Production optimization method and system combined with logistics management

Abstract

The invention discloses a production optimization method and system combined with logistics management, and relates to the technical field of production optimization. The invention discloses a production optimization system combined with logistics management. The production optimization system comprises a personalized order analysis module, a logistics state acquisition module, an internal logistics modeling module, an external logistics optimization module, a supply strategy management module, a joint scheduling optimization module, a scheduling execution module and a congestion monitoring and simulation feedback module. According to the invention, an external logistics resource supply strategy is structured and expressed as a transport time window matrix and a dynamic transport capacity allocation table, and the transport time window matrix and the dynamic transport capacity allocation table are used as boundary conditions to be introduced into a joint scheduling optimization model, so that a scheduling system can dynamically adjust a production and logistics collaborative plan according to an external resource state; and the adaptive capacity of the scheduling strategy to external supply changes is effectively enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of production optimization, and in particular to a production optimization method and system combined with logistics management. Background Art

[0002] Against the backdrop of increasing complexity in manufacturing systems and increasingly personalized order requirements, production optimization is gradually shifting from static scheduling to collaborative optimization that dynamically adapts resources and demand. Existing production scheduling methods are mostly based on order data and process flows, combining the capacity constraints of manufacturing resources to allocate and prioritize tasks, and provide relatively mature support for internal factory scheduling logic. In some high-end manufacturing scenarios, production process flow charts based on graph-based modeling are already available, enabling dynamic scheduling optimization at the process and equipment levels.

[0003] At the same time, factory logistics systems, as a crucial link connecting various production nodes, are gradually evolving towards automation and real-time capabilities. For internal factory logistics, some studies have proposed using Petri nets and queue networks to model node channels and transportation resources to assess equipment occupancy and path accessibility. For external logistics scheduling, existing transportation management systems or supply chain visualization platforms enable real-time coordination between vehicle scheduling and route planning. However, these systems often operate as standalone systems, relying on pre-set rules or manual intervention to interface with workshop scheduling systems.

[0004] In the manufacturing process of customized products, the heterogeneity of orders and the pressure of real-time delivery pose challenges to traditional scheduling systems. In high-variety, small-batch, and lead-time-sensitive manufacturing scenarios, existing solutions still need to improve the integrated collaboration between logistics management and production scheduling. In particular, when external transportation resource supply strategies influence production scheduling decisions, there is a lack of a mechanism to dynamically incorporate external supply strategies into the production system's optimization process, enabling closed-loop collaboration across the modeling, scheduling, and feedback layers. This has become one of the core challenges in integrated manufacturing logistics scheduling. Summary of the Invention

[0005] A production optimization method combined with logistics management, comprising: Receive personalized customer orders and extract order feature data; For personalized customer orders for customized products, called customized orders, before the actual customized order is fully confirmed, the customer's configuration intention is obtained and cluster analysis is performed to generate typical configuration clusters. These are used to construct corresponding shadow tasks. These shadow tasks are introduced as pre-scheduling inputs into the joint scheduling optimization model. Real-time collection of factory internal and external logistics resource status information, referred to as internal logistics information and external logistics information, and modeling of internal logistics information to generate the factory internal logistics capacity map; Factories, logistics providers, and their external transport fleets are modeled as agents. A dynamic game model based on a multi-agent reinforcement learning framework is used to generate the optimal response strategy for each agent using a hierarchical game strategy, which then outputs an external logistics resource supply strategy. A joint scheduling optimization model based on graph structure modeling is established based on order feature data, factory internal logistics capacity map, external logistics resource supply strategy and shadow tasks to coordinate the scheduling optimization of production tasks and internal logistics routes. Generate work instructions and transportation scheduling tasks for each production unit and logistics node based on the collaborative scheduling optimization results; During the execution of the transport scheduling task, internal logistics information is collected in real time to obtain the congestion index of the transport scheduling task. When the congestion index is abnormal, the transport scheduling task is dynamically adjusted; When the congestion indicator of the transport scheduling task is still abnormal after dynamic adjustment, a path conflict simulation prediction of the transport scheduling task is performed, and the simulation prediction results are fed back to the joint scheduling optimization model to trigger adaptive rescheduling and form a decision-making closed loop.

[0006] As a preferred technical solution of the present invention, generating a factory internal logistics capability diagram includes: The Petri net modeling tool is used to construct an internal logistics capability diagram of the factory, which includes the topological connection relationship and resource transfer rules between logistics nodes. The internal logistics information includes the occupancy and service capacity of logistics nodes. Logistics nodes include fixed facilities, mobile equipment and external interfaces within the factory. The resource transfer rules are constructed based on the service capacity of each logistics node, including task processing rhythm, path smoothness and accessibility conditions.

[0007] As a preferred technical solution of the present invention, the hierarchical game strategy is used to generate the optimal response strategy of each agent, including: The factory, logistics provider, and their external transport fleet are each modeled as an intelligent agent for reinforcement learning. Each agent constructs a set of observation states based on order feature data and external logistics information, and defines a preset set of optional actions corresponding to its role. A dynamic game model is constructed, with the factory acting as the leader agent to generate the leader's strategy. The logistics provider and its external transport fleet serve as follower agents. After acquiring the leader's strategy, they use reinforcement learning to optimize their own response strategies based on a predefined reward function. This reward function is a weighted combination of transportation cost, on-time delivery rate, and path load balancing. The optimization results are output as external logistics resource supply strategies, including transportation route allocation, vehicle scheduling, supply time windows, and task acceptance relationships.

