Knowledge-driven intelligent production scheduling method for workshops
Through the knowledge-driven intelligent production scheduling method in workshops, using knowledge graphs and improved NSGA-II algorithms, the problems of data management and manual adjustment in the production process in workshops are solved, achieving more efficient production scheduling and data utilization, and improving the problem of data management in workshops is achieved. Production efficiency and intelligent production scheduling level.
Patent Information
- Application Number
- CN202210559945.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-23
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2042-05-23
AI Technical Summary
The existing workshop production scheduling methods are difficult to meet complex actual production needs, and the lack of unified data management models and data knowledge utilization methods, resulting in the need to manually adjust the computer production scheduling plan during the production process.
Using knowledge-driven workshop intelligent production scheduling method, by constructing a multi-objective multi-constraint flexible work workshop scheduling model, using knowledge graphs and improved NSGA-II algorithm, combined with hierarchical analysis method, Pareto's optimal solution set and optimal scheduling solution are obtained. And verify the feasibility of the scheduling scheme through the online knowledge graph and make manual adjustments.
It realizes unified management and production guidance of workshop data, improves production efficiency, reduces the dependence of manual adjustments, obtains a scheduling plan that is more in line with actual production, and improves the workshop's intelligent production scheduling level and feasibility of scheduling plan.
Smart Images

Figure CN114879618B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of intelligent production scheduling methods for production processes, and specifically relates to a knowledge-driven intelligent production scheduling method for workshops. Background Art
[0002] Intelligent manufacturing is the main direction for promoting the deep integration of informatization and industrialization. Intelligent manufacturing includes three levels: the intelligentization of production objects, the intelligentization of production processes, and the intelligentization of production tools. Among them, the intelligentization of production processes is the key link in realizing intelligent manufacturing. The intelligentization of production processes actually uses computers to simulate the intelligent activities of manufacturing experts such as analysis, judgment, reasoning, conception, and decision-making, and organically integrates these intelligent activities with intelligent machines. To realize the intelligentization of production processes, it is necessary to perceive the actual production environment, complete decisions on intelligently issuing production tasks and adjusting production scheduling plans, and use equipment to run the production process according to the scheduling results to complete order tasks. The distribution results are used to complete processing using equipment. Workshop scheduling is a key step in the production process, and the results of workshop scheduling have an important impact on the production process and efficiency. Therefore, obtaining an efficient and practical scheduling plan is one of the key links in the intelligentization of the production process. Optimizing the workshop scheduling process and results is crucial to improving the production efficiency and competitiveness of enterprises.
[0003] At present, although there are many mathematical models and optimization algorithms for workshop scheduling, it is difficult to build and solve models for relatively complex real production, resulting in scheduling solutions that rely solely on computer scheduling algorithms cannot fully meet actual production needs, and manual adjustments to computer scheduling are required. At the same time, a large amount of data is generated during the production process, but companies currently lack a unified data management model and methods for utilizing data knowledge, resulting in a lack of utilization of data knowledge in the production process. In order to make full use of manual scheduling adjustment experience and workshop data knowledge, a method is needed to achieve data management and application.
[0004] As an important branch of artificial intelligence, knowledge graph provides technical support for data management and application and the establishment of connections between data. The knowledge graph is composed of elements such as nodes, relationships and attributes. It can establish connections between various data in the form of triples, and form a complete database of multi-source data to facilitate the management and application of data knowledge. The use of knowledge graph can make up for the shortcomings of workshop data management, establish connections between various data, and manage manual experience in the model layer to provide guidance for subsequent production, reduce the dependence of workshop scheduling on manual adjustments, and lay the foundation for the realization of intelligent production processes. Summary of the invention
[0005] The purpose of the present invention is to provide a knowledge-driven intelligent workshop production scheduling method to solve the problems existing in the above-mentioned workshop production scheduling.
