A method and system for multi-objective decision making in manufacturing systems based on predictive models

Abstract

The application discloses a kind of manufacturing system multi-objective decision-making method and system based on prediction model, belong to intelligent manufacturing resource layout and planning decision-making technical field, according to enterprise production demand, space restriction and market condition construction manufacturing system multi-objective optimization problem model;Optimization target includes maximizing productivity, minimizing machine total cost and minimizing buffer total capacity;According to manufacturing system actual product processing flow, add fault and repair attribute to machine node, construct manufacturing system generalized stochastic Petri net model;Topology enhanced random forest regression prediction model is constructed;Improved NSGA-II algorithm with embedded local enhancement strategy is used to solve the manufacturing system multi-objective optimization problem model, and the Pareto frontier is obtained;Using decision output device, decision-making is carried out in the Pareto frontier according to the current production demand of enterprise, space restriction and fund situation, and the manufacturing system resource allocation scheme satisfying the current production demand is output.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent manufacturing resource layout and planning decision, and particularly relates to a manufacturing system multi-objective decision method and system based on a prediction model. BACKGROUND

[0002] Under the impetus of the global intelligent manufacturing wave, modern manufacturing is facing many challenges, including production efficiency improvement, cost control, and rapid response to market demand changes. Manufacturing enterprises need to ensure production efficiency while making intelligent analysis and decision-making to allocate resources leanly, thereby stabilizing the production process, achieving automated production, and improving the core competitiveness of enterprises.

[0003] Currently, most manufacturing enterprises still rely on traditional single-objective optimization, such as ensuring production efficiency or controlling costs. Such optimization methods often result in a single resource allocation scheme that lacks flexibility, making it difficult to respond quickly to changes in market demand, leading to production plan adjustment lag, consuming a large amount of time and cost, and possibly causing economic losses due to delayed response.

[0004] In addition, existing research and practice mostly focus on the analysis of single or single unreliable resources, such as preventive maintenance. However, there is little research on the simultaneous failure of multiple devices. In actual production processes, the sudden failure of multiple devices may occur throughout the production cycle of the manufacturing system, causing the production line to stall or collapse, and thus triggering large-scale production disruptions and severe economic losses. These problems may prevent manufacturing enterprises from achieving effective balance among delivery time, production efficiency, cost, and production stability. SUMMARY

[0005] The purpose of the present application is to overcome the problem of how to respond to changes in market demand, balance production efficiency and cost, and ensure production stability, and to provide more flexible and efficient decision support for manufacturing enterprises. A manufacturing system multi-objective decision method and system based on a prediction model are proposed.

[0006] To achieve the above purpose, the present application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a manufacturing system multi-objective decision method based on a prediction model, comprising the following steps:

[0008] A manufacturing system multi-objective optimization problem model is constructed according to enterprise production demand, space limitations, and market conditions. The optimization objectives of the manufacturing system multi-objective optimization problem model include maximizing production rate, minimizing total machine cost, and minimizing total buffer capacity.

[0009] Based on the actual product processing flow of the manufacturing system, fault and repair attributes are added to machine nodes to construct a generalized stochastic Petri net model of the manufacturing system; a topology-enhanced random forest regression prediction model is constructed based on random forest regression improvement.

[0010] An improved NSGA-II algorithm with an embedded local enhancement strategy is adopted, combined with a generalized stochastic Petri net model of the manufacturing system and a topology-enhanced random forest regression prediction model, to solve the multi-objective optimization problem model of the manufacturing system and obtain the Pareto front.

[0011] Using the decision outputter, decisions are made at the Pareto front based on the company's current production needs, space constraints, and financial situation, and the output is a manufacturing system resource allocation scheme that meets the current production needs.

[0012] Furthermore, a multi-objective optimization problem model for the manufacturing system is constructed based on the enterprise's production needs, space constraints, and market conditions, specifically as follows:

[0013]

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] in, This represents the total cost of the machine. Indicates the total capacity of the buffer. Indicates productivity. This indicates the total number of processing steps in the manufacturing system. This indicates the total number of buffers in the manufacturing system. Indicates the process, Indicates a buffer. Indicates the minimum required productivity. Indicates each process The maximum number of machines that can process in parallel. This indicates the maximum capacity of each buffer. Indicates each process The maximum number of machine types available. Indicates the machine type. Indicates machine cost, Represents positive integers. the number of intermediate variables of the machining process of the manufacturing system; the number of intermediate variables of the buffer zone of the manufacturing system.

[0021] Further, according to the actual product machining process of the manufacturing system, the machine node is added with the fault and repair attribute, and the manufacturing system generalized stochastic Petri net model is constructed, specifically:

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028] wherein, represents the Petri net, represents the initial resource configuration of the manufacturing system; is a set of libraries, is a set of raw material libraries, is a set of machine processing operation libraries, is a set of buffer zone libraries, is a set of machine state detection libraries, is a set of fault machine repair libraries; is a set of transitions, is a set of delay transitions, is a set of instantaneous transitions, represents the topological structure of the manufacturing system, represents the connection relationship between the process and the buffer zone, is a delay function.

[0029] Further, based on the random forest regression, a topological enhanced random forest regression prediction model is constructed, specifically:

[0030] The historical data and simulation data are obtained to obtain a data set, and the simulation data is generated by a hypercube sampling method; each piece of data in the data set is composed of a system feature enhanced expression input, a complete manufacturing system resource configuration scheme and a real production rate label, and the complete manufacturing system resource configuration scheme is composed of the resource configuration scheme in the initial input manufacturing system generalized stochastic Petri net model and the selected machine type scheme of each process;

[0031] splitting the dataset into a training set, a validation set and a test set;

[0032] inputting the system-level feature representation into a first random forest regression prediction model and inputting the system feature enhanced representation into a second random forest regression prediction model as a topology-enhanced random forest regression prediction model;

[0033] training the first random forest regression prediction model and the topology-enhanced random forest regression prediction model in parallel, adjusting hyperparameters using Bayesian optimization, and finding the best combination of hyperparameters on the validation set;

[0034] evaluating the performance of the first random forest regression prediction model and the topology-enhanced random forest regression prediction model on the test set using the coefficient of determination, the mean squared error and the root mean squared error as performance evaluation indicators; comparing the performance of the first random forest regression prediction model and the topology-enhanced random forest regression prediction model, selecting the random forest regression prediction model with better performance as the residual correction model, and calculating the residual between the productivity prediction value of the residual correction model and the true productivity label;

[0035] updating the label of each data in the dataset of the residual correction model from the true productivity to the residual to obtain a trained topology-enhanced random forest regression prediction model;

[0036] evaluating the performance of the trained topology-enhanced random forest regression prediction model on the test set using the coefficient of determination, the mean squared error and the root mean squared error as performance evaluation indicators; inputting the complete manufacturing system resource configuration scheme into the trained topology-enhanced random forest regression prediction model and outputting a single predicted productivity value.

[0037] Further, the system-level feature representation input and the system feature enhanced representation input are as follows:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044] wherein, represents the static feature input of the i-th process, i represents the number of parallel machines of the i-th process, represents the number of parallel machines of the i-th process, i represents the number of parallel machines of the i-th process, represents the number of parallel machines of the i-th process,i average processing time of the i-th machine, representing the i-th i average processing rate of the i-th machine, representing the i-th i probability of machine failure of the i-th machine, representing the i-th i average repair time of the i-th machine in failure state, representing the i-th i buffer capacity after the i-th machine, representing system-level feature expression input, representing the i-th i enhanced feature input of the i-th machine, machine availability, representing system feature enhanced expression input, buffer pressure index.