[0008] As a preferred technical solution of the present invention, the external logistics resource supply strategy is structured and expressed as a transportation time window matrix and a dynamic transportation capacity allocation table, which serve as input boundary conditions of the joint scheduling optimization model and are updated in real time through the intelligent agent communication interface.

[0009] As a preferred technical solution of the present invention, the joint scheduling optimization model includes an order-driven production process flow chart constructed based on order feature data, and an internal logistics network chart constructed based on the factory internal logistics capacity chart; the process flow chart uses order operation units and production equipment as nodes and process dependency relationships as edges; the internal logistics network chart uses logistics nodes as nodes and topological connection relationships as edges; the production equipment nodes in the production process flow chart and the nearest logistics nodes in the internal logistics network chart are automatically mapped to form a joint graph model; the objective function is set to minimize production task delays and maximize equipment and transportation resource utilization, and the resource transfer rules of the factory internal logistics capacity chart and the external logistics resource supply strategy are combined as boundary conditions to perform collaborative scheduling optimization; When the customized order has not been fully confirmed but there is a typical configuration cluster generated by configuration intention clustering, shadow tasks are introduced into the joint graph model as pre-scheduling tasks for resource pre-occupancy and path pre-scheduling.

[0010] As a preferred technical solution of the present invention, the generation and scheduling of the shadow task includes: When a custom order has not yet been fully confirmed, the customer's configuration intent is constructed as a vector and clustered. This generates representative typical configuration clusters. For each configuration cluster, a typical process flow and resource requirement template is generated, and shadow tasks are constructed accordingly. Shadow tasks are introduced into the joint scheduling optimization model as soft constraints to implement resource pre-occupation and path pre-scheduling before the custom order is confirmed. When the actual order is placed, the corresponding shadow tasks are matched and activated. Unmatched shadow tasks are released after a set time window to improve adaptability to the diversity and uncertainty of custom orders. The configuration intention includes the customer's configuration behavior data and historical order data on the customization platform.

[0011] As a preferred technical solution of the present invention, the method for determining the abnormality of the congestion index includes task congestion and local logistics capacity bottleneck, and the congestion index is calculated in batches at fixed time intervals; the content of the dynamic adjustment includes job order priority, path selection and batch division, and the dynamic adjustment must meet the constraints of the process dependency of the production process flow chart.

[0012] As an optimal technical solution of the present invention, the method for determining task congestion is to obtain the occupancy status of each logistics node in the internal logistics capacity diagram of the factory, calculate the variance of the resource occupancy rate in the occupancy status, and when the resource occupancy rate variance exceeds the set threshold and the duration exceeds the set time, it is determined to be task congestion; the method for determining local logistics capacity bottlenecks is to identify logistics nodes where resource request blocking occurs continuously in the internal logistics capacity diagram of the factory, and detect whether the task waiting time of the node exceeds the set threshold in multiple consecutive scheduling cycles. If so, it is determined to be a local logistics capacity bottleneck.

[0013] A production optimization system combined with logistics management, comprising: Personalized order parsing module: used to receive personalized customer orders and extract order feature data; Customization Intention Clustering Module: This module is used to obtain and cluster customers' configuration intentions before a customized order is fully confirmed, generate typical configuration clusters, construct corresponding shadow tasks, and manage the introduction, activation, and release of shadow tasks. Logistics status collection module: used to collect status information of internal and external logistics resources of the factory respectively, and generate internal logistics information and external logistics information; Internal logistics modeling module: used to build the factory internal logistics capacity diagram based on internal logistics information, using Petri net modeling tools to describe the topological structure of logistics nodes and resource transfer rules; External logistics optimization module: used to model factories, logistics suppliers and their subordinate transport fleets as intelligent agents, build dynamic game models, and generate external logistics resource supply strategies; Supply strategy management module: used to express the external logistics resource supply strategy in a structured manner, implement its regular update through the communication interface, and generate boundary conditions for the joint scheduling model to call; The joint scheduling optimization module establishes a joint scheduling optimization model based on order feature data, factory internal logistics capacity diagram, external logistics resource supply strategy and shadow tasks, and performs collaborative scheduling optimization of production tasks and internal logistics routes; The scheduling execution module generates operation instructions and transportation scheduling tasks for each production unit and logistics node based on the collaborative scheduling optimization results; The congestion monitoring and simulation feedback module is used to monitor whether the congestion indicators of the transportation scheduling tasks are abnormal, and perform path conflict simulation prediction when the anomaly persists, and feed the simulation results back to the joint scheduling optimization module to trigger rescheduling operations.

[0014] The present invention has the following advantages: The present invention models factories, logistics suppliers and their transport fleets as reinforcement learning agents, constructs a dynamic game model, and realizes the generation of external logistics resource supply strategies including transportation route allocation, vehicle scheduling, supply time windows and task acceptance relationships, thus solving the problems of static settings, delayed response and lack of intelligent optimization of external logistics scheduling in traditional systems.

[0015] Through configuration intention clustering and shadow task pre-scheduling mechanism, the present invention targets typical manufacturing scenarios in which customized product order characteristics are highly uncertain, orders are concentrated, and impacts are sudden. It identifies configuration trends and constructs representative shadow tasks before the order is confirmed, participates in resource scheduling and path planning in advance, effectively alleviates the problem of post-scheduling response, realizes the pre-scheduling strategy for customized orders, and improves the scheduling response speed and resource adaptability in a diversified and high-frequency customization environment.

