A method for modeling and module division of elevator mechanical structure based on complex network

Abstract

The present invention discloses a method for modeling and module division of elevator mechanical structure based on complex network. Firstly, the present invention takes elevator parts as the smallest functional unit, and uses complex network modeling method to abstract parts into nodes of the network, and uses the description information of parts as the attribute of nodes, and abstracts the physical connection relationship between parts as the edge of the network, so as to construct the network model of elevator mechanical structure. Secondly, according to the network topology structure information, it is divided into communities to obtain preliminary module division; then, according to the description information of all nodes in the community, the community theme word vector is established, and then the community similarity is calculated; finally, the community is merged according to the community similarity threshold to obtain the final module division result. The present invention considers adding text description information to the node data, and further merging the communities according to the description information and using the analysis method of text data to improve the accuracy of the division result.

Description

Technical Field

[0001] The invention proposes a complex network-based elevator mechanical structure modeling and module partitioning method, which belongs to the field of elevator mechanical structure design. Background Art

[0002] With the continuous emergence of high-rise buildings, elevators, as a vertical transportation equipment, have brought great convenience to people's lives and work. The rapid development of modern industry, the continuous shortening of the replacement cycle of elevator products, and the personalization and diversification of customer needs have brought great challenges to elevator manufacturers. How companies can adapt to these needs and develop fast, efficient, reliable products that can meet the personalized and diversified needs of customers has become the top priority for their survival and development.

[0003] As an effective design method, modular design can design efficient and reliable elevator products in a fast and economical way to meet customer needs. The basic purpose of modularization is to continuously develop new products and improve product quality and reliability under extremely fierce market competition and diversified, personalized and multi-variety conditions. Modules are the basic components of modular products, and module division is the basis of modular product design. Whether the modules are scientifically and reasonably divided, and whether the level, scale and quantity of the modules can be reasonably handled are all crucial factors affecting the success of modular design. Therefore, it is very necessary to develop a functional module division software for elevator mechanical structure. Summary of the invention

[0004] The present invention proposes a method for modeling and module partitioning of elevator mechanical structure based on complex network. Different from the traditional method of taking elevator functional components as the smallest unit and single complex network community partitioning algorithm, the present invention first constructs the elevator mechanical structure network model with elevator parts as network nodes and physical connection relationships of parts as edges. Then, the network is partitioned into communities based on modularity as the evaluation standard, the similarity between communities is analyzed, and based on this, the communities are further merged to obtain the elevator functional module partitioning result.

[0005] The present invention comprises the following steps:

[0006] (1) Based on the elevator 3D model assembly drawing in Solidworks software, the elevator parts list and the physical connection relationships between the parts are extracted; description methods for standard parts and non-standard parts are established respectively, and then the description information of each part is obtained by combining the knowledge of domain experts.

[0007] (2) All mechanical parts of the elevator are abstracted as network nodes, the description information of the parts is used as the attribute of the node, and the physical connection relationship between the parts is abstracted as the edge of the network, so as to establish a complex network model of the mechanical structure; the constructed complex network model is mathematically expressed and stored using the adjacency matrix to obtain an n×n matrix M, where n is the number of nodes. If there is an edge between nodes i and j, then M[i][j]=1, otherwise it is 0.

[0008] (3) Based on the complex network model obtained in step (2), the Louvain community partitioning algorithm with modularity as the evaluation criterion is used to obtain the preliminary partitioning result of the elevator mechanical structure. The modularity is defined as follows:

[0009]

[0010] In formula (1), m represents the edge connecting the nodes; i and j represent any two nodes in the network; A ij Represents the weight of the edge between node i and node j. When the network is an undirected graph model, A ij =1; k i represents the degree of node i, k j represents the degree of node j; δ(c i , c j ) is a binary function. If node i and node j are in the same community, then δ(c i , c j )=1, otherwise, δ(c i , c j )=0.

