CN118627387B - A method for clustering optimization design of system actions - Google Patents

Abstract

The invention discloses a system action clustering optimization design method which comprises the steps of decomposing and obtaining detailed actions required for realizing system functions through function-action analysis, constructing action label vectors, performing action clustering based on similarity of the action label vectors to obtain action subsets to be designed integrally, and performing integrated design on actions by adopting an integrated action combination optimization method based on branch delimitation so as to obtain integrated physical domain design parameters/components through corresponding mapping. The invention can fully analyze the relation among functional requirements, behaviors and actions in the system design process, can realize the integration of physical domain design parameters through the clustering of similar actions, effectively reduce the design redundancy, and can consider the constraint in the design process and the cost brought by the integrated design in the integrated optimization process based on branch delimitation so as to select the design scheme with optimal comprehensive performance.

Description

System action clustering optimization design method

Technical Field

The invention belongs to the field of system engineering, and relates to a system action clustering optimization design method.

Background

Axiom design is a novel product design concept theory, and the core content of the axiom design is the mapping design process among a user domain, a functional domain, a physical domain and a process domain. The system design method provides a set of standards and specifications for the system design process through the effect of design axiom, can effectively reduce the influence of human factors of designers in the design process, guides the generation of innovative design, and is a research hotspot in the fields of current system engineering, industrial design and the like.

At present, research on axiom design methods comprises expansion of design domains, combination with other design methods, application research of axiom design theory and the like, and the research further expands the application range of axiom design, but still has certain problems. For example, in the axiom design process, the functional requirements and the physical domain design parameters show a one-to-one correspondence, and when the functional requirements of the system are numerous and cross, a great amount of design redundancy is caused, which is not beneficial to fully utilizing design resources and space. Therefore, an innovative design method is needed to fully analyze the interaction relation of actions among the intra-system and the inter-component systems and to perform clustering optimization on similar actions so as to realize integrated design.

Disclosure of Invention

In order to solve the problems in the background technology, the invention provides a systematic action clustering optimization design method.

The invention aims at realizing the following technical scheme:

a systematic action clustering optimization design method comprises the following steps:

Step 1, decomposing and obtaining detailed actions required for realizing system functions through function-behavior-action analysis, and providing a basis for action clustering optimization, wherein the specific steps are as follows:

Step 1.1, demand-function analysis:

Starting from the top-level user demand, decomposing from top to bottom to obtain a realization way-behavior of the function;

step 1.2, behavior-action analysis:

for the obtained behaviors, an 'input and output' black box model is established, and the realization approach of the behaviors is obtained by decomposition based on a functional system analysis technology, so that the system actions are obtained;

step 2, constructing an action tag vector, and performing action clustering based on the similarity of the action tag vector to obtain an action subset to be designed integrally, wherein the specific steps are as follows:

step 2.1, construction of action tag vectors:

summarizing and analyzing the types of the action labels according to the results of the function-behavior-action analysis;

step 2.2, calculating action similarity:

① Normalizing the number of the labels;

② The word frequency is used for representing the occurrence frequency of the tag in the action, and a specific calculation formula of the word frequency tf _w corresponding to the tag w in a certain action is as follows:

③ The reverse text frequency idf of a specific label is calculated, and a specific calculation formula is as follows:

Wherein D is the total number of actions, |{ j: t _i∈d_j } | represents the number of actions comprising the tag t;

④ For an action j, the TF-IDF value w _ij of tag i is:

w_ij＝tf_ij·idf_i

weighting the label vector of the action to obtain a weighted label vector;

⑤ Based on a cosine similarity calculation formula, calculating the similarity between every two actions;

Step 2.3, action clustering based on K-means algorithm:

① Selecting an action cluster number initial value k;

② Selecting k clustering centers, and performing action clustering based on a traditional k-means algorithm;

③ Calculating the maximum Euclidean distance between the actions in each cluster and the central point in the action clustering result;

④ Judging whether the maximum Euclidean distance change of two adjacent times is smaller than a threshold value, if so, turning to ⑤, otherwise, turning to ② by making k=k+1;

⑤ Ending the action clustering and outputting an action clustering result;

step 3, aiming at the action subset obtained by clustering, adopting an integrated action combination optimization method based on branch delimitation to carry out integrated design on actions, thereby obtaining integrated physical domain design parameters/components by corresponding mapping, and specifically comprising the following steps:

step 3.1, constructing an action integrated optimization model:

① According to the action TF-IDF label vector, after clustering based on label similarity, action clustering with certain similarity is realized, a plurality of action subsets are obtained, and the actions in each action subset have the possibility of integrated design;