[0045] Further, an improved NSGA-II algorithm with embedded local enhancement strategy is adopted to solve the multi-objective optimization problem model of the manufacturing system in combination with the generalized stochastic Petri net model of the manufacturing system and the topologically enhanced random forest regression prediction model, and the Pareto front is obtained, specifically:

[0046] Step 4.1, encoding and decoding of the chromosome is performed to obtain encoding information and decoding information. In the encoding, the chromosome is divided into three parts, which respectively represent the number of machines in parallel on the process, the buffer capacity allocated to each buffer zone, and the type of machine selected on each process.

[0047] In the decoding, the elements in the chromosome are corresponded to the elements in the generalized stochastic Petri net model of the manufacturing system, wherein the type of machine selected on each process is decoded into failure rate, processing time distribution function, repair time distribution function and cost.

[0048] Step 4.2, after initializing parameters including population size, maximum number of iterations, crossover probability and mutation probability according to the scale of the manufacturing system, an initial population is randomly generated under boundary constraints. Each individual in the initial population represents the configuration of resources in the system, and the complete resource configuration scheme is obtained in combination with the known quantitative information. The decoding information is combined with the known quantitative information and the generalized stochastic Petri net model of the manufacturing system to calculate the total cost of the machine, the total capacity of the buffer zone and evaluate the production rate. The topologically enhanced random forest regression prediction model is used to predict the production rate.

[0049] The individuals in the initial population consider three fitness target values including total machine cost, total buffer capacity and production rate.

[0050] Taking into account the importance of the three fitness target values, the judgment results of the importance of the three fitness target values ​​are obtained, including: judging whether the productivity is less than the minimum productivity requirement. If so, the importance is marked as productivity > (total machine cost = total buffer capacity); otherwise, the importance is marked as (total machine cost = total buffer capacity) > productivity.

[0051] Based on the judgment results of the importance of the three fitness target values, a dynamic domination criterion is designed, and non-dominated ranking and crowding distance calculation are performed to classify individuals in the initial population.

[0052] Step 4.3: The genetic operator includes a selection phase, a crossover phase, and a mutation phase; the selection phase uses a classic multi-objective binary tournament; the crossover phase uses partially matched crossover; the mutation phase uses equal-probability selection with single-point and two-point mutation; each individual in the population consists of three gene sequences, and each sequence is independently probabilistically judged during the execution of the genetic operator and the local enhancement strategy.

[0053] The initial population is used as the parent population and merged with the population after the genetic operator is executed. After non-dominated sorting and crowding distance calculation, a new offspring population is generated.

[0054] Step 4.4: Obtain the non-dominated solution set from the new offspring population. Each solution The three-part structure representing chromosome coding; for each non-dominated solution Perform local superbox search and variable neighborhood search;

[0055] right of , , Perform independent local hyperbox searches, defining the hyperbox extent for each part, for the first... j In part, the step size of the superbox is As shown in the following formula:

[0056]

[0057] in, Indicates the first i The first non-dominated solution j All elements of the part, each element at a step size Internal adjustments will be made. For the first j Partial search step size; [] indicates a closed interval; The number of non-dominated solutions;

[0058] For each solution Sparsely sample several candidate solutions within the superbox. , forming the solution set ; wherein, k denotes the current solution 's sample number, K denotes the sample number, F denotes the feasible solution space;

[0059] A variable neighborhood search is performed on each part of the candidate solution generated by the local hyperbox search, using three equally probable neighborhood structures:

[0060] The first considers randomly selecting a gene for perturbation, the second considers selecting a contiguous segment of genes in a part and reversing their order, and the third considers moving the entire contiguous segment of genes to a randomly selected new location;

[0061] After each variable neighborhood search, a number of candidate new solutions are sparsely sampled within the hyperbox to form a new solution set ;

[0062] The non-dominated solution set , the solution set and the new solution set are combined to perform non-dominated sorting and crowdedness calculation to obtain new Pareto front individuals as high-quality individuals, and the high-quality individuals are retained to obtain a new population;

[0063] Step 4.5, judge whether the current iteration meets the termination condition, if yes, output the optimal population and a Pareto front, otherwise update the current population and the iteration number and continue iteration.

[0064] Further, using a decision output device, a manufacturing system resource configuration scheme that meets the current production demand is output according to the current production demand, space limitation and fund situation in the Pareto front, specifically:

[0065]

[0066]

[0067] wherein, denotes the Pareto front solution set, denotes the th individual in the solution set, Z denotes the number of solutions in the solution set; denotes the composition of the current individual chromosome, which is mapped to a manufacturing system resource configuration scheme, , , respectively denote the total cost of machines, the total capacity of buffers and the productivity under the current scheme ;

[0068] In According to the current production demand of the enterprise , space limitation and fund situation , a preliminary feasible Pareto frontier solution set is obtained ;

[0069]

[0070] If , according to the current solution, the critical indicators of the current minimum feasible solution are provided, and the constraint relaxation suggestion is given

[0071] If , for the solution in , a layer-by-layer screening decision method is adopted, if the importance degree of the enterprise is cost> productivity> space, then the lowest solution set Cost is screened out in ;

[0072] If , the next layer screening is continued according to the importance degree, until the three-layer screening is finished to obtain the final solution , otherwise the final solution is directly output , and the manufacturing system resource allocation scheme meeting the current production demand is obtained.

[0073] Secondly, the application provides a manufacturing system multi-objective decision system based on a prediction model, comprising:

[0074] A manufacturing system multi-objective optimization problem model module is used to construct a manufacturing system multi-objective optimization problem model according to the production demand of the enterprise, space limitation and market condition; the optimization objectives of the manufacturing system multi-objective optimization problem model include maximizing productivity, minimizing total machine cost and minimizing total buffer capacity;

[0075] A Petri net model and a prediction model construction module is used to add fault and repair attributes to the machine node according to the actual product processing flow of the manufacturing system, and construct a manufacturing system generalized stochastic Petri net model; a topological enhanced random forest regression prediction model is constructed based on random forest regression improvement;

[0076] A manufacturing system multi-objective optimization problem solving module is used to adopt an improved NSGA-II algorithm embedded with a local enhancement strategy, combine the manufacturing system generalized stochastic Petri net model and the topological enhanced random forest regression prediction model, solve the manufacturing system multi-objective optimization problem model, and obtain a Pareto frontier

[0077] The output manufacturing system resource configuration scheme module is configured to utilize the decision outputter to make a decision in the Pareto frontier according to the current production demand, the space limitation and the fund condition of the enterprise, and output a manufacturing system resource configuration scheme satisfying the current production demand.

[0078] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements a multi-objective decision-making method for a manufacturing system based on a prediction model when executing the computer program.

[0079] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program, and the computer program implements a multi-objective decision-making method for a manufacturing system based on a prediction model when executed by a processor.

