Clean room logistics multi-agent cooperative scheduling method

Abstract

The application provides a clean room logistics multi-agent collaborative scheduling method, three types of physical agents of production equipment, buffer zone and logistics handling equipment and a coordination agent responsible for scheduling control are constructed when a system is modeled, a global benefit function model is constructed, production continuity, resource utilization rate and disturbance loss are considered in the global benefit function model at the same time, when the coordination agent calculates a global comprehensive benefit value G, all factors influencing the processing process in the system can be covered at the same time, and it is ensured that the coordination agent can dynamically set reasonable scheduling decisions which can be quantified according to system changes; meanwhile, through adjustment of a dynamic weight coefficient, a production line can automatically adjust a scheduling strategy according to different production line switching mode instructions, and through resetting of the dynamic weight coefficient in an emergency recovery mode, a major disturbance is solved, and an abnormal state of the production line is quickly solved.

Description

Technical Field

[0001] This invention relates to the field of production line management technology, specifically to a method for collaborative scheduling of multiple intelligent agents in cleanroom logistics. Background Technology

[0002] In semiconductor manufacturing (such as silicon carbide production lines), photovoltaic manufacturing (such as TopCon production lines), and other precision manufacturing industries with high cleanliness requirements, high-precision products are produced using ultra-clean workshops and cleanroom production lines. Under the control of the control system, intelligent agents based on the production equipment on the production line execute automatic or semi-automatic product processing procedures. Intelligent logistics handling agents, such as Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs), move autonomously between the production line and smart warehouses to complete the tasks of handling and transporting goods.

[0003] Existing technologies include numerous studies on AVG vehicle scheduling and material distribution methods between production lines and warehouses. For example, Peng Yiheng's paper, "Design of a Multi-Agent Collaborative Control System Based on Automated Warehousing" (Jianghan University, journal article), proposes a solution for a multi-agent collaborative control system in automated warehousing scenarios. Through single-agent hardware and software development, collaborative scheduling algorithm design, and verification using the Unity simulation platform, a complete technology chain of "hardware construction—algorithm optimization—simulation verification" is formed. Its core structure includes: a single-agent hardware system (sensors, controllers, execution units), a software control layer (localization, mapping, navigation), a collaborative scheduling algorithm module (cultural asexual reproduction genetic algorithm), and a simulation verification platform. This solution enables multiple agents to collaboratively complete package handling tasks, effectively improving sorting efficiency and resource utilization in automated warehousing scenarios. For example, application document CN101944201A discloses a simulation method for overhead crane scheduling in a steelmaking workshop based on multi-agent intelligence. For overhead crane scheduling in a steelmaking workshop, the method abstracts the overhead crane, workstation, and coordination management unit into intelligent agents, constructing a multi-agent simulation model. The system consists of three types of intelligent agents: coordination management agent, workstation agent, and overhead crane agent. A set of transportation tasks is generated through task occurrence rules. The coordination management agent determines and issues local objectives, the workstation agent selects the target workstation, and the overhead crane agent completes the transportation tasks and performs conflict resolution and optimization. This method fully considers time, space, and resource constraints, can eliminate spatial and temporal conflicts between overhead cranes, shorten the average material transit time, and improve the rationality of the scheduling strategy by introducing global objectives. For example, patent document CN120707048A discloses an automated material delivery method based on delivery vehicles between workstations and warehouses; patent document CN113219983B discloses an automated material transport method between processes based on AGVs; and patent document CN121580877A discloses another simulation optimization method for in-warehouse sorting operation systems.

[0004] However, practical applications have revealed shortcomings in the aforementioned technologies: 1. Existing technologies primarily employ fixed-path and static-priority scheduling logic, lacking a real-time response mechanism to dynamic disturbances affecting the processing, such as equipment failures, replenishment fluctuations, buffer overflows, material mismatches, and congestion of logistics agents, resulting in insufficient system practicality; 2. The scheduling modes in warehousing and workshops are mainly based on unidirectional push or single-task allocation, leading to the contradiction of upstream backlog and downstream material shortages in high-cleanliness production lines; 3. Optimization in existing technologies largely relies on rule settings or single evaluation functions, without establishing a global benefit quantification model, resulting in a lack of quantitative basis for the scheduling decisions made by the agents in response to system changes. These shortcomings mean that existing production line agent control methods cannot meet actual production needs. Summary of the Invention

[0005] To address the problem that existing intelligent agent mobilization methods cannot meet the actual production needs of cleanroom production line logistics, this invention provides a multi-agent collaborative scheduling method for cleanroom logistics. This method can realize autonomous prediction of conflicts and dynamic collaborative adjustment of scheduling strategies during the production line logistics allocation process, thereby opening up a push-pull scheduling closed loop between the production line and logistics, and significantly improving the overall resource utilization rate of the cleanroom production line.

[0006] The technical solution of this invention is as follows: a multi-agent collaborative scheduling method for cleanroom logistics, characterized by comprising the following steps: S1: System Modeling Construct a scheduling system; the scheduling system includes: a coordinating agent and an executing agent; The execution agents include: production equipment agents, buffer agents, and logistics handling agents; The buffer agent is configured for the buffer area, which is used to store materials and includes: a workstation buffer area and a common buffer area; the workstation buffer area is set up separately for each processing workstation to temporarily store materials for each processing workstation; the common buffer area is a shared buffer area for all workstations, which temporarily stores materials when there is no space in the workstation buffer area. The coordinating agent, acting as the core of global coordination, is responsible for issuing scheduling instructions and agent parameters to the three types of executing agents. The agent parameters include: dynamic weight coefficients and a dynamic trigger threshold θ. pull ; The three types of execution agents upload their own status data and disturbance warning information to the coordinating agent in real time, and at the same time receive the scheduling instructions from the coordinating agent to complete autonomous decision-making and collaborative execution. The scheduling behaviors performed by the buffer agent include push scheduling and pull scheduling. When the production unit completes the processing task, the push scheduling behavior is triggered, which notifies the coordinating agent that the materials processed at this workstation have entered the logistics mobilization ready state, allowing them to be transported to the downstream workstation. When the material in the buffer area corresponding to the production unit is lower than the dynamic pull trigger threshold θ pull When the pull scheduling behavior is triggered, the coordinating agent is notified that materials need to be replenished at this workstation. S2: Construct a global profit function model; G = α·P + β·R - γ·L; Where G is the global comprehensive benefit value, α is the dynamic weight coefficient of production continuity, β is the dynamic weight coefficient of resource utilization, and γ is the dynamic weight coefficient of disturbance loss, satisfying α≥0, β≥0, γ≥0, and α+β+γ=1. P represents the production continuity benefit, which is calculated from the equipment failure and downtime and the cumulative time of depletion of buffer materials. R represents the resource utilization benefit, which is the percentage of the average working time of the logistics handling agent. AGV and the average time occupied by the cache T buffer Statistics show that; L represents the disturbance loss, which is calculated using real-time monitoring data from the production line. S3: Based on the aforementioned scheduling system and the global comprehensive benefit value G, monitor the logistics scheduling process of the production line; S4: At the beginning of each statistical period, the coordinating agent summarizes the global information, reads the latest dynamic weight coefficients α, β, γ, and calculates the global comprehensive benefit value G. When the coordinating agent controls the material movement behavior, it sets the dynamic pull trigger threshold θ for each buffer agent based on the real-time global comprehensive benefit value G. pull ; S5: Construct an adaptive weight adjustment mechanism, in which the coordinating agent adjusts the dynamic weight coefficients in real time according to the actual production situation during the scheduling process; the adaptive weight adjustment mechanism specifically includes the following operations: a1: Calculate the total disturbance duration percentage D in real time and compare D with the preset disturbance duration percentage threshold η1; D=ΣT disturb / T total , Among them, T total ΣT is the preset statistical period duration. disturb This indicates the cumulative duration of all disturbance events in the system during this statistical period; When D>η1, proceed to step S6; a2: Calculate the fluctuation range FR of the global comprehensive return value G in real time over N consecutive statistical periods, and compare FR with the preset fluctuation range threshold η2; FR=G max -G min ; In the formula, G max G is the maximum value of G over N consecutive statistical periods. min The minimum G value in N consecutive statistical periods; When FR>η2, the system automatically backtracks and analyzes all disturbance events that caused the fluctuation of the G value, and calculates the cumulative duration percentage D of all disturbance events. retro Proceed to step S7; a3: Real-time monitoring of production line switching mode instructions; if a high disturbance response mode or capacity sprint mode instruction is received, then execute step S8; a4: Monitor the global comprehensive benefit value G in real time. If G < 0, then execute step S9. S6: Calculate the temporary value of the dynamic weighting coefficient based on the real-time value of the total disturbance duration percentage D: α temp =α·(1+k·D); β temp =β·(1-k·D); γ temp =γ·(1+2k·D); In the formula, α temp For the temporary values ​​of the dynamic weighting coefficients for continuous production, β temp γ is a temporary value for the dynamic weighting coefficient of resource utilization. temp This represents the temporary value of the dynamic weighting coefficient for the disturbance loss; k is the adjustment sensitivity, k∈(0, 0.5]; Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. S7: Based on the proportion of total disturbance duration D retro The real-time value is used to calculate the temporary value of the dynamic weighting coefficient. α temp =α·(1+k·D retro ); β temp =β·(1-k·D retro ); γ temp =γ·(1+2k·D retro ); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. S8: Determine whether the received instruction to switch production line modes is a high-disturbance response mode; If so, then based on the preset high disturbance ratio benchmark value D high Calculate the temporary values ​​of the dynamic weighting coefficients: α temp =α·(1+k·D high ); β temp =β·(1-k·D high ); γ temp =γ·(1+2k·D high ); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. Otherwise, it indicates that the received production line switching instruction is in capacity sprint mode, and the adjustment sensitivity k is restricted to k∈(0, 0.3], and a temporary value of the dynamic weighting coefficient is calculated: α temp =α·(1+k); β temp =β·(1-k); γ temp =γ·(1+2k); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. S9: Start the emergency recovery mode to reset the weights; after exiting the emergency recovery mode, use the temporary value of the dynamic weight coefficient obtained in the emergency recovery mode as the new dynamic weight coefficient to execute step S10; S10: Based on the latest dynamic weight coefficients, repeat steps S4 to S5.

