Dynamic scheduling method and system for AGV carrying path in chip production

Abstract

This invention discloses a dynamic scheduling method and system for AGV transport paths in chip manufacturing, relating to the field of industrial automation technology. The method includes: acquiring multi-source scene data of the target scenario in real time through a central scheduling terminal and initializing corresponding scheduling baseline information; monitoring and judging scheduling trigger events in real time based on a preset scheduling trigger event library; if the real-time trigger judgment passes, making AGV transport path decisions based on multi-objective optimization based on the scheduling trigger event monitoring results, the scheduling baseline information, and a pre-built disturbance proxy evaluation model to obtain a dynamic scheduling scheme; comparing the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario, and selecting the better one for execution. This invention effectively improves the scheduling reliability of AGV transport paths in chip manufacturing.

Description

Technical Field

[0001] This invention relates to the field of industrial automation technology, specifically to a dynamic scheduling method and system for AGV transport paths in chip manufacturing. Background Technology

[0002] With the rapid iteration of intelligent manufacturing technologies and the increasing refinement of chip manufacturing processes, automated material handling has become a core support for chip factories to improve production efficiency and ensure production continuity. AGVs (Automated Guided Vehicles), as key handling equipment, have seen widespread application of path scheduling technology. Existing technologies mostly rely on preset scenario parameters to achieve static planning or simple dynamic adjustment of AGV paths, which initially meets the scheduling needs of basic material handling in chip production.

[0003] However, in the face of frequent task insertions, equipment status changes, sudden congestion or path blockages in the chip manufacturing environment, traditional scheduling methods are slow to respond and difficult to coordinate and optimize globally and in a forward-looking manner. This can easily lead to decreased handling efficiency and longer waiting time, resulting in insufficient scheduling reliability of AGV handling paths in chip manufacturing, which directly affects the continuity and capacity of chip production lines. Summary of the Invention

[0004] This invention provides a dynamic scheduling method and system for AGV transport paths in chip manufacturing, aiming to solve the technical problem of insufficient scheduling reliability of AGV transport paths in existing chip manufacturing.

[0005] In view of the above problems, the present invention provides a method and system for dynamic scheduling of AGV transport paths in chip manufacturing.

[0006] In a first aspect, the present invention provides a dynamic scheduling method for AGV transport paths in chip manufacturing, comprising: The central dispatch terminal acquires multi-source scene data of the target scene in real time and initializes the corresponding dispatch baseline information. Based on the preset scheduling trigger event library, perform scheduling trigger event monitoring and real-time trigger judgment; If the real-time trigger judgment passes, then based on the scheduling trigger event monitoring results, the scheduling baseline information and the pre-built disturbance proxy evaluation model, a multi-objective optimization-based AGV transport path decision is made to obtain a dynamic scheduling scheme. Compare the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario, and select the better one to execute.

[0007] Secondly, the present invention provides a dynamic scheduling system for AGV transport paths in chip manufacturing, including: The scene data initialization module is used to acquire multi-source scene data of the target scene in real time through the central dispatch terminal, and initialize the corresponding dispatch background information. The event monitoring module is used to monitor and determine real-time trigger events based on a preset event database. The dynamic path optimization decision module is used to make AGV transport path decisions based on multi-objective optimization, and obtain a dynamic scheduling scheme, based on the scheduling trigger event monitoring results, the scheduling baseline information and the pre-built disturbance proxy evaluation model, if the real-time trigger judgment is passed. The scheduling scheme optimization and execution module is used to compare the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario and select the better one to execute.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a dynamic scheduling method and system for AGV transport paths in chip manufacturing. By acquiring multi-source scene data in real time and initializing scheduling baseline information, it solves the problems of lagging data perception and inaccurate baseline constraint construction in existing technologies, providing a precise data foundation for subsequent scheduling decisions. Based on a preset event library, it conducts real-time monitoring and trigger judgment, realizing rapid identification and response to various disturbance events in chip manufacturing, breaking the limitation of untimely trigger response in existing technologies. Combining trigger results, baseline information, and disturbance proxy evaluation models for multi-objective optimization decisions, it effectively takes into account multiple requirements such as transportation efficiency, airflow disturbance control, and task priority, making up for the shortcomings of single optimization dimensions in existing technologies. By comparing and selecting the best solution, it further ensures the adaptability and reliability of the scheduling scheme, improving the overall reliability of AGV transport scheduling in chip manufacturing scenarios and adapting to the refined and dynamic needs of chip manufacturing. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating the dynamic scheduling method for AGV transport paths in chip manufacturing provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a dynamic scheduling system for AGV transport paths in chip manufacturing provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Scene data initialization module 11, trigger event monitoring module 12, dynamic path optimization decision module 13, and scheduling scheme optimization execution module 14. Detailed Implementation

[0011] This invention provides a dynamic scheduling method and system for AGV transport paths in chip manufacturing, which addresses the technical problem of insufficient scheduling reliability of AGV transport paths in existing chip manufacturing processes.

[0012] Example 1, as Figure 1 As shown, this invention provides a dynamic scheduling method for AGV transport paths in chip manufacturing, the method comprising: S100: Acquires multi-source scene data of the target scene in real time through the central dispatch terminal, and initializes the corresponding dispatch baseline information.

[0013] In this embodiment of the invention, multi-source scene data of the target scenario is acquired in real time through a central scheduling terminal, and corresponding scheduling baseline information is initialized. Chip manufacturing scenarios have extremely high requirements for the scheduling accuracy, cleanliness, and process adaptability of AGV transport paths. Existing technologies lack real-time integration of multi-source scene data and accurate baseline information construction, leading to a disconnect between scheduling schemes and actual production needs. Therefore, it is necessary to achieve comprehensive data perception and constraint rule establishment of the chip manufacturing scenario through multi-source data acquisition and baseline information initialization, providing accurate data support and a logical foundation for subsequent dynamic scheduling decisions.

[0014] Step S100 in the method provided in this embodiment of the invention includes: This includes acquiring multi-source scene data of the target scene in real time through the central dispatch terminal, including: Load a digital map of the target scene, and determine the type of each area and the airflow disturbance coefficient based on the digital map; Obtain the current production plan and real-time scheduling scheme for the target scenario from the Manufacturing Execution System; Read the safe operation parameters of the AGV from the vehicle specification library of the target scenario.

[0015] First, load a digital map of the target scene, and determine the type of each area and the airflow disturbance coefficient based on the digital map.

[0016] This includes loading a digital map of the target scene and determining the type of each region and the airflow disturbance coefficient based on the digital map, including: A digital model of the AGV is established, and simulations are performed in each region based on the default operating parameters of each region. Based on the simulation results and the preset disturbance trigger threshold, the first disturbance range is determined. Based on the digital model and the default operating parameters of each region, an ideal blank space is established to perform simulation, and the reference disturbance range is determined based on the simulation results and the preset disturbance trigger threshold. Historical disturbance data for each region is obtained, and the second disturbance range is determined by weighting the type probability of the transported object type corresponding to the historical disturbance data and combining it with a normalized weighting method. The inherent airflow disturbance coefficient is determined based on the ratio of the weighted fusion range of the first disturbance range and the second disturbance range to the reference disturbance range. The inherent airflow disturbance coefficient is then corrected based on the preset area cleanliness level of each area to obtain the airflow disturbance coefficient.

[0017] First, a digital model of the AGV is established. Simulations are performed across each region based on its default operating parameters. The first disturbance range is determined based on the simulation results and a preset disturbance trigger threshold. The digital model of the AGV includes its physical properties and operating parameters, such as load capacity, dimensions, and maximum operating speed. Operating parameters include maximum operating speed and turning radius. Simulations are performed across each region based on its default operating parameters to simulate the AGV's operating state in different regions. Based on the simulation results and a preset disturbance trigger threshold, the first disturbance range is determined, representing the degree of airflow disturbance to the AGV in different regions.

[0018] The AGV digital model is a digital model built based on the physical properties of the AGV, used to simulate the AGV's operating state and environmental parameters. Default operating parameters refer to the AGV's operating parameters under ideal conditions, such as maximum operating speed and turning radius, providing a benchmark for AGV operation. The first disturbance range refers to the airflow disturbance range determined based on the AGV digital model and default operating parameters, simulating the AGV's operating state in different areas. The disturbance trigger threshold is a threshold used to determine whether airflow disturbance has reached the trigger condition; when airflow disturbance exceeds this threshold, that location in the space is considered to be affected by airflow disturbance.

[0019] For example, taking the lithography area of ​​a chip factory as an example, the AGV digital model includes the AGV's physical attributes such as a load capacity of 50kg and dimensions of 1m×1m×0.5m, and its operating parameters such as a maximum operating speed of 1m / s and a turning radius of 0.5m. Based on the default operating parameters of the lithography area, the simulation is performed by traversing the lithography area to simulate the AGV's operating state in the lithography area. Based on the simulation results and the preset disturbance trigger threshold of 0.3, the first disturbance range is determined to be 0.15~0.22.