[0006] The technical solution adopted by the present invention is a knowledge-driven workshop intelligent production scheduling method, which is specifically implemented according to the following steps:
[0007] Step 1: Taking the multi-objective and multi-constraint flexible job shop scheduling problem as the research object, a mathematical model is constructed;
[0008] Step 2: Use the manual scheduling adjustment experience and historical data of the production workshop to build the model layer and data layer of the offline workshop knowledge graph respectively, and complete the construction of the data management model;
[0009] Step 3: Based on the offline knowledge graph data layer, the production data of the workpiece to be processed is obtained, and the Pareto optimal solution set and optimal scheduling plan of the workpiece to be processed are obtained by using the improved NSGA-II algorithm and hierarchical analysis method;
[0010] Step 4: Build an online workshop knowledge graph based on actual production, compare the optimal scheduling solution obtained by the algorithm with the resources in actual production, and determine the feasibility of the scheduling solution. If feasible, output the scheduling solution; otherwise, adjust the output based on the manual scheduling adjustment experience in the offline knowledge graph model layer and actual production.
[0011] The present invention is also characterized in that:
[0012] The model of step 1 is as follows: There are n workpieces to be processed in the production workshop, and m optional equipment. Each workpiece to be processed contains one or more processes O i,j (i=1,2,…,n;j=1,2,…,m), where each process can have one or more optional devices, and the processing time of a process on each device may be different.
[0013] In step 1, the multi-objectives include maximum completion time, total delay, total equipment load and total energy consumption, and the multi-constraints mainly include process constraints, equipment constraints and time constraints.
[0014] The multi-objective model is as follows:
[0015] Maximum completion time: f1 = maxC i (1)
[0016] Among them, C i is the completion time of the last process of job i, i = 1, 2, …, n;
[0017] Total extension:
[0018] Among them, D i is the delivery date of workpiece i;
[0019] Total equipment load:
[0020] Among them, n i is the number of steps for job i, P i,j,h It is process O i,j On device M h Processing time on X i,j,h is a decision variable, process O i,j On device M h During processing, X i,j,h =1; otherwise, X i,j,h =0;
[0021] Total energy consumption of the workshop:
[0022]
[0023] Among them, E idle is the idling energy consumption of the equipment, E busy is the processing energy consumption of the equipment, is the idling power of the device h, is the idle time of device h, is the processing power of equipment h, is the processing time of equipment h, W inhe It is the fixed power of the workshop.
[0024] The multi-constrained model is:
[0025] Each process can only be processed on one device.
[0026] Among them, Y i,j,h,k It is process O i,j On device M h Upper processing, process O i,j+1 On device M k processing, i∈(1,2,…,n), j∈(1,2,…,n i );
[0027] Each workpiece process has a sequence, and the next process can only be processed after the previous process is completed. i,j ≥FT i,j-1 (6)
[0028] Among them, ST i,j It is process O i,j Time to start processing, FT i,j-1 It is process O i,j-1 The time to end processing, i∈(1,2,…,n), j∈(1,2,…,n i );
[0029] The processing equipment for the process must be selected from the optional equipment.
[0030] Among them, Z i,j,k It is process O i,j On device M k processing, i∈(1,2,…,n), j∈(1,2,…,n i ),
[0031] k∈(1,2,…,m);
[0032] The maximum completion time is not less than the completion time of each workpiece.
[0033] in, is the completion time of job i, i∈(1,2,…,n);
[0034] The completion time of each process is not less than 0.
[0035] in, is the completion time of job l, i∈(1,2,…,n), l∈(1,2,…,n).
[0036] The improvement of NSGA-II algorithm in step 3 is the determination of crossover and mutation operators and the elite retention strategy.