[0007] Its further features are: In step S4, the dynamic pull trigger threshold θ pull The setup method is as follows: b1: Read the pre-set global high-return threshold G in the system. high Global return low threshold G low Buffer stable inventory threshold Q stable and safety stock threshold Q robust ; The global return high threshold G high The threshold for determining the efficient operation of the production line; The global return low threshold G low Production line operation early warning threshold; The stable inventory threshold Q of the buffer area stable To maintain the minimum baseline inventory for regular continuous production of downstream equipment in the workstation buffer area; The safety stock threshold Q robust The redundancy safety stock threshold for the buffer to cope with short-term sudden disturbances; b2: Combine the value of the global comprehensive return G with G high and G low A comparison was made to confirm the dynamic pull trigger threshold θ. pull The possible values ​​of ; When G≥G high When, θ pull =round(Q stable ); When G <G low When, θ pull =round(Q robust ); When G low ≤G <G high When, θ pull =round(Qstable +(( Q robust- Q stable )*(GG high ) / (G low -G high ))); In the formula, round() is the rounding function; The method for calculating the production continuity benefit P is as follows: P=(T total -T stop -T lack ) / (T total +ε); In the formula, T total T is the preset statistical period duration. stop T is the cumulative duration of equipment failures and downtime collected from the PLCs of various production equipment. lack ε is the cumulative duration of material depletion as monitored by sensors in the downstream buffer area; ε is a minimal constant. The method for calculating the resource utilization rate benefit R is as follows: R=ω agv ×U AGV +ω buf ×T buffer ; Where, ω agv ω represents the weighting coefficient of logistics handling equipment in resource utilization. buf This represents the weighting coefficient of the cache in resource utilization. U AGV For n agv The average working time percentage of each logistics handling intelligent agent; T buffer For n buffer The average time spent in each cache area; ; T i agv,busy Let be the working time of the i-th logistics handling agent within the statistical period; ; T i buffer,occupied This represents the time the i-th buffer is occupied within the statistical period; The method for calculating the disturbance loss L is as follows: L=(λ1·T fault +λ2·T congest +λ3·T error ) / T total ; In the formula, λ1 is the disturbance weight coefficient corresponding to the cumulative duration of equipment failure, λ2 is the disturbance weight coefficient corresponding to the cumulative duration of congestion in the logistics intelligence agent, and λ3 is the disturbance weight coefficient corresponding to the cumulative duration of material mismatch; λ1≥0, λ2≥0, λ3≥0, and λ1+λ2+λ3=1; T fault T represents the cumulative duration of equipment failure. congest T represents the cumulative congestion time for logistics intelligent systems. error Cumulative duration of material mismatch; In step S5, when calculating the percentage of each total disturbance duration D, D is restricted to the range of [0,1]. If D>1, then D is assigned the value 1. Calculate the cumulative duration percentage D of all disturbance events over N periods. retro For each total disturbance duration D within each period, the percentage of D is calculated separately, and each D is restricted to the range [0,1]. If D>1, then D is assigned the value 1. Finally, the N D values ​​are added together to obtain D. retro ; It also includes a lower bound protection mechanism; In the lower bound protection mechanism, the temporary value β is the dynamic weight coefficient of the resource utilization rate. temp Set the minimum weight lower bound value β min ,β min >0; Each calculation yields a β temp After that, β will be temp With β min Compare; If β temp <β min Then β min The value is assigned to β temp ; In step S9, the emergency recovery mode includes the following operations: c1: Calculate the negative depth Depth corresponding to the global comprehensive benefit value G; Depth = |G|; c2: Determine the emergency level based on the negative depth, and apply differentiated adjustment ranges based on the emergency level; Depth ≥ D H At this time, the corresponding emergency level is: Critical Emergency; proceed to step c3; D L ≤ Depth <D H At this time, the corresponding emergency level is: moderate emergency; proceed to step c4; Depth <D L At this time, the corresponding emergency level is: mild emergency; proceed to step c5; In the formula, D HD is the preset high-level judgment threshold. L The preset low-level judgment threshold; c3: After calculating the temporary value of the dynamic weight coefficient using an aggressive adjustment, proceed to step c6; α temp =min(α×1.8,0.75); γ temp =min(γ×1.5,0.40); β temp =1-α temp -γ temp ; c4: After calculating the temporary value of the dynamic weight coefficient using the standard adjustment range, proceed to step c6; α temp =min(α×1.5,0.70); γ temp =min(γ×1.3,0.35); β temp =1-α temp -γ temp ; c5: After calculating the temporary value of the dynamic weighting coefficient using a moderate adjustment range, proceed to step c6; α temp =min(α×1.3,0.65); γ temp =min(γ×1.2,0.30); β temp =1-α temp -γ temp ; c6: Performs weight boundary protection and normalization processing for dynamic weight coefficients; c7: Force overwrite system parameters, specifically including the following operations: The threshold θ for dynamically pulling all caches pull Set to the preset safety stock threshold Q robust ; The statistical period duration T total The value is updated to be adjusted to a temporary statistical period T. total =T total / 2; c8: Based on the latest dynamic weighting coefficients and dynamic pull trigger threshold θ pull and temporary statistical period T total ', Repeat steps S4~S5; The system will monitor its operation in real time and exit the emergency recovery mode when the following conditions are met simultaneously. Condition 1: Two consecutive temporary statistical periods T total All satisfy: G≥0.3; Condition 2: Two consecutive temporary statistical periods T total 'All satisfy: (T) fault +T congest ) / ΣT disturb <0.5; In the formula, T fault T represents the cumulative duration of equipment failure. congest Cumulative congestion time for logistics intelligent agents; ΣT disturb This indicates the cumulative duration of all disturbance events in the system during this statistical period; Step c6 specifically includes the following operations: c61: Read the preset β temp The lower bound β min , If β temp <β min Then β min Assigned to β temp 'Then execute step c62;' Otherwise, proceed to step c63; c62: Readjust weights and calculate α temp 'and γ temp ': α temp '=α temp ×((1-β temp ') / (α temp +γ temp )); γ temp '=γ temp ×((1-β temp ') / (α temp +γ temp )); and with α temp '、 β temp 'and γ temp 'Execute step c7 as the latest dynamic weight coefficient;' c63: α is obtained after normalization. temp '、 β temp 'and γ temp ': α temp '=α temp / S; β temp '=β temp / S; γ temp '=γ temp / S; S=α temp +β temp +γ temp ; and with α temp '、 β temp 'and γ temp 'Execute step c7 as the latest dynamic weight coefficient;' Before step S10 is executed, the following operations also need to be performed: d1: α is obtained after normalization. new , β new and γ new ; α new =α temp / S; β new =β temp / S; γ new =γ temp / S; S=α temp +β temp +γ temp ; d2: Read the preset amplitude change threshold δ max If any of the following conditions are met, then proceed to step d3; otherwise, proceed with step α. new , β new and γ new Step S9 is executed as the latest dynamic weighting coefficient; Amplitude judgment condition 1: |α new -α |>δ max ; Amplitude judgment condition 2: |β new -β |>δ max ; Amplitude judgment condition 3: |γ new -γ |>δ max ; d3: Perform anti-oscillation processing on each of the aforementioned dynamic weight coefficients, specifically including the following steps; When α new -α>δ max At that time, α new '=α new +δ max ; When α new –α<-δ max At that time, α new '=α new -δ max ; When |α new –α|≤δ maxWhen, α new ’ = α new ; When β new - β > δ max When, β new ’ = β new + δ max ; When β new – β < - δ max When, β new ’ = β new - δ max ; When |β new – β| ≤ δ max When, β new ’ = β new ; When γ new - γ > δ max When, γ new ’ = γ new + δ max ; When γ new – γ < - δ max When, γ new ’ = γ new - δ max ; When |γ new – γ| ≤ δ max When, γ new ’ = γ new ; d4: Weight normalization processing: α norm = α new ’ / (α new ’ + β new ’ + γ new ’); β norm = β new ’ / (α new ’ + β new ’ + γ new ’); γ norm = γ new ’ / (α new ’ + β new ’ + γ new ’); d5: Using α norm , β norm and γ norm as the latest dynamic weight coefficients to execute step S9; In step S3, when the system runs for the first time, it reads the initial value of the preset dynamic weight coefficient and performs subsequent calculations.

