Disassembly control method and device for electric power material equipment

Abstract

The invention provides an electric power material equipment disassembly control method and device, and the method comprises the steps: collecting basic information data of standard electric power material equipment, constructing a standard electronic material equipment structure knowledge graph, and building a knowledge base; the method comprises the following steps: constructing a standard electronic material equipment disassembling rule based on electric power material equipment domain knowledge, and establishing a disassembling rule base; according to the identity information of the electric power material equipment to be disassembled, respectively obtaining a corresponding standard electronic material equipment structure knowledge graph and a standard electronic material equipment disassembling rule from the knowledge base and the disassembling rule base; a prediction model based on graph neural coding and a prediction model based on Transform decoding are constructed, and training is carried out; and inputting the disassembling target, the initial state of the to-be-disassembled electric power material equipment, the corresponding standard electronic material equipment structure knowledge graph and the standard electronic material equipment disassembling rule into the trained prediction model, and outputting an optimal disassembling scheme. According to the method, the disassembling efficiency of the electric power material equipment can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power material disassembly, and in particular to an electric power material equipment disassembly control method and device. BACKGROUND

[0002] To build a recycling system for waste materials in the electric power industry, waste electric power material equipment is generally required to be disassembled, and the disassembled parts that can be used are recycled.

[0003] Patent CN20744469B discloses a method and system for collaborative optimization of classification and green disassembly of power grid waste materials, relating to the technical field of material classification and disassembly, comprising the following steps: obtaining multi-source feature data of the waste materials to be disposed; constructing a residual value prediction model based on the multi-source feature data, outputting a component residual value mapping atlas; analyzing the structural information of the target materials, and generating a disassembly path decoupling graph in combination with the residual value mapping atlas; establishing a double-objective optimization scheduling model, and solving the scheduling model to generate a green disassembly operation sequence that meets the first constraint condition; collecting real-time feedback data when executing the disassembly operation sequence, and dynamically correcting the residual value prediction model parameters and the scheduling model constraint weights.

[0004] The above method aims to achieve green disassembly, minimize pollution risk and carbon emissions, so the generated disassembly operation sequence may be tedious and complex, resulting in low production efficiency. SUMMARY

[0005] The present application provides an electric power material equipment disassembly control method and device, which can effectively improve the production efficiency of electronic material equipment disassembly.

[0006] An electric power material equipment disassembly control method, comprising: Collecting standard electric power material equipment basic information data and constructing a standard electronic material equipment structure knowledge graph to establish a knowledge base; Based on the knowledge in the field of electric power material equipment, constructing standard electronic material equipment disassembly rules and establishing a disassembly rule base; According to the identity information of the electric power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and standard electronic material equipment disassembly rules are obtained from the knowledge base and disassembly rule base, respectively; Constructing a prediction model based on graph neural coding and Transformer decoding and training it; The disassembly target, the initial state of the electric power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and standard electronic material equipment disassembly rules are input into the trained prediction model, and the optimal disassembly scheme is output.

[0007] Further, the standard electric power material equipment basic information data is CAD drawing data or BOM list; constructing a standard electronic material equipment structure knowledge graph, comprising: identifying components and assembly relationships based on the standard electric power material equipment basic information data; extracting constraint conditions in the assembly relationship and identifying assembly features; taking the standard electric power material equipment as a root node, adding all component nodes, and creating relationship edges for each assembly relationship to connect the corresponding component nodes, adding constraint conditions and assembly features to each relationship edge, and obtaining a standard electronic material equipment structure knowledge graph.

[0008] Further, based on the electric power material equipment domain knowledge, a standard electronic material equipment disassembly rule is constructed, comprising: extracting disassembly operation steps from the text of the electric power material equipment domain knowledge, and taking each disassembly operation step as an operation root node, obtaining safety rules, sequence rules, operation rules, tool rules and quality rules and their constraints under each operation root node and structuring them, and adding them to the operation root node.

[0009] Further, the prediction model includes an input layer, a processing layer and an output layer; the processing layer includes a graph neural network coding module, a rule constraint fusion module and a Transformer sequence decoding module; The input layer is used to input the disassembly target, the initial state of the electric power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and the standard electronic material equipment disassembly rule and perform preprocessing; the graph neural network coding module is used to perform structural coding on the preprocessed standard electronic material equipment structure knowledge graph, outputting node embedding matrix and graph-level representation vector; the rule constraint fusion module is based on the preprocessed standard electronic material equipment disassembly rule, the disassembly target and the initial state of the electric power material equipment to be disassembled to perform dynamic integration, outputting conditional constraint guidance; the Transformer sequence decoding module is based on the preprocessed disassembly target, the initial state of the electric power material equipment to be disassembled, the node embedding matrix, the graph-level representation vector and the conditional constraint guidance to generate an optimal disassembly scheme autoregressively, and the output layer is used to output the optimal disassembly scheme.