[0011] Formula (1) can be simplified to formula (2),

[0012]

[0013] In formula (2), ∑ in represents the number of edges in community c; ∑ tot Represents the sum of the degrees of the nodes in community c.

[0014] The value of modularity is between [-1, 1]. If all nodes are divided into one community, the modularity is 1. If all nodes are in one community, the modularity is -1. When the value of modularity is between 0.3 and 0.7, it means that the community partitioning algorithm is more effective.

[0015] From formula (2), we can get the expression (3) of modularity gain ΔQ:

[0016]

[0017] In formula (3), ∑ inrepresents the number of edges in community c; ∑ tot represents the sum of the degrees of the nodes in community c; k i represents the degree of node i; k i，in Represents the sum of the number of connecting edges between node i and the nodes in community c.

[0018] The algorithm mainly consists of two stages: modularity optimization and network aggregation. The completion of the two stages completes one iteration of the algorithm. When any node movement cannot increase the modularity, the iteration terminates. The specific steps of the Louvain algorithm are as follows:

[0019] (3-1) Each node in the network is considered as an independent community. The number of communities is the same as the number of nodes.

[0020] (3-2) For each node i, try to assign the i node to the communities of all its neighboring nodes in turn, calculate the modularity gain ΔQ before and after the assignment, and record the neighboring node with the largest ΔQ. If maxΔQ>0, assign the node i to the community of the neighboring node, otherwise keep the communities of the two nodes unchanged;

[0021] (3-3) Repeat step (3-2) until the communities to which all nodes belong no longer change;

[0022] (3-4) Compress the network, compress all nodes in the same community into a supernode, convert the edge weights between nodes in the community into the ring weights of the new node, and convert the edge weights between communities into the edge weights between new nodes;

[0023] (3-5) Repeat step (3-1) until any node movement cannot increase the modularity.

[0024] (4) Based on the preliminary community division results obtained in step (3), combined with the attributes of the nodes in step (2), the part description information of all nodes contained in each community is extracted, and the community keyword extraction algorithm is used to obtain the theme word vector of each community to describe the community. The specific steps of the community keyword extraction algorithm are as follows:

[0025] (4-1) Read the part descriptions of all nodes contained in the i-th community as a document (the size of i is determined by the number of communities divided by the predecessor Louvain algorithm);

[0026] (4-2) Discretize the entire document into a list of component words using commas as separators;

[0027] (4-3) removing the numeric suffixes of the part names and standard parts (bolts, nuts, etc.) from the word list in step (4-2) to obtain a new word list;

[0028] (4-4) counting the number of occurrences of parts with the same name in the word list in step (4-3) and calculating the frequency;

[0029] (4-5) If the frequency of occurrence of the part in step (4-4) is greater than the word frequency threshold and the number of keywords currently used to describe the community is less than the word quantity requirement, then the part name is added as a keyword to the community description subject word vector;

[0030] (4-6) Repeat steps (4-1) to (4-5) until all communities have been read.

[0031] (5) Based on the topic word vectors of all communities obtained in step (4), the topic similarity between all vectors is calculated based on the Levenshtein distance, and the result is the similarity between communities. Levenshtein distance, also known as edit distance, refers to the minimum number of edits required to convert a source string into a target string. The allowed edit operations are: deletion, insertion, and replacement. Generally speaking, the smaller the edit distance, the greater the similarity between the two strings. The specific steps of the Levenshtein algorithm and similarity conversion are as follows:

[0032] (5-1) Define a two-dimensional array LD[i][j] to store the minimum number of operations required to convert the first i characters of the source string to the first j characters of the target string. Assume that the source string is S and the target string is T, then 0≤i≤len(S), 0≤j≤len(T);

[0033] (5-2) Initialize the two-dimensional array LD[i][j] according to formula (4), where 0≤i≤len(S), 0≤j≤len(T),

[0034]

[0035] (5-3) Calculate the remaining elements of the two-dimensional array LD[i][j] according to formula (5), where 0≤i≤len(S), 0≤j≤len(T),

[0036]

[0037] (5-4) Take LD[len(S)][len(T)] as the edit distance from source string S to target string T;

[0038] (5-5) According to formula (6), the conversion from the edit distance of the source string S and the target string T to the similarity is completed.

