Assembly sequence planning method of improved flower pollination algorithm based on multiple strategies

Abstract

The invention discloses an assembly sequence planning method based on a multi-strategy improved flower pollination algorithm, and relates to the research field of assembly sequence planning. According to the method, on the basis of assembly body information, the assembly direction transformation cost, the assembly tool transformation cost and the assembly operation difficulty cost are set as assembly cost evaluation indexes of the assembly sequence, and a sequence evaluation system is constructed through quantitative evaluation of the constructed assembly sequence cost. The improved flower pollination algorithm based on multiple strategies is provided, the convergence efficiency of assembly sequence planning optimization can be improved, the method has more advantages than a single algorithm, the problem that sequence optimization falls into a local optimal solution can be solved, and a new thought and an efficient method are provided for assembly planning research. The method aims at solving the problems that in industrial product assembly sequence planning, a flower pollination algorithm is weak in solution capacity in a discrete spatial domain, prone to falling into local optimum, low in solution efficiency and the like, the optimal assembly sequence is rapidly solved, then the production time and cost of industrial products are reduced, and the assembly quality is improved.

Description

Technical Field

[0001] The present invention relates to the research field of assembly sequence planning, and particularly to a multi-strategy improved flower pollination algorithm sequence planning method for the assembly sequence planning of gas turbine products. Background Technique

[0002] Assembly sequence planning, abbreviated as assembly planning, is a necessary step in virtual assembly and also the basis and important content of assembly technology. Assembly sequence planning refers to screening out feasible sequences under the premise of meeting the product assembly conditions and quantitatively evaluating them according to the sequence evaluation system to find the optimal assembly sequence. Currently, the traditional assembly of complex products such as gas turbines mainly relies on manual experience to carry out assembly design. The generated assembly sequences are largely affected by the level of workers, making it difficult to ensure the rationality and effectiveness of the assembly process and easily leading to cost waste. Therefore, how to use computer technology to improve the assembly efficiency of products and improve the assembly quality of products has always been a problem that the industry is committed to solving. To explore the optimal assembly sequence that simultaneously meets the geometric feasibility between parts and the assembly operation constraints during the assembly process, it is necessary to construct an intelligent assembly sequence planning model and deeply study the generation problem of assembly sequences.

[0003] There are some problems when using a single intelligent planning algorithm to solve the optimization of the assembly optimal sequence. First, due to too many variables, complex processing procedures, and large amounts of calculation, it takes a long operation time to obtain the final optimization result. Second, the intelligent optimization algorithm is prone to falling into local optimal solutions, with weak overall search capabilities and difficulty in finding global optimal solutions. In addition, the optimization result is also affected by the initial value. Therefore, when conducting sequence planning optimization design, the above problems need to be comprehensively considered, and an effective intelligent optimization algorithm should be selected in the research of assembly sequence planning to generate the optimal assembly sequence. The flower pollination algorithm is a relatively new meta-heuristic algorithm, and its design inspiration comes from cross-pollination and self-pollination of pollen, corresponding to local search and global search respectively. It is determined whether to perform cross-pollination or self-pollination according to the switching probability P, and has the advantages of fast convergence speed, few parameters to be adjusted, and strong global search capabilities. According to the NFL theorem, no single algorithm can handle all optimization problems, that is, each algorithm has its own advantages and limitations. The flower pollination algorithm lacks a corresponding mutation mechanism. As the evolution progresses, the diversity of the population decreases, which may lead to lower convergence accuracy when the algorithm optimizes some functions, and may converge prematurely. At the same time, its mutation mechanism can be improved to avoid the population falling into local optimality during the exploration stage.

[0004] The present invention applies the flower pollination algorithm to the assembly sequence planning on the basis of the virtual assembly of a certain type of gas turbine. To address the shortcomings of the flower pollination algorithm, such as premature convergence and easy entrapment in local optima, a multi-strategy improved flower pollination algorithm is proposed. First, the global search strategy is adjusted by utilizing the Lévy flight mechanism of two groups of random individuals to increase population diversity and expand the search scope, making the algorithm more likely to jump out of local optima and enhancing its exploitation ability. Second, an elite mutation strategy is introduced in the local search part and combined with the random individual mutation mechanism to form a new local pollination strategy. The evolution direction of other individuals is guided by elite individuals, and the population diversity is maintained through the random individual mutation strategy, enhancing the continuous optimization ability of the algorithm. To more accurately solve the assembly sequence that conforms to actual operations, a sequence evaluation system based on the assembly direction transformation cost, assembly tool transformation cost, and assembly operation difficulty cost is established. The fitness value of the corresponding pollen is obtained by quantitatively evaluating the sequence and used as the evaluation criterion. Summary of the Invention

[0005] Aiming at the problems of low calculation efficiency and poor quality of the generated sequence existing in the existing single intelligent assembly planning algorithm, the present invention proposes an assembly sequence planning method based on a multi-strategy improved flower pollination algorithm. It aims to solve the problems of weak solving ability in the discrete space domain, easy entrapment in local optima, and low solving efficiency of the flower pollination algorithm in the assembly sequence planning of industrial products, so as to quickly solve the optimal assembly sequence, thereby reducing the production time and cost of industrial products and improving the assembly quality.

[0006] The technical solution of the present invention is an assembly sequence planning method based on a multi-strategy improved flower pollination algorithm, and the method includes the following steps:

[0007] S1: Establish an information model of the assembly body and obtain the three-dimensional model information of the assembly body and parts;

[0008] For an assembly body P = {p1, p2,..., p n}, there are several assembly sequences S = {s1, s2,..., s n}, and the assembly sequence S describes the part assembly order for realizing the assembly process of the assembly body P. Among them, the i-th element s i (1, 2,..., n) in the assembly sequence refers to the part number of the part to be assembled in the assembly body in the i-th assembly order operation in the assembly process;

[0009] The assembly information includes: the interference matrix and contact matrix of the assembly, the volume of the parts, the number of geometric constraints, the number of connection constraints, the assembly direction, the type of assembly tool, and the assembly operation difficulty; during the assembly process, for parts with larger volume and more geometric constraints, they are used as the basic parts of the assembly to be assembled first or given priority in assembly. The assembly operation difficulty cost is set as an evaluation index, and the parts with greater assembly operation difficulty are assembled first; the assembly direction transformation cost and assembly tool transformation cost are set as evaluation indexes to reduce the number of assembly direction transformations and assembly tool transformations during the assembly process.

