Part manufacturing carbon emission prediction method and system based on heterogeneous attribute graph

Abstract

The application belongs to the technical field of green intelligent manufacturing. A part manufacturing carbon emission prediction method and system based on a heterogeneous attribute graph are provided. A part heterogeneous attribute graph of a part to be manufactured is constructed according to design feature nodes, process feature nodes and resource activity nodes of the part to be manufactured, and the carbon emission value of each carbon emission path of the part heterogeneous attribute graph is determined. Graph convolution is used to perform initial feature propagation and representation learning on the design feature nodes, process feature nodes and resource activity nodes, and the node embedding representation is determined. The attention weight corresponding to the design feature is obtained according to the node embedding representation. The final representation of each design feature node is determined according to the carbon emission value and the attention weight. A double-layer graph neural network model is used to predict the carbon emission prediction result of each design feature node by taking the final representation of each design feature node as input. Higher-precision carbon emission prediction and attribution analysis are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of green intelligent manufacturing, and in particular to a part manufacturing carbon emission prediction method and system based on a heterogeneous attribute graph. BACKGROUND

[0002] The statements in this section merely provide background technology related to the present application and do not necessarily constitute prior art.

[0003] Predicting the manufacturing carbon emissions of a part in the design phase is a key link to achieve low-carbon design. If high-carbon design elements can be identified in the structure or process design phase, the designer can adjust the product structure configuration, material or process planning according to the prediction results, thereby effectively reducing the total carbon emissions of the manufacturing process.

[0004] The existing part design phase carbon emission prediction scheme has the following problems: (1) it is difficult to finely model and accurately predict the part structure features in the design phase, and there is a lack of low-carbon guidance for early design; (2) it is difficult to effectively integrate multi-source heterogeneous information such as design, process, and resources, and it is difficult to establish semantic association relationships between part structures and processing paths; (3) there is a lack of explainable analysis methods for carbon emission contribution paths, which cannot identify and attribute high-emission features, limiting the exploration of carbon emission reduction optimization space. SUMMARY

[0005] In order to solve the problems of the prior art, the present application provides a part manufacturing carbon emission prediction method and system based on a heterogeneous attribute graph, which introduces a graph neural network and a heterogeneous graph joint modeling mechanism in the modeling of product part manufacturing carbon emissions, making up for the key technical deficiencies of existing carbon emission prediction methods in terms of product part information modeling capability, carbon emission prediction capability, and result explainability. Through graph structure expression and double-layer graph neural network prediction, the multiple influences of local geometric features, process selection, and resource differences on carbon emissions can be captured simultaneously, achieving higher-precision carbon emission prediction and attribution analysis.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] In the first aspect, the present application provides a part manufacturing carbon emission prediction method based on a heterogeneous attribute graph.

[0008] A part manufacturing carbon emission prediction method based on a heterogeneous attribute graph includes the following processes:

[0009] According to the design feature nodes, process feature nodes, and resource activity nodes of the part to be manufactured, a part heterogeneous attribute graph of the part to be manufactured is constructed, and the carbon emission value of each carbon emission path of the part heterogeneous attribute graph is determined;

[0010] The design feature node, the process feature node and the resource activity node are subjected to initial feature propagation and representation learning by using graph convolution to determine node embedding representation, and the attention weight corresponding to the design feature is obtained according to the node embedding representation;

[0011] According to the carbon emission value and the attention weight, the final representation of each design feature node is determined;

[0012] The final representation of each design feature node is taken as input, and a double-layer graph neural network model is used to predict the carbon emission prediction result of each design feature node.

[0013] In a second aspect, the present application provides a part manufacturing carbon emission prediction system based on a heterogeneous attribute graph.

[0014] A part manufacturing carbon emission prediction system based on a heterogeneous attribute graph comprises:

[0015] A heterogeneous attribute graph generation unit is configured to construct a part heterogeneous attribute graph of a part to be manufactured according to a design feature node, a process feature node and a resource activity node of the part to be manufactured, and determine the carbon emission value of each carbon emission path of the part heterogeneous attribute graph;

[0016] An attention weight generation unit is configured to perform initial feature propagation and representation learning on the design feature node, the process feature node and the resource activity node by using graph convolution to determine node embedding representation, and obtain the attention weight corresponding to the design feature according to the node embedding representation;

[0017] A design feature node representation unit is configured to determine the final representation of each design feature node according to the carbon emission value and the attention weight;

[0018] A carbon emission result generation unit is configured to take the final representation of each design feature node as input, and use a double-layer graph neural network model to predict the carbon emission prediction result of each design feature node.