[0080] Compared with the prior art, the present application has the following beneficial technical effects:

[0081] The application provides a manufacturing system multi-objective decision-making method based on a prediction model, which comprises the following steps: firstly, obtaining enterprise related production information to construct a manufacturing system multi-objective optimization problem model, and converting an actual product processing flow into a manufacturing system generalized stochastic Petri net model, which can describe the manufacturing system production and processing topological structure relationship and dynamic behavior changing with time, is helpful for systematic analysis, adds machine fault and repair attributes, and enables the model to describe and analyze inevitable machine fault behavior in the production and processing process; considering that traditional production rate single evaluation of the multi-machine fault manufacturing system is often tedious and time-consuming and state explosion, a topology-enhanced random forest regression (TERFR) prediction model is constructed to replace the traditional production rate evaluation mode, which greatly reduces the time cost within the allowable prediction error in the same processing situation; a local enhancement improved NSGA-II (Non-dominated Sorting Genetic Algorithm II) embedded with local hyperbox search and variable neighborhood search is adopted to realize the coordinated exploration of the breadth and depth in the feasible solution space and obtain the Pareto front; the decision output enables the enterprise to weigh between the quantitative indicators of different schemes, flexibly make decisions according to actual needs, and finally select a resource allocation scheme suitable for the enterprise's instant needs; for the manufacturing system multi-objective optimization problem, the efficient trade-off between the production rate, machine cost and buffer capacity of the enterprise is realized; the manufacturing system generalized stochastic Petri net is established, the potential production abnormality in the production process is considered, and the system robustness analysis is realized, so that the production demand can still be met in the case of sudden machine failure; the TERFR prediction model can replace the production rate evaluation under the same production situation within the error tolerance, ensuring the efficiency of subsequent iterative optimization; the improved NSGA-II embedded with the local enhancement strategy ensures that the effective global Pareto front is obtained in the maximum feasible solution space; through the decision output device, the enterprise can make flexible decisions according to the real-time production demand, cost budget and space condition, and obtain a resource allocation scheme suitable for instant production; even if the production demand changes, the enterprise can still quickly decide on a new scheme in the Pareto front through the decision output device, ensuring the efficient adjustment and operation of the manufacturing system. BRIEF DESCRIPTION OF DRAWINGS

[0082] The drawings described herein are only for the purpose of explanation and are not intended to limit the scope of the present application in any way. In addition, the shapes and scale sizes of the components in the drawings are only illustrative and are used to help understand the present application, and are not specific limitations on the shapes and scale sizes of the components. In the drawings:

[0083] Figure 1 The flowchart of the manufacturing system multi-objective decision-making method based on the prediction model.

[0084] Figure 2 A structural diagram of a manufacturing system multi-objective decision system based on a prediction model according to an embodiment of the present application.

[0085] Figure 3 An electronic device diagram of a manufacturing system multi-objective decision method based on a prediction model according to an embodiment of the present application.

[0086] Figure 4 A flowchart of a manufacturing system multi-objective decision method based on a prediction model according to an embodiment of the present application.

[0087] Figure 5 A structural diagram of a manufacturing system multi-objective decision system based on a prediction model according to an embodiment of the present application.

[0088] Figure 6 A flowchart of a manufacturing system multi-objective decision method based on a prediction model according to an embodiment of the present application.

[0089] Figure 7 A manufacturing system generalized stochastic Petri net model diagram according to an embodiment of the present application. DETAILED DESCRIPTION

[0090] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor should fall within the protection scope of the present application.

[0091] Embodiment one

[0092] Referring to Figure 1 A manufacturing system multi-objective decision method based on a prediction model, comprising the following steps:

[0093] According to enterprise production demand, space limitation and market condition, a manufacturing system multi-objective optimization problem model is constructed; the optimization objectives of the manufacturing system multi-objective optimization problem model include maximizing productivity, minimizing total machine cost and minimizing total buffer capacity;

[0094] According to the actual product processing flow of the manufacturing system, a fault and repair attribute is added to the machine node, and a manufacturing system generalized stochastic Petri net model is constructed; a topological enhanced random forest regression prediction model is constructed based on random forest regression improvement;

[0095] An improved NSGA-II algorithm with embedded local enhancement strategy is used to solve the multi-objective optimization problem model of the manufacturing system, combined with the generalized stochastic Petri net model of the manufacturing system and the topological enhanced random forest regression prediction model, to obtain the Pareto front;

[0096] A decision output device is used to make decisions in the Pareto front according to the current production needs, space limitations, and financial conditions of the enterprise, and output a manufacturing system resource allocation scheme that meets the current production needs.

[0097] The method of this embodiment comprehensively considers key factors of the manufacturing system, including production efficiency, cost control, and space utilization, and can effectively trade off between multiple dimensions. This multi-objective optimization is closer to the needs in actual production and manufacturing. By fusing machine failure and repair attributes to construct a generalized stochastic Petri net model of the manufacturing system, the production topology relationship and sudden failure behavior are accurately described, the depth and robustness of system behavior modeling are improved, and the production stability under sudden failure is ensured. The TERFR prediction is used to achieve efficient prediction within a controllable error range, which is embedded in the improved NSGA-II algorithm to replace the traditional production rate evaluation method in the iteration process. The algorithm further integrates a local enhancement strategy to efficiently generate a Pareto front solution set with strong convergence. The decision output device can dynamically adapt to the production of the enterprise manufacturing system, ensure the response capability to sudden demand changes, and output the resource allocation scheme after trade-off, providing a decision support method for intelligent manufacturing that combines agility and scientificity.

[0098] Embodiment Two

[0099] Referring to Figure 2 A multi-objective decision system for a manufacturing system based on a prediction model includes:

[0100] A manufacturing system multi-objective optimization problem model module is configured to construct a manufacturing system multi-objective optimization problem model based on enterprise production needs, space limitations, and market conditions. The optimization objectives of the manufacturing system multi-objective optimization problem model include maximizing production rate, minimizing total machine cost, and minimizing total buffer capacity.

[0101] A Petri net model and prediction model construction module is configured to add failure and repair attributes to machine nodes based on actual product processing flows of the manufacturing system, and construct a generalized stochastic Petri net model of the manufacturing system. A topological enhanced random forest regression prediction model is constructed based on an improved random forest regression.

[0102] A manufacturing system multi-objective optimization problem solving module is configured to use an improved NSGA-II algorithm with embedded local enhancement strategy, combined with the generalized stochastic Petri net model of the manufacturing system and the topological enhanced random forest regression prediction model, to solve the manufacturing system multi-objective optimization problem model, and obtain the Pareto front.

[0103] The output manufacturing system resource configuration scheme module is configured to utilize the decision outputter to make decisions in the Pareto frontier according to the current production demand, space limitations, and financial conditions of the enterprise, and output a manufacturing system resource configuration scheme that meets the current production demand.

[0104] The system considers multi-objective optimization of the manufacturing system, obtains information such as production demand, space limitations, and market conditions of the enterprise, quickly provides intelligent decision support for the manufacturing system of the enterprise, and outputs a resource configuration scheme that balances multiple objectives. By using TERFR prediction as a productivity evaluation method in the improved NSGA-II algorithm, the multi-objective optimization operation time is greatly reduced. The system describes a robust manufacturing system, taking into account possible machine failure situations that may occur in actual production in advance, not only meeting the production demand of the enterprise, but also ensuring stable production. The system also provides comprehensive decision support, which can make optimal decisions efficiently according to the actual demand and current conditions of the enterprise. The output resource configuration scheme can be directly applied to actual production layout. The generalized stochastic Petri net model of the manufacturing system has good graphical representation ability and scalability, and can describe manufacturing systems of different production and processing processes and scales. The system can adjust the generalized stochastic Petri net model according to the actual situation of the enterprise to adapt to new manufacturing system production and processing modes. Through fast and efficient decision support, the system can improve production stability, enhance the enterprise's ability to respond to changing production demands, help it achieve the best trade-off in multi-objective optimization, thereby reducing costs and increasing efficiency, and effectively improving the competitiveness of the enterprise.

[0105] Embodiment Three

[0106] Referring to Figure 3 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements a manufacturing system multi-objective decision-making method based on a prediction model when executing the computer program.

[0107] Embodiment Four

[0108] A computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a manufacturing system multi-objective decision-making method based on a prediction model.

[0109] Embodiment Five

[0110] Referring to Figure 4 , Figure 6 A manufacturing system multi-objective decision-making method based on a prediction model includes the following steps:

[0111] Based on information obtained from the enterprise's production needs, physical space constraints, and existing market conditions, a multi-objective optimization problem model for the manufacturing system is constructed, aiming to balance the trade-offs between maximizing productivity, minimizing total machine cost, and minimizing total buffer capacity.