[0008] This application provides a multi-agent collaborative scheduling method for cleanroom logistics. During system modeling, three types of physical agents—production equipment, buffer zones, and logistics handling equipment—are constructed, along with a coordinating agent responsible for scheduling and control. A global benefit function model is also built, simultaneously considering production continuity, resource utilization, and disturbance losses. This ensures that when the coordinating agent calculates the global comprehensive benefit value G, it can cover all factors affecting the processing process, enabling the coordinating agent to dynamically set reasonable and quantifiable scheduling decisions based on system changes. During operation, the three types of physical agents report in real-time the factors affecting the processing process. The disturbance events of the process are included in the calculation of the G value, which effectively improves the practicality of scheduling decisions. In this method, two forms of buffer material scheduling behavior are set up: push scheduling behavior and pull scheduling behavior are set up for the production unit. That is, once the processing operation of a certain workstation is completed, the coordinating agent is immediately notified so that the material can be scheduled to the downstream workstation. When the material in the buffer corresponding to the production unit is lower than the dynamic pull trigger threshold, the coordinating agent is immediately notified that the material needs to be replenished. By setting push scheduling behavior and pull scheduling behavior, the scheduling efficiency of materials on the production line can be effectively improved, and the problem of upstream backlog and downstream material shortage can be avoided. This method monitors the scheduling process on the production line based on the scheduling system and the global comprehensive benefit value G. It also designs an adaptive weight adjustment mechanism. By monitoring the fluctuation range of the total disturbance duration percentage D and the global comprehensive benefit value G, the dynamic weight coefficients in the system can be adjusted in a timely manner, ensuring that the optimization direction of the scheduling strategy can flexibly adapt to different actual production conditions and external disturbances. Furthermore, this application also enables the production line to automatically adjust its scheduling countermeasures according to different production line switching mode commands by adjusting the dynamic weight coefficients, and resolves major disturbances and quickly addresses abnormal production line states by resetting the dynamic weight coefficients in emergency recovery mode. Attached Figure Description

[0009] Figure 1 A flowchart illustrating a multi-agent collaborative scheduling method for cleanroom logistics; Figure 2 Here is a diagram of the scheduling system architecture; Figure 3 This is a flowchart illustrating the emergency recovery mode. Figure 4 This is a schematic diagram of the system structure for an embodiment. Detailed Implementation

[0010] This method does not involve the processing control of products by production equipment; it serves only as a technical solution for monitoring and adjusting logistics allocation. It aims to improve the efficiency and practicality of logistics allocation in existing product processing lines and is applicable to various types of production lines, not limited to a single product type. This application aims to achieve autonomous prediction of conflicts and dynamic collaborative adjustment of scheduling strategies during logistics allocation by constructing a monitoring method for intelligent production equipment agents, buffer zone agents, and logistics handling agents with autonomous decision-making and collaborative optimization capabilities. This will open up a closed loop for push-pull scheduling in production line logistics, significantly improving the stability, continuity, and overall resource utilization of cleanroom production line logistics allocation in the face of dynamic disturbances such as equipment failure, material mismatch, and path congestion.

[0011] like Figure 1 As shown, this application includes a method for collaborative scheduling of multiple agents in cleanroom logistics, which includes the following steps.

[0012] S1: System modeling.

[0013] like Figure 2 As shown, the scheduling system constructed in this application includes: a coordinating agent and an executing agent; the executing agent includes: a production equipment agent, a buffer agent, and a logistics handling agent.

[0014] The intelligent agent for production equipment is deployed on the programmable logic controller (PLC) or host computer of each piece of production equipment. It collects real-time data on equipment operating status (normal or abnormal), process completion signals, and the real-time capacity of downstream buffer areas. Its core decision-making mechanism dynamically adjusts material flow based on the real-time status of downstream processes: when there is space in the downstream buffer, materials flow directly to the downstream workstation; when the downstream buffer is full, materials are temporarily stored in a common temporary storage area to prevent congestion. Simultaneously, the intelligent agent generates a unique traceability code for each batch of materials and binds it to processing information. When an equipment malfunction is detected, the intelligent agent immediately broadcasts a warning event to the system and automatically suspends the material pushing process for its own equipment to prevent invalid materials from flowing in.

[0015] The buffer zone intelligent agent is implemented based on the inventory management system or reader / writer terminal of each buffer zone, and monitors the inventory quantity, material batches, and available capacity in real time. The buffer zone intelligent agent is set up for the buffer zone, which is used to store materials, including: workstation buffer zone and common buffer zone; the workstation buffer zone is set up separately for each processing workstation, and temporarily stores materials for each processing workstation; the common buffer zone is a shared buffer zone for all workstations, and temporarily stores materials when the workstation buffer zone has no space.

[0016] The scheduling behaviors performed by the buffer agent in this application include push scheduling behavior and pull scheduling behavior; When a production unit completes its processing task, it triggers a push scheduling behavior, notifying the coordinating agent that the materials processed at this workstation have entered the logistics mobilization ready state, allowing them to be transported to downstream workstations. When the material level in the buffer area corresponding to the production unit is lower than the dynamic pull trigger threshold θ pull When this occurs, a pull scheduling behavior is triggered, notifying the coordinating agent that materials need to be replenished at this workstation.

[0017] When the buffer agent acts as the pull trigger, to overcome the risk of resource misallocation or material shortage caused by fluctuations in production line status due to fixed inventory thresholds, this method adopts a dynamic threshold triggering mechanism, with a trigger threshold θ. pull It is linked to the global benefit G, which reflects the overall efficiency of the system. When the buffer agent acts as the trigger (such as in a common temporary storage area), it autonomously decides to trigger the return flow of materials according to the first-in-first-out (FIFO) principle and in combination with the availability and priority of each downstream workstation.

[0018] The logistics handling intelligent agent is the final execution unit of scheduling instructions, deployed on intelligent logistics handling equipment or intelligent transportation equipment, such as Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs). It provides real-time feedback on its own status, location, and task details. Its decision-making follows a multi-level task priority rule, with emergency material replenishment having the highest priority, followed by push tasks, and routine flow tasks having the lowest priority. Tasks of the same priority follow a "first-come, first-served" principle. When executing multi-AGV collaborative tasks, it performs rapid game theory based on local real-time traffic information, autonomously optimizing path selection to avoid congestion. When disturbances such as mechanical failure, insufficient power, or severe path congestion occur, it immediately reports its status to the coordinating intelligent agent. After completing the transportation task, the logistics transportation intelligent agent sends a completion signal to the destination buffer and the coordinating intelligent agent, triggering an atomic update of the inventory status. Throughout the entire process, the unique traceability code of the material is tightly bound to the IDs of all interacting intelligent agents, action timestamps, and geographical location information, forming a complete digital twin trajectory, which is persistently stored in the database, achieving full-process traceability.

[0019] The coordinating agent, acting as the core of overall coordination, is responsible for issuing scheduling instructions and agent parameters to the three types of executing agents. Agent parameters include: dynamic weight coefficients and dynamic trigger threshold θ. pull .

[0020] The three types of execution agents upload their own state data and disturbance warning information to the coordinating agent in real time, and at the same time receive scheduling instructions from the coordinating agent to complete autonomous decision-making and collaborative execution. The local information interaction between the buffer agent and the logistics handling agent, and between the logistics handling agents, as well as the implementation details of the structure, transportation and handling actions and information interaction methods of each type of agent, can be implemented based on existing technologies.

[0021] S2: Construct a global revenue function model.

[0022] To address the challenge of multi-dimensional trade-offs and dynamic evaluation of scheduling effectiveness, this application constructs a system where a coordinating agent (deployed on a production line scheduling server) operates based on a preset statistical periodicity T. total A global comprehensive benefit function model is calculated, serving as the unified optimization objective. In this model, a configurable statistical period duration T is set. total , serving as a unified calculation benchmark for all time cumulative amounts. total It can be set according to production line cycle time and management needs, such as 15 minutes, 1 hour, or one production shift. Unless otherwise specified, all time-related variables below are cumulative values ​​or ratios calculated based on this period.

[0023] The global benefit function model is: G = α·P + β·R - γ·L; Wherein, G is the global comprehensive benefit value, which is used to comprehensively evaluate the overall scheduling efficiency within a statistical period; α is the dynamic weighting coefficient for production continuity, β is the dynamic weighting coefficient for resource utilization, and γ is the dynamic weighting coefficient for disturbance loss. The three dynamic weighting coefficients satisfy: α≥0, β≥0, γ≥0, and α+β+γ=1. The three dynamic weighting coefficients represent the real-time emphasis of the system on the three major performance indicators of production continuity, resource utilization, and disturbance loss.

[0024] P represents the production continuity benefit, calculated from the equipment failure and downtime and the cumulative time of material depletion in the buffer area. Its value range is controlled within [0,1] through parameter design.

[0025] The method for calculating the production continuity benefit P is as follows: P=(T total -T stop -T lack ) / (T total +ε); In the formula, T total T is the preset statistical period duration. stop T is the cumulative duration of equipment failures and downtime collected from the PLCs of various production equipment. lack This represents the cumulative duration of material depletion as monitored by sensors in the downstream buffer area; ε is a minimal constant with a value of 1×10⁻⁶. -6 .

[0026] R represents the resource utilization benefit, which is the percentage of the average working time of the logistics handling agent. AGV and the average time occupied by the cache T buffer Statistics show that the value of R is controlled within the range of [0,1].

[0027] The method for calculating the resource utilization rate benefit R is as follows: R=ω agv ×U AGV +ω buf ×T buffer ; Where, ω agv ω represents the weighting coefficient of logistics handling equipment in resource utilization. buf This represents the weighting coefficient of the cache in resource utilization. U AGV For n agv The average working time percentage of each logistics handling intelligent agent; T buffer For n buffer The average time spent in each cache area; n agv n represents the number of intelligent logistics handling agents participating in production within the system. buffer This represents the number of caches participating in production within the system.

[0028] ; T i agv,busy The working time of logistics handling intelligent agent numbered i within the statistical period; ; T i buffer,occupied This represents the time that cache numbered i has been occupied during the statistical period.

[0029] L represents the disturbance loss, which is calculated from real-time monitoring data of the production line. The value of L is controlled within the range of [0,1].

