Process manufacturing factory material scheduling system and scheduling method based on intelligent agent

Abstract

The invention discloses a process manufacturing factory material scheduling system based on intelligent agents. The system comprises a central coordination intelligent agent, a plurality of process intelligent agents, a video subsystem and a trolley subsystem. The scheduling method based on the system comprises the following steps: S1, initializing the system; s2, real-time state monitoring; s3, dolly scheduling decision making: the process agent calculates the priority of each transport task according to the priority function; s4, according to the queue capacity of the downstream workshop and the current outlet material quantity, S5, path decision making; s6, returning a path decision; s7, performing global optimization adjustment; s8, performing exception handling and cooperative adjustment; according to the invention, organic combination of local autonomy and global coordination is realized, the real-time response capability of the system is ensured, global optimization is also realized, and compared with traditional centralized control, the system is faster in response; multiple constraint conditions and optimization targets are considered, the average waiting time of materials can be reduced by one third, and the trolley utilization rate can be improved by one fourth.

Description

Technical Field

[0001] This invention relates to the field of intelligent factory process scheduling technology, specifically to a material scheduling system and method for process manufacturing plants based on intelligent agents. Background Technology

[0002] Factories typically involve multiple processes. Raw materials need to be processed sequentially through each process according to a predetermined process route, ultimately producing finished products.

[0003] With the integration of digital and information technologies with industrialization, the construction of smart factories places great emphasis on the efficient transfer of materials between processes, as it is crucial for improving production efficiency and reducing inventory costs. Currently, material scheduling in process manufacturing plants primarily employs a centralized control architecture, where a central control system manages the scheduling of all transportation equipment. This approach suffers from the following technical problems: First, the central control system needs to process large amounts of real-time data, resulting in a heavy computational burden and slow response time; second, when local anomalies occur on the production floor, the centralized system struggles to respond quickly, leading to the amplification of the anomaly's impact; third, the system lacks adaptive capabilities and cannot automatically adjust scheduling strategies based on dynamic changes in production status. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a material scheduling system and method for process manufacturing plants based on intelligent agents.

[0005] The technical solution of the present invention to solve the above problems is: a material scheduling system for a process manufacturing plant based on intelligent agents, including a central coordinating intelligent agent, multiple process intelligent agents, a video subsystem, and a vehicle subsystem;

[0006] The number of process agents is N, which correspond to the N processes of the process manufacturing. The process agent of the nth process is responsible for managing the M(n) workshops of the process and is used to control the scheduling of vehicles, path planning and return decision within the process segment.

[0007] The central coordinating agent is connected to multiple process agents to optimize the allocation of the vehicle among the processes from a global perspective, coordinate the flow of materials among multiple processes, and handle system-level abnormal events.

[0008] The video subsystem includes camera devices deployed at the entrances and exits of each workshop, used to identify the length of the material queue, the unloading status of the trolley, or the loading status in real time.

[0009] The vehicle subsystem includes S unmanned transport vehicles and their control units. Each vehicle is equipped with an onboard camera and a communication module for receiving material transport instructions, providing feedback on path status information, and reporting abnormal information.

[0010] Furthermore, the process intelligence agent includes a vehicle scheduling module, a loading decision module, a path decision module, a return decision module, a video analysis module, and a communication interface module that are interconnected. The vehicle scheduling module is used to calculate the vehicle scheduling priority and generate a scheduling plan based on the exit material queue status of each workshop in the current process and the entrance queue status of each workshop in the downstream process. The loading decision module is used to determine the quantity of goods loaded by the vehicle. The path decision module is used to select the optimal path for the vehicle from the current workshop to the downstream workshop. The return decision module is used to select the optimal return target workshop based on the real-time queue status of each workshop in the current process after the vehicle has completed unloading in the downstream process. The video analysis module is used to obtain the queue length information of each workshop by performing target detection on the queue videos at the workshop entrance and exit, and to identify the status of the unloading and loading vehicles by analyzing the videos. The communication interface module is used to interact with the central coordinating intelligence agent.

[0011] Furthermore, the central coordinating intelligent agent includes a globally optimized module, a bottleneck detection module, a collaborative scheduling module, an emergency response module, and a fault takeover module that are interconnected. The globally optimized module is used to establish a global multi-objective optimization model and dynamically adjust the number of vehicles allocated to each process segment. The bottleneck detection module is used to monitor the material flow status of each process in real time and identify production bottlenecks. The collaborative scheduling module is used to coordinate relevant processes to adjust their scheduling strategies when an anomaly occurs in a process. The emergency response module is used to manage mobile vehicle resources and temporarily deploy them in emergency situations. The fault takeover module is used to temporarily take over the scheduling function of a process intelligent agent when it fails. The fault takeover module interacts with the intelligent agents of each process.

[0012] A material scheduling method for a process manufacturing plant based on intelligent agents includes the following steps:

[0013] S1. System initialization: Establish a hierarchical control architecture containing N process agents and 1 central coordinating agent, and set the initial number of vehicles allocated to each process.