[0016] The present invention expresses the external logistics resource supply strategy in a structured manner as a transportation time window matrix and a dynamic capacity allocation table, and introduces a joint scheduling optimization model as boundary conditions, so that the scheduling system can dynamically adjust the production and logistics collaboration plan according to the external resource status, effectively enhancing the scheduling strategy's adaptability to changes in external supply.

[0017] The present invention builds a joint scheduling optimization model based on order feature data, the factory's internal logistics capacity diagram and external supply strategy, integrates the production process flow chart and the internal logistics network diagram, introduces a dynamic external supply strategy and an internal logistics status feedback mechanism, and realizes an intelligent closed loop of the entire process from order input to scheduling output to feedback adjustment, so as to meet the real-time response needs of highly personalized production.

[0018] The present invention designs a congestion identification mechanism for the internal logistics capacity diagram of the base factory, utilizes resource occupancy rate variance and task waiting time indicators to dynamically identify task congestion and local logistics capacity bottleneck nodes, and performs path conflict simulation prediction when the anomaly persists. The feedback is fed back to the scheduling model to form a scheduling closed loop, thereby improving the real-time response capability to deal with local imbalances and path conflicts. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only schematic diagrams of the present invention. Those skilled in the art can also derive other drawings based on the provided drawings without inventive effort. Figure 1 This is a schematic diagram of the structure of a production optimization system combined with logistics management adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Example 1, a production optimization method combined with logistics management, comprising the following steps: Step S1: Receive personalized customer orders and extract order feature data, including but not limited to the order's process requirements, delivery deadlines, material requirements, and resource constraints; Specifically, they include: Process requirements include but are not limited to customer-specified product processing flow, key process sequences, customized parameters (such as specific heat treatment curves, specific tool paths), and other information. These are parsed from the MES system or order input interface and serve as the basis for node and edge composition when constructing the production process flow chart. Delivery deadlines include but are not limited to expected delivery times and tolerable early / late tolerance windows. These information is sourced from the order management system or the customer-side configuration platform. This information is directly used in setting production task priorities and generating path scheduling timing constraints in the joint scheduling model. Material requirements include, but are not limited to, the raw material codes, specifications, expected quantities, and supply cycle constraints required for the product. These requirements are extracted from the material master data in the ERP system or explicitly set by the customer during the order acceptance process. This field is used in subsequent scheduling to ensure that the material arrival sequence matches the logistics node task time. Resource constraints include, but are not limited to, customer-specified production lines, equipment preferences, exclusive processing resources (such as a certain CNC model), transportation mode preferences (such as limited cold chain transportation or nighttime delivery), and other constraints. This type of data is set as an optional parameter in the order configuration platform and can also be derived from learning rules of historical orders to limit the candidate solution space of the joint scheduling model.

[0022] For example, consider order feature data for a custom mechanical parts order with a process requirement specifying "grinding after heat treatment," a delivery deadline of "June 20th, with one day earlier or two days later permitted," a material requirement of "20 45# steel Φ100mm bars," and a resource constraint of "processing only on Line 4, no nighttime delivery." This order feature data forms an order feature vector, which serves as the core input for downstream multi-agent modeling and scheduling optimization.

[0023] Step S2: For personalized customer orders for customized products, known as customized orders, before the actual customized order is fully confirmed, the customer's configuration intention is obtained and cluster analysis is performed to generate typical configuration clusters. These are used to construct corresponding shadow tasks. These shadow tasks are introduced as pre-scheduling inputs into the joint scheduling optimization model. The generation and scheduling of the shadow task includes: When a customized order has not yet been fully confirmed, the customer's configuration intention is constructed as a configuration vector. Cluster analysis is performed on multiple configuration vectors to obtain several representative typical configuration clusters. A typical process flow and resource requirement template is generated for each configuration cluster, and shadow tasks are constructed based on this. Shadow tasks are introduced into the joint scheduling optimization model in the form of soft constraints to achieve resource pre-occupation and path pre-scheduling before the customized order is confirmed. When the customized order is officially confirmed and matched with a certain configuration cluster, the system automatically activates the corresponding shadow task and incorporates it into the formal scheduling process. If the shadow task is not matched to an actual order within the set time window, the system releases or clears the shadow task to free up system resources.

[0024] The configuration intention includes the customer's configuration behavior data and historical order data on the customization platform. The configuration behavior data includes the configuration parameter information and click behavior records selected or preferred by the customer on the customization platform. This mechanism significantly enhances the proactive response capability and scheduling flexibility in the face of the diversity and uncertainty of customized products through the structured expression of potential customization needs and the introduction of a joint scheduling optimization model.

[0025] Shadow tasks are virtual scheduling tasks generated by configuration intent clustering before a custom order is officially confirmed. These tasks are used to proactively participate in the optimization calculations of the joint scheduling optimization model. Shadow tasks are not directly tied to actual materials or actions, but they share a similar process path and resource requirement description to future orders, allowing them to be identified and processed by the joint scheduling optimization model.

[0026] The specific form of the shadow task includes the following key fields: task ID, used to uniquely identify the shadow task; source configuration cluster ID, which identifies the configuration cluster number corresponding to the shadow task; estimated process flow, which is a typical process sequence generated by the configuration cluster, such as [roughing → heat treatment → finishing → inspection]; estimated resource requirements, including the required production equipment type, logistics node type, processing time, beat rate, etc.; virtual time window, which is the predicted start and completion time interval used to plan the occupancy window; execution priority weight, which is set lower than that of the formal order task and can be used as a soft constraint in the scheduling objective function; status tag, which includes three states: unmatched, activated, and released, for lifecycle management; Shadow tasks are added to the joint graph model as variable scheduling entities and participate in scheduling solutions together with formal order tasks, but they will not be executed by the scheduling system and are only used to influence path planning and resource allocation strategies.