[0112] Based on the actual product processing flow of the manufacturing system, a generalized stochastic Petri net model of the manufacturing system is constructed by adding fault and repair attributes to machine nodes;

[0113] Construct a topology-enhanced random forest regression (TERFR) prediction model, and output the predicted stable operating productivity after inputting a complete resource allocation scheme;

[0114] The NSGA-II algorithm, which employs an embedded local enhancement strategy, replaces the traditional performance evaluation method with TERFR prediction as its productivity evaluation method to solve the multi-objective optimization problem and obtain a Pareto front.

[0115] The design decision output device makes decisions based on the company's current production needs, space constraints, and financial situation at the Pareto front, and outputs a manufacturing system resource allocation scheme that meets the current production needs.

[0116] Information is obtained based on the company's production needs, physical space constraints, and existing market conditions, specifically:

[0117] The manufacturing system includes multiple processing steps, and different steps Use a buffer between them Establish a connection. Based on the company's production needs, obtain the minimum required productivity. Obtain each process based on physical space constraints. The maximum number of machines that can process in parallel. And the maximum capacity of each buffer. Obtain each process based on existing market conditions. Maximum number of machine types available Different machine types Determine its cost same process The machines that perform the processing operations are the same.

[0118] The manufacturing system has Processing steps, Given several buffers, the multi-objective optimization problem model is constructed as follows:

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126] where the objective functions are to minimize the total cost of machines, the total capacity of buffers and maximize the production rate respectively; the constraints are the minimum production rate requirement, the number of machines limit, the capacity of buffers limit, the number of machine types limit and the decision variable value limit.

[0127] According to the actual product processing flow, the generalized stochastic Petri net model of manufacturing system is constructed by adding the failure and repair attributes to the machine nodes as follows:

[0128]

[0129]

[0130]

[0131]

[0132]

[0133]

[0134] where, represents the Petri net; represents the initial resource configuration of manufacturing system; is the set of places, is the set of raw material places, is the set of machine processing operation places, is the set of buffer places, is the set of machine state detection places, is the set of places for repairing the failed machine;

[0135] is the set of transitions, is the set of delay transitions, usually represented by ( ); is the set of instantaneous transitions, usually represented by ( ); such as Figure 7 ;

[0136] and are the pre and post functions, respectively, describing the topology of the manufacturing system and the connection between processes and buffers;

[0137] is the delay function, which is the definition of the delay distribution of each transition; for the delay transition , is the probability distribution describing the time required for the execution of the transition, which can be any random distribution such as exponential distribution, normal distribution, etc., describing the randomness of machine processing, mechanical arm loading and unloading, and machine fault repair time; instantaneous transition is used to determine whether the machine is in a fault state without delay, is an empty function;

[0138] is the initial state identifier, indicating the initial resource configuration of the manufacturing system.

[0139] A topological enhanced random forest regression (TERFR) prediction model is constructed, which outputs the predicted stable running productivity after inputting the complete resource configuration scheme. The specific construction process is as follows:

[0140] For a manufacturing system containing processing procedures, buffers, the complete resource configuration scheme expresses a series of information, which is convenient for constructing feature input;

[0141] number of machines represents the number of parallel machines of the ith processing procedure;

[0142] processing time represents the machine processing time distribution function of the ith processing procedure, represents the average processing time of the machine of the ith processing procedure;

[0143] failure rate represents the machine failure probability (%) of the ith processing procedure;

[0144] repair time represents the machine fault repair time distribution function of the ith processing procedure, represents the average repair time of the machine in the fault state of the ith processing procedure;

[0145] buffer capacity represents the buffer capacity after the ith processing procedure;

[0146]

[0147]

[0148] represents the static feature input of the ith process, represents the system-level feature expression input;

[0149] Enhanced processing is performed on the interactive features constructed using the manufacturing system topology structure;

[0150]

[0151]

[0152] wherein, is the machine availability, directly reflecting the proportion of the machine of the ith process that can normally work in the long-term operation, for quantifying the equipment stability; is the buffer pressure index, reflecting the balance between fault impact and buffer carrying capacity, capturing the blocking risk between the ith process and the ith+1 process, represents the average processing rate of the machine of the ith process; i

[0153]

[0154]

[0155] represents the enhanced feature input of the ith process, represents the system feature enhanced expression input; A topology-enhanced random forest regression (TERFR) prediction model is obtained by improving the random forest regression (RFR) as the basis, and the specific construction, training and verification process is as follows:

[0156] The data set includes historical data and simulation data, and the specific scheme of the simulation data is generated by a hypercube sampling method to ensure uniform coverage of different sample areas in a multi-dimensional feature space. The data set is divided into a training set, a validation set and a test set, and each data in the data set is composed of a complete manufacturing system resource configuration scheme and its corresponding real productivity label

[0157] ; wherein the complete manufacturing system resource configuration scheme is composed of the resource configuration scheme input into the Petri model and the machine type scheme selected by each process.

[0158] respectively, and and ​​As input, train the first random forest regression prediction model RFR1 and the second random forest regression prediction model RFR2 two prediction models in parallel. Adjust the hyperparameters (number of trees n_estimators, maximum depth max_depth, and minimum sample split min_samples_split, etc.) using Bayesian optimization to find the best combination of hyperparameters on the validation set;

[0159] On the test set, evaluate the performance of RFR1 and RFR2 as evaluation indicators with the coefficient of determination , mean square error MSE, and root mean square error RMSE;

[0160] Train the residual correction model RFR3, select the model with better performance of RFR1 and RFR2, assume RFR1, calculate the production rate prediction value of RFR1 And the real value, that is, the real production rate label The residual Re1 between them;

[0161]

[0162] Update the label of each data in the data set of RFR3 to the residual instead of the real production rate, and the remaining training steps are the same as above;

[0163]

[0164] For the final prediction result of the topological enhanced random forest regression (TERFR) prediction model, The production prediction value of RFR1, The production prediction value of RFR3;

[0165] On the test set, evaluate the performance of the final TERFR prediction model as evaluation indicators with the coefficient of determination , mean square error MSE, and root mean square error RMSE; input the complete manufacturing system resource configuration scheme into the trained TERFR prediction model to obtain an output predicted production rate value.

[0166] The improved NSGA-II algorithm includes the following steps:

[0167] Step 4.1, encoding and decoding of chromosomes;

[0168] When encoding, the chromosome is divided into three segments, which respectively represent the number of machines in parallel on the process , the buffer capacity allocated to each buffer , and the selected machine type on each process ; as shown in Table 1:

[0169] Table 1 contains the processing sequence of the track, the decoding table of the manufacturing system with one buffer

[0170]

[0171] corresponding the elements in the chromosome to the elements in the manufacturing system generalized stochastic Petri net model at decoding, wherein the decoding is the failure rate , the processing time distribution function , the repair time distribution function and the cost ; as shown in Table 2:

[0172] Table 2 contains the processing sequence of the track, the decoding table of the manufacturing system with one buffer

[0173]

[0174] Step 4.2, initialization parameter setting, population generation, fitness evaluation and grading;

[0175] The initialization parameters consider the population size Pop, the maximum number of iterations Maxiter, the crossover probability and the mutation probability ; after designing reasonable parameters according to the size of the manufacturing system, the initial population is randomly generated under the boundary constraint, and each individual in the population represents the configuration of resources in the system. Combined with existing conditions, a complete resource configuration scheme can be obtained. The information obtained after decoding can be combined with the existing conditions and the established manufacturing system generalized stochastic Petri net model to calculate the total cost of the machine, the total capacity of the buffer and evaluate the production rate. Among them, the production rate is predicted by the TERFR prediction model. Because each individual contains the number of machines, the type of buffer, the type of machine, and the failure probability, processing time distribution function, repair time distribution function, and cost information of the currently selected machine. The existing conditions refer to the known quantitative information before selecting these variables, including the assumption that the raw materials are supplied infinitely and the time distribution function obeyed by the mechanical arm loading / unloading is known. With the existing conditions, combined with the information of each individual, a complete resource configuration scheme of the manufacturing system can be obtained.