[0010] S2: Extract the interference matrix and contact matrix of the assembly, analyze the factors affecting the assembly according to the information of the assembly, construct a sequence evaluation index, quantify the evaluation sequence to obtain the corresponding fitness function, and select decimal coding for the components of the assembly. Each part corresponds to the pollen on the same kind of plant;

[0011] S3: Generate an initial population, generate an initial population through the initial population generation rule and screen out feasible pollen, that is, the assembly sequence;

[0012] S4: Calculate the fitness values of each feasible pollen, record the feasible pollen in the feasible sequence set FP, generate an offspring population from the feasible sequence set FP using the crossover operation, and generate new pollen through the mutation operation;

[0013] S5: Determine whether the termination condition is met. If not, use the latest sequence as the initial population for flower pollination and repeat S4 for iteration; if so, obtain the final global optimal result, that is, the required assembly sequence.

[0014] Furthermore, in step S2, the interference matrix is used to describe the interference situation of part p in the assembly P i when moving along the {±x, ±y, ±z} directions in the Cartesian coordinate system with part p j It reflects the spatial constraint relationship between the parts of the assembly and is used to judge the geometric feasibility of the assembly sequence;

[0015] For an assembly P = {p1, p2,..., p n} composed of n parts, the interference situation of part p i (i ∈ n) when moving along the positive direction k (k ∈ {±x, ±y, ±z}) of the space rectangular coordinate system O-XYZ with part p j (i ∈ n) is expressed as:

[0016]

[0017] In formula (1), I ijk is a 0-1 variable, indicating the interference situation of the part when moving with other parts;

[0018] Accordingly, part p j When moving along the negative direction -k (-k ∈ {±x, ±y, ±z}) of the space rectangular coordinate system O-XYZ, the interference situation with part p i is expressed as I ji-k ;

[0019] I is obtained from the kinematic relationship of the components ji-k = I ijk Therefore, the interference situations of the components moving along the 6 directions of +x, +y, +z, -x, -y, -z are transformed into the interference matrix of moving along the 3 directions of {+x, +y, +z}. Then the moving interference matrix I is:

[0020]

[0021] In formula (2), when part p i moves along direction k and does not interfere with the other n - 1 components, the movement in this direction is feasible; when and only when all the matrix elements of the n parts in formula (2) are 0, I ijk = 0; that is:

[0022] I ijk = {I i1k |I i2k |…|I ijk |…|I ink}(3)

[0023] At this time, the component is preferentially moved in this direction. The elements d in in the i-th row and the elements d ni in the i-th column of the interference matrix I are both replaced by 0, and the movement direction and sequence at this time are recorded; the movement feasibility of other components is judged in the same way. Then all the elements of the interference matrix are 0 after the final movement, and the finally obtained interference matrix is a zero matrix.

[0024] Furthermore, in step S2, the contact matrix describes the contact situation between the components of the assembly and is represented in matrix form;

[0025] If part p i contacts part p j , the contact relationship Q ij = 1, otherwise Q ij = 0. Its expression is:

[0026]

[0027] In formula (4): Q ij is a 0-1 variable representing the contact relationship between parts; then the contact matrix Q of n parts is:

[0028]

[0029] In formula (5), replace the \(i\)-th row element \(q\) of the contact matrix \(Q\) and the \(i\)-th column element \(q\) with 0; assuming the sequence is feasible, except for the change in the contact relationship between the moved part and the unmoved parts, the relative contact relationships between other parts remain unchanged; during the moving process, it is necessary to ensure that there is at least one contact relationship between the unmoved parts and the assembly, and the finally obtained contact matrix is a zero matrix. in and the \(i\)-th column element \(q\) ni are both replaced with 0; assuming the sequence is feasible, except for the change in the contact relationship between the moved part and the unmoved parts, the relative contact relationships between other parts remain unchanged; during the moving process, it is necessary to ensure that there is at least one contact relationship between the unmoved parts and the assembly, and the finally obtained contact matrix is a zero matrix.

[0030] As the number of parts increases, the theoretical number of assembly sequences grows exponentially. Therefore, under the condition of meeting the geometric feasibility of the assembly sequence, it is extremely crucial to screen and evaluate the assembly sequences. Reasonably constructing an evaluation method for the assembly sequence is a prerequisite for outputting the optimal or approximate optimal assembly sequence. The factors affecting the product assembly order are multifaceted. The assembly sequence cost reflects the efficiency and rationality of the assembly sequence. The assembly sequence planning essentially measures the assembly cost incurred in the entire assembly process.

[0031] Furthermore, the basic part of the assembly is the part that should be assembled first in the assembly, which has the characteristics of large mass, large volume, and a large number of geometric constraints with other parts. The basic part of the assembly, Base, is selected according to formula (6):

[0032] Base = argmax(B i )(6)

[0033] In formula (6), B i represents the evaluation score of the basic part of the assembly corresponding to the part \(p\) i in the assembly, as shown in formula (7):

[0034]

[0035] In formula (7), \(v\) i , \(c\) i respectively represent the volume and the number of geometric constraints of the \(i\)-th part \(p\) n in the assembly \(P=\{p1, p2, \ldots, p\) i \};

[0036] Furthermore, in step S2, three evaluation indicators of the following three assembly sequences are defined respectively from three dimensions: assembly direction transformation, assembly tool transformation, and assembly operation difficulty:

[0037] The assembly direction transformation cost AOCC is described by formula (8):

[0038]

[0039] In formula (8), aocc i represents the cost of changing the assembly direction of the i-th part s n to be assembled in the assembly sequence S = {s1, s2,..., s i}. During the assembly process, the more times the assembly direction changes, the more complex the assembly process is, and the higher the assembly time and cost required to complete the assembly body. Therefore, the number of assembly direction changes should be minimized as much as possible. aocc i is calculated as shown in formula (9):

[0040]

[0041] In formula (9), o i ∈ {±x, ±y, ±z} is the assembly direction adopted when assembling part s i ;

[0042] The cost of changing the assembly tool, ATCC, is described by formula (10):

[0043]

[0044] In formula (10), atcc i represents the cost of changing the assembly tool of the i-th part s n to be assembled in the assembly sequence S = {s1, s2,..., s i}. During the assembly process, the more times the assembly tool changes, the more complex the assembly process is, and the higher the assembly time and cost required to complete the assembly body. Therefore, the number of tool changes should be minimized as much as possible. atcc i is calculated as shown in formula (11):