[0014] S2. Real-time status monitoring: Each process agent continuously monitors the entry queue length Q_in(n,i) and exit queue length Q_out(n,i) of each workshop in its process by analyzing the video stream from the video subsystem in real time, where n is the process number and i is the workshop number; the path congestion coefficient c_cong is updated by the congestion events reported by the vehicle-mounted camera.

[0015] S3, Car dispatching decision: The process agent calculates the priority of each transportation task according to the priority function;

[0016] S4. Loading quantity decision: Based on the queue capacity of the downstream workshop and the current quantity of exported materials, determine the optimal loading quantity p*=min{P(n),Q_out(n,i),Q_max-Q_in(n+1,j)}, where P(n) is the maximum loading capacity of the trolley in the nth process segment;

[0017] S5. Path Decision: The process agent uses distance and congestion system as metrics and uses the Dijkstra algorithm to calculate the shortest path as the selected transportation path for the vehicle.

[0018] S6. Return Path Decision: After the trolley completes unloading, the process agent calculates and returns the cost function to each workshop.

[0019] S7. Global Optimization and Adjustment: The central coordinating agent performs global optimization on a periodic T_globe basis, adjusting the vehicle allocation scheme according to the operating efficiency and queue status of each process.

[0020] S8. Anomaly Handling and Coordinated Adjustment: When an anomaly is detected, a coordinated adjustment mechanism is triggered. The central coordinating agent coordinates the relevant processes to adjust the scheduling strategy and deploys mobile vehicles when necessary.

[0021] Furthermore, the priority function in step S3 is:

[0022] P(n,i,j)=α·Q_out(n,i)+β·(Q_max-Q_in(n+1,j))+γ·exp(-in d(n,i,j))

[0023] Where Q_out(n,i) is the length of the exit queue of the i-th workshop in the n-th process, Q_in(n+1,j) is the length of the entrance queue of the j-th workshop in the (n+1)-th process, Q_max is the upper limit of the queue capacity, d(n,i,j) is the distance from workshop (n,i) to workshop (n+1,j), and α, β, γ, and λ are weight parameters.

[0024] Furthermore, the specific method for the process agent to calculate and return the cost function for each workshop in step S6 is as follows:

[0025] C_return(i) = d(current,i)·[1+Num_to(i) / Num_v(n)] / [1+Q_out(n,i) / Q_out_avg], where d(current,i) is the distance from the current position to workshop i, Q_out_avg is the average length of the exit queue of each workshop, Num_to(i) is the number of cars heading to workshop i, and Num_v(n) is the total number of cars in the nth process segment; the workshop with the lowest cost is selected as the return target.

[0026] Furthermore, the specific method for adjusting the car allocation scheme in step S7 is as follows:

[0027]

[0028] stNum_v(n)≥MinNum_v(n), and

[0029] Where T_wait(n) is the average material waiting time for process n, T_idle(k) is the idle time of trolley k, Var(Q_out) is the variance of the length of the exit queue for each workshop, Num_v(n) is the number of trolleys allocated to process n, MinNum_v(n) is the minimum number of trolleys allocated to process n, and S is the total number of trolleys.

[0030] Furthermore, in step S8, abnormal situations include, but are not limited to: the length of the exit queue of a certain workshop exceeds the threshold θ1; the stopping time of a certain car exceeds the threshold θ2; the communication interruption of a certain process agent exceeds the threshold θ3; the average processing time of a certain process deviates from the normal value by more than 3 times the standard deviation.

[0031] Furthermore, step S8 specifically includes the following steps:

[0032] S801: Each process's intelligent agent continuously monitors the status of its respective process;

[0033] S802: Determine if an anomaly has occurred. Anomaly types include, but are not limited to: Type A: Exit queue too long; Type B: Trolley status abnormality; Type C: Equipment failure; Type D: Processing efficiency abnormality.

[0034] If no abnormalities are found, return to S801 for continued monitoring;

[0035] If an anomaly is detected, execute S803;

[0036] S803: Send alarm information to the central coordinating agent, including but not limited to the type, location, and severity level of the anomaly;

[0037] S804: The central coordinating agent assesses the scope of the anomaly's impact, and preferably, also predicts the duration of the anomaly and assesses the losses caused by the anomaly;

[0038] S805: The central coordinating agent determines whether coordinated adjustments are needed.

[0039] If the anomaly only affects a single process, execute S806 for local adjustment; otherwise, execute S807 for global coordinated adjustment.

[0040] S806: Partial adjustments, such as temporarily adding motorized trolleys or adjusting the loading quantity;

[0041] S807: Global coordinated adjustment, including: adjusting upstream processes, such as reducing the output rate; adjusting this process, such as deploying mobile resources and optimizing internal scheduling; adjusting downstream processes, such as increasing buffer capacity and adjusting the receiving strategy.

