CN111445079B - A multi-fidelity simulation optimization method and equipment applied to workshop planning and production - Google Patents

Abstract

The invention belongs to the technical field of production control, and in particular relates to a multi-fidelity simulation optimization method and equipment applied to workshop planning and production. This method optimizes the problem of workshop planning and production by using two fidelity levels of simulation models: first, a high-fidelity and low-fidelity simulation model is established based on the actual production system; then a low-fidelity simulation model is used, combined with the optimization iteration of the GA algorithm Search the solution space of the production planning problem; use the sampling method with ordinal transformation and optimal sampling to sample according to the results obtained by running the low-fidelity model to obtain the optimal sampling set; finally run the optimal sampling through the high-fidelity simulation model Since the high-fidelity model is very similar to the actual production system of the workshop, using the high-fidelity model to run the production plan in the optimal sampling set can obtain relatively accurate results, so this result can be used as a basis to select the optimal production plan.

Description

A multi-fidelity simulation optimization method and equipment applied to workshop planning and production

technical field

The invention belongs to the field of production control, and relates to a multi-fidelity simulation optimization method and equipment applied to workshop planning and production, more particularly, to a DBR theory-based multi-fidelity simulation optimization method applied to workshop planning and production.

Background technique

Making a reasonable production plan can make workshop processing smoother and make full use of workshop resources. Taking the problem of workshop production planning as an example, an inappropriate workpiece production plan may cause workshop processing blockage or a large number of idle machines. Therefore, optimizing the workshop production plan can improve processing efficiency. Because the production process of the production system is irreversible, the workshop managers often use the simulation method to judge the feasibility of the plan before executing the production decision, and even find the optimal solution to the problem through the simulation operation method. Traditional simulation technology usually uses a single simulation model to simulate the physical space. When faced with large-scale problems, the simulation speed is slow and it is difficult to apply to today's increasingly complex processing scenarios. However, the results obtained by running the simplified simulation model may be very different from the actual production system and have no reference significance.

However, it is often possible to build multiple simulation models of different fidelity levels for the system under study. Among them, the high-fidelity simulation model with a high degree of system reduction can accurately predict the effect of the solution, but the simulation calculation of complex systems is expensive and may be very time-consuming; the low-fidelity simulation model is faster to solve, but the simulation results are reliable. The performance is relatively low, and may have significant bias and randomness, making it impossible to accurately assess the effect of candidate solutions.

In order to ensure the model accuracy and solution speed as much as possible, it can be considered to solve the problem quickly by using different fidelity models. This method is called multi-precision modeling or multi-fidelity modeling, that is, in a reasonable abstraction. On the basis of and simplification, a low-fidelity model that is easy to calculate is established to solve the entire solution space, and the solutions are sorted and then sampled. , this method may reduce the time to find the optimal solution using a high-fidelity model.

At present, the research and application of multi-fidelity simulation optimization methods are mostly concentrated in continuous system simulation, and there are very few related applications and research in discrete system simulation. On the other hand, for complex production planning problems, even if a low-fidelity simulation model is used, the running time of the entire solution space is difficult to ignore, which leads to a long search time in the solution space and affects decision-making efficiency.

SUMMARY OF THE INVENTION

In view of the above defects or improvement needs of the prior art, the present invention provides a multi-fidelity simulation optimization method applied to workshop planning and production, the purpose of which is to use the simulation optimization method based on the multi-fidelity simulation model to optimize the production planning of workshop products , so as to solve the technical problem that the actual production workshop product launch plan is only based on the experience of the management personnel.

In order to achieve the above object, according to one aspect of the present invention, a multi-fidelity simulation optimization method applied to workshop planning and production is provided, and the method specifically includes the following steps:

Step 1. Establish a high-fidelity simulation model and a low-fidelity simulation model of the actual production system commissioning problem. The operation result obtained by the high-fidelity simulation model running the scheme X is denoted as h(X), and the low-fidelity simulation model The operation obtained by running the scheme x The result is recorded as l(x);

Step 2. Use the low-fidelity simulation model to perform a GA search, so as to perform a low-fidelity rough solution space search for the specific production problem, and obtain a low-fidelity search solution space set;

Step 3. Reorder the low-fidelity search solution space set in step 2 according to the pros and cons of the solutions;

Step 4. Sampling from the reordered solution space set to form the best sampling subset;

Step 5. Use the high-fidelity model to run the best sampling subset N _j and select the best plan to obtain the best production plan.

