Lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm

Abstract

The invention relates to a lean production multi-objective optimization intelligent control method based on an adaptive simulation genetic algorithm, and the method specifically comprises the following steps: 1, constructing a multi-objective optimization algorithm under a constraint condition that n decision vectors belong to an n * m cube R matrix space in an X decision unit, all sub-objective functions of the material-to-weight ratio, the temperature, the CO2, the NH3 and the H2S reach the minimum lean production multi-objective optimization premise as much as possible; a Pareto optimal solution of multi-target minimization is determined by adopting segmentation-parallel selection-combination in a swarm intelligence hybrid optimization algorithm; measuring the concentration ratio of individuals in the population by adopting three parameters, namely a maximum adaptive value fitmax, a minimum adaptive value fitmin and an average adaptive value fitave of the population, and exploring by adopting a first maximum variation rate operation, a second minimum crossover rate operation and a second copy operation; performing development by firstly performing minimum variation rate operation, then performing maximum crossover rate operation and then performing copying operation; the improved Monte Carlo simulation is adopted, the global optimization ability of the genetic algorithm is improved, and the core of the adaptive simulation genetic algorithm is integrated.

Description

Background Art

[0001] The present invention relates to the technical field of a multi-objective optimization intelligent control method for lean production based on an adaptive simulation genetic algorithm, and in particular to a multi-objective optimization intelligent control method in lean product production based on an adaptive simulation algorithm.

[0002] The existing single swarm intelligence algorithm is a single genetic algorithm (GA) and particle swarm optimization algorithm (PSO) based on the biological evolution inspiration theory. Due to the slow convergence speed in the later stage, premature maturity and easy to fall into local optimization, the optimization accuracy is not ideal. We want to use a mixed mode of multiple methods, split-parallel selection-merge operation, to overcome the low convergence efficiency and easy to fall into local optimization, and improve the multi-objective optimization accuracy. For traditional extensive production: relying on excessive addition of vitamins A, E, B2, glutamic acid, serine, invertase, illegal use of hydroxymethylepinephrine, sex hormones, gonadotropins, zearalanol (estrogen), hypnotic sedatives (diazepines, methaqualone), and heavy metal powder to promote the production of products, the heavy metal chemical drugs and antibiotics are seriously exceeded; the zinc content of production waste exceeds the standard by 300 times, and nitrate ammonia and heavy metal residues pollute the environment. The use of this essential product seriously endangers people's physical and mental health. There is an urgent need to develop an intelligent control system for multi-objective optimization in lean product production. The multi-objective control minimization of production precision feeding, temperature and humidity of the production environment, and carbon dioxide (CO2), ammonia (NH3), and hydrogen sulfide (H2S) gas concentrations is applied to the optimal integrated environment of lean product production, avoiding the complex state of data set control nonlinearity, time-varying, and saturation, which can not only solve the bottleneck problem of traditional production efficiency and improve the intrinsic quality of lean products. Therefore, in response to the above problems, a lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm is proposed Summary of the invention

[0003] In order to make up for the shortcomings of the prior art and solve the complexity implied by nonlinear multi-objective optimization, the intelligent control of the physical space integrated environment with relatively independent objects of product production and the inability to effectively parallel distributed optimization processing tasks for the frequent regulation of multi-objective variables and the relatively independent physical space integration environment of product production objects, the multi-modal complex information leads to a large feedback delay in the automatic production control system, a low feed-to-output ratio in lean product production, and low efficiency of human-machine intervention and control. The present invention proposes a lean production multi-objective optimization intelligent control method based on an adaptive simulation genetic algorithm.

[0004] The technical solution adopted by the present invention to solve the technical problem is: the lean production multi-objective optimization intelligent control method based on the adaptive simulation genetic algorithm described in the present invention comprises the following steps:

[0005] Step 1: Build satisfaction Under the constraints, V-min vector is minimized, so that the vector objective function f(x) = [f1(x), f2(x), …, f n (x)] T Each sub-objective function in is minimized as much as possible, and the lean production multi-objective optimization problem is:

[0006]

[0007] Where x∈X represents n decision vectors in X decision units. It indicates that the objective function satisfies the mapping relationship from X to f(x) of multi-objective optimization under the constraint of n×m cube R matrix space; constructs an algorithm to analyze whether the project has the conditions for achieving the minimum value of multi-objective optimization, divides the sub-objective function into some sub-groups, and assigns a sub-objective function to each sub-group. Each sub-group independently selects some individuals with high fitness to form a new sub-group, merges all the newly generated sub-groups into a complete group, and then performs a series of crossover and mutation operations in this group to generate the next generation of complete groups. In this way, the "split-parallel selection-merge" operation is continuously performed, and finally the Pareto optimal solution of the multi-objective optimization problem can be obtained.

[0008] Step 2: When V-minf(x) can find the minimum value for multiple objectives, and there is more than one minimum value, create the objective function set for the application as follows:

[0009]

[0010] The lean production multi-objective optimization intelligent control method based on the adaptive simulation algorithm is integrated to support intelligent operation control, and the final multi-objective lean product production intelligent control minimum value is output; D(h) is the target weight gain, B(h) is the weight of the consumed material, and the first objective function f1(x) is optimized to obtain the minimum value. The denominator is the weight of the product quality increase in lean product production compared to the weight of the material input to obtain the maximum value, indicating that more product quality can be obtained per unit input; σ and ρ are the influence coefficients of the target weight gain and the consumed material under the temperature and suitable air, h=1,2…V, V is the sampling time; H set is the temperature setting value, H i is the system output temperature value, i=1,2,…,W, W is the number of sampling times, and f2(x) is the computer system intelligent control objective function to find the minimum value. This fitness function comprehensively considers multiple performance indicators such as the system's rise time, overshoot, and steady-state error. Similarly, K set 、N set , S set are the setting values ​​of the carbon dioxide, ammonia and hydrogen sulfide concentration calculation models, K i 、N i , Si For the output values of the computer system optimized and controlled for the concentrations of carbon dioxide / ammonia / hydrogen sulfide respectively, j = 1, 2, …, Q, where Q is the number of sampling times of carbon dioxide; t = 1, 2, …, M, where M is the number of sampling times of ammonia; p = 1, 2, …, R, where R is the number of sampling times of hydrogen sulfide concentration, f3(x), f4(x), and f5(x) respectively obtain the minimum values through optimization of the objective functions for the concentrations of carbon dioxide, ammonia, and hydrogen sulfide.

[0011] Preferably, in the third step: Only by gradually integrating the fuzzy adaptive simulated genetic algorithm can the multi-objective optimization minimum value solution be obtained; aiming at the drawbacks of the existing adaptive genetic algorithm, the calculation methods of the crossover rate P d and the mutation rate P e are improved to overcome the defect that in the early stage, the relatively simple calculation of P d and P e , and the algorithm of excellent individuals tending not to change their own states cannot jump out of the local optimization defect; when fit ave and fit max , the closer the two are, the more convergent the algorithm tends to be; when fit ave / fit max > a, and 0.5 < a < 1, when a → 0.5, it is easier to determine that the individuals are concentrated; (fit max -fit min ) is closer to 0, the more concentrated the population is, and the greater the possibility of the local optimal solution of the algorithm, otherwise it is more dispersed; when fit min / fit max > b and 0 < b < 1; the closer b is to 0, the easier it is to judge that the individuals are concentrated; when fit ace / fit max > a and fit min / fit max > b, it is judged that the population is concentrated. At this time, the crossover rate P d and the mutation rate P e change adaptively according to the concentration degree of the population, so that the algorithm automatically adjusts to the natural selection mechanism; when this condition is not met, it is judged that the population is dispersed; the calculation formulas for the values of the improved crossover rate P d and the mutation rate P e are as follows:

[0012]

[0013]

[0014]

[0015] In the formula, fit max , fit min , fit aveare the maximum, minimum and average values ​​of fitness respectively, c is the conservative estimate of the limit of the objective function, c>0, f(x) is the objective function, k1 and k2 are adaptive control parameters, and they take fixed values ​​in the interval (0, 1); in the early stage of the improved adaptive genetic algorithm (IAGA), the individuals are more dispersed and the mutation probability is lower, which can ensure that the population converges to the global optimal solution; when the individuals in the population tend to be concentrated, the larger mutation rate (P e ), which increases the diversity of the population and ensures the evolution of the population while enhancing the local search capability of the algorithm.