[0027] For example, a configuration cluster represents the frequently occurring combination of "body + premium interior + large wheels." The system generates a process path for it: "painting → interior → final assembly → quality inspection." The resource requirements are equipment E2 / E5, AGV-2 lane, and buffer C1, with a time window of tomorrow's 2:00 PM to 5:00 PM. This shadow task is scheduled with a low weight in the joint scheduling model. Once the system detects that a customer has placed an order for the corresponding configuration, it automatically activates the task and becomes a formally scheduled item.

[0028] Step S3: Real-time collection of status information of the factory's internal and external logistics resources, referred to as internal logistics information and external logistics information, respectively, and modeling of the internal logistics information to generate a factory internal logistics capacity map; Generating the factory internal logistics capability diagram includes: The Petri net modeling tool is used to construct an internal logistics capability diagram of the factory, which includes the topological connection relationship and resource transfer rules between logistics nodes. The internal logistics information includes the occupancy and service capacity of logistics nodes. Logistics nodes include fixed facilities, mobile equipment and external interfaces within the factory. The resource transfer rules are constructed based on the service capacity of each logistics node, including task processing rhythm, path smoothness and accessibility conditions.

[0029] The fixed facilities include transportation channels and loading and unloading sites; mobile equipment includes but is not limited to automatic guided vehicles (AVGs), forklifts, manual transport vehicles and other in-plant transportation tools; the external interface is the loading and unloading area that connects to the transportation fleet.

[0030] Internal logistics information is as follows: The occupancy status of logistics nodes refers to the current occupancy status of each logistics unit (such as loading and unloading areas, conveyor belt starting and ending points, buffer areas, and transfer platforms), such as "idle", "in execution", "queued", etc. This status is obtained in real time through embedded sensors, RFID readers, MES interfaces, etc., and is used as the status mapping of each location in the Petri net; The service capacity of a logistics node includes parameters such as the number of tasks it can handle per unit time (e.g., an AGV line can handle 6 tasks per hour), the processing unit cycle time (e.g., the time required for each load), and the maximum buffer capacity (e.g., the ability of a conveyor belt to simultaneously accommodate 5 pallets). This data can be sourced from a process master data system, a historical operation analysis system, or manual calibration. Topological connection relationships refer to the accessibility and path connectivity structure between logistics nodes. For example, "from the raw material warehouse to the injection molding workshop, you need to pass through channel A2 and elevator B1." These relationships are automatically generated by importing CAD layout drawings or PLC configuration drawings into modeling tools. Resource transfer rules represent the conditions for migrating tasks between different nodes. For example, "transfer from node X to node Y can only be triggered when the AGV is idle and the path is unoccupied." This logic is embedded as a transition trigger rule in the Petri net for dynamic simulation and state evolution. For example: In a factory AGV system, node A is the loading station, node B is the processing area, and node C is the unloading station. If the current queue length of node B is 2, the maximum processing capacity is 3, the processing time per unit is 10 minutes, the current number of AGVs is 5, and the loading capacity of each AGV is 1 vehicle / time, then the current effective service capacity of the A→B→C path can be obtained through modeling, and the path availability and task load can be evaluated accordingly; the factory internal logistics capacity diagram formed by the above modeling results will serve as the basic structural input for subsequent path planning, scheduling optimization and bottleneck identification.

[0031] Step S4: Model the factory, logistics supplier, and its external transport fleet as intelligent agents. A dynamic game model based on a multi-agent reinforcement learning framework is used to generate the optimal response strategy for each agent using a hierarchical game strategy, and output an external logistics resource supply strategy. The hierarchical game strategy is used to generate the optimal response strategy of each agent, including: The factory, logistics provider, and their external transport fleet are each modeled as an intelligent agent for reinforcement learning. Each agent constructs a set of observation states based on order feature data and external logistics information, and defines a preset set of optional actions corresponding to its role. A dynamic game model is constructed, with the factory acting as the leader agent to generate the leader's strategy. The logistics provider and its affiliated external transport fleet serve as follower agents. After acquiring the leader's strategy, they use reinforcement learning to optimize their own response strategies. This optimization process is based on a set reward function, calculated only based on the local state of their current transport mission. The reward function is a weighted combination of transportation cost, on-time delivery rate, and path load balancing. The optimization results are output as an external logistics resource supply strategy, including transportation route allocation, vehicle scheduling, supply time windows, and task acceptance relationships. This external logistics resource supply strategy is structured as a transportation time window matrix and a dynamic capacity allocation table, which serve as input boundary conditions for the joint scheduling optimization model and are updated in real time via the agent communication interface.