[0039]

[0040] Here, similarity represents the similarity between the source string S and the target string T.

[0041] (6) Based on the similarity obtained in step (5), the communities obtained in step (3) are further merged to obtain the result of the elevator mechanical structure functional module division. The specific steps are as follows:

[0042] (6-1) Compress all communities into supernodes (hereinafter referred to as nodes), read the data in the i-th row of the similarity matrix (i.e., the similarity information between the i-th node and other nodes), and first determine whether the node can be used as the head node of the search chain. There are two criteria for determining whether it cannot be used as the head node: ① The node has been divided into a certain module in the previous chain search; ② The similarity between the node and any previous node is greater than the similarity threshold.

[0043] (6-2) If the node can be used as the head node, the chain search algorithm is performed; otherwise, the program is executed according to the criteria for determining whether the node can be used as the head node.

[0044] (6-3) Repeat steps (6-1) to (6-2) until all nodes are divided.

[0045] The specific steps of the chain search algorithm in step (6-2) are as follows:

[0046] (6-2-1) If a node can be used as the head node of the search chain, find the node a1 with the greatest similarity and merge it into the module to which the head node belongs;

[0047] (6-2-2) Repeat step (6-2-1) with a1 as the head node;

[0048] (6-2-3) Repeat steps (6-2-1) to (6-2-2) until the traversal is completed. The end mark is that the maximum similarity is lower than the similarity threshold or the last node has been visited.

[0049] Beneficial effects of the present invention: Since it is impossible to accurately divide the functional modules of the elevator according to the actual situation by directly using the community division algorithm based on the complex network, it is considered to add text description information to the node data, and further merge the communities according to the description information and the analysis method of the text data to improve the accuracy of the division result. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is the overall flow chart of the method of the present invention;

[0051] Figure 2 It is a flowchart for obtaining community nodes;

[0052] Figure 3 It is a flowchart for obtaining the community similarity matrix;

[0053] Figure 4 It is a flow chart of community merger. DETAILED DESCRIPTION

[0054] The present invention proposes a method for modeling and module partitioning of elevator mechanical structure based on complex network, the flow chart of which is shown in FIG1 , and includes the following steps:

[0055] (1) Based on the elevator 3D model assembly drawing in Solidworks software, the elevator parts list and the physical connection relationships between the parts are extracted; description methods for standard parts and non-standard parts are established respectively, and then the description information of each part is obtained by combining the knowledge of domain experts.

[0056] (2) All mechanical parts of the elevator are abstracted as network nodes, the description information of the parts is used as the attribute of the node, and the physical connection relationship between the parts is abstracted as the edge of the network, so as to establish a complex network model of the mechanical structure; the constructed complex network model is mathematically expressed and stored using the adjacency matrix to obtain an n×n matrix M, where n is the number of nodes. If there is an edge between nodes i and j, then M[i][j]=1, otherwise it is 0.

[0057] In order to deepen the understanding of step (2), an example is given here to explain the process of building a complex network model of the elevator mechanical structure.

[0058] The network nodes are all the mechanical parts of the elevator, and each node is required to be numbered; the node attributes are the description information of the parts, which are used as the input of the subsequent algorithm and are required to have a unified description format. The part description of the non-standard parts is defined as "the location of the part + the orientation (up, down, left, right) + the part name + the serial number of the parts with the same name", such as "the left door machine mounting seat 1 of the car"; the part description of the standard parts is defined as "connected part 1, connected part 2, standard part", such as "the middle plate of the car, the left top plate of the car, nut 1"; the physical connection relationship between the parts is abstracted as the edge of the network. If the two part nodes numbered α1 and α2 have physical contact, a set of node pairs (α1, α2) is defined, and so on, all the node pairs are defined, and then a complex network model of the elevator mechanical structure is constructed.

