay is reduced, the improvement result is accepted; otherwise, the original individual is retained. After all poor individuals complete the knowledge feedback, a new population is formed. The specific operation method of the knowledge base update sub-step is as follows: Select the top-ranked individuals in the evaluation function from the new population as the elite solution set, extract high-quality local structural fragments from the material allocation layer, batch loading layer and path optimization layer of the elite individuals respectively, and check the newly extracted fragments against the existing fragments in the knowledge base for deduplication. If there are completely identical or highly similar fragments, they are skipped. If the knowledge base capacity exceeds the preset limit, the fragments with the lowest score or those that have not been adopted for several consecutive generations are eliminated. The new fragments that pass the deduplication check are stored in the corresponding knowledge base.

9. The distributed production material distribution and scheduling method according to claim 1, characterized in that, The specific implementation method of repeating the execution and outputting the results in step (4) is as follows: Repeat the global exploration phase, knowledge-driven local search phase, and knowledge base feedback and update phase in step (3). The three phases form a closed loop iteration: the new generation of population output by the global exploration phase becomes the processing object of the knowledge-driven local search phase; the elite individuals output by the knowledge-driven local search phase enter the knowledge base feedback and update phase after local optimization; the new population output by the knowledge base feedback and update phase is used for the global exploration phase of the next iteration; and the updated knowledge base is fed back to the global exploration and local search of the next iteration, forming a closed loop of continuous evolution. When a preset termination condition is met, the iteration stops and the optimal scheduling scheme is output. The optimal scheduling scheme specifically includes the following: The allocation results of each warehouse to each factory for various types of materials, i.e., the specific quantity of each type of material supplied by each warehouse to each factory; the loading plan of each vehicle and each batch in each warehouse, i.e. the specific quantity of each type of material transported by each vehicle in each batch to each factory; the access path sequence of each batch of vehicles, i.e. the complete node sequence of each vehicle starting from the warehouse, visiting the factory in sequence, and returning to the warehouse in each batch; and the actual arrival time and material supply delay of each factory, i.e. the final arrival time of each type of material in each factory and the delay time relative to the planned deadline.

Images