[0019] In a third aspect, the present application provides a computer device comprising a processor and a computer readable storage medium.

[0020] The processor is adapted to execute a computer program;

[0021] The computer readable storage medium has a computer program stored therein, and the computer program is executed by the processor to implement the part manufacturing carbon emission prediction method based on a heterogeneous attribute graph according to the first aspect of the present application.

[0022] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program, the computer program being adapted to be loaded and executed by a processor to implement the part manufacturing carbon emission prediction method based on a heterogeneous attribute graph according to the first aspect of the present application.

[0023] Compared with the prior art, the present application has the following beneficial effects:

[0024] 1. In the expression of product information, the present application introduces a heterogeneous attribute graph, which models design features, process features and processing resources as three types of heterogeneous nodes, and forms multiple processing paths through the "design-process-resource" links between the nodes. The system expresses part design, process and carbon emission information, breaking through the limitation of the prior art that multiple source heterogeneous information cannot be integrated, and improving the product information structure modeling capability and semantic association expression capability.

[0025] 2. The present application designs a two-layer graph neural network architecture combining graph convolution (GCN) and graph attention mechanism (GAT) to jointly represent the nodes in the heterogeneous attribute graph. GCN is used to capture the adjacency dependence of the nodes, and GAT is used to assign attention weights to the nodes and paths, thereby realizing deep modeling of the carbon emission paths and identification of the key nodes.

[0026] 3. The present application realizes carbon emission prediction based on design feature nodes and high-carbon path explainability analysis. Finally, the design feature nodes of the part are taken as input, and the corresponding carbon emission prediction results are output. The visualization analysis of the high-carbon emission contribution paths and key nodes is realized through the weight distribution of the graph attention mechanism, which provides geometric structure optimization basis for carbon emission reduction in the design stage and early optimization suggestions for low-carbon design.

[0027] The advantages of the additional aspects of the present application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0028] The accompanying drawings, which form a part of the present application, are used to provide further understanding of the present application, and the illustrative embodiments of the present application and their description are used to explain the present application, and do not constitute improper limitations on the present application.

[0029] Figure 1 A schematic diagram of the part manufacturing carbon emission prediction method based on a heterogeneous attribute graph provided for an exemplary embodiment of the present application;

[0030] Figure 2 A schematic diagram of the part heterogeneous attribute graph and node association relationship provided for an exemplary embodiment of the present application;

[0031] Figure 3A schematic diagram of a part heterogeneous attribute graph carbon emission path provided for an exemplary embodiment of the present application;

[0032] Figure 4 A schematic diagram of a graph neural network prediction model workflow provided for an exemplary embodiment of the present application;

[0033] Figure 5 A schematic diagram of a part carbon emission prediction model based on a graph neural network provided for an exemplary embodiment of the present application;

[0034] Figure 6 A schematic diagram of a part manufacturing carbon emission prediction system based on a heterogeneous attribute graph provided for an exemplary embodiment of the present application;

[0035] Figure 7 A schematic diagram of a computer device provided for an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0036] The present application will be further described below in conjunction with the accompanying drawings and embodiments.

[0037] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0038] Existing part design stage carbon emission prediction schemes mainly include empirical method, analytical model method and data-driven method (machine learning), but these schemes all have their own problems:

[0039] The empirical method is based on existing processing data or expert knowledge to construct empirical rules or regression models to estimate carbon emissions per process or per unit time, for example, linear fitting according to processing type, machine type and cutting parameters. This method has the advantages of simple implementation and no need for complex calculations, but also has obvious defects: poor generalization, dependent on empirical parameters of specific machine or material, difficult to migrate to other parts or processes; ignores design structure information, does not consider the influence of part geometry on machining path and carbon emission; cannot systematically integrate multi-source information affecting carbon emission.

[0040] The analytic model method is based on physical formula and energy consumption law to deduce energy consumption, and then combines carbon emission factor to calculate carbon emission. Its representative models include cutting energy model, spindle power model, etc. Some researches also introduce time-energy integral model. This kind of method is relatively rigorous in theory, and is suitable for carbon emission analysis of single process. However, it is complex in actual application, has strong parameter dependence, needs to accurately obtain multiple cutting and machine tool parameters, is not suitable for early design prediction, and is difficult to express complex structure and variable process route of product parts. It is only suitable for single process prediction, and cannot adapt to carbon emission analysis of multi-process manufacturing process.