[0059] (3) Based on the complex network model obtained in step (2), the Louvain community partitioning algorithm with modularity as the evaluation criterion is used to obtain the preliminary partitioning result of the elevator mechanical structure, see Figure 2 . The modularity is defined as follows:

[0060]

[0061] In formula (1), m represents the edge connecting the nodes; i and j represent any two nodes in the network; A ij Represents the weight of the edge between node i and node j. When the network is an undirected graph model, A ij =1; ki represents the degree of node i, k j represents the degree of node j; δ(c i , c j ) is a binary function. If node i and node j are in the same community, then δ(c i , c j )=1, otherwise, δ(c i , c j )=0.

[0062] Formula (1) can be simplified to formula (2),

[0063]

[0064] In formula (2), ∑ in represents the number of edges in community c; ∑ tot Represents the sum of the degrees of the nodes in community c.

[0065] The value of modularity is between [-1, 1]. If all nodes are divided into one community, the modularity is 1. If all nodes are in one community, the modularity is -1. When the value of modularity is between 0.3 and 0.7, it means that the community partitioning algorithm is more effective.

[0066] From formula (2), we can get the expression (3) of modularity gain ΔQ:

[0067]

[0068] In formula (3), ∑ in represents the number of edges in community c; ∑ tot represents the sum of the degrees of the nodes in community c; k i represents the degree of node i; k i，in Represents the sum of the number of connecting edges between node i and the nodes in community c.

[0069] The algorithm mainly consists of two stages: modularity optimization and network aggregation. The completion of the two stages completes one iteration of the algorithm. When any node movement cannot increase the modularity, the iteration terminates. The specific steps of the Louvain algorithm are as follows:

[0070] (3-1) Each node in the network is considered as an independent community. The number of communities is the same as the number of nodes.

[0071] (3-2) For each node i, try to assign the i node to the communities of all its neighboring nodes in turn, calculate the modularity gain ΔQ before and after the assignment, and record the neighboring node with the largest ΔQ. If maxΔQ>0, assign the node i to the community of the neighboring node, otherwise keep the communities of the two nodes unchanged;

[0072] (3-3) Repeat step (3-2) until the communities to which all nodes belong no longer change;

[0073] (3-4) Compress the network, compress all nodes in the same community into a supernode, convert the edge weights between nodes in the community into the ring weights of the new node, and convert the edge weights between communities into the edge weights between new nodes;

[0074] (3-5) Repeat step (3-1) until any node movement cannot increase the modularity.

[0075] According to the preliminary community division results obtained in step (3), combined with the attributes of the nodes in step (2), the part description information of all nodes contained in each community is extracted, and the community keyword extraction algorithm is used to obtain the theme word vector of each community to describe the community. The specific steps of the community keyword extraction algorithm are as follows:

[0076] (4-1) Read the part descriptions of all nodes contained in the i-th community as a document (the size of i is determined by the number of communities divided by the predecessor Louvain algorithm);

[0077] (4-2) Discretize the entire document into a list of component words using commas as separators;

[0078] (4-3) removing the numeric suffixes of the part names and standard parts (bolts, nuts, etc.) from the word list in step (4-2) to obtain a new word list;

[0079] (4-4) counting the number of occurrences of parts with the same name in the word list in step (4-3) and calculating the frequency;

[0080] (4-5) If the frequency of occurrence of the part in step (4-4) is greater than the word frequency threshold and the number of keywords currently used to describe the community is less than the word quantity requirement, then the part name is added as a keyword to the community description subject word vector;

[0081] (4-6) Repeat steps (4-1) to (4-5) until all communities have been read.

[0082] In order to deepen the understanding of steps (3) and (4), an example is given here to explain the process of obtaining a community keyword vector.