Further, the step 1 includes the following sub-steps:

Step 1.1: Establish a high-fidelity simulation model of the real scene according to the production operation and technological process characteristics of the actual production system, and set the scheduling rules and related parameters in the running process in the high-fidelity simulation model according to the actual production;

Step 1.2: Simplify the production operation and technological process characteristics of the actual production system, retain the modeling of bottleneck problems in the system, and replace other non-critical processes and resources with unlimited capacity to obtain a low-fidelity simulation model; The scheduling rules and related parameters are set to be the same as the high-fidelity simulation model.

Further, the throughput of the non-bottleneck process is replaced by the throughput time in the low-fidelity simulation model.

Further, the step 2 includes the following sub-steps:

Step 2.1: Set the total budget number M _max for low-fidelity search, the number P of GA populations, and initialize the evolutionary algebra as r=0;

Step 2.2: Generate the solution set {x _r1 , x _r2 , ..., x _rp } of the initial population, where x _rp represents the p-th solution obtained by the evolution of the r-th generation; use the low-fidelity simulation model to run and get the solution The set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )}, l(x _rp ) is the low-fidelity simulation solution corresponding to x _rp ;

Step 2.3: Set the evolutionary algebra as r=r+1; the solution set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )} uses the elite selection rule to select the cross population 1, and then the solution Set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )}, shuffle the order, use the elite selection rule to select cross population 2, use cross population 1 and cross population 2 for genetic evolution, and obtain The new solution set x _r ={x _r1 , x _r2 , . . . , x _rp };

Step 2.4: Determine whether the low-fidelity search budget has been used up, that is, determine whether there is r*P<M _max , if the inequality is established, repeat step 2.3, otherwise go to step 2.5;

Step 2.5: Collect all schemes in the evolution process to form a low-fidelity search scheme set {x ₁ , x ₂ , ..., x _r , ..., x _M }, M is the total number of solutions, and obtain The corresponding set of low-fidelity search solution spaces {l(x ₁ ), l(x ₂ ), ..., l(x _r ), ..., l(x _M )}.

Further, the elite selection rule in step 2.3 is as follows: select two unselected individuals in the population in sequence, compare the results of the two individuals, and select a better individual until all individuals in the population are selected. Pass;

The genetic evolution method in step 2.3 is as follows: using the idea of chromosome crossover and mutation in the genetic algorithm, use the two-point crossover method to crossover the two schemes to achieve global search, and use the single-point mutation method to mutate the two schemes. , to implement neighborhood search.

Further, step 3 includes the following substeps:

Step 3.1: Ordinal Conversion

Sort the low-fidelity search solution space set {l(x ₁ ), l(x ₂ ), ..., l(x _M )} according to the size of the solutions, and form the ordinal-transformed solutions from the best to the next space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )};

Step 3.2: Divide the ordinal transformed solution space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )} into K subsets Θ _j , j=1, ... , K, then each subset Θ _j contains N solutions, M=N*K.

Further, in step 4, the following optimal sampling strategy is adopted for sampling:

Step 4.1: Set the high-fidelity search budget N _max , the number of initial samples N ₀ and the total number of incremental samples Δ; set the evolutionary algebra to r=0;

个样本并通过高保真仿真模型运行得到运行结果；Step 4.2: Randomly choose in each subset Θ _j

samples and run the high-fidelity simulation model to get the running results;

则跳转到步骤4.5，否则增加Δ个预算，根据公式

j≠l≠b和

计算获得每个子集新的预算结果

j，n，l＝1，...，K；每个子集新增的预算数量为

δ_b，j表示子集Θ_b和Θ_j之间的平均差，δ_b，l表示子集Θ_b和Θ_l之间的平均差，σ_j表示Θ_j的标准差，σ_l表示Θ_l的标准差，σ_b表示Θ_b的标准差，N_l表示分配给Θ_l的高保真度评估预算的数量；Step 4.3: If