[0041] The data-driven method (machine learning) can learn the complex nonlinear relationship between variables affecting carbon emission from historical data, overcoming some limitations of traditional methods. However, it still has the following limitations: current researches have a single product structure information expression, and cannot explicitly represent the correlation between part geometric structure features, process and resource allocation, making it difficult to accurately model complex structure and variable process route of parts, and also difficult to support reasoning of possible machining path and corresponding carbon emission from design features. At the same time, machine learning models generally lack interpretability, making it difficult to identify high-carbon emission nodes or paths, limiting their application value in green design optimization.

[0042] In view of the problems existing in the current part design stage carbon emission prediction scheme, the implementation mode proposes a part manufacturing carbon emission prediction method based on heterogeneous attribute graph, as shown in Figure 1 The user inputs the design information of the product part; according to the input design information, the heterogeneous attribute graph of the part is constructed; the constructed part heterogeneous attribute graph is input into the double-layer graph neural network prediction model; the double-layer graph neural network prediction model processes and predicts the input information; the double-layer graph neural network prediction model outputs two results: part manufacturing carbon emission and high-carbon node and path. This method first introduces graph neural network and heterogeneous graph joint modeling mechanism in product part manufacturing carbon emission modeling, making up for the key technical deficiencies of existing carbon emission prediction methods in product part information modeling ability, carbon emission prediction ability and result interpretability. In terms of method, through graph structure expression and double-layer graph neural network prediction, the multiple influences of local geometric features, process selection and resource differences on carbon emission can be captured, providing an effective means for carbon emission prediction and attribution analysis. In terms of task, facing typical parts with complex structure, multi-stage process and multiple paths (such as shafts), this method shows good adaptability and expansibility. The present application provides a new technical route for realizing "structure-driven green design" and "manufacturing process emission reduction optimization based on graph learning", and has broad application prospects in the fields of intelligent manufacturing and green design.

[0043] The application establishes a heterogeneous attribute graph composed of three types of nodes (design feature nodes, process feature nodes and resource feature nodes), extracts path-level features by combining graph convolution (GCN) and graph attention mechanism (GAT), and finally realizes carbon emission prediction of the design feature level through a deep regression network.

[0044] The application constructs a part multi-source information heterogeneous attribute graph G, which is a directed connected graph with the following definitions:

[0045] (1);

[0046] (2);

[0047] (3);

[0048] Wherein, is a node set; N is the total number of nodes, and the node type is a design feature, a process feature and a resource activity feature respectively; is an edge set, and the edge type is Wherein: Wherein, represents the edge between the design feature nodes, represents the edge between the design feature nodes and the process feature nodes, represents the edge between the process feature nodes and the resource activity nodes; represents a node type set, which contains three types of nodes: , represents a design feature node set, represents a process feature node set, represents a resource activity node set, represents an edge type set, which contains three semantic connections: "design-design", "design-process" and "process-resource".

[0049] Each node corresponds to an original feature vector:

[0050] (4);

[0051] Wherein, , and represent the specific attributes of the design feature, the process feature and the resource activity feature respectively.

[0052] Edge attribute function:

[0053] (5);

[0054] where, , and represent the specific attributes of DF-DF edge, DF-PF edge and PF-RA edge respectively.

[0055] According to the three semantic connections in the defined edge type set: "Design-Design", "Design-Process" and "Process-Resource", three kinds of adjacency matrices , and are constructed, which represent the connection relationship between nodes. If two nodes have a connection, it is 1, and if there is no connection, it is 0.

[0056] For example: if there are three design feature nodes , the adjacency matrix is:

[0057] (6);

[0058] which means: and have edge connections, and have edge connections.

[0059] Make specific definitions and explanations for the above nodes:

[0060] (1) Design feature (DF, Design Feature) refers to the basic unit of a product, such as holes, keyways, chamfers, etc. A part can be regarded as a collection of multiple design features. The feature vector of the design feature node is:

[0061] (7);

[0062] F represents shape information, which refers to a shaped entity composed of a set of geometric elements; S is the size information, including the basic dimensions of geometric elements such as length, width, height, radius, angle, etc.; T represents tolerance information, which refers to the technical requirements of roughness; M refers to material information, which is non-geometric related information with material type and material properties.