[0045]

[0046] In formula (11), t i ∈ {tool1, tool2,..., tool n} represents the type of assembly tool used when assembling part s i , and n represents the total number of types of assembly tools used in the assembly process, and its value is determined by the total number of tool types used in the actual assembly process;

[0047] The calculation steps of the cost of the assembly operation difficulty include:

[0048] S21: Establish assembly rules, including the first assembly rule, the second assembly rule, the third assembly rule, the fourth assembly rule, the fifth assembly rule, the sixth assembly rule, and the seventh assembly rule;

[0049] The first assembly rule is "assemble the basic parts first"; the second assembly rule includes "assemble the heavy and large parts first, and then the light and small parts"; the third assembly rule includes "assemble the symmetric parts first, and then the asymmetric parts"; the fourth assembly rule includes "assemble the parts with more connection relationships first, and then the parts with fewer connection relationships"; the fifth assembly rule includes "assemble the parts with interference fit first, and then the parts with transition fit"; the sixth assembly rule includes "assemble the parts at the bottom layer of the assembly tree first, and then assemble layer by layer upwards"; the seventh assembly rule includes "assemble the internal parts first, and then the external parts".

[0050] S22: According to the above assembly rules, define the assembly operation difficulty pr n of the i-th part s i to be assembled in the assembly sequence S = {s1, s2,..., s i . Specifically, it includes:

[0051] When the part s i meets the first assembly rule, pr i = 1;

[0052] When the part s i does not meet the first assembly rule, but meets three to six of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 2;

[0053] When the part s i does not meet the first assembly rule, but meets one or two of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 3;

[0054] When the part s i does not meet the first assembly rule and does not meet any of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 4;

[0055] S23: According to the method in S22, obtain the assembly operation difficulty sequence Pr = {pr1, pr2,..., pr n} corresponding to the assembly sequence S = {s1, s2,..., s n . During the assembly process, the parts with greater assembly operation difficulty are assembled first. Calculate the assembly operation difficulty cost AODC of the assembly sequence S according to formula (12):

[0056]

[0057] In formula (12), σ aodc (i, u) is calculated as shown in formula (13):

[0058]

[0059] Furthermore, for the assembly sequence set SS = [S1, S2, …, S n , the method for calculating the assembly sequence cost of one assembly sequence is as shown in formula (14):

[0060] Cost = ω1 × AOCC + ω2 × ATCC + ω3 × AODC (14)

[0061] In formula (14), ω1, ω2, and ω3 are the weight coefficients corresponding to the assembly direction transformation cost, the assembly tool transformation cost, and the assembly operation difficulty cost respectively, and satisfy:

[0062] ω1 + ω2 + ω3 = 1 (15)

[0063] Construct a sequence evaluation system based on the quantitative evaluation of the constructed assembly sequence cost, and add a penalty factor ω4 to the infeasible sequence to obtain the final objective function F(a) as shown in formula (16):

[0064]

[0065] According to the needs of engineering practice, determine the values of the weight coefficients of each index, and evaluate the quality of the assembly sequence according to the magnitude of the fitness value. The smaller the fitness value F(a), the smaller the assembly cost corresponding to the assembly sequence, and the better the assembly quality, that is, the better the assembly sequence; the smaller the calculated fitness value F(a) of the assembly sequence, the lower the cost of the assembly sequence and the better the precision quality of the assembly sequence.

[0066] Furthermore, the improved flower pollination algorithm based on multiple strategies is specifically as follows:

[0067] Step S301: Initialize the algorithm parameters, and specify the total number of parts i, the initial population size of pollen n, and the maximum number of algorithm iterations N;

[0068] Step S302: Generate the initial population. Define the conversion parameter P according to the strategy, and generate the initial population and screen the feasible pollen through the initial population generation rule, that is, the assembly sequence;

[0069] Step S303: Calculate the fitness values of each feasible pollen, record the feasible pollen in the feasible sequence set FP, and generate the offspring population from the feasible sequence set FP using the mutation operation;

[0070] Step S304: Generate a random number Rand, and determine whether the random number Rand is greater than the conversion parameter P. If so, perform cross-pollination and update the current pollen position; if not, perform self-pollination and update the pollen position.

[0071] Step S305: Generate new pollen according to the mutation strategy and calculate the latest fitness value;

[0072] Step S306: Determine whether the termination condition is reached. If so, terminate the algorithm and output the optimal sequence; otherwise, execute Step S303.

[0073] Furthermore, each part of the gas turbine is set as the pollen on the same kind of plant, and the generation of the assembly sequence is realized through the cross-pollination process of organisms, i.e., the global search link, and the non-biological self-pollination process, i.e., the local search link;

[0074] Assume that each flowering plant has only one flower, each flower has a pollen gamete, and one gamete corresponds to a solution to the problem to be solved. Assume the following 4 regulations:

[0075] The cross-pollination of flowering plants corresponds to the exploration behavior of the algorithm. Global pollination is carried out by pollinators such as bees carrying pollen and using Lévy flight; biological self-pollination corresponds to the exploitation behavior of the algorithm, and local pollination is realized by pollination between different flowers of the same kind of plant; the reproduction probability corresponds to the constancy of the flower, and there is a certain proportional relationship between the similarity between two flowers in evolution and its value; the parameter adjusts the mutual conversion between the global pollination and local pollination of the flower pollination algorithm, and due to the influence of the proximity of the positions of flower individuals and the wind, the pollination is more biased towards self-pollination;

[0076] As can be seen from the above, cross-pollination and self-pollination are the core of the flower pollination algorithm. The global pollination of the algorithm is realized by Equation (17), and the local pollination is realized by Equation (21);

[0077] The formula for global pollination is as follows:

[0078]

[0079] Among them, correspond to the solutions obtained in the t-th and (t + 1)-th iterations respectively, x best is the optimal solution obtained after each iteration, γ is the scaling factor for controlling the step size, L(λ) is the Lévy flight displacement corresponding to the flower individual, and the calculation of L(λ) is as shown in Equation (18):

[0080]

[0081] Among them, s0 represents the set minimum step size, λ = 3 / 2, G(λ) is the standard gamma function, and s is obtained from Equation (19):

[0082]

[0083] Among them, the values of μ and v follow a Gaussian distribution, μ ~ N(0, σ 2 ), v ~ N(0, 1), and σ 2 is obtained from Equation (20):

[0084]

[0085] The local pollination optimization of individual flowers is locally searched by Equation (21):

[0086]

[0087] Among them, and are two different random solutions in the optimization process, and ε is a random number uniformly distributed on the interval [0 - 1].