[0032] The observation state set is derived from the order feature data in step S1 and the external logistics information in step S3 (such as vehicle idle status, task waiting queue length, and current road traffic load). For example, the observation state of transport fleet A is [fleet idle rate 60%, number of waiting tasks for route X 3, and next period transport window opening time t = 5 minutes]. The set of optional actions is set according to the agent type: for factories, the optional actions include at least whether to split orders, set task release time, select task merging strategy, and adjust resource priority allocation method; for transport fleets, the optional actions include at least whether to accept the current order, which transportation route to choose, the planned departure time, and whether to participate in task reallocation; for logistics providers, the optional actions include at least allocating the number of vehicles, specifying the service time window, and adjusting the loading strategy of each fleet; The reward function design comprehensively considers objectives such as logistics costs (unit transportation costs (obtained through actual logistics operators), empty trip rate), delivery timeliness (whether the task is completed within the specified window), and route balance (whether the same section is overloaded). For example, if a transport fleet chooses route Y and completes the task 2 minutes early but incurs an additional 30 yuan in fuel costs, the strategy score is calculated based on the weighted values ​​(cost accounts for 0.4, on-time performance accounts for 0.6); The leader-follower model is based on Stackelberg game logic. The factory, as the publisher of production plans, imposes a priori constraints on downstream agents through its production schedule and demand release window. Logistics providers and transport fleets then optimize resources and allocate resources accordingly within these constraints. This hierarchical structure ensures overall system consistency. The data structure of the external logistics resource supply strategy includes but is not limited to: transportation route allocation: such as "Fleet A is assigned to route P1, and the route load is expected to be 80%"; vehicle scheduling: such as "Vehicle 3 is scheduled to depart at 15:00 and is expected to arrive at 18:20"; supply time window: such as "The available window of loading and unloading area B is 16:00~17:00"; task acceptance relationship: such as "Order O1 is accepted by logistics supplier L and assigned to fleet T1 for transportation."

[0033] The external logistics resource supply strategy is structured as follows: The transportation time window matrix is ​​used to describe the service time range of each external logistics resource (such as transportation fleet, loading and unloading port, and supplier scheduling platform) for different orders or tasks. It is defined in the form of a two-dimensional matrix, including: row index, corresponding to the order number or loading and unloading node; column index, corresponding to the external logistics unit (such as fleet number, loading and unloading port number); cell value, the serviceable time interval (such as [08:30,09:00]).

[0034] Example: For order O101, T_O101,T1=[10:00,10:30] in the transportation time window matrix indicates that fleet T1 can provide transportation services for O101 during this period.

[0035] The dynamic capacity allocation table is used to represent the capacity arrangements of various external transportation resources in different time periods, including the current available capacity, the upper limit of task acceptance, historical load trends, etc. It is usually modeled with time as the main axis and defined as: time segments (such as every 30 minutes); the maximum number of dispatchable times, number of tasks undertaken, average load, etc. of each fleet or logistics unit in each time period.

[0036] Example: The capacity allocation table for transport fleet T3 is shown below:

[0037] Step S5: Based on the order feature data, the factory internal logistics capacity map, the external logistics resource supply strategy and the shadow tasks, a joint scheduling optimization model based on graph structure modeling is established to coordinate the scheduling optimization of production tasks and internal logistics routes; The joint scheduling optimization model includes an order-driven production process flow chart constructed based on order feature data and an internal logistics network chart constructed based on the factory's internal logistics capacity chart; the process flow chart uses order operation units and production equipment as nodes and process dependency relationships as edges; the internal logistics network chart uses logistics nodes as nodes and topological connection relationships as edges; the production equipment nodes in the production process flow chart are automatically mapped to the nearest logistics nodes in the internal logistics network chart to form a joint graph model; the objective function is set to minimize production task delays and maximize equipment and transportation resource utilization, and the resource transfer rules of the factory's internal logistics capacity chart and the external logistics resource supply strategy are combined as boundary conditions for collaborative scheduling optimization; When the customized order has not been fully confirmed but there is a typical configuration cluster generated by configuration intention clustering, shadow tasks are introduced into the joint graph model as pre-scheduling tasks for resource pre-occupancy and path pre-scheduling.

[0038] The order-driven production process flow chart is a multi-stage task dependency graph constructed based on the process requirement fields in the order feature data. Nodes represent specific processes or required equipment, and edges represent the sequence of processes. For example, if order O3 requires "rough milling → fine milling → heat treatment → inspection," four process nodes and their dependency edges are constructed accordingly. The internal logistics network graph is derived from the factory internal logistics capacity graph constructed in step S3. The logistics nodes are abstracted into graph nodes, and the accessible paths between nodes are mapped into edges. At the same time, edge weight attributes (path processing time, channel width level, etc.) are added as a reference for scheduling optimization. External logistics resource supply strategies are introduced as input boundary conditions, including a transportation time window matrix and a dynamic capacity allocation table. These constraints limit the earliest and latest arrival times of certain tasks and the transportation resources they can use. For example, order O5 must be shipped before 5:00 PM and can only use fleet T3 assigned by logistics provider L1. This condition will bind the time window constraint and capacity matching between the task node and the logistics exit node in the joint model. The joint graph model is a multi-layered heterogeneous graph structure generated by automatically mapping equipment nodes in the process flow diagram to the nearest logistics nodes in the internal logistics diagram. This mapping is generated by combining nearest neighbor logic (such as minimizing spatial distance and shortest accessible paths) with pre-set rules (such as binding equipment to fixed docking points), ensuring that production tasks can be fully mapped to feasible paths. Objective functions include but are not limited to minimizing total task delay, maximizing production equipment utilization, minimizing transportation resource conflict rate, minimizing the number of job batch changes, and other multi-objective combinations, supporting weighted settings or staged optimization; In addition to traditional production scheduling constraints (such as equipment non-overlapping and process sequence restrictions), the following key boundary conditions are also introduced: task start time window restrictions based on external supply strategies; path capacity and task scheduling rhythm based on logistics capability graphs; and task migration conditions based on resource transfer rules. For example, for order O6 with the process of "assembly → inspection → packaging", if its delivery time is T+6h, combined with the internal logistics capacity diagram, it is known that the "assembly area → inspection area" channel is often congested between T+3h and T+4h. In this case, the scheduling optimization model will move the inspection node forward or backward within the allowable range, or adjust the transportation route to avoid high-load sections.