② The integrated optimization design is carried out aiming at the actions in each action subset, and the integrated design problem is converted into a mixed integer programming problem, namely, under the condition that a certain constraint condition is met, the actions taking extreme values of a given objective function are solved, namely:

Find x=[x₁,x₂,…,x_n],(x_i=0/1)

min J=αM+βC+ηW

φ_i(X_i)≤0

Wherein, the parameter x= [ x ₁,x₂,…,x_n ] to be designed is a characteristic variable of whether the actions are combined, and n represents n components of the parameter to be designed:

Wherein i is an action number, M is a structural mass, C is cost, W is power, alpha, beta and eta are weights, Phi _i(X_i) is a single-component constraint;

step 3.2, mixed integer programming solution based on branch delimitation:

pruning operations are performed when the following occurs:

Pruning condition 1, constraint conflict corresponding to actions cannot be combined, and the branches are eliminated;

pruning 2, namely, merging the actions at excessive cost, so that the performance index value becomes large, and discarding the branches;

pruning, 3, namely, the performance index is not obviously reduced, temporary reservation is selected, or after a certain number of iterations, the branch is eliminated.

Compared with the prior art, the invention has the following advantages:

The invention can fully analyze the relation among functional requirements, behaviors and actions in the system design process, can realize the integration of physical domain design parameters through the clustering of similar actions, effectively reduce the design redundancy, and can consider the constraint in the design process and the cost brought by the integrated design in the integrated optimization process based on branch delimitation so as to select the design scheme with optimal comprehensive performance.

Drawings

FIG. 1 is a general scheme diagram of a systematic action clustering optimization design method;

FIG. 2 is a schematic diagram of action branch delimitation;

FIG. 3 is a schematic diagram of sub-set 1 action branch delimitation.

Detailed Description

The following description of the present invention is provided with reference to the accompanying drawings, but is not limited to the following description, and any modifications or equivalent substitutions of the present invention should be included in the scope of the present invention without departing from the spirit and scope of the present invention.

The invention provides a systematic action clustering optimization design method, as shown in figure 1, which comprises the following steps:

and step 1, decomposing and obtaining detailed actions required for realizing system functions through function-behavior-action analysis, and providing a basis for action clustering optimization. The method comprises the following specific steps:

Step 1.1, demand-function analysis:

Starting from the top-level user demand, the realization approach-behavior of the function is obtained by decomposing from top to bottom.

Step 1.2, behavior-action analysis:

For the obtained behaviors, an 'input and output' black box model is established, and based on a functional system analysis technology, the realization path of the behaviors is obtained through decomposition, so that the system actions are obtained.

And 2, constructing an action tag vector, and performing action clustering based on the similarity of the action tag vector to obtain an action subset to be designed integrally. The method comprises the following specific steps:

step 2.1, construction of action tag vectors:

And summarizing and analyzing the types of the action labels according to the results of the function-behavior-action analysis. The action tags include input type, output type, constraints, etc. of the action. For a certain action, its corresponding tag vector is:

{ input information, input energy, output information, output energy a. Constraint (data volume) }

The elements in the tag vector should cover all features of the action, and the element types can be freely defined, and are determined by a designer according to specific application scenarios. When the element in the tag vector is 1, this indicates that the action has the tag. For example, when the tag vector is {1,0,1,0,1}, it indicates that its input type is "information", its output type is "information", and its constraint type is "data amount", and at the same time, the input and output do not contain "energy".

Step 2.2, calculating action similarity:

Based on the action tag vector, each action in the design process can be represented in a unified form. The label vector of each action does not have the clustering necessity if the labels exist in most actions, and has good class distinguishing capability and can reflect the characteristics of the actions if the labels exist in less actions. In order to more objectively judge the similarity between actions, the importance degree of the labels needs to be measured and weighted in the similarity calculation process.

First, the frequency of occurrence of a tag (keyword) in this action is expressed by a word frequency (tf). Therefore, the number of labels is normalized (number of label occurrences/total number of labels) first, and the number of labels is prevented from being biased. The specific calculation formula of the word frequency tf _w corresponding to the label w in a certain action is as follows:

the inverse text frequency idf for a particular tag is then calculated, and the number of actions that comprise that tag is divided by the total number of actions, and the resulting quotient is then logarithmically derived.

If fewer documents containing a tag t, the larger idf, the tag is said to have good category discrimination. The specific calculation formula is as follows:

where D is the total number of actions and { j: t _i∈d_j } | represents the number of actions that contain tag t.