Then jump to step 4.5, otherwise add Δ budget, according to the formula

j≠l≠b and

Compute to get new budget results for each subset

j,n,l=1,...,K; the new budget amount for each subset is

_{δb , j} is the mean difference between subsets Θb and Θj, _δb , _l is the mean difference between subsets Θb and _Θl , _σj is the standard deviation of _Θj , _σl is _Θl The standard deviation of , σ _b is the standard deviation of Θ _b , and N _l is the amount of high-fidelity evaluation budget allocated to Θ _l ;

Step 4.4: Continue to randomly select N _rj samples from the subset Θ _j ; set the evolutionary algebra as r=r+1;

Step 4.5: Obtain the best sampled subset N _{j corresponding to each subset Θ j .}

Further, step 5 includes the following substeps:

j＝1，...，K，使用步骤1.1中的高保真模型及相应规则和参数运行，获得高保真的解空间集

Step 5.1: Collect N _j in step 4.5 to form a high-fidelity search optimal sampling set

j=1,...,K, run using the high-fidelity model from step 1.1 and the corresponding rules and parameters to obtain a high-fidelity solution space set

Step 5.2: From the solution space set in step 5.1, the solution with the smallest running time is selected as the final solution, that is, the best commissioning plan.

In order to achieve the above object, according to another aspect of the present invention, a computer-readable storage medium is provided, and a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the above-mentioned items are realized. Methods.

In order to achieve the above object, according to another aspect of the present invention, there is provided a multi-fidelity simulation and optimization device applied to workshop planning and production, comprising the aforementioned computer-readable storage medium and a processor, and the processor is used for calling and A computer program stored in a computer-readable storage medium is processed.

In general, compared with the prior art, the above technical solution conceived by the present invention firstly establishes high-fidelity and low-fidelity simulation models according to the actual production system; The solution space of the production planning problem is searched, and the sampling method with ordinal transformation and optimal sampling is used for sampling according to the results obtained by running the low-fidelity model to obtain the optimal sampling set. Finally, run the optimal sampling set through the high-fidelity simulation model. Since the high-fidelity model is very similar to the actual production system in the workshop, using the high-fidelity model to run the production plan in the optimal sampling set can obtain relatively accurate results, so this result can be used as According to the selection of the optimal production plan. Based on the above concept, the present invention can achieve the following beneficial effects.

(1) A multi-fidelity simulation optimization method applied to workshop planning and production provided by the present invention has strong practicability. On the basis of the traditional multi-fidelity simulation optimization framework, the genetic algorithm is used to accelerate the low-fidelity model solution space It can improve the efficiency of low-fidelity simulation model search problem solution space, improve decision-making efficiency, and help to optimize decision-making in discrete manufacturing industry.

(2) The present invention designs a low-fidelity simulation model based on the DBR theory, which has been verified to have strong consistency with the operation results of the high-fidelity simulation model under the same problem, so it is a very convenient and effective low-fidelity modeling method. .

(3) The present invention successfully applies the improved multi-fidelity simulation optimization method to the discrete production planning optimization problem. The method framework is easy to adjust according to the actual production problems of the enterprise, and is flexible and practical.

Description of drawings

Figure 1 is a flow chart of the multi-fidelity simulation optimization method applied to workshop planning and production;

(a) and (b) of FIG. 2 are schematic diagrams of two crossover processes of GA applied to the multi-fidelity optimization method in the solution space search process;

(a) and (b) of Figure 3 are schematic diagrams of two mutation processes of GA applied to the multi-fidelity optimization method in the solution space search process;

Fig. 4 is the workshop data report figure of a practical case provided by the present invention;

FIG. 5 is a schematic diagram of a high-fidelity simulation model abstracted by the present invention based on the actual case of FIG. 4 .