[0063] (2) Process feature (Process Feature) is a feature set used to express the part process constraint conditions, which contains the processing information of part process decision requirements and the manufacturing resource information of parts. Process feature is a multi-dimensional feature set that carries manufacturing knowledge. The mathematical description of the process feature node is:

[0064] (8)

[0065] Wherein, MF represents process meta-feature, and expresses process method processing stage information; MA represents machining allowance; CP represents process parameter, and includes cutting speed vc, feed speed f, cutting depth ap or width; and A represents accuracy.

[0066] (3) Resource activity node (RA) is determined by specific processing scheme. It is defined as the operation information set related to physical and functional changes of product entity, and causes environmental impact, and it contains selection of machine tool, cutting tool, cutting fluid, etc., and feature vector of operation feature node:

[0067] (9);

[0068] Wherein, E represents selected machining equipment, including machine tool type and spindle power; D is cutting tool used in machining, including tool type, tool hardness and tool life; P is selected cutting fluid type.

[0069] Part isomorphism attribute graph and attribute information and associated relationship of each node are as shown in Figure 2 , wherein the red circle is a resource activity node (including spindle speed, machine tool type, tool hardness, tool life, tool material and cutting fluid type), the blue circle is a process feature node (including cutting parameter, machining stage, machining allowance, dimensional accuracy, shape accuracy, machining method and geometric accuracy), and the green circle is a design feature node (including shape, size, material density, roughness and material hardness).

[0070] The present application carries out carbon emission calculation mode, and each design feature node (representing the first design feature node) can be machined through multiple process-resource paths:

[0071] (10);

[0072] Carbon emission value of each path is composed of two parts: material carbon emission and energy consumption carbon emission :

[0073] (11).

[0074] Part isomorphism attribute graph carbon emission path calculation is as shown in Figure 3 , and the green represents a design feature node, which can be completed by rough machining-semi-finishing-finishing, and related information is expressed by connected blue process feature nodes, each process feature node can select different color resource activity nodes, such as different machine tools and different cutting tools, more specifically, including:​

[0075] (12);

[0076] (13);

[0077] (14);

[0078] (15);

[0079] (16);

[0080] wherein, represents the carbon emission of the path formed by the 1st design feature node, the 2nd process feature node and the 5th resource activity node; represents the carbon emission of the path formed by the 1st design feature node, the 3rd process feature node and the 6th resource activity node; represents the carbon emission of the path formed by the 1st design feature node, the 4th process feature node and the 7th resource activity node; represents the carbon emission of the path formed by the 1st design feature node, the 4th process feature node and the 8th resource activity node; represents the carbon emission of the 1st design feature node; , , and respectively represent weight coefficients; represents the 2nd process feature node; represents the 5th resource activity feature node; represents the 3rd process feature node; represents the 6th resource activity feature node; represents the 4th process feature node; represents the 7th resource activity feature node; represents the 8th process feature node; represents the material carbon emission between represents the material carbon emission between represents the material carbon emission between represents the material carbon emission between represents the material carbon emission between ​​​​Energy carbon emission between represent and Energy carbon emission between represent and Energy carbon emission between represent and Energy carbon emission between

[0081] Material carbon emission Including tool wear carbon emission And cutting fluid carbon emission , Is the material carbon emission of the tool, Is the tool, Is the cutting fluid. The life cycle of the tool is related to the cutting condition, and when the tool wear condition reaches the tool wear criterion, the tool is regarded as waste cutting tool. The tool wear carbon emission depends on the contribution of the tool to the tool retirement during the machining, and the cutting fluid is an indispensable auxiliary material in the manufacturing process, and the common types include oil-based and water-based cutting fluid.

[0082] (17);

[0083] (18);

[0084] (19);

[0085] (20).

[0086] Where, Is the mass of the cutting tool, Is the carbon emission factor of tool production, Is the time of tool sharpening, Is the cutting time of using a specific cutting tool under specific cutting conditions, Is the life cycle of the cutting tool under specific cutting conditions, and the tool life is calculated by the number of sharpening times N and the tool service life H is the cutting fluid replacement period, generally 2-3 months; δ is the concentration of cutting fluid, and the concentration of water-based cutting fluid after dilution is generally 5%; , Cutting fluid preparation and waste liquid treatment carbon emission factors, respectively; V is the amount of cutting fluid.