[0088] Furthermore, according to the bionic principle of the basic flower pollination algorithm, it can be known that it is composed of global search and local search fused by the parameter p, and the mathematical modeling is realized by using Formula (17) and Formula (21), laying a theoretical foundation for solving a series of complex optimization problems; while the qualitative analysis of the search mechanism of the basic flower pollination algorithm shows that the deficiencies existing in the search strategy of the basic flower pollination algorithm restrict the convergence effect of the algorithm, including the disadvantages of slow convergence speed and easy to fall into local optimum. For these adverse factors affecting the algorithm performance, this patent makes the following multi-strategy improvements to the flower pollination algorithm:

[0089] In order to balance the global search and local search of the flower pollination algorithm, the fixed conversion probability p is changed to a dynamic conversion probability P that changes adaptively with the number of iterations α , that is, Equation (22):

[0090] P α = w max -(w max - w min )α

[0091] α is calculated by Equation (23), that is:

[0092]

[0093] In the formula: w max and w min are the maximum and minimum values of w respectively; t is the current number of iterations, and t max is the maximum number of iterations; set w max = 0.9, w min = 0.1 at the beginning of the iteration, and P αThe value of is relatively large, so the algorithm focuses on global search, effectively enhancing the global search ability and making the individuals in the population closer to the optimal solution; as the iteration progresses, P α becomes smaller and smaller, making the algorithm more inclined to local fine search, which is conducive to quickly finding the optimal solution in the later stage of the algorithm.

[0094] In the basic flower pollination algorithm, the coefficient of the maternal pollen is 1 during the global search process. To be able to adjust the dependence of the pollen position on the maternal pollen position, a new dynamic factor ω is introduced, as shown in Equation (24):

[0095]

[0096] In the formula, ω max and ω min represent the maximum and minimum values of ω, taking 0.9 and 0.2;

[0097] The improved global search formula by Equation (17) is Equation (25):

[0098]

[0099] In the early stage of iteration, ω t is smaller, weakening the influence of the maternal pollen position, enabling the pollen to perform Levy search more freely. In the later stage of iteration, it mainly focuses on local search and is prone to falling into the local optimal value. At this time, ω t is smaller, and the influence of the maternal pollen position is smaller, increasing the ability of the algorithm to jump out of the local optimal value.

[0100] During the algorithm iteration process, the population diversity also decreases accordingly, and the algorithm is prone to falling into the local optimal solution, which is a common problem of swarm intelligence algorithms. To improve the ability of the algorithm to jump out of the local optimal value, a mutation strategy is introduced. Regarding the current optimal pollen position as a fulcrum, each pollen is updated to a random position in the symmetric direction of the fulcrum, increasing the population diversity. The mathematical expression is as shown in Equation (26):

[0101]

[0102] In the formula, λ1 and λ2 are two random numbers in the interval (0, 1), S is the somersault factor, and g b is the current optimal solution.

[0103] It can be seen from the formula that each pollen finds a random position between the current position and the symmetric position and searches within a new search range closer to the optimal solution. As the number of iterations increases, the position of each pollen gradually approaches the optimal solution, the position fluctuation and the search space gradually shrink, and the search process gradually becomes more accurate and efficient. Therefore, as the number of iterations increases, the mutation range will also automatically shrink, which means that the search process will become more accurate and efficient.

[0104] In each iteration, the fitness values are compared between the current pollen and the mutated pollen, and the position of the better pollen is selected for update. When the pollen falls into a local optimum, it may be replaced by the position of the mutated pollen in the mutation strategy, thus jumping out of the local optimum. The mutation strategy is updated around the position of the optimal pollen, making the algorithm have stronger convergence, being more likely to approach the global optimum, and at the same time helping to accelerate the convergence process of the algorithm and enabling it to find the optimal solution faster.

[0105] Compared with the prior art, the present invention has the following technical effects:

[0106] (1) A feasible regularization strategy is designed, so that the assembly sequences obtained by the algorithm are all feasible assembly sequences without interference. Based on the information of the assembly body, the assembly direction transformation cost, the assembly tool transformation cost, and the assembly operation difficulty cost are set as the evaluation indexes of the assembly cost of the assembly sequence, and a quantitative evaluation of the assembly sequence cost is constructed to build a sequence evaluation system.

[0107] (2) The advantages and disadvantages of applying the flower pollination algorithm to the assembly sequence planning are studied, and an improved flower pollination algorithm based on multiple strategies is proposed to optimize the assembly sequence for the random assembly sequence. The improved algorithm can improve the convergence efficiency of the assembly sequence planning optimization, has more advantages than a single algorithm, and can solve the problem of the sequence optimization falling into a local optimum, providing new ideas and efficient methods for the research of assembly planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 It is a flowchart of the improved flower pollination algorithm in the present invention.

[0109] Figure 2 It is a schematic diagram of the gas turbine component assembly example in the present invention.

[0110] Figure 3 It is a schematic diagram of the part assembly information of the gas turbine components of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0111] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0112] The core idea of the technical solution of the present invention is: design a feasible regularization strategy, so that the assembly sequences obtained by the algorithm are all feasible assembly sequences without interference. Based on the information of the assembly body, the assembly direction transformation cost, the assembly tool transformation cost, and the assembly operation difficulty cost are set as the evaluation indexes of the assembly cost of the assembly sequence, and a quantitative evaluation of the assembly sequence cost is constructed to build a sequence evaluation system.

[0113] Study the advantages and disadvantages of the flower pollination algorithm applied to assembly sequence planning, propose an improved flower pollination algorithm based on multiple strategies, and optimize the assembly sequence for random assembly sequences. The improved algorithm can improve the convergence efficiency of assembly sequence planning optimization, has more advantages than a single algorithm, and can solve the problem that the sequence optimization falls into a local optimal solution, providing new ideas and efficient methods for assembly planning research.