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

Figure 1 is a flow diagram of a multi-fidelity simulation optimization method applied to shop floor planning and production, which can be used more specifically in a manufacturing environment. The problem of workshop planning and production is described as follows: There are n workpieces in the workshop that need to be processed. Due to the limitation of workpiece delivery and workshop resources, different workpieces need to choose different delivery times. The workshop planning and production problem needs to select the release time of each workpiece. And use the known scheduling rules for scheduling to form a complete scheduling scheme. The goal of this problem is generally to minimize the total machining time. For the basic concepts of high-fidelity models and low-fidelity models, please refer to Xu J, Zhang S, Huang E, et al. MO 2 TOS: Multi-Fidelity Optimization with Ordinal Transformation and Optimal Sampling [J]. Asia Pacific Journal of Operational Research, 2016, 33(3):1650017. The present invention will not be further introduced.

Preferably, the high-fidelity model is constructed using simulation modeling software SIMIO. Taking the mixed-flow processing workshop of shaft parts as an example, according to the workshop production status and equipment, the data structure diagram of the simulation model of the mixed-flow processing workshop of shaft parts as shown in the figure below is established, which reflects the basic information of each object in the workshop and its association. In order to facilitate real-time adjustment according to workshop production data, layout changes, equipment changes, etc. in practical applications, the simulation model can be consistent with the real scene of the workshop.

For ease of understanding, a more detailed introduction to the system objects and attributes of simulation modeling is given below with an actual case shown in Figure 4:

(1) Order object

The order object is the order entity in the workshop. It is processed and processed as a collection of production objects. The main attributes of the order object are:

Table 1 - Order Object Properties Table

(2) Parts products

The part product can be integrated by the order object in the data model, but its concept object and related data still exist, which are used to represent the type of processed product and associated with the process route. Its main attributes are:

Table 2 - Parts Product Attribute Table

(3) Process route

The process route is the processing process route table of the product, including all the process routes in the workshop and outside the workshop. On-site scheduling requires the relevant requirements in the sequential process route. The main attributes of the process route table are:

Table 3 - Routing attribute table

(4) Production line

The workshop production mode is mixed flow production, but there are many mixed production lines in the actual process, and the products and production lines do not strictly correspond. The main attributes of the production line are:

Table 4 - Production line attribute table

property name type of data illustrate ★Production line name String Namely the production line name, which is unique workstation equipment String The workstation equipment belonging to the production line Production line equipment String Equipment belonging to the line management

(5) Logistics resources

Since other auxiliary resources in the workshop have less constraints on production, only one kind of logistics resource object is considered in the data model. Workstation equipment is a transport tool and container for material transfer in the workshop. Its main attributes are:

Table 5-Logistics resource attribute table

(6) Processing equipment

The processing equipment object is the processing equipment entity inside and outside the workshop. This entity includes processing equipment, testing equipment, testing area, fitter area, other outsourcing workshops and other actual or conceptual objects identified by the process. Its main attributes are:

Table 6 - Processing Equipment Attribute Table

On the basis of the above simulation framework and data relationship table, a high-fidelity digital simulation model can be established for the mixed-flow machining workshop of shaft parts.

Due to the complex production status of the workshop, the incomplete statistics of various types of information, and the loss of information caused by abstract modeling of workshop objects, it is unrealistic to fully restore the real scene. Therefore, it is necessary to make a certain degree of abstraction and reasonable assumptions about the mixed-flow machining workshop of shaft parts.

Therefore, the following basic assumptions are made for the simulation model of the mixed-flow machining workshop system for shaft parts:

1) Assuming that the equipment and workers can continue to work during the work period, under normal circumstances, there will be no shutdown due to various reasons; if there is an abnormal situation, another scenario arrangement will be considered;

2) It is assumed that the processing time of all products is the same as the standard working hours set by the workshop, and the influence of other factors such as workers' skill proficiency and equipment conditions on the working hours is normally not considered; if necessary, special arrangements shall be made to consider;

3) Assuming that the defective product rate is zero, the series of impacts brought about by the output of defective products are not considered for the time being; under certain circumstances, other scenarios will be considered for the impact of quality fluctuations;

4) It is assumed that the outsourcing process can be delivered and returned in time, or the service level is in a certain distribution range; the capacity and scheduling of the outsourcing unit are not considered for the time being;

5) Assuming that the capacity of the buffer area in the workshop can be changed, the strict restrictions on the actual space in the workshop are not considered for the time being; when there are rigid constraints on space resources, other scenarios should be considered.

