An intelligent solution method for configuration scheme based on product configuration diagram

Abstract

The present invention discloses an intelligent solution method for configuration schemes based on product configuration diagrams. The method comprises the following steps: obtaining historical customer order data and corresponding bills of materials, establishing a set of all configuration modules, a set of configuration structures, and a set of configuration results; constructing a node feature matrix based on the specific customer demand information of the order, and constructing an adjacency matrix based on the configuration template information of the order; constructing a label matrix based on the configuration results of the order; inputting the node feature data set, the adjacency matrix data set, and the label data set into an intelligent solution model for customized product configuration schemes for training to obtain a trained intelligent solution model; when new order data is created, first determining the feature matrix based on the configuration structure and configuration parameters, and then using the intelligent solution model to solve the complete configuration scheme. The present invention avoids the process of acquiring, representing, maintaining, and updating configuration knowledge, and directly outputs a complete bill of materials, thereby improving the efficiency of customized product configuration design.

Description

Technical Field

[0001] The present invention relates to a configuration solution technology in the field of customized product configuration design, and specifically relates to an intelligent solution method for configuration schemes based on product configuration diagrams. Background Art

[0002] Configuration design is the process of selecting and combining optional components and modules to determine the final product's composition and properties, while meeting specific requirements and constraints, based on established design rules and experience. The goal is to design products that meet user needs and market competition requirements while maximizing design efficiency and minimizing design costs. Configuration design often requires consideration of multiple factors, including product functionality, performance, reliability, safety, usability, and environmental friendliness.

[0003] Traditional configuration solving methods suffer from difficulties in representing and maintaining configuration knowledge, and low configuration solving efficiency. Existing configuration design methods typically require the manual construction of configuration templates and configuration structure trees. For complex product families with numerous parameters, building configuration templates and maintaining configuration structure trees can become very difficult and time-consuming. Existing configuration solving methods based on deep learning have improved configuration efficiency to a certain extent, but they do not fully utilize information such as configuration templates. To address this issue, this method proposes an intelligent solution method for configuration solutions based on product configuration diagrams. This method comprehensively considers configuration parameter information and configuration template information, ultimately directly obtaining the bill of materials for customized products. Summary of the Invention

[0004] In order to solve the problems in the background technology, the present invention provides an intelligent solution method for configuration schemes based on product configuration diagrams, which uses configuration structure and configuration parameter information to construct a node feature matrix and an adjacency matrix, and directly solves the configuration material list of customized products through GCN and FC models.

[0005] The present invention adopts the following technical solutions:

[0006] 1. An intelligent solution method for configuration schemes based on product configuration diagrams

[0007] The specific steps include:

[0008] (1) Obtain historical customer order data from a customized product database. For each order data item, identify all configuration modules and their corresponding configuration parameters in the order data and construct a configuration module set. Search all configuration structures in the order data and construct a configuration structure set. Find all instances and the number of instances corresponding to all configuration modules in the order data and construct a configuration instance set.

[0009] (2) Establishing a node feature matrix based on the configuration module set;

[0010] Construct an adjacency matrix based on the configuration structure set, and use the adjacency matrix to represent the relationship between modules;

[0011] Encode each instance in the configuration instance set and establish a label matrix;

[0012] (3) For each type of configuration structure, an intelligent solution model for customized product configuration solutions is constructed. The input of the intelligent solution model for customized product configuration solutions includes the node feature matrix and the adjacency matrix, and the output is the label vector of each configuration module;

[0013] (4) Classifying historical order data according to configuration structure, obtaining training data sets of different configuration structures and performing data preprocessing, inputting the training data sets into the model corresponding to step (3) for training; the trained intelligent solution model can directly generate a list of customized product instances based on the input configuration parameters and configuration structure;

[0014] (5) Based on the new customer order data, the adjacency matrix and node feature matrix are constructed according to step (2), and the node feature matrix and adjacency matrix are input into the intelligent solution model trained in step (4). The label vector of the customized product is output, and the label vector is decoded and converted to obtain the corresponding instance. The configuration results of the customized product are integrated and displayed.

[0015] In the step (1):

[0016] The order data includes customer requirements, configuration templates, and instance lists; customer requirements are performance parameter requirements for configuration modules in customized products; configuration templates include configuration structures, which are complex product structure trees (complexity indicates coupling relationships between configuration modules); and instance lists are instances corresponding to all modules to be configured.

[0017] The configuration parameters are the performance parameters of the product.

[0018] The step 2) is specifically as follows:

[0019] 2.1) Construct the adjacency matrix between the modules to be configured according to the following formula:

[0020]

[0021] Among them, m i , m j are two configuration modules in the structure tree, i and j are the numbers of the configuration modules, and a ij Configure module m in the adjacency matrix i and m j the relationship between

[0022] 2.2) The node feature matrix where nNode Indicates the number of configuration modules included in the order data; d Node Indicates the number of parameter dimensions of the configuration module, which is obtained by concatenating the configuration parameters of all configuration modules.

[0023] A row of feature vectors in the node feature matrix represents the configuration parameters of a configuration module;

[0024] The label format described in 2.3) is one-hot encoding, which is obtained by concatenating the instance model codes of each module;

[0025] The configuration modules are numbered in sequence, and the unique hot encoding of each module instance is obtained in sequence and recorded as L i , concatenate the one-hot encodings of each module instance in numerical order to obtain the label vector.

[0026] The adjacency matrix in step (2) reflects the configuration structure information of the order task, and the node feature matrix reflects the configuration parameter information of the order task;

[0027] The customized product configuration solution intelligent solution model takes the form of a graph as input, and comprehensively considers the parameters of the configuration module itself and the relationship between each configuration module; the customized product configuration solution intelligent solution can be regarded as a multi-label classification prediction process, and the model finally outputs the instance models of all involved modules.

[0028] The customized product configuration solution intelligent solution model in step (3) is composed of two sub-networks: a graph neural network (GNN) model and a fully connected layer (FC) model. The node feature matrix F and the adjacency matrix A are used to generate low-dimensional dense module embeddings through the GNN model. The generated embeddings of each module are spliced ​​together as the input of the fully connected layer FC, and the instance encoding of each module is output through the FC model.

[0029] The GNN model uses the general graph convolutional neural network model GCN (Graph Convolution Networks) to obtain the embedding of each node. The GCN model consists of a series of message passing layers MPLs (Message Passing Layers). Each MPL uses an update function that updates the feature vector of each node based on its own feature vector and the feature vectors of adjacent nodes.

[0030] The FC model takes the concatenated node embeddings as input and outputs the instance encoding of each module. The calculation formula of one layer of neurons in the FC model is:

[0031]

[0032] Among them, v i is the neuron of the current layer, u j is the neuron of the previous layer, σ(x) is the ReLU activation function, w ij is the weight, b i For bias.

[0033] In the step (4):

[0034] The training dataset includes node feature matrix, adjacency matrix and label vector;

[0035] Data preprocessing is to fill in the default values ​​for modules without configuration parameters in the order.

[0036] In the step (5): when new order information enters the pre-trained solution model for prediction, the module parameters not defined by the customer in the node feature matrix are filled with preset default values.

[0037] In the step (5): the prediction result of the intelligent solution model is post-processed. When the confidence of the module configuration result is less than the threshold, it is deemed that no instance that meets the customer's needs is found, and the module for which no instance is found is marked as a "mismatched module" waiting for manual operation.

[0038] 2. A computer storage medium

[0039] The computer storage medium includes a stored program, wherein when the program is run, the device where the computer storage medium is located is controlled to execute the above-mentioned intelligent solution method for configuration solutions based on product configuration drawings.

[0040] 3. A processor

[0041] The processor is used to run a program, wherein the program executes the above-mentioned intelligent solution method for configuration solutions based on product configuration drawings when running.

[0042] Beneficial effects of the present invention:

[0043] Based on a graph neural network, this invention achieves intelligent solution for customized product configuration schemes. It uses the graph information of the configuration structure to represent the relationship between modules and expresses the customer's personalized needs through configuration parameters. It uses the GCN model to obtain the node embedding of each module, so that the configuration structure and configuration parameter information are simultaneously retained in the low-dimensional embedding. It implements classification prediction of each module based on the FC model and directly outputs the bill of materials of the customized product. For modules with low prediction confidence, the invention automatically marks them as "mismatched modules." This invention effectively utilizes the structure and parameter information of the configuration task, more directly solving the configuration scheme of the customized product, reducing the cost of acquiring, representing, maintaining, and updating complex configuration knowledge, and greatly improving the efficiency of customized product configuration design. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flowchart of an intelligent solution method for configuration schemes based on product configuration diagrams;

[0045] Figure 2 A schematic diagram of the product configuration;

[0046] Figure 3 Schematic diagram of the intelligent solution model for product configuration solutions. DETAILED DESCRIPTION

[0047] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0048] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a," "the," and "the" used in this invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0049] It should be understood that although the terms "first," "second," "third," etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0050] The specific embodiments of the present invention will be described more completely and clearly below with reference to the accompanying drawings. Figure 1 This is an intelligent solution method for configuration schemes based on product configuration diagrams, which includes the following steps:

[0051] (1) Based on the historical customer order data in the customized product database, identify all involved configuration modules and corresponding configuration parameters and construct a configuration module set; search all configuration structures and construct a configuration structure set. For specific configuration structures, number each module sequentially according to the rules; search all configuration results and construct a configuration instance set, and count the number of instances of each module;

[0052] For the module numbering rule in step (1), a specific method is given. The configuration structure is a tree diagram with complex coupling relationships. The module set of the diagram is V, and the associated edge set is E. Then the configuration structure can be described as G = (V, E); take v i ∈V represents a node (configuration module), e ij =(v i , v j )∈E indicates that there is an association relationship between node i and node j, and node v i The neighborhood of is defined as Nei(v i )={u∈V|(v i ,u)∈E};

[0053] Nodes are numbered recursively from top to bottom according to the number of neighbors.

[0054] (2) Obtain the specific customer requirements for each order and establish a node feature matrix based on the configuration parameters of each configuration module; establish an adjacency matrix based on the configuration structure information of the order, and the adjacency matrix represents the relationship between each module; encode each instance of the order configuration result and establish a label matrix;

[0055] The intelligent solution model of the present invention takes the form of a graph as input, comprehensively considers the parameters of the configuration module itself and the relationship between each configuration module, and finally outputs the instance model of all involved modules;

[0056] The node feature matrix in step (2) where n Node Indicates the number of configuration modules included in the order, d Node Indicates the number of module parameter dimensions, which is obtained by concatenating the configuration parameters of all modules; the configuration structure information is as follows: Figure 2 As shown in Figure 1, it is a tree diagram with complex coupling relationships, which is specifically manifested as mutual constraints between modules. The value of the adjacency matrix is ​​defined as follows:

[0057]

[0058] Among them, m i , m j are two configuration modules in the structure tree, i and j are the numbers of the configuration modules, and a ij Configure module m in the adjacency matrix i and m j The relationship between them.

[0059] exist Figure 2In the figure, node pairs connected by solid lines are parent-child nodes, while node pairs connected by dashed lines are associated nodes. The practical significance of associated nodes is that the parameters of a module instance can indirectly affect the configuration and selection of its associated nodes. For example, the configuration of the "traction rope" and "traction sheave" modules will affect each other. The solution model of the present invention can effectively take such invisible information into account when configuring.

[0060] The label matrix in step (2) is obtained by concatenating the instance model codes of each module. The specific implementation is: according to the node number (module number), the unique hot code of each module is obtained in sequence and recorded as L i , the one-hot encoding of each module is spliced ​​in the order of node number. The configuration solution process can be regarded as a multi-label classification prediction process. The splicing formula is as follows:

[0061]

[0062] Among them L all For the one-hot encoding after splicing, assume that there are m configuration modules in the configuration structure.

[0063] (3) Repeat steps (1) and (2) for each historical order data to obtain a training data set, and fill in the default values ​​for the modules without configuration parameters in the order; input the preprocessed data set of each order into the intelligent solution model for training;

[0064] (4) According to the new customer order data, the corresponding configuration structure is obtained and the adjacency matrix is ​​constructed. The node feature matrix is ​​constructed for the customer's configuration parameter requirements. If the customer does not put forward demand constraints for some configuration templates, the preset default values ​​will be filled in the corresponding positions; the node feature matrix and the adjacency matrix are input into the trained customized product configuration solution intelligent solution model, and the category vector of each node is output; according to the node number and configuration structure, it is converted into the corresponding configuration instance, and the confidence threshold α is set. For each module with a low prediction probability, the instantiation operation will not be performed and it will be marked as a "mismatched module", and the configuration results of the customized product will be integrated and displayed.

[0065] Figure 3 This is the model structure of the present invention. Figure 3 As shown in the figure, the working principle of the model is explained by taking 6 nodes as an example. For the 6 modules of "lower car top", "door machine", "car wall", "car door", "car bottom" and "car floor", there are d Node Configuration parameters are used to establish a node feature matrix based on the order customer demand information. If the required configuration parameters of the module are missing, the default values ​​are filled in to obtain the initial node feature matrix. in Represents the node feature vector of the i-th module. According to the correlation coefficient between modules, the adjacency matrix A is obtained, where the correlation coefficient between the i-th module and the j-th module is recorded as a ij .

[0066] The node feature matrix and adjacency matrix are first converted into node embeddings through the GCN model. The GCN model consists of a series of MPLs. Each MPL uses an update function that updates the feature vector of each node based on its own feature vector and the feature vectors of adjacent nodes. After the nth MPL, the graph F (n) The input feature matrix is ​​updated to F (n+1) , the following formula is given:

[0067]

[0068] Where σ(x) is the ReLU activation function, is the sum of the adjacency matrix and the identity matrix, The elements of is a diagonal matrix, and its diagonal elements are denoted by d ii ,have W (n) is a trainable weight matrix. Depending on the dimension of W(n), the dimension of the feature matrix may change after passing through MPL.

[0069] Embedding the output of the nth MPL as a node, we have the feature matrix in After a series of MPL transformations, the configuration structure and configuration parameter information are retained. These embeddings are concatenated as the input of the multi-label classification model. Here, the FC model is taken as an example. The FC model input is:

[0070]

[0071] The neurons in the current layer of the FC model are denoted as v i , the neurons in the previous layer are denoted as u j , there is a calculation formula

[0072]

[0073] Among them, σ(x) is the ReLU activation function, w ij is the weight, b i For the last layer output, to ensure that each output label is an independent probability, we use the sigmoid function as the activation function, and the calculation formula is:

[0074]

[0075] Where z is the number of layers of the FC model, R (z) is the vector of the output layer. A threshold α is set to verify whether each module is successfully instantiated according to the node number. For unsuccessful modules, the model marks them as "mismatched". The training loss function of the customized product configuration solution intelligent solution model is:

[0076]

[0077] Among them, y i is the true value of the module category label, is the predicted value of the module category label. In practice, the category encoding of each module is predicted, and one-hot encoding is used here. If the maximum probability of a module is less than 0.5, the current module is considered to have failed instantiation and is marked as a "mismatched" module.

[0078] The above embodiments are intended only to illustrate the design concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. The scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed in the present invention are within the scope of protection of the present invention.

Claims

1. An intelligent solution method for configuration scheme based on product configuration diagram, characterized in that: The specific steps include: (1) Obtain historical customer order data from a customized product database. For each order data item, identify all configuration modules and their corresponding configuration parameters in the order data and construct a configuration module set. Search all configuration structures in the order data and construct a configuration structure set. Find all instances and the number of instances corresponding to all configuration modules in the order data and construct a configuration instance set. (2) Establishing a node feature matrix based on the configuration module set; Construct an adjacency matrix based on the set of configuration structures; Encode each instance in the configuration instance set and establish a label matrix; (3) For each type of configuration structure, an intelligent solution model for customized product configuration solutions is constructed. The input of the intelligent solution model for customized product configuration solutions includes the node feature matrix and the adjacency matrix, and the output is the label vector of each configuration module; (4) Classifying historical order data according to the configuration structure, obtaining training data sets with different configuration structures and performing data preprocessing, and inputting the training data sets into the model corresponding to step (3) for training; (5) Based on the new customer order data, the adjacency matrix and node feature matrix are constructed according to step (2), and the node feature matrix and adjacency matrix are input into the intelligent solution model trained in step (4). The label vector of the customized product is output, and the label vector is decoded and converted to obtain the corresponding instance. The configuration results of the customized product are integrated and displayed.

2. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 1 is characterized in that: In the step (1): The order data includes customer requirements, configuration templates, and instance lists; customer requirements are performance parameter requirements for configuration modules in customized products; configuration templates include configuration structures, which are complex product structure trees; and instance lists are instances corresponding to all modules to be configured. The configuration parameters are the performance parameters of the product.

3. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 1 is characterized in that: The step 2) is specifically as follows: 2.1) Construct the adjacency matrix between the modules to be configured according to the following formula: Among them, m i ,m j are two configuration modules in the structure tree, i and j are the numbers of the configuration modules, and a ij Configure module m in the adjacency matrix i and m j the relationship between 2.2) The node feature matrix where n Node Indicates the number of configuration modules included in the order data; d Node Indicates the parameter dimension number of the configuration module, which is obtained by concatenating the configuration parameters of all configuration modules; The label format described in 2.3) is one-hot encoding, which is obtained by concatenating the instance model codes of each module; The configuration modules are numbered in sequence, and the unique hot encoding of each module instance is obtained in sequence and recorded as L i , concatenate the one-hot encodings of each module instance in numerical order to obtain the label vector.

4. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 1 is characterized in that: The customized product configuration solution intelligent solution model in step (3) is composed of two sub-networks: a graph neural network GNN model and a fully connected layer FC model. The node feature matrix F and the adjacency matrix A are used to generate low-dimensional dense module embeddings through the GNN model. The generated embeddings of each module are spliced ​​together as the input of the fully connected layer FC, and the instance encoding of each module is output through the FC model.

5. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 4 is characterized in that: The GNN model uses a graph convolutional neural network model GCN to obtain the embedding of each node. The GCN model consists of a series of message passing layers MPLs. Each MPL uses an update function that updates the feature vector of each node based on its own feature vector and the feature vectors of adjacent nodes. The FC model takes the concatenated node embeddings as input and outputs the instance encoding of each module. The calculation formula of one layer of neurons in the FC model is: Among them, v i is the neuron of the current layer, u j is the neuron of the previous layer, σ(x) is the ReLU activation function, w ij is the weight, b i For bias.

6. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 1 is characterized in that: In the step (4): The training dataset includes node feature matrix, adjacency matrix and label vector; Data preprocessing is to fill in the default values ​​for modules without configuration parameters in the order.

7. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 1 is characterized in that: In the step (5): when new order information enters the pre-trained solution model for prediction, the module parameters not defined by the customer in the node feature matrix are filled with preset default values.

8. The intelligent solution method for configuration scheme based on product configuration diagram according to claim 1 is characterized in that: In the step (5): The prediction results of the intelligent solution model are post-processed. When the confidence level of the module configuration result is less than the threshold, it is considered that no instance that meets the customer's requirements has been found. The module for which no instance has been found is marked as a "mismatched module" and awaits manual operation.

9. A computer storage medium, characterized in that The computer storage medium includes a stored program, wherein when the program is running, the device where the computer storage medium is located is controlled to execute the intelligent solution method for configuration scheme based on product configuration diagram according to any one of claims 1 to 8.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the intelligent solution method for configuration scheme based on product configuration diagram described in any one of claims 1 to 8.

Images