[0042] S808: Execute the adjustment plan: Issue instructions to the process agent, allocate resources, and monitor the execution effect;

[0043] S809: Determine whether the exception has been resolved. If not, return to S804 for re-evaluation. If resolved, execute S810.

[0044] S810: Restore normal operating mode and recover motor resources.

[0045] The present invention has the following beneficial effects:

[0046] 1. By constructing a hierarchical intelligent agent architecture, the organic combination of local autonomy and global coordination is achieved, which not only ensures the real-time response capability of the system, but also realizes global optimization. Compared with traditional centralized control, the system responds faster.

[0047] 2. The introduction of machine vision technology for real-time monitoring of queue status and path congestion provides accurate on-site information perception capabilities and reliable data support for optimization decisions;

[0048] 3. By adopting intelligent optimization algorithms to dynamically adjust the scheduling and path planning of the trolleys, and taking into account various constraints and optimization objectives, the average waiting time for materials can be reduced by one-third and the utilization rate of the trolleys can be increased by one-quarter.

[0049] 4. A return path optimization mechanism for the vehicle was designed to avoid excessive empty driving distance and significantly reduce the average empty driving distance of the vehicle.

[0050] 5. It enables coordinated adjustment among multiple processes. When an anomaly occurs in a certain process, the scheduling strategy of related processes can be adjusted in a timely manner to avoid the impact of the anomaly from escalating, and the overall robustness of the system is significantly improved. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the overall architecture of a material scheduling system for a process manufacturing plant based on intelligent agents.

[0052] Figure 2 A layout diagram of process manufacturing plant A;

[0053] Figure 3 This is a functional module structure diagram of the process intelligent agent;

[0054] Figure 4 This is a schematic diagram of the optimized return path of the vehicle provided in an embodiment of the present invention;

[0055] Figure 5 Functional module structure diagram of the central coordinating intelligent agent;

[0056] Figure 6 The following is a flowchart of the decision-making process of the vehicle scheduling module: (a) is the outbound scheduling process, (b) is the inbound scheduling process, and (c) is the queue / vehicle data interaction process.

[0057] Figure 7 The timing relationship for the central coordinating intelligent agent to perform global optimization and adjustment;

[0058] Figure 8 This refers to the process by which the central coordinating agent coordinates the process agents to handle anomalies and make collaborative adjustments. Detailed Implementation

[0059] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0060] Example 1

[0061] like Figure 1 As shown, a material scheduling system for a process manufacturing plant based on intelligent agents includes a central coordinating intelligent agent, multiple process intelligent agents, a video subsystem, and a vehicle subsystem.

[0062] The number of process agents is N, which correspond to the N processes of the process manufacturing. The process agent of the nth process is responsible for managing the M(n) workshops of the process and is used to control the scheduling of vehicles, path planning and return decision within the process segment. There are 3 process agents (A1, A2 and A3), and A1, A2 and A3 correspond to process 1, process 2 and process 3 respectively.

[0063] The central coordinating agent is connected to multiple process agents to optimize the allocation of the vehicle among the processes from a global perspective, coordinate the flow of materials among multiple processes, and handle system-level abnormal events.

[0064] The video subsystem includes camera devices deployed at the entrances and exits of each workshop, used to identify the length of the material queue, the unloading status of the trolley, or the loading status in real time.

[0065] The vehicle subsystem includes S automated guided vehicles (AGVs) and their control units. Each vehicle is equipped with an onboard camera and a communication module for receiving material transport instructions, providing feedback on path status information, and reporting abnormal information.

[0066] Figure 2A floor plan of a process manufacturing plant, designated as an example, is provided. This is an automotive parts manufacturing plant comprising three processes (N=3): process 1 (machining), process 2 (heat treatment), and process 3 (surface treatment). Process 1 comprises four workshops (M(1)=4), process 2 comprises three workshops (M(2)=3), and process 3 comprises two workshops (M(3)=2). A video subsystem is equipped with high-definition cameras at the entrance and exit of each workshop, with the cameras' field of view covering the queue area. The system is configured with 15 unmanned vehicles (S=15) for material transportation.

[0067] The process intelligence agent includes the following functional modules, such as Figure 3 As shown:

[0068] (1) Car Scheduling Module 301: This module maintains a real-time task queue, which is updated periodically. When an idle car is detected, the priority of each candidate task is calculated according to the following priority function:

[0069] P(n,i,j)=α·Q_out(n,i)+β·(Q_max-Q_in(n+1,j))+γ·exp(-λd(n,i,j))

[0070] Wherein, P(n,i,j) is the priority for providing material transportation from workshop i of process n to workshop j of process n+1, Q_out(n,i) is the length of the exit queue of workshop i of process n (its value comes from video analysis module 305), Q_in(n+1,j) is the length of the entrance queue of workshop j of process n+1 (its value is received from the central coordinating agent through communication interface module 306), Q_max is the upper limit of queue capacity, d(n,i,j) is the distance from workshop i of process n to workshop j of process n+1, and α, β, γ, and λ are weight parameters.