Based on the above basic assumptions and the simulation framework, the workshop is simulated and modeled. The layout of the simulation model is basically arranged according to the CAD layout provided by the workshop and combined with the actual situation on site. The logistics path in the model is planned according to the actual situation of the workshop. There is only one entrance in the workshop, and a main channel is formed vertically leading to the precision room. There is one horizontal logistics channel inside the precision room, and two horizontally outside the precision room. There are two longitudinal channels at the end. The high-fidelity simulation model of the workshop is shown in Figure 5, and the black solid line segment in the figure represents the logistics path of the workshop.

Preferably, the low-fidelity simulation model is usually obtained by simplifying and abstracting the low-fidelity model. This patent provides a low-fidelity simulation model establishment method based on constraint theory. Taking the mixed-flow processing workshop for shaft parts as an example, through statistical analysis of the equipment load in the workshop, the bottleneck equipment and the equipment that may become a temporary bottleneck are retained. Other equipment that has sufficient capacity or is idle most of the time can be reduced to a single unit of infinite capacity, using a fixed processing time instead. In this patent, a low-fidelity simulation model is established according to the above-mentioned method for establishing a high-fidelity simulation model combined with the bottleneck theory.

Aiming at the problem of the planned production of the workshop, the multi-fidelity simulation optimization process applied to the planned production of the workshop mainly includes the following implementation steps:

Step 1. Establish a high-fidelity model and a low-fidelity model of the actual production system commissioning problem

Step 1.1 Establish a high-fidelity simulation model that is almost the same as the real scene according to the production operation and technological process characteristics of the actual production system, and set the scheduling rules and related parameters in the running process in the model according to the actual production. The running result obtained by the high-fidelity simulation model running scheme x is denoted as h(x).

Step 1.2 simplifies the production operation and technological process characteristics of the actual production system according to the DBR theory, and highlights the modeling of the bottleneck problem in the system. Other non-critical processes and resources can be replaced with unlimited capacity. Set the scheduling rules and related parameters in the low-fidelity simulation model to be the same as the high-fidelity simulation model. The running result obtained by running the scheme x of the low-fidelity simulation model is denoted as l(x).

Step 2. Using the low-fidelity simulation model established in Step 1.2, combined with the search process of GA, perform a low-fidelity general solution space search for the specific production problem.

Step 2.1 Set parameters such as the total budget number M _max of low-fidelity search, the number of GA populations P and so on, and set the evolutionary algebra as r=0.

Step 2.2 Generate solution set {x _r1 , x _r2 , ..., x _rp } of initial population, run with low fidelity simulation model and get solution set {l(x _r1 ), l(x _r2 ), .. ., l(x _rp )}.

Step 2.3 Set the evolutionary algebra as r=r+1. The solution set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )} uses the elite selection rule to select cross population 1, then the solution set {l(x _r1 ), l(x _r2 ) , ..., l(x _rp )} order is scrambled, use the elite selection rule to select cross population 2, use cross population 1 and cross population 2 and perform genetic evolution to obtain a new solution set {x _r1 , x _r2 , ..., x _rp }.

Step 2.4 judges whether the low-fidelity search budget has been used up, that is, judges whether there is rP<M _max , if the inequality is established, repeat step 2.3, otherwise go to step 2.5.

Step 2.5 Collect all schemes in the evolution process to form a low-fidelity search scheme set {x ₁ , x ₂ , ..., x _M }, and obtain the corresponding low-fidelity search solution space set {l(x ₁ ) , l(x ₂ ), ..., l(x _M )}.

Step 3. The solution space set {l(x ₁ ), l( _{x 2} ), .

Step 3.1 Ordinal conversion: Sort the low-fidelity search solution space set {l(x ₁ ), l(x ₂ ), ..., l(x _M )} according to the size of the solution, forming the results from the best to the next Ordinal transformed solution space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )}.

Step 3.2 Divide the ordinal transformed solution space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )} into K subsets Θ _j (j=1, ..., K), then each subset Θ _j contains N solutions.

