Search

CN-121981469-A - Flexible production-oriented AI scheduling optimization method

CN121981469ACN 121981469 ACN121981469 ACN 121981469ACN-121981469-A

Abstract

The invention discloses an AI scheduling optimization method for flexible production, which belongs to the technical field of data processing and system management, and comprises the steps of collecting the requirement attribute of process nodes in a manufacturing system, the state attribute of resource nodes and the topological constraint relation in real time, constructing a dynamic relation diagram of the manufacturing system for dynamic value evaluation, calculating and generating dynamic value evaluation parameters to simulate virtual bidding of multiple process nodes on the resource nodes and the bidding pressure diffusion process of the resource nodes on the dynamic relation diagram of the manufacturing system, screening and outputting a simulated balanced configuration scheme to convert the resource scheduling instruction, transmitting the resource scheduling instruction to the manufacturing system for execution and collecting actual production performance data, and updating the dynamic value evaluation process. By constructing a dynamic relation diagram of a manufacturing system, performing value evaluation based on diagram attention and multiple virtual bidding, and performing model updating by combining closed-loop feedback, the scheduling configuration of dynamic, self-adaptive and global optimization of production resources can be realized.

Inventors

  • SHEN XIANGJIN

Assignees

  • 江苏云机汇软件科技有限公司

Dates

Publication Date
20260505
Application Date
20260123

Claims (10)