[0071] For example, for the transportation task from workshop 2 of process 1 to workshop 1 of process 2, Q_out(1,2)=12 (there are 12 materials in the exit queue of workshop 2 of process 1), Q_in(2,1)=5 (there are 5 materials in the entrance queue of workshop 1 of process 2), Q_max=20 (the upper limit of queue capacity), d(1,2,1)=50 meters (the distance between the two workshops), then the priority calculation result is: P(1,2,1)=0.4×12+0.3×(20-5)+0.2×exp(-0.01×50)=9.42.

[0072] Idle vehicles are assigned to the highest priority candidate tasks.

[0073] Preferably, the frequency of task queue updates is set according to the generation speed of the nth process. For example, if a single workshop can complete the processing of 100 materials per hour, then the task queue can be updated every 30 seconds.

[0074] (2) Loading Decision Module 302: This module determines the optimal loading quantity p* based on the queue capacity of the downstream workshop and the current quantity of materials exported from the workshop. The calculation method is as follows:

[0075] p*=min{P(n),Q_out(n,i),Q_max-Q_in(n+1,j)}

[0076] Wherein, P(n) is the maximum loading capacity of the trolley in the nth process segment;

[0077] (3) Path Decision Module 303: This module maintains a path topology graph, where nodes represent workshop locations and critical path points, and edges represent feasible road segments. The road segment weight W_edge considers distance and congestion factors.

[0078] W_edge = L_edge × (1 + C_cong)

[0079] Where L_edge is the physical length of the edge; C_cong is the congestion coefficient of the path. Because the AGV may encounter people or other objects on the road it travels on, the AGV needs to slow down to give way or even go around obstacles. This situation is considered congestion. The AGV will report congestion events (including the location of the event and the resulting delay) to the process agent through the communication interface module. The process agent dynamically updates the C_cong value for each edge based on the congestion events reported by each AGV in the most recent window time using a preset algorithm (such as the Long Short-Term Memory Network algorithm). A value of 0 corresponds to completely smooth flow, and a larger value corresponds to more severe congestion.

[0080] Based on the road segment weights W_edge that simultaneously consider distance and congestion factors, the process agent uses the Dijkstra algorithm to calculate the shortest path, which is then used as the selected vehicle's transportation route.

[0081] (4) Return to decision module 304: After the trolley completes unloading, the process agent calculates and returns the cost function for each workshop:

[0082] C_return(i)=d(current,i)·[1+Num_to(i) / Num_v(n)] / [1+Q_out(n,i) / Q_out_avg]

[0083] Where d(current,i) is the distance from the current position to workshop i, Q_out_avg is the average length of the exit queue for each workshop, Num_to(i) is the number of carts heading to workshop i, and Num_v(n) is the total number of carts in the nth process segment; the workshop with the lowest cost is selected as the return target; for example... Figure 4 As shown, after the trolley V1 completes unloading in workshop 1 of process 2, it needs to decide which workshop in process 1 to return to. The cost of returning to workshop 1 is C_return(1)=45×(1+1 / 5) / (1+8 / 10)=30; the cost of returning to workshop 2 is C_return(2)=50×(1+0 / 5) / (1+12 / 10)=22.7; the cost of returning to workshop 3 is C_return(3)=55×(1+0 / 5) / (1+6 / 10)=34.4; the cost of returning to workshop 4 is C_return(4)=60×(1+1 / 5) / (1+14 / 10)=30. The system selects workshop 2, which has the lowest cost, as the return target.

[0084] (5) Video analysis module 305: Performs target detection on the video of the material queue collected by the cameras deployed at the material inlet and outlet of each workshop in the process section, and calculates the quantity of materials in real time; performs video analysis on the trolley at the unloading position at the material inlet of the workshop to identify the status of the trolley (including but not limited to: arrival, heavy load, light load, empty, departure); performs video analysis on the loading position of the trolley at the material outlet of the workshop to identify the status of the trolley (including but not limited to: arrival, empty, light load, heavy load, departure).

[0085] (6) Communication Interface Module 306: Communicates with the central coordinating agent using the MQTT protocol. The message format is JSON, containing fields such as timestamp, message type, and data content. Under normal circumstances, a status report is sent periodically (default value is 1 second), and an alarm message is sent immediately in case of an anomaly.

[0086] The central coordinating intelligent agent includes the following functional modules, such as Figure 5 As shown:

[0087] (1) Global Optimization Module 501: This module establishes a global multi-objective optimization model and dynamically adjusts the number of trolleys allocated to each process segment.