Step 4 uses the optimal sampling strategy to sample the ordinal transformed solution space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )} in step 3.1 to form the optimal sampling set.

Step 4.1 Set the high-fidelity search budget N _max , the initial number of samples N ₀ and the total number of incremental samples Δ. Set the evolutionary algebra to r=0.

个样本并运行得到结果Step 4.2 Randomly select in each subset Θ _j

samples and run to get the result

则跳转到步骤4.5，否则增加Δ个预算，根据公式

和

计算获得每个子集新的预算结果

每个子集新增的预算数量为

Step 4.3 If

Then jump to step 4.5, otherwise add Δ budget, according to the formula

and

Compute to get new budget results for each subset

The number of new budgets added to each subset is

Step 4.4 continues to randomly select N _rj samples from the subset Θ _j . Set the evolutionary algebra as r=r+1.

Step 4.5 obtains the best sampling subset N _{j corresponding to each subset Θ j .}

Step 5 uses the high-fidelity model to run the best sampled subset N _j and select the best solution.

使用步骤1.1中的高保真模型及相应规则和参数运行，获得较精确的解空间集

Step 5.1 Collect N _j (j=1,...,K) in step 4.5 to form the best sampling set for high-fidelity search

Run with the high-fidelity model from step 1.1 and the corresponding rules and parameters to obtain a more accurate set of solution spaces

Step 5.2 selects the solution with the smallest running time as the final solution from the solution space set in step 5.1.

Figure 2 is a schematic diagram of the crossover process of GA applied to the multi-fidelity optimization method in the solution space search process. Individuals 1 and 2 shown in the figure are solutions to the workshop production planning problem. Each cell and the number in the cell indicate a type of workpiece and the point in time at which it was selected for production. The crossover process is shown in the figure. Two crossover points are randomly selected, and the solution segments between the two individual crossover points are crossed to form a new solution.

Figure 3 is a schematic diagram of the mutation process of GA applied to the multi-fidelity optimization method in the solution space search process. The individual 1 shown in the figure is the solution in the shop production planning problem. The mutation process is shown in the figure. A mutation point is randomly selected, and the production time point corresponding to the mutation point is randomly modified to other production points to form a new solution.

Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, etc., All should be included within the protection scope of the present invention.

Claims (5)