[0010] Further, the input layer preprocesses the disassembly target, the initial state of the electric power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and the standard electronic material equipment disassembly rule, comprising: based on the standard electronic material equipment structure knowledge graph, extracting component node features and edge features to generate a component feature matrix; Encode the initial state of the power material equipment to be disassembled, generate an initial state description; Parse the disassembly target and convert it to vector form to obtain a target vector; Extract safety rules, sequence rules, operation rules, tool rules, and quality rules and their constraints from the standard electronic material equipment disassembly rules.

[0011] Further, the graph neural network encoding module encodes the preprocessed standard electronic material equipment structure knowledge graph, outputs a node embedding matrix and a graph-level representation vector, including: The graph neural network encoding module aggregates component node information in the component feature matrix based on multi-layer graph convolution, generates a multi-dimensional vector representation containing each component node, obtains the node embedding matrix, and perceives the connection relationship between component nodes, captures local and global information, and generates a graph-level representation vector containing device global feature representation.

[0012] Further, the rule constraint fusion module dynamically integrates based on the preprocessed standard electronic material equipment disassembly rules, the disassembly target, and the current state of the power material equipment to be disassembled, and outputs a conditional constraint guide, including: Iterate through each rule in the preprocessed standard electronic material equipment disassembly rules, determine whether it is compatible with the initial state and disassembly target of the power material equipment to be disassembled, and add the compatible rule to the conditional rule list; Based on the constraints of the corresponding rules in the conditional rule list, a constraint representation matrix is established; The conditional rule list and the constraint representation matrix are output as a conditional constraint guide.

[0013] Further, the Transformer sequence decoding module generates an optimal disassembly scheme based on the preprocessed disassembly target, the initial state of the power material equipment to be disassembled, the node embedding matrix, the graph-level representation vector, and the conditional constraint guide autoregressively, including: Iterate the following operations until the stop condition is met: based on the conditional rule list in the conditional constraint guide, calculate the executable operation set, for each operation in the executable operation set, calculate the comprehensive score according to the current state of the power material equipment to be disassembled, the target vector, the node embedding matrix, the graph-level representation vector, the constraint representation matrix in the conditional constraint guide, and the optimization target, and select the operation with the highest comprehensive score as the output, update the current state of the power material equipment to be disassembled; After the stop condition is met, output each disassembly component sequence and its operation to obtain the optimal disassembly scheme.

[0014] Furthermore, the objective function of the prediction model includes minimizing the number of operation steps, maximizing component integrity, and minimizing the number of tool switching operations.

[0015] A power equipment dismantling control device, comprising: The knowledge base building module is used to collect basic information data of standard power materials and equipment and construct a knowledge graph of the structure of standard electronic materials and equipment to build a knowledge base. The rule base creation module is used to build standard electronic equipment disassembly rules based on knowledge of the power equipment field, and to establish a disassembly rule base; The acquisition module is used to acquire the corresponding standard electronic equipment structure knowledge graph and standard electronic equipment dismantling rules from the knowledge base and dismantling rule base, respectively, based on the identity information of the power equipment to be dismantled. The model building module is used to build and train prediction models based on graph neural coding and Transformer decoding. The analysis module is used to input the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules into the trained prediction model, and output the optimal dismantling scheme.

[0016] Furthermore, the basic information data for standard power equipment and materials is CAD drawing data or BOM list; The knowledge base building module constructs a standard electronic materials and equipment structure knowledge graph, including: Identification of components and assembly relationships is performed based on the aforementioned standard power equipment and materials basic information data; Extract the constraints in the assembly relationship and identify the assembly features; Using standard power equipment as the root node, add all component nodes, create relationship edges for each assembly relationship, connect the corresponding component nodes, add constraints and assembly features to each relationship edge, and obtain a knowledge graph of the structure of standard electronic equipment.