[0037] The determination of crossover and mutation operators is as follows:
[0038] The crossover and mutation operators change with the population, where the maximum and minimum crossover probability values are randomly selected from 0.4 to 0.8. Similarly, the maximum and minimum mutation probability values are randomly selected from 0.1 to 0.2. By effectively controlling the crossover and mutation probabilities, the algorithm's solution performance is improved. The final crossover probability is solved as follows:
[0039] Among them, P c(i) is the crossover probability when the number of iterations is i, P minc is the random minimum crossover probability, P maxc is the random maximum crossover probability, i is the number of iterations, and N is the maximum number of iterations;
[0040] When selecting the mutation probability, it is considered that the evolution process is divided into the early stage and the late stage. In the early stage of evolution, in order to ensure the global optimization ability, the smaller the mutation probability is, the better. The calculation formula is as follows:
[0041]
[0042] Among them, P m(i) is the mutation probability when the number of iterations is i, P minm is the minimum random mutation probability, Pmaxm is the maximum random mutation probability;
[0043] In order to avoid the algorithm from falling into the local optimum in the later stage of evolution, a larger mutation probability value is taken. The calculation formula is as follows:
[0044] The specific elite retention strategy is: the original elite retention strategy is used for the first N / 3, and the distribution function is introduced for the last N / 3~N to limit the probability of them entering the new parent population, ensuring that the proportion of each level of non-dominated frontier entering the new parent population decreases linearly with the increase in the number of iterations, avoiding directly placing excellent individuals into the new parent population, and ensuring population diversity. The calculation formula is as follows: m k =F k *r k (13)
[0045] Among them, m k is the total number of individuals retained by the elite strategy, F k represents the total number of non-dominated frontier individuals at level k, r k =rand(0.7,1) represents a random real number in the range of 0.7 to 1.
[0046] The beneficial effects of the present invention are:
[0047] 1) The present invention uses knowledge graph technology to integrate the historical processing data of the production workshop and complete the construction of the offline workshop knowledge graph. It provides data support such as processing technology and optional equipment for the subsequent optimization algorithm to solve the workshop scheduling problem. At the same time, it manages the historical data of the production workshop, extracts artificial experience knowledge, and provides a unified model and production guidance for workshop data management;
[0048] 2) The present invention improves the NSGA-II algorithm and improves the algorithm solving performance. At the same time, the algorithm solving is based on the offline knowledge graph data layer. The method can use the offline graph data layer to obtain the production data of the workpiece to be processed, which improves the workshop production efficiency to a certain extent;
[0049] 3) The present invention constructs an online workshop knowledge graph based on actual production, and compares the scheduling solution obtained by the algorithm with the production resources of the online knowledge graph to verify the feasibility of the scheduling solution. At the same time, the scheduling solution obtained by the algorithm can be adjusted and output based on the manual scheduling adjustment experience in the offline knowledge graph model layer and the online graph. This method replaces manual adjustment, improves the level of intelligent scheduling in the workshop, obtains a scheduling solution that is more in line with actual production, and improves the feasibility of the scheduling solution. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 Flow chart of the knowledge-driven NSGA-II algorithm method of the present invention;
[0051] Figure 2 It is a model of the machining process of the workpiece to be machined and related information according to an embodiment of the present invention;
[0052] Figure 3 is a flow chart of an improved NSGA-II algorithm according to an embodiment of the present invention;
[0053] Figure 4 is a Gantt chart of a flexible workshop scheduling case in an embodiment of the present invention;
[0054] Figure 5 It is a diagram of an online knowledge graph information system according to an embodiment of the present invention. DETAILED DESCRIPTION
[0055] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0056] The knowledge-driven intelligent workshop production scheduling method of the present invention is as follows: Figure 1 , follow the steps below:
[0057] Step 1: Taking the multi-objective and multi-constraint flexible job shop scheduling problem as the research object, a mathematical model is constructed;
[0058] The model of step 1 is as follows: There are n workpieces to be processed in the production workshop, and m optional equipment. Each workpiece to be processed contains one or more processes O i,j (i=1,2,…,n;j=1,2,…,m), where each process can have one or more optional equipments, and the processing time of one process on each equipment may be different;
[0059] In order to ensure the superiority of the workshop scheduling results, multiple objectives and multiple constraints are set. The multiple objectives include the maximum completion time, total delay, total equipment load and total energy consumption. The multiple constraints mainly include process constraints, equipment constraints and time constraints.
[0060] The multi-objective model is as follows:
[0061] Maximum completion time: f1 = maxC i (1)
[0062] Among them, C i is the completion time of the last process of job i, i = 1, 2, …, n;
[0063] Total extension:
[0064] Among them, D i is the delivery date of workpiece i;
[0065] Total equipment load:
[0066] Among them, n i is the number of steps for job i, P i,j,h It is process O i,j On device M h Processing time on X i,j,h is a decision variable, process O i,j On device M h During processing, X i,j,h =1; otherwise, X i,j,h =0;
[0067] Total energy consumption of the workshop:
[0068]
[0069] Among them, E idle is the idling energy consumption of the equipment, E busy is the processing energy consumption of the equipment, is the idling power of the device h, is the idle time of device h, is the processing power of equipment h, is the processing time of equipment h, W inhe It is the fixed power of the workshop.