[0020] Secondly, based on the digital model and the default operating parameters of each region, an ideal blank space is established for simulation. Based on the simulation results and a preset disturbance trigger threshold, a reference disturbance range is determined. The ideal blank space refers to an ideal space within the chip manufacturing workshop without AGV operation or equipment interference, serving as a reference benchmark for AGV operation. The reference disturbance range refers to the airflow disturbance range determined based on the simulation results of the ideal blank space, used for comparison with the airflow disturbance range in actual production. Based on the AGV digital model and the default operating parameters of each region, an ideal blank space is established for simulation, simulating the AGV's operating state in the ideal blank space. Based on the simulation results and a preset disturbance trigger threshold, a reference disturbance range is determined, representing the degree of airflow disturbance in the ideal blank space. Combined with the process requirements of chip manufacturing, the reference disturbance range is optimized to ensure its comparability with the airflow disturbance range in actual production.

[0021] For example, taking the lithography area of ​​a chip factory as an example, its ideal blank space is a space without AGV operation and without equipment interference. Based on the AGV digital model and the default operating parameters of the lithography area, an ideal blank space is established and simulation is performed to simulate the operating state of the AGV in the ideal blank space. Based on the simulation results and the preset disturbance trigger threshold of 0.3, the reference disturbance range is determined to be 0.10~0.16.

[0022] Next, historical disturbance data for each region is acquired, and the second disturbance range is defined using the type probability of the transported object corresponding to the historical disturbance data as weight, combined with a normalized weighting method. This historical disturbance data includes disturbance data for different transported object types, such as wafers and photoresist. The second disturbance range is defined using the type probability of the transported object corresponding to the historical disturbance data as weight, combined with a normalized weighting method. This range represents the degree of airflow disturbance in actual production.

[0023] Historical disturbance data refers to airflow disturbance data recorded during historical production processes within the chip manufacturing workshop, used to provide a reference for AGV operation. Type probability is the probability distribution of the type of transported object, reflecting the degree of influence of the object's shape on airflow disturbance. The second disturbance range refers to the airflow disturbance range determined based on historical disturbance data and type probability, combined with a normalized weighting method. The normalized weighting method involves normalizing historical disturbance data of different types and then weighting them according to the type probability to obtain the second disturbance range.

[0024] For example, taking the lithography area of ​​a chip factory as an example, its historical disturbance data includes disturbance data of handling objects such as wafers and photoresist. Extracting historical data, the probability of wafer box handling is 0.6 and the disturbance value is 0.18~0.25, the probability of photoresist bottle handling is 0.4 and the disturbance value is 0.20~0.28. After normalization and weighting, the second disturbance range is calibrated to be 0.188~0.262.

[0025] Furthermore, based on the ratio of the weighted fusion range of the first disturbance range and the second disturbance range to the reference disturbance range, an inherent airflow disturbance coefficient is determined. This inherent airflow disturbance coefficient is then corrected according to the preset cleanliness level of each region to obtain the final airflow disturbance coefficient. The weight of the first disturbance range is set to 0.4, and the weight of the second disturbance range is set to 0.6. The weighted fusion range of the two is calculated. The ratio of the median of the weighted fusion range to the median of the reference disturbance range is taken as the inherent airflow disturbance coefficient. The inherent coefficient is then corrected according to the region's cleanliness level to finally obtain the airflow disturbance coefficient.

[0026] The inherent airflow disturbance coefficient is a coefficient determined by the ratio of the weighted fusion range of the first and second disturbance ranges to the reference disturbance range. It reflects the degree of airflow disturbance in actual production. The area cleanliness level refers to the cleanliness level of different areas within the chip manufacturing workshop, such as Class 100 and Class 1000. Different cleanliness levels have different impacts on AGV operation. The airflow disturbance coefficient is a parameter characterizing the degree of influence of airflow movement within the chip manufacturing workshop on AGV handling; a larger value indicates a more significant interference from airflow disturbances generated by AGV operation.

[0027] For example, taking the photolithography area of ​​a chip factory as an example, the first perturbation range is 0.10~0.16, the second perturbation range is 0.188~0.262, and the reference perturbation range is 0.10~0.16. The weighted fusion range is 0.1728~0.2452, with a median of 0.209. The median of the reference perturbation range is 0.13. 0.209 / 0.13≈1.608. The correction factor corresponding to the cleanliness level Class 100 of the area is 0.3. The airflow perturbation coefficient = 1.608×0.3=0.482. The final airflow perturbation coefficient is 0.482.

[0028] Subsequently, the current production plan and real-time scheduling scheme for the target scenario are obtained from the Manufacturing Execution System (MES). The MES is the core management system responsible for production planning, progress monitoring, and task allocation in the chip manufacturing process. The current production plan refers to the phased production tasks issued by the MES system, including task quantity, process sequence, completion deadline, and priority. The real-time scheduling scheme refers to the currently executing AGV handling arrangements, including task allocation, path planning, and execution progress. A real-time data interface is established between the central scheduling terminal and the workshop MES system to retrieve the current production plan and extract key information such as process name, task quantity, completion deadline, workpiece type, and corresponding production equipment. The real-time scheduling scheme is read from the AGV scheduling module of the MES system, breaking down the task allocation relationship, the coordinates of each AGV's running path, and the current execution status, such as whether AGV1 is en route and the number of completed tasks. The acquired data is cleaned, and invalid data, such as duplicate path coordinates and obsolete production tasks, is removed. The data is then categorized and organized by region, equipment, and task dimensions to form a standardized data list.

[0029] For example, based on the lithography area scenario of a chip factory mentioned above, the MES system is accessed through the central scheduling terminal: The current production plan is retrieved: 10 wafers will be moved for the lithography process within 1 hour, and 8 wafers will be moved for the inspection process. The equipment corresponding to the lithography process is lithography machine A / B, and the equipment corresponding to the inspection process is inspection machine C. The real-time scheduling plan is read: AGV1 is responsible for wafer movement for the lithography process, with the path being wafer storage area, lithography machine A, lithography machine B, wafer storage area. Currently, 3 wafers have been moved and are waiting in the wafer storage area. The data is cleaned and organized to form a list: Area: Lithography area, Equipment: Lithography machine A / B, Inspection machine C, Task: Lithography movement of 10 wafers / 1 hour, Inspection movement of 8 wafers / 1 hour, AGV1 status: waiting, 3 wafers have been moved, Path: Storage area, Lithography machine A / B, Storage area.

[0030] In addition, the safe operating parameters of the AGVs are retrieved from the vehicle specification library of the target scenario. The vehicle specification library is a database that stores the specification parameters of all AGV equipment in the workshop, including hardware parameters, operating parameters, safety thresholds, and other information. AGV safe operating parameters are the core parameters for ensuring the stable and safe operation of AGVs, including maximum operating speed, turning radius, maximum load capacity, emergency stop response time, and safe distance. Exceeding the parameter range can easily lead to equipment failure or safety accidents. The central dispatch terminal accesses the workshop vehicle specification library, retrieves the corresponding specification parameters by AGV number, filters key safe operating parameters, and eliminates irrelevant parameters, such as equipment manufacturing date and maintenance records. The parameters are verified by comparing them with the historical average operating parameters of the AGV to confirm that the parameters have not been tampered with and are within a reasonable range, ultimately forming a list of AGV safe operating parameters.

[0031] For example, based on the above AGV1 scenario: the central dispatch terminal accesses the vehicle specification library and retrieves the complete specification parameters of AGV1; filters safe operating parameters: maximum operating speed 1m / s, turning radius 0.5m, maximum load capacity 50kg, emergency stop response time 0.2s, and safe distance from equipment / other AGVs 0.3m; verifies parameters: compares the historical operating parameter averages of AGV1, such as maximum average operating speed 0.95m / s and load capacity 45kg, confirms that the current parameters are reasonable, and forms a list of safe operating parameters.

[0032] The corresponding initialization scheduling baseline information includes: Based on the area type and the digital map, construct clean area constraint rules; Based on the aforementioned safe operating parameters, vehicle safety constraint rules are constructed; Define process constraint rules based on the current production plan.

[0033] First, clean area constraint rules are constructed based on the area type and the digital map. These rules are adapted to the cleanliness level of the area and prevent AGV operation from damaging the clean environment. They include speed limits, access ranges, and operating procedures to ensure that workpieces are not disturbed by airflow or contaminated by dust. Based on the area type and digital map, the clean area boundaries are clearly defined; for example, the area within the lithography area fence is a clean area, and the area outside the fence is a non-clean area. Combined with the area's cleanliness level, AGV operating speed limits are set; the higher the cleanliness level, the stricter the speed limit. Areas within the clean area where AGVs are prohibited from passing are delineated, such as directly below equipment and around airflow recirculation vents, and the priority of access channels is clearly defined. Integrating these requirements forms the clean area constraint rules.

[0034] For example, based on the above lithography area scenario: clearly define the clean area boundary: the area inside the lithography area fence is the clean area, and AGVs are prohibited from entering outside the fence; set speed limits: the maximum operating speed of AGVs in the clean area shall not exceed 0.5m / s, which is lower than 1m / s in the non-clean area; delineate prohibited areas: the area within a radius of 0.5m directly below the lithography machine A / B and the area within a radius of 0.8m around the airflow circulation vent are prohibited from entering, and the main channel has higher priority than the auxiliary channel; form clean area constraint rules: when AGVs operate in the clean area of ​​the lithography area, they must travel along the main channel at a speed ≤0.5m / s, and are prohibited from entering the area below the equipment and around the vent.

[0035] Secondly, based on the aforementioned safe operating parameters, vehicle safety constraint rules are constructed. These rules, based on the AGV's safe operating parameters, ensure the AGV's stability and prevent collisions with equipment or other AGVs, including load limits, turning specifications, and safe distance requirements. Based on the AGV's safe operating parameters, upper load limits and lower turning radius limits are set; the safe distance requirements for AGV operation are clearly defined, i.e., the minimum distance between the AGV and equipment, other AGVs, and workshop walls; abnormal response rules are set, such as automatic speed reduction when the operating speed exceeds the limit; triggering an emergency stop when the distance to an obstacle is less than the safety threshold; and rules are associated with AGV numbers to ensure that the rules are compatible with the corresponding AGV models.

[0036] For example, based on the above-mentioned AGV1 safe operation parameters scenario: set load constraints: the AGV1 load capacity shall not exceed 50kg, and it is prohibited to transport overweight workpieces; the turning radius shall not be less than 0.5m, and sharp turns at the corners of the passage are prohibited; set safety distance constraints: the distance from fixed equipment such as lithography machines and testing equipment during operation shall be ≥0.3m, and the distance from other AGVs shall be ≥0.5m; set abnormal response rules: when the speed in the clean area exceeds 0.5m / s, or the speed in the non-clean area exceeds 1m / s, the speed shall be automatically reduced to the compliant range; when the distance from an obstacle is <0.2m, an emergency stop shall be triggered; the above rules are only applicable to AGV1 and shall be synchronously entered into the central dispatch terminal constraint rule library.

[0037] Next, based on the current production plan, define process constraint rules. Process constraint rules refer to rules that adapt to the chip manufacturing process sequence and task priority, ensuring the coordination between AGV handling and the production process. These rules include task execution order, priority ranking, and time limits. Based on the current production plan, clarify the process sequence of each step, such as requiring wafers to complete lithography handling before inspection handling; set AGV task execution priorities according to the task priorities in the production plan, prioritizing high-priority tasks; set AGV task completion time limits in conjunction with production time limits, such as requiring the handling of 10 wafers for lithography to be completed within 1 hour; and clarify task connection rules, such as requiring AGVs to prioritize connecting to the next process equipment after completing the handling of the previous process to avoid process delays.

[0038] For example, based on the current production plan scenario of the lithography area described above: Clearly define process sequence constraints: Wafers must first be transported from the storage area to lithography machines A / B by AGV1, and after lithography is completed, they must be transported to inspection equipment C. Reversing the order is prohibited. Set priority constraints: Lithography transport tasks have higher priority than inspection transport tasks. AGV1 must complete lithography transport first, and then undertake inspection transport. Set time limit constraints: AGV1 must complete the lithography transport of 10 wafers within 1 hour, triggering an alert for every 10-minute delay. Set connection rules: After AGV1 completes the lithography transport of one wafer, it must immediately report back to the MES system to synchronously confirm the transport instruction for the next wafer, avoiding idle lithography machines.

[0039] In this embodiment of the invention, by real-time and comprehensive collection of multi-source data from chip manufacturing scenarios, key environmental parameters such as airflow disturbance coefficients in various regions are accurately quantified, and a three-dimensional constraint rule system covering cleanroom adaptation, vehicle safety, and process coordination is constructed. This clarifies the environmental boundaries, safety thresholds, and process requirements for AGV operation, solving the problems of insufficient data support and ambiguous constraint rules in existing scheduling. It provides a precise data foundation and rigid constraint basis for subsequent scheduling trigger event monitoring and path optimization decisions, ensuring that subsequent scheduling schemes can meet the actual needs of chip manufacturing scenarios and avoid risks such as exceeding cleanliness standards, process delays, and equipment collisions.

[0040] S200: Based on the preset scheduling trigger event library, perform scheduling trigger event monitoring and real-time trigger judgment.

[0041] In this embodiment of the invention, scheduling trigger events are monitored and real-time trigger judgments are performed based on a preset scheduling trigger event library. In chip manufacturing scenarios, AGV handling tasks need to adapt to production disturbances in real time. Existing technologies lack dynamic monitoring and rapid response mechanisms for critical events such as equipment failures, order changes, and timing deviations, which can easily lead to problems such as AGV path conflicts, task delays, and equipment idling. Therefore, event monitoring and trigger judgment are needed to achieve real-time capture and accurate identification of various disturbance events during the production process, providing a trigger basis for subsequent dynamic scheduling decisions and ensuring the timeliness and adaptability of the scheduling scheme.

[0042] Step S200 in the method provided in this embodiment of the invention includes: Real-time monitoring of the equipment automation program interface and the AGV body data interface to obtain event monitoring data; The system iterates through the preset scheduling trigger event library, compares and matches the event monitoring data, and outputs a real-time trigger judgment result based on the comparison and matching results. The real-time trigger judgment result includes a trigger discrimination Boolean value and a trigger event category label. The scheduling trigger event library includes at least resource scheduling events, task scheduling events, and time-series scheduling events.

[0043] First, real-time monitoring of the equipment automation program interface and the AGV body data interface is used to acquire event monitoring data. The equipment automation program interface refers to the communication interface opened by the automation control program of the production equipment in the chip manufacturing workshop, used to transmit data such as equipment operating status and fault alarms in real time. The AGV body data interface refers to the communication interface between the built-in sensors and control modules of the AGV, used to transmit data such as AGV operating status and task execution progress. Event monitoring data refers to real-time data collected from the equipment automation program interface and the AGV body data interface, including information such as equipment fault alarms, AGV status anomalies, and production plan changes. The scheduling trigger event library refers to a set of pre-defined event types that need to be monitored in the chip manufacturing scenario, including characteristic parameters and triggering conditions for resource, task, and time-series events.

[0044] Specifically, the central dispatch terminal establishes a real-time connection with the automation program interface of all production equipment in the workshop through network communication protocols such as TCP / IP, and starts a continuous monitoring mode. At the same time, the central dispatch terminal establishes a real-time connection with the body data interface of all AGV equipment, starts a continuous monitoring mode, and obtains the status data of AGV operation. The collected raw data is initially filtered to remove duplicate and invalid data, retain the valid data containing event characteristics, and form an event monitoring data list.

[0045] For example, the central dispatch terminal establishes a connection with the interfaces of the lithography machine and AGV1 to monitor in real time: alarm data is obtained from the lithography machine automation program interface: lithography machine A has a sudden failure; status data is obtained from the AGV1 body data interface: AGV1 battery level is 15%, which is lower than the 20% threshold; task change data is obtained from the MES system interface: an emergency order has been added, and wafers need to be moved first; after filtering duplicate heartbeat signals, an event monitoring data list containing 3 valid data points is formed.

[0046] Secondly, the pre-defined scheduling trigger event library is traversed, and the event monitoring data is compared and matched. Based on the comparison and matching results, a real-time trigger judgment result is output. The real-time trigger judgment result includes a trigger discrimination Boolean value and a trigger event category flag. The scheduling trigger event library includes at least resource scheduling events, task scheduling events, and time-series scheduling events.

[0047] The scheduling trigger event library is a pre-defined event feature library containing specific definitions and matching rules for three types of events: resource scheduling events: AGV malfunctions, production equipment malfunctions, AGV power shortages, and other resource anomaly events; task scheduling events: emergency order placements, order cancellations, process path changes, and other task-related events; time-series scheduling events: upstream equipment processing completion time deviations exceeding thresholds, AGV waiting times being too long, and other time-series deviation events; and comparison matching rules: logical rules for accurately matching the features of event monitoring data with the pre-defined event features in the event library.

[0048] Specifically, the system iterates through the pre-defined scheduling trigger event library, extracting the characteristic parameters of each event in the library according to the priority order of resource scheduling events, task scheduling events, and time-series scheduling events. These parameters include the fault type of resource events, the order type of task events, and the deviation threshold of time-series events. Each data point in the event monitoring data list is compared item by item with the corresponding event characteristic parameters in the event library to determine whether the data meets the event triggering conditions. For successfully matched events, information such as event type, trigger time, and scope of impact is recorded. Events that fail to match are marked as non-triggering events and removed from the subsequent judgment process.

[0049] For example, based on the above monitoring data list, the event database is traversed and compared: the monitoring data of sudden failure of lithography machine A matches the feature parameters of resource scheduling event - production equipment failure in the event database, the trigger condition is equipment EAP alarm, and the match is successful; the monitoring data of AGV1 battery 15% matches the feature parameters of resource scheduling event - AGV battery insufficient in the event database, the trigger condition is battery below 20% threshold, and the match is successful; the monitoring data of newly added emergency order matches the feature parameters of task scheduling event - emergency order placement in the event database, the trigger condition is MES issuing emergency order, and the match is successful; the monitoring data of upstream equipment processing time deviation matches the feature parameters of timing scheduling event - processing time deviation in the event database, the trigger condition is deviation exceeding threshold, and the match is successful; the AGV1 normal operation data that does not match is marked as non-triggering event and removed from the subsequent process.

[0050] In this embodiment of the invention, by monitoring, comparing and matching, and outputting results, the real-time capture and accurate identification of three types of key events in the chip manufacturing scenario are achieved. This enables rapid response to disturbances such as equipment failures, order changes, and timing deviations, providing clear triggering basis for subsequent dynamic scheduling decisions. It effectively improves the real-time performance and adaptability of AGV handling path scheduling, avoids production interruptions or task delays caused by disturbance events, and ensures the smooth operation of the chip manufacturing process.

[0051] S300: If the real-time trigger judgment passes, then based on the scheduling trigger event monitoring results, the scheduling baseline information and the pre-built disturbance proxy evaluation model, a multi-objective optimization-based AGV transport path decision is made to obtain a dynamic scheduling scheme.