  1. 1. An AI scheduling optimization method for flexible production, which is characterized by comprising the following steps: acquiring the demand attribute of a process node and the state attribute of a resource node in a manufacturing system in real time, acquiring the topological constraint relation among the nodes, and constructing a dynamic relation diagram of the manufacturing system; Carrying out dynamic value evaluation based on a graph attention mechanism on the manufacturing system dynamic relation graph, and calculating and generating dynamic value evaluation parameters reflecting each resource node to each process node under the current system state; Simulating virtual bidding of a plurality of process nodes to resource nodes and bidding pressure diffusion process of the resource nodes on the dynamic relation diagram of the manufacturing system by taking the dynamic value evaluation parameter as a virtual bidding initial price signal, screening and outputting a simulated balanced configuration scheme; Converting the simulated balanced configuration scheme into an actually executable resource scheduling instruction, and sending the actually executable resource scheduling instruction to a manufacturing system for execution; Monitoring the production process after executing the resource scheduling instruction, collecting actual production performance data, comparing the expected performance of the simulated balanced configuration scheme, and generating a configuration efficiency signal; And updating the dynamic value evaluation process by using the configuration efficiency signal.
  2. 2. The flexible production-oriented AI scheduling optimization method of claim 1, wherein constructing the manufacturing system dynamic relationship graph comprises: Acquiring the requirement attribute of a process node in a manufacturing system in real time, wherein the requirement attribute at least comprises a process type, an emergency degree and a quality requirement; Synchronously acquiring state attributes of resource nodes in a manufacturing system in real time, wherein the state attributes at least comprise resource types, idle states, health degrees and geographic positions; calculating dynamic association strength parameters between the process nodes and between the resource nodes based on the requirement attribute of the process nodes and the state attribute of the resource nodes and in combination with a preset process route, wherein the dynamic association strength parameters represent the connectivity and the adaptation degree between the nodes; And taking the process nodes and the resource nodes as graph vertexes, fusing the dynamic association strength parameters with the physical connection relationship to generate graph edges describing the topological constraint relationship among the nodes, and fusing all the graph vertexes and the graph edges to generate a manufacturing system dynamic relationship graph.
  3. 3. The flexible production-oriented AI scheduling optimization method of claim 2, wherein the calculating to generate the dynamic value evaluation parameter reflecting each resource node for each process node in the current system state includes: extracting original attribute characteristics of each node in the dynamic relation diagram of the manufacturing system, and calculating attention weight coefficients among the nodes according to the original attribute characteristics of adjacent nodes and the dynamic association strength parameters; According to the attention weight coefficient, carrying out weighted aggregation on the original attribute characteristics of each node neighbor to generate a context enhancement characteristic vector of each node fused with local topology information; and carrying out feature integration and mapping by utilizing the context enhanced feature vector to obtain a two-dimensional value matrix, wherein each element in the two-dimensional value matrix is the dynamic value evaluation parameter of the corresponding resource node to the corresponding process node.
  4. 4. The flexible production-oriented AI scheduling optimization method of claim 3, wherein the screening and outputting the simulated equalization configuration scheme comprises: Allocating virtual budget for each process node, calculating initial bidding bids of each process node on all feasible resource nodes by taking the dynamic value evaluation parameters as initial price signals, and generating initial bidding pressure parameters; Loading the initial bidding pressure parameters onto corresponding resource nodes, performing multi-round diffusion along the graph edges in the dynamic relation graph of the manufacturing system, updating price signals by the resource nodes according to the received bidding pressure parameters in each round of diffusion, and re-calculating bidding bids by process nodes according to the updated price signals to iterate the process; When the price signal change rate of all the resource nodes is lower than a preset price change threshold value and the bid change rate of all the process nodes is lower than a preset bid change threshold value, judging that the process nodes reach a stable state, and decoding according to the matching relation between the final bid and the price signal to generate a simulated balanced configuration scheme.
  5. 5. The flexible production-oriented AI scheduling optimization method of claim 4, wherein the resource node updating the price signal based on the received bid pressure parameter comprises: monitoring convergence speed indexes and price fluctuation indexes of each round in the spreading process of price signals and the bidding pressure parameters in real time, and generating process dynamic characteristics; Comparing and analyzing the process dynamic characteristics with historical multi-round bidding data, and calculating real-time adjustment quantity of bidding step length parameters and diffusion attenuation coefficients based on analysis results; And updating the price signal by applying the real-time adjustment before the next round or the next scheduling period of bidding simulation starts.
  6. 6. The flexible production-oriented AI schedule optimization method of claim 4, wherein converting the simulated equalization configuration scheme into an actual executable resource schedule instruction, and sending to a manufacturing system for execution comprises: analyzing the simulated balanced configuration scheme, extracting matching pairs of process nodes and resource nodes and planning time information, and generating a preliminary scheduling sequence; Checking and time-adjusting the preliminary scheduling sequence and the real-time state of the resources in the manufacturing system, eliminating gap conflict, and generating a scheduling instruction sequence; and issuing the scheduling instruction sequence to a corresponding physical resource controller for execution, and completing resource configuration.
  7. 7. The flexible production-oriented AI schedule optimization method of claim 6, wherein generating the configuration efficiency signal comprises: During and after the execution of the resource allocation, acquiring actual production performance data in real time, wherein the actual production performance data at least comprises actual working procedure finishing time, actual resource utilization rate and product qualification rate; based on the simulated balanced configuration scheme, simulating and calculating the expected process completion time and the expected resource utilization rate under the preset execution condition to be used as expected performance; calculating a difference index between the actual production performance data and the expected performance, and normalizing the difference index to generate a configuration efficiency signal representing the deviation between the actual effect and the simulation effect of the current resource configuration.
  8. 8. The flexible production-oriented AI scheduling optimization method of claim 7, wherein updating the dynamic value evaluation process with the configuration efficiency signal comprises: correlating the configuration efficiency signal with the dynamic value evaluation parameter output by the scheduling period, and constructing a loss function for measuring the value evaluation accuracy; and updating the dynamic value evaluation process by using the loss function.
  9. 9. The flexible production-oriented AI scheduling optimization method of claim 8, further comprising: collecting an emergency signal of a resource node and a progress deviation signal of a process node in a manufacturing system in real time, and generating a rescheduling triggering instruction when the emergency signal exceeds a preset event threshold or the progress deviation signal exceeds a preset deviation threshold; Responding to the rescheduling triggering instruction, and carrying out dynamic value evaluation and virtual bidding based on a graph attention mechanism by utilizing a current manufacturing system dynamic relation graph and an affected but unfinished process node set to generate a local rescheduling scheme; And fusing and conflict resolution is carried out on the local rescheduling scheme and the original scheduling instruction which is being executed, and an updated resource scheduling instruction is generated and issued.
  10. 10. An AI-scheduling optimization system for flexible production, the system comprising: The map construction module is used for acquiring the demand attribute of the process node and the state attribute of the resource node in the manufacturing system in real time, acquiring the topological constraint relation among the nodes and constructing a dynamic relation map of the manufacturing system; The value evaluation module is used for carrying out dynamic value evaluation based on a graph attention mechanism on the manufacturing system dynamic relation graph, and calculating and generating dynamic value evaluation parameters reflecting each resource node to each process node under the current system state; The bargained decision module is used for taking the dynamic value evaluation parameter as a virtual bidding initial price signal, simulating virtual bidding of a plurality of process nodes on a resource node and bidding pressure diffusion process of the resource node on the manufacturing system dynamic relation diagram, screening and outputting a simulated balanced configuration scheme; the instruction generation module is used for converting the simulated balanced configuration scheme into an actual executable resource scheduling instruction and sending the actual executable resource scheduling instruction to a manufacturing system for execution; The feedback monitoring module is used for monitoring the production process after executing the resource scheduling instruction, collecting actual production performance data, comparing the expected performance of the simulated balanced configuration scheme and generating a configuration efficiency signal; and the closed-loop optimization module is used for updating the dynamic value evaluation process by utilizing the configuration efficiency signal.