[0039] Step S6: Generate work instructions and transportation scheduling tasks for each production unit and logistics node based on the collaborative scheduling optimization results; Among them, the operation instruction refers to the specific execution action sequence generated for the production unit, including but not limited to: task number, operation start and end time, required tooling and process parameters, task priority and delivery deadline, and logistics interface information. This instruction is sent to the equipment control layer or operation terminal through the MES system to drive the production equipment to execute according to the schedule; For example: "Task T1001, start execution at 08:00 on June 20, equipment number CNC-05, use fixture A23, program path 'P12345.NC', after completion, the material will be transferred to loading and unloading port L4"; Transportation scheduling tasks refer to the instruction plans generated for in-plant mobile logistics resources (such as AGVs, carts, and forklifts) and shipment scheduling with external interfaces. The content includes: task number, pickup and delivery points, allowed window time period, allocation resource number, urgency tag, route preference, etc. For example: "AGV scheduling task D233, pick up items from buffer S5 to loading and unloading point L4, task priority is medium, execution window 10:00~10:20, recommended path is Route-A"; Operation instructions and transportation scheduling tasks are centrally scheduled and issued through the scheduling platform, forming a task execution queue. When the external logistics resource supply strategy includes supply window restrictions or scheduling time recommendations, the system will simultaneously embed such constraints into the transportation tasks to ensure overall rhythm matching. Supports the joint issuance of task bundling and task splitting strategies: certain work tasks and their corresponding transportation tasks will be bundled and issued as a group of concurrent tasks to avoid idle equipment waiting or route congestion; it also allows some bulk tasks to be split into multiple transportation batches, and scheduling tasks are generated separately; The scheduling granularity and execution accuracy in this process are limited by the actual availability of current internal logistics resources and system feedback delays. Therefore, the execution system supports dynamic adjustment and rescheduling interfaces to ensure that the scheduling results are adaptable and robust in a dynamic environment.

[0040] Step S7: During the execution of the transport scheduling task, internal logistics information is collected in real time to obtain the congestion index of the transport scheduling task. When the congestion index is abnormal, the transport scheduling task is dynamically adjusted; The methods for determining abnormal congestion indicators include task congestion and local logistics capacity bottlenecks, and the congestion indicators are calculated in batches at fixed time intervals; the content of the dynamic adjustment includes job order priority, path selection and batch division, and the dynamic adjustment must meet the constraints of the process dependency of the production process flow chart.

[0041] The method for determining task congestion is to obtain the occupancy status of each logistics node in the internal logistics capacity diagram of the factory, calculate the variance of the resource occupancy rate in the occupancy status, and when the resource occupancy rate variance exceeds the set threshold and the duration exceeds the set time, it is determined to be task congestion; the method for determining local logistics capacity bottlenecks is to identify the logistics nodes where resource request blocking occurs continuously in the internal logistics capacity diagram of the factory, and detect whether the task waiting time of the node exceeds the set threshold in multiple consecutive scheduling cycles. If so, it is determined to be a local logistics capacity bottleneck.

[0042] The congestion index is calculated based on the factory internal logistics capacity diagram constructed in step S3. The status of each logistics node is abstracted into a Petri net identifier. The following key indicators are calculated through periodic sampling: resource occupancy rate, which is the percentage of the node's effective load per unit time; task waiting time, which is the average waiting time of tasks in the queue; occupancy rate fluctuation, which is the variance of resource utilization rate, used to identify traffic imbalance; blocking frequency, which is whether resource request failures occur in multiple consecutive cycles; The congestion index is automatically calculated at fixed scheduling intervals (such as 5 minutes). If any index exceeds the set threshold, the adjustment module is triggered. Specific adjustments include: job order priority adjustment, increasing the resource queue priority occupied by urgent orders, and prioritizing scheduling; route selection optimization, switching to logistics channels or backup routes with lighter current loads; batch division and reconstruction, splitting large tasks into multiple small batches for scheduling to reduce instantaneous channel pressure, or merging multiple microtasks to reduce the frequency of handling; adjustment constraints indicate that all adjustment operations must meet the process dependencies of the production process flow chart, that is, ensuring that the subsequent process can be scheduled only after the previous process is completed to avoid logical errors.

[0043] Example: The system detects that task T203 has waited for more than 12 minutes on the assembly→inspection path for three consecutive cycles. The resource utilization variance reaches the upper threshold, indicating task congestion. The platform responds by postponing the inspection process by 5 minutes and adjusting it to backup path B. At the same time, the priority of task T203 is increased by one level. The congestion is expected to be eliminated within two cycles.

[0044] Step S8: When the congestion index of the transport scheduling task is still abnormal after dynamic adjustment, a path conflict simulation prediction of the transport scheduling task is performed, and the simulation prediction results are fed back to the joint scheduling optimization model to trigger adaptive rescheduling and form a decision closed loop.