High tag frequencies within a particular action, and low tag frequencies for that tag in the entire set of actions, may yield a high weighted TF-IDF value. Thus, TF-IDF values tend to filter out common tags, preserving important tags. Then for an action j, where the TF-IDF value w _ij for tag i is:

w_ij＝tf_ij·idf_i

And weighting the label vector of the action to obtain a weighted label vector. And further calculating the similarity between every two actions based on a cosine similarity calculation formula.

Step 2.3, action clustering based on K-means algorithm:

① Selecting an action cluster number initial value k;

② Selecting k clustering centers, and performing action clustering based on a traditional k-means algorithm;

③ Calculating the maximum Euclidean distance between the actions in each cluster and the central point in the action clustering result;

④ Judging whether the maximum Euclidean distance change of two adjacent times is smaller than a threshold value, if so, turning to ⑤, otherwise, turning to ② by making k=k+1;

⑤ And (5) after the action clustering is finished, outputting an action clustering result.

And 3, aiming at the action subset obtained by clustering, adopting an integrated action combination optimization method based on branch delimitation to carry out integrated design on actions, thereby obtaining integrated physical domain design parameters/components by corresponding mapping. The method comprises the following specific steps:

step 3.1, constructing an action integrated optimization model:

According to the action TF-IDF label vector, after clustering based on label similarity, action clustering with certain similarity is achieved, a plurality of action subsets are obtained, and the actions in each action subset have the possibility of integrated design. Next, an integrated optimization design is performed for the actions within each action subset. The integrated design problem is converted into a mixed integer programming problem, namely, the action integrated combination for taking the extremum of the given objective function under the condition that a certain constraint condition is met is solved. Namely:

The parameter x= [ x ₁,x₂,…,x_n ] to be designed is a characteristic variable of whether the actions are combined or not:

Wherein i is an action number. The purpose of the integrated design is that the structure is light, low cost, low power, etc., so the objective function is a weighting of the structure mass M, cost C, power W, where the weights α, β, η are predetermined by the designer. In addition, constraints of the optimization problem include a combined inequality constraint (a sum of a plurality of component indices ) And a single component constraint phi _i(X_i) (ensuring proper operation).

Step 3.2, mixed integer programming solution based on branch delimitation:

In connection with fig. 2, an integrated optimization design method is given by taking an integrated optimization design process of three actions as an example. Here, the operation 1 (1) indicates that the operation 1 is integrated with other operations, whereas the operation 1 (0) indicates that the operation 1 is not integrated with other operations.

First, it is determined whether or not the operations 1 and 2 are integrated, and when the operations 1 and 2 are integrated, the physical domain component is mapped and the performance index is calculated, and the performance index value is better than the case where the operations 1 and 2 are not integrated, so that branches where the operations 1 and 2 are not integrated are cut off. Further, with the operation 2 (1) as a starting point, it is determined whether or not the operation 2 and the operation 3 should be integrally fused. When the actions 2 and 3 are fused, the design cost is excessive, so that the branch is abandoned, and the actions 1 and 2 are fused into the current optimal design scheme.

In summary, pruning is performed when the following occurs:

Pruning condition 1, constraint conflict corresponding to actions cannot be combined, and the branches are eliminated;

pruning 2, namely, merging the actions at excessive cost, so that the performance index value becomes large, and discarding the branches;

pruning, 3, namely, the performance index is not obviously reduced, temporary reservation is selected, or after a certain number of iterations, the branch is eliminated.

Examples:

in this embodiment, a spacecraft power system including ascending, track-in and track-on functions is designed as an example, and specific steps of the method are described:

Step 1, decomposing and obtaining detailed actions required by 'power capable of providing ascending, track entering and track running' for realizing system functions through function-behavior-action analysis, and providing a basis for action clustering optimization. The method comprises the following specific steps:

Step 1.1, demand-function analysis:

Starting from the top-level user requirement of being capable of providing ascending, track entering and on-track running power, 3 specific functional requirements are obtained that { FR1: the system can provide ascending section gesture track control power, FR2: the system can provide track entering section gesture track control power, and FR3: the system can provide on-track section gesture track control power }. Where FR represents the functional requirement.

The realization way of each function is obtained by decomposition from top to bottom, wherein { BE1 is used for providing ascending section gesture track control power, BE2 is used for providing track entering section gesture track control power, and BE3 is used for providing track on track section gesture track control power }. Wherein BE represents behavior.

Step 1.2, behavior-action analysis:

For the obtained behaviors, an 'input and output' black box model is established, and the realization way of the behaviors is obtained by decomposition based on a functional system analysis technology, namely, the system acts as follows:

BE1 'provides the ascending section gesture control power' the corresponding actions are { AC1-1: outputting ascending section trajectory control thrust }, AC1-2: outputting ascending section gesture control moment }. Wherein AC represents an action.