[0052] In this embodiment of the invention, if the real-time trigger judgment passes, a multi-objective optimization-based AGV transport path decision is made based on the scheduling trigger event monitoring results, the scheduling baseline information, and the pre-built disturbance proxy evaluation model to obtain a dynamic scheduling scheme. In chip manufacturing scenarios, after a scheduling trigger event occurs, global scheduling of all AGVs and transport tasks in the workshop would result in excessive computation and decision delays. Furthermore, traditional path planning does not accurately couple the disturbance evaluation results and struggles to simultaneously consider multiple objectives such as transport efficiency, airflow disturbance control, task priority, and equipment utilization. A single optimization algorithm cannot balance the efficiency and accuracy of path planning. Therefore, it is necessary to first construct an accurate disturbance proxy evaluation model to provide a basis for cost quantification, then accurately locate the tasks and AGVs affected by the event, and finally achieve path decision-making under multi-objective constraints through a hierarchical hybrid optimization algorithm. This reduces the scheduling scope and computational load while ensuring the multi-dimensional optimality of the scheduling scheme, adapting to the high-precision and high-coordination requirements of chip manufacturing.

[0053] Step S300 in the method provided in this embodiment of the invention includes: The construction of the disturbance agent evaluation model includes: Establish a digital model of the AGV and construct a corresponding task load model based on the historical task information of the AGV in the scenario. Couple the digital model with the task load model to obtain the disturbance evaluation model of the AGV, and perform multi-condition simulation analysis based on the disturbance evaluation model and the safe operation parameters of the AGV. Based on the results of multi-condition simulation analysis, and combined with prior knowledge, operational indicators and key disturbance indicators are extracted to obtain disturbance assessment sample data. Based on the perturbation assessment sample data, a perturbation proxy assessment model containing an autoencoder layer and a regression analysis layer is constructed and trained. The autoencoder layer is used to map the input operational indicators into low-dimensional vectors, and the regression analysis layer is used to map the low-dimensional vectors into the key perturbation indicators.

[0054] First, a digital model of the AGV is established, and a corresponding task load model is constructed based on the AGV's historical task information. The AGV digital model is a digital simulation model built upon the AGV's physical properties and motion characteristics. It accurately reproduces the AGV's actual operating state and supports the adjustment of operating parameters and real-time output of operating data. Historical task information refers to all the handling tasks previously performed by the AGV in the chip manufacturing scenario, including the type of object being handled, load weight, operating path, regional environment, and disturbance correlation data. The task load model is a digital model that quantifies the correlation between handling load and AGV operating disturbances based on historical task information, reflecting the degree of impact of different loads on AGV disturbances.

[0055] Secondly, the digital model and the task load model are coupled to obtain the AGV disturbance assessment model. Based on this disturbance assessment model, multi-condition simulation analysis is performed in conjunction with the AGV's safe operating parameters. Coupling refers to the data linkage and fusion of the AGV digital model and the task load model, enabling the input and output parameters of the two models to be interconnected, forming an integrated model that simultaneously reflects the AGV's own operation and the impact of load disturbances. The disturbance assessment model is the integrated model after coupling the AGV digital model and the task load model, which can simulate and analyze the comprehensive disturbance degree of the AGV under different loads and operating conditions. Multi-condition simulation analysis refers to simulating the AGV's operating state under different operating parameters, load conditions, and regional environments based on the disturbance assessment model, achieving full-scenario, full-variable disturbance simulation, and obtaining operating and disturbance data corresponding to multiple sets of operating conditions. AGV safe operating parameters are core parameters ensuring the safe and compliant operation of the AGV, including maximum operating speed, minimum turning radius, and load limit, providing constraint boundaries for setting operating conditions.

[0056] Specifically, the AGV disturbance assessment model can output disturbance data such as comprehensive airflow disturbance value and path operation deviation based on the input AGV operating parameters and load characteristic parameters; retrieve the AGV's safe operating parameters, and set multiple sets of independent operating condition variables with the safe operating parameters as the constraint boundary, covering AGV operating speed, load characteristic parameters, operating area, and path segment length; control the disturbance assessment model to traverse all combinations of operating condition variables, perform simulation analysis group by group, and synchronously record the AGV operating parameters, load parameters, and corresponding comprehensive disturbance data under each group of operating conditions to form a full-condition simulation dataset.

[0057] For example, the digital model of AGV1 is coupled with the task load model. The load parameters output by the task load model are used as the basis for the disturbance calculation of the digital model to generate the AGV1 disturbance evaluation model. With the safe operation parameters of AGV1 as constraints, the working condition variables are set as follows: 4 operating speeds: 0.3m / s, 0.4m / s, 0.6m / s, 0.8m / s; 2 load parameters: 0.3, 0.54; 1 operating area: lithography area; airflow disturbance coefficient: 0.482; 3 path lengths: 5m, 10m, 15m, for a total of 24 working condition combinations. The AGV1 disturbance evaluation model is controlled to perform simulations one by one, and the operating speed, load parameters, path length and corresponding comprehensive airflow disturbance value, clean area disturbance diffusion range and other data of each working condition are recorded to form 24 sets of initial simulation data.

[0058] Furthermore, based on the results of multi-condition simulation analysis, and combined with prior knowledge, operational indicators and key disturbance indicators are extracted to obtain disturbance assessment sample data. Operational indicators refer to the core characteristic parameters extracted from the multi-condition simulation analysis results that characterize the AGV's operating state; they are the input features for disturbance assessment and are directly related to the degree of disturbance. Key disturbance indicators refer to the disturbance quantification parameters extracted from the multi-condition simulation analysis results that are strongly correlated with chip production quality; they are the output targets for disturbance assessment and directly reflect the disturbance impact of AGV operation on the production scenario. Prior knowledge refers to professional common sense and industry experience in chip production and AGV scheduling, such as the more significant the impact of airflow disturbance at higher cleanroom levels, and the positive correlation between AGV operating speed and airflow disturbance. Disturbance assessment sample data refers to structured data composed of a set of operational indicators and a corresponding set of key disturbance indicators; it serves as the foundational dataset for subsequently training the disturbance proxy assessment model.

[0059] Specifically, based on prior knowledge of chip production and AGV scheduling, parameters directly related to the degree of disturbance are selected from the multi-condition simulation analysis results as operational indicators. The selection criteria are that the parameters are quantifiable, have a clear correlation with the disturbance, and are easy to collect in real time. Based on the standard of not affecting chip production quality and meeting the clean area operation requirements, disturbance parameters strongly related to chip production are extracted from the simulation analysis results as key disturbance indicators, focusing on quantitative indicators related to airflow disturbance and clean area disturbance. The operational indicators in each set of simulation data are matched one by one with the corresponding key disturbance indicators to form a structured disturbance assessment sample data. After deduplication and outlier removal of all sample data, they are divided into model training set and test set in a 7:3 ratio.

[0060] For example, based on prior knowledge, five operational indicators are extracted from 24 sets of simulation data of AGV1: AGV operating speed, load characteristic parameters, path segment length, regional airflow disturbance coefficient, and regional cleanliness level; three key disturbance indicators are extracted: comprehensive airflow disturbance quantification value, clean area disturbance diffusion range, and AGV path operation deviation value; each set of operational indicators, such as [0.3m / s, 0.3, 5m, 0.482, Class100], is matched with the corresponding key disturbance indicators, such as [0.12, 0.5m, 0.03], to form a single sample data. After deduplication and outlier removal, 2000 valid sample data are obtained, of which 1400 are the training set and 600 are the test set.

[0061] Furthermore, based on the perturbation assessment sample data, a perturbation surrogate assessment model comprising an autoencoder layer and a regression analysis layer is constructed and trained. The autoencoder layer maps the input operational indicators into low-dimensional vectors, and the regression analysis layer maps the low-dimensional vectors into the key perturbation indicators. The autoencoder layer, a feature dimensionality reduction layer in deep learning networks, consists of an input layer, hidden layers, and an output layer. It can map high-dimensional, multi-dimensional operational indicators into low-dimensional vectors while retaining the core feature information of the operational indicators. The low-dimensional vector refers to the output of the autoencoder layer, with a dimension far lower than the input operational indicators, representing the digital representation of the core features of the operational indicators. The regression analysis layer is a deep learning prediction layer that takes low-dimensional vectors as input and maps them into continuous predicted values ​​of key perturbation indicators through regression calculations. The perturbation surrogate assessment model is a neural network model composed of an autoencoder layer and a regression analysis layer. It can quickly and accurately output predicted values ​​of key perturbation indicators based on the input operational indicators, replacing the complex simulation process of traditional perturbation assessment models.

[0062] Specifically, a neural network model is constructed, consisting of an autoencoder layer and a regression analysis layer. The input dimension of the autoencoder layer is the same as the number of operational indicators, while the hidden layer dimension is set to 1 / 3 to 1 / 2 of the input dimension to achieve feature dimensionality reduction. The output layer outputs a low-dimensional vector. The regression analysis layer takes the low-dimensional vector output by the autoencoder layer as input, and its output dimension is the same as the number of key perturbation indicators, responsible for outputting the predicted values ​​of the key perturbation indicators. The training set of perturbation assessment sample data is input into the model. The autoencoder layer extracts and reduces the dimensionality of the high-dimensional operational indicators, mapping them to low-dimensional core feature vectors. The regression analysis layer performs regression calculations based on these low-dimensional vectors, outputting the predicted values ​​of the key perturbation indicators. The model is iteratively trained using gradient descent with the mean squared error between the predicted values ​​of the key perturbation indicators and the true values ​​in the sample data as the loss function. After each training round, the model's prediction accuracy is verified using a test set. Training stops when the model's prediction accuracy on the test set reaches a preset threshold (≥95%), or when the number of iterations reaches a preset value, resulting in the final perturbation surrogate assessment model.