[0087] Energy carbon emission Refers to the carbon emission caused by the power consumption of machine tools during the machining process, and the specific energy consumption (SEC) is introduced to quantify the energy carbon emission, that is, the electric energy required to remove unit volume or mass of material.

[0088] (21);

[0089] (22);

[0090] (23);

[0091] wherein, represents the carbon emission factor of electric energy; represents the specific energy consumption; represents the removal volume of machining material; respectively represent the cutting speed, feed speed and cutting depth of machining.

[0092] The total carbon emission of each design feature is:

[0093] (24);

[0094] wherein, represents the importance weight of the path (which can be learned through an attention mechanism).

[0095] The present application constructs a double-layer graph neural network model for prediction, as shown in Figure 4 , based on the graph convolution mechanism, initial feature propagation and representation learning are performed on each type of node; an attention mechanism is introduced to model the carbon emission contribution at the path level and embed carbon emission information; a multi-layer perceptron (MLP) structure is used to predict the design node embedding information; a loss value is calculated to identify high-carbon nodes. As shown in Figure 5 , it is a schematic diagram of a part carbon emission prediction model based on a graph neural network, the node feature matrix represents the combination of the initial feature vector of the node, the adjacency matrix represents the connection relationship between the nodes, the user inputs the product part design parameter information, according to the user input information, a heterogeneous attribute graph of the part is constructed, then the corresponding data is extracted to form a node feature matrix (information) and an adjacency matrix (connection relationship), the extracted node feature matrix and adjacency matrix are input into the prediction model, the prediction model outputs the part manufacturing carbon emission, through the MLP and the regression method, the output carbon emission is analyzed, the high-carbon nodes and paths are identified, and the part manufacturing carbon emission and the identified high-carbon nodes and paths are output to the user.

[0096] In the present implementation, the graph convolution mechanism is used to perform initial feature propagation and representation learning on each type of node, separate message passing channels are constructed for different edge types, and normalized adjacency matrices are used to realize feature aggregation, preliminary information fusion between different types of nodes is realized in this stage, specifically, including:

[0097] For design feature nodes:

[0098] (25);

[0099] For process feature nodes:

[0100] (26);

[0101] For resource feature nodes:

[0102] (27);

[0103] Where: are normalized adjacency matrices for different edge types (i.e. , and ); , and , , and are learnable projection matrices for corresponding edge types; the activation function is uniformly ReLU: .

[0104] After the graph convolution obtains the base representation, an attention mechanism is introduced to model the path-level carbon emission contribution, realizing the carbon emission attribution of design feature nodes. For each design-process-resource path , its corresponding attention weight is calculated:

[0105] (28);

[0106] Where is a learnable scoring vector; , and represent the node embedding representation of the i-th design feature node, the j-th process feature node, and the k-th resource feature node, respectively; represents vector concatenation; the activation function uses LeakyReLU with a negative slope of 0.01, represents the node embedding representation of the i-th design feature node, the node embedding representation of the j-th process feature node, and the node embedding representation of the k-th resource feature node.

[0107] Then, the carbon emission contribution embeddings on all paths are weighted and summed according to the attention weight as the final representation of the design feature node:

[0108] (29);

[0109] Where ​​The nonlinear mapping function for the path carbon emission value is a two-layer perceptron as follows:

[0110] (30);

[0111] wherein, represents the weight matrix of the first layer perceptron, represents the weight matrix of the second layer perceptron, represents the bias vector of the first layer perceptron, represents the bias vector of the second layer perceptron, represents the activation function.

[0112] To enhance the ability of the model to fit complex carbon emission distribution, a multi-layer perceptron (MLP) structure is used for the design node embedding for prediction, and the complete structure is as follows:

[0113] (31);

[0114] (32);

[0115] (33);

[0116] wherein, represents the weight of the first layer (learnable parameter), represents the bias of the first layer (learnable parameter), represents the weight of the second layer (learnable parameter), represents the bias of the second layer (learnable parameter), represents the weight of the third layer (learnable parameter), represents the bias of the third layer (learnable parameter), represents the carbon emission prediction result of the i design feature node, which can be used for mean square error regression training with the true value.

[0117] In order to effectively predict the carbon emission value of the design feature node, the double-layer graph neural network model proposed by the present application adopts a supervised learning method for training, and the training target is to minimize the error between the predicted value and the true carbon emission value, including the following processes:

[0118] A plurality of part samples are modeled as a heterogeneous graph structure, which includes design feature nodes (such as structural features such as holes, grooves, etc.), corresponding processing technology nodes, and resource nodes used by the processing technology, and a true carbon emission label corresponding to each design feature node is prepared, including the sum of material carbon emission and energy consumption carbon emission.

[0119] The output of the double-layer graph neural network model is a carbon emission prediction value for each design node , and the objective is to minimize the Mean Squared Error (MSE) loss between the prediction value and the true value:

[0120] (34);

[0121] wherein, denotes the total number of design feature nodes in the training set.

[0122] During the training process of the double-layer graph neural network model, the Adam optimizer is used for parameter updating, with an initial learning rate of 0.001, and the learning rate can be dynamically adjusted according to the performance of the validation set to improve the convergence effect. Graph-level mini-batch training strategy is adopted for training, i.e., each training batch consists of a subgraph, so as to realize efficient computation and sample expansion on graph structure data. To prevent overfitting of the model, an Early Stopping mechanism based on the validation set is introduced: when the validation set loss does not decrease for a certain number of consecutive training rounds, the training is automatically terminated, thereby improving the generalization ability. In terms of model performance evaluation, Mean Squared Error (MSE) and Mean Absolute Error (MAE) are used to measure the difference between the carbon emission prediction value and the true value. Through the above training optimization strategies and evaluation system, this method can achieve high-precision prediction of carbon emissions at the part design stage.

[0123] After the training of the graph neural network model is completed, a carbon emission attribution visualization module is further introduced to analyze and display the key influencing factors of different design features, processing paths, and resource combinations in the carbon emission results. The design of the carbon emission attribution visualization module is based on the weight coefficients learned in the graph attention mechanism. Through attribution analysis and path highlighting, high-carbon emission contribution paths are identified and displayed in a graphical manner. Specifically, it includes:

[0124] After the model training is completed, the attention weight of each design node related path is retained :

[0125] For any processing path starting from a design feature node , the path contribution value is defined as the average value of all edge weights:

[0126] (35);

[0127] A contribution threshold T is set, or the top N% paths are selected as "high-carbon emission contribution paths". These paths often result from specific processing parameter combinations or resource usage methods, leading to increased carbon emissions and having significant optimization potential.

[0128] (36).

[0129] wherein, represent the path contribution value of the i-th design feature node. i

[0130] For the nodes and edges identified as high contribution paths, a node heat map is used to color code each node according to its maximum contribution in the path, with colors ranging from green (low) to red (high), and output the explanatory text automatically summarized by the system, such as "large hole depth and high machining precision requirements lead to the need for precision grinding equipment, resulting in a significant increase in carbon emissions".

[0131] Figure 6 A part manufacturing carbon emission prediction system based on a heterogeneous attribute graph is shown, comprising:

[0132] The heterogeneous attribute graph generation unit 601 is configured to: construct a part heterogeneous attribute graph of a part to be manufactured according to design feature nodes, process feature nodes and resource activity nodes of the part to be manufactured, and determine a carbon emission value of each carbon emission path of the part heterogeneous attribute graph;

[0133] The attention weight generation unit 602 is configured to: perform initial feature propagation and representation learning on the design feature nodes, process feature nodes and resource activity nodes using graph convolution, determine node embedding representation, and obtain attention weights corresponding to the design features according to the node embedding representation;

[0134] The design feature node representation unit 603 is configured to: determine a final representation of each design feature node according to the carbon emission value and the attention weight;

[0135] The carbon emission result generation unit 604 is configured to: input the final representation of each design feature node, and use a double-layer graph neural network model to predict a carbon emission prediction result of each design feature node.

[0136] It can be understood that the above-mentioned units can be combined into one or several other units respectively or all, or some of the units can be further divided into a plurality of units with smaller functions to constitute, which can realize the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions, and in actual application, the functions of one unit can also be realized by multiple units, or the functions of multiple units can be realized by one unit. In other embodiments of the present application, the system can also include other units, and in actual application, these functions can also be realized by other units, and can be realized by multiple units. ​

[0137] According to another embodiment of the present application, the system described in the embodiment can be constructed by running a computer program (including program codes) capable of performing each step involved in the corresponding method described in Embodiment 1 on a general computing device such as a computer including processing elements and storage elements such as a Central Processing Unit (CPU), a Random Access Memory (RAM), a Read Only Memory (ROM), etc., the computer program can be recorded on a computer readable recording medium, loaded into the above-mentioned computing device through the computer readable recording medium, and run therein.