1. a multi-fidelity simulation optimization method that is applied to workshop planning and put into production, is characterized in that, this method specifically comprises the following steps: Step 1. Establish a high-fidelity simulation model and a low-fidelity simulation model of the actual production system commissioning problem. The operation result obtained by the high-fidelity simulation model running the scheme X is denoted as h(X), and the low-fidelity simulation model The operation obtained by running the scheme x The result is recorded as l(x); Step 2. Use the low-fidelity simulation model to perform a GA search, so as to perform a low-fidelity rough solution space search for the specific production problem, and obtain a low-fidelity search solution space set; Step 3. Reorder the low-fidelity search solution space set in step 2 according to the pros and cons of the solutions; Step 4. Sampling from the reordered solution space set to form the best sampling subset; Step 5. Use the high-fidelity model to run the best sampling subset N _j and select the best plan to obtain the best production plan; The step 1 includes the following sub-steps: Step 1.1: Establish a high-fidelity simulation model of the real scene according to the production operation and technological process characteristics of the actual production system, and set the scheduling rules and related parameters in the running process in the high-fidelity simulation model according to the actual production; Step 1.2: Simplify the production operation and technological process characteristics of the actual production system, retain the modeling of bottleneck problems in the system, and replace other non-critical processes and resources with unlimited capacity to obtain a low-fidelity simulation model; The scheduling rules and related parameters are set to be the same as the high-fidelity simulation model; The step 2 includes the following sub-steps: Step 2.1: Set the total budget number M _max for low-fidelity search, the number P of GA populations, and initialize the evolutionary algebra as r=0; Step 2.2: Generate the solution set {x _r1 , x _r2 , ..., x _rp } of the initial population, where x _rp represents the p-th solution obtained by the evolution of the r-th generation; use the low-fidelity simulation model to run and get the solution The set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )}, l(x _rp ) is the low-fidelity simulation solution corresponding to x _rp ; Step 2.3: Set the evolutionary algebra as r=r+1; the solution set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )} uses the elite selection rule to select the cross population 1, and then the solution Set {l(x _r1 ), l(x _r2 ), ..., l(x _rp )}, shuffle the order, use the elite selection rule to select cross population 2, use cross population 1 and cross population 2 for genetic evolution, and obtain The new solution set x _r ={x _r1 , x _r2 , . . . , x _rp }; Step 2.4: Determine whether the low-fidelity search budget has been used up, that is, determine whether there is r*P<M _max , if the inequality is established, repeat step 2.3, otherwise go to step 2.5; Step 2.5: Collect all schemes in the evolution process to form a low-fidelity search scheme set {x ₁ , x ₂ , ..., x _r , ..., x _M }, M is the total number of solutions, and obtain the corresponding low-fidelity search solution space set {l(x ₁ ), l(x ₂ ), ..., l(x _r ), ..., l(x _M )}; The elite selection rules in step 2.3 are as follows: select two unselected individuals in the population in sequence, compare the results of the two individuals, and select a better individual until all individuals in the population have been selected; The genetic evolution method in step 2.3 is as follows: using the idea of chromosome crossover and mutation in the genetic algorithm, use the two-point crossover method to crossover the two schemes to achieve global search, and use the single-point mutation method to mutate the two schemes. , to achieve neighborhood search; Step 3 includes the following sub-steps: Step 3.1: Ordinal Conversion Sort the low-fidelity search solution space set {l(x ₁ ), l(x ₂ ), ..., l(x _M )} according to the size of the solutions, and form the ordinal-transformed solutions from the best to the next space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )}; Step 3.2: Divide the ordinal transformed solution space set {l(x _OT1 ), l(x _OT2 ), ..., l(x _OTM )} into K subsets Θ _j , j=1, ... , K, then each subset Θ _j contains N solutions, M=N*K; In step 4, the following optimal sampling strategy is used for sampling: Step 4.1: Set the high-fidelity search budget N _max , the number of initial samples N ₀ and the total number of incremental samples Δ; set the evolutionary algebra to r=0; Step 4.2: Randomly choose in each subset Θ _j

A total of K groups of running results are obtained by running the high-fidelity simulation model; Step 4.3: If

Then jump to step 4.5, otherwise add Δ budget, according to the formula

and

Compute to get new budget results for each subset

j, b, l = 1, ..., K; the number of new budgets added to each subset is

_{δb , j} is the mean difference between subsets Θb and Θj, _δb , _l is the mean difference between subsets Θb and _Θl , _σj is the standard deviation of _Θj , _σl is _Θl The standard deviation of , σ _b is the standard deviation of Θ _b , and N _l is the amount of high-fidelity evaluation budget allocated to Θ _l ; Step 4.4: Continue to randomly select N _rj samples from the subset Θ _j , one for each

Add N _rj samples; set the evolutionary algebra to r=r+1, repeat step 4.3; Step 4.5: Obtain the best sampled subset N _{j corresponding to each subset Θ j .} 2. a kind of multi-fidelity simulation optimization method that is applied to workshop planning and put into production as claimed in claim 1 is characterized in that, step 5 comprises following substep: Step 5.1: Collect N _j in step 4.5 to form a high-fidelity search optimal sampling set

j=1,...,K, run using the high-fidelity model from step 1.1 and the corresponding rules and parameters to obtain a high-fidelity solution space set

Step 5.2: From the solution space set in step 5.1, the solution with the smallest running time is selected as the final solution, that is, the best commissioning plan. 3 . The multi-fidelity simulation optimization method applied to workshop planning and production as claimed in claim 1 , wherein in the low-fidelity simulation model, the production capacity of the non-bottleneck process is replaced by the throughput time. 4 . 4. A computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented. 5. a kind of multi-fidelity simulation optimization equipment applied to workshop planning and put into production, is characterized in that, comprises the computer-readable storage medium and processor as claimed in claim 4, and the processor is used for calling and processing in the computer-readable storage medium. Stored computer program.

Images