[0017] Furthermore, the rule base establishment module, based on knowledge of the power equipment and materials domain, constructs standard electronic equipment and materials disassembly rules, including: The dismantling operation steps are extracted from the text of knowledge in the field of power materials and equipment. Each dismantling operation step is taken as the operation root node. The safety rules, sequence rules, operation rules, tool rules and quality rules and their constraints under each operation root node are obtained, structured and added to the operation root node.

[0018] Furthermore, the prediction model includes an input layer, a processing layer, and an output layer; the processing layer includes a graph neural network encoding module, a rule constraint fusion module, and a Transformer sequence decoding module; The input layer is used to input the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules, and performs preprocessing. The graph neural network encoding module is used to perform structural encoding on the preprocessed standard electronic equipment structure knowledge graph and output a node embedding matrix and a graph-level representation vector. The rule constraint fusion module dynamically integrates the preprocessed standard electronic equipment dismantling rules, the dismantling target, and the initial state of the power equipment to be dismantled, and outputs conditional constraint guidance. The Transformer sequence decoding module generates the optimal dismantling scheme through autoregression based on the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and the conditional constraint guidance. The output layer is used to output the optimal dismantling scheme.

[0019] Furthermore, the input layer preprocesses the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules, including: Based on the standard electronic materials and equipment structure knowledge graph, component node features and edge features are extracted to generate a component feature matrix; The initial state of the power materials and equipment to be dismantled is encoded to generate an initial state description. The decomposed target is analyzed and converted into vector form to obtain the target vector; Safety rules, sequence rules, operation rules, tool rules, and quality rules and their constraints are extracted from the standard electronic equipment dismantling rules.

[0020] Furthermore, the graph neural network encoding module performs structural encoding on the preprocessed standard electronic material equipment structure knowledge graph, outputting a node embedding matrix and a graph-level representation vector, including: The graph neural network encoding module aggregates the component node information in the component feature matrix based on multi-layer graph convolution, generates a multi-dimensional vector representation containing each component node, obtains the node embedding matrix, perceives the connection relationship between component nodes, captures local and global information, and generates a graph-level representation vector containing global feature representation of the device.

[0021] Furthermore, the rule constraint fusion module dynamically integrates the preprocessed standard electronic equipment dismantling rules, dismantling targets, and the current state of the power equipment to be dismantled, outputting conditional constraint guidance, including: Iterate through each rule in the preprocessed standard electronic material and equipment dismantling rules, determine whether it matches the initial state and dismantling target of the power material and equipment to be dismantled, and add the matching rules to the conditional rule list. Based on the constraints of the corresponding rules in the conditional rule list, a constraint representation matrix is ​​established; The conditional rule list and the constraint representation matrix are output as constraint guidance for conditionalization.

[0022] Furthermore, the Transformer sequence decoding module, based on the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and conditional constraints, guides autoregression to generate the optimal dismantling scheme, including: Iteratively perform the following operations until the stopping condition is met: Based on the list of conditional rules in the conditional constraint guidance, calculate the set of executable operations; for each operation in the set of executable operations, calculate the comprehensive score based on the current state of the power equipment to be dismantled, the target vector, the node embedding matrix, the graph-level representation vector, the constraint representation matrix in the conditional constraint guidance, and the optimization target; and select the operation with the highest comprehensive score to output and update the current state of the power equipment to be dismantled. After the stopping condition is met, output the sequence of each disassembled component and its operation to obtain the optimal disassembly scheme.

[0023] Furthermore, the objective function of the prediction model includes minimizing the number of operation steps, maximizing component integrity, and minimizing the number of tool switching operations.

[0024] The method and apparatus for controlling the dismantling of power equipment provided by this invention have at least the following beneficial effects: (1) A prediction model based on graph neural network coding and Transformer decoding is adopted to minimize the number of operation steps, maximize the integrity of components and minimize the number of tool switching. The generated optimal disassembly scheme can ensure the accuracy of disassembly steps while effectively improving the production efficiency of electronic material equipment disassembly. (2) The prediction model based on graph neural coding and Transformer decoding can effectively capture the spatial relationship between components, understand the assembly topology, identify physical constraints and perform global optimization, thus ensuring the accuracy of the disassembly steps; (3) The construction of knowledge base and rule base is applicable to the dismantling of different power materials and equipment, and has stronger universality. Attached Figure Description

[0025] Figure 1 This is a flowchart of one embodiment of the power equipment dismantling control method provided by the present invention.

[0026] Figure 2 This is a schematic diagram of the structure of a prediction model in the power equipment dismantling control method provided by the present invention.