[0138] Figure 7 A computer device is shown, which includes a processor 701, a communication interface 702, and a computer readable storage medium 703. Among them, the processor 701, the communication interface 702, and the computer readable storage medium 703 can be connected through a bus or other means.

[0139] Among them, the communication interface 702 is used to receive and send data, the computer readable storage medium 703 can be stored in the memory of the electronic device, the computer readable storage medium 703 is used to store a computer program, the computer program includes program instructions, and the processor 701 is used to execute the program instructions stored by the computer readable storage medium 703.

[0140] The processor 701 is the computing core and control core of the electronic device, which is suitable for implementing one or more instructions, and is particularly suitable for loading and executing one or more instructions to realize a corresponding method flow or a corresponding function.

[0141] The processor 701 is configured to perform the following process:

[0142] According to the design feature nodes, process feature nodes and resource activity nodes of the part to be manufactured, a part heterogeneous attribute graph of the part to be manufactured is constructed, and the carbon emission value of each carbon emission path of the part heterogeneous attribute graph is determined;

[0143] The design feature nodes, process feature nodes and resource activity nodes are subjected to initial feature propagation and representation learning by using graph convolution, and node embedding representation is determined, and the attention weight corresponding to the design feature is obtained according to the node embedding representation;

[0144] According to the carbon emission value and the attention weight, the final representation of each design feature node is determined;

[0145] Taking the final representation of each design feature node as input, a double-layer graph neural network model is used to predict the carbon emission prediction result of each design feature node.

[0146] The application further provides a computer readable storage medium, which is a memory device in the electronic device and is used for storing programs and data. It can be understood that the computer readable storage medium herein can include a built-in storage medium in the electronic device, and of course can include an extended storage medium supported by the electronic device. The computer readable storage medium provides a storage space, and the storage space stores a processing system of the electronic device.

[0147] In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and the instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium herein can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory; optionally, the computer readable storage medium can be at least one computer readable storage medium located away from the aforementioned processor.

[0148] In one embodiment, the computer readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer readable storage medium to implement the following process:

[0149] According to the design feature node, the process feature node and the resource activity node of the part to be manufactured, a part heterogeneous attribute graph of the part to be manufactured is constructed, and a carbon emission value of each carbon emission path of the part heterogeneous attribute graph is determined;

[0150] Initial feature propagation and representation learning are performed on the design feature node, the process feature node and the resource activity node by using graph convolution, a node embedding representation is determined, and an attention weight corresponding to the design feature is obtained according to the node embedding representation;

[0151] According to the carbon emission value and the attention weight, a final representation of each design feature node is determined;

[0152] The final representation of each design feature node is taken as input, and a double-layer graph neural network model is used to predict a carbon emission prediction result of each design feature node.

[0153] Those of ordinary skill in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0154] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in or transmitted by a computer readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (for example, coaxial cable, optical fiber, digital line) or wireless (for example, infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data processing device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (for example, floppy disk, hard disk, magnetic tape), optical media (for example, DVD), or semiconductor media (for example, solid state disk) and the like.

[0155] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for part manufacturing carbon emission prediction based on heterogeneous attributed graphs, characterized in that, The method comprises the following steps: According to the design feature nodes, process feature nodes and resource activity nodes of the part to be manufactured, a part heterogeneous attribute graph of the part to be manufactured is constructed, and the carbon emission value of each carbon emission path of the part heterogeneous attribute graph is determined; Initial feature propagation and representation learning are performed on the design feature nodes, process feature nodes and resource activity nodes by using graph convolution to determine node embedding representation, and the attention weight corresponding to the design feature is obtained according to the node embedding representation; Each design feature node is completed by multiple process-resource paths, and the carbon emission value of each path includes material carbon emission and energy consumption carbon emission; According to the carbon emission value and the attention weight, the final representation of each design feature node is determined; a final representation of the design feature node: wherein, represents a corresponding attention weight, represents a carbon emission of a path formed by the first design feature node, the first process feature node, and the first resource activity node; A double-layer graph neural network model is used to predict the carbon emission prediction result of each design feature node by taking the final representation of each design feature node as input. The feature vector of the design feature node is represented as: where F represents shape information, S represents size information, T represents tolerance information, and M represents material information. The characteristic vector of the process feature node is represented as: wherein MF represents a process meta-feature, MA represents a machining allowance, CP represents a process parameter, and A represents accuracy. The feature vector of the resource activity node is represented as: , E representing a selected machining equipment, D representing a cutting tool used in the machining, P representing a selected cutting fluid type.