[0070] The multi-constrained model is:
[0071] Each process can only be processed on one device.
[0072] Among them, Y i,j,h,k It is process O i,j On device M h Upper processing, process O i,j+1 On device M k On processing,
[0073] i∈(1,2,…,n), j∈(1,2,…,n i );
[0074] Each workpiece process has a sequence, and the next process can only be processed after the previous process is completed. i,j ≥FT i,j-1 (6)
[0075] Among them, ST i,j It is process O i,j Time to start processing, FT i,j-1 It is process O i,j-1 The time to end processing, i∈(1,2,…,n), j∈(1,2,…,n i );
[0076] The processing equipment for the process must be selected from the optional equipment.
[0077] Among them, Z i,j,k It is process O i,j On device M k processing, i∈(1,2,…,n), j∈(1,2,…,n i ), k∈(1,2,…,m);
[0078] The maximum completion time is not less than the completion time of each workpiece.
[0079] in, is the completion time of job i, i∈(1,2,…,n);
[0080] The completion time of each process is not less than 0.
[0081] in, is the completion time of job l, i∈(1,2,…,n), l∈(1,2,…,n).
[0082] Step 2: Use the manual scheduling adjustment experience and historical data of the production workshop to build the model layer and data layer of the offline workshop knowledge graph respectively, and complete the construction of the data management model;
[0083] The model layer mainly integrates the factors and corresponding relationships that affect production, and guides the construction of the data layer. At the same time, the manual scheduling adjustment experience in the model layer can provide guidance for subsequent plan adjustments and production. The data layer represents the specific data of the production workshop in the form of a triple of head entity-relationship-tail entity. Cypher statements can be used to search for relevant information of the workpiece to be processed in the offline graph to provide data support for the workpiece to be processed.
[0084] Step 3: Based on the offline knowledge graph data layer, the production data of the workpiece to be processed is obtained, and the Pareto optimal solution set and optimal scheduling plan of the workpiece to be processed are obtained by using the improved NSGA-II algorithm and hierarchical analysis method;
[0085] The improvement of NSGA-II algorithm in step 3 is the determination of crossover and mutation operators and the elite retention strategy;
[0086] Improvement of crossover and mutation operators: In the current research on the NSGA-II algorithm, the crossover and mutation operators are assigned values in a fixed way, which limits the algorithm's solution performance to a certain extent. The present invention adopts an adaptive random strategy, and the crossover and mutation operators change with the population, where the maximum and minimum crossover probability values are randomly selected from 0.4 to 0.8, and the maximum and minimum probability values of mutation are randomly selected from 0.1 to 0.2. By effectively controlling the crossover and mutation probabilities, the algorithm's solution performance is improved, and the final crossover probability is solved as follows:
[0087] Among them, P c(i) is the crossover probability when the number of iterations is i, P minc is the random minimum crossover probability, P maxc is the random maximum crossover probability, i is the number of iterations, and N is the maximum number of iterations;
[0088] When selecting the mutation probability, it is considered that the evolution process is divided into the early stage and the late stage. In the early stage of evolution, in order to ensure the global optimization ability, the smaller the mutation probability is, the better. The calculation formula is as follows:
[0089]
[0090] Among them, P m(i) is the mutation probability when the number of iterations is i, P minm is the minimum random mutation probability, P maxm is the maximum random mutation probability;
[0091] In order to avoid the algorithm from falling into the local optimum in the later stage of evolution, a larger mutation probability value is taken. The calculation formula is as follows:
[0092] Improvement of elite retention strategy: The traditional elite retention strategy uses non-dominated sorting and crowding distance calculation to retain the first N generations of individuals as the new parent population, which limits the population diversity. The present invention adopts the method of distribution ratio elite retention strategy. The original elite retention strategy is used for the first N / 3, and the distribution function is introduced for the last N / 3~N to limit the probability of entering the new parent population, ensuring that the proportion of each level of non-dominated frontier entering the new parent population decreases linearly with the increase of the number of iterations, avoiding directly placing excellent individuals into the new parent population, and ensuring population diversity. The calculation formula is as follows: m k =F k *r k (13)
[0093] Among them, m k is the total number of individuals retained by the elite strategy, F k represents the total number of non-dominated frontier individuals at level k, rk =rand(0.7,1) represents a random real number in the range of 0.7 to 1.