[0045] Path conflict simulation prediction refers to building a time-driven simulation model based on the current transportation scheduling plan, internal logistics resource status and path topology relationship to predict and analyze path conflicts, equipment resource preemption, task accumulation and other issues that may occur in multiple future scheduling cycles. The simulation inputs include: the current set of tasks to be executed and their time windows (from step S6); the factory internal logistics capacity diagram (Petri net structure); the current status and historical traffic statistics of internal logistics nodes (from step S7); and the delivery rhythm and window restrictions set by the external supply strategy. The prediction model uses event-driven scheduling simulation (Discrete-Event Simulation), with a simulation granularity as accurate as 5 seconds for transportation scheduling, to identify in advance: Path conflict: if two vehicles request the same road section at the same time; Channel congestion: The accumulation of tasks causes the logistics channel to be continuously blocked for a period exceeding the tolerance period; Blocking loop: Due to unreasonable path allocation, tasks cross and mutually exclude each other, forming a locked state; Feedback mechanisms include: If the conflict is unavoidable, a "high-risk path set" and a "task rescheduling suggestion list" will be automatically generated; The system inputs this feedback information into the joint scheduling optimization model as a constraint enhancement or penalty item to trigger "adaptive rescheduling"; The joint model automatically regenerates the job plan and scheduling path based on the current scheduling status, boundary conditions and feedback information; Rescheduling optimization maintains the original scheduling dependency basis, focusing on adjusting variables such as conflict paths, resource utilization density, task start and end times, to ensure that the goal is achieved with the minimum change based on the original plan.

[0046] Example: The simulation system detects that tasks T402 and T403 are both scheduled to pass through logistics node L7 between 9:00 and 9:10, and only one forklift resource is available. If they are executed according to the original path, traffic will be blocked. The system generates a path conflict warning and recommends changing the path of T403 to "L5→L8→L9" and adjusting the start time to 3 minutes later. After entering the joint scheduling model for optimization, the new scheduling task is issued.

[0047] Example 2, a production optimization system combined with logistics management, see Figure 1 As shown, it includes the following modules: Personalized order parsing module: used to receive personalized customer orders and extract order feature data; Customization Intention Clustering Module: This module is used to obtain and cluster customers' configuration intentions before a customized order is fully confirmed, generate typical configuration clusters, construct corresponding shadow tasks, and manage the introduction, activation, and release of shadow tasks. Logistics status collection module: used to collect status information of internal and external logistics resources of the factory respectively, and generate internal logistics information and external logistics information; Internal logistics modeling module: used to build the factory internal logistics capacity diagram based on internal logistics information, using Petri net modeling tools to describe the topological structure of logistics nodes and resource transfer rules; External logistics optimization module: used to model factories, logistics suppliers and their subordinate transport fleets as intelligent agents, build dynamic game models, and generate external logistics resource supply strategies; Supply strategy management module: used to express the external logistics resource supply strategy in a structured manner, implement its regular update through the communication interface, and generate boundary conditions for the joint scheduling model to call; The joint scheduling optimization module establishes a joint scheduling optimization model based on order feature data, factory internal logistics capacity diagram, external logistics resource supply strategy and shadow tasks, and performs collaborative scheduling optimization of production tasks and internal logistics routes; The scheduling execution module generates operation instructions and transportation scheduling tasks for each production unit and logistics node based on the collaborative scheduling optimization results; The congestion monitoring and simulation feedback module is used to monitor whether the congestion indicators of the transportation scheduling tasks are abnormal, and perform path conflict simulation prediction when the anomaly persists, and feed the simulation results back to the joint scheduling optimization module to trigger rescheduling operations.

[0048] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A production optimization method combined with logistics management, characterized in that: include: Receive personalized customer orders and extract order feature data; For personalized customer orders for customized products, called customized orders, before the actual customized order is fully confirmed, the customer's configuration intention is obtained and cluster analysis is performed to generate typical configuration clusters. These are used to construct corresponding shadow tasks. These shadow tasks are introduced as pre-scheduling inputs into the joint scheduling optimization model. Real-time collection of factory internal and external logistics resource status information, referred to as internal logistics information and external logistics information, and modeling of internal logistics information to generate the factory internal logistics capacity map; Factories, logistics providers, and their external transport fleets are modeled as agents. A dynamic game model based on a multi-agent reinforcement learning framework is used to generate the optimal response strategy for each agent using a hierarchical game strategy, which then outputs an external logistics resource supply strategy. A joint scheduling optimization model based on graph structure modeling is established based on order feature data, factory internal logistics capacity map, external logistics resource supply strategy and shadow tasks to coordinate the scheduling optimization of production tasks and internal logistics routes. Generate work instructions and transportation scheduling tasks for each production unit and logistics node based on the collaborative scheduling optimization results; During the execution of the transport scheduling task, internal logistics information is collected in real time to obtain the congestion index of the transport scheduling task. When the congestion index is abnormal, the transport scheduling task is dynamically adjusted; When the congestion indicator of the transport scheduling task is still abnormal after dynamic adjustment, a path conflict simulation prediction of the transport scheduling task is performed, and the simulation prediction results are fed back to the joint scheduling optimization model to trigger adaptive rescheduling and form a decision-making closed loop.

2. A production optimization method combined with logistics management according to claim 1, characterized in that: Generating the factory internal logistics capability diagram includes: The Petri net modeling tool is used to construct an internal logistics capability diagram of the factory, which includes the topological connection relationship and resource transfer rules between logistics nodes. The internal logistics information includes the occupancy and service capacity of logistics nodes. Logistics nodes include fixed facilities, mobile equipment and external interfaces within the factory. The resource transfer rules are constructed based on the service capacity of each logistics node, including task processing rhythm, path smoothness and accessibility conditions.