[0016] Optimally, the fourth step: improve the fitness scale transformation, solve the problem that the proportional selection strategy is easy to select "super individuals" to cause premature convergence and affect the convergence speed of the algorithm, and calculate the fitness transformation f'(x i )as follows:

[0017]

[0018] Among them, when Z→1, the new fitness gap between different individuals is smaller than the original fitness gap, which reduces the selection probability of the super individual and increases the selection probability of other individuals, which is conducive to maintaining the diversity of the group. When the downward trend of Z slows down, the group diversity in the early stage of evolution can be better maintained; when Z quickly approaches 0, f′(x i ) selection sensitivity, increasing the probability of excellent individuals being selected and promoting the algorithm to accelerate convergence; where f′(x i ) is the fitness transformation, also called the transformed fitness, fit max 、fit min 、fit ave They are the maximum, minimum and average values ​​of the population fitness respectively. Their calculation is the same as that in the third step. The present invention solves the problem of minimizing the objective function: fit(x i ) is the fitness function = negative objective function; g represents the current evolutionary generation, G represents the total evolutionary generation, a is 6, T is the temperature of the simulated annealing method, is the attenuation coefficient.

[0019] Preferably, the fifth step: after the public population performs a crossover operation on the excellent individuals, all individuals search according to the local search strategy, and the fitness is updated as follows:

[0020] fit(x i ) = fit(x i )+ω(1-K λ )(f m (x i )-fit max )

[0021] In the formula, ω represents the weight of evolutionary potential, K represents the temperature attenuation coefficient, λ represents the number of new solution searches, and f m (x i ) represents chromosome x i The optimal solution after local search; fit max is the maximum value of population fitness, and its calculation is the same as that in the third step; the present invention solves the problem of minimizing the objective function: fit(x i ) is the fitness function = negative objective function; by updating the fitness, the solution space around the individual can be explored, and a small number of chromosomes with a large population can be selected to enter the next generation, individuals with evolutionary potential can be discovered, and new solutions can be discovered; it not only enriches the diversity of the population to maintain the search range of the population, but also enables other excellent individuals to cross and hopefully produce better new individuals.

[0022] Preferably, step 6: the population enhances the local optimization ability of the algorithm by searching the neighborhood structure, and the neighborhood structure is calculated as follows:

[0023]

[0024] When an individual conducts a local search, the phenotype changes while its own genotype remains unchanged; if there are multiple neighborhood structures that make the current individual better than the global optimal solution, the neighborhood structure with the largest acceptance rate is selected to generate a new solution to enter the next hop; when the neighborhood structure of each hop is determined to generate a new solution to replace the current solution, when the individual finds a solution that is better than the global optimal solution through local search, it is determined that the individual has a greater evolutionary potential, and the probability of being able to search for the global optimal solution through evolution is also greater, and the fitness of the individual with evolutionary potential is adjusted to make it have a greater chance of entering the next generation; P i (max), P i (min) represent the decision variables P i The upper and lower bounds of the domain, i = 1, 2, ..., μ, μ is the dimension of the decision variable (i.e., the dimension of the problem), L i represents the neighborhood structure problem of 1 to μ-dimensional decision variables, l represents the gene length, δ is the range coefficient of the search space, and θ is the coefficient that controls the size of the neighborhood structure. When μ = 1, the function f(P i ) of chromosomes with a gene length of l, the binary code string represented by the decision variable P i The domain of [P i (min), P j (max)] is discretized to 2 l -1 uniform region. Each of these regions has a neighborhood structure, i.e., a length of L(P i ) segment; when μ=2, the task function f(P i , P i+1 ) represents the decision variable P i With Pi+1 The domain of is discretized into 2 l -1 uniform region, there are (2 l -1) 2 neighborhood structures, each of which has an area of ​​L(P i )×L(P i+1 ) plane matrix. Similarly, the k-dimensional problem includes (2 l -1) k Neighborhood structures, each of which represents a volume of A k-dimensional cube is constructed, thereby establishing a one-to-one correspondence between the target solution space of the problem being solved and the search space of the algorithm.

[0025] Preferably, the seventh step: if the population is stagnant or stuck in a local optimal solution during evolution, the evolution strategy should be changed according to the population evolution situation. The present invention designs an evolutionary generation R in which the optimal solution remains unchanged. f To measure the evolutionary state of the population, and calculate the probability of population switching strategy as follows:

[0026]

[0027] The probability formula of population transformation strategy fully explains the P ed Following the law of stagnant algebraic transformation, it is revealed that nonlinear probability can more reasonably control the population transformation evolution strategy; R t is the current evolutionary generation (also the generation of population stagnation), R is the total evolutionary generation, β is the control parameter, which is 6 in this case, R max is the maximum stagnation algebra.

[0028] Preferably, the eighth step: To make the adaptive simulation algorithm play a better role in the multi-objective control of lean product production, the Monte Carlo simulation algorithm is improved and integrated into the whole algorithm, in order to solve the complex parameter optimization problems in practical applications, enhance the algorithm's ability to identify data structure parameters, and transform the solution space equilibrium and parameter optimization of complex problems into constrained nonlinear programming problems:

[0029] Min f(X)X∈E n

[0030] stg i (X)≥0i=1,2,…,m

[0031] a j ≤x i ≤ b j j=1,2,…,n

[0032] Monte Carlo simulation is used to estimate the region {(x1,x2…xn )|x j ∈[a j ,b j ],j=1,2,…,n} randomly select several experimental point sets, then find the feasible point solution set from the experimental point set, and then select the minimum solution set that meets the conditions from the feasible point solution set;

[0033] The jth component x of the test point set j obey j ,b j ] uniform distribution within; that is:

[0034] x i =a j +S(b j -a j ),j=1,2,…,n

[0035] S is a random number uniformly distributed between [0,1]; let: P: total number of trials; MaxP: maximum number of trial points; K: total number of feasible points; MaxK: maximum number of feasible points; X * : the optimal point set generated by iteration; Q: the minimum value of the optimal feasible solution generated by iteration f(X * ), whose initial value is the maximum number of experiments that the computer can represent; first generate a random number as the initial test point set, and then randomly generate the jth component of the previous test point, that is, x i =a j +S(b j -a j ), the other components remain unchanged and a new test point is generated; in this way, only one new random component is needed to generate each new test point; the iteration ends when K>MaxK or P>MaxP.

[0036] Preferably, the ninth step: in order to make the lean production multi-objective optimization intelligent control method based on the adaptive simulation algorithm fully play its role, the local optimization search strategy process, the improved Monte Carlo simulation process, and the improved genetic algorithm process are organically integrated to integrate the technical core of the present invention; the adaptive simulation genetic algorithm for multi-objective optimization effectively solves the multi-objective optimization intelligent control tasks in lean product production.

[0037] Preferably, in the tenth step, during the operation of the ASGA algorithm, the present invention does not directly operate on the actual decision variables of the problem to be solved, but applies various genetic operations to the individual codes representing feasible solutions to achieve the purpose of optimization; the intrinsic quality improvement and quantity growth of lean production products are highly dependent on the centralized control environment. In the computer system for optimal control, the parameters of the lean production centralized control environment to be optimized are temperature, and the concentration setting values ​​of CO2, NH3, and H2S, which are required to achieve a certain accuracy, so a binary encoding method is adopted, and the mapping relationship between each parameter and the binary encoding is:

[0038] H i =H imin +binar / 2 L (H imax -H imin )

[0039] K i =K imin +binar / 2 L (K imax -K imin )

[0040] N i =N imin +binar / 2 L (N imax -N imin )

[0041] S i =S imin +binar / 2 L (N iamx -B imin )

[0042] Among them, H i is the temperature to be set, [H imin ,H imax ] is the temperature range, binar is a 2 L The binary integer represented by the longest binary string. Similarly, K i is the carbon dioxide concentration to be set, [K imin ,K imax ] is the minimum and maximum range of carbon dioxide concentration, N i is the ammonia concentration to be set, [N imin ,N imax ] is the minimum and maximum variation range of ammonia concentration; S i is the concentration of hydrogen sulfide to be set, [S imin ,S imax ] is the minimum and maximum variation range of hydrogen sulfide concentration; binar and 2 L Same as above.