[0027] Figure 3 This is a schematic diagram of one embodiment of the power equipment dismantling control device provided by the present invention. Detailed Implementation

[0028] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0029] refer to Figure 1 In some embodiments, a method for controlling the dismantling of power equipment is provided, comprising: S1. Collect basic information data on standard power materials and equipment and construct a knowledge graph of the structure of standard electronic materials and equipment to establish a knowledge base; S2. Based on knowledge of the power equipment and materials field, construct standard electronic equipment disassembly rules and establish a disassembly rule library; S3. Based on the identity information of the power materials and equipment to be dismantled, obtain the corresponding standard electronic materials and equipment structure knowledge graph and standard electronic materials and equipment dismantling rules from the knowledge base and dismantling rule base, respectively. S4. Construct and train a prediction model based on graph neural coding and Transformer decoding; S5. Input the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules into the trained prediction model, and output the optimal dismantling scheme.

[0030] Furthermore, in step S1, the basic information data of standard power materials and equipment is CAD drawing data or BOM list; Construct a knowledge graph of the structure of standard electronic materials and equipment, including: S11. Identify components and assembly relationships based on the standard power equipment basic information data; S12. Extract the constraints in the assembly relationship and identify the assembly features; S13. Using standard power equipment as the root node, add all component nodes, create relationship edges for each assembly relationship, connect the corresponding component nodes, add constraints and assembly features to each relationship edge, and obtain the standard electronic equipment structure knowledge graph.

[0031] Specifically, for CAD drawing data, format conversion is performed beforehand, for example, converting it to STEP AP242 format. Then, CAD software APIs or professional parsing tools are used to identify component and assembly relationships. Assembly relationships refer to the assembly methods and connections between components; for example, component A and component B are connected using a screw connection. Constraints in assembly relationships include rules that must be followed during assembly; for example, component A and component B must fit concentrically. Assembly features can include connection features and process features. Connection features include threaded holes, slots, and welded edges, while process features include fillets and chamfers.

[0032] Using standard power equipment as the root node, add all component nodes, and form a tree structure based on the relationships between component nodes. The leaf nodes of each component node are its constituent components. Create relationship edges for each assembly relationship, connect the corresponding component nodes, and add constraints and assembly features to each relationship edge to obtain a knowledge graph of standard electronic equipment structure.

[0033] For the BOM (Bill of Materials) list, the flat BOM list is converted into a tree structure. First, the component and assembly relationship is identified based on the component code. Similarly, the constraints in the assembly relationship are extracted and the assembly features are identified. With the standard power materials and equipment as the root node, all component nodes are added. Based on the relationship between the component nodes, a tree structure is formed. In addition, attribute enhancement can be performed based on the BOM list, adding associated technical parameters, whether it contains hazardous substances, and other attribute information to each component node.

[0034] Furthermore, in step S2, based on knowledge of the power equipment and materials field, standard electronic equipment and materials disassembly rules are constructed, including: The dismantling operation steps are extracted from the text of knowledge in the field of power materials and equipment. Each dismantling operation step is taken as the operation root node. The safety rules, sequence rules, operation rules, tool rules and quality rules and their constraints under each operation root node are obtained, structured and added to the operation root node.

[0035] Specifically, sources of knowledge in the field of power equipment and materials can include the manufacturer's official maintenance manual, disassembly sections in the product's user manual, engineering standards and specifications, and work instructions.

[0036] The dismantling operation steps are extracted from the text of knowledge in the field of power materials and equipment. Each dismantling operation step is taken as the operation root node. The safety rules, sequence rules, operation rules, tool rules and quality rules and their constraints under each operation root node are obtained and structured, and added to the operation root node to form a tree structure.

[0037] For example, for a specific step in disassembling component A and component B, considered as the root node, the next level nodes include safety rules, sequence rules, operation rules, tool rules, and quality rules. The safety rules node further includes electrical safety constraints, mechanical safety constraints, chemical safety constraints, and environmental safety constraints. The sequence rules node includes the order of operations, mandatory constraints, and recommended constraints, such as the requirement that the current operation must be performed after another operation. The operation rules node includes standard operations, special operations, and exception handling. Tool rules include tool selection constraints, tool usage constraints, and tool management constraints. The quality rules node includes component protection constraints, accuracy maintenance constraints, and reusability constraints.