BE 2' provides the gesture control power of the track-in section and the corresponding actions are { AC2-1: output track-in section track control thrust }, AC2-2: output track-in section gesture control moment }.

BE 3' provides the corresponding actions of the on-orbit attitude and orbit control power } { AC3-1: outputting orbit transfer control thrust, AC3-2: outputting attitude machine power moment }.

And 2, constructing an action tag vector, and performing action clustering based on the similarity of the action tag vector to obtain an action subset to be designed integrally. The method comprises the following specific steps:

step 2.1, construction of action tag vectors:

And summarizing and analyzing to obtain the action label type of the power system according to the result of the function-behavior-action analysis. The power system is characterized in that each action involves the conversion of information and energy in different stages, so that the time period of the action and the type of information energy conversion are used as action labels, and the obtained labels are { time-ascending section, input information type-thrust instruction, time-track entering section, output-energy, time-track entering section, output energy type-moment, input-information, output energy type-thrust, input information type-moment instruction }.

The label vector for each action is obtained as follows:

where each row represents an action, and when an element is 1, this indicates that the action has the tag.

Step 2.2, calculating action similarity:

For the label vector of each action of the power system, if the labels exist in most actions, the clustering necessity is not provided, and if the labels exist in less actions, the label vector has good class distinguishing capability and can reflect the characteristics of the actions. In order to more objectively judge the similarity between actions, the importance degree of the labels needs to be measured and weighted in the similarity calculation process.

1) The word frequency (tf) of the tag (keyword) is calculated. The word frequency tf _w corresponding to the tag w in each action is:

2) Calculating the inverse text frequency idf of each specific label:

where D is the total number of actions and { j: t _i∈d_j } | represents the number of actions that contain tag t.

3) For each action, the TF-IDF value of the tag therein (i.e., the weighted tag vector) is calculated. For an action j, the TF-IDF value w _ij of tag i is:

w_ij＝tf_ij·idf_i

For the system, the TF-IDF value is calculated as follows:

step 2.3, performing action clustering based on a K-means algorithm:

① Selecting an action cluster number initial value k=1, and setting a clustering calculation threshold value to be 1%;

② Selecting k clustering centers, and performing action clustering based on a traditional k-means algorithm;

③ Calculating the maximum Euclidean distance between the actions in each cluster and the central point in the action clustering result;

④ Judging whether the maximum Euclidean distance change of two adjacent times is smaller than a set threshold value of 1%, if so, turning ⑤, otherwise, turning ② to k=k+1;

⑤ And (5) after the action clustering is finished, outputting an action clustering result.

In the calculation process, the distances obtained by each iteration are 0.3618, 0.3545 and 0.3542 in sequence. When the iteration is carried out for the 3 rd time, the distance change from the 2 nd time is within 1%, and the jump is calculated, and the number of the cluster types is 2. Therefore, the clustering obtains 2 action sets, namely { set 1: output ascending track control thrust (AC 1), output orbit transfer control thrust (AC 5), output orbit entering track control thrust (AC 3) } { set 2: output ascending track control moment (AC 2), output orbit entering track control moment (AC 4), and output attitude machine power moment (AC 6) }.

And 3, aiming at the power system action subset obtained by clustering, adopting an integrated action combination optimization method based on branch delimitation to carry out integrated design on actions, thereby obtaining integrated physical domain design parameters/components by corresponding mapping. Take subset 1 as an example for illustration.

Step 3.1, constructing an action integrated optimization model:

For the power system, the purpose of the integrated design is mainly to make the structure light, so the weight of the mass is alpha=1, and the other weights are 0. Then for the power system, the optimization model is:

The parameter x= [ x ₁,x₂,…,x₆ ] to be designed is a characteristic variable of whether the actions are combined or not:

Wherein i is an action number. Phi _i(X_i) is less than or equal to 0, and whether the force/moment provided for the action meets the requirements of the spacecraft.

Step 3.2, mixed integer programming solution based on branch delimitation:

Referring to fig. 3, taking the integrated optimization design process of the subset 1 as an example, an integrated optimization design method is given. Firstly, whether the motion 1 and the motion 3 are integrated or not is judged, and when the motion 1 and the motion 3 are integrated, only one set of propulsion system is adopted in the whole process of the ascending section and the track entering section, so that the structural quality can be saved. However, because the difference of the thrust magnitude constraint of the two flight stages is large, the prior art cannot realize the large-range thrust variation, and the situation 1 accords with the pruning situation, so that the actions 1 and 3 are not combined, and the combined branches are sheared. Similarly, actions 1 and 5 should not be combined. Continuing to start with action 3, it is determined whether action 5 can be combined with it. After the actions 3 and 5 are combined, the required physical domain parts are reduced to 1 set from 2 sets of propulsion systems, the quality is reduced to 14kg from 28kg, and the performance index J is better, so that the actions 3 and 5 are combined.