[0043] The present invention is beneficial in that:

[0044] 1. The present invention provides technical support for finding optimal solutions to complex distributed problems by developing the potential of cooperation among implicit non-intelligent entities in solving distributed problems without centralized control or providing a global model. It is crucial to the application potential of situational awareness of complex information and multi-objective optimization control.

[0045] 2. The present invention overcomes the defects of existing single swarm intelligence algorithms such as genetic algorithm, ant colony algorithm, particle swarm algorithm, simulated annealing, hill climbing algorithm, etc., which are prone to local optimization and weak self-learning / self-organization / self-adaptation capabilities. In addition to being more capable of solving nonlinear, complex, multi-objective coexisting bottleneck problems that are difficult to solve with conventional optimization algorithms, it also has the following advantages:

[0046] (1) It is one or two orders of magnitude superior in terms of multi-objective optimization accuracy and iteration efficiency.

[0047] (2) Strong capabilities in distributed, parallel processing, and solving discrete problems.

[0048] (3) Strong global search, system analysis, and model expansion capabilities.

[0049] (4) It is more universal. We believe that in future work, it will be able to fully demonstrate its strong vitality in multi-objective optimization, distributed intelligent system modeling, multimodal information processing, robot control, decision support and simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0051] Figure 1 Multi-objective optimization parallel selection merge diagram.

[0052] Figure 2 Plot of the probability adaptation curve for changing policy.

[0053] Figure 3 Block diagram of the improved Monte Carlo simulation iteration steps.

[0054] Figure 4 Flowchart of the adaptive simulation-genetic hybrid optimization algorithm.

[0055] Figure 5(a) to Figure 5(f)Graphs of 1 to 6 test functions.

[0056] Figure 6(a) to Figure 6(f) Evolution curves of the three algorithms verified by 1 to 6 test functions.

[0057] Figure 7 Graph showing daily weight gain curve of lean products at different temperatures.

[0058] Figure 8 Curve of daily material consumption of lean products at different temperatures.

[0059] Fig. 9 Schematic diagram of optimal temperature control for lean products at four different weight stages.

[0060] Fig.10 Graph showing temperature control curve for lean product production.

[0061] Fig.11 CO2 control curve for lean product production.

[0062] Fig.12 Ammonia control curve for lean product production.

[0063] Fig.13 Hydrogen sulfide control curve for lean product production. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The lean production multi-objective optimization intelligent control method based on the adaptive simulation algorithm is to integrate the multi-objective factors of production objects and production environment required for the production of lean products. Its core supporting technology is the adaptive simulation genetic algorithm, which can represent:

[0066] ASGA=(C,E,H0,M,Φ,Γ,ψ,f′,R,P ed ,T)

[0067] Where C is the population encoding method, E is the population fitness evaluation function, H0 is the initial population, M is the population size, Φ is the selection operator, Γ is the crossover operator, ψ is the mutation operator, f′ is the proportional operation operator, R is the local search strategy operator, P ed is the probability of population switching strategy, and T is the termination condition of adaptive simulated genetic operation.

[0068] 1. Lean production multi-objective optimization problem:

[0069]

[0070] Where V-min represents vector minimization, that is, the vector objective function f(x) = [f1(x), f2(x), …, f n (x)] T Each sub-objective function in is minimized as much as possible.

[0071] Among them, x∈X, represents n decision vectors in X as decision units, It indicates that the objective function satisfies the mapping relationship of multi-objective optimization from X to f(x) under the constraint of n×m cube R matrix space; the objective function of integrated intelligent control of lean product production is as follows:

[0072]

[0073] The first objective function f1(x) is optimized to obtain the minimum value, and the denominator is maximized, indicating that a larger lean product weight is obtained per unit input. When f2(x) reaches the minimum value, the difference between the set temperature and the actual output temperature is the smallest, and the temperature fluctuation of the system intelligent control is not large; The optimization of each objective function has achieved the minimum value, indicating that the difference between the set CO2, NH3, H2S and the actual output CO2, NH3, H2S is the smallest, the system intelligent control CO2, NH3, H2S fluctuations are not large, and the integrated environment is excellent. Where D(l) is the target weight gain, D(h) is the weight of the consumed material, σ, ρ are the influence coefficients of the target weight gain and the consumed material under the temperature and suitable air, h = 1,2...V, V is the sampling time; H set is the temperature setting value, H i is the system output temperature value, i=1,2,…,W, W is the number of sampling times; this fitness function comprehensively considers multiple performance indicators such as the system's rise time, overshoot and steady-state error. Similarly, K set 、N set , S set are the setting values ​​of the carbon dioxide, ammonia and hydrogen sulfide concentration calculation models, K j 、N t , S p are the computer system output values ​​for optimizing the concentration control of CO2, NH3, and H2S, respectively; j = 1, 2, …, Q, where Q is the sampling number of carbon dioxide; t = 1, 2, …, M, where M is the sampling number of ammonia; and p = 1, 2, …, R, where R is the sampling number of hydrogen sulfide concentration.

[0074] 2. Perform parallel selection and merging of sub-goals to complete the output of multi-objective optimization integration results

[0075] Divide into several sub - populations according to the average of the sub - objective functions, and correspondingly assign a sub - objective function. Independently perform operations in their respective sub - populations to select some individuals with high fitness to form a new sub - population. Combine all the newly generated sub - populations into a complete population, and then perform a series of operations such as crossover and mutation in this population to generate the next - generation complete population. Continuously perform the operations of splitting - parallel selection - merging in this way, and finally the Pareto optimal solution of the multi - objective optimization problem can be obtained; see Figure 1 as shown.

[0076] 3. Gradually establish an adaptive simulated genetic algorithm. To obtain the overall optimal solution with multiple objectives minimized and the optimal solution of multi - objective optimization, gradually integrate the fuzzy adaptive simulated genetic algorithm to address the drawbacks of the existing algorithms; improve the variable crossover rate P d and the mutation rate P e to enhance the population's adaptive ability.

[0077] (1) Improve P d and P e to enhance the adaptive ability

[0078] to overcome the defect that the earlier relatively simple calculation of P d and P e , and the algorithm that excellent individuals tend not to change their own states cannot jump out of the local optimization defect; when fit ave and fit max , the closer the two are, the more the algorithm tends to converge; when fit ave / fit max >a, and 0.5 < a < 1, when a → 0.5, it is easier to determine that the individuals are concentrated; (fit max -fit min ) is closer to 0, the more concentrated the population is, and the greater the possibility of the local optimal solution of the algorithm, otherwise it is more dispersed; when fit min / fit max >b and 0 < b < 1; the closer b is to 0, the easier it is to judge that the individuals are concentrated; when fit ace / fit max >a and fit min / fit max >b, it is judged that the population is concentrated. At this time, the crossover rate P d and the mutation rate P e change adaptively according to the concentration degree of the population, so that the algorithm automatically adjusts to the natural selection mechanism; when this condition is not met, it is judged that the population is dispersed; the calculation formulas for the values of the improved crossover rate P d and the mutation rate P e are shown in (3) and (4):

[0079]

[0080]

[0081]

[0082] Where fit max 、fit min 、fit ave are the maximum, minimum and average values ​​of fitness respectively, c is the conservative estimate of the limit of the objective function, c>0, f(x) is the objective function, k1 and k2 are adaptive control parameters, and they take fixed values ​​in the interval (0, 1); in the early stage of the improved adaptive genetic algorithm (IAGA), the individuals are more dispersed and the mutation probability is lower, which can ensure that the population converges to the global optimal solution; when the individuals in the population tend to be concentrated, the larger mutation rate (P e ), which increases the diversity of the population and ensures the evolution of the population while enhancing the local search ability of the algorithm.

[0083] (2) Optimal preservation strategy

[0084] If Fmax i-1 >Fmax i ,(Fmax i is the highest fitness value in the contemporary population, Fmax i-1 is the highest fitness value of the previous generation population), randomly eliminate one individual in the contemporary group, and then add the individual with the highest fitness value of the previous generation to the contemporary; ensuring that excellent individuals are not eliminated, the algorithm can converge well.

[0085] (3) Improve the adaptive parallel mechanism

[0086] ① Adopting the neighborhood model: dividing a single population into multiple sub-populations, connecting them on the boundary according to their topological structure to exchange information, and handing each population over to different processors for independent evolution operations; this feature improves the iterative efficiency of the algorithm.

[0087] ②Parallel evolution strategy: Use multiple populations instead of a single population for parallel evolution. Each population evolves independently in a different way. After a certain iteration period, a certain number of excellent individuals are exchanged between various populations to expand the search range.

[0088] ③Parallel mechanism of exploration and development strategies

[0089] Table 1 Parallel mechanism strategy

[0090]

[0091] (4) Improved population competition strategy

[0092] The principle of survival of the fittest and death of the inferior is adopted to eliminate inferior individuals or populations and reproduce superior individuals or populations. First, the similarity of all individuals in the group is calculated, and individuals with fitness lower than the average fitness of the population and individual similarity higher than the similarity threshold η are selected with a smaller mutation probability P. ed When adjusting individuals, the population has evolved to a certain extent. The method of randomly generating new chromosomes is used to adjust similar individuals to solve the slow convergence speed and explore new areas of the solution space to create conditions for obtaining the optimal solution. The Hamming distance formula is used to determine the individual similarity calculation:

[0093]

[0094] Where ω k represents the weight of gene position k, and the present invention takes ω k =2 k ; Respectively represent x i and x j For the two codewords at positions k (k = 1, 2, ... l), the numerator of equation (6) is x i and x j The two vertices of the hypercube of two codewords are the weighted sum of the shortest distances of l groups in the hypercube of l-dimensional space formed by the codeword of l bits, and the denominator is x i and x j The two vertices of the hypercube of two codewords are in the (l-1)-bit codewords, which constitute the sum of the (l-1) groups of maximum distances in the hypercube of the (l-1)-dimensional space; since it is difficult for the hybridization of two similar individuals to produce relatively good individuals, in order to increase the diversity of hybrid offspring and produce better individuals, and promote the improvement of hybrid offspring; therefore, the Hamming distance judgment of the two selected individuals is required to better identify the differences between individuals. The individual similarity judgment is determined by the following formula:

[0095]

[0096]

[0097] Where g is the current evolutionary generation, G is the total evolutionary generation, γ is the distance threshold of individual similarity, γ2 and γ1 represent the larger threshold of individual similarity distance and the smaller threshold of individual similarity distance respectively. If the threshold is lower than γ, x is considered i ,x j If the approximation is similar, otherwise it is not similar. In the early stage of approximation, the γ value is large, and in the later stage of approximation, the γ value keeps decreasing to avoid affecting the convergence speed.

[0098] (5) Improve the transformation of adaptive scale

[0099] ①Choose strategy

[0100] Due to individual x i The probability of choosing is fit(x i ) is individual x i The fitness function is n, where n is the number of individuals in the population. The fitness scale transformation is to solve the problem that the proportional selection strategy is prone to selecting "super individuals" and causing premature convergence and affecting the convergence speed of the algorithm. The transformation f′(x i )as follows:

[0101]

[0102] In the formula, f′(x i ) is the fitness transformation, also called the transformed fitness, fit max 、fit min 、fit ave are the maximum, minimum and average values ​​of the population fitness respectively. Their calculation is the same as that of formula (5). The objective function of the present invention solves the minimization problem: fit(x i ) is the objective function of fitness function = negative, g represents the current evolutionary generation, G represents the total evolutionary generation, a is 6, T is the temperature of the simulated annealing method, is the attenuation coefficient. The fitness transformation avoids the algorithm from falling into "premature" convergence due to too many super individuals selected in the early stage. max -fit min becomes smaller, using f′(x i ) = fit(x i )+Zfit ave Solve the task of achieving convergence in the later stage of the algorithm and ensure that the population evolution is more stable. Z is the fitness transformation parameter. When Z→1, the new fitness gap between different individuals is smaller than the original fitness gap, which reduces the selection probability of super individuals and increases the selection probability of other individuals, which is conducive to maintaining population diversity. When the downward trend of Z slows down, the population diversity in the early stage of evolution can be better maintained; when Z quickly approaches 0, f′(x i ) selection sensitivity, increasing the probability of excellent individuals being selected and promoting the accelerated convergence of the algorithm.

[0103] ② Local search strategy

[0104] After the public population performs crossover operations on excellent individuals, all individuals search according to the local search strategy and update the fitness according to formula (10):

[0105] fit(x i ) = fit(x i )+ω(1-K λ )(f m (x i )-fit max ) (10)

[0106] In the formula, ω represents the weight of evolutionary potential, K represents the temperature attenuation coefficient, λ represents the number of new solution searches, and f m (x i ) represents chromosome x i The optimal solution after local search; fit max is the maximum value of population fitness, and its calculation is the same as that of formula (5); the present invention solves the problem of minimizing the objective function: fit(x i ) is the fitness function = negative objective function; by updating the fitness, the solution space around the individual can be explored, and a small number of chromosomes with a large population can be selected to enter the next generation, individuals with evolutionary potential can be mined, and new solutions can be discovered. It not only enriches the diversity of the population to maintain the search range of the population, but also enables other excellent individuals to cross and hopefully produce better new individuals for leapfrog search. By strengthening the local search through the neighborhood structure, the potential for solutions that are better than the global optimal solution can be discovered. The neighborhood structure Y is calculated as follows:

[0107]

[0108] Where P i (max), P i (min) represent the decision variables P i The upper and lower bounds of the domain, i = 1, 2, ..., μ, μ represents the dimension of the decision variable (i.e., the dimension of the problem), L i represents the neighborhood structure problem of 1 to μ-dimensional decision variables, l represents the gene length, δ is the range coefficient of the search space, and θ is the coefficient that controls the size of the neighborhood structure. When μ = 1, the function f(P i ) of chromosomes with a gene length of l, the binary code string represented by the decision variable P i The domain of [P i (min), P j (max)] is discretized to 2 l -1 uniform region. Each of these regions has a neighborhood structure, i.e., a length of L(P i ) segment; when μ=2, the task function f(P i , P i+1 ) represents the decision variable P i With P i+1 The domain of is discretized into 2 l -1 uniform region, there are (2 l -1) 2 neighborhood structures, each of which has an area of ​​L(P i )×L(P i+1 ) plane matrix. Similarly, the k-dimensional problem includes (2 l -1) kNeighborhood structures, each of which represents a volume of A k-dimensional cube is constructed, thereby establishing a one-to-one correspondence between the target solution space of the problem being solved and the search space of the algorithm.

[0109] When an individual conducts a local search, the phenotype changes while the genotype remains unchanged; if there are multiple neighborhood structures that make the current individual better than the global optimal solution, the neighborhood structure with the largest acceptance rate is selected to generate a new solution to enter the next hop. When the neighborhood structure of each hop is determined to generate a new solution to replace the current solution. When an individual finds a solution that is better than the global optimal solution through local search, it is determined that the individual has greater evolutionary potential, and the probability of being able to search for the global optimal solution through evolution is also greater. The fitness of individuals with evolutionary potential is adjusted to give them a greater chance of entering the next generation.

[0110] ③Adaptive decision transformation strategy

[0111] If the population is stagnant or stuck in a local optimal solution during evolution, the evolution strategy should be changed according to the population evolution situation. The present invention designs an evolutionary generation R in which the optimal solution remains unchanged. f To measure the evolutionary state of the population, the probability of the population changing strategy is calculated as follows:

[0112]

[0113] In the formula, R t is the current evolutionary generation (also the generation of population stagnation), R is the total evolutionary generation, β is the control parameter, the median value in this case is 6, R max is the maximum stagnation algebra.

[0114] Figure 2 The solid curve in the middle indicates that when the evolutionary algebra remains unchanged, P ed The law of algebraic transformation with stagnation; P ed In the early stage of evolutionary stagnation, the population grows slowly, leaving a certain amount of time for the population to rely on the existing evolutionary strategy to get rid of stagnation. As the number of stagnation generations increases, the algorithm determines that the population is in a stagnant state, which significantly improves the probability of P compared to linear control. ed , to avoid the population wasting time on the existing evolutionary strategy. Finally, in order to maintain the balance between population exploration and algorithm convergence speed, when the stagnation number remains unchanged, P ed With the further increase of evolutionary generations, it shows a linear downward trend.

[0115] (6) Integrated adaptive simulated genetic algorithm

[0116] ① Local optimization search strategy process

[0117] 1) Initialize strategy parameters: search space coefficient δ, number of decision variables μ, Markov chain length m = (2δ) μ, initial temperature T, temperature attenuation coefficient K, temperature attenuation times k, new solution search times λ, C={n1,n2,…,n m} represents the set of neighborhood structures, f(x i ) is chromosome x i The initial fitness of the local search for the optimal value f′(x i )=fit max , fit max is the global optimal fitness.

[0118] 2) For chromosome x i Adopting neighborhood structure z j Perform a local search and calculate z according to the improved Metropolis criterion of formula (13) j Probability of being selected:

[0119]

[0120] In the formula, Represents x i Use z j The new fitness after local search. j=j+1.

[0121] 3) If j<m, go to 2), otherwise go to 4).

[0122] 4) If there exists P(z j )>1, go to 5), otherwise go to 6).

[0123] 5) Take P(z j ) value of the largest neighborhood structure produces a new solution, Reset j to 7).

[0124] 6) Normalize the selection probability P(z) of the neighborhood structure within N j ), select a neighborhood structure to generate a new solution by proportional selection method, Reset j to 7).

[0125] 7) If the termination condition is met, go to 8), otherwise go to 2).

[0126] 8) Local search ends, output f′(x i ).

[0127] ②Improving Monte Carlo simulation algorithm and process

[0128] In order to solve the complex parameter optimization problems in practical applications and enhance the algorithm's ability to identify structural parameters, we transform the complex problem of seeking solution space equilibrium and parameter optimization into a constrained nonlinear programming problem:

[0129]

[0130] Monte Carlo simulation is used to estimate the region {(x1,x2…x n )|x j ∈[a j ,b j ],j=1,2,…,n}, randomly select several experimental point sets, then find the feasible point solution set from the experimental point set, and then select the minimum solution set that meets the conditions from the feasible point solution set.

[0131] The jth component x of the test point set j obey j ,b j ] is uniformly distributed within. That is:

[0132] x i =a j +S(b j -a j ),j=1,2,…,n (15)

[0133] S is a random number uniformly distributed between [0,1], let:

[0134] P: total number of trials;

[0135] MaxP: maximum total number of test points;

[0136] K: total number of feasible point sets;

[0137] MaxK: maximum number of feasible points;

[0138] X * : The optimal point set generated by iteration;

[0139] Q: The optimal feasible solution minimum value f(X * ), whose initial value is the maximum number of experiments that the computer can represent.

[0140] First, a random number is generated as the initial test point set, and then the jth component of the previous test point is randomly generated, that is, x i =a j +S(b j -a j ), and the other components remain unchanged to generate a new test point. In this way, only a new random component is needed to generate a new test point. The iteration stops when K>MaxK or P>MaxP; the iteration steps are shown in Figure 3 Block diagram of the improved Monte Carlo simulation iteration steps.

[0141] ③Integrated adaptive simulation genetic algorithm process

[0142] The process of adaptive simulation genetic hybrid optimization algorithm is as follows Figure 4shown.

[0143] Step 1. Initialize the parameters. Population size M, chromosome number N, chromosome length l, neighborhood structure range coefficient δ, neighborhood structure size coefficient θ, decision variable dimension u, initial temperature T, temperature decay coefficient K, maximum stagnation generation G max Initializing the population by diversity is the basic condition for the GA algorithm to search for the global optimal solution. Effectively controlling the diversity of the population is an important way to improve the performance of the algorithm. The population should have rich diversity when it is established. ASGA first generates the initial population. Each individual in the population needs to calculate the individual similarity h(x i ), and randomly generate new chromosomes to replace similar individuals until there are no similar individuals in the population, and randomly assign evolutionary strategies.

[0144] Step 2. Calculate the fitness of each individual in each population and perform fitness transformation according to formula (9).

[0145] Step 3. Calculate the probability of each population switching strategy according to formula (12).

[0146] Step 4. Each population performs genetic operations independently according to the evolutionary strategy and shares the excellent individuals within the population. The public population first performs a crossover operation on the individuals, and then performs a local search according to the method described in (5)②. The individuals with evolutionary potential correct their fitness in a timely manner.

[0147] Step 5. Determine the termination criteria. If the conditions are met, go to step 7, otherwise go to step 6.

[0148] Step 6. The population completes the constraints of the competition strategy through equations (6)(7)(8), and then goes to step 2.

[0149] Step 7. Achieve the population adjustment evolution strategy goal and switch to the Monte Carlo simulation module.

[0150] Step 8. Analog Output X * ,Q, transfer genetic operation module.

[0151] Step 9. After selection, crossover and variation operations, local search-global optimization.

[0152] Step 10. Determine whether the output result meets the overall optimal solution and automatically control it, otherwise go to step 3.

[0153] 4. Simulation experiment and test of ASGA algorithm performance

[0154] In order to verify the effectiveness of the ASGA algorithm, we use 6 relatively complex multi-peak functions for testing. Table 2 lists the 6 function structures.

[0155] The first measure function f1(x) is a popular and complex multi-modal optimization Rosenbrock function. As shown in Figure 5(a), its global optimal point is in a continuous and smooth but narrow valley. Because it can only provide a small amount of information, it is difficult for traditional single optimization search algorithms to find the right direction. Even if the global minimum point can be found, the convergence speed is very slow.

[0156] The global optimal solution of the second measure function f2(x) is the Schaffer function (Schaffer) located at the center of the search area, and the optimal solution value is very close to the value of the suboptimal solution in the adjacent area. There is a very large search area between the local minimum and the global minimum, and the local minimum is distributed around the global minimum, which is very deceptive.

[0157] The third measure function f3(x) has many local minima, called the Rastrigin function. A single swarm intelligence algorithm can easily fall into the local optimal solution and cannot search for the global optimal solution of the function.

[0158] The fourth measure function f4(x) is of order 2 and can be extended to higher orders. It is called the Griewank function. Its global minimum is 0 at (100, 100), but it also has many local minima, which increase exponentially with the increase of the function order.

[0159] The global minimum of the fifth measure function f5(x) is located at the origin and is called the needle function. It looks like a needle from Figure 5 (f5), but it has four local minima. When a single group intelligent algorithm selects different values ​​for parameters a and b, it can form deceptive test results to varying degrees.

[0160] Table 2 Six test function tables

[0161]

[0162] The sixth measure function f6(x) is a multimodal scalable parameter optimization, called the Rana function. It can utilize the weighted bag method of the extended function to expand from the two-dimensional original function to a higher dimension. It is a very typical multi-modal function with strong deceptiveness.

[0163] The results of the six measurement functions fully reflect the superiority of the FASGA algorithm. For each test function, the three algorithms were run 50 times, with the parameter settings of population size N = 50, population number M = 6, chromosome length u × 20, u is the problem dimension, and the algorithm execution generation number is 400 generations. The three algorithms all use proportional selection operations, two-point crossover operations, and multi-point mutation operations. The evolutionary strategy follows the aforementioned ASGA hybrid optimization algorithm process, and the specific parameters are P c1 =0.7, Pm1 =0.1, P c2 =0.5, P m2 =0.3, P c3 =0.85, P m3 =0.05. In ASGA, δ = 3, θ = 105, T = 100, K = 0.9, R max =15, η = N / 5. All simulation experiments were carried out on the windows11 computer operating system and Matlab7.11 simulation platform.

[0164] The evolution curves of the six functions are shown in Figure 5(a), Figure 5(b), Figure 5(c), Figure 5(d), Figure 5(e), and Figure 5(f). The maximum values ​​(F max ), minimum value (F min ), mean value (F ave ) as shown in Table 3.

[0165] Table 3 Comparison of maximum / minimum / average values ​​of three algorithms after 50 iterations of 6 test functions

[0166]

[0167] The analysis shows that the convergence speed and the optimal solution obtained are the most important indicators for measuring the performance of the algorithm. Figure 6(a)-Figure 6(f) As can be seen from Table 3, for the test functions f1(x) and f4(x), the three algorithms can find the optimal solution, and the extreme value of the ASGA algorithm is significantly better than that of the GA and PSO. At the same time, the convergence speeds of f2(x), f5(x) and f6(x) are different, and the difference between the optimal values ​​of the ASGA algorithm is larger than that of the other two algorithms. As shown in Table 3, the average optimal value obtained by the ASGA algorithm is better than that of the GA and PSO; the number of iterations of the ASGA algorithm is better than that of the GA and PSO algorithms. The simulation results verify the effectiveness of the ASGA algorithm in optimizing the test function; thus, the ASGA algorithm has better performance than the GA and PSO algorithms.

[0168] 5. Application of ASGA algorithm in the intelligent control of the minimum overall optimization of each sub-goal in lean product production

[0169] Since lean product production is a very complex intelligent control system, it is necessary to solve multi-objective computer system optimization problems. Practice has proved that a single swarm intelligence optimization algorithm can no longer meet the needs. Therefore, the present invention uses an ASGA algorithm to support a computer system of a specific calculation model for multi-objective nonlinear precise feeding, temperature, and optimal control of CO2, NH3, and H2S concentrations, and designs an identification method for the FASGA algorithm to avoid the nonlinear, time-varying, and saturated complex states of multiple objectives. The overall optimization and generalization advantages of the trained FASGA model are used in a computer system for integrated optimization control of multi-objective precise feeding, temperature, CO2, NH3, and H2S concentrations in key scientific research projects to shorten the production cycle and improve the quality and efficiency of the project.

[0170] (1) ASGA algorithm coding

[0171] ASGA algorithm coding is a conversion method that converts feasible solutions to the problem to be solved from the solution space to the search space that can be handled by the adaptive simulation genetic algorithm. During the operation of the ASGA algorithm, it does not directly operate on the actual decision variables of the problem to be solved, but applies various genetic operations to the individual codes representing feasible solutions to achieve the purpose of optimization through such genetic operations. The intrinsic quality improvement and quantity growth of lean production products are highly dependent on the centralized control environment. In the computer system of optimization control, the parameters to optimize the lean production centralized control environment are temperature, and the concentration setting values ​​of CO2, NH3, and H2S. The amount of data is not large, but it must have a certain degree of accuracy, so the binary coding method is adopted. The mapping relationship between each parameter and the binary code is:

[0172] H i =H i min +binar / 2 L (H i max -H i min ) (16)

[0173] K i =K i min +binar / 2 L (K i max -K i min ) (17)

[0174] N i =N i min +binar / 2 L (N i max -Ni min ) (18)

[0175] S i =S i min +binar / 2 L (N i amx -N i min ) (19)

[0176] Among them, H i is the temperature to be set, [H imin ,H imax ] is the temperature range, binar is a 2 L The binary integer represented by the longest binary string. Similarly, K i is the carbon dioxide concentration to be set, [K imin ,K imax ] is the minimum and maximum range of carbon dioxide concentration, N i is the ammonia concentration to be set, [N imin ,N imax ] is the minimum and maximum variation range of ammonia concentration; S i is the concentration of hydrogen sulfide to be set, [S imin ,S imax ] is the minimum and maximum variation range of hydrogen sulfide concentration; binar and 2 L Same as above.

[0177] (2) Computer system for intelligent control of the minimum overall optimization of each sub-goal based on ASGA I

[0178] In the computer system I of the lean production optimization intelligent control method based on ASGA, it is composed of three hybrid optimization algorithms: adaptive, Monte Carlo simulation, and genetic. The excellent performance of the adaptive simulation genetic algorithm is fully verified through six multi-peak test function simulation experiments. The multi-objective intelligent control optimization scheme for improving the lean product production environment and the input-output quality ratio is obtained through the ASGA algorithm, so that the lean production intelligent control system obtains better economic benefits. The present invention first selects the inverse of the weight gain of lean product production and the inverse of the feed ratio as the fitness function, which represents the maximization of the benefit of material input and output at this moment, that is:

[0179]

[0180] Where: D(h) is the target weight gain, B(h) is the weight of the consumed materials, the first fitness function fit1(x) is equal to the inverse of the objective function f1(x), fit1(x) obtains the maximum value, and the objective function f1(x) obtains the minimum value. The denominator is the weight of the product quality increase in lean product production, which is the maximum value of the material input weight, indicating that more product quality can be obtained per unit input; σ and ρ are the influence coefficients of the target weight gain and the consumed materials under the temperature and suitable air, h = 1,2…V, V is the sampling time.

[0181] (3) Computer system for intelligent control of the minimum overall optimization of each sub-goal based on ASGA II

[0182] The present invention states that in order to achieve the best input-output quality benefit for lean production, the multi-objective environmental factors of the production must be intelligently controlled to fully meet the computer system requirements for the overall minimum optimization intelligent control of each sub-objective. The ASGA algorithm is mainly used to optimize the quantization factor and the proportional factor, the membership function width, the Monte Carlo simulation rule, and the adaptive genetic algorithm to find the optimal values ​​of the simulation controller parameters respectively, to ensure the environmental temperature of the production target, and the accuracy of the optimal control of CO2, NH3, and H2S concentrations. The present invention uses the inverse of the absolute value of the error between the setting value of the computer system for optimal control and the system output value as the fitness function, that is:

[0183]

[0184]

[0185]

[0186]

[0187] Where: H set is the temperature setting value, H i is the system output temperature value, i=1,2,…,W, W is the sampling number, the second adaptation function of fit2(x) is equal to the inverse of the objective function f2(x), and the fit2(x) adaptation function obtains the maximum value, and the objective function f2(x) obtains the minimum value, indicating that the difference between the ambient temperature setting value and the output temperature value obtains the minimum value through system optimization and intelligent control. It comprehensively considers multiple performance indicators such as the system temperature rise time, overshoot and steady-state error. Similarly, K set 、N set , S set are the setting values ​​of the carbon dioxide, ammonia and hydrogen sulfide concentration calculation models, K i 、N i , S iare the computer system output values ​​for optimizing the concentrations of carbon dioxide, ammonia and hydrogen sulfide, j = 1, 2, …, Q, Q is the sampling times of carbon dioxide; t = 1, 2, …, M, M is the sampling times of ammonia; p = 1, 2, …, R, R is the sampling times of hydrogen sulfide, fit3(x), fit4(x) and fit5(x) are the fitness functions for optimizing the concentrations of carbon dioxide, ammonia and hydrogen sulfide, respectively, and the objective functions f3(x), f4(x) and f5(x) obtain their minimum values; these three environmental control objective functions and the temperature control objective function, supported by the FASGA algorithm, constitute a computer system for the integrated optimal control environment of lean product production.

[0188] (4) Results analysis of the implementation of the lean production multi-objective optimization intelligent control method based on ASGA

[0189] In order to verify the effect of fuzzy adaptive simulation genetic algorithm and its application in multi-objective optimization intelligent control system, we chose to conduct experiments on the intelligent optimization control link of the production physical space of lean products, which is more difficult to grasp, in the ongoing projects. First, an independent production unit was selected to test the production feed consumption and weight gain ratio: by establishing refined management of variable control of the whole life cycle such as endogenous characteristics, variety type, daily feed structure, climate environment, temperature and humidity, the quality of lean production in different production stages is guaranteed to be matched automatically with the inherent needs; product weight within 16 catties: consumption of 7 catties of fine feed, net weight gain of 14 catties, feed-to-weight ratio of 0.5:1; weight within 17-30 catties: consumption of 24 catties of fine feed, net weight gain of about 15 catties, feed-to-weight ratio of 1.6:1; weight within 31-70 catties: consumption of about 74 catties of low fine feed, net weight gain of about 40 catties, feed-to-weight ratio of 1.85:1; weight within 71-14 Within 0 catties: consumption of medium-quality materials is about 196 catties, net weight gain is about 70 catties, and feed-to-weight ratio is 2.8:1; self-weight is within 141-200 catties: consumption of medium-quality materials is about 180 catties, net weight gain is about 60 catties, and feed-to-weight ratio is 3:1; self-weight is 200-240 catties: consumption of medium-quality materials is about 128 catties, net weight gain is about 40 catties, and feed-to-weight ratio is 3.2:1; comparison between the centralized control method and the company's original production method: due to the refined process control in the production stage, the quality value of lean product production has increased by more than 25%. The weighted average results show that the output weight gain and feed input ratio is 1:2.55, which is more than 15% higher than the industry level of 1:3.

[0190] The effect of the experimental temperature on the production quality performance of the product was studied. 160 target objects of lean production were selected and the experiment was carried out indoors. There were multiple independent small rooms in the room, which could automatically adjust the indoor temperature according to the needs. The adjustable range was 0-40℃. The experimental design adopted a 4×5 factorial experimental design. The weight of the lean products was divided into 4 production stages of 20-40Kg, 40-60Kg, 60-80Kg, and 80-100Kg. The temperature was set at 5 levels of 23, 26, 29, 32 and 35℃, with a total of 40 treatments, and 4 matchings in each treatment for a total of 160 lean products. The experimental results showed that the curves of daily weight gain and daily material consumption at different temperatures in different production stages are as follows Figure 7 and Figure 8 .

[0191] According to the different production stages of lean products, the optimal temperature setting value at different growth stages is obtained through the computer system I based on the ASGA-based minimum overall optimization intelligent control of each sub-goal. Fig. 9 As shown, the four growth stages are 28, 22, 28, and 22 degrees respectively.

[0192] Through the optimal controller in the computer system II based on the ASGA's minimum overall optimization intelligent control of each sub-goal and the experiment in the test field of this case, we obtained the temperature response curve when the temperature is 25 degrees at a certain initial moment in winter and the set value is 28 degrees. Fig. 9 By comparing the results of the control computer output without optimizing the fuzzy controller, it can be seen that the temperature control error is smaller, which can provide a comfortable integrated environment for the company's lean product production.

[0193] Furthermore, to build a lean product production multi-objective intelligent control computer system, it is necessary to collect air temperature and humidity online through intelligent sensors, and the general content of CO2 is about 0.03%-0.04% (when ventilation is poor, it can reach 0.4%, that is, 7920mg / M 3 ), NH3 in the production workshop is not higher than 20mg / M 3 , H2S concentration is not higher than 15mg / M 3In the integrated scenarios, the Internet of Things is used to upload information: the computer control system will intelligently judge whether the control environment is suitable based on the area of ​​the lean product production space, the density of product targets and other big data, and take corresponding measures according to the specific situation. For example, if the concentration of CO2 or NH3 in the air of the production workshop is too high, the computer system will issue a command to drive the intelligent control equipment to turn on the fan to increase the air exchange volume; if the feedback temperature is too high and the humidity is too low, the amount of water in the indoor water curtain will be increased to play a role in cooling and humidifying. At the same time, the integrated control computer system will also collect data through sensors, analyze the response of lean products to environmental comfort, establish a parameter model and threshold for the computer system to judge the comprehensive environmental comfort, and analyze and establish the relationship between environmental parameters and material conversion rate, production performance, etc. to improve the quality and efficiency of lean products.

[0194] Because the waste in lean product production will decompose ammonia, its content is related to the ventilation of the physical space, waste treatment, and material density. The ammonia content in the physical space of lean product production is generally 5-27 mg / m 3 , required to be no higher than 20mg / M 3 The hydrogen sulfide in the workshop comes from the chemical decomposition of sulfur-containing organic matter (waste from lean product production and organic materials in production, etc.); the hydrogen sulfide content in the safe production workshop is generally 15mg / M 3 The following, too much material or untimely waste treatment can increase its content. The optimal control change results of CO2, NH3, and H2S concentrations in a 24-hour cycle are obtained through the optimized computer control system II. Fig.11 , Fig.12 and Fig.13 shown.

[0195] Analysis of computer system for optimized intelligent control of lean product production. Aiming at improving the quality and efficiency of lean product input-output, the adaptive simulated genetic algorithm (ASGA) is used to support the multi-objective optimization intelligent control system in the face of parameter optimization problems in the computer system data information processing of the centralized control environment of lean product production. Experiments on the intelligent optimization control computer system with the minimum target weight gain and feed ratio of lean product production, as well as the minimum target of ambient temperature, CO2, NH3, and H2S gas concentration show that the ASGA algorithm has certain reliability in the multi-objective optimization intelligent control computer system with the minimum target product weight and feed ratio inverse value, ambient temperature, and harmful gas concentration in lean product production. The computer system for optimized control of the specific calculation model established by the present invention not only lays the foundation for the computer system of multi-objective optimal automatic control, completes data collection, processing and analysis, and realizes refinement and scientificization, but also can effectively block the spread of risks that are unfavorable to production, reduce material input, increase the output of lean products, reduce labor and material consumption costs, and improve the quality and efficiency of project implementation.

[0196] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0197] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A lean production multi-objective optimization intelligent control method based on an adaptive simulation genetic algorithm, characterized by: The technique includes the following steps: Step 1: Construct f that satisfies x∈X: Under the constraints, V-min vector is minimized, so that the vector objective function f(x) = [f1(x), f2(x), …, f n (x)] T Each sub-objective function in is minimized as much as possible; Lean production multi-objective optimization problem: Where x∈X represents n decision vectors in X decision units, f: It indicates that the objective function satisfies the mapping relationship from X to f(x) of multi-objective optimization under the constraint of n×m cube R matrix space; construct an algorithm to analyze whether the project has the conditions for achieving the minimum value of multi-objective optimization, divide the sub-objective function into some sub-groups, and assign a sub-objective function to each sub-group. Each sub-group independently selects some individuals with high fitness to form a new sub-group, and merges all the newly generated sub-groups into a complete group. Then, a series of crossover and mutation operations are performed in this group to generate the next generation of complete groups. In this way, the "split-parallel selection-merge" operation is continuously performed, and finally the Pareto optimal solution of the multi-objective optimization problem can be obtained.

2. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 1, characterized in that: Step 2: V-minf(x) can help multi-objective optimization to obtain the minimum value. When there is more than one minimum value, create the objective function set for the application as follows: The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm is integrated to support project intelligent operation and control, and output the final multi-objective lean product production intelligent control minimum value; D(h) is the target weight gain, B(h) is the weight of the consumed material, f1(x) is the first objective function to obtain the minimum value, then the denominator is the weight of the product quality increase in lean product production compared to the weight of the material input to obtain the maximum value, indicating that more product quality can be obtained per unit input, σ, ρ are the influence coefficients of the target weight gain and the consumed material under the temperature and suitable air, h = 1, 2...V, V is the sampling time; H set is the temperature setting value, H i is the system output temperature value, i = 1, 2, ..., W, W is the number of sampling times, f2(x) is the optimization of the computer system intelligent control target to make the objective function minimum, the fitness function comprehensively considers the system's rise time, overshoot and steady-state error and other performance indicators; similarly, K set 、N set , S set are the setting values ​​of the carbon dioxide, ammonia and hydrogen sulfide concentration calculation models, K i 、N i , S i are the computer system output values ​​for optimizing the concentrations of carbon dioxide / ammonia / hydrogen sulfide, respectively; j=1, 2, ..., Q, where Q is the sampling number of carbon dioxide; t=1, 2, ..., M, where M is the sampling number of ammonia; p=1, 2, ..., R, where R is the sampling number of hydrogen sulfide concentration; f3(x), f4(x), and f5(x) are the objective functions for optimizing the concentrations of carbon dioxide, ammonia, and hydrogen sulfide to obtain their minimum values.

3. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 2 is characterized by: The third step is to obtain the optimal solution of multi-objective optimization and gradually integrate the fuzzy adaptive simulation genetic algorithm as a prerequisite; in view of the shortcomings of the existing adaptive genetic algorithm, improve the calculation method of crossover rate and mutation rate to enhance the algorithm's fitness and overcome the crossover rate (P d ) and mutation rate (P e ) The calculation is simpler, excellent individuals tend not to change their own state, and the algorithm cannot jump out of the local optimization defect; the crossover rate P d and mutation rate P e The calculation is as follows: Where fit max 、fit min 、fit ave are the maximum, minimum and average values ​​of fitness respectively, c is the conservative estimate of the limit of the objective function, c>0, f(x) is the objective function, k1 and k2 are adaptive control parameters, which take fixed values ​​in the interval (0, 1), 0.5<a<1, 0<b<1, and the analysis is detailed in the third step of the description; in the early stage of the improved adaptive genetic algorithm (TAGA), the individuals are more dispersed and the mutation probability is lower, which can ensure that the population converges to the global optimal solution; when the individuals in the population tend to be concentrated, the larger mutation rate (P e ), which increases the diversity of the population and ensures the evolution of the population while enhancing the local search capability of the algorithm.

4. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 3 is characterized by: Step 4: Improve the fitness scale transformation, solve the problem that the proportional selection strategy is prone to selecting "super individuals" and causing premature convergence and affecting the convergence speed of the algorithm, and calculate the fitness transformation f′(x i )as follows: f′(x i ) is the fitness transformation, also called the transformed fitness, fit max 、fit min 、fit ave are the maximum, minimum and average values ​​of the population fitness respectively. Their calculation is the same as that in claim 3. The present invention solves the problem of minimizing the objective function: fit(x i ) is the objective function of fitness function = negative, g represents the current evolutionary generation, G represents the total evolutionary generation, a is 6, T is the temperature of the simulated annealing method, is the attenuation coefficient; when Z approaches 1, it changes, either reducing the super individual selection rate and increasing the selection rate of other individuals to maintain the population diversity in the early stage of evolution; or increasing f′(x i ) selection sensitivity, increasing the probability of excellent individuals being selected and promoting the accelerated convergence of the algorithm.

5. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 4, characterized in that: Step 5: After the public population performs crossover operations on excellent individuals, all individuals search according to the local search strategy and update the fitness according to the following formula: fit(x i )=fit(x i )+ω(1-K λ )(f m (x i )-fit max ) In the formula, ω represents the weight of evolutionary potential, K represents the temperature attenuation coefficient, λ represents the number of new solution searches, and f m (x i ) represents chromosome x i The optimal solution after local search; fit max is the maximum value of population fitness, and its calculation is the same as that of right 3; the present invention solves the problem of minimizing the objective function: fit(x i ) is the fitness function = negative objective function; by updating the fitness, the solution space around the individual can be explored, a small number of chromosomes with a large population can be selected to enter the next generation, individuals with evolutionary potential can be discovered, and new solutions can be discovered; it not only enriches the diversity of the population to maintain the search range of the population, but also enables other excellent individuals to cross and hopefully produce better new individuals.

6. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 5, characterized in that: Step 6: The population enhances the local optimization ability of the algorithm by searching the neighborhood structure. The neighborhood structure is calculated as follows: Where P i (max), P i (min) represent the decision variables P i The upper and lower bounds of the domain, i = 1, 2, ..., μ, μ represents the dimension of the decision variable (i.e., the dimension of the problem), L i It represents the neighborhood structure problem of 1 to μ dimensional decision variables, l represents the gene length, δ is the range coefficient of the search space, θ is the coefficient that controls the size of the neighborhood structure, and the detailed analysis of the parameters is detailed in the description; when the individual conducts a local search, the phenotype changes while its own genotype remains unchanged; if there are multiple neighborhood structures that make the current individual better than the global optimal solution, the neighborhood structure with the largest acceptance rate is selected to generate a new solution and enter the next hop; when the neighborhood structure of each hop is determined to generate a new solution to replace the current solution, when the individual finds a solution that is better than the global optimal solution through local search, it is determined that the individual has a greater evolutionary potential, and the greater the probability of finding the global optimal solution through evolutionary search, the fitness of the individual with evolutionary potential is adjusted to make it have a greater chance of entering the next generation.

7. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 6, characterized in that: Step 7: If the population is stagnant or stuck in a local optimal solution during evolution, the evolution strategy should be changed according to the population evolution situation. The present invention designs an evolutionary generation R that maintains the optimal solution unchanged. f To measure the evolutionary state of the population, and calculate the probability of population switching strategy as follows: In the formula, R t is the current evolutionary generation (also the generation of population stagnation), R is the total evolutionary generation, β is the control parameter, which is 6 in this case, R max is the maximum stagnation generation; the probability formula of population transformation strategy fully explains the evolutionary generation when P ed Following the law of stagnant algebraic transformation, it is revealed that nonlinear probability can more reasonably control the population transformation evolution strategy.

8. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 7, characterized in that: Step 8: To make the adaptive simulation algorithm play a better role in the multi-objective control of lean product production, the Monte Carlo simulation algorithm is improved and integrated into the whole algorithm. In order to solve the complex parameter optimization problems in practical applications, the algorithm's ability to identify data structure parameters is enhanced, and the search for solution space equilibrium and parameter optimization of complex problems is transformed into a constrained nonlinear programming problem: Min f(X)X∈E n s.t.g i (X)≥0i=1,2,…,m has j ≤x i ≤b j j=1,2,…,n Monte Carlo simulation is used to estimate the region {(x1, x2…x n )|x j ∈[a j , b j ], j = 1, 2, ..., n} randomly select a number of experimental point sets, then find a feasible point solution set from the experimental point set, and then select the minimum solution set that meets the conditions from the feasible point solution set; The jth component x of the test point set j obey j , b j ] uniform distribution within; that is: x i =a j +S(b j -a j ),j=1,2,…,n S is a random number uniformly distributed between [0, 1]; set up: P: total number of trials; MaxP: maximum number of trial points; K: total number of feasible points; MaxK: maximum number of feasible points; X * : The optimal point set generated by iteration; Q: The minimum value of the optimal feasible solution generated by iteration f(X * ), whose initial value is the maximum number of experiments that the computer can represent; first generate a random number as the initial test point set, and then randomly generate the jth component of the previous test point, that is, x i =a j +S(b j -a j ), the other components remain unchanged and a new test point is generated; thus, only one new random component is needed to generate each new test point; the iteration ends when K>MaxK or P>MaxP.

9. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 8, characterized in that: Step 9: To make the adaptive simulation algorithm fully play its role in the multi-objective control of lean product production, the local optimization search strategy process, the improved Monte Carlo simulation process, and the improved genetic algorithm process are organically integrated to integrate the technical core of the present invention; the adaptive simulation genetic algorithm for multi-objective optimization can effectively solve the multi-objective optimization intelligent control tasks in lean product production.

10. The lean production multi-objective optimization intelligent control method based on adaptive simulation genetic algorithm according to claim 9, characterized in that: In the tenth step, during the operation of the ASGA algorithm, the actual decision variables of the problem to be solved are not directly operated, but various genetic operations are applied to the individual codes representing feasible solutions to achieve the purpose of optimization; the intrinsic quality improvement and quantity growth of lean production products are highly dependent on the centralized control environment. In the computer system of optimization control, the parameters of the lean production centralized control environment to be optimized are temperature, and the concentration setting values ​​of CO2, NH3, and H2S, which are required to achieve a certain accuracy. Therefore, the binary coding method is adopted. The mapping relationship between each parameter and the binary code is: H i =H i min +binar / 2 L (H i max -H i min ) K i =K i min +binar / 2 L (K i max -K i min ) N i N i min +binary / 2 L (N i max -N i min ) S i =S i min +binary / 2 L (N i amx -N i min ) Among them, H i is the temperature to be set, [H i min , H i max ] is the temperature range, binar is a 2 L The binary integer represented by the longest binary string. Similarly, K i is the carbon dioxide concentration to be set, [K i min , K i max ] is the minimum and maximum range of carbon dioxide concentration, N i is the ammonia concentration to be set, [N i min , N i max ] is the minimum and maximum variation range of ammonia concentration; S i is the concentration of hydrogen sulfide to be set, [S i min , S i max ] is the minimum and maximum variation range of hydrogen sulfide concentration; binar and 2 L Same as above.

Images