[0176] The individuals in the population consider three fitness target values, namely the total cost of the machine , the total capacity of the buffer and the production rate ;

[0177] Considering their importance, it is judged whether < , if yes, the importance degree is marked as ( = ), otherwise, the importance degree is marked as = ) , i.e. to make the productivity reach the minimum requirement first, and then to optimize the multiple targets gradually;

[0178] According to the above judgment, the dynamic dominance rule is designed, and the subsequent fast non-dominant sorting and crowded distance calculation are performed to rank the individuals in the population;

[0179] Step 4.3, genetic operator design and execution;

[0180] The genetic operator includes three stages of selection, crossover and mutation;

[0181] The selection part adopts the classic multi-objective binary tournament; the crossover part adopts the partial matching crossover; the mutation part selects the equal probability selection of single-point and two-point mutation for disturbance;

[0182] The individuals in the population are composed of three gene sequences, each of which is independently judged in the genetic operator execution and the subsequent local enhancement strategy, to ensure the controllable search direction;

[0183] The parent population and the population after executing the genetic operator are merged to generate a new child population after non-dominant sorting and crowded distance calculation;

[0184] Step 4.4, local enhancement strategy

[0185] The non-dominant solution set is obtained from the new child population , wherein each solution represents the three-part structure of the chromosome code; for each non-dominant solution , the local hyperbox search and variable neighborhood search strategies are executed;

[0186] For , the , , independent local hyperbox search is executed, and for each part, the hyperbox range is defined; for the first j part, the step of the hyperbox is , i.e.:

[0187]

[0188] wherein represents all elements of the first i part in the first j non-dominant solution, and each element is disturbed at a step of Adjusting the inner loop, the search step of the part;

[0189]

[0190] For each solution , sparse sampling several candidate solutions in the hyper-box range , sampling ensures uniformity, and must meet the requirements of feasible solutions, forming a solution set ; Wherein, k the sampling number of the current solution , the sampling number of the current solution K , the sampling number of the current solution F , the sampling number of the current solution

[0191] Perform variable neighborhood search on each part of the candidate solution generated by local hyper-box search, and design three equally probable neighborhood structures to be executed. The first considers randomly selecting a gene for disturbance, the second considers selecting a certain part of a certain continuous gene segment, and reversing the gene sequence between them. The third considers migrating the entire continuous gene sequence to a random new location.

[0192]

[0193] After each variable neighborhood search, ensure that each new solution is a feasible solution, forming a solution set ;

[0194] Merge the non-dominated solution set , the solution set obtained after local strategy search, and to perform non-dominated sorting and congestion calculation, and retain high-quality individuals to obtain a new population; High-quality solutions are new Pareto front individuals obtained after non-dominated sorting and congestion calculation in the new solution set after merging, and historical high-quality solutions are supplemented to meet the individual quantity demand of the new population for subsequent iteration.

[0195] Step 4.5, iteration termination judgment

[0196] Determine whether the current iteration meets the termination condition. If yes, output the optimal population and a Pareto front, otherwise update the current population and iteration number and continue iteration.

[0197] The purpose of the decision output device is to output a manufacturing system resource configuration scheme that meets the current production demand, specifically:

[0198]

[0199]

[0200] Wherein, the Pareto front solution set, represents an individual in the solution set, Z represents the number of solutions in the solution set; represents the composition of the chromosome of the current individual, mapped to the manufacturing system resource configuration scheme, , , respectively represent the total cost of machines, the total capacity of buffers and the productivity under the current scheme ;

[0201] In , according to the current production demand of the enterprise , the space limit and the financial situation , a preliminary feasible scheme is screened to obtain a preliminary feasible Pareto frontier solution set ;

[0202]

[0203] If , according to the current solution, the critical indicators of the current minimum feasible solution are provided and the constraint relaxation suggestions are given;

[0204] If , for the solution in , a layer-by-layer screening decision method is adopted, if the enterprise's emphasis degree is "cost> productivity> space", the lowest solution set is first screened out in , if , the next layer screening is continued according to the emphasis degree as the standard, until the three-layer screening ends to obtain the final solution , otherwise the final solution is directly output ;

[0205] The final solution includes the manufacturing system resource configuration scheme and the corresponding productivity, total cost of machines and total capacity of buffers indicator values;

[0206] Through the decision output device, the enterprise can weigh between the quantitative indicators under different schemes, make flexible decisions, and select the final resource configuration scheme suitable for the enterprise's demand.

[0207] Referring to Figure 5 , a manufacturing system multi-objective decision system based on a prediction model, comprising:

[0208] An information acquisition and problem establishment module is used to acquire information according to the production demand of the enterprise, the space limit and the existing market conditions; a manufacturing system multi-objective optimization problem model is constructed, aiming to balance between maximizing productivity, minimizing total cost of machines and minimizing total capacity of buffers;

[0209] The module for building a generalized stochastic Petri net model for manufacturing systems is used to construct a generalized stochastic Petri net model for manufacturing systems by adding fault and repair attributes to machine nodes based on the actual product processing flow.

[0210] The prediction model building module is used to build a topology-enhanced random forest regression (TERFR) prediction model. After inputting a complete resource allocation scheme, it outputs the predicted stable operating productivity.

[0211] The multi-objective optimization module is used to solve the multi-objective optimization problem by replacing the traditional performance evaluation method with TERFR prediction using the improved NSGA-II algorithm, and obtain a Pareto front.

[0212] The decision output module is used to design a decision outputter that makes decisions based on the company's current production needs, space constraints, and financial situation in the Pareto front, and outputs a manufacturing system resource allocation scheme that meets the current production needs.

[0213] The following example, a manufacturing system comprising five processing steps and four buffer zones, further illustrates a multi-objective decision-making method for manufacturing systems based on a predictive model. Figure 7 The main steps include:

[0214] Step 1: Construct a multi-objective optimization problem model based on the enterprise's production needs, space constraints, and market conditions;

[0215] In this example, see Figure 7 The manufacturing system comprises 5 processing steps and 4 buffer zones. It determines the minimum required productivity based on the company's production needs. =0.02 pieces / second. Obtaining each process based on physical space constraints. The maximum number of machines that can process in parallel. And the maximum capacity of each buffer. , specific =3, i=1,...,5, =5, j=1,2,3,4. Each process The machines on the machines are of the same type, but the available machine types differ. Each process can be obtained based on current market conditions. Maximum number of machine types available =[1,2,2,1,2], for different machine types The specific parameters differ, and are expressed as follows: .

[0216] in Indicates cost, This represents the current machine processing time distribution function. Machine failure probability (%), Machine repair time distribution function after machine failure. In this example, the time distribution function is exponential distribution function, for convenience, the average processing time (sec) and the average repair time (sec) are used to represent the specific parameters of the machine, i.e. the parameters of the machine type are .

[0217] The specific parameters of the machine type available for each process are given. For , the specific parameters are [9, 20, 3, 500]; for , [4, 25, 3, 250], [6, 22, 3, 150]; for , [7, 30, 5, 200], [10, 20, 4, 150]; for , [10, 9, 3, 200]; for , [5, 20, 3, 200], [9, 13, 1, 140].