2. The part manufacturing carbon emission prediction method based on a heterogeneous attribute graph according to claim 1, wherein The part heterogeneous attribute graph is expressed as: wherein, represents a set of nodes, represents a set of edges, represents a set of node types, represents a set of edge types.

3. The part manufacturing carbon emission prediction method based on a heterogeneous attribute graph according to claim 1, wherein The corresponding attention weights are: ,in, is a learnable scoring vector, Representative The node embedding representation of the design feature nodes, Representative The node embedding representation of process feature nodes, Representative The node embedding representation of the resource activity node, LeakyReLU is the activation function, represents transpose, represents vector concatenation, Representative process feature nodes, Representative Resource activity nodes, Representative The node embedding representation of process feature nodes, Representative Node embedding representation of resource activity nodes.

4. The part manufacturing carbon emission prediction method based on a heterogeneous attribute graph according to any one of claims 1-3, wherein A multi-layer perceptron is used to predict the carbon emission prediction result of each design feature node, including: ； ； ； in, Representative i The final representation of the design feature nodes, represents the weight of the first layer, represents the bias of layer 1, represents the weight of the second layer, represents the bias of layer 2, represents the weight of the third layer, represents the bias of layer 3, Representative i Carbon emission prediction results for each design feature node.

5. The part manufacturing carbon emission prediction method based on a heterogeneous attribute graph according to any one of claims 1-3, wherein After the training of the double-layer graph neural network model is completed, the attention weight of each design node related path is retained, the path contribution value of any one processing path starting from a design feature node is defined as the average value of all edge weights, a path with a contribution value greater than a set contribution threshold is selected as a high-carbon-emission contribution path, or the top N% of paths according to the average value are selected as high-carbon-emission contribution paths, and a node heat map is used to represent the high-carbon-emission contribution path.

6. A heterogeneous attribute graph-based part manufacturing carbon emission prediction system, characterized in that, The method comprises the following steps: The heterogeneous attribute graph generation unit is configured to construct a part heterogeneous attribute graph of a part to be manufactured according to design feature nodes, process feature nodes and resource activity nodes of the part to be manufactured, and determine the carbon emission value of each carbon emission path of the part heterogeneous attribute graph; The attention weight generation unit is configured to perform initial feature propagation and representation learning on the design feature nodes, process feature nodes and resource activity nodes by using graph convolution to determine node embedding representation, and obtain the attention weight corresponding to the design feature according to the node embedding representation; Each design feature node is completed by multiple process-resource paths, and the carbon emission value of each path includes material carbon emission and energy consumption carbon emission; The design feature node representation unit is configured to determine the final representation of each design feature node according to the carbon emission value and the attention weight; a final representation of the design feature nodes: wherein, represent a corresponding attention weight, represent a path formed by the first design feature node, the first process feature node, and the first resource activity node; The carbon emission result generation unit is configured to predict the carbon emission prediction result of each design feature node by taking the final representation of each design feature node as input and using a double-layer graph neural network model. The feature vector of the design feature node is represented as: where F represents shape information, S represents size information, T represents tolerance information, and M represents material information. The feature vector of the process feature node is represented as: wherein MF represents a process meta-feature, MA represents a machining allowance, CP represents a process parameter, and A represents accuracy. The feature vector of the resource activity node is represented as: E representing a selected machining equipment, D representing a selected cutting tool used in the machining, P representing a selected cutting fluid type.​ 7. A computer device, comprising: The method comprises the following steps: A processor and a computer readable storage medium; A processor adapted to execute a computer program; A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and the computer program, when executed by a processor, implements the method for predicting carbon emissions of part manufacturing based on a heterogeneous attribute graph according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by a processor to implement the method for predicting carbon emissions of part manufacturing based on a heterogeneous attribute graph according to any one of claims 1 to 5.

Images