[0083] Assume that the nodes (node ​​number + node attribute) contained in a community divided in step (3) are as shown in Table 1:

[0084] Table 1 Nodes included in a community

[0085]

[0086] After steps (4-1) to (4-4), the number of occurrences and frequencies of each part of the community are obtained as shown in Table 2:

[0087] Table 2 The number and frequency of occurrence of each part name in a certain community

[0088]

[0089]

[0090] Assuming that the word frequency threshold is 0.05 and the maximum number of keywords is 5, the subject word vector used to describe the community is [traction machine end cover, traction machine shaft, traction machine base, traction machine pressure cover, traction machine encoder mounting plate].

[0091] (5) Based on the topic word vectors of all communities obtained in step (4), the topic similarity between all vectors is calculated based on the Levenshtein distance. The result is the similarity between communities, see Figure 3 . Levenshtein distance, also known as edit distance, refers to the minimum number of edits required to convert a source string into a target string. The allowed edit operations are: deletion, insertion, and replacement. Generally speaking, the smaller the edit distance, the greater the similarity between the two strings. The specific steps of the Levenshtein algorithm and similarity conversion are as follows:

[0092] (5-1) Define a two-dimensional array LD[i][j] to store the minimum number of operations required to convert the first i characters of the source string to the first j characters of the target string. Assume that the source string is S and the target string is T, then 0≤i≤len(S), 0≤j≤len(T);

[0093] (5-2) Initialize the two-dimensional array LD[i][j] according to formula (4), where 0≤i≤len(S), 0≤j≤len(T),

[0094]

[0095] (5-3) Calculate the remaining elements of the two-dimensional array LD[i][j] according to formula (5), where 0≤I≤len(S), 0≤j≤len(T),

[0096]

[0097] (5-4) Take LD[len(S)][len(T)] as the edit distance from source string S to target string T;

[0098] (5-5) According to formula (6), the conversion from the edit distance of the source string S and the target string T to the similarity is completed.

[0099]

[0100] Here, similarity represents the similarity between the source string S and the target string T.

[0101] (6) Based on the similarity matrix obtained in step (5), the communities obtained in step (3) are further merged to obtain the result of the elevator mechanical structure functional module division, see Figure 4 The specific steps are as follows:

[0102] (6-1) Compress all communities into supernodes (hereinafter referred to as nodes), read the data in the i-th row of the similarity matrix (i.e., the similarity information between the i-th node and other nodes), and first determine whether the node can be used as the head node of the search chain. There are two criteria for determining whether it cannot be used as the head node: ① The node has been divided into a certain module in the previous chain search; ② The similarity between the node and any previous node is greater than the similarity threshold.

[0103] (6-2) If the node can be used as the head node, the chain search algorithm is performed; otherwise, the program is executed according to the criteria for determining whether the node can be used as the head node.

[0104] (6-3) Repeat steps (6-1) to (6-2) until all nodes are divided.

[0105] The specific steps of the chain search algorithm in step (6-2) are as follows:

[0106] (6-2-1) If a node can be used as the head node of the search chain, find the node a1 with the greatest similarity and merge it into the module to which the head node belongs;

[0107] (6-2-2) Repeat step (6-2-1) with a1 as the head node;

[0108] (6-2-3) Repeat steps (6-2-1) to (6-2-2) until the traversal is completed. The end mark is that the maximum similarity is lower than the similarity threshold or the last node has been visited.

[0109] In order to deepen the understanding of steps (5) and (6), an example is given here to explain the community merging process. Assume that after step (5), the community similarity matrix shown in Table 3 is obtained:

[0110] Table 3 Community similarity matrix

[0111] Club 1 Club 2 Club 3 Club 4 Club 5 Club 6 Club 7 Club 8 Club 9 Club 1 0 0.88 0.58 0.54 0.61 0.43 0.65 0.53 0.75 Club 2 0 0 0.46 0.89 0.73 0.77 0.46 0.28 0.74 Club 3 0 0 0 0.45 0.57 0.93 0.54 0.67 0.33 Club 4 0 0 0 0 0.66 0.62 0.45 0.76 0.78 Club 5 0 0 0 0 0 0.33 0.87 0.67 0.35 Club 6 0 0 0 0 0 0 0.45 0.56 0.60 Club 7 0 0 0 0 0 0 0 0.78 0.57 Club 8 0 0 0 0 0 0 0 0 0.35 Club 9 0 0 0 0 0 0 0 0 0