[0063] For example, a disturbance proxy evaluation model is built for AGV1: the autoencoder layer has a 5-dimensional input layer corresponding to 5 operational indicators, the hidden layer is set to 2-dimensional, and the output layer outputs a 2-dimensional low-dimensional vector; the regression analysis layer takes the 2-dimensional low-dimensional vector as input and outputs a 3-dimensional layer corresponding to 3 key disturbance indicators; 1400 training set samples are input into the model, and iterative training is performed using the mean squared error as the loss function and gradient descent method. After each iteration, 600 test sets are used for verification. When the iteration reaches 50 rounds, the prediction accuracy reaches 96.5%, which meets the preset threshold, and training is stopped, thus completing the construction of the AGV1 disturbance proxy evaluation model. This disturbance proxy evaluation model can directly input operational indicators [0.5m / s, 0.3, 8m, 0.482, Class100], which are reduced to 2-dimensional low-dimensional vectors by the autoencoder layer, and the regression analysis layer quickly outputs the predicted values ​​of key disturbance indicators [0.21, 0.78m, 0.04].

[0064] If the real-time trigger check passes, the following steps will then be taken: Based on the monitoring results of the scheduling trigger events, identify the tasks that directly affect them; Based on the current production plan and real-time scheduling scheme in the aforementioned scheduling baseline information, the associated transportation tasks that have a process sequence dependency relationship with the directly affected tasks are determined. The directly impacting tasks and the associated transportation tasks are defined as the target transportation task set, and the associated automated guided vehicles set is determined. The AGV transport path decision based on multi-objective optimization is executed for the target transport task set and the automated guided vehicle set.

[0065] First, based on the monitoring results of the scheduling trigger events, directly affected tasks are identified. The monitoring results of the scheduling trigger events refer to the event trigger judgment results output by S200, including information such as event type, trigger time, scope of impact, and degree of interference with the production process. Directly affected tasks refer to AGV transport tasks that cannot be executed according to the original scheduling plan or require adjustment of the execution strategy due to the occurrence of scheduling trigger events. The monitoring results of the scheduling trigger events output by S200 are retrieved to extract the event type, scope of impact, and degree of interference with the production process; combined with current production status data such as equipment operating status and task execution progress, transport tasks directly interfered with by the events and unable to be executed according to the original plan are screened out; the screened tasks are marked, and key information such as task name, original AGV, original planned execution time, and associated production equipment is recorded.

[0066] For example, based on the monitoring results of the scheduling trigger event of a sudden failure of lithography machine A and the MES issuing an urgent order for lithography of 10 wafers: the event type is extracted as resource scheduling event - production equipment failure, the affected area is the lithography area, and the interference level is high; combined with the production status data: lithography machine A is in a stopped state, the wafer storage area originally planned to be executed by AGV1 and the lithography machine A handling task cannot be executed; the tasks directly affected are screened out as: wafer storage area and lithography machine A handling task, denoted as task T1, the original executing AGV is marked as AGV1, the original planned execution time is within 0.5h, and the associated equipment is lithography machine A.

[0067] Secondly, based on the current production plan and real-time scheduling scheme in the aforementioned scheduling baseline information, associated transportation tasks with process sequence dependencies on the directly affected tasks are determined. The scheduling baseline information refers to the multi-source scenario data and scheduling constraint rules output by S100, including the current production plan, real-time scheduling scheme, AGV safety operation parameters, etc. The current production plan refers to the phased production task arrangement in the chip manufacturing workshop, including task volume, process sequence, completion deadline, priority, etc. The real-time scheduling scheme refers to the currently executing AGV handling task allocation and path planning scheme. Process sequence dependency refers to the logical relationship between the handling tasks of each process in chip manufacturing; if the preceding task is not completed, the subsequent task cannot be executed, such as the sequence of wafer handling, lithography, and inspection. Associated transportation tasks refer to handling tasks that have a process sequence dependency on the directly affected tasks; if the directly affected task is delayed, it will be simultaneously affected.

[0068] Specifically, the scheduling baseline information output by S100 is retrieved, and the core data in the current production plan and real-time scheduling scheme are extracted, including the process sequence, task priority, completion time limit, and associated production equipment of each handling task; the position of the directly affected task in the production process is analyzed, and the incomplete preceding task and the pending subsequent task are identified; all handling tasks that have a process sequence dependency relationship with the directly affected task are screened out, including the preceding task, the subsequent task, and the parallel associated task, forming a set of associated transportation tasks.

[0069] For example, based on the background information of tasks and scheduling that directly affect the production plan, the process sequence information in the current production plan and real-time scheduling scheme is extracted: the subsequent task of T1 is the handling task of lithography machine A and inspection equipment C, denoted as T2, and the parallel related task of T1 is the wafer storage area and lithography machine B handling task added by MES, denoted as T3; taking T1 as the core, tasks T2 and T3 that have process sequence dependencies with it are selected and determined as related transportation tasks; the priority of related transportation tasks is defined as: T3 (urgent task) > T1 (original task) > T2 (subsequent task).

[0070] Furthermore, the directly impacting tasks and the associated transportation tasks are defined as a target transportation task set, and the associated automated guided vehicles (AGVs) set is determined. The target transportation task set is a task set composed of the directly impacting tasks and the associated transportation tasks, and is the core execution object of this dynamic scheduling. The associated AGV set refers to the set of equipment consisting of the AGVs responsible for executing the target transportation task set in the original scheduling plan, and the idle AGVs in the workshop that can be allocated to this task set. The directly impacting tasks and associated transportation tasks are integrated to form the target transportation task set, which is uniformly numbered and the priority, completion time limit, and associated equipment of each task in the set are clearly defined; the AGVs responsible for executing the target transportation task set in the real-time scheduling plan are extracted, and their operating status, power consumption, and location information are recorded; AGVs in the workshop that are currently idle and whose safe operating parameters are adapted to the target transportation task set are retrieved; the above AGVs are integrated into an associated AGV set, and the executable task types and operating constraints of each AGV in the set are clearly defined.

[0071] For example, based on the directly affected task T1 and the associated transportation tasks T2 and T3, T1, T2 and T3 are integrated to form a target transportation task set, and the priority of tasks in the set is defined as T3 > T1 > T2, with completion time limits of 1 hour, 0.5 hours and 0.3 hours respectively; the AGV responsible for executing this task set in the real-time scheduling plan is extracted as AGV1; the idle AGV in the workshop is retrieved as AGV2; the associated AGV set is determined as {AGV1, AGV2}, and it is determined that AGV1 can execute T1 and T2, and AGV2 can execute T3.

[0072] Specifically, based on the monitoring results of scheduling trigger events, the aforementioned scheduling baseline information, and the pre-built disturbance proxy evaluation model, a multi-objective optimization-based AGV transport path decision is made to obtain a dynamic scheduling scheme, including: Based on the monitoring results of the scheduling trigger events, initialize and define scheduling decision variables, wherein the scheduling decision variables include task allocation relationships and task path series; Based on the disturbance proxy evaluation model, a target cost function is constructed, wherein the target cost function includes at least a transportation time cost sub-item, an airflow disturbance cost sub-item, an urgent order delay penalty cost sub-item, and a critical equipment idle cost sub-item; With the goal of minimizing the objective cost function and the scheduling baseline information as a constraint, a hierarchical hybrid optimization solution is performed.

[0073] First, based on the monitoring results of the scheduling trigger events, scheduling decision variables are initialized and defined. These variables include task allocation relationships and task path series. Scheduling decision variables are quantitative variables representing the core decision content of this dynamic scheduling, serving as the solution object for multi-objective optimization, and are set only for the target transportation task set and the associated AGV set. The task allocation relationship refers to the one-to-one mapping between each task in the target transportation task set and the AGVs in the associated AGV set, specifying which AGV will execute each task. The task path series refers to the specific travel path corresponding to each task, composed of multiple continuous path segments divided in the workshop digital map according to the execution order, serving as the route basis for AGV task execution. The monitoring results of the scheduling trigger events are structured data containing event type, scope of impact, and task priority adjustment requirements, providing boundary conditions for setting decision variables.

[0074] Specifically, the system retrieves the monitoring results of scheduling trigger events, the set of target transportation tasks, and the set of associated AGVs. Based on the task-AGV correspondence, it defines task allocation relationship variables: represented by a two-dimensional 0-1 matrix. Rows correspond to the unique numbers of the target transportation tasks, and columns correspond to the unique numbers of the associated AGVs. A matrix value of 1 indicates that the task is assigned to the corresponding AGV, and 0 indicates that it is not assigned. Each row of the matrix contains only one 1. Based on the constraints in the workshop digital map and scheduling baseline information, the workshop access routes are divided into standardized path segments and assigned unique numbers. Task path series variables are defined, using an ordered array to represent the path series for each task. The elements in the array are consecutive path segment numbers, and the array order corresponds to the AGV's passage order. The path series can only select path segments that comply with clean area and vehicle safety constraints. Considering the impact of scheduling trigger events, such as disabling path segments around faulty equipment and prioritizing urgent task paths, the system sets initial value ranges for two types of decision variables, eliminating variable values ​​that do not conform to the event impact and constraint rules. For example, the path segment corresponding to a faulty lithography machine cannot be included in the task path series.