[0094] Step 4: Build an online workshop knowledge graph based on actual production, compare the optimal scheduling solution obtained by the algorithm with the resources in actual production, and determine the feasibility of the scheduling solution. If feasible, output the scheduling solution; otherwise, adjust the output based on the manual scheduling adjustment experience in the offline knowledge graph model layer and actual production.
[0095] Example 1
[0096] Combined with Figure 2 , the construction of offline workshop knowledge graph is explained in detail.
[0097] Applying knowledge graph technology to workshop scheduling has improved production efficiency to a certain extent, replaced manual adjustments, and provided a unified data management model. The construction of the offline workshop knowledge graph data layer can be completed through the following steps: First, the workshop processing data needs to be collected, including the definition of the workshop knowledge graph, the specification of workpieces and related processing information; then, knowledge fusion is performed to determine entities and relationships; finally, the obtained data is stored.
[0098] The data knowledge in the offline workshop knowledge graph can be obtained through Cypher statements. This method is simple and easy to operate, and can effectively improve the work efficiency of dispatchers.
[0099] The order information data for constructing the graph is obtained through the above method, as shown in Table 1:
[0100]
[0101] Table 1
[0102] The data format for constructing the offline workshop knowledge graph data layer is shown in Table 2:
[0103]
[0104] Table 2
[0105] The data in the above format is stored in the Neo4j graph database to complete the offline workshop knowledge graph construction. The Cypher statement is used to obtain the process and related information of the workpiece to be processed. The search statement is as follows:
[0106] match(n:parts)<-->(b)<-->(c)-[r1:Has_technology]-(d)-
[0107] (r2:Operating_equipment)-(e)
[0108] where n.name='1AP08423-2101-13z9A#CSL'
[0109] return n,b,c,d,r1,r2,e
[0110] Combined with Figure 3 , the improved NSGA-II algorithm for solving scheduling problems is explained in detail.
[0111] (1) Initialization parameter settings
[0112] According to the requirements, after multiple experiments, the parameter population size pop = 100 and the maximum number of iterations gen = 200 are initialized.
[0113] (2) Encoding and decoding
[0114] A two-layer coding mechanism is used to solve the workshop scheduling problem, encoding both the workpiece process and the processing equipment. This coding method can obtain two chromosomes, process and equipment, and obtain the final scheduling plan based on the chromosome operation. The length of the process chromosome is the sum of the number of processes corresponding to each workpiece. Each value in the process coding layer chromosome represents the number of the workpiece, the number of occurrences represents the number of processes of the workpiece, and the order of occurrence represents the number of processes of the processing. Each value in the equipment coding layer chromosome represents the processing equipment corresponding to each process, and its chromosome length is the same as the length of the process coding layer chromosome. The coding process determines the processing equipment for each process.
[0115] Decoding is the process of generating a scheduling plan based on the encoded information. The decoding process needs to pay attention to the size of the idle time and the processing time. If the idle time is greater than the processing time, the process can be processed within the idle time of the device; otherwise, it will be processed after the last process processed by the device.
[0116] (3) Non-dominated cumulative sorting and crowding distance calculation
[0117] The non-dominated cumulative sorting is obtained by improving the fast non-dominated sorting. The cumulative sorting takes into account both the individual density information and the Pareto sorting value. This method obtains the Pareto sorting value of all individuals by performing Pareto sorting on each individual. The calculation formula is as follows:
[0118] Among them, r(x,t) is the sum of the Pareto ranking values of individual x in generation t, r(x k ,t) is the individual x of generation t k The Pareto ranking value of .
[0119] In order to evaluate the degree of crowding of adjacent solutions of a solution, the concept of crowding distance is introduced. It is used as an evaluation criterion for individuals in the non-dominated front of the same level. The sum of the distance difference between two adjacent individuals of an individual in the non-dominated front of the same level in each sub-objective function is called the crowding distance of the individual. Its calculation formula is as follows:
[0120] Among them, D j is the crowding distance of individual j, D j .f k is the individual j in sub-goal f k The function value on , t is the total number of sub-goals.
[0121] According to the previous non-dominated sorting and crowding distance calculation, each individual j in the population has its own non-dominated sorting j rank and crowding distance j d Based on these two attributes, the crowding comparison operator is defined: individual j is compared with individual i. If j wants to be better than i, it needs to meet any of the following two conditions:
[0122] j rank ramk , that is, when the rankings of individual j and individual i are different, the individual with a smaller ranking is selected;
[0123] j rank =i ramk And D j >D i , that is, when two individuals have the same ranking, the individual with a larger crowding distance is selected.
[0124] (4) Selection, crossover, and mutation
[0125] Selection operation: The tournament selection mechanism is adopted. The idea is to randomly select individuals from the population, use the crowding comparison operator to select better individuals to enter the next parent population, and finally select N better individuals to perform crossover and mutation operations.
[0126] An adaptive strategy is used to determine the crossover and mutation probabilities, which change with the number of iterations. This method is conducive to retaining excellent individuals and avoiding premature and local optimal imagination.
[0127] Crossover operation: Crossover operation includes process crossover and equipment crossover. Process crossover adopts the method of partial crossover mapping, and equipment crossover operation adopts the crossover method of random process corresponding to equipment. In the crossover process, it should be noted that when the process is crossover, the equipment used for process processing remains unchanged; similarly, when the equipment is crossover, the chromosome encoding of the process remains unchanged.
[0128] Process crossover operation: The idea is to randomly select two chromosomes from the chromosomes retained by the selection operation for partial crossover operation; then, randomly select two points within the length of the chromosome, take the chromosome segment between the two points, and exchange the segments between the two points taken from the two chromosomes; finally, compare the sub-segments, delete the parts with the same numbers in the sub-segments, and retain the different parts to obtain new sub-segments. If the new sub-segment is greater than 0, the new sub-segment part is compared with the chromosome to perform chromosome code replacement; otherwise, the new chromosome after crossover is directly obtained.
[0129] Equipment crossover operation: The idea is to perform crossover operation on the equipment corresponding to a certain process, randomly select a process from all processes, obtain the processing equipment corresponding to the process, compare the chromosome fragments of the two devices, obtain different fragments of the equipment, exchange the fragments, and obtain the chromosome encoding after crossover.
[0130] The formula for determining the crossover probability is as follows:
[0131]
[0132] Among them, P c(i) is the crossover probability when the number of iterations is i, P minc is the random minimum crossover probability, P maxc is the random maximum crossover probability, i is the number of iterations, and N is the maximum number of iterations.
[0133] Mutation operation: Like the crossover operation, both the process and the equipment are subjected to mutation operation. The process mutation operation adopts the reverse operation, and the equipment mutation adopts the random mutation operation. For the process coding layer, when performing the mutation operation, the order of the processing equipment for each process must remain unchanged; for the equipment coding layer, when performing the mutation operation, the process code does not change.
[0134] Process variation operation: First, randomly select a chromosome; then, randomly select two points on the chromosome, sort the two points, and take out the segment between the two points; finally, reverse the segment to complete the process variation operation. It should be noted that the processing equipment and process of the corresponding process should be consistent.
[0135] Equipment mutation operation: Determine the chromosome positions in the process coding layer where the optional equipment for the process is greater than 1, randomly select one of these points, obtain the remaining optional equipment for the process, replace the equipment at the corresponding position in the parent chromosome, and complete the equipment mutation operation.
[0136] When selecting the mutation probability, it is considered that the evolution process is divided into the early stage and the late stage. In the early stage of evolution, in order to ensure the global optimization ability, the smaller the mutation probability is, the better. The calculation formula is as follows:
[0137]
[0138] Among them, P m(i) is the mutation probability when the number of iterations is i, P minm is the minimum random mutation probability, P maxm is the maximum random mutation probability.