[0030] The method for calculating the disturbance loss L is as follows: L=(λ1·T fault +λ2·T congest +λ3·T error ) / T total ; In the formula, λ1 is the disturbance weight coefficient corresponding to the cumulative duration of equipment failure, λ2 is the disturbance weight coefficient corresponding to the cumulative duration of congestion in the logistics intelligence agent, and λ3 is the disturbance weight coefficient corresponding to the cumulative duration of material mismatch; λ1≥0, λ2≥0, λ3≥0, and λ1+λ2+λ3=1; T fault T represents the cumulative duration of equipment failure. congest T represents the cumulative congestion time for logistics intelligent systems. error This represents the cumulative duration of material mismatch. In practice, it is the cumulative duration of equipment failure, T. fault The cumulative duration of congestion in logistics intelligent systems (T) congest Cumulative duration T of material mismatch errorThe system utilizes existing intelligent processing systems on production lines, along with physical equipment such as PLC interfaces, RFID readers, and AGVs, to collect and report statistics.

[0031] S3: Based on the scheduling system and the global comprehensive benefit value G, monitor the logistics scheduling process of the production line.

[0032] When the system runs for the first time, it reads the initial values ​​of the preset dynamic weight coefficients and performs subsequent calculations.

[0033] The global comprehensive benefit value G designed in this application is based on the above normalization parameters, and the theoretical range of G is [-1, 1].

[0034] Theoretical maximum value G max =1: This is achieved when production is completely continuous (P=1), resources are fully utilized (R=1), there are no disturbance losses (L=0), and α+β=1.

[0035] Theoretical minimum value G min =-1: This is reached when production is completely interrupted (P=0), resources are completely idle (R=0), and disturbance loss is at its maximum (L=1), while γ=1.

[0036] In actual scheduling optimization, this application also sets up a multi-layer protection mechanism, namely: an adaptive weight adjustment mechanism, to actively suppress G from sliding into the negative region. Under normal and controllable production conditions, the actual observed value of G is usually distributed in the range of [0.3, 0.9]. Only when encountering uncontrollable major disturbances, such as a plant-wide power outage or complete failure of core equipment, and the duration exceeds the system's recovery capacity, can G temporarily drop to a negative value. At this time, the system will trigger the highest level alarm and enter the emergency recovery mode.

[0037] S4: At the beginning of each statistical period, the coordinating agent summarizes global information, reads the latest dynamic weight coefficients α, β, and γ, and calculates the global comprehensive benefit value G; then, based on the real-time global comprehensive benefit value G, it sets a dynamic pull trigger threshold θ for each buffer. pull .

[0038] Specifically, the dynamic pull trigger threshold θ pull The setup method is as follows: b1: Read the pre-set global high-return threshold G in the system. high Global return low threshold G low Buffer stable inventory threshold Q stable Safety stock threshold Q robust ; Global return high threshold G high The threshold for determining the efficient operation of the production line is used to characterize the critical value in which the overall production scheduling is in a good and stable state. Global return low threshold Glow Production line operation early warning thresholds are used to identify states where overall scheduling efficiency declines and there is a risk of production interruption. Buffer stable inventory threshold Q stable The minimum baseline inventory in the workstation buffer area is the core reference indicator for triggering regular replenishment requests to maintain continuous production of downstream equipment. When the real-time inventory in the buffer area falls below this threshold, the buffer area agent automatically initiates a basic replenishment request to the coordinating agent to prevent downstream equipment from stopping due to inventory depletion and to ensure the stability of the regular production cycle. Safety stock threshold Q robust This serves as a redundancy safety stock threshold for the buffer to cope with short-term sudden disturbances, thereby improving the production line's resistance to interference.

[0039] Among them G high G low Q stable Q robust All four parameters are configurable. Initial values ​​are set based on historical data or expert opinions during system initialization. These four parameters can be flexibly adjusted according to the production cycle time, equipment configuration, material characteristics, and on-site conditions of different production lines. This allows the solution to adapt to various conductor packaging and testing production line scenarios, ensuring the versatility and adaptability of this technical solution.

[0040] b2: Combine the value of the global comprehensive return G with G high and G low Compare and confirm the dynamic pull trigger threshold θ pull The value of .

[0041] When G≥G high When, θ pull =round(Q stable In the formula, round() is the rounding function; When G≥G high When θ indicates that the system determines the current production line resource allocation is reasonable and there is no significant risk of disruption, a lenient strategy is adopted. pull The value is set to Q. stable Replenishment is triggered with a lower threshold, and a push-type logistics mode is preferred. Processed materials are directly transferred to the corresponding downstream workstation buffer area, reducing the frequency of replenishment, simplifying logistics transfer links, maximizing overall circulation efficiency, and avoiding redundant temporary storage that occupies resources.

[0042] When G < G low When, θ pull =round(Q robust ); When G < G low When the system determines that the production line is under resource strain or potential disruption, it will then set θ. pullThe value is set to Q. robust A strict replenishment strategy is adopted, with a higher threshold to trigger replenishment, thereby increasing the priority of pull replenishment, focusing on ensuring the supply of materials to downstream equipment, realizing predictive adjustment of logistics requests, and reserving redundant inventory in advance to avoid equipment downtime due to insufficient inventory in downstream equipment and ensure production continuity.

[0043] When G low ≤G <G high When, θ pull =round(Q stable +(( Q robust- Q stable )*(GG high ) / (G low -G high ))); θ is achieved through linear interpolation. pull In Q stable and Q robust The system achieves a smooth transition between different thresholds, balancing efficiency and stability, and adapting to stable production line operation. By combining this threshold with a stable inventory threshold, the system dynamically calculates the real-time replenishment critical value, reserving reasonable inventory redundancy. This effectively addresses unexpected situations such as short-term equipment failures, AGV path congestion, and material delivery delays, ensuring the robustness of the scheduling process.

[0044] This method involves θ pull Adjusting the values ​​allows for regulation of material allocation in the buffer zone. In conjunction with the simultaneous adjustment of the three dynamic weight coefficients, the global comprehensive benefit value G is recalculated, effectively adjusting the task allocation scheme and material flow suggestions of the logistics transportation agent, thereby regulating the production line status.

[0045] S5: To enable the optimization direction of the scheduling strategy to flexibly adapt to different production stages and external disturbances, this method constructs an adaptive weight adjustment mechanism to coordinate the agent to adjust the dynamic weight coefficients in real time according to the actual production situation during the scheduling process; the adaptive weight adjustment mechanism specifically includes the following operations in sequence.

[0046] a1: Monitoring of total disturbance duration exceeding the standard: Calculate the total disturbance duration percentage D in real time and compare D with the preset disturbance duration percentage threshold η1; D=ΣT disturb / T total , Among them, T total ΣT is the preset statistical period duration. disturbThis represents the cumulative duration of all disturbance events in the system during the current statistical period. In actual calculation, the D value is limited to the range of [0,1], meaning that the maximum proportion of disturbances in a single statistical period is 100%. If the calculated D value exceeds 1, the excess part will be automatically truncated to make D=1, thus avoiding excessive weight expansion caused by D exceeding the limit. When D>η1, proceed to step S6.

[0047] The disturbance events in this application include any event in the processing process, such as equipment failure, replenishment fluctuations, buffer overflow, material mismatch, congestion of the logistics agent, etc. The specific method for defining disturbance events can be to statistically analyze abnormal events in historical data that affect the processing process (such as causing processing delays or interruptions), or it can be obtained through expert advice and pre-stored in the system during system construction.

[0048] In practice, the triggering time for monitoring the total disturbance duration exceeding the standard is set in each scheduling cycle T. total After the operation is completed, the total disturbance ratio D during the cycle is calculated. If D > η1, a weight adjustment is triggered at the beginning of the next scheduling cycle to avoid scheduling oscillations caused by frequent adjustments during the cycle.

[0049] a2: Monitoring of excessive fluctuations in global return value G: The fluctuation range FR of the global comprehensive return value G over N consecutive statistical periods is calculated in real time, and FR is compared with the preset fluctuation range threshold η2. FR=G max -G min ; In the formula, G max G is the maximum value of G over N consecutive statistical periods. min The minimum G value in N consecutive statistical periods; When FR>η2, the system automatically backtracks and analyzes all disturbance events that caused the fluctuation of the G value, and calculates the cumulative duration percentage D of all disturbance events. retro Proceed to step S7.

[0050] Calculate the cumulative duration percentage D retro When there are N statistical periods T total If there are multiple disturbance events, the durations of all disturbance events within the statistical period need to be summed up. In actual calculation, the percentage of the cumulative duration D of all disturbance events across N periods is calculated. retro For each total disturbance duration D within each period, the percentage of D is calculated separately, and each D is restricted to the range [0,1]. If D>1, then D is assigned the value 1. Finally, the N D values ​​are added together to obtain D. retro To avoid D retro Exceeding limits.

[0051] In practice, the triggering time for monitoring the fluctuation of the global return value G is set after the end of N consecutive statistical periods. The fluctuation range is calculated, and if it exceeds the limit, a weight adjustment is triggered at the beginning of the next period.

[0052] In this method, the fluctuation of the global return value G exceeds the standard, and the data of N consecutive periods are monitored to identify the impact of abnormal events across periods, thereby improving the reliability and practicality of the method.

[0053] a3: Production Line Mode Switching Monitoring: Monitor production line switching mode instructions in real time; if a high disturbance response mode or capacity sprint mode instruction is received, execute step S8.

[0054] In practice, the triggering timing for production line mode switching monitoring is set so that after receiving the mode switching instruction, a weight adjustment is triggered at the beginning of the next scheduling cycle to adapt to the production line operation requirements.

[0055] a4: Global negative return warning: Monitor the global comprehensive benefit value G in real time. If G < 0, then execute step S9.