[0112] For the similarity matrix shown in Table 3, 9 communities are merged. Assuming the similarity threshold is 0.69, first, community 1 is used as the search head node, and community 2 with the highest similarity is searched, and the similarity is greater than the similarity threshold 0.69, then it is merged with community 1; then it is determined that community 2 can be used as the next search head node, and community 4 is searched; and so on, until community 9 is searched and the current round of traversal is stopped (search to the last community), and the result of this round of traversal merges communities 1, 2, 4, and 9 into the first category.

[0113] The second round of traversal begins. First, it is determined that community 2 cannot be used as the search head node (community 2 has been merged). Then, it is determined that community 3 can be used as the search head node. Community 6 with the highest similarity is searched, and the similarity is greater than the similarity threshold of 0.69. Then, it is merged with community 3. Then, it is determined that community 6 cannot be used as the next search head node (the community with the highest similarity to community 6 has a similarity lower than the similarity threshold of 0.69). This round of traversal is stopped. The result of this round of traversal merges communities 3 and 6 into the second category.

[0114] The third round of traversal begins. First, it is determined that community 4 cannot be used as the search head node (community 4 has been merged). It is further determined that community 5 cannot be used as the search head node (the similarity between community 5 and community 2 is greater than the similarity threshold of 0.69, and community 5 will be merged into the category where community 2 is located). Similarly, community 6 cannot be used as the search head node (community 6 has been merged). It is further determined that community 7 can be used as the search head node. Community 8 with the highest similarity is searched, and the similarity is greater than the similarity threshold of 0.69, so it is merged with community 8. It is further determined that community 8 cannot be used as the search head node (the community with the greatest similarity to community 9 has a similarity lower than the similarity threshold of 0.69), and this round of traversal is stopped. The result of this round of traversal merges community 5 into the first category, and communities 7 and 8 into the third category.

[0115] The fourth round of traversal begins, and it is determined that community 9 has been merged, so the traversal is stopped directly.

[0116] Because community 9 is the last community, all traversals are completed. Finally, the algorithm merges the 9 communities into three categories, which are described in the form of a list: [community 1, community 2, community 4, community 5, community 9] is the first category, [community 3, community 6] is the second category, and [community 7, community 8] is the third category.

[0117] The specific implementation steps of the present invention are described in detail below in conjunction with the accompanying drawings:

[0118] The core of the present invention is to build a complex network model based on the acquired elevator mechanical parts nodes, node attributes and node connection relationships, perform a community partitioning algorithm on it and obtain the keyword vector used to describe each community, perform a similarity analysis on it and then use the community merging algorithm to obtain the various functional modules of the elevator.

[0119] 1. Elevator mechanical structure data acquisition

[0120] According to step (1), the elevator parts list and parts description information are obtained as nodes and node attributes of the complex network model respectively. The obtained results are shown in Table 4, with a total of 3024 nodes.

[0121] Table 4 Elevator parts node data acquisition

[0122]

[0123] According to the physical connection relationship between parts, the part nodes with physical connection relationship are defined as node pairs, which are used as the edges of the complex network model. After analysis, the part node pairs of the elevator mechanical system are obtained as shown in Table 5, with a total of 3900 node pairs (edges).

[0124] Table 5 Elevator mechanical system parts node pair

[0125]

[0126] 2. Build a complex elevator network model, use the Louvain algorithm to preliminarily divide the network model and obtain the keyword vector of each community

[0127] According to the input data obtained in step (1), the model parameters are initialized with 3024 nodes and 3900 edges, and the elevator complex network model is constructed. Then, according to step (3), the obtained network model is preliminarily divided into communities, and the theme word vector of each community is obtained. When obtaining the community theme word vector, the keyword frequency threshold is set to 0.05, and the number of keywords used to describe the community is limited to 5. The preliminary division of the community and the theme word vector of each community are shown in Table 6.

[0128] Table 6 Preliminary division of communities and keyword extraction

[0129]

[0130]

[0131] 3. Calculate the similarity matrix between communities and merge them

[0132] According to step (5), the preliminary division results are calculated by the Levenshtein algorithm and similarity conversion is performed to obtain a similarity matrix (32×32), as shown in Table 7.

[0133] Table 7 Community similarity matrix

[0134] Club 1 Club 2 Club 3 Club 4 … Club 31 Club 32 Club 1 0 0.85 0.72 0.68 … 0.57 0.45 Club 2 0 0 0.69 0.60 … 0.51 0.40 Club 3 0 0 0 0.71 … 0.57 0.52 Club 4 0 0 0 0 … 0.56 0.56 … … … … … … … … Club 31 0 0 0 0 … 0 0.70 Club 32 0 0 0 0 … 0 0

[0135] According to step (6), the communities are merged, where the similarity threshold is set to 0.69. The elevator function module division results are shown in Table 8, and the division results are consistent with the actual situation.

[0136] Table 8 Elevator functional module division results

[0137]

Claims

1. A complex network-based elevator mechanical structure modeling and module division method, characterized in that The method comprises the following steps: (1) Elevator mechanical structure data acquisition Based on the elevator 3D model assembly drawing in Solidworks software, extract the elevator parts list and the physical connection relationship between the parts; Establish description methods for standard parts and non-standard parts respectively, and then combine domain expert knowledge to obtain description information for each part; (2) Constructing a complex network model of the elevator mechanical structure All mechanical parts of the elevator are abstracted as network nodes, the description information of the parts is used as the attributes of the nodes, and the physical connection relationship between the parts is abstracted as the edges of the network, thereby establishing a complex network model of the mechanical structure; (3) Community division of mechanical structure complex network model According to the complex network model obtained in step (2), the Louvain community partitioning algorithm with modularity as the evaluation standard is used to obtain the preliminary partitioning result of the elevator mechanical structure; wherein the modularity is defined as follows: In formula (1), m represents the edge connecting the nodes; i and j represent any two nodes in the network; A ij Represents the weight of the edge between node i and node j. When the network is an undirected graph model, A ij =1; k i represents the degree of node i, k j represents the degree of node j; δ(c i , c j ) is a binary function. If node i and node j are in the same community, then δ(c i , c j )=1, otherwise, δ(c i , c j )=0; Formula (1) is simplified to formula (2), In formula (2), ∑ in represents the number of edges in community c; ∑ tot represents the sum of the degrees of the nodes in community c; The value of modularity is between [-1, 1]. If all nodes are divided into one community, the modularity is 1. If all nodes are divided into one community, the modularity is -1. From formula (2), we get the expression (3) of modularity gain ΔQ: In formula (3), ∑ in represents the number of edges in community c; ∑ tot represents the sum of the degrees of the nodes in community c; k i represents the degree of node i; k i，in Represents the sum of the number of connecting edges between node i and the nodes in community c; The Louvain community partitioning algorithm mainly includes two stages: modularity optimization and network aggregation. The completion of the two stages completes one iteration of the algorithm. When any node movement cannot increase the modularity, the iteration terminates. The specific steps of the Louvain community partitioning algorithm are as follows: (3-1) Each node in the network is considered as an independent community. The number of communities is the same as the number of nodes. (3-2) For each node i, try to assign the i node to the communities of all its neighboring nodes in turn, calculate the modularity gain ΔQ before and after the assignment, and record the neighboring node with the largest ΔQ. If maxΔQ>0, assign the node i to the community of the neighboring node, otherwise keep the communities of the two nodes unchanged; (3-3) Repeat step (3-2) until the communities to which all nodes belong no longer change; (3-4) Compress the network, compress all nodes in the same community into a supernode, convert the edge weights between nodes in the community into the ring weights of the new node, and convert the edge weights between communities into the edge weights between new nodes; (3-5) Repeat step (3-1) until any node movement cannot increase the modularity; (4) Community keyword vector extraction Based on the preliminary community division results obtained in step (3), combined with the attributes of the nodes in step (2), the part description information of all nodes contained in each community is extracted, and the community keyword extraction algorithm is used to obtain the theme word vector of each community to describe the community; (5) Similarity calculation Based on the topic word vectors of all communities obtained in step (4), the topic similarity between all vectors is calculated based on the Levevshtein distance, and the result is the similarity between communities; (6) Merger of associations Based on the similarity obtained in step (5), the communities obtained in step (3) are further merged to obtain the result of the elevator mechanical structure functional module division; the specific steps are as follows: (6-1) All communities are compressed into super nodes, hereinafter referred to as nodes, and the data of the i-th row of the similarity matrix is ​​read to first determine whether the node can be used as the head node of the search chain; (6-2) If the node can be used as the head node, the chain search algorithm is performed; otherwise, the program is executed according to the judgment criteria for not being able to be the head node; (6-3) Repeat steps (6-1) to (6-2) until all nodes are divided.