Description

Flexible production-oriented AI scheduling optimization method Technical Field The invention relates to the technical field of data processing and system management, in particular to an AI scheduling optimization method for flexible production. Background Flexible production scheduling is a core management and administration task in modern manufacturing industry, and is essentially to reasonably allocate time and space for limited production resources such as machine tools, robots, material handling equipment and the like according to order demands, process constraints and real-time changing production environments. The execution efficiency of the task is directly related to the production cost, the delivery period, the product quality and the market competitiveness of enterprises, so that the efficient and intelligent production scheduling system is a key management tool for realizing lean and intelligent manufacturing. Currently, manufacturing enterprises mainly rely on various technical means when performing production scheduling. Some enterprises still adopt a manual scheduling mode based on personal experience of a dispatcher, and tasks are manually distributed through tools such as Gantt charts. Another part of enterprises adopts a scheduling system based on preset rules, for example, the execution sequence of the working procedures is determined according to fixed scheduling rules such as first come first served, shortest processing time priority or critical ratio. In addition, there are methods that use conventional operation study optimization algorithms, such as genetic algorithm, simulated annealing, integer programming, etc., to solve the scheduling problem offline by creating a mathematical model. However, the above-described existing scheduling techniques all exhibit certain limitations in dealing with highly complex and dynamic flexible production environments. The manual scheduling mode has strong subjectivity and low efficiency, and is difficult to deal with large-scale and multi-constraint scheduling scenes. The scheduling system based on the fixed rule is simple and quick, but the decision basis is often limited to local information, and the lack of global field of view easily causes uneven resource allocation and production bottleneck. While the traditional centralized optimization algorithm can theoretically seek a global optimal solution, the traditional centralized optimization algorithm generally depends on a static and idealized system model, has slow response to real-time dynamic changes of a production site, faces a combined explosion problem when the problem scale is increased, is excessively long in calculation time, and is difficult to meet the real-time requirement of production management. Disclosure of Invention In order to solve the problems, the invention provides an AI scheduling optimization method for flexible production, which can realize the scheduling configuration of dynamic, self-adaptive and global optimization of production resources by constructing a dynamic relation diagram of a manufacturing system, performing value evaluation based on diagram attention and multiple rounds of virtual bidding, and performing model updating by combining closed-loop feedback. The above object can be achieved by the following scheme: A flexible production-oriented AI scheduling optimization method comprises the steps of collecting demand attributes of process nodes and state attributes of resource nodes in a manufacturing system in real time, obtaining topological constraint relations among the nodes, constructing a manufacturing system dynamic relation diagram, carrying out dynamic value evaluation based on a graph attention mechanism on the manufacturing system dynamic relation diagram, calculating and generating dynamic value evaluation parameters reflecting each resource node to each process node in the current system state, simulating virtual bidding of the process nodes to the resource nodes and bidding pressure diffusion processes of the resource nodes on the manufacturing system dynamic relation diagram by using the dynamic value evaluation parameters as virtual bidding initial price signals, screening and outputting a simulation balanced configuration scheme, converting the simulation balanced configuration scheme into an actual executable resource scheduling instruction, sending the actual executable resource scheduling instruction to the manufacturing system for execution, monitoring the production process after executing the resource scheduling instruction, collecting actual production performance data, comparing expected performance of the simulation balanced configuration scheme, generating configuration efficiency signals, and updating the dynamic value evaluation process by using the configuration efficiency signals. The method comprises the steps of establishing a dynamic relation graph of a manufacturing system, wherein the dynamic relation graph comprises the st