[0038] Furthermore, in step S3, the established knowledge base includes standard electronic equipment structure knowledge graphs for multiple power equipment, each with a unique identifier for easy searching. Similarly, the established rule base contains standard electronic equipment disassembly rules for multiple power equipment, each with a unique identifier.

[0039] Based on the identity information of the power materials and equipment to be dismantled, a matching judgment is made with the unique identifier information in the knowledge base and rule base to find the corresponding standard electronic materials and equipment structure knowledge graph and standard electronic materials and equipment dismantling rules, which are used as inputs for the prediction model.

[0040] Further, refer to Figure 2 In step S4, the prediction model includes an input layer 1, a processing layer 2, and an output layer 3; the processing layer 2 includes a graph neural network encoding module 21, a rule constraint fusion module 22, and a Transformer sequence decoding module 23. Input layer 1 is used to input the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules, and to perform preprocessing. Graph neural network encoding module 21 is used to perform structural encoding on the preprocessed standard electronic equipment structure knowledge graph and output node embedding matrix and graph-level representation vector. Rule constraint fusion module 22 dynamically integrates the preprocessed standard electronic equipment dismantling rules, the dismantling target, and the initial state of the power equipment to be dismantled, and outputs conditional constraint guidance. Transformer sequence decoding module 23 generates the optimal dismantling scheme through autoregression based on the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and the conditional constraint guidance. Output layer 3 outputs the optimal dismantling scheme.

[0041] The objective function of the prediction model includes minimizing the number of operation steps, maximizing component integrity, and minimizing the number of tool switching operations.

[0042] Furthermore, input layer 1 preprocesses the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules, including: Based on the standard electronic materials and equipment structure knowledge graph, component node features and edge features are extracted to generate a component feature matrix; The initial state of the power materials and equipment to be dismantled is encoded to generate an initial state description. The decomposed target is analyzed and converted into vector form to obtain the target vector; Safety rules, sequence rules, operation rules, tool rules, and quality rules and their constraints are extracted from the standard electronic equipment dismantling rules.

[0043] Specifically, the input to input layer 1 is the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules. The dismantling target is the degree of dismantling, such as complete dismantling or partial dismantling. The initial state of the power equipment to be dismantled can be, for example, a complete power equipment or a power equipment that has already been partially dismantled. This initial state is the starting point of this dismantling.

[0044] Furthermore, the preprocessing of the standard electronic material and equipment structure knowledge graph in input layer 1 includes extracting component node features and edge features to generate a component feature matrix. For example, component node features can include geometric features, structural features, functional features, disassembly features, and safety and environmental features. Geometric features can include spatial location, size information, volume, mass, complexity, etc. Structural features can include the level depth of the component node in the tree structure and the number of connected component nodes. Functional features can include the type encoding of the component node, such as structural components, moving parts, electrical components, sealing components, etc. Disassembly features can include disassembly capability (e.g., disassembly difficulty rating 1-10), tool requirements, operation actions, disassembly time estimation, etc. Safety and environmental features can include hazard level (high temperature, high pressure, toxic, etc.), environmental classification (recyclable, hazardous, etc.), and material type (metal, plastic, etc.). Edge features can include connection type features, direction features, and sequence dependency features. Connection features can include connection method, connection strength, and number of connections. Direction features can include disassembly direction. Sequence dependency features can include mandatory disassembly in sequence and recommended disassembly in sequence. Based on the aforementioned component node features and edge features, a component feature matrix is ​​generated and used as an output of the input layer.

[0045] Furthermore, input layer 1 encodes the initial state of the power equipment to be dismantled and generates an initial state description, which describes the initial components of the power equipment to be dismantled.

[0046] Furthermore, input layer 1 parses the disassembly target and converts it into vector form to obtain the target vector. For example, for complete disassembly, the identifier of each component in the target vector is 1. For partial disassembly, the identifier of the component that needs to be disassembled in the target vector is 1, and the identifier of the component that does not need to be disassembled is 0.

[0047] Furthermore, input layer 1 extracts safety rules, sequence rules, operation rules, tool rules, and quality rules and their constraints from the standard electronic material and equipment dismantling rules, encodes them, and converts them into a form that the next module can recognize.

[0048] Furthermore, the graph neural network encoding module 21 performs structural encoding on the preprocessed standard electronic material equipment structure knowledge graph, outputting a node embedding matrix and a graph-level representation vector, including: The graph neural network encoding module aggregates the component node information in the component feature matrix based on multi-layer graph convolution, generