A production line configuration structure diagram recognition matching method

Abstract

The application discloses a kind of identification matching methods of line configuration structure diagram, comprising: when the selection decision of enterprise in line equipment supplier, the individualized line that enterprise wants to build and corresponding function are presented by the mode of line configuration structure diagram (i.e. query subgraph) to enterprise.The connected various line equipment supplier nodes in graph need to have cooperative relationship for the reason of production transportation convenience, so a line configuration structure diagram with clear function and clear supply and demand relationship can be built.Then the line configuration structure diagram is matched with the existing line equipment supplier database of manufacturing enterprise community, so as to obtain multiple possible result matching subgraph, for enterprise to make next accurate decision, so most of decision interference factors can be excluded, and decision efficiency is improved.

Description

Technical Field

[0001] This invention relates to the technical field of supply and demand matching, and in particular to a method for identifying and matching production line configuration structure diagrams. Background Technology

[0002] When customers are looking for a production line that can meet their personalized needs, the description of their functional requirements is often vague or has multiple possibilities in the early stages. This results in a large number of historical case studies of production line designs to choose from. The challenge is how to select a series of production line equipment suppliers that meet the required functions (and require these suppliers to have cooperative relationships) from this massive amount of data, and how to filter out most of the suppliers that do not meet the requirements.

[0003] Currently, most manufacturing companies still follow the traditional supply-demand matching approach when searching for production line equipment suppliers. This involves finding individual suppliers for each piece of equipment based on the functional requirements of the production line, and then combining all the suppliers to build their customized production line. While this traditional approach is acceptable for small businesses with limited production line needs and simple functional structures, it significantly increases the decision-making workload for large enterprises with substantial production line needs and complex functional structures. This leads to a marked decrease in production line construction efficiency and an increase in R&D costs. Furthermore, this discrete supply-demand matching method can result in inadequate overall considerations for the production line and a strong reliance on the experience of designers. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying and matching production line configuration structure diagrams to solve the problem of manufacturing enterprises finding production line equipment suppliers for building customized production lines.

[0005] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0006] A method for identifying and matching production line configuration structure diagrams, comprising:

[0007] The query requirements submitted by the demand side are transformed into a query subgraph, which represents the production line that the enterprise needs to build.

[0008] Represent the query subgraph as a factor graph;

[0009] Construct a matching model between enterprise production lines and production line equipment suppliers;

[0010] The matching model is solved using factor graphs and belief propagation algorithms, thereby finding matching results, i.e., matching subgraphs, from a data network containing a database of the company's existing production line equipment suppliers.

[0011] Transform the matching subgraph into query results described in text.

[0012] Furthermore, the query subgraph is represented as a factor graph, including:

[0013] The query subgraph is represented as Q = (V Q E Q ,vt Q ,et Q The data network is represented as D = (V D E D ,vt D ,et D A match represents a set of mappings from the query subgraph to the nodes and edges of the data network, and this mapping is defined as a vector X = [X...]. m ], where each query node m∈V Q Each has a mapping node X in the data network D. m ∈X D Each query edge e = (m, n) ∈ E Q All are mapped to edge X in the data network m X n )∈E D ;

[0014] Among them, V Q V represents the set of nodes in subgraph Q. D E represents the set of nodes in data network D; Q E represents the set of edges in subgraph Q. D Represents the set of edges in the data network D; vt Q This represents the set of attribute vectors of nodes in the query subgraph Q, vt D et represents the set of attribute vectors of nodes in data network D; Q Let et represent the set of attribute vectors of edges in the query subgraph Q. D X represents the set of attribute vectors of edges in the data network D; D E represents the set of all mapped nodes in data network D; D Let D represent the set of all mapped edges in the data network D;

[0015] The query subgraph is represented as a factor graph; a factor graph contains two types of nodes: one type represents the edge e = (m,n) ∈ E in the query subgraph. Q Factor nodes, a type of node m∈V representing a query subgraph. Q Variable nodes.

[0016] Furthermore, a matching model is constructed between the enterprise's production line and its equipment suppliers, including:

[0017] By maximizing the marginal probability bm (i)=P(X m =i|D) to find the mapping; the LBP algorithm can find marginal probabilities to maximize the joint probability distribution of the mapping:

[0018]

[0019] In the formula:

[0020] Let X represent the mapping vector X when the joint probability distribution is maximized; X represents the mapping vector, which is the subset of nodes in the data network D corresponding to node m in the query subgraph Q; P(X|D) represents the joint probability distribution of the mapping conditioned on the data network D; D represents the data network.

[0021] The joint probability distribution P(X|D), i.e., the matching objective function, is expressed as:

[0022]

[0023] In the formula:

[0024] Represents the standardization factor; X = [X m [m∈Q] represents the mapping vector, that is, the subset of nodes in the data network D corresponding to node m in the query subgraph Q; X m x represents an element in the mapping vector X; m Represents node X m The variable's possible values; x n In data network D, x represents the relationship between x and x. m Adjacent node X n The variable's possible values; Functions representing the similarity relationships between individuals; Functions that represent bidirectional dependencies between nodes;

[0025] Due to the structural differences in the mapping variables, the objective function cannot be directly applied to message passing in the LBP algorithm. Therefore, the discrete mapping vector X is replaced with a 0-1 mapping matrix T = [T mi If X m =i, then T mi =1; otherwise, T mi =0;

[0026] The matching objective function is further transformed to obtain:

[0027]

[0028] In the formula:

[0029] Represents the standardization factor; vt Q(m) represents the attribute vector of node m in the query subgraph Q; vt D (i) represents the attribute vector of node i in data network D; et Q (e mn ) indicates querying edge e in subgraph Q mn Attribute vector; et D (e ij ) represents edge e in data network D ij Attribute vector; p(vt) D (i)|vt Q (m)) represents the conditional probability distribution of data node i given query node m; p(et D (e ij )|et Q (e mn )) indicates that in the query edge e mn Under the condition of mapping edge e ij The conditional probability distribution of occurrence; T represents the mapping matrix; T mi / T nj represents an element in the mapping matrix T; m and n represent any node in the query subgraph Q; i and j represent any node in the data network D;

[0030] In data network D, the attributes of edges represent semantic relationships, temporal and spatial dependencies, and interaction and influence relationships, while the attributes of nodes represent metadata, state, and semantic information.

[0031] To simplify computation by transforming the product operation in the logarithmic field into a sum operation, we take the logarithm of the matching objective function P(T|D,Q) and transform it into O(T):

[0032]

[0033] M mi =log[p(vt) D (i)|vt Q (m))] (1-5)

[0034] M mnij =log[p(et)] D (e ij )|et Q (e mn (1-6)

[0035] In the formula:

[0036] O(T) represents the matching degree function; M mi M represents the semantic similarity function between query node m and data node i; mnij Indicates query edge e mn With data edge eij Semantic similarity function between; e mn This indicates a query for any edge in subgraph Q; e ij Let represent any edge in the data network D.

[0037] Furthermore, the matching model is solved using factor graphs and belief propagation algorithms to find matching results from a data network containing the company's existing production line equipment supplier database. This includes finding the best match by passing belief messages of the mapping probabilities of query nodes / edges and data nodes / edges between factor nodes and variable nodes, ultimately obtaining the marginal probability distribution b of the query node. m (i) is used to generate the mapping vector X.

[0038] Furthermore, the optimal match is found by passing belief messages of the mapping probabilities of query nodes / edges and data nodes / edges between factor nodes and variable nodes, ultimately yielding the marginal probability distribution b of the query node. m (i) Used to generate the mapping vector X, including:

[0039] A1. Initialize the factor graph. Construct an initial potential function, i.e., an individual matching similarity relationship function, based on the attribute features in the factor graph. bidirectional dependency function between nodes Initialize the belief values ​​of all nodes;

[0040] A2. Perform the first to nth iterations, i.e. message passing loop, traversing all factor nodes and variable nodes, passing messages from factor node m to variable node i, and passing messages from variable node i to all adjacent factor nodes.

[0041] The belief propagation algorithm updates the belief values ​​of each node by maximizing the summation operation, and its propagation and update involve two directions:

[0042] (1) Passing messages from variable node m to factor node i;

[0043] (2) Pass messages from factor node i to its connected variable nodes;

[0044] The message passing in the simplified confidence propagation algorithm consists of two steps:

[0045] The first step is to update the two messages passed from each factor node to the connected variable node:

[0046] f mn (j)∝max i (M mnij +μ m (i)-r mn (i)) (1-7)

[0047] f mn (i)∝max j (M mnij +μ n (j)-r mn (j)) (1-8)

[0048] In the formula:

[0049] f mn (j) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Ends at node j; r mn (i) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Start at node i; r mn (j) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Start at node j; f mn (i) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Ends at node i; μ m (i) represents the marginal log probability that query node m maps to data node i; μ n (j) represents the marginal log probability that query node n maps to data node j; M mnij Indicates query edge e mn With data edge e ij Semantic similarity function between them;

[0050] If a query edge corresponds to an empty data edge, i.e., no mapping, then M will be... mnij The value is replaced with a predetermined M. null value;

[0051] The second step is to use the messages received by the variable nodes to update the belief values ​​of the factor nodes:

[0052]

[0053] In the formula:

[0054] μ m (i) represents the marginal logarithmic probability of query node m mapping to data node i; f lm (i) indicates that query edge e lm The marginal log probability is mapped to the data edge, where the data edge terminates at node i; r lm(i) indicates that query edge e lm The marginal log probability mapped to the data edge, where the data edge starts at node i; E Q The query represents the set of edges in the subgraph; l represents a query for a node in subgraph Q that is adjacent to node m; e lm This indicates a query for an edge in subgraph Q that is connected to nodes m and l;

[0055] A3. In each iteration, the belief value of all nodes is evaluated, and the evaluation condition is: for all nodes, does |v| exist? (t) -v (t-1) |<δ con In other words, whether the difference between the belief values ​​of the two most recent iterations is less than a set threshold, thus approximating no change; if so, it indicates that the belief values ​​of each node tend to stabilize, then... Obtain the marginal probability distribution and calculate the maximum marginal probability argmax b. m (i) can then obtain the mapping node X in the data network corresponding to the query node m. m If the condition is not met, continue iterating until it is met.

[0056] Among them, v (t) δ represents the belief value in each iteration. con Indicates the convergence threshold; This indicates that the belief value μ of the query node m will be used. m (i) The marginal probability distribution of the query node m mapped to the data node i obtained by re-taking the exponent;

[0057] By doing so, the mapping nodes of all query nodes in the query subgraph can be obtained, thus forming the matching subgraph.

[0058] Furthermore, when the confidence propagation algorithm begins message passing, we have:

[0059]

[0060] Furthermore, to further improve the algorithm's performance and capture larger message changes in each iteration, the original message update method is changed to an incremental method:

[0061] v = v (t-1) +α(v (t) -v (t-1) (1-11)

[0062] In the formula:

[0063] v (t) This represents the belief value, or μ. m (i), f mn (i) and r mn(i); α represents the convergence speed control coefficient to avoid the probability of getting trapped in a local optimum; v (t-1) This represents the belief value from the previous iteration.

[0064] Compared with existing technologies, the principles and advantages of this solution are as follows:

[0065] In this solution, when enterprises make decisions regarding production line equipment suppliers, they present their desired customized production line and corresponding functions through a production line configuration structure diagram (i.e., a query subgraph). For ease of production and transportation, the connected production line equipment supplier nodes in the diagram must have cooperative relationships. This creates a production line configuration structure diagram with clear functions and well-defined supply and demand relationships. This diagram is then matched against the existing production line equipment supplier database within the manufacturing enterprise community, resulting in multiple possible matching subgraphs for the enterprise to use for further precise decision-making. This process eliminates most decision-making interference factors and improves decision-making efficiency. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the services required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a flowchart illustrating the principle of a production line configuration structure diagram identification and matching method according to the present invention.

[0068] Figure 2 This is a schematic diagram of network subgraph matching;

[0069] Figure 3 To query subgraphs and their corresponding factor graphs;

[0070] Figure 4 A diagram illustrating the elements and their relationships in the target function;

[0071] Figure 5 This is a schematic diagram of message propagation in a factor graph. Detailed Implementation

[0072] To address the challenge of manufacturing enterprises finding equipment suppliers for building customized production lines, this invention proposes a method for identifying and matching production line configuration structure diagrams as an auxiliary decision-making tool for equipment selection. This method first identifies a rough solution space from a large-scale network of production line equipment suppliers, and then considers the evaluation and coordination of specific candidate suppliers. In other words, given the massive amount of production line equipment supplier network data, this new approach to "demand-capability" matching first performs fuzzy supply-demand matching at the network level, rather than performing precise supply-demand matching at the individual node level in the initial network. This not only reduces decision-making time and improves efficiency but also reduces or even eliminates the influence of subjective human decision-making through computer-aided analysis, resulting in more objective and valuable algorithm-matched results.

[0073] This invention utilizes an existing database of production line equipment suppliers within a manufacturing enterprise community. It employs computer and machine learning algorithms as decision-making tools to help manufacturing enterprises find suppliers for their customized production lines. The strategy involves presenting the desired customized production line and its corresponding functions through a production line configuration structure diagram. This diagram is then matched against the existing database of production line equipment suppliers within the manufacturing enterprise community, generating multiple possible matching sub-diagrams for the enterprise to use for further precise decision-making. This reduces the decision-making workload of personnel and improves decision-making efficiency, especially for large enterprises with significant decision-making needs, undoubtedly reducing R&D costs considerably. Furthermore, this method largely eliminates subjective interference from decision-makers; the matching extracted by the computer is more objective and exhibits better stability.

[0074] The present invention will be further described below with reference to specific embodiments:

[0075] like Figure 1 As shown in this embodiment, a method for identifying and matching a production line configuration structure diagram includes the following steps:

[0076] S1. When a company is looking for production line equipment suppliers to build a customized production line, it needs to transform its query target into a query subgraph. This query subgraph can represent the production line functions that a group of companies can build, and then match them in a data network (the existing database of production line equipment suppliers in the manufacturing enterprise community). For example... Figure 2 This illustrates a fuzzy supply-demand matching process at a network level, where the demand side needs to find a combination of an assembly line and a component production line. For example... Figure 2As shown in the left half, the demand side's query target is: a company that can process mobile phones.<mobile phone> Assembly line<assembly line> And its production status<production state> In a state of surplus<surplus state> And a chip supplier that can supply chips to assembly lines. <chip>Parts production line<parts production line> . Figure 2 The data network shown on the right is a database of existing production line equipment suppliers within the manufacturing enterprise community.

[0077] S2. Represent the query subgraph as a factor graph:

[0078] The query subgraph is represented as Q = (V Q E Q ,vt Q ,et Q The data network is represented as D = (V D E D ,vt D ,et D A match represents a set of mappings from the query subgraph to the nodes and edges of the data network, and this mapping is defined as a vector X = [X...]. m ], where each query node m∈V Q Each has a mapping node X in the data network D. m ∈X D Each query edge e = (m, n) ∈ E Q All are mapped to edges (X) in the data network. m ,X n )∈E D ;

[0079] Among them, V Q V represents the set of nodes in subgraph Q. D E represents the set of nodes in data network D; Q E represents the set of edges in subgraph Q. D Represents the set of edges in the data network D; vt Q This represents the set of attribute vectors of nodes in the query subgraph Q, vt D et represents the set of attribute vectors of nodes in data network D; Q Let et represent the set of attribute vectors of edges in the query subgraph Q. D X represents the set of attribute vectors of edges in the data network D; D E represents the set of all mapped nodes in data network D; D Let D represent the set of all mapped edges in the data network D;

[0080] Represent the query subgraph as a factor graph; for example... Figure 3 As shown, a factor graph contains two types of nodes: one type represents the edge e = (m,n) ∈ E in the query subgraph. Q Factor nodes, a type of node m∈V representing a query subgraph. Q Variable nodes.

[0081] S3. Construct a matching model between the enterprise's production line and its equipment suppliers, specifically including:

[0082] By maximizing the marginal probability b m (i)=P(X m =i|D) to find the mapping; the LBP algorithm can find marginal probabilities to maximize the joint probability distribution of the mapping:

[0083]

[0084] In the formula:

[0085] Let X represent the mapping vector X when the joint probability distribution is maximized; X represents the mapping vector, which is the subset of nodes in the data network D corresponding to node m in the query subgraph Q; P(X|D) represents the joint probability distribution of the mapping conditioned on the data network D; D represents the data network.

[0086] The joint probability distribution P(X|D), i.e., the matching objective function, is expressed as:

[0087]

[0088] In the formula:

[0089] Represents the standardization factor; X = [X m [m∈Q] represents the mapping vector, that is, the subset of nodes in the data network D corresponding to node m in the query subgraph Q; X m x represents an element in the mapping vector X; m Represents node X m The variable's possible values; x n In data network D, x represents the relationship between x and x. m Adjacent node X n The variable's possible values; Functions representing the similarity relationships between individuals; Functions that represent bidirectional dependencies between nodes;

[0090] Due to the structural differences in the mapping variables, the objective function cannot be directly applied to message passing in the LBP algorithm. Therefore, as follows: Figure 4 As shown, the discrete mapping vector X is replaced with a 0-1 mapping matrix T = [T mi If X m =i, then T mi =1; otherwise, T mi =0;

[0091] The matching objective function is further transformed to obtain:

[0092]

[0093] In the formula:

[0094] Represents the standardization factor; vt Q (m) represents the attribute vector of node m in the query subgraph Q; vt D (i) represents the attribute vector of node i in data network D; et Q (e mn ) indicates querying edge e in subgraph Q mn Attribute vector; et D (e ij ) represents edge e in data network D ij Attribute vector; p(vt) D (i)|vt Q (m)) represents the conditional probability distribution of data node i given query node m; p(et D (e ij )|et Q (e mn )) indicates that in the query edge e mn Under the condition of mapping edge e ij The conditional probability distribution of occurrence; T represents the mapping matrix; T mi / T nj represents an element in the mapping matrix T; m and n represent any node in the query subgraph Q; i and j represent any node in the data network D;

[0095] The probability calculation in this invention is based on the similarity between query nodes (edges) and data nodes (edges). In the data network D, the attributes of edges can represent semantic relationships, temporal and spatial dependencies, interactions, and influence relationships, while the attributes of nodes can represent metadata, state, and semantic information. Therefore, these semantic similarities are used to initialize the matching probability of nodes and edges.

[0096] To simplify computation by transforming the product operation in the logarithmic field into a sum operation, we take the logarithm of the matching objective function P(T|D,Q) and transform it into O(T):

[0097]

[0098] M mi =log[p(vt) D (i)|vt Q (m))] (1-5)

[0099] M mnij =log[p(et)] D (e ij )|et Q (e mn (1-6)

[0100] In the formula:

[0101] O(T) represents the matching degree function; M mi M represents the semantic similarity function between query node m and data node i; mnij Indicates query edge e mn With data edge e ij Semantic similarity function between; e mn This indicates a query for any edge in subgraph Q; e ij Let represent any edge in the data network D.

[0102] O(T) defines the matching degree between the query subgraph and the nodes and edges of the corresponding data network. The larger the O(T) value, the more accurate the nodes and edges of the obtained data network. It can be seen that the matching objective function has been transformed from maximizing the joint probability distribution P(X|D) to maximizing the matching degree function O(T), but their ultimate goal is the same, that is, to find the result that best matches the query subgraph in the data network.

[0103] Ultimately, the mapping matrix with the maximum O(T) value can be used as the matching result between the query subgraph and the data network; that is, the mapping matrix T with the maximum O(T) value represents the required matching subgraph. Furthermore, the mapping matrix T can also be converted back into the form of the mapping vector X.

[0104] S4. Use factor graphs and belief propagation algorithms to solve the matching model, thereby finding matching results from the data network containing the company's existing production line equipment supplier database, i.e., the matching subgraph;

[0105] In this step, matching results are found from a data network containing the company's existing production line equipment supplier database. This includes finding the best match by passing belief messages about the mapping probabilities of query nodes / edges and data nodes / edges between factor nodes and variable nodes, ultimately obtaining the marginal probability distribution b of the query node. m (i) The process for generating the mapping vector X is as follows:

[0106] A1. Initialize the factor graph. Construct an initial potential function, i.e., an individual matching similarity relationship function, based on the attribute features in the factor graph. bidirectional dependency function between nodes Initialize the belief values ​​of all nodes;

[0107] A2. Perform the first to nth iterations, i.e. message passing loop, traversing all factor nodes and variable nodes, passing messages from factor node m to variable node i, and passing messages from variable node i to all adjacent factor nodes.

[0108] The belief propagation algorithm updates the belief values ​​of each node by maximizing the summation operation. Its propagation and updating involve two directions, such as... Figure 5 As shown:

[0109] (1) Passing messages from variable node m to factor node i;

[0110] (2) Pass messages from factor node i to its connected variable nodes;

[0111] The message passing in the simplified confidence propagation algorithm consists of two steps:

[0112] The first step is to update the two messages passed from each factor node to the connected variable node:

[0113] f mn (j)∝max i (M mnij +μ m (i)-r mn (i)) (1-7)

[0114] f mn (i)∝max j (M mnij +μ n (j)-r mn (j)) (1-8)

[0115] In the formula:

[0116] f mn (j) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Ends at node j; r mn (i) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Start at node i; r mn (j) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Start at node j; f mn (i) indicates that query edge e mn Mapped to data edge e ij Marginal log probability, data edge e ij Ends at node i; μ m (i) represents the marginal log probability that query node m maps to data node i; μ n (j) represents the marginal log probability that query node n maps to data node j; M mnij Indicates query edge e mn With data edge e ij Semantic similarity function between them;

[0117] If a query edge corresponds to an empty data edge, i.e., no mapping, then M will be... mnij The value is replaced with a predetermined M. null value;

[0118] The second step, as Figure 5 The lower half shows how the message received by the variable node is used to update the belief value of the factor node:

[0119]

[0120] In the formula:

[0121] μ m (i) represents the marginal logarithmic probability of query node m mapping to data node i; f lm (i) indicates that query edge e lm The marginal log probability is mapped to the data edge, where the data edge terminates at node i; r lm (i) indicates that query edge e lm The marginal log probability mapped to the data edge, where the data edge starts at node i; E Q The query represents the set of edges in the subgraph; l represents a query for a node in subgraph Q that is adjacent to node m; e lm This indicates a query for an edge in subgraph Q that is connected to nodes m and l;

[0122] It is important to note that when the confidence propagation algorithm begins message passing, the messages need to be normalized to ensure that the sum of the corresponding probabilities is 1.

[0123]

[0124] To further improve the algorithm's performance and capture larger message changes in each iteration, the original message update method was changed to an incremental method:

[0125] v = v (t-1) +α(v (t) -v (t-1) (1-11)

[0126] In the formula:

[0127] v (t) This represents the belief value, or μ. m (i), f mn (i) and r mn (i); α represents the convergence speed control coefficient to avoid the probability of getting trapped in a local optimum; v (t-1) This represents the belief value from the previous iteration.

[0128] A3. In each iteration, the belief value of all nodes is evaluated, and the evaluation condition is: for all nodes, does |v| exist? (t) -v (t-1) |<δ con In other words, whether the difference between the belief values ​​of the two most recent iterations is less than a set threshold, thus approximating no change; if so, it indicates that the belief values ​​of each node tend to stabilize, then... Obtain the marginal probability distribution and calculate the maximum marginal probability argmax b. m (i) can then obtain the mapping node X in the data network corresponding to the query node m. m If the condition is not met, continue iterating until it is met.

[0129] Among them, v (t) δ represents the belief value in each iteration. con Indicates the convergence threshold; This indicates that the belief value μ of the query node m will be used. m (i) The marginal probability distribution of the query node m mapped to the data node i obtained by re-taking the exponent;

[0130] By doing so, the mapping nodes of all query nodes in the query subgraph can be obtained, thus forming the matching subgraph.

[0131] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.< / chip>

Claims

1. A method for identifying and matching production line configuration structure diagrams, characterized in that, include: The query requirements submitted by the demand side are transformed into a query subgraph, which represents the production line that the enterprise needs to build. Represent the query subgraph as a factor graph; Construct a matching model between enterprise production lines and production line equipment suppliers; The matching model is solved using factor graphs and belief propagation algorithms, thereby finding matching results, i.e., matching subgraphs, from a data network containing a database of the company's existing production line equipment suppliers. Transform the matching subgraph into query results described in text; Construct a matching model between enterprise production lines and production line equipment suppliers, including: By maximizing marginal probabilities To find the mapping; the LBP algorithm can discover marginal probabilities to maximize the joint probability distribution of the mapping: (1-1) In the formula: The mapping vector representing the maximization of the joint probability distribution. ; Represents a mapping vector, i.e., a data network. In and query subgraph Middle node The corresponding subset of nodes; Indicates data network The joint probability distribution of the mappings given the condition; Represents a data network; Among them, the joint probability distribution The matching objective function is expressed as: (1-2) In the formula: Indicates the standardization factor; Represents a mapping vector, i.e., a data network. In and query subgraph Middle node The corresponding subset of nodes; Represents a mapping vector One of the elements; Represents a node The variable's possible values; Represents data network Zhongyu Adjacent nodes The variable's possible values; Functions representing the similarity relationships between individuals; Functions that represent bidirectional dependencies between nodes; Due to the structural differences in the mapping variables, the matching objective function (1-2) cannot be directly applied to message passing in the LBP algorithm. Therefore, the discrete mapping vector is used... Replace with a 0-1 mapping matrix ;if ,but ;otherwise, ; The target function is transformed to obtain: ， In the formula: Indicates the standardization factor; Represents a query subgraph Middle node Attribute vectors; Represents data network Nodes in Attribute vectors; Represents a query subgraph Middle Attribute vectors; Represents data network Middle Attribute vectors; This indicates a data node under the condition of a query node. The conditional probability distribution of occurrence; Indicates the query edge Mapping edges under the condition The conditional probability distribution of occurrence; Represents the mapping matrix; Represents the mapping matrix One of the elements; Represents a query subgraph Any node in the list; Represents data network Any node in the list; Data Network The attributes of the middle edge represent semantic relationships, temporal and spatial dependencies, and interaction and influence relationships, while the attributes of the node represent metadata, state and semantic information; To simplify computation by transforming the product operation in the logarithmic field of equation (1-3) into a sum operation, the matching objective function is... Take the logarithm and transform it into : (1-4) (1-5) (1-6) In the formula: Represents the matching degree function; Indicates the query node With data nodes Semantic similarity function between them; Represents the query edge With data edge Semantic similarity function between them; Represents a query subgraph Any edge in the array; Represents data network Any edge in the array.

2. The method for identifying and matching a production line configuration structure diagram according to claim 1, characterized in that, Representing the query subgraph as a factor graph includes: Represent the query subgraph as Representing the data network as = A match represents a set of mappings from the query subgraph to the nodes and edges of the data network, and this mapping is defined as a vector. Each query node All in the data network There is a mapping node Each query edge All are mapped to edges in the data network ; in, Represents a query subgraph The set of nodes in Represents data network The set of nodes in; Represents a query subgraph The set of edges in, Represents data network The set of edges in; Represents a query subgraph The set of attribute vectors of the middle node. Represents data network The set of attribute vectors for the nodes; Represents a query subgraph The set of attribute vectors in the middle edge, Represents data network The set of attribute vectors of the middle edge; Represents data network The set of all mapped nodes in the dataset; Represents data network The set of all mapped edges in the array; Represent the query subgraph as a factor graph; a factor graph contains two types of nodes: one representing edges in the query subgraph. Factor nodes, a type of node representing a query subgraph. Variable nodes.

3. The method for identifying and matching a production line configuration structure diagram according to claim 1, characterized in that, The matching model is solved using factor graphs and belief propagation algorithms to find matching results from a data network containing a database of existing production line equipment suppliers. This includes finding the best match by passing belief messages about the mapping probabilities of query nodes / edges and data nodes / edges between factor nodes and variable nodes, ultimately obtaining the marginal probability distribution of the query nodes. Used to generate mapping vectors .

4. The method for identifying and matching a production line configuration structure diagram according to claim 3, characterized in that, The optimal match is found by passing belief messages about the mapping probabilities of query nodes / edges and data nodes / edges between factor nodes and variable nodes, ultimately yielding the marginal probability distribution of the query node. Used to generate mapping vectors ,include: A1. Initialize the factor graph. Construct an initial potential function, i.e., an individual matching similarity relationship function, based on the attribute features in the factor graph. bidirectional dependency function between nodes Initialize the belief values ​​of all nodes; A2. Perform the first to nth iterations, i.e., the message passing loop, traversing all factor nodes and variable nodes, starting from the factor node. To variable node Passing messages from variable nodes Send messages to all adjacent factor nodes; The belief propagation algorithm updates the belief values ​​of each node by maximizing the summation operation, and its propagation and update involve two directions: (1) By variable node To factor nodes Passing messages; (2) By factor nodes Pass messages to the variable nodes connected to it; The message passing in the simplified confidence propagation algorithm consists of two steps: The first step is to update the two messages passed from each factor node to the connected variable node: (1-7) (1-8) In the formula: Represents the query edge Mapping to data edges Marginal log probability, data edge End at node ; Represents the query edge Mapping to data edges Marginal log probability, data edge Start at node ; Represents the query edge Mapping to data edges Marginal log probability, data edge Start at node ; Represents the query edge Mapping to data edges Marginal log probability, data edge End at node ; Indicates the query node Mapped to data nodes Marginal logarithmic probability; Indicates the query node Mapped to data nodes Marginal logarithmic probability; Represents the query edge With data edge Semantic similarity function between them; If a query edge corresponds to an empty data edge, i.e., no mapping, then... The value is replaced with a predetermined value. value; The second step is to use the messages received by the variable nodes to update the belief values ​​of the factor nodes: (1-9) In the formula: Indicates the query node Mapped to data nodes Marginal logarithmic probability; Represents the query edge The edge log probability is mapped to the data edge, and the data edge terminates at a node. ; Represents the query edge The edge log probability is mapped to the data edge, where the data edge starts at node 1. ; This represents the set of edges in the query subgraph; Represents a query subgraph In and nodes An adjacent node; Represents a query subgraph In and nodes A connected edge; A3. In each iteration, the belief value of all nodes is evaluated. The evaluation condition is: for all nodes, is there a belief value? This involves determining whether the difference between the belief values ​​of the two most recent iterations is less than a set threshold, thus approximating no change. If so, it indicates that the belief values ​​of each node tend to stabilize, and then... = Obtain the marginal probability distribution and calculate the maximum marginal probability. This will give you the query node. Mapping nodes in the corresponding data network If the condition is not met, continue iterating until it is met. in, This represents the belief value in each iteration; Indicates the convergence threshold; This indicates that the query node will be used. Belief value The query node obtained by retrieving the index Mapped to data nodes The marginal probability distribution; By doing so, the mapping nodes of all query nodes in the query subgraph can be obtained, thus forming the matching subgraph.

5. The method for identifying and matching a production line configuration structure diagram according to claim 4, characterized in that, When the confidence propagation algorithm begins message passing, we have: (1-10)。 6. The method for identifying and matching a production line configuration structure diagram according to claim 4, characterized in that, To further improve the algorithm's performance and capture larger message changes in each iteration, the original message update method was changed to an incremental method: (1-11) In the formula: Represents belief value, i.e. 、 and ; This represents the convergence speed control coefficient to avoid the probability of getting trapped in a local optimum. This represents the belief value from the previous iteration.

Images