[0075] For example, based on the target transportation task set {T1: wafer storage area, lithography machine B; T2: lithography machine B, inspection equipment C; T3: wafer storage area, lithography machine B} and the associated AGV set {AGV1, AGV2}, the scheduling trigger events are: lithography machine A failure and MES issuing an urgent lithography order. A 0-1 matrix of task allocation relationships is defined, with rows: T1, T2, T3; columns: AGV1, AGV2. The initial matrix value range is one 1 per row. The task allocation is then divided based on a digital map of the lithography area. Define the path segments as follows: P1: Wafer storage area, intersection 1; P3: Intersection 1, lithography machine B; P4: Lithography machine B, inspection equipment C. Define the task path series variables as ordered arrays, which can only consist of P1, P3, and P4. For example, the initial range of the path series for T3 is [P1, P3], [P1, intersection 2, P3], etc. In combination with the influence of events, remove all path series values ​​containing P2. Set the task allocation relationship variable for urgent order T3 to only take the values ​​of idle high-power AGVs in AGV1 and AGV2.

[0076] Secondly, based on the aforementioned disturbance proxy evaluation model, a target cost function is constructed. This target cost function includes at least the following sub-items: transportation time cost, airflow disturbance cost, urgent order delay penalty cost, and critical equipment idle cost. The target cost function quantifies the multiple demands of chip production on AGV scheduling into a single comprehensive cost index; a smaller function value indicates a better scheduling scheme. The transportation time cost sub-item is a standardized index quantifying the total transportation time of all tasks in the target transportation task set by the associated AGVs, reflecting the execution efficiency of the scheduling scheme. The airflow disturbance cost sub-item quantifies the total airflow disturbance generated by all AGVs across all travel path segments. It is proportional to the airflow disturbance sensitivity coefficient of each path segment and the travel time of the AGVs on the corresponding path segment; the total cost is the sum of the calculated values ​​of all AGVs across all path segments. The urgent order delay penalty cost sub-item quantifies the penalty cost incurred when an urgent order task is not completed within the preset time limit; the cost is 0 if completed on time, and the longer the delay, the greater the penalty cost. The critical equipment idle cost sub-item is the cost incurred by critical equipment in quantification chip production due to task delays. The longer the idle time and the higher the importance of the equipment, the greater the cost.

[0077] First, construct the transportation time cost sub-item F1: calculate the basic travel time of each path segment based on the workshop digital map, determine the transportation time of a single AGV performing a single task by combining the AGV safety operation parameters, sum up to obtain the total transportation time of all AGVs completing the assigned tasks, introduce a time standardization coefficient (total transportation time / preset shortest transportation time), and normalize F1 to the 0-1 interval.

[0078] Secondly, construct the airflow disturbance cost sub-item F2: input the AGV operation indicators (operating speed, path segment, area parameters, etc.) into the disturbance proxy evaluation model, and output the airflow disturbance sensitivity coefficient of each AGV in each path segment; calculate the actual travel time of each AGV in each path segment, and the disturbance cost of a single path segment = sensitivity coefficient × travel time. Then sum the disturbance costs of all AGVs in all path segments to obtain the total airflow disturbance cost, which is also normalized to the 0-1 interval as F2.

[0079] Next, construct the urgent order delay penalty cost sub-item F3: set the preset completion time limit for urgent orders and the unit delay penalty coefficient, calculate the difference between the actual completion time of the urgent order task and the preset time limit. If the difference is ≤0, then F3=0; if the difference is >0, then F3=delay time×unit delay penalty coefficient, and after normalization, it is included in the total function.

[0080] In addition, a key equipment idle cost sub-item F4 is constructed: set the idle weight coefficient for each key equipment, the higher the equipment importance, the greater the weight, calculate the actual idle time of the equipment due to task delay, that is, the difference between the task execution time and the actual execution time, the single equipment idle cost = idle time × idle weight coefficient, sum up the idle costs of all key equipment and normalize them to obtain F4.

[0081] Finally, the overall objective cost function is constructed: based on the priority of chip production needs, the weight coefficients ω1, ω2, ω3, and ω4 of the four sub-items are set, and the sum of the weights is 1. The overall function is: overall objective cost F = ω1×F1 + ω2×F2 + ω3×F3 + ω4×F4. The optimization objective is to minimize the overall objective cost F.

[0082] For example, based on the above target task, associated AGVs, and disturbance agent evaluation model, weights are set as ω1=0.2, ω2=0.3, ω3=0.3, and ω4=0.2. The preset time limit for urgent order T3 is 1 hour, the idle weight coefficient for lithography machine B is 2, and the unit delay penalty coefficient is 5. Calculate F1: AGV2 executes T3 transportation time of 0.3 hours, AGV1 executes T1 and T2, total transportation time is 1.0 hour, preset minimum time is 0.8 hours, after standardization F1=1.0 / 0.8=1.25, after normalization it is 1.0. Calculate F2: Sensitivity coefficients output by the disturbance proxy evaluation model: P1=0.2, P3=0.482, P4=0.3; AGV2 travels through P1 (0.1h) and P3 (0.2h), disturbance cost = 0.2×0.1+0.482×0.2=0.1164; AGV1 travels through P1 (0.1h), P3 (0.2h), and P4 (0.2h), disturbance cost = 0.2×0.1+0.482×0.2+0.3×0.2=0.1764; total airflow disturbance cost = 0.1164+0.1764=0.2928, normalized F2=0.2928; Calculate F3: T3 is actually completed in 0.3h, with no delay, F3=0. Calculate F4: Lithography machine B is idle for only 0.3 hours due to execution T3, with no task delay, idle time is 0, F4=0. The total objective cost function F=0.2×1.0+0.3×0.2928+0.3×0+0.2×0≈0.2878.

[0083] Furthermore, with the goal of minimizing the objective cost function and the scheduling baseline information as constraints, a hierarchical hybrid optimization solution is performed.

[0084] The solution involves a hierarchical hybrid optimization process, with the objective of minimizing the target cost function and the scheduling baseline information as constraints. This includes: Task allocation and coarse path selection are performed using a higher-level genetic algorithm, where the fitness function is the target cost function. A lower-level conflict-based search algorithm is used to refine the path planning with time windows based on the task allocation and coarse path selection results output by the upper-level genetic algorithm.

[0085] First, a genetic algorithm is used for task allocation and coarse path selection, where the fitness function is the objective cost function. The genetic algorithm is a global optimization algorithm that simulates natural selection in biological evolution. It iteratively selects the optimal solution through selection, crossover, and mutation to adapt to multi-variable combinations of tasks and path decisions. The fitness function directly adopts the objective cost function; a smaller function value indicates a better scheduling scheme. Coarse path selection only determines the core path segments for the task, clarifying the starting point, key areas traversed, and the destination, without involving precise travel timing or detailed node planning.

[0086] Specifically, the task allocation relationship 0-1 matrix and the ordered array of task paths are first uniformly encoded and merged into a one-dimensional chromosome, with each chromosome corresponding to a complete preliminary scheduling scheme. Then, a reasonable population size is set according to the target task and the number of AGVs, and an initial population that meets the scheduling background information constraints is randomly generated. Next, the target cost function is used as the fitness function to calculate and sort the fitness value of each chromosome in the population, and select high-quality chromosomes. Then, selection, crossover, and mutation genetic operations are performed on the high-quality chromosomes in sequence to generate the next generation population. Finally, the calculation and genetic operations are continuously iterated until the population converges or reaches the preset number of iterations. After decoding the optimal chromosome, the optimal task allocation relationship and coarse path selection results are output.

[0087] For example, for the target transportation task set {T1, T2, T3} and the associated AGV set {AGV1, AGV2}, the population size is set to 50 and the number of iterations is set to 50. The coding rules are formulated according to the task number, AGV number, and path segment number, and 50 initial chromosomes are generated. The fitness value of each chromosome is calculated using the objective cost function. The top 10 high-quality chromosomes are selected and subjected to selection, crossover, and mutation operations. After 50 iterations, the population converges. The optimal chromosome is decoded to obtain the task allocation result as AGV1 executes T1 and AGV2 executes T2+T3. The rough path is T1[P1, P3], T2[P3, P4], T3[P1, P3].

[0088] Secondly, a conflict-based search algorithm is employed at the lower level to refine the task allocation and coarse path selection results output by the upper-level genetic algorithm using a time-window-based path planning approach. The conflict-based search algorithm refers to a local optimization algorithm for path planning, which resolves spatial and temporal conflicts in paths by detecting conflicts and making targeted adjustments to paths or timing. The time window refers to the execution time range set for each task, including task start / end limits, equipment availability time, and AGV passage timing constraints; it is the core basis for path refinement. The time-window-based path refinement planning involves precise timing and node adjustments to the upper-level coarse path under the constraints of the time window and scheduling baseline information, ensuring that the path is conflict-free and compliantly executable.