In summary, the design result is that the output ascending track control thrust (AC 1) is realized by one component, the output track entering track control thrust (AC 3) and the output track changing control thrust (AC 5) are realized by one component.

Claims (4)

1. A systematic action clustering optimization design method is characterized by comprising the following steps:

step 1, decomposing and obtaining detailed actions required for realizing system functions through function-behavior-action analysis, and providing a basis for action clustering optimization;

step 2, constructing an action tag vector, and performing action clustering based on the similarity of the action tag vector to obtain an action subset to be designed integrally, wherein the specific steps are as follows:

step 2.1, construction of action tag vectors:

summarizing and analyzing the types of the action labels according to the results of the function-behavior-action analysis;

step 2.2, calculating action similarity:

① Normalizing the number of the labels;

② The word frequency is used for representing the occurrence frequency of the tag in the action, and a specific calculation formula of the word frequency tf _w corresponding to the tag w in a certain action is as follows:

③ Calculating the reverse text frequency idf of a specific label;

④ For an action j, the TF-IDF value w _ij of tag i is:

w_ij＝tf_ij·idf_i

weighting the label vector of the action to obtain a weighted label vector;

⑤ Based on a cosine similarity calculation formula, calculating the similarity between every two actions;

Step 2.3, action clustering based on K-means algorithm:

① Selecting an action cluster number initial value k;

② Selecting k clustering centers, and performing action clustering based on a traditional k-means algorithm;

③ Calculating the maximum Euclidean distance between the actions in each cluster and the central point in the action clustering result;

④ Judging whether the maximum Euclidean distance change of two adjacent times is smaller than a threshold value, if so, turning to ⑤, otherwise, turning to ② by making k=k+1;

⑤ Ending the action clustering and outputting an action clustering result;

And 3, aiming at the action subset obtained by clustering, adopting an integrated action combination optimization method based on branch delimitation to carry out integrated design on actions, thereby obtaining integrated physical domain design parameters/components by corresponding mapping.

2. The system action clustering optimization design method according to claim 1, wherein the specific steps of the step 1 are as follows:

Step 1.1, demand-function analysis:

Starting from the top-level user demand, decomposing from top to bottom to obtain a realization way-behavior of the function;

step 1.2, behavior-action analysis:

For the obtained behaviors, an 'input and output' black box model is established, and based on a functional system analysis technology, the realization path of the behaviors is obtained through decomposition, so that the system actions are obtained.

3. The system action clustering optimization design method according to claim 1, wherein the calculation formula of the reverse text frequency idf is:

where D is the total number of actions and { j: t _i∈d_j } | represents the number of actions that contain tag t.

4. The system action clustering optimization design method according to claim 1, wherein the specific steps of the step 3 are as follows:

step 3.1, constructing an action integrated optimization model:

① According to the action TF-IDF label vector, after clustering based on label similarity, action clustering with certain similarity is realized, a plurality of action subsets are obtained, and the actions in each action subset have the possibility of integrated design;

② The integrated optimization design is carried out aiming at the actions in each action subset, and the integrated design problem is converted into a mixed integer programming problem, namely, under the condition that a certain constraint condition is met, the actions taking extreme values of a given objective function are solved, namely:

Find x=[x₁,x₂,…,x_n],(x_i=0/1)

min J=αM+βC+ηW

φ_i(X_i)≤0

Wherein, the parameter x= [ x ₁,x₂,…,x_n ] to be designed is a characteristic variable of whether the actions are combined, and n represents n components of the parameter to be designed:

Wherein i is an action number, M is a structural mass, C is cost, W is power, alpha, beta and eta are weights, Phi _i(X_i) is a single-component constraint;

step 3.2, mixed integer programming solution based on branch delimitation:

pruning operations are performed when the following occurs:

Pruning condition 1, constraint conflict corresponding to actions cannot be combined, and the branches are eliminated;

pruning 2, namely, merging the actions at excessive cost, so that the performance index value becomes large, and discarding the branches;

pruning, 3, namely, the performance index is not obviously reduced, temporary reservation is selected, or after a certain number of iterations, the branch is eliminated.