[0218] The multi-objective optimization problem model is constructed as:

[0219]

[0220]

[0221]

[0222]

[0223]

[0224]

[0225]

[0226] The constraints that the multi-objective optimization needs to meet are the minimum production rate requirement of the enterprise production , the machine number limit of each process , the buffer capacity limit , the available machine type limit , and the decision variable value limit, aiming to maximize the production rate , minimize the total machine cost , and minimize the total buffer capacity , and balance between them.

[0227] Step two: construct a generalized stochastic Petri net model according to the product processing flow of the manufacturing system;

[0228] The manufacturing system and the actual product processing flow described in step one are used to build a manufacturing system generalized stochastic Petri net model with machine failure and repair attributes , see Figure 7 .

[0229] wherein, represents a Petri net; represents an initial state identifier, mapped to an initial manufacturing system resource configuration scheme.

[0230] is a set of places. For example, Figure 7 , is a set of raw material places, wherein the token number represents the number of raw materials; is a set of machine processing operation places, wherein , , respectively represent the idle machine place of the ith process, the machine place being processed, and the finished product unloading place; is a set of buffer places, wherein , represents the remaining buffer place and the used buffer place of the jth buffer; is a set of machine state detection places, represents the detection place set of the ith process machine in normal working or fault state; is a set of fault machine repair places, represents the repair place set of the ith process machine in fault state.

[0231] is a set of transitions. is a set of delay transitions, usually represented by ( ); is a set of instantaneous transitions, usually represented by ( ); for example Figure 7 , the transition of each process has similar meaning, taking the first process as an example, is a delay transition with a certain time meaning, representing the time for the mechanical arm to transport the raw material to the idle machine, the time for the machine to process the raw material into a semi-finished product, the time for the mechanical arm to transport the semi-finished product to the idle buffer, and the time for the machine to be repaired from fault to good; is an instantaneous transition without time meaning, which occurs instantaneously but has a certain probability of triggering, representing the machine in normal working state with a high probability and the machine in fault state with a low probability. Therefore, in the system is a mechanical arm loading transition, with a time meaning of the time for the mechanical arm to load the raw material or semi-finished product onto the machine; is a machine processing transition, whose time meaning is the time of processing raw materials or semi-finished products by machines; is a robot unloading transition, whose time meaning is the time of unloading semi-finished products or finished products by robots to buffer zones; is a machine maintenance transition, whose time meaning is the time required for maintenance when the machine fails; indicates that the current machine is probably working normally; indicates that the current machine has a very small probability of failure.

[0232] indicates the connection relationship between the warehouse and the transition, which is mapped to the manufacturing system production process. For example, indicates that the current state is that the machine is processing raw materials, and after a certain processing time, it reaches the processing completion state waiting for semi-finished product unloading. is a delay function, which gives each delay transition a certain time meaning by obtaining the selected machine type parameters. For example, the time distribution function of the above step one delay transition adopts an exponential distribution function.

[0233] , where the element represents the number of tokens in the current warehouse, which is mapped to the number of raw materials, the number of machines, and the capacity of the buffer zone represented by the relevant warehouse.

[0234] Step three: build a topological enhanced random forest regression (TERFR) prediction model;

[0235] The data set in this embodiment contains 2050 groups of historical data and simulation data, and the simulation data is generated by a hypercube sampling method, and is simulated in a generalized random Petri net simulation software to ensure uniform coverage of different sample areas in a multi-dimensional feature space;

[0236] The data set is divided into a training set, a validation set and a test set in the proportions of 70%, 15% and 15%. Each piece of data in the data set is composed of a complete manufacturing system resource configuration scheme and its corresponding real productivity label ; wherein the complete manufacturing system resource configuration scheme is composed of the resource configuration scheme initially input into the Petri model and the machine type scheme selected for each process. According to the data, the following contents are obtained and constructed:

[0237]

[0238]

[0239]

[0240] ​

[0241]

[0242]

[0243] wherein, represents the number of parallel machines of the ith process; represents the average processing time (seconds) of the ith process machine; represents the average processing rate of the ith process machine; i represents the average processing rate of the ith process machine; represents the machine failure probability (%) of the ith process; represents the average repair time (seconds) of the ith process machine under failure state; represents the buffer capacity after the ith process; is the machine availability, directly reflecting the proportion of the ith process machine that can work normally in long-term operation, and is used to quantify the equipment stability; is the buffer pressure index, reflecting the balance between failure impact and buffer carrying capacity, and capturing the congestion risk between the ith process and the ith+1 process;

[0244] represents the static feature input of the ith process, represents the system-level feature expression input; represents the enhanced feature input of the ith process, represents the system feature enhancement expression input;

[0245] and and are taken as inputs, respectively, and two prediction models RFR1 and RFR2 are trained in parallel. The hyperparameter search space is defined, specifically the number of trees n_estimators [50, 300], the maximum depth max_depth [5, 30], the minimum sample split min_samples_split [2, 20], and the minimum sample leaf node min_samples_leaf [1, 10]. Bayesian optimization is used to adjust the hyperparameters on the validation set to find the best combination. For RFR1, the final hyperparameter results {n_estimators, max_depth, min_samples_split, min_samples_leaf} are {220, 18, 8, 2}; for RFR2, the results are {185, 22, 5, 1}.

[0246] The performance of the two prediction models RFR1 and RFR2 is evaluated on the test set to determine the coefficient of determination R-squared (R2) and the mean absolute error (MAE). , mean square error (MSE) and root mean square error (RMSE) as evaluation indexes. For RFR1, the results are = 0.81, MSE = 15.21, RMSE = 4.75; for RFR2, the results are = 0.86, MSE = 8.72, RMSE = 3.32. The comprehensive analysis selects the prediction results of RFR2 The residual error is generated by subtracting the true production rate from the predicted production rate of RFR3 , and the training process is the same as above, and the prediction results are .

[0247] The final prediction results of the TERFR prediction model. The performance index evaluation results on the test set are = 0.89, MSE = 5.53, RMSE = 3.47. When facing the complex manufacturing system environment, the traditional production rate evaluation only evaluates the single scheme, and quickly obtains an output prediction production rate value.

[0248] Step four: adopt the improved NSGA-II algorithm with embedded local enhancement strategy and take TERFR prediction as its production rate evaluation method to solve the multi-objective problem and obtain the Pareto frontier solution set;

[0249] When encoding, the chromosome is divided into three segments, which respectively represent the number of machines in parallel on the process , the buffer capacity allocated to each buffer and the selected machine type on each process ; as shown in Table 3:

[0250] Table 3 Encoding table of manufacturing system containing 5 processing procedures and 4 buffers

[0251]

[0252] When decoding, the elements in the chromosome are corresponding to the elements in the generalized stochastic Petri net model of the manufacturing system, wherein the decoding is the failure rate , the processing time distribution function , the repair time distribution function and the cost ; as shown in Table 4:

[0253] Table 4 Decoding table of manufacturing system containing 5 processing procedures and 4 buffers

[0254]

[0255] The initialization parameters are designed as follows: population size Pop = 50, maximum number of iterations Maxiter = 300, and crossover probability... The value is 0.85 and the mutation probability. It is 0.1.

[0256] Individuals in the population consider three fitness target values, namely, total machine cost. Total buffer capacity and productivity The encoded and decoded information corresponds to the complete manufacturing system resource configuration scheme in step three above, therefore it can be directly calculated. and Predict using a pre-trained TERFR prediction model .

[0257] A dynamic dominance criterion is designed to classify individuals based on their non-dominated ranking and crowding distance calculations using three fitness target values.

[0258] The genetic operator consists of three stages: selection, crossover, and mutation. The selection part adopts a multi-objective binary tournament; the crossover part adopts partial matching crossover; the mutation part adopts equal probability selection with single-point and two-point mutation; the individuals in the population are referred to in the encoding and decoding part, which consists of three gene sequences. Each sequence is independently probabilistically judged during the execution of the genetic operator and in the subsequent local enhancement strategy to ensure that the search direction is controllable.

[0259] The parent population is merged with the population after the genetic operator is executed. After non-dominated sorting and crowding distance calculation, a new offspring population is generated by elimination based on the population size.

[0260] Local enhancement strategies focus on non-dominated solution sets in the new offspring population. Each solution The three-part structure representing chromosome coding; for each non-dominated solution The strategy consists of two parts: local superbox search and variable neighborhood search.

[0261] right of , , Perform independent local hyperbox searches, defining the hyperbox extent for each part; for the first part... j In part, the step size of the superbox is ,Right now:

[0262]

[0263] in, Indicates the first i The first non-dominated solutionj All elements of the part are adjusted respectively in step , The search step of the part;

[0264]

[0265] For each solution , sparse sampling several candidate solutions in the hyper-box range , sampling ensures uniformity, and must meet the requirements of feasible solutions, forming the solution set ; Wherein, k Indicates the sampling number of the current solution , indicates the number of samples, K Indicates the feasible solution space; F Perform variable neighborhood search on each part of the candidate solution generated by local hyper-box search, and design three equally probable neighborhood structures to be executed. The first considers randomly selecting a gene for disturbance, the second considers selecting a certain continuous gene segment in a certain part, and reversing the gene sequence between them. The third considers migrating the entire continuous gene sequence to a random new location;

[0266]

[0267] After each variable neighborhood search, ensure that each new solution is a feasible solution, forming the solution set

[0268] ;

[0269] Merge the non-dominated solution set , the solution set obtained after local strategy search and Non-dominated sorting and congestion calculation, and retain high-quality individuals to get new population;

[0270] Repeat the above process until the iteration termination condition is met, output the optimal population and Pareto front solution set.

[0271] Step five: adopt decision output device to filter out resource allocation scheme that meets the enterprise's immediate production needs from the Pareto front solution set.

[0272]

[0273]

[0274] Wherein, Indicates the Pareto front solution set, Indicates the individual in the solution set, and Z indicates the number of solutions in the solution set; Indicates the composition of the current individual chromosome, which is mapped to the resource allocation scheme of the manufacturing system, ,​ , , ,

[0275]

[0276] In , according to the current production demand of the enterprise , space limit and fund condition , the preliminary feasible solution is screened to obtain ;

[0277] From the above step four, the number of solutions in the Pareto frontier solution set is 17. The current production demand of the enterprise is =0.04 pieces per second, the space limit is =5, the fund condition is =1 million yuan, the preliminary screening obtains the number of solutions is 2.

[0278]

[0279]

[0280]

[0281] , ,

[0282]

[0283] , ,

[0284] Further to the solution in , the layer-by-layer screening decision method is adopted. If the emphasis of the enterprise is "cost> production rate> space", the final solution is obtained after screening; if the emphasis of the enterprise is "production rate> cost> space" or "space> production rate> cost", the final solution is obtained after screening. When the demand of the enterprise changes, the quantitative indicators under different solutions can be weighed through the decision output device, and the resource allocation scheme suitable for the current demand can be selected through flexible decision.

[0285] Those skilled in the art will appreciate that embodiments of the application can be devised for a method, a system, or a computer program product. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0286] The present application is described in reference to the flowchart and / or block diagrams of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0287] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0288] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0289] Finally, it should be noted that the above-mentioned embodiments are merely intended for describing the technical solutions of the present application, but not for limiting it. Although the present application is described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the present application.

Claims

1. A multi-objective decision-making method for manufacturing systems based on predictive models, characterized in that, The method comprises the following steps: constructing a manufacturing system multi-objective optimization problem model according to enterprise production requirements, space limitations and market conditions; the optimization objectives of the manufacturing system multi-objective optimization problem model include maximizing productivity, minimizing total machine cost and minimizing total buffer capacity; adding fault and repair attributes to machine nodes according to actual product processing flows of the manufacturing system, and constructing a manufacturing system generalized stochastic Petri net model; and improving and constructing a topological enhanced random forest regression prediction model based on random forest regression; adopting an improved NSGA-II algorithm with an embedded local enhancement strategy, combining the manufacturing system generalized stochastic Petri net model and the topological enhanced random forest regression prediction model, solving the manufacturing system multi-objective optimization problem model, and obtaining a Pareto front; using a decision output device to make a decision in the Pareto front according to current enterprise production requirements, space limitations and financial conditions, and outputting a manufacturing system resource configuration scheme meeting the current production requirements; the improved topological enhanced random forest regression prediction model based on random forest regression is constructed as follows: inputting system-level feature expression into a first random forest regression prediction model and inputting system feature enhanced expression into a second random forest regression prediction model as the topological enhanced random forest regression prediction model; the system-level feature expression input and the system feature enhanced expression input are as follows: in, Indicates the first i The static feature input of each process, Indicates the first i The number of parallel machines in each process step. Indicates the first i The average processing time of the machine in each process step Indicates the first i The average processing speed of the machine in each process step Indicates the first i The probability of machine failure in each process step. Indicates the first i Mean time to repair when a machine fails during a process step. Indicates the first i The buffer capacity after each process This represents the system-level feature representation input. Indicates the first i Enhanced feature input for each process step For machine availability, This indicates that the system features enhance the representation of the input. This represents the buffer zone pressure index.

2. The method of claim 1, wherein, the manufacturing system multi-objective optimization problem model is constructed according to enterprise production requirements, space limitations and market conditions as follows: in, This represents the total cost of the machine. Indicates the total capacity of the buffer. Indicates productivity. This indicates the total number of processing steps in the manufacturing system. This indicates the total number of buffers in the manufacturing system. Indicates the process, Indicates a buffer. Indicates the minimum required productivity. Indicates each process The maximum number of machines that can process in parallel. This indicates the maximum capacity of each buffer. Indicates each process The maximum number of machine types available. Indicates the machine type. Indicates machine cost, Represents positive integers. This is an intermediate variable representing the number of processing steps in a manufacturing system. The number of buffers in the manufacturing system is an intermediate variable.

3. The method of claim 1, wherein, the manufacturing system generalized stochastic Petri net model is constructed by adding fault and repair attributes to machine nodes according to actual product processing flows of the manufacturing system as follows: wherein, represents a Petri net, represents an initial resource configuration of the manufacturing system; is a set of places, is a set of raw material places, is a set of machine processing operation places, is a set of buffer places, is a set of machine state detection places, is a set of places for a machine under repair, is a set of transitions, is a set of delay transitions, is a set of instantaneous transitions, represents a topological structure of the manufacturing system, represents a connection relationship between a process and a buffer, is a delay function.

4. The method of claim 1, wherein, the improved topological enhanced random forest regression prediction model based on random forest regression is constructed as follows: obtain a data set from historical data and simulation data, the simulation data is generated by a hypercube sampling method; each piece of data in the data set is composed of system feature enhanced expression input, complete manufacturing system resource configuration scheme and real productivity label, the complete manufacturing system resource configuration scheme is composed of initial input manufacturing system resource configuration scheme in the generalized stochastic Petri net model and machine type scheme selected for each process; split the data set into a training set, a validation set and a test set; train the first random forest regression prediction model and the topological enhanced random forest regression prediction model in parallel, use Bayesian optimization to adjust hyperparameters, and find the best hyperparameter combination on the validation set; evaluate the performance of the first random forest regression prediction model and the topological enhanced random forest regression prediction model on the test set by taking the coefficient of determination, mean square error and root mean square error as performance evaluation indexes; compare the performance of the first random forest regression prediction model and the topological enhanced random forest regression prediction model, select the random forest regression prediction model with better performance as a residual correction model, and calculate the residual between the productivity prediction value of the residual correction model and the real productivity label; update the label of each piece of data in the data set of the residual correction model from the real productivity to the residual, and obtain the trained topological enhanced random forest regression prediction model; The trained topological enhanced random forest regression prediction model is evaluated in terms of coefficient of determination, mean square error and root mean square error on the test set; and the complete manufacturing system resource configuration scheme is input into the trained topological enhanced random forest regression prediction model, and a single predicted productivity value is output.