[0056] In practice, the global negative return warning is triggered in real time without waiting for the cycle to end, serving as a fallback mechanism for extreme scenarios to ensure the safe operation of the production line.

[0057] S6: Based on the real-time value of the total disturbance duration percentage D, calculate the temporary value of the dynamic weight coefficient, and use the temporary value of the dynamic weight coefficient as the new dynamic weight coefficient to execute step S10. α temp =α·(1+k·D); β temp =β·(1-k·D); γ temp =γ·(1+2k·D); In the formula, α temp β temp and γ temp These are temporary values ​​for the dynamic weighting coefficients α, β, and γ; k is used to adjust sensitivity, k∈(0, 0.5], ensuring 1-k·D≥0.5 to eliminate weight risk, γ temp The coefficient is set to 2κ to enhance the perturbation suppression effect.

[0058] S7: Based on the proportion of total disturbance duration D retro The real-time value is used to calculate the temporary value of the dynamic weight coefficient, and the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient to execute step S10. α temp=α·(1+k·D retro ); β temp =β·(1-k·D retro ); γ temp =γ·(1+2k·D retro ); In the formula, α temp β temp and γ temp These are temporary values ​​for the dynamic weighting coefficients α, β, and γ; k is the sensitivity adjustment, k∈(0, 0.5], ensuring 1-k·D retro ≥0.5 eliminates the risk of negative weighting, γ temp The coefficient is set to 2κ to enhance the disturbance suppression effect and improve the system's anti-interference capability.

[0059] In this method, by analyzing the D value and D... retro The value is set to truncate the constraint, triggering lower bound protection only in extreme scenarios with multiple perturbations, while maintaining the validity of the original constraint in normal scenarios. This allows the D value and D to be combined... retro The value is controlled within the range of [0,1], while the value of k is controlled within the range of (0, 0.5], ensuring 1-k·D retro The relationship between ≥0.5 and 1-k·D≥0.5 holds true.

[0060] S8: Determine whether the received instruction to switch production line modes is a high-disturbance response mode; If so, then based on the preset high disturbance ratio benchmark value D high Calculate the temporary values ​​of the dynamic weighting coefficients: α temp =α·(1+k·D high ); β temp =β·(1-k·D high ); γ temp =γ·(1+2k·D high ); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. In the high-disturbance response mode, to enhance stability, in this embodiment, D high The default value is 0.3. Adjusting the sensitivity k in this mode applies to its upper limit range, k∈(0,0.5]. Otherwise, it indicates that the received production line switching instruction is in capacity sprint mode, and the sensitivity k is restricted to k∈(0,0.3], and a temporary value of the dynamic weighting coefficient is calculated: α temp =α·(1+k); β temp =β·(1-k); γ temp =γ·(1+2k); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient to execute step S10.

[0061] In the capacity sprint mode, efficiency is enhanced, and the sensitivity k is adjusted to a more stringent range k∈(0,0.3] to ensure the perturbation suppression weight γ. temp >0, to avoid over-emphasizing production and neglecting stability.

[0062] S9: Start emergency recovery mode to reset the weights; after exiting emergency recovery mode, use the temporary value of the dynamic weight coefficient obtained in emergency recovery mode as the new dynamic weight coefficient to execute step S10.

[0063] Based on the global comprehensive benefit value G constructed in this application, when G < 0, it indicates that the system is in a state of severe imbalance, and the disturbance loss has completely outweighed the production benefit. At this time, the system immediately starts the "emergency recovery mode" and executes a rapid recovery process that is distinct from the regular adjustment logic and has a clear objective.

[0064] The adjustment objectives of the emergency recovery mode are: to rapidly increase the weight of production continuity (α) and disturbance suppression (γ), force the system to prioritize the restoration of basic production order, and at the same time retain appropriate resource utilization attention (β) to ensure recovery efficiency and avoid the system from falling into a complete standstill.

[0065] The emergency adjustment is triggered as follows: when the coordinating agent detects that G < 0, this mechanism is immediately triggered to suspend the regular weight adjustment logic.

[0066] In step S9, the emergency recovery mode includes the following operations: c1: Calculate the negative depth corresponding to the global comprehensive return value G using the absolute value of G; Depth = |G|.

[0067] c2: Determine the emergency level based on the depth of negative values, and apply differentiated adjustment ranges according to the emergency level; Read the preset high-level judgment threshold D H And the preset low-level judgment threshold D L The specific threshold is obtained by summarizing historical data; in this embodiment, D H =0.5, D L =0.2; When Depth ≥ 0.5, the corresponding emergency level is: Severe Emergency; proceed to step c3. When 0.2 ≤ Depth < 0.5, the corresponding emergency level is: Moderate Emergency; proceed to step c4. When Depth < 0.2, the corresponding emergency level is: mild emergency; proceed to step c5.

[0068] c3: After calculating the temporary value of the dynamic weight coefficient using an aggressive adjustment, proceed to step c6; α temp =min(α×1.8,0.75); γ temp =min(γ×1.5,0.40); β temp =1-α temp -γ temp ; In severe emergencies, the adjustment aims to significantly enhance α and ensure α dominance.

[0069] c4: After calculating the temporary value of the dynamic weight coefficient using the standard adjustment range, proceed to step c6; α temp =min(α×1.5,0.70); γ temp =min(γ×1.3,0.35); β temp =1-α temp -γ temp ; In cases of moderate emergency, the standard adjustment range shall be used; c5: After calculating the temporary value of the dynamic weighting coefficient using a moderate adjustment range, proceed to step c6; α temp =min(α×1.3,0.65); γ temp =min(γ×1.2,0.30); β temp =1-α temp -γ temp ; In cases of mild emergency, a moderate adjustment range is adopted, approaching the high-disturbance response mode.

[0070] c6: Performs weight boundary protection and normalization processing on temporary values ​​of dynamic weight coefficients.

[0071] Step c6 specifically includes the following operations: c61: Read the system's preset β temp The lower bound β min The following threshold β min Ensure β temp The lowest value; If βtemp <β min Then β min Assigned to β temp 'Then execute step c62;' Otherwise, proceed to step c63; In this embodiment, β min The value is set to 0.05.

[0072] c62: Readjust weights and calculate α temp 'and γ temp ': α temp '=α temp ×((1-β temp ') / (α temp +γ temp )); γ temp '=γ temp ×((1-β temp ') / (α temp +γ temp )); and with α temp '、 β temp 'and γ temp 'Execute step c7 as the latest dynamic weight coefficient.'

[0073] In this embodiment, if β temp Adjusted, i.e., β temp =0.05; then α is redistributed proportionally. temp and γ temp To maintain a total weight of 1, the remaining weight is 0.95; α temp '=α temp ×(0.95 / (α temp +γ temp )); γ temp '=γ temp ×(0.95 / (α temp +γ temp )); c63: α is obtained after normalization. temp '、 β temp 'and γ temp ': α temp '=α temp / S; β temp '=β temp / S; γ temp '=γ temp / S; S=α temp +β temp +γ temp ; and with α temp '、 β temp 'and γ temp 'Execute steps c7~c8 as the latest dynamic weight coefficients.'

[0074] c7: System parameter linkage adjustment. When the weight adjustment takes effect, the following system-level parameters are forcibly overridden to form a combined recovery force. Forced overriding of system parameters includes the following operations: The threshold θ for dynamically pulling all caches pull Set to the preset safety stock threshold Q robust Ensure that materials are replenished in advance; The statistical period duration T total The value is updated to be adjusted to a temporary statistical period T. total =T total / 2, to accelerate the frequency of system status perception and response.

[0075] c8: Based on the latest dynamic weighting coefficients and dynamic pull trigger threshold θ pull and temporary statistical period T total ', Repeat steps S4~S5; During execution, the system operation status is monitored in real time. When the following conditions are met simultaneously, the emergency recovery mode is exited. Condition 1: Two consecutive temporary statistical periods T total All satisfy: G≥0.3; Condition 2: Two consecutive temporary statistical periods T total 'All satisfy: (T) fault +T congest ) / ΣT disturb <0.5; In the formula, T fault T represents the cumulative duration of equipment failure. congest Cumulative congestion time for logistics intelligent agents; ΣT disturb This indicates the cumulative duration of all disturbance events in the system during this statistical period.

[0076] Before step S10 is executed, the following operations also need to be performed: d1: α is obtained after normalization. new , β new and γ new ; α new =α temp / S; βnew =β temp / S; γ new =γ temp / S; S=α temp +β temp +γ temp ; d2: Read the preset amplitude change threshold δ max If any of the following conditions are met, then proceed to step d3; otherwise, proceed with step α. new , β new and γ new As the latest dynamic weighting coefficient, step S9 is executed; in this embodiment, δ max =0.15; Amplitude judgment condition 1: |α new -α |>δ max ; Amplitude judgment condition 2: |β new -β |>δ max ; Amplitude judgment condition 3: |γ new -γ |>δ max ; d3: Perform anti-oscillation processing on each dynamic weight coefficient to avoid drastic fluctuations in the scheduling strategy, which could lead to excessive overall energy consumption; Specifically, it includes the following steps; When α new -α>δ max At that time, α new '=α new +δ max ; When α new –α<-δ max At that time, α new '=α new -δ max ; When |α new –α|≤δ max At that time, α new '=α new ; When β new -β>δ max At that time, β new '=β new +δ max ; When β new –β<-δ max At that time, β new '=β new -δ max ; When |β new –β|≤δ max At that time, β new '=β new ; When γ new -γ>δ max At that time, γ new '=γ new +δ max ; When γ new –γ<-δ max At that time, γ new '=γ new -δ max ; When |γ new –γ|≤δ max At that time, γ new '=γ new ; d4: Weight normalization processing: α temp =α new ' / (α new '+β new '+γ new '); β temp =β new ' / (α new '+β new '+γ new '); γ temp =γ new ' / (α new '+β new '+γ new '); d5: with α temp β temp and γ temp Step S9 is executed as the latest dynamic weight coefficient.