[0114] Taking Figure 3 the three-dimensional model diagram of the "gas turbine component assembly" example shown as the implementation example, this invention illustrates the method for obtaining the optimal assembly sequence. The implementation process mainly can be divided into the following steps:

[0115] S1: Establish an assembly information model to obtain the three-dimensional model information of the assembly and parts;

[0116] According to Figure 3 the "gas turbine component assembly" shown as the implementation example, use the structural hierarchy model of the assembly and the method of unified coding of similar parts to code the parts of the gas turbine components. The 67 parts of the gas turbine components can be numbered 20, greatly reducing the number of parts. The components are coded in the decimal system of real number coding, and each digital code represents the corresponding component.

[0117] S2: Extract the interference matrix and contact matrix of the assembly, analyze the factors affecting the assembly according to the information of the assembly, construct a sequence evaluation system, quantify the evaluation sequence to obtain the corresponding objective function value, select the decimal coding for the components of the assembly, set the initial parameters of the algorithm, and randomly generate an initial sequence;

[0118] First, extract the interference matrix and contact matrix:

[0119] The interference matrix represents the interference situation between each part and other parts when assembling along a certain coordinate axis direction in the Cartesian coordinate system, reflecting the spatial constraint relationship between the parts of the assembly, and is used to judge the geometric feasibility of the assembly sequence.

[0120] Open the three-dimensional model of the gas turbine components that needs assembly planning in CAD software, turn on the "stop operation when colliding" function, select the corresponding parts in turn, and move them in the +x, +y, +z directions in the Cartesian coordinate system in the disassembly manner until they are separated from the whole assembly. Detect the dynamic collision between parts during this disassembly process, observe whether there is interference with other parts during the dragging process, and obtain the interference matrix I of the assembly in the directions of k ∈ {±x, ±y, ±z} MK ;

[0121] The contact matrix represents the contact situation of a certain part with other parts in the assembly, reflecting the connection constraint relationship between components. Open the 3D model of the gas turbine components that need assembly planning in CAD software, use the "check collision function", and detect contact collisions among all components to obtain the contact matrix CM.

[0122] Design an assembly sequence evaluation method based on the cost of the assembly sequence:

[0123] Select the base part in the assembly sequence. The volume, assembly tool category, assembly direction, and geometric constraint quantity information of each part in the assembly can be directly extracted from the CAD model file. The connection constraint quantity of each part in the assembly can be obtained from the contact matrix CM of the assembly. According to the model information and the formula mentioned above, calculate the evaluation score of the base part corresponding to the part with different part numbers in the gas turbine component assembly, and it can be obtained that the base part of this gas turbine component assembly is the front bearing, and the part number is 1.

[0124] Construct an evaluation index for the cost of the assembly sequence

[0125] According to the following seven assembly rules, the assembly operation difficulty of each part in the assembly can be obtained: including the first assembly rule, the second assembly rule, the third assembly rule, the fourth assembly rule, the fifth assembly rule, the sixth assembly rule, and the seventh assembly rule:

[0126] The first assembly rule: Give priority to assembling the base part;

[0127] The second assembly rule: Give priority to assembling heavy and large parts, and then assemble light and small parts;

[0128] The third assembly rule: Give priority to assembling symmetric parts, and then assemble asymmetric parts;

[0129] The fourth assembly rule: Give priority to assembling parts with more connection relationships, and then assemble parts with fewer connection relationships;

[0130] The fifth assembly rule: Give priority to assembling parts with interference fit, and then assemble parts with transition fit;

[0131] The sixth assembly rule: Give priority to assembling the parts at the bottom layer of the assembly tree, and then assemble layer by layer upwards;

[0132] The seventh assembly rule: Give priority to assembling internal parts, and then assemble external parts;

[0133] According to the above 7 assembly rules, the assembly operation difficulty of different parts in the gas turbine component assembly can be obtained. Through the volume, geometric constraint quantity, connection constraint quantity, assembly direction, assembly tool category, and assembly operation difficulty of each part in the assembly, the cost of the assembly sequence can be calculated.

[0134] According to the aforementioned assembly rules, define the assembly sequence S = {s1, s2,..., s n} and the assembly operation difficulty pr i of the i-th part s i to be assembled, specifically including:

[0135] When the part s i meets the first assembly rule, pr i = 1,

[0136] When the part s i does not meet the first assembly rule, but meets three to six of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 2,

[0137] When the part s i does not meet the first assembly rule, but meets one or two of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 3,

[0138] When the part s i does not meet the first assembly rule and does not meet any one of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 4,

[0139] Obtain the assembly operation difficulty sequence Pr corresponding to the assembly sequence. During the assembly process, the assembly process of the part with a greater assembly operation difficulty is more complex, and the assembly time and cost consumed for assembling such parts are also higher. Therefore, it should be assembled preferentially. The assembly operation difficulty cost AODC of the assembly sequence S is calculated by the following formula:

[0140]

[0141] In the formula, σ aodc (i, u) is calculated as shown in the following formula:

[0142]

[0143] Furthermore, for the set of assembly sequences SS, the method for calculating the assembly sequence cost of one of the assembly sequences is shown in the following formula:

[0144] Cost = ω1 × AOCC + ω2 × ATCC + ω3 × AODC

[0145] Where ω1, ω2, and ω3 are the weight coefficients corresponding to the assembly direction transformation cost, the assembly tool transformation cost, and the assembly operation difficulty cost, respectively, and satisfy:

[0146] ω1 + ω2 + ω3 = 1

[0147] Construct a sequence evaluation system based on the quantitative evaluation of the assembly sequence cost constructed by the present invention, add a penalty factor ω4 to the infeasible sequence, and obtain the final objective function F(a) as shown in the formula:

[0148]

[0149] According to the needs of engineering practice, reasonably determine the values of the weight coefficients of each index, and evaluate the quality of the assembly sequence according to the magnitude of the fitness value. The smaller the fitness value F(a), the smaller the assembly cost corresponding to the assembly sequence, and the better the assembly quality, that is, the better the assembly sequence.

[0150] S3: Improve the flower pollination algorithm with multiple strategies. Solve and screen the initial pollen population according to the assembly feasibility criterion, and perform the global search link through the biological cross-pollination process. In order to prevent the algorithm from falling into local optimality and not finding the optimal assembly sequence, a mutation operation is adopted to generate a new population. Before each iteration of the algorithm, perform a mutation operation on the individuals in the population, thereby expanding the search space of the population, increasing diversity, helping to avoid falling into local optimal solutions and improving the global optimization ability. Generate offspring assembly sequences from the pollen generated by local pollination using partial matching crossover operation. However, the new pollen sometimes does not represent a meaningful sequence.