[0089] Specifically, the process begins by retrieving data such as task deadlines, equipment availability, and AGV status from the baseline scheduling information, and setting a time window containing core constraints for each task. Then, based on the task allocation and coarse path output from the upper layer, the passage information of each AGV is extracted, and temporal conflicts, spatial conflicts, and constraint conflicts in the path are comprehensively detected and marked, forming a list of conflict points. Next, conflicting paths are adjusted according to a priority order: temporal, spatial, and constraint. Conflicts are resolved primarily by adjusting the AGV passage timing; if timing adjustment is ineffective, compliant alternative path segments are used, while simultaneously optimizing AGV operating parameters to meet constraints. Finally, the adjusted refined path undergoes full constraint verification. After confirming no conflicts and compliance with the time window and baseline scheduling information requirements, the final refined path is output and integrated with the upper-layer task allocation results to form a complete scheduling scheme.

[0090] For example, based on the task allocation and coarse path output from the upper layer, time windows [0, 1h] are set for T3, [0.5, 1.2h] for T1, and [1.0, 1.5h] for T2. A speed constraint of 0.5 m / s in the clean area is defined. It is found that there is an overlap and conflict in the passage time of AGV1 and AGV2 in the P1 path segment. The passage time of AGV2 in P1 is adjusted to 0-0.1h and the passage time of AGV1 in P1 is adjusted to 0.5-0.6h. After the adjustment, no other conflicts are detected, and all AGV running parameters and passage time meet the constraints and time window requirements. The final refined path output is T3 [P1, P3] (completed in 0-0.3h), T1 [P1, P3] (completed in 0.5-1.0h), and T2 [P3, P4] (completed in 1.0-1.2h).

[0091] In this embodiment of the invention, by constructing a disturbance proxy evaluation model, the precise quantification of AGV operation disturbances is achieved, providing a reliable basis for multi-objective cost calculation and solving the problem of inaccurate disturbance evaluation in traditional scheduling. Furthermore, by locating tasks that directly affect and are related to tasks, the scope of dynamic scheduling is narrowed, the computational load of scheduling decisions is reduced, and decision-making efficiency is improved. Finally, by initializing scheduling decision variables and constructing a multi-dimensional objective cost function, multi-objective coordination of transportation efficiency, airflow disturbance, urgent order priority, and equipment utilization is achieved. Combined with hierarchical hybrid optimization solution, the upper-level genetic algorithm ensures the global optimum of task allocation and coarse path, while the lower-level conflict-based search algorithm achieves refined path and conflict-free planning. The final output dynamic scheduling scheme not only conforms to all constraints of chip production but also minimizes multi-objective costs, adapting to production needs after scheduling trigger events and providing high-quality candidate schemes for subsequent scheme comparison.

[0092] S400: Compare the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario, and select the better one to execute.

[0093] In this embodiment of the invention, the dynamic scheduling scheme is compared with the real-time scheduling scheme of the target scenario, and the superior one is selected for execution. The dynamic scheduling scheme is an optimization scheme generated based on scheduling trigger events, but the real-time scheduling scheme of the target scenario is still in operation and may still be adaptable to some production stages. If the dynamic scheduling scheme is directly replaced, it may lead to production interruptions, task delays, and other problems because the new scheme is not fully adapted to the current instantaneous production state or there are potential path conflicts. To ensure the feasibility, multi-objective optimality, and continuity of the chip production process of the scheduling scheme, it is necessary to systematically compare the dynamic scheduling scheme and the real-time scheduling scheme, accurately determine the advantages and disadvantages of the two, and select the scheme that better fits the current production needs for execution, avoiding the production risks caused by blind replacement.

[0094] Specifically, the core dimensions and judgment criteria for scheme comparison are first clarified. The comparison dimensions must correspond one-to-one with the sub-items of the target cost function, including transportation time, airflow disturbance cost, urgent order delay penalty, and key equipment idle cost. At the same time, two auxiliary dimensions, scheme feasibility and execution continuity, are added. The weights of each dimension are set according to production needs, with urgent order response and feasibility having the highest weights, each accounting for 30%, and the remaining dimensions each accounting for 10%. Then, the core parameters of the dynamic scheduling scheme and the real-time scheduling scheme are extracted, including task allocation relationship, path planning details, cost values ​​of each dimension, execution sequence, and constraint compliance. Subsequently, based on the set dimensions and weights, the comprehensive score of the two schemes is calculated. The scheme with the higher comprehensive score and meeting the core dimension standards is judged as the better scheme. Finally, the optimal scheme execution instruction is issued to the central dispatch terminal, and the scheduling information of the MES system, AGV control module, and production equipment is updated simultaneously to ensure that all links execute the optimal scheme in a coordinated manner. At the same time, the scheme comparison results and execution logs are recorded to accumulate data for subsequent scheduling optimization.

[0095] For example, based on the lithography area scenario described above, the real-time scheduling scheme for the target scenario is: AGV1 executes T1 and T2, while AGV2 remains in standby mode. This scheme fails to respond to the newly added urgent order T3 from the MES system and carries execution risk due to a malfunction in lithography machine A, resulting in a comprehensive score of 0.4. The dynamic scheduling scheme is: AGV2 executes T3, and AGV1 executes T1 and T2 after charging. This scheme meets all constraints, responds to urgent orders on time, and achieves optimal cost across all dimensions, resulting in a comprehensive score of 0.92. Through dimensional comparison and comprehensive score calculation, the dynamic scheduling scheme achieves a higher comprehensive score than the real-time scheduling scheme, and all core dimensions meet the standards. Therefore, the dynamic scheduling scheme is deemed superior, and an execution command is sent to the central scheduling terminal. The control commands for the MES system and AGV1 and AGV2 are simultaneously updated, and the optimal scheme is initiated.

[0096] In this embodiment of the invention, by comparing and selecting the optimal solution, the production risk of blindly replacing the scheduling solution is effectively avoided, ensuring that the executed scheduling solution achieves optimal performance in multiple dimensions such as emergency order response, transportation efficiency, disturbance control, and equipment utilization, while also ensuring the feasibility of the solution and the continuity of the production process.

[0097] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a dynamic scheduling method and system for AGV transport paths in chip manufacturing. First, by comprehensively acquiring multi-source scenario data and accurately initializing scheduling baseline information, it solves the problems of data fragmentation and coarse constraint rules in traditional scheduling. Second, through real-time event monitoring, comparison and matching, and trigger judgment, it achieves rapid capture and accurate identification of production disturbances such as equipment failures, order changes, and timing deviations, providing timely triggering conditions for dynamic scheduling. Furthermore, based on a disturbance proxy evaluation model and a hierarchical hybrid optimization algorithm, it focuses on the affected tasks and AGV range, achieving comprehensive optimization of multiple objectives such as transport efficiency, airflow disturbance, urgent order response, and equipment utilization, generating a dynamic scheduling scheme adapted to the disturbance scenario, solving the problems of imbalance between accuracy and efficiency and inaccurate disturbance quantification in single optimization algorithms. Finally, by comparing and verifying the dynamic scheme with the real-time scheme, it avoids the production risk of blindly replacing the scheme, ensuring the multi-dimensional optimality of the execution scheme and production continuity. This invention improves the adaptability of AGV handling scheduling to dynamic disturbances in chip manufacturing scenarios, effectively reduces risks such as task delays, path conflicts, and equipment idling, and ensures high-precision, high-efficiency, and high-stability operation of chip manufacturing, providing technical support for the automated and intelligent scheduling of the entire chip manufacturing process.

[0098] Example 2, as Figure 2 As shown, this invention provides a dynamic scheduling system for AGV transport paths in chip manufacturing, the system comprising: The scene data initialization module 11 is used to obtain multi-source scene data of the target scene in real time through the central dispatch terminal, and initialize the corresponding dispatch background information. The trigger event monitoring module 12 is used to monitor and determine the trigger events in real time based on the preset scheduling trigger event library. The dynamic path optimization decision module 13 is used to make a multi-objective optimization AGV transport path decision based on the scheduling trigger event monitoring results, the scheduling baseline information and the pre-built disturbance proxy evaluation model if the real-time trigger judgment is passed, and obtain a dynamic scheduling scheme. The scheduling scheme selection and execution module 14 is used to compare the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario and select the better one to execute.

[0099] In one embodiment, the scene data initialization module 11 is further configured to: This includes acquiring multi-source scene data of the target scene in real time through the central dispatch terminal, including: Load a digital map of the target scene, and determine the type of each area and the airflow disturbance coefficient based on the digital map; Obtain the current production plan and real-time scheduling scheme for the target scenario from the Manufacturing Execution System; Read the safe operation parameters of the AGV from the vehicle specification library of the target scenario.

[0100] The corresponding initialization scheduling baseline information includes: Based on the area type and the digital map, construct clean area constraint rules; Based on the aforementioned safe operating parameters, vehicle safety constraint rules are constructed; Define process constraint rules based on the current production plan.

[0101] This includes loading a digital map of the target scene and determining the type of each region and the airflow disturbance coefficient based on the digital map, including: A digital model of the AGV is established, and simulations are performed in each region based on the default operating parameters of each region. Based on the simulation results and the preset disturbance trigger threshold, the first disturbance range is determined. Based on the digital model and the default operating parameters of each region, an ideal blank space is established to perform simulation, and the reference disturbance range is determined based on the simulation results and the preset disturbance trigger threshold. Historical disturbance data for each region is obtained, and the second disturbance range is determined by weighting the type probability of the transported object type corresponding to the historical disturbance data and combining it with a normalized weighting method. The inherent airflow disturbance coefficient is determined based on the ratio of the weighted fusion range of the first disturbance range and the second disturbance range to the reference disturbance range. The inherent airflow disturbance coefficient is then corrected based on the preset area cleanliness level of each area to obtain the airflow disturbance coefficient.