[0139] In order to avoid the algorithm from falling into the local optimum in the later stage of evolution, a larger mutation probability value is taken. The calculation formula is as follows:
[0140] (5) Elite retention strategy
[0141] The distribution variable ratio elite retention strategy is adopted. The idea is to use the original elite retention strategy for the first N / 3, and introduce a distribution function for the last N / 3~N to limit the probability of them entering the new parent population, ensuring that the proportion of each level of non-dominated frontier entering the new parent population decreases linearly with the increase in the number of iterations, avoiding directly placing excellent individuals into the new parent population, and ensuring the diversity of the population. The calculation formula is as follows:
[0142] m k =F k *r k
[0143] Where: m k is the total number of individuals retained by the elite strategy, F k represents the total number of non-dominated frontier individuals at level k, r k =rand(0.7,1) represents a random real number in the range of 0.7 to 1.
[0144] (6) Analytical Hierarchy Process
[0145] The improved NSGA-II algorithm is used to obtain the Pareto optimal solution set. Since the present invention solves multi-objective problems, it is impossible to obtain a unique optimal solution, but an optimal solution set. In order to select a unique optimal solution from the optimal solution set, the hierarchical analysis method is used with four objectives as factor layers. According to the importance of the four objectives in the scheduling plan, the indicator allocation weight matrix is constructed, and the consistency test is performed to finally obtain the optimal scheduling plan.
[0146] The present invention uses 3×3 cases to verify the effectiveness of the algorithm. Table 3 shows the equipment selection and processing time for each workpiece process. Among them, * indicates that the equipment cannot be selected for processing in this process. The scheduling results are as follows Figure 4 .
[0147]
[0148]
[0149] Table 3
[0150] Combined with Figure 5 , a detailed explanation is given on the problem of guiding workshop production scheduling using online workshop knowledge graph.
[0151] The construction of online workshop knowledge graph relies on various monitoring devices to integrate and transmit workshop information in real time, providing data support for the online graph. The online graph update speed can be updated in real time according to the distribution of processing order tasks, mainly updating the equipment usage status, personnel allocation, inventory quantity and other factors affecting production in the graph.
[0152] The adjustment of the scheduling plan is mainly aimed at the change of processing equipment. At present, most production adjustments rely on manual work, which has a low degree of intelligence, is time-consuming and labor-intensive. Secondly, manual adjustments rely on experience. Everyone has different experience in the adjustment process, and there is no unified basis for adjustment. To address this problem, a knowledge graph model is introduced. This model can improve the utilization of data knowledge, integrate and summarize the experience of manual scheduling adjustment, and form a common experience to guide the adjustment of the workshop scheduling plan. The steps are as follows: First, observe whether the equipment selected in the scheduling plan is idle through the online graph. If it meets the actual production, no adjustment is required. Otherwise, the plan is adjusted according to the actual alternative equipment in the workshop. This method relies on actual production and makes adjustments according to actual production and demand. It can effectively and reasonably utilize various production resources. At the same time, when the number of processed workpieces is too large, it relies on manual scheduling or adjustment, resulting in low scheduling efficiency, which affects the production efficiency of the enterprise and causes production delays.
[0153] The present invention overcomes the problem that the scheduling solution obtained by simply using optimization models and algorithms does not conform to actual production. It integrates knowledge graph technology and scheduling optimization algorithms to solve the problem that the scheduling mode of algorithm optimization and manual adjustment step by step has limitations in manual adjustment, data knowledge management and application.
Claims
1. The knowledge-driven intelligent workshop production scheduling method is characterized by: Follow the steps below to implement it: Step 1: Taking the multi-objective and multi-constraint flexible job shop scheduling problem as the research object, a mathematical model is constructed; Step 2: Use the manual scheduling adjustment experience and historical data of the production workshop to build the model layer and data layer of the offline workshop knowledge graph respectively, and complete the construction of the data management model; Step 3: Based on the offline knowledge graph data layer, the production data of the workpiece to be processed is obtained, and the Pareto optimal solution set and optimal scheduling plan of the workpiece to be processed are obtained by using the improved NSGA-II algorithm and hierarchical analysis method; Step 4: Build an online workshop knowledge graph based on actual production, compare the optimal scheduling solution obtained by the algorithm with the resources in actual production, determine the feasibility of the scheduling solution, and output the scheduling solution if it is feasible; Otherwise, the output is adjusted based on the manual scheduling adjustment experience and actual production in the offline knowledge graph model layer; The improvement of the NSGA-II algorithm in step 3 is the determination of the crossover and mutation operators and the elite retention strategy; The determination of crossover and mutation operators is as follows: The crossover and mutation operators change with the population, where the maximum and minimum crossover probability values are randomly selected from 0.4 to 0.