2. The elevator mechanical structure modeling and module division method based on complex network according to claim 1, characterized in that: The constructed complex network model is mathematically expressed and stored using the adjacency matrix, resulting in an n×n matrix M, where n is the number of nodes. If there is an edge between nodes i and j, M[i][j]=1, otherwise it is 0.

3. The elevator mechanical structure modeling and module division method based on complex network according to claim 1 is characterized by: The specific steps of the community keyword extraction algorithm are as follows: (4-1) Read the part descriptions of all nodes contained in the i-th community as a document; (4-2) Discretize the entire document into a list of component words using commas as separators; (4-3) removing the numeric suffixes of the part names and standard parts in the word list in step (4-2) to obtain a new word list; (4-4) counting the number of occurrences of parts with the same name in the word list in step (4-3) and calculating the frequency; (4-5) If the frequency of occurrence of the part in step (4-4) is greater than the word frequency threshold and the number of keywords currently used to describe the community is less than the word quantity requirement, then the part name is added as a keyword to the community description subject word vector; (4-6) Repeat steps (4-1) to (4-5) until all communities have been read.

4. The elevator mechanical structure modeling and module division method based on complex network according to claim 1 is characterized by: The similarity calculation process is: (5-1) Define a two-dimensional array LD[i][j] to store the minimum number of operations required to convert the first i characters of the source string to the first j characters of the target string. Assume that the source string is S and the target string is T, then 0≤i≤len(S), 0≤j≤len(T); (5-2) Initialize the two-dimensional array LD[i][j] according to formula (4), where 0≤i≤len(S), 0≤j≤len(T), (5-3) Calculate the remaining elements of the two-dimensional array LD[i][j] according to formula (5), where 0≤i≤len(S), 0≤j≤len(T), (5-4) Take LD[len(S)][len(T)] as the edit distance from source string S to target string T; (5-5) According to formula (6), the conversion of the edit distance of the source string S and the target string T to similarity is completed; Here, similarity represents the similarity between the source string S and the target string T.

5. The elevator mechanical structure modeling and module division method based on complex network according to claim 1 is characterized by: There are two criteria for determining whether a node cannot be used as a head node: ① the node has been divided into a module in the previous chain search; ② the similarity between the node and any previous node is greater than the similarity threshold.

6. The elevator mechanical structure modeling and module division method based on complex network according to claim 1 is characterized by: The specific steps of the chain search algorithm described in step (6-2) are as follows: (6-2-1) If a node can be used as the head node of the search chain, find the node a1 with the greatest similarity to it and merge it into the module to which the head node belongs; (6-2-2) Repeat step (6-2-1) with a1 as the head node; (6-2-3) Repeat steps (6-2-1) to (6-2-2) until the traversal is completed. The end mark is that the maximum similarity is lower than the similarity threshold or the last node has been visited.

Images