5. The method of claim 1, wherein, The improved NSGA-II algorithm adopting the embedding local enhancement strategy is combined with the manufacturing system generalized stochastic Petri net model and the topological enhanced random forest regression prediction model to solve the manufacturing system multi-objective optimization problem model and obtain a Pareto front, specifically as follows: Step 4.1, encoding and decoding of chromosomes are performed to obtain encoding information and decoding information; in the encoding, the chromosomes are divided into three parts, which represent the number of machines in parallel on the process, the buffer capacity allocated to each buffer zone and the selected machine type on each process; in the decoding, the elements in the chromosomes are matched with the elements in the manufacturing system generalized stochastic Petri net model, wherein the selected machine type on each process is decoded into failure rate, processing time distribution function, repair time distribution function and cost; Step 4.2, after the initialization parameters including population size, maximum iteration number, crossover probability and mutation probability are set according to the manufacturing system scale, an initial population is randomly generated under boundary constraints, each individual in the initial population represents the resource configuration in the system, and a complete resource configuration scheme is obtained in combination with the known quantitative information; the decoding information is combined with the known quantitative information and the manufacturing system generalized stochastic Petri net model to calculate the total machine cost, the total buffer capacity and evaluate the productivity; wherein the topological enhanced random forest regression prediction model is used to predict the productivity; The individuals in the initial population consider three fitness target values including total machine cost, total buffer capacity and productivity; The importance of the three fitness target values is comprehensively considered to obtain a judgment result of the importance of the three fitness target values, including: judging whether the productivity is less than the minimum productivity requirement, if yes, the importance is marked as productivity > (total machine cost = total buffer capacity), otherwise the importance is marked as (total machine cost = total buffer capacity) > productivity; According to the judgment result of the importance of the three fitness target values, a dynamic dominance criterion is designed, and the non-dominated sorting and crowding distance calculation are performed to classify the individuals in the initial population; Step 4.3, the genetic operators include selection stage, crossover stage and mutation stage; the classical multi-objective binary tournament is used in the selection stage; the partial match crossover is used in the crossover stage; the single-point and two-point mutation are selected with equal probability in the mutation stage; the individuals in the population are composed of three gene sequences, and each sequence is independently judged in the execution of the genetic operator and the local enhancement strategy; The initial population is combined with the population after the execution of the genetic operator to generate a new offspring population after the non-dominated sorting and crowding distance calculation; The variable neighborhood search is performed on each part of the candidate solution generated by the local hyper-box search, and three neighborhood structures are selected with equal probability: Step 4.

4. Obtain a set of non-dominated solutions from the new offspring population where each solution represents a three-part structure encoded by a chromosome; for each non-dominated solution perform local hypercube search and variable neighborhood search; To , , , perform independent local hypercube search, define the hypercube range for each part, for the first j part, the step size of the hypercube is as shown in the following formula: wherein, denotes the first i part of the j th non-dominated solution, each element of which is adjusted within a step size , is the search step size for the first j part; [] denotes a closed interval; is the number of non-dominated solutions; for each solution sparsely sample several candidate solutions within the hyperbox , form a solution set ; wherein, k denotes the sampling number of the current solution , denotes the sampling number of the current solution K denotes the sampling number of the current solution F denotes the feasible solution space; ​ The first consideration randomly selects a gene for disturbance, the second consideration selects a certain part of a certain continuous gene segment, reverses the gene sequence therebetween, and the third consideration migrates the continuous gene sequence as a whole to a random new location. After each variable neighborhood search, a number of candidate new solutions are sparsely sampled within the hyperbox The new solution set is constructed ; The non-dominated solution set , the solution set and the new solution set are combined to carry out non-dominated sorting and crowdedness calculation, and new Pareto front individuals are obtained as high-quality individuals, and the high-quality individuals are reserved to obtain a new population; Step 4.5, judging whether the current iteration meets the termination condition, yes, outputting the optimal population and a Pareto front, otherwise, updating the current population and the iteration number and continuing iteration.

6. The method of claim 1, wherein, The decision outputter is used to make decisions in the Pareto front according to the current production demand, space limitation and fund situation of the enterprise, and output a manufacturing system resource configuration scheme meeting the current production demand, specifically: in, This represents the Pareto front solution set. Denotes the first in the solution set Individual, Z Indicates the number of solutions in the solution set; This represents the composition of an individual's chromosomes, mapped to a resource allocation scheme for the manufacturing system. , , They represent the current schemes respectively. The total cost of the machines, the total buffer capacity, and the productivity; In the current production needs of the enterprise , space constraints and financial situation are screened to obtain a preliminary feasible Pareto frontier solution set ; If , according to the current solution, provide the criticality index of the current minimal feasible solution and give a constraint relaxation suggestion; If , for the solution of , the layer-by-layer screening decision method is adopted, if the degree of attention of the enterprise is cost>productivity>space, the solution set Cost with the lowest cost is screened out in firstly; ; If , continue the next layer screening according to the importance degree as the standard until the end of the three-layer screening to get the final solution , otherwise directly output the final solution , get the manufacturing system resource configuration scheme that meets the current production demand.

7. A multi-objective decision-making system for manufacturing systems based on predictive models, characterized in that, The manufacturing system multi-objective optimization problem model module is used to construct a manufacturing system multi-objective optimization problem model according to the production demand, space limitation and market condition of the enterprise; The optimization objectives of the manufacturing system multi-objective optimization problem model include maximizing productivity, minimizing total machine cost and minimizing total buffer capacity; The Petri net model and prediction model construction module is used to add fault and repair attributes to the machine node according to the actual product processing flow of the manufacturing system, and construct a manufacturing system generalized stochastic Petri net model; and the topological enhanced random forest regression prediction model is constructed based on the improved random forest regression; The manufacturing system multi-objective optimization problem solving module is used to solve the manufacturing system multi-objective optimization problem model by using the improved NSGA-II algorithm embedded with a local enhancement strategy, in combination with the manufacturing system generalized stochastic Petri net model and the topological enhanced random forest regression prediction model, to obtain a Pareto front; The output manufacturing system resource configuration scheme module is used to make decisions in the Pareto front according to the current production demand, space limitation and fund situation of the enterprise by using the decision outputter, and output a manufacturing system resource configuration scheme meeting the current production demand; The topological enhanced random forest regression prediction model is constructed based on the improved random forest regression, specifically as follows: The system-level feature expression is input into a first random forest regression prediction model, and the system feature enhanced expression is input into a second random forest regression prediction model as a topological enhanced random forest regression prediction model; The system-level feature expression input and the system feature enhanced expression input are as follows: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the manufacturing system multi-objective decision method based on the prediction model in any one of claims 1-6. wherein, represents the number of machines in the i static feature input of the represents the number of machines in the i parallel machines of the represents the average processing time of the i machines of the represents the average processing rate of the i machines of the represents the probability of machine failure of the i machines of the represents the average repair time of the i machines of the represents the buffer capacity after the i machines of the represents the system level feature expression input, represents the enhanced feature input of the i machines of the is the machine availability, represents the system feature enhanced expression input, is the buffer pressure index.

8. An electronic device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the manufacturing system multi-objective decision method based on the prediction model in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, ​

Images