8. Similarly, the maximum and minimum mutation probability values are randomly selected from 0.1 to 0.
2. By effectively controlling the crossover and mutation probabilities, the algorithm's solution performance is improved. The final crossover probability is solved as follows: (10) in, The number of iterations is The crossover probability when is the random minimum crossover probability, is the random maximum crossover probability, is the number of iterations, is the maximum number of iterations; When selecting the mutation probability, it is considered that the evolution process is divided into the early stage and the late stage. In the early stage of evolution, in order to ensure the global optimization ability, the smaller the mutation probability is, the better. The calculation formula is as follows: (11) in, The number of iterations is The probability of mutation when is the minimum random mutation probability, is the maximum random mutation probability; In order to avoid the algorithm from falling into the local optimum in the later stage of evolution, a larger mutation probability value is taken. The calculation formula is as follows: (12); The specific elite retention strategy is: Adopting the original elite retention strategy, A distribution function is introduced to limit the probability of entering the new parent population, ensuring that the proportion of each level of non-dominated frontier entering the new parent population decreases linearly with the increase in the number of iterations, avoiding directly placing excellent individuals into the new parent population, and ensuring population diversity. The calculation formula is as follows: (13) in, is the total number of individuals retained by the elite strategy, express The total number of non-dominated frontier individuals of level , express A random real number in the range.
2. The knowledge-driven intelligent workshop production scheduling method according to claim 1, characterized in that: The model of step 1 is as follows: the production workshop has Workpieces to be processed, optional equipment Each workpiece to be processed contains one or more processes , where each process can have one or more optional devices, and the processing time of a process on each device may be different.
3. The knowledge-driven intelligent workshop production scheduling method according to claim 1, characterized in that: In step 1, the multiple objectives include four aspects: maximum completion time, total delay, total equipment load and total energy consumption; and the multiple constraints include three aspects: process constraints, equipment constraints and time constraints.
4. The knowledge-driven intelligent workshop production scheduling method according to claim 3 is characterized in that: The multi-objective model is specifically: Maximum completion time: (1) in, It is a workpiece The last process completion time, ; Total extension: (2) in, It is a workpiece Delivery period; Total equipment load: (3) in, It is a workpiece The number of processes, It is a process On the device The processing time on is a decision variable, process On the device During processing, ;on the contrary, ; Total energy consumption of the workshop: (4) in, C max is the maximum completion time, is the energy consumption of equipment idling, is the processing energy consumption of the equipment, Yes Equipment Idle power, Yes Equipment Idle time, Yes Equipment The processing power, Yes Equipment The processing time, It is the fixed power of the workshop.
5. The knowledge-driven intelligent workshop production scheduling method according to claim 3, characterized in that: The multi-constrained model is: Each process can only be processed on one device. (5) Among them, It is a process On the device Processing On the device On processing, , ; Each workpiece process has a sequence, and the next process can only be processed after the previous process is completed. (6) in, It is a process The time to start processing, It is a process The time to finish processing, , ; The processing equipment for the process must be selected from the optional equipment. (7) in, It is a process On the device On processing, , , ; The maximum completion time is not less than the completion time of each workpiece. (8) in, It is a workpiece The completion time, ; The completion time of each process is not less than 0. (9) in, It is a workpiece The completion time, , .
Citation Information
Patent Citations
Method for obtaining optimal dispatching scheme of part machining based on one-dimensional particle swarm algorithm
CN103809506A
Flexible flow shop energy consumption optimization scheduling method
CN104808636A
Flexible job-shop scheduling multi-objective method
CN104914835A
Low-energy-consumption flexible workshop scheduling method
CN110619437A
Industrial Internet of Things equipment collaborative management and control system based on data and knowledge driving
CN112651520A