[0088] The central coordinating agent periodically performs global multi-objective optimization, adjusting the vehicle allocation scheme based on the operational efficiency and queue status of each process. The optimization model is as follows:

[0089]

[0090] stNum_v(n)≥MinNum_v(n), and

[0091] Where T_wait(n) is the average material waiting time for process n, T_idle(k) is the idle time of trolley k, Var(Q_out) is the variance of the length of the exit queue for each workshop, Num_v(n) is the number of trolleys allocated to process n, MinNum_v(n) is the minimum number of trolleys allocated to process n, S is the total number of trolleys, and w1, w2, and w3 are the weight coefficients of the three items respectively, used to adjust the importance of different objectives in the overall optimization;

[0092] Optionally, the optimization model can be solved using methods such as genetic algorithms, particle swarm optimization, and simulated annealing to adjust the car allocation scheme. Taking the genetic algorithm as an example, the chromosome encoding is [s1, s2, s3], representing the number of cars allocated to the 3 processes. The fitness function is Fitness = 1 / (w1 × T_wait_total + w2 × T_idle_total + w3 × var(Q_out), where w1 = 0.5, w2 = 0.3, and w3 = 0.2. After 100 generations of evolution, the optimal allocation scheme [6, 5, 4] is obtained.

[0093] (2) Bottleneck Detection Module 502: This module is used to monitor the material flow status of each process in real time and identify production bottlenecks. The bottleneck index B(n) of the nth process is calculated according to the following formula:

[0094]

[0095] Where Q_out_avg(n) is the average exit queue length of the nth process, Q_capacity is the maximum queue length, and ρ(n) is the overall equipment utilization rate of the nth process. When B(n) > 0.8, the nth process can be identified as the bottleneck.

[0096] (3) Cooperative Scheduling Module 503: This module is used to coordinate the adjustment of scheduling strategies for related processes when an anomaly occurs in a certain process. For example, when process 2 is detected as a bottleneck, the following cooperative adjustments are performed: 1) Temporarily add 1 trolley to process 2; 2) Notify process agent A3 to reserve more entry buffer space for process 3 and improve equipment utilization; 3) Notify process agent A2 to reduce equipment utilization in process 1.

[0097] (4) Emergency Response Module 504: This module is used to manage mobile vehicle resources and temporarily deploy them in emergency situations. For example, maintain 3 mobile vehicles as emergency resources. When the length of the exit queue of a certain process exceeds 90% of Q_capacity, deploy the mobile vehicle immediately.

[0098] (5) Fault takeover module 505: This module is used to temporarily take over the scheduling function of a process agent when it fails, until the failed process agent resumes operation. The fault takeover module 505 interacts with each process agent.

[0099] Example 2: This example combines Figure 6 To further explain the workflow of the vehicle scheduling module 301 of the process intelligence agent, it includes three types of processes: outbound scheduling process, inbound return scheduling process, and queue / vehicle data interaction process. Taking the second process as an example, the key steps of the three types of processes executed by the process intelligence agent A2 are as follows:

[0100] (1) Outbound Scheduling Process: This process is executed at the exit of each workshop. Taking workshop 1 (W21) as an example, this process runs in a cycle T_21_out as follows (e.g. Figure 6 (a)):

[0101] S611: The outbound scheduling period begins; check if there are any idle vehicles. If not, wait until the end of the current period; if so, execute S612.

[0102] S612: Obtain the entry queue length data Q_in(j) of all workshops in the downstream process segment, filter out all workshops with Q_in(j)≤15, and select the car transportation task from workshop W21 to W3j as the candidate task of the process;

[0103] S613: Calculate the priority P(2,1,j) of all candidate tasks;

[0104] S614: Select the highest priority candidate task;

[0105] S615: Calculate the optimal loading quantity p*, taking into account the trolley capacity (e.g., maximum loading of 10 materials), the existing quantity in the exit queue, and the remaining capacity at the downstream inlet;

[0106] S616: Plans routes for vehicles and estimates transportation time;

[0107] S617: Send a dispatch instruction to the vehicle, which must include at least the destination, the quantity of goods loaded, and the transportation route;

[0108] S618: Monitor the loading and departure process of the trolley and update its status data.

[0109] (2) Inbound return scheduling process: This process runs in a loop according to the period T_2_in as follows (e.g. Figure 6 (b)):

[0110] S621: The inlet return trip scheduling period begins. The system retrieves information on whether a trolley is currently unloaded and in an empty state in the downstream process segment, and the location of that empty trolley (at the entrance of which workshop in the downstream process segment). If not, it waits until the end of the current period; if so, it executes S622.

[0111] S622: The cost functions C_return(1), C_return(2), and C_return(3) are used to calculate the cost of idle trolleys returning from downstream workshops to workshops in this process segment.

[0112] S623: Select the workshop with the minimum cost function as the destination for the idle trolley to return to.

[0113] (3) Queue / Cart Data Interaction Process: The process agent reads real-time video data collected by cameras deployed at the entrances and exits of each workshop, and runs the following process (e.g. Figure 6 (c)):

[0114] S631: The process intelligent agent video analysis module performs target detection and counting based on the material shape in the video of the exit queue; performs video analysis on the loading position of the trolley at the material exit of the workshop to identify the status of the trolley; performs target detection and counting based on the material shape in the video of the entrance queue; and performs video analysis on the trolley at the unloading position of the material entrance of the workshop to identify the status of the trolley.

[0115] S632: The length of the entry queue and the status data of the vehicle at the entry point are sent to the central coordinating agent via the communication interface module;

[0116] S633: The process intelligence agent and the central coordination intelligence agent periodically exchange and synchronize data. The latest data obtained by the process intelligence agent is stored in the vehicle scheduling module.

[0117] Preferably (Option 1), the process intelligent agent A2 is deployed with 3 outbound scheduling processes (1 per workshop), 1 inbound return scheduling process, and 1 queue / cart data interaction process.

[0118] Optionally (Solution 2), the process agent can also perform outbound scheduling for each workshop sequentially in a round-robin manner within a single outbound scheduling process. This results in a total of one outbound outbound scheduling process, one inbound return scheduling process, and one queue / cart data interaction process. In Solution 2, a failure in the outbound outbound scheduling process will affect the transportation of materials from all workshops to downstream processes, disrupting production continuity. In Solution 1, a failure in the outbound outbound scheduling process of a particular workshop only affects the transportation of materials from that workshop to downstream processes, having a smaller impact on production. Therefore, Solution 1 is the preferred option.

[0119] Example 3: This example further illustrates the process of global optimization and adjustment by the central coordinating agent. In this example, the process manufacturing plant B has N=5 processes. Under normal operating conditions, 20 vehicles are deployed, with an additional 5 vehicles for maneuver. One central coordinating agent and five process agents are deployed. Process agent A1 schedules vehicles traveling between processes 1 and 2; process agent A2 schedules vehicles traveling between processes 2 and 3; process agent A3 schedules vehicles traveling between processes 3 and 4; and process agent A4 schedules vehicles traveling between processes 4 and 5. Figure 7 The following is an example of the timing relationship of a global optimization adjustment performed by the central coordinating agent in this case:

[0120] At time T0: The system starts up, and the initial allocation scheme for the number of dispatching cars A1, A2, A3, and A4 is [5,5,5,5].

[0121] T0+10 minutes: First global optimization, low efficiency of process 2 was detected, adjusted to [5,6,4,5];

[0122] T0+20 minutes: Second global optimization, material backlog occurred in process 1, adjusted to [6,5,4,5];

[0123] T0+26 minutes: A sudden bottleneck was detected in process 2. Two mobile trolleys were urgently deployed. The temporary solution is [6,7,4,5].

[0124] T0+30 minutes: Third global optimization, bottleneck alleviated, 1 motor vehicle recycled, adjusted to [6,6,4,5];

[0125] T0+40 minutes: Fourth global optimization, bottleneck removed, 1 motor vehicle recycled, adjusted to [6,5,4,5].

[0126] Through dynamic adjustments, the overall efficiency of the system was improved by 12%, and the average waiting time for materials was reduced by 15%.

[0127] Example 4: This example further illustrates the process of anomaly handling and collaborative adjustment performed by the central coordinating agent and the process agents. See also... Figure 8 The key steps in this process are as follows:

[0128] S801: Each process's intelligent agent continuously monitors the status of its respective process;

[0129] S802: Determine if an anomaly has occurred. Anomaly types include, but are not limited to: Type A: Exit queue is too long (e.g., Q_out>18); Type B: Cart status is abnormal (e.g., the car stops for more than 1 minute while driving); Type C: Equipment failure (e.g., the agent loses connection); Type D: Processing efficiency is abnormal (e.g., deviation from the mean>3σ).

[0130] If no abnormalities are found, return to S801 for continued monitoring;

[0131] If an anomaly is detected, execute S803;

[0132] S803: Send alarm information to the central coordinating agent, including but not limited to the type, location, and severity level of the anomaly;

[0133] S804: The central coordinating agent assesses the scope of the anomaly's impact, and preferably, also predicts the duration of the anomaly and assesses the losses caused by the anomaly;

[0134] S805: The central coordinating agent determines whether coordinated adjustments are needed.

[0135] If the anomaly only affects a single process, execute S806 for local adjustment; otherwise, execute S807 for global coordinated adjustment.

[0136] S806: Local adjustments (limited to a single process), such as temporarily adding a motorized trolley or adjusting the loading quantity;

[0137] S807: Global coordinated adjustment, including: adjusting upstream processes, such as reducing the output rate; adjusting this process, such as deploying mobile resources and optimizing internal scheduling; adjusting downstream processes, such as increasing buffer capacity and adjusting the receiving strategy.

[0138] S808: Execute the adjustment plan: Issue instructions to the process agent, allocate resources, and monitor the execution effect;

[0139] S809: Determine whether the exception has been resolved. If not, return to S804 for re-evaluation. If resolved, execute S810.