[0102] In one embodiment, the event monitoring module 12 is further configured to: Real-time monitoring of the equipment automation program interface and the AGV body data interface to obtain event monitoring data; The system iterates through the preset scheduling trigger event library, compares and matches the event monitoring data, and outputs a real-time trigger judgment result based on the comparison and matching results. The real-time trigger judgment result includes a trigger discrimination Boolean value and a trigger event category label. The scheduling trigger event library includes at least resource scheduling events, task scheduling events, and time-series scheduling events.

[0103] In one embodiment, the dynamic path optimization decision module 13 is further configured to: The construction of the disturbance agent evaluation model includes: Establish a digital model of the AGV and construct a corresponding task load model based on the historical task information of the AGV in the scenario. Couple the digital model with the task load model to obtain the disturbance evaluation model of the AGV, and perform multi-condition simulation analysis based on the disturbance evaluation model and the safe operation parameters of the AGV. Based on the results of multi-condition simulation analysis, and combined with prior knowledge, operational indicators and key disturbance indicators are extracted to obtain disturbance assessment sample data. Based on the perturbation assessment sample data, a perturbation proxy assessment model containing an autoencoder layer and a regression analysis layer is constructed and trained. The autoencoder layer is used to map the input operational indicators into low-dimensional vectors, and the regression analysis layer is used to map the low-dimensional vectors into the key perturbation indicators.

[0104] If the real-time trigger check passes, the following steps will then be taken: Based on the monitoring results of the scheduling trigger events, identify the tasks that directly affect them; Based on the current production plan and real-time scheduling scheme in the aforementioned scheduling baseline information, the associated transportation tasks that have a process sequence dependency relationship with the directly affected tasks are determined. The directly impacting tasks and the associated transportation tasks are defined as the target transportation task set, and the associated automated guided vehicles set is determined. The AGV transport path decision based on multi-objective optimization is executed for the target transport task set and the automated guided vehicle set.

[0105] Specifically, based on the monitoring results of scheduling trigger events, the aforementioned scheduling baseline information, and the pre-built disturbance proxy evaluation model, a multi-objective optimization-based AGV transport path decision is made to obtain a dynamic scheduling scheme, including: Based on the monitoring results of the scheduling trigger events, initialize and define scheduling decision variables, wherein the scheduling decision variables include task allocation relationships and task path series; Based on the disturbance proxy evaluation model, a target cost function is constructed, wherein the target cost function includes at least a transportation time cost sub-item, an airflow disturbance cost sub-item, an urgent order delay penalty cost sub-item, and a critical equipment idle cost sub-item; With the goal of minimizing the objective cost function and the scheduling baseline information as a constraint, a hierarchical hybrid optimization solution is performed.

[0106] The solution involves a hierarchical hybrid optimization process, with the objective of minimizing the target cost function and the scheduling baseline information as constraints. This includes: Task allocation and coarse path selection are performed using a higher-level genetic algorithm, where the fitness function is the target cost function. A lower-level conflict-based search algorithm is used to refine the path planning with time windows based on the task allocation and coarse path selection results output by the upper-level genetic algorithm.

[0107] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic scheduling method for AGV transport paths in chip manufacturing, characterized in that, include: The central dispatch terminal acquires multi-source scene data of the target scene in real time and initializes the corresponding dispatch baseline information. Based on the preset scheduling trigger event library, perform scheduling trigger event monitoring and real-time trigger judgment; If the real-time trigger judgment passes, then based on the scheduling trigger event monitoring results, the scheduling baseline information and the pre-built disturbance proxy evaluation model, a multi-objective optimization-based AGV transport path decision is made to obtain a dynamic scheduling scheme. Compare the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario, and select the better one to execute.

2. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 1, characterized in that, Real-time acquisition of multi-source scene data of the target scene through the central dispatch terminal, including: Load a digital map of the target scene, and determine the type of each area and the airflow disturbance coefficient based on the digital map; Obtain the current production plan and real-time scheduling scheme for the target scenario from the Manufacturing Execution System; Read the safe operation parameters of the AGV from the vehicle specification library of the target scenario.

3. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 2, characterized in that, Corresponding initialization scheduling baseline information, including: Based on the area type and the digital map, construct clean area constraint rules; Based on the aforementioned safe operating parameters, vehicle safety constraint rules are constructed; Define process constraint rules based on the current production plan.

4. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 2, characterized in that, Load a digital map of the target scene, and determine the type of each area and the airflow disturbance coefficient based on the digital map, including: A digital model of the AGV is established, and simulations are performed in each region based on the default operating parameters of each region. Based on the simulation results and the preset disturbance trigger threshold, the first disturbance range is determined. Based on the digital model and the default operating parameters of each region, an ideal blank space is established to perform simulation, and the reference disturbance range is determined based on the simulation results and the preset disturbance trigger threshold. Historical disturbance data for each region is obtained, and the second disturbance range is determined by weighting the type probability of the transported object type corresponding to the historical disturbance data and combining it with a normalized weighting method. The inherent airflow disturbance coefficient is determined based on the ratio of the weighted fusion range of the first disturbance range and the second disturbance range to the reference disturbance range. The inherent airflow disturbance coefficient is then corrected based on the preset area cleanliness level of each area to obtain the airflow disturbance coefficient.

5. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 1, characterized in that, Based on a pre-defined database of scheduling trigger events, the system monitors and determines which events will trigger in real time, including: Real-time monitoring of the equipment automation program interface and the AGV body data interface to obtain event monitoring data; The system iterates through the preset scheduling trigger event library, compares and matches the event monitoring data, and outputs a real-time trigger judgment result based on the comparison and matching results. The real-time trigger judgment result includes a trigger discrimination boolean value and a trigger event category label. The scheduling trigger event library includes at least resource scheduling events, task scheduling events, and time-series scheduling events.

6. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 1, characterized in that, The construction of the perturbation agent evaluation model includes: Establish a digital model of the AGV and construct a corresponding task load model based on the historical task information of the AGV in the scenario. Couple the digital model with the task load model to obtain the disturbance evaluation model of the AGV, and perform multi-condition simulation analysis based on the disturbance evaluation model and the safe operation parameters of the AGV. Based on the results of multi-condition simulation analysis, and combined with prior knowledge, operational indicators and key disturbance indicators are extracted to obtain disturbance assessment sample data. Based on the perturbation assessment sample data, a perturbation proxy assessment model containing an autoencoder layer and a regression analysis layer is constructed and trained. The autoencoder layer is used to map the input operational indicators into low-dimensional vectors, and the regression analysis layer is used to map the low-dimensional vectors into the key perturbation indicators.

7. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 1, characterized in that, If the real-time trigger check passes, the following steps will follow: Based on the monitoring results of the scheduling trigger events, identify the tasks that directly affect them; Based on the current production plan and real-time scheduling scheme in the aforementioned scheduling baseline information, the associated transportation tasks that have a process sequence dependency relationship with the directly affected tasks are determined. The directly impacting tasks and the associated transportation tasks are defined as the target transportation task set, and the associated automated guided vehicles set is determined. The AGV transport path decision based on multi-objective optimization is executed for the target transport task set and the automated guided vehicle set.

8. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 1, characterized in that, Based on the monitoring results of scheduling trigger events, the aforementioned scheduling baseline information, and the pre-built disturbance proxy evaluation model, a multi-objective optimization-based AGV transport path decision is made to obtain a dynamic scheduling scheme, including: Based on the monitoring results of the scheduling trigger events, initialize and define scheduling decision variables, wherein the scheduling decision variables include task allocation relationships and task path series; Based on the disturbance proxy evaluation model, a target cost function is constructed, wherein the target cost function includes at least a transportation time cost sub-item, an airflow disturbance cost sub-item, an urgent order delay penalty cost sub-item, and a critical equipment idle cost sub-item; With the goal of minimizing the objective cost function and the scheduling baseline information as a constraint, a hierarchical hybrid optimization solution is performed.

9. The dynamic scheduling method for AGV transport paths in chip manufacturing as described in claim 8, characterized in that, With the goal of minimizing the objective cost function and the scheduling baseline information as constraints, a hierarchical hybrid optimization solution is performed, including: Task allocation and coarse path selection are performed using a higher-level genetic algorithm, where the fitness function is the target cost function. A lower-level conflict-based search algorithm is used to refine the path planning with a time window based on the task allocation and coarse path selection results output by the upper-level genetic algorithm.

10. A dynamic scheduling system for AGV transport paths in chip manufacturing, characterized in that, The system is used to implement the dynamic scheduling method for AGV transport paths in chip manufacturing according to any one of claims 1-9, the system comprising: The scene data initialization module is used to acquire multi-source scene data of the target scene in real time through the central dispatch terminal, and initialize the corresponding dispatch background information. The event monitoring module is used to monitor and determine real-time trigger events based on a preset event database. The dynamic path optimization decision module is used to make AGV transport path decisions based on multi-objective optimization, and obtain a dynamic scheduling scheme, based on the scheduling trigger event monitoring results, the scheduling baseline information and the pre-built disturbance proxy evaluation model, if the real-time trigger judgment is passed. The scheduling scheme optimization and execution module is used to compare the dynamic scheduling scheme with the real-time scheduling scheme of the target scenario and select the better one to execute.

Images