[0077] In this application, to ensure that the system maintains a minimum level of efficiency even in emergency situations, the dynamic weighting coefficient β for the resource utilization rate is a temporary value. temp A lower bound protection mechanism is implemented. In this lower bound protection mechanism, β is... temp Set the minimum weight lower bound value β min ,β min >0; Each calculation yields a β temp After that, β will be temp With β min Compare; If β temp <β min Then βmin The value is assigned to β temp In this embodiment, β min The value is set to 0.05.

[0078] S10: Based on the latest dynamic weight coefficients, repeat steps S4 to S5.

[0079] The coordinating agent writes the updated weight set {α,β,γ} into the system's core parameter table, which takes effect at the start of the next scheduling cycle, ensuring the temporal consistency of scheduling decisions and system stability.

[0080] The following section uses a semiconductor packaging and testing cleanroom production line as an example to describe the specific implementation process, parameter configuration, and application effects of the cleanroom logistics multi-agent collaborative scheduling method described in this invention. The specific simulation environment is implemented using existing software such as Plant Simulation.

[0081] like Figure 4 As shown, the production line in this embodiment includes two die bonders and two wire bonders: die bonder A1, die bonder A2, wire bonder B1, and wire bonder B2; four workstation buffer areas: buffer areas C1-C4, corresponding to A1, A2, B1, and B2 respectively; one common temporary storage area: buffer area C5; and three automated guided vehicles (AGVs): logistics handling intelligent agents AGV1-AGV3, and one production line scheduling server deployed with a coordination intelligent agent. The overall workstation and intelligent agent layout adopts a conventional cleanroom layout with upstream and downstream process partitioning and the logistics channel (marked as AVG path in the figure) centered.

[0082] The intelligent entities for production equipment are embedded in the PLC controllers of A1, A2, B1, and B2, collecting real-time equipment operating status data at a frequency of 1 second per instance, with an anomaly warning broadcast delay not exceeding 100ms. They are responsible for reporting equipment start / stop, process completion, and fault information. The intelligent entities for the buffer area are deployed in the material RFID reader terminals of C1-C5, with C1-C4 having a rated capacity of 10 baskets of material and C5 having a rated capacity of 30 baskets. They are configured with 100% accuracy for material traceability code recognition, a locking response time of ≤500ms for material mismatch, and real-time reporting of buffer area material inventory, overflow, and shortage status. The intelligent entities for logistics handling are integrated into the onboard controllers of AGV1-AGV3, configured with a task priority rule of "emergency replenishment > push-type transfer tasks > regular transfer tasks." The communication frequency for local path game between AGV devices is 500ms per instance to avoid congestion in the logistics channel. The coordinating agent is deployed on the production line scheduling server, with a CPU frequency of no less than 2.8GHz and memory of ≥16GB. It is responsible for global data aggregation, global revenue calculation, weight adjustment, and scheduling instruction issuance. The latency of a single round of global decision calculation does not exceed 1 second.

[0083] In this method, all weight coefficients and threshold parameters are dynamically adjustable configuration values, not fixed values. They can be flexibly modified according to the reliability of production line equipment, logistics pressure, and the frequency of disturbances. In this embodiment, the core parameter configurations used are as follows: Statistical period: T total =30min, ε=1×10 -6 ; Global reward weights (satisfying α+β+γ=1, non-negativity constraint), initial weight settings: α=0.4, β=0.3, γ=0.3; Resource utilization weight (satisfying ω) agv +ω buf =1 (non-negative constraint), initial weight setting: AGV utilization weight ω agv =0.5, Cache utilization weight ω buf =0.5; Disturbance loss weights (satisfying λ1+λ2+λ3=1, non-negative constraint), initial weight settings: equipment failure weight λ1=0.4, AGV congestion weight λ2=0.4, material mismatch weight λ3=0.2; Buffer dynamic threshold parameter: G high =0.7, G low =0.4; Q stable =3 baskets, Q robust =5 baskets; Weighting parameters: perturbation adjustment sensitivity k=0.3, threshold for magnitude change in word weight adjustment δ max =0.15; Disturbance duration threshold η1=10%, global return fluctuation threshold η2=0.2; Emergency adjustment parameters: severe emergency (Depth≥0.5), moderate emergency (0.2≤Depth<0.5), mild emergency (Depth<0.2).

[0084] After the system starts, the coordinating agent enters the initialization running state, and the agents corresponding to each production equipment, buffer area, and logistics transportation execution unit go online simultaneously. The system loads production line layout information, equipment capacity parameters, buffer capacity parameters, and logistics scheduling rules, establishes communication relationships between agents, and completes initial state synchronization to ensure that the scheduling process is executed based on unified and real-time system status data.

[0085] After initialization, the system enters the normal closed-loop scheduling operation phase; in this phase, material distribution and production line supply are completed in coordination through push and pull scheduling mechanisms. When the upstream production unit completes its processing task, it triggers push scheduling behavior, and the upstream agent publishes a material ready event with a material traceability identifier.

[0086] Example of global benefit G calculation: Select the normal operating cycle of the production line, and the statistical period is T. total =30min, the specific calculation process is as follows: Production continuity benefit P: With all 4 machines operating stably and each machine having an effective operating time of 27 minutes, P = 0.9 is calculated. Resource utilization benefit R: AGV average utilization rate is 0.82, buffer area average utilization rate is 0.78, calculated as R = 0.5 × 0.82 + 0.5 × 0.78 = 0.8; Disturbance loss L: With only a slight AGV congestion of 1.2 minutes and no equipment failure or material mismatch, the calculated L = 0.016 ≈ 0.02; Global return G: Substituting into the formula for calculating G, we get G = 0.4 × 0.9 + 0.3 × 0.8 - 0.3 × 0.02 = 0.594 ≈ 0.6; The result of this calculation is G low <G<G high The system triggers intermediate-state dynamic material replenishment logic, balancing production efficiency and operational stability.

[0087] Example of a push logistics process: Production equipment A1 completes die bonding of 10 baskets of semiconductor chips. The A1 agent receives a "process completed" signal and simultaneously detects that the current inventory in the downstream buffer C1 is 2 baskets, which is lower than Q. stable =3 baskets.

[0088] The A1 agent generates a unique traceability code for this batch of materials (code: GJ20240520A101), binds the material processing information to the traceability code, and sends a "material ready" event to the coordinating agent.

[0089] Based on the real-time capacity status of downstream workstations and buffer zones, the coordinating agent determines the flow path of materials and generates corresponding logistics transportation tasks.

[0090] The coordinating agent combines the current global benefit G=0.75, because G high =0.7, therefore G≥G high The decision is made that the material flows directly into C1; a "routine transfer task" is issued to AGV1, instructing it to pick up the material from A1 and transport it to C1.

[0091] The logistics and transportation execution unit accepts scheduling tasks according to task priority, completes loading, transportation, and unloading operations, and feeds back the execution results after task completion to update the operating status of relevant intelligent agents, thus forming a continuously iterative closed-loop scheduling process. During normal closed-loop scheduling operation, the system continuously performs disturbance detection.

[0092] After receiving the task, AGV1 reports its own status (idle, location near A1), performs material picking, transportation, and unloading actions, and sends a "task completed" signal to the A1 agent, C1 agent, and coordinating agent after completion; the C1 agent updates the inventory to 12 baskets, triggering an overflow warning, and subsequent materials will be dispatched to the public temporary storage area C5.

[0093] Example of a pull logistics process: When downstream workstations consume materials during production, their buffer agents continuously monitor inventory levels. When inventory falls below a dynamically set pull threshold, a pull scheduling behavior is triggered, and the coordinating agent or buffer agent locks available material resources and generates a replenishment transportation task.

[0094] Production equipment B1 continuously consumes materials from the C2 buffer, and the C2 agent monitors the inventory in real time as it drops to 4 baskets. At this point, the global benefit calculated by the coordinating agent is G = 0.6 (at which point G...). low =0.4 and G high =between 0.7), calculated according to the dynamic threshold formula: θ pull =round(3+((5-3)×(0.6-0.7) / (0.4-0.7)))=round(3.67)=4(basket); C2 inventory has fallen to the threshold θ pull =4 baskets, C2 agent sends a "pull replenishment request" to the coordinating agent.

[0095] The coordinating agent queries the inventory in the public temporary storage area C5 and finds matching materials (traceability code: GJ20240520A201). It then issues an "emergency replenishment task" (highest priority) to AGV2, instructing it to take 5 baskets of materials from C5 and transport them to C2.

[0096] During the execution of a task, AGV2 engages in local path game with AGV3, which is performing a regular task. Based on real-time electronic map congestion information, it autonomously selects the optimal path to avoid potential congestion points, reducing task completion time by 15% compared to traditional scheduling.

[0097] The intelligent agent embeds a material traceability code binding module, linking traceability information with scheduling decisions in real time. The agent's sensing data is synchronized every 100ms, triggering trajectory locking in case of anomalies. When any intelligent agent detects an abnormal operating state, it generates a disturbance warning event and reports it to the coordinating agent. These abnormal operating states include, but are not limited to, equipment failure, logistics congestion, or material mismatch. When a disturbance event is detected, the system enters emergency scheduling mode. The coordinating agent suspends the affected logistics links, dynamically adjusts scheduling weight parameters based on current available resources, and reconstructs the scheduling scheme based on remaining resources to generate emergency scheduling tasks and organize their execution. During emergency scheduling execution, the system periodically checks whether the operating state meets the conditions for returning to normal scheduling. When the anomaly is detected and eliminated and the system's operating state is stable, the system exits the emergency scheduling mode and re-enters the normal closed-loop scheduling operation phase.