[0140] S810: Restore normal operating mode and recover motor resources.

[0141] Example 5: In this example, the number of processes N is 5, the total number of VGA carts is 23, and the global optimization cycle T_globe of the central coordinating agent is 1 hour. The following is a list of the first complete scheduling cycle after the system starts up:

[0142] At 08:00:00, the system starts: five process agents A1, A2, A3, A4, and A5 are initialized, along with one central coordinating agent CA; 20 cars are allocated in the order [5,5,5,5] between processes 1-2, 2-3, 3-4, and 4-5, with an additional 3 cars in operation; all workshop queues are cleared, and production begins.

[0143] At 08:05:00, the first batch of materials is produced: 8 materials have accumulated in the exit queue of workshop 2 of process 1; after A1 calculates the priority, it dispatches trolley V3 to load 8 materials and go to workshop 1 of process 2;

[0144] At 08:07:30, V3 arrives at its destination: V3 unloads at Workshop 1 of Process 2, A1 calculates the return path, and V3 returns to Workshop 3 of Process 1;

[0145] At 08:30:00, CA detected a bottleneck: the exit queue of workshop 2 in process 2 reached 16 materials, the bottleneck index B(2) = 0.85 > 0.8, and CA triggered a collaborative adjustment;

[0146] At 08:30:10, the following coordinated adjustments were performed: 1 mobile trolley was deployed to process 2, the inlet buffer capacity of A3 was increased, and the output rate of A1 was reduced by 15%.

[0147] At 08:45:00, the bottleneck was alleviated: the number of materials in the exit queue of process 2 decreased to 10, the motorized trolley was withdrawn, and the normal production rate was restored.

[0148] Global optimization at 09:00:00: Adjust the car allocation to [6,5,4,5].

[0149] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A material scheduling system for a process manufacturing plant based on intelligent agents, characterized in that: This includes a central coordinating intelligent agent, multiple process intelligent agents, a video subsystem, and a vehicle subsystem. The number of process agents is N, which correspond to the N processes of the process manufacturing. The process agent of the nth process is responsible for managing the M(n) workshops of the process and is used to control the scheduling of vehicles, path planning and return decision within the process segment. The central coordinating agent is connected to multiple process agents to optimize the allocation of the vehicle among the processes from a global perspective, coordinate the flow of materials among multiple processes, and handle system-level abnormal events. The video subsystem includes camera devices deployed at the entrances and exits of each workshop, used to identify the length of the material queue, the unloading status of the trolley, or the loading status in real time. The vehicle subsystem includes S unmanned transport vehicles and their control units. Each vehicle is equipped with an onboard camera and a communication module for receiving material transport instructions, providing feedback on path status information, and reporting abnormal information.

2. The agent-based material scheduling system for process manufacturing plants as described in claim 1, characterized in that: The process intelligence agent includes interconnected vehicle scheduling, loading decision, path decision, return decision, video analysis, and communication interface modules. The vehicle scheduling module calculates vehicle scheduling priorities and generates scheduling schemes based on the exit material queue status of each workshop in the current process and the entrance queue status of each workshop in the downstream process. The loading decision module determines the quantity of goods loaded by the vehicle. The path decision module selects the optimal path for the vehicle from the current workshop to the downstream workshop. The return decision module selects the optimal return target workshop based on the real-time queue status of each workshop in the current process after the vehicle has finished unloading in the downstream workshop. The video analysis module obtains queue length information for each workshop by performing target detection on the queue videos at workshop entrances and exits, and analyzes and identifies the status of the unloading and loading vehicles. The communication interface module interacts with the central coordinating intelligence agent.

3. The material scheduling method and system for a process manufacturing plant based on intelligent agents as described in claim 1, characterized in that: The central coordinating intelligent agent includes a globally optimized module, a bottleneck detection module, a collaborative scheduling module, an emergency response module, and a fault takeover module, all interconnected. The globally optimized module establishes a global multi-objective optimization model and dynamically adjusts the number of vehicles allocated to each process segment. The bottleneck detection module monitors the material flow status of each process in real time and identifies production bottlenecks. The collaborative scheduling module coordinates relevant processes to adjust their scheduling strategies when an anomaly occurs in a process. The emergency response module manages mobile vehicle resources and deploys them temporarily in emergencies. The fault takeover module temporarily takes over the scheduling function of a process agent when it fails. The fault takeover module interacts with the intelligent agents of each process.