3. The production optimization method combined with logistics management according to claim 1, characterized in that: The hierarchical game strategy is used to generate the optimal response strategy of each agent, including: The factory, logistics provider, and their external transport fleet are each modeled as an intelligent agent for reinforcement learning. Each agent constructs a set of observation states based on order feature data and external logistics information, and defines a preset set of optional actions corresponding to its role. A dynamic game model is constructed, with the factory acting as the leader agent to generate the leader's strategy. The logistics provider and its external transport fleet serve as follower agents. After acquiring the leader's strategy, they use reinforcement learning to optimize their own response strategies based on a predefined reward function. This reward function is a weighted combination of transportation cost, on-time delivery rate, and path load balancing. The optimization results are output as external logistics resource supply strategies, including transportation route allocation, vehicle scheduling, supply time windows, and task acceptance relationships.

4. The production optimization method combined with logistics management according to claim 3, characterized in that: The external logistics resource supply strategy is structured and expressed as a transportation time window matrix and a dynamic transportation capacity allocation table, which serve as the input boundary conditions of the joint scheduling optimization model and are updated in real time through the intelligent agent communication interface.

5. The production optimization method combined with logistics management according to claim 1, characterized in that: The joint scheduling optimization model includes an order-driven production process flow chart based on order feature data and an internal logistics network chart based on the factory's internal logistics capacity chart. The process flow chart uses order units and production equipment as nodes and process dependency relationships as edges. The internal logistics network chart uses logistics nodes as nodes and topological connection relationships as edges. Automatically map the production equipment nodes in the production process flow diagram to the nearest logistics nodes in the internal logistics network diagram to form a joint graph model. Minimizing production task delays and maximizing equipment and transportation resource utilization are set as the objective functions. Collaborative scheduling optimization is performed by combining the resource transfer rules of the factory's internal logistics capacity diagram and the external logistics resource supply strategy as boundary conditions. When the customized order has not been fully confirmed but there is a typical configuration cluster generated by configuration intention clustering, shadow tasks are introduced into the joint graph model as pre-scheduling tasks for resource pre-occupancy and path pre-scheduling.

6. A production optimization method combined with logistics management according to claim 5, characterized in that: The generation and scheduling of the shadow task includes: When a custom order has not yet been fully confirmed, the customer's configuration intent is constructed as a vector and clustered. This generates representative typical configuration clusters. For each configuration cluster, a typical process flow and resource requirement template is generated, and shadow tasks are constructed accordingly. Shadow tasks are introduced into the joint scheduling optimization model as soft constraints to implement resource pre-occupation and path pre-scheduling before the custom order is confirmed. When the actual order is placed, the corresponding shadow tasks are matched and activated. Unmatched shadow tasks are released after a set time window to improve adaptability to the diversity and uncertainty of custom orders. The configuration intention includes the customer's configuration behavior data and historical order data on the customization platform.

7. The production optimization method combined with logistics management according to claim 1, characterized in that: The methods for determining abnormal congestion indicators include task congestion and local logistics capacity bottlenecks, and the congestion indicators are calculated in batches at fixed time intervals; the content of the dynamic adjustment includes job order priority, path selection and batch division, and the dynamic adjustment must meet the constraints of the process dependency of the production process flow chart.

8. The production optimization method combined with logistics management according to claim 7, characterized in that: The method for determining task congestion is to obtain the occupancy status of each logistics node in the internal logistics capacity diagram of the factory, calculate the variance of the resource occupancy rate in the occupancy status, and when the resource occupancy rate variance exceeds the set threshold and the duration exceeds the set time, it is determined to be task congestion; the method for determining local logistics capacity bottlenecks is to identify the logistics nodes where resource request blocking occurs continuously in the internal logistics capacity diagram of the factory, and detect whether the task waiting time of the node exceeds the set threshold in multiple consecutive scheduling cycles. If so, it is determined to be a local logistics capacity bottleneck.

9. A production optimization system combined with logistics management, characterized in that: The system applies any one of claims 1 to 8 of the production optimization combined with logistics management, including: Personalized order parsing module: used to receive personalized customer orders and extract order feature data; Customization Intention Clustering Module: This module is used to obtain and cluster customers' configuration intentions before a customized order is fully confirmed, generate typical configuration clusters, construct corresponding shadow tasks, and manage the introduction, activation, and release of shadow tasks. Logistics status collection module: used to collect status information of internal and external logistics resources of the factory respectively, and generate internal logistics information and external logistics information; Internal logistics modeling module: used to build the factory internal logistics capacity diagram based on internal logistics information, using Petri net modeling tools to describe the topological structure of logistics nodes and resource transfer rules; External logistics optimization module: used to model factories, logistics suppliers and their subordinate transport fleets as intelligent agents, build dynamic game models, and generate external logistics resource supply strategies; Supply strategy management module: used to express the external logistics resource supply strategy in a structured manner, implement its regular update through the communication interface, and generate boundary conditions for the joint scheduling model to call; The joint scheduling optimization module establishes a joint scheduling optimization model based on order feature data, factory internal logistics capacity diagram, external logistics resource supply strategy and shadow tasks, and performs collaborative scheduling optimization of production tasks and internal logistics routes; The scheduling execution module generates operation instructions and transportation scheduling tasks for each production unit and logistics node based on the collaborative scheduling optimization results; The congestion monitoring and simulation feedback module is used to monitor whether the congestion indicators of the transportation scheduling tasks are abnormal, and perform path conflict simulation prediction when the anomaly persists, and feed the simulation results back to the joint scheduling optimization module to trigger rescheduling operations.

Images