[0098] This embodiment simulates a disturbance scenario involving a fault in wire bonding machine B2 to verify the responsiveness of the scheduling method.

[0099] When production equipment B2 experiences a sudden mechanical failure, its intelligent agent detects the "equipment abnormality" signal and immediately broadcasts the warning event to the coordinating intelligent agent and upstream and downstream intelligent agents. At the same time, it automatically suspends its own material pushing process.

[0100] The coordinating agent counts the duration of disturbances within 1 T. total =The cumulative duration T of B2 fault within a 30-minute period fault =5min, the perturbation ratio D = 5 / 30≈16.7%, because D exceeds η1=10%, triggering adaptive weight adjustment.

[0101] The system detected an anomaly in the monitoring of the total disturbance duration exceeding the standard. The total disturbance duration percentage was calculated to be D=16.7%. Substituting this into the formula for temporary weights, the following calculations were performed: α temp =0.4×(1+0.3×0.167)=0.42; β temp =0.3×(1-0.3×0.167)=0.285; γ temp =0.3×(1+2×0.3×0.167)=0.33; Verification weight change magnitude: |α temp -α|=0.02<δ max =0.15, which meets the constraint conditions; after normalization, the new weight is α. new =0.423, β new =0.287, γ new =0.29, effective in the next scheduling cycle.

[0102] The coordinating agent suspends all logistics tasks destined for B2 and relocates the materials originally planned to be transported to B2 (traceability code: GJ20240520A102) to the public temporary storage area C5.

[0103] When the agent in buffer zone C4 (corresponding to B2) detects that no material has been consumed, it automatically locks the inventory; the coordinating agent switches the regular task of AGV3 to "B1 emergency replenishment task", prioritizing replenishment of B1 to ensure the production continuity of B1.

[0104] Assuming the fault repair of B2 takes 40 minutes, its agent sends a "device ready" signal to the system.

[0105] The coordinating agent initiates the recovery process, adjusting the weights to a "mild emergency" level. At this point, Depth = |G| = 0.18 < 0.2, and α is calculated. temp =0.4×1.3=0.52, γ temp =0.3 × 1.2 = 0.36, β temp =0.12, because β temp ≥0.05, no lower bound protection required.

[0106] The coordinating agent schedules AGV2 to transport materials from C5 to C4, prioritizing the resumption of production at B2; simultaneously, it adjusts the global task allocation to ensure the system's overall benefit is within 2T. total It rebounded to G=0.72 within the cycle, returning to normal operating levels.

[0107] Once the semiconductor packaging and testing order for this batch is completed, the system enters standby mode, coordinates the intelligent agents to summarize the data of the entire process, and generates a production report, which includes: Material traceability chain: traceability codes, transportation routes, processing equipment and timestamps of all materials; AGV running trajectory: task execution order, travel path and idle / working time ratio of AGV1-AGV3; Intelligent agent decision log: state changes, disturbance response actions and weight adjustment records of each intelligent agent.

[0108] The above data is persistently stored in the production line database, enabling full-process auditability and traceability.

[0109] The following is a description of this embodiment and the comparative example: 1. Simulation Model Declaration: Based on the simulation software, a simulation model of the actual semiconductor packaging and testing cleanroom production line is constructed. The system deployment, scheduling process, parameter configuration and disturbance scenarios described in this embodiment are all fully reproduced and tested in this simulation model. 2. Testing cycle and verification method: The unified testing cycle is to complete 100 batches of semiconductor packaging testing orders (each batch contains 50 baskets of materials, corresponding to 8 hours of full-load operation of the production line). This embodiment verifies the technical effect by comparing the running data of the traditional centralized scheduling method and the method of the present invention. 3. Comparative Example: The traditional centralized multi-agent cooperative scheduling method is adopted, with the following specific characteristics: ①Task allocation: A static task allocation mode is adopted, referring to the logic of the cultural asexual reproduction genetic algorithm. The task allocation scheme is preset in advance and is not dynamically adjusted according to global benefits, equipment and cache status during operation. ② Path planning: The AGV path planning adopts the traditional A* global path planning algorithm in the paper, without local path game mechanism.

[0110] Based on the above simulation model, unified testing cycle, and traditional method settings, two methods were used to conduct tests, and the core comparative verification results are as follows: 1. Dynamic disturbance response time: The average response time of traditional methods is 5 minutes, while the method of this invention shortens it to 45 seconds, improving the response efficiency by 85%; 2. Equipment downtime: Under fault disturbance, the overall downtime of the production line using traditional methods is 15 minutes, while the method of this invention is only 3 minutes, a reduction of 80%; 3. Resource utilization: The average working time of AGVs increased from 60% in the traditional method to 82%, and the average occupancy time of the buffer area was optimized from 55% to 70%. 4. Global return stability: The fluctuation range of the global return G value of the method of the present invention is controlled within 0.1, while the fluctuation range of the traditional method is 0.35, which improves the stability by 71.4%.

Claims

1. A method for multi-agent collaborative scheduling of logistics in cleanrooms, characterized in that, It includes the following steps: S1: System Modeling Construct a scheduling system; the scheduling system includes: a coordinating agent and an executing agent; The execution agents include: production equipment agents, buffer agents, and logistics handling agents; The buffer agent is configured for the buffer area, which is used to store materials and includes: a workstation buffer area and a common buffer area; the workstation buffer area is set up separately for each processing workstation to temporarily store materials for each processing workstation; the common buffer area is a shared buffer area for all workstations, which temporarily stores materials when there is no space in the workstation buffer area. The coordinating agent, acting as the core of global coordination, is responsible for issuing scheduling instructions and agent parameters to the three types of executing agents. The agent parameters include: dynamic weight coefficients and a dynamic trigger threshold θ. pull ; The three types of execution agents upload their own status data and disturbance warning information to the coordinating agent in real time, and at the same time receive the scheduling instructions from the coordinating agent to complete autonomous decision-making and collaborative execution. The scheduling behaviors performed by the buffer agent include push scheduling and pull scheduling. When the production unit completes the processing task, the push scheduling behavior is triggered, which notifies the coordinating agent that the materials processed at this workstation have entered the logistics mobilization ready state, allowing them to be transported to the downstream workstation. When the material in the buffer area corresponding to the production unit is lower than the dynamic pull trigger threshold θ pull When the pull scheduling behavior is triggered, the coordinating agent is notified that materials need to be replenished at this workstation. S2: Construct a global profit function model; G = α·P + β·R - γ·L; Where G is the global comprehensive benefit value, α is the dynamic weight coefficient of production continuity, β is the dynamic weight coefficient of resource utilization, and γ is the dynamic weight coefficient of disturbance loss, satisfying α≥0, β≥0, γ≥0, and α+β+γ=1. P represents the production continuity benefit, which is calculated from the cumulative time of equipment failures and downtime and the depletion of buffer materials. R represents the resource utilization benefit, which is the percentage of the average working time of the logistics handling agent. AGV and the average time occupied by the cache T buffer Statistics show that; L represents the disturbance loss, which is calculated using real-time monitoring data from the production line. S3: Based on the aforementioned scheduling system and the global comprehensive benefit value G, monitor the logistics scheduling process of the production line; S4: At the beginning of each statistical period, the coordinating agent summarizes the global information, reads the latest dynamic weight coefficients α, β, γ, and calculates the global comprehensive benefit value G. When the coordinating agent controls the material movement behavior, it sets the dynamic pull trigger threshold θ for each buffer agent based on the real-time global comprehensive benefit value G. pull ; S5: Construct an adaptive weight adjustment mechanism, in which the coordinating agent adjusts the dynamic weight coefficients in real time according to the actual production situation during the scheduling process; the adaptive weight adjustment mechanism specifically includes the following operations: a1: Calculate the total disturbance duration percentage D in real time and compare D with the preset disturbance duration percentage threshold η1; D=ΣT disturb / T total , Among them, T total ΣT is the preset statistical period duration. disturb This indicates the cumulative duration of all disturbance events in the system within the current statistical period; When D>η1, proceed to step S6; a2: Calculate the fluctuation range FR of the global comprehensive return value G in real time over N consecutive statistical periods, and compare FR with the preset fluctuation range threshold η2; FR=G max -G min ; In the formula, G max G is the maximum value of G over N consecutive statistical periods. min The minimum G value in N consecutive statistical periods; When FR>η2, the system automatically backtracks and analyzes all disturbance events that caused the fluctuation of the G value, and calculates the cumulative duration percentage D of all disturbance events. retro Proceed to step S7; a3: Real-time monitoring of production line switching mode commands; If a high-disturbance response mode or capacity sprint mode instruction is received, then step S8 is executed; a4: Real-time monitoring of the overall global benefit value G; If G < 0, then proceed to step S9; S6: Calculate the temporary value of the dynamic weighting coefficient based on the real-time value of the proportion of total disturbance duration D: a temp =α·(1+k·D); b temp =β·(1-k·D); c temp =γ·(1+2k·D); In the formula, α temp For the temporary values ​​of the dynamic weighting coefficients for continuous production, β temp γ is a temporary value for the dynamic weighting coefficient of resource utilization. temp This is a temporary value for the dynamic weighting coefficient of the disturbance loss; k is the adjustment sensitivity, k∈(0, 0.5]; Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. S7: Based on the proportion of total disturbance duration D retro The real-time value is used to calculate the temporary value of the dynamic weighting coefficient. a temp =α·(1+k·D retro ); b temp =β·(1-k·D retro ); c temp =γ·(1+2k·D retro ); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. S8: Determine whether the received instruction to switch production line modes is a high-disturbance response mode; If so, then based on the preset high disturbance ratio benchmark value D high Calculate the temporary values ​​of the dynamic weighting coefficients: a temp =α·(1+k·D high ); b temp =β·(1-k·D high ); c temp =γ·(1+2k·D high ); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. Otherwise, it indicates that the received production line switching instruction is in capacity sprint mode, and the adjustment sensitivity k is restricted to k∈(0,0.3], and a temporary value of the dynamic weighting coefficient is calculated: α temp =α·(1+k)； β temp =β·(1-k)； c temp =γ·(1+2k); Then, the temporary value of the dynamic weight coefficient is used as the new dynamic weight coefficient, and step S10 is executed. S9: Start the emergency recovery mode to reset the weight coefficients; after exiting the emergency recovery mode, use the temporary value of the dynamic weight coefficients obtained in the emergency recovery mode as the new dynamic weight coefficients to execute step S10; S10: Based on the latest dynamic weight coefficients, repeat steps S4 to S5.

2. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: In step S4, the dynamic pull trigger threshold θ pull The setup method is as follows: b1: Read the pre-set global high-return threshold G in the system. high Global return low threshold G low Buffer stable inventory threshold Q stable and safety stock threshold Q robust ; The global return high threshold G high The threshold for determining the efficient operation of the production line; The global return low threshold G low Production line operation early warning threshold; The stable inventory threshold Q of the buffer area stable To maintain the minimum baseline inventory for regular continuous production of downstream equipment in the workstation buffer area; The safety stock threshold Q robust The redundancy safety stock threshold for the buffer to cope with short-term sudden disturbances; b2: Combine the value of the global comprehensive return G with G high and G low A comparison was made to confirm the dynamic pull trigger threshold θ. pull The possible values ​​of ; When G≥G high When, θ pull =round(Q stable ); When G <G low When, θ pull =round(Q robust ); When G low ≤G <G high When, θ pull =round(Q stable +(( Q robust- Q stable )*(GG high ) / (G low -G high ))); In the formula, round() is the rounding function.

3. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: The method for calculating the production continuity benefits is as follows: P=(T total -T stop -T lack ) / (T total +ε); In the formula, T total T is the preset statistical period duration. stop T is the cumulative duration of equipment failures and downtime collected from the PLCs of various production equipment. lack ε is the cumulative duration of material depletion as monitored by sensors in the downstream buffer area; ε is a minimal constant. The method for calculating the resource utilization rate benefit is as follows: R=ω agv ×U AGV +oh buf ×T buffer ; Where, ω agv ω represents the weighting coefficient of logistics handling equipment in resource utilization. buf This represents the weighting coefficient of the cache in resource utilization. U AGV For n agv The average percentage of working time for each logistics handling intelligent agent; T buffer For n buffer The average time spent in each cache area; ； T i agv,busy The working time of logistics handling intelligent agent numbered i within the statistical period; ； T i buffer,occupied This represents the time that the buffer with ID i is occupied within the statistical period; The method for calculating the disturbance loss is as follows: L=(λ1·T fault +λ2·T congest +λ3·T error ) / T total ; In the formula, λ1 is the disturbance weight coefficient corresponding to the cumulative duration of equipment failure, λ2 is the disturbance weight coefficient corresponding to the cumulative duration of congestion in the logistics intelligence agent, and λ3 is the disturbance weight coefficient corresponding to the cumulative duration of material mismatch; λ1≥0, λ2≥0, λ3≥0, and λ1+λ2+λ3=1; T fault T represents the cumulative duration of equipment failure. congest T represents the cumulative congestion time for logistics intelligent systems. error This represents the cumulative duration of material mismatch.

4. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: In step S5, when calculating the percentage of each total disturbance duration D, D is restricted to the range of [0,1]. If D>1, then D is assigned the value 1. Calculate the cumulative duration percentage D of all disturbance events over N periods. retro For each total disturbance duration D within each period, the percentage of D is calculated separately, and each D is restricted to the range [0,1]. If D>1, then D is assigned the value 1. Finally, the N D values ​​are added together to obtain D. retro .

5. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: It also includes a lower bound protection mechanism; In the lower bound protection mechanism, the temporary value β is the dynamic weight coefficient of the resource utilization rate. temp Set the minimum weight lower bound value β min ,β min >0; Each calculation yields a β temp After that, β will be temp With β min Compare; If β temp <β min Then β min The value is assigned to β temp .

6. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: In step S9, the emergency recovery mode includes the following operations: c1: Calculate the negative depth Depth corresponding to the global comprehensive benefit value G; Depth = |G|; c2: Determine the emergency level based on the negative depth, and apply differentiated adjustment ranges based on the emergency level; Depth ≥ D H At this time, the corresponding emergency level is: Critical Emergency; proceed to step c3; D L ≤ Depth < D H At this time, the corresponding emergency level is: moderate emergency; proceed to step c4; Depth < D L At this time, the corresponding emergency level is: mild emergency; proceed to step c5; In the formula, D H D is the preset high-level judgment threshold. L The preset low-level judgment threshold; c3: After calculating the temporary value of the dynamic weight coefficient using an aggressive adjustment, proceed to step c6; a temp =min(α×1.8,0.75); c temp =min(γ×1.5,0.40); b temp =1-a temp -c temp ? c4: After calculating the temporary value of the dynamic weight coefficient using the standard adjustment range, proceed to step c6; a temp =min(α×1.5,0.70); c temp =min(γ×1.3,0.35); b temp =1-a temp -c temp ? c5: After calculating the temporary value of the dynamic weighting coefficient using a moderate adjustment range, proceed to step c6; a temp =min(α×1.3,0.65); c temp =min(γ×1.2,0.30); b temp =1-a temp -c temp ? c6: Performs weight boundary protection and normalization processing for dynamic weight coefficients; c7: Force overwrite system parameters, specifically including the following operations: The threshold θ for dynamically pulling all caches pull Set to the preset safety stock threshold Q robust ; The statistical period duration T total The value is updated to be adjusted to a temporary statistical period T. total =T total / 2; c8: Based on the latest dynamic weighting coefficients and dynamic pull trigger threshold θ pull and temporary statistical period T total ', Repeat steps S4~S5; The system will monitor its operation in real time and exit the emergency recovery mode when the following conditions are met simultaneously. Condition 1: Two consecutive temporary statistical periods T total All satisfy: G≥0.3; Condition 2: Two consecutive temporary statistical periods T total 'All satisfy: (T) fault +T congest ) / ΣT disturb <0.5; In the formula, T fault T represents the cumulative duration of equipment failure. congest Cumulative congestion time for logistics intelligent agents; ΣT disturb This indicates the cumulative duration of all disturbance events in the system during this statistical period.

7. The cleanroom logistics multi-agent collaborative scheduling method according to claim 6, characterized in that: Step c6 specifically includes the following operations: c61: Read the preset β temp The lower bound β min , If β temp <β min Then β min Assigned to β temp 'Then execute step c62;' Otherwise, proceed to step c63; c62: Readjust weights and calculate α temp 'and γ temp ': a temp '=a temp ×((1-β temp ') / (a temp +g temp )); c temp '=c temp ×((1-β temp ') / (a temp +g temp )); and with α temp '、 β temp 'and γ temp 'Execute subsequent steps as the latest dynamic weighting coefficients;' c63: α is obtained after normalization. temp '、 β temp 'and γ temp ': α temp ’=α temp / S; β temp ’=β temp / S; c temp '=c temp / S? S=a temp +b temp +g temp ; and with α temp '、 β temp 'and γ temp 'Execute subsequent steps as the latest dynamic weight coefficients.' 8. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: Before step S10 is executed, the following operations also need to be performed: d1: α is obtained after normalization. new , β new and γ new ; α new =α temp / S； β new =β temp / S； c new =c temp / S; S=a temp +b temp +g temp ; d2: Read the preset amplitude change threshold δ max If any of the following conditions are met, then proceed to step d3; otherwise, proceed with step α. new , β new and γ new Step S9 is executed as the latest dynamic weighting coefficient; Amplitude judgment condition 1: |α new -α |>δ max ; Amplitude judgment condition 2: |β new -β |>δ max ; Amplitude judgment condition 3: |γ new -γ |>δ max ; d3: Perform anti-oscillation processing on each of the aforementioned dynamic weight coefficients, specifically including the following steps; When α new -α > δ max then, α new ’ = α new + δ max ; When α new –α < -δ max then, α new ’ = α new - δ max ; When |α new – α| ≤ δ max then, α new ’ = α new ; When β new -β > δ max then, β new ’ = β new + δ max ; When β new –β < -δ max then, β new ’ = β new - δ max ; When |β new – β| ≤ δ max then, β new ’ = β new ; When γ new -γ > δ max then, γ new ’ = γ new + δ max ; When γ new –γ < -δ max then, γ new ’ = γ new -δ max ; When |γ new – γ| ≤ δ max then, γ new ’ = γ new ; d4: Weight normalization processing: a norm =a new ' / (a new '+b new '+g new '); b norm =b new ' / (a new '+b new '+g new '); c norm =c new ' / (a new '+b new '+g new '); d5: with α norm β norm and γ norm Step S9 is executed as the latest dynamic weight coefficient.

9. The cleanroom logistics multi-agent collaborative scheduling method according to claim 1, characterized in that: In step S3, when the system runs for the first time, it reads the initial value of the preset dynamic weight coefficient and performs subsequent calculations.

Images