4. A material scheduling method for a process manufacturing plant based on intelligent agents, characterized in that: Includes the following steps: S1. System initialization: Establish a hierarchical control architecture containing N process agents and 1 central coordinating agent, and set the initial number of vehicles allocated to each process. S2. Real-time status monitoring: Each process agent continuously monitors the entry queue length Q_in(n,i) and exit queue length Q_out(n,i) of each workshop in its process by analyzing the video stream from the video subsystem in real time, where n is the process number and i is the workshop number; the path congestion coefficient c_cong is updated by the congestion events reported by the vehicle-mounted camera. S3, Car dispatching decision: The process agent calculates the priority of each transportation task according to the priority function; S4. Loading quantity decision: Based on the queue capacity of the downstream workshop and the current quantity of exported materials, determine the optimal loading quantity p*=min{P(n),Q_out(n,i),Q_max-Q_in(n+1,j)}, where P(n) is the maximum loading capacity of the trolley in the nth process segment; S5. Path Decision: The process agent uses distance and congestion system as metrics and uses the Dijkstra algorithm to calculate the shortest path as the selected transportation path for the vehicle. S6. Return Path Decision: After the trolley completes unloading, the process agent calculates and returns the cost function to each workshop. S7. Global Optimization and Adjustment: The central coordinating agent performs global optimization on a periodic T_globe basis, adjusting the vehicle allocation scheme according to the operating efficiency and queue status of each process. S8. Anomaly Handling and Coordinated Adjustment: When an anomaly is detected, a coordinated adjustment mechanism is triggered. The central coordinating agent coordinates the relevant processes to adjust the scheduling strategy and deploys mobile vehicles when necessary.

5. The agent-based material scheduling method for process manufacturing plants as described in claim 4, characterized in that: The priority function in step S3 is: P(n,i,j)=α·Q_out(n,i)+β·(Q_max-Q_in(n+1,j))+γ·exp(-λd(n,i,j)) Where Q_out(n,i) is the length of the exit queue of the i-th workshop in the n-th process, Q_in(n+1,j) is the length of the entrance queue of the j-th workshop in the (n+1)-th process, Q_max is the upper limit of the queue capacity, d(n,i,j) is the distance from workshop (n,i) to workshop (n+1,j), and α, β, γ, and λ are weight parameters.

6. The agent-based material scheduling method for process manufacturing plants as described in claim 4, characterized in that: The specific method for the process agent to calculate and return the cost function for each workshop in step S6 is as follows: C_return(i) = d(current,i)·[1+Num_to(i) / Num_v(n)] / [1+Q_out(n,i) / Q_out_avg], where d(current,i) is the distance from the current position to workshop i, Q_out_avg is the average length of the exit queue of each workshop, Num_to(i) is the number of cars heading to workshop i, and Num_v(n) is the total number of cars in the nth process segment; the workshop with the lowest cost is selected as the return target.

7. The agent-based material scheduling method for process manufacturing plants as described in claim 4, characterized in that: The specific method for adjusting the car allocation scheme in step S7 is as follows: s.t. Num_v(n)≥MinNum_v(n), and Where T_wait(n) is the average material waiting time for process n, T_idle(k) is the idle time of trolley k, Var(Q_out) is the variance of the length of the exit queue for each workshop, Num_v(n) is the number of trolleys allocated to process n, MinNum_v(n) is the minimum number of trolleys allocated to process n, and S is the total number of trolleys.

8. The agent-based material scheduling method for process manufacturing plants as described in claim 4, characterized in that: In step S8, abnormal situations include, but are not limited to: the length of the exit queue of a certain workshop exceeds the threshold θ1; the stopping time of a certain car exceeds the threshold θ2; the communication interruption of a certain process agent exceeds the threshold θ3; the average processing time of a certain process deviates from the normal value by more than 3 times the standard deviation.

9. A material scheduling method for a process manufacturing plant based on intelligent agents as described in claim 4 or 8, characterized in that: Step S8 specifically includes the following steps: S801: Each process's intelligent agent continuously monitors the status of its respective process; S802: Determine if an anomaly has occurred. Anomaly types include, but are not limited to: Type A: Exit queue too long; Type B: Trolley status abnormality; Type C: Equipment failure; Type D: Processing efficiency abnormality. If no abnormalities are found, return to S801 for continued monitoring; If an anomaly is detected, execute S803; S803: Send alarm information to the central coordinating agent, including but not limited to the type, location, and severity level of the anomaly; S804: The central coordinating agent assesses the scope of the anomaly's impact, and preferably, also predicts the duration of the anomaly and assesses the losses caused by the anomaly; S805: The central coordinating agent determines whether coordinated adjustments are needed. If the anomaly only affects a single process, execute S806 for local adjustment; otherwise, execute S807 for global coordinated adjustment. S806: Partial adjustments, such as temporarily adding motorized trolleys or adjusting the loading quantity; S807: Global coordinated adjustment, including: adjusting upstream processes, such as reducing the output rate; adjusting this process, such as deploying mobile resources and optimizing internal scheduling; adjusting downstream processes, such as increasing buffer capacity and adjusting the receiving strategy. S808: Execute the adjustment plan: Issue instructions to the process agent, allocate resources, and monitor the execution effect; S809: Determine whether the exception has been resolved. If not, return to S804 for re-evaluation. If resolved, execute S810. S810: Restore normal operating mode and recover motor resources.

Images