[0151] The process of the improved flower pollination algorithm is as follows:

[0152] Step S301: Initialize the algorithm parameters, and specify the total number of parts i, the initial pollen population size n, and the maximum number of algorithm iterations N;

[0153] Step S302: Generate the initial population. Define the conversion parameter P according to the strategy, generate the initial population according to the initial population generation rule, and screen the feasible pollen, that is, the assembly sequence;

[0154] Step S303: Calculate the fitness values of each feasible pollen, record the feasible pollen in the feasible sequence set FP, and generate an offspring population from the feasible sequence set FP using the mutation operation;

[0155] Step S304: Generate a random number Rand, and determine whether the random number Rand is greater than the conversion parameter P. If so, perform cross-pollination and update the current pollen position; if not, perform self-pollination and update the pollen position.

[0156] Step S305: Generate new pollen according to the mutation strategy and calculate the latest fitness value;

[0157] Step S306: Determine whether the termination condition is reached. If so, terminate the algorithm and output the optimal sequence; otherwise, execute step S303;

[0158] Through the improvement of the flower pollination algorithm by integrating multiple strategies, each part of the gas turbine component is set as the pollen on the same kind of plant. Then, the generation of the assembly sequence is mainly achieved through the biotic cross-pollination process, i.e., the global search link, and the abiotic self-pollination process, i.e., the local search link.

[0159] The initial quality and diversity of the population will affect the time required for the intelligent algorithm to reach the optimal solution and the quality of the optimal solution. The generation of the initial population of the flower pollination algorithm is achieved by randomly generating solutions, resulting in the random distribution of population individuals in the entire solution space without considering the feasibility of these random solutions in the assembly sequence planning problem. Therefore, the present invention proposes an initial population generation rule to ensure the feasibility of the initial assembly sequence. The initial pollen is randomly selected by roulette wheel. However, to generate an initial population that is feasible for assembly, the assembly feasibility criterion is used to judge during the generation of the initial pollen population, and only those that meet the assembly feasibility criterion can be used as the initial population. The feasibility of the assembly sequence can be listed as follows: First, in a sequence, if the assembly precedence relationship is not satisfied, then the sequence is infeasible; second, in a sequence, if the assembly interference of each part is greater than 0, then the sequence is infeasible; avoiding the above two conditions, the sequence is feasible. The assembly sequence feasibility f seq The judgment expression is as follows:

[0160]

[0161] In the formula, pos i represents the assembly order of the i-th part P i ; pos j represents the assembly order of the j-th part P j ; p ij <0 indicates that there is an assembly precedence relationship between the i-th part P i and the j-th part P j ; represents that the interference times are greater than 0.

[0162] S4: Solve using the improved flower pollination algorithm:

[0163] Based on the improved flower pollination algorithm described above, use a computer programming language to implement a gas turbine assembly sequence planning method based on the multi-strategy improved flower pollination algorithm, and obtain the optimal assembly sequence. The optimal gas turbine assembly sequence includes the number of times of assembly direction change, the number of times of assembly tool change, and the minimum fitness value of each part;

[0164] The solution of the improved flower pollination algorithm is to repeatedly iterate the steps S302 - S306 of the algorithm with multi - strategy improvement through Python programming. The iteration number N = N + 1. It is judged whether the current iteration number is greater than the maximum iteration number N. If so, the iteration process ends and the optimal gas turbine assembly sequence is output, including the number of changes in the assembly direction of each part, the number of changes in the assembly tool, and the minimum fitness value; if not, continue to iterate until the iteration number is greater than the maximum iteration number until the optimal assembly sequence is output.

[0165] The fitness value of the optimal gas turbine assembly sequence will change correspondingly with the iteration number. The improved flower pollination algorithm is compared with four optimization algorithms, namely genetic algorithm (GA), basic flower pollination algorithm (FPA), ant colony algorithm (ACO), and particle swarm algorithm (PSO) in terms of the minimum fitness value, running time, the number of changes in the assembly direction, and the number of changes in the assembly tool to verify the effectiveness of the improved flower pollination algorithm in the gas turbine assembly sequence planning. As shown in the following table:

[0166]

[0167] The experimental results of the gas turbine assembly sequence planning method based on the improved multi - strategy flower pollination algorithm proposed in the present invention are shown in the following figure, which characterizes the change of the fitness value of the objective function with the iteration number. It can be seen from the figure that the optimal fitness value of the improved flower pollination algorithm is 5.74, the optimal fitness value of the genetic algorithm is 12.78, and the optimal fitness value of the traditional flower pollination algorithm is also 7.52. However, the convergence speed of the genetic algorithm and the traditional flower pollination algorithm is not as fast as that of the improved flower pollination algorithm.

[0168] To further compare and analyze the optimization effects of each method, this article compares with the other four algorithms. The experimental results show that the number of changes in the assembly direction of the improved flower pollination algorithm is 4 times and the number of changes in the assembly tool is 6 times, which are the smallest among the other algorithms and are all optimal. The present invention improves the assembly sequence optimization algorithm for the assembly sequence planning of gas turbine assembly parts, significantly improving its convergence speed and the generation of the optimal assembly sequence.

Claims

1. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies, the method comprising the following steps: S1: Establish an assembly information model and obtain the three-dimensional model information of the assembly and parts; For an assembly P = {p1, p2,..., p n} composed of n parts, there are several assembly sequences S = {s1, s2,..., s n}, and the assembly sequence S describes the part assembly order for realizing the assembly process of the assembly P. Among them, The i-th element s in the assembly sequence i (1, 2, …, n) refers to the part numbers of the parts to be assembled in the assembly during the i-th assembly sequence operation in the assembly process; The assembly information includes: the interference matrix and contact matrix of the assembly, the volume of the parts, the number of geometric constraints, the number of connection constraints, the assembly direction, the type of assembly tool, and the difficulty of assembly operation; during the assembly process, the parts with larger volume and more geometric constraints are used as the basic parts of the assembly to be assembled first or preferentially assembled, and the cost of assembly operation difficulty is set as an evaluation index. The parts with greater assembly operation difficulty are assembled first; the cost of assembly direction change and the cost of assembly tool change are set as evaluation indexes to reduce the number of assembly direction changes and the number of assembly tool changes during the assembly process. S2: Extract the interference matrix and contact matrix of the assembly, analyze the factors affecting the assembly according to the information of the assembly, construct a sequence evaluation index, quantify the evaluation sequence to obtain the corresponding fitness function, and encode the components of the assembly in decimal. Each part corresponds to the pollen on the same plant; S3: Generate an initial population, generate an initial population through the initial population generation rule and screen the feasible pollen, that is, the assembly sequence; S4: Calculate the fitness values of each feasible pollen, record the feasible pollen in the feasible sequence set FP, generate an offspring population from the feasible sequence set FP using the crossover operation, and generate new pollen through the mutation operation; S5: Determine whether the termination condition is met. If not, use the latest sequence as the initial population of flower pollination and repeat S4 for iteration; if so, obtain the final global optimal result, that is, the required assembly sequence.

2. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies according to claim 1, characterized in that In step 1, the basic part of the assembly is the part that should be assembled first in the assembly, and the basic part Base of the assembly is selected according to formula (6): Base=argmax(B i )(6) In formula (6), B i represents the evaluation score of the basic part of the assembly corresponding to part p i in the assembly, as shown in formula (7): In formula (7), v i , c i respectively represent the volume and the number of geometric constraints of the n ith part p i in the assembly P = {p1, p2,..., p n}.

3. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies according to claim 1, characterized in that In the step S2, the interference matrix is used to describe the interference situation between the part p in the assembly P i when moving along the directions of {±x, ±y, ±z} in the Cartesian coordinate system and the part p j The interference situation reflects the spatial constraint relationship between the parts of the assembly and is used to judge the geometric feasibility of the assembly sequence; For an assembly P = {p1, p2, …, p n}, when the part p i (i ∈ n) moves along the positive direction k (k ∈ {±x, ±y, ±z}) of the space rectangular coordinate system O-XYZ, the interference situation with the part p j (i ∈ n) is expressed as: In formula (1), I ijk is a 0-1 variable, representing the interference situation between parts when the part moves and other parts; Accordingly, part p j When moving along the negative direction -k (-k ∈ {±x, ±y, ±z}) of the spatial rectangular coordinate system O-XYZ, the interference situation with part p i is represented as I ji-k ; Obtained from the kinematic relationship of components I ji-k = I ijk , so the interference conditions of the components moving in the six directions of +x, +y, +z, -x, -y, -z are transformed into the interference matrix of moving in the three directions of {+x, +y, +z}, and the moving interference matrix I is: In formula (2), part p i moves along direction k, and the movement in this direction is feasible only when there is no interference with the other n - 1 components; if and only if all the matrix elements of the n parts in formula (2) are 0, I ijk = 0; That is: I ijk = {I i1k | I i2k | … | I ijk | … | I ink}(3) At this time, this component is preferentially moved in this direction, and the element d in the i-th row of the interference matrix I in and the element d in the i-th column ni are both replaced with 0, and the moving direction and sequence at this time are recorded; the moving feasibility of other components is judged in the same way, and finally all elements of the interference matrix are 0 after the movement ends, and the finally obtained interference matrix is a zero matrix.

4. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies according to claim 1, characterized in that In step S2, the contact matrix describes the contact situation between the components of the assembly and is represented in the form of a matrix; If part p i contacts part p j and the contact relationship Q ij = 1; otherwise Q ij = 0. Its expression is: In formula (4): Q ij is a 0-1 variable representing the contact relationship between parts; then the contact matrix Q of n parts is: In formula (5), replace the \(i\)-th row element \(q\) in and the \(i\)-th column element \(q\) ni of the contact matrix \(Q\) with 0; assuming the sequence is feasible, except for the change in the contact relationship between the moved part and the unmoved parts, the relative contact relationships between other parts remain unchanged; During the movement process, it is necessary to ensure that there is at least one contact relationship between the unmoved components and the assembly, and the finally obtained contact matrix is a zero matrix.

5. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies according to claim 1, characterized in that, In step S2, the following three evaluation indexes of the assembly sequence are defined from three dimensions of assembly direction change, assembly tool change, and assembly operation difficulty: The assembly direction change cost AOCC is described by formula (8): In formula (8), aocc i represents the cost of the assembly direction transformation of the i-th part s n to be assembled in the assembly sequence S = {s1, s2,..., s i}. During the assembly process, the more times the assembly direction is transformed, the more complex the assembly process is, and the higher the assembly time and cost consumed to complete the assembly. Therefore, the number of assembly direction transformations should be minimized as much as possible. aocc i is calculated as shown in formula (9): In formula (9), o i ∈ {±x, ±y, ±z} is the assembly direction adopted when assembling part s i at that time; The assembly tool change cost ATCC is described by formula (10): In formula (10), atcc i represents the assembly tool change cost of the i-th part s n to be assembled in the assembly sequence S = {s1, s2,..., s i}. During the assembly process, the more times the assembly tool changes, the more complex the assembly process becomes, and the higher the assembly time and cost consumed to complete the assembly. Therefore, the number of tool changes should be minimized. atcc i is calculated as shown in formula (11): In formula (11), t i ∈ {tool1, tool2, …, tool n} represents the type of assembly tool used when assembling part s i , n represents the total number of types of assembly tools used in the assembly process, and its value is determined by the total number of tool types used in the actual assembly process; The calculation steps of the assembly operation difficulty cost include: S21: Establish assembly rules, including the first assembly rule, the second assembly rule, the third assembly rule, the fourth assembly rule, the fifth assembly rule, the sixth assembly rule, and the seventh assembly rule; The first assembly rule is "assemble the basic parts first"; the second assembly rule includes "assemble the heavy and large parts first, and then the light and small parts"; the third assembly rule includes "assemble the symmetric parts first, and then the asymmetric parts"; the fourth assembly rule includes "assemble the parts with more connection relationships first, and then the parts with fewer connection relationships"; the fifth assembly rule includes "assemble the parts with interference fit first, and then the parts with transition fit"; the sixth assembly rule includes "assemble the parts at the bottom layer of the assembly tree first, and then assemble layer by layer upward"; the seventh assembly rule includes "assemble the internal parts first, and then the external parts". S22: According to the above assembly rules, define the assembly operation difficulty pr n of the i-th part s i to be assembled in the assembly sequence S = {s1, s2,..., s i}, specifically including: When part s i meets the first assembly rule, pr i = 1; When part s i does not meet the first assembly rule, but meets three to six of the second, third, fourth, fifth, sixth, and seventh assembly rules, pr i = 2; When part s i does not meet the first assembly rule, but meets one or two of the second assembly rule, the third assembly rule, the fourth assembly rule, the fifth assembly rule, the sixth assembly rule, and the seventh assembly rule, pr i = 3; When part s i does not meet the first assembly rule and does not meet any one of the second assembly rule, the third assembly rule, the fourth assembly rule, the fifth assembly rule, the sixth assembly rule, and the seventh assembly rule, pr i = 4; S23: According to the method in S22, obtain the assembly operation difficulty sequence Pr = {pr1, pr2,..., pr n} corresponding to the assembly sequence S = {s1, s2,..., s n}. During the assembly process, the parts with greater assembly operation difficulty are assembled first. Calculate the assembly operation difficulty cost AODC of the assembly sequence S according to formula (12): In formula (12), σ aodc (i, u) is calculated as shown in formula (13):

6. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies as claimed in claim 1, characterized in that, For the set of assembly sequences SS = [S1, S2, …, S n , the method for calculating the assembly sequence cost of one of the assembly sequences is shown in Equation (14): Cost = ω1×AOCC + ω2×ATCC + ω3×AODC(14) In formula (14), ω1, ω2, and ω3 are the weight coefficients corresponding to the assembly direction transformation cost, the assembly tool transformation cost, and the assembly operation difficulty cost respectively, and satisfy: ω1 + ω2 + ω3 = 1(15) Construct a sequence evaluation system based on the quantitative evaluation of the assembly sequence cost. Add a penalty factor ω4 to the infeasible sequence to obtain the final objective function F(a) as shown in formula (16): According to the needs of engineering practice, determine the values of the weight coefficients of each index, and evaluate the quality of the assembly sequence according to the size of the fitness value. The smaller the fitness value F(a), the smaller the assembly cost corresponding to the assembly sequence, and the better the assembly quality, that is, the better the assembly sequence; the smaller the calculated fitness value F(a) of the assembly sequence, the lower the cost of the assembly sequence and the better the accuracy quality of the assembly sequence.

7. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies according to claim 1, characterized in that, The improved flower pollination algorithm based on multiple strategies is specifically as follows: Step S301: Initialize the algorithm parameters, and specify the total number of parts i, the initial population size of pollen n, and the maximum number of algorithm iterations N; Step S302: Generate the initial population. Define the conversion parameter P according to the strategy, and generate the initial population and screen the feasible pollen through the initial population generation rule, that is, the assembly sequence; Step S303: Calculate the fitness values of each feasible pollen, record the feasible pollen in the feasible sequence set FP, and generate the offspring population from the feasible sequence set FP using the mutation operation; Step S304: Generate a random number Rand, and determine whether the random number Rand is greater than the conversion parameter P. If so, perform cross-pollination and update the current pollen position; if not, perform self-pollination and update the pollen position. Step S305: Generate new pollen according to the mutation strategy and calculate the latest fitness value; Step S306: Determine whether the termination condition is reached. If so, the algorithm terminates and outputs the optimal sequence; otherwise, execute step S303.

8. An assembly sequence planning method based on an improved flower pollination algorithm with multiple strategies as claimed in claim 7, characterized in that, Set each part of the gas turbine as the pollen on the same plant, then the generation of the assembly sequence is realized through the biological cross-pollination process, that is, the global search link, and the non-biological self-pollination process, that is, the local search link; Assume that each flowering plant has only one flower, each flower has one pollen gamete, and one gamete corresponds to a solution to the problem. Assume the following 4 regulations: Cross-pollination of flowering plants corresponds to the exploration behavior of the algorithm, and global pollination is carried out by pollinators such as bees carrying pollen and adopting Lévy flight; self-pollination of organisms corresponds to the exploitation behavior of the algorithm, and local pollination is achieved by pollination between different flowers of the same plant; the reproduction probability corresponds to the constancy of the flower, and there is a certain proportional relationship between the similarity between two flowers in evolution and its value; the parameter adjusts the mutual conversion between global pollination and local pollination of the flower pollination algorithm, and due to the influence of the proximity of the positions of individual flowers and the wind, the pollination is more biased towards self-pollination; As can be seen from the above, cross-pollination and self-pollination are the cores of the flower pollination algorithm. The global pollination of the algorithm is realized by formula (17), and the local pollination is realized by formula (21); The formula for global pollination is as follows: Among them, correspond to the solutions obtained in the t-th and (t + 1)-th iterations respectively, and x best is the optimal solution obtained after each iteration, γ is a scaling factor for controlling the step size, L(λ) is the Lévy flight displacement corresponding to the flower individual, and the calculation of L(λ) is as shown in Equation (18): Among them, s0 represents the set minimum step size, λ = 3 / 2, G(λ) is the standard gamma function, and s is obtained from formula (19): wherein, the values of μ and v follow a Gaussian distribution, μ ~ N(0, σ 2 ), v ~ N(0, 1), and σ 2 is obtained from Equation (20): The local pollination optimization of flower individuals is locally searched by formula (21): Among them, and are two different random solutions in the optimization process, and ε is a random number uniformly distributed on the interval [0 - 1].

9. An assembly sequence planning method for an improved flower pollination algorithm based on multiple strategies according to claim 7, characterized in that To balance the global search and local search of the flower pollination algorithm, the fixed transition probability p is changed to a dynamic transition probability P that adaptively changes with the number of iterations α , that is, Equation (22): P α = w max -(w max - w min )α α is calculated by formula (23), that is: where: w max and w min are the maximum and minimum values of w, respectively; t is the current iteration number, t max is the maximum iteration number; set w max = 0.9, w min = 0.1 at the beginning of the iteration; The coefficient of the maternal pollen in the global search process of the basic flower pollination is 1. In order to be able to adjust the dependence of the pollen position on the maternal pollen position, a new dynamic factor ω is introduced, as shown in formula (24): where ω max and ω min represent the maximum and minimum values of ω, taking 0.9 and 0.2; The improved global search formula from formula (17) is formula (25): A mutation strategy is introduced. The current optimal pollen position is regarded as a fulcrum, and each pollen is updated to a random position in the symmetric direction of the fulcrum to increase the population diversity. The mathematical expression is as shown in formula (26): where λ1 and λ2 are two random numbers in the interval (0, 1), S is the somersault factor, and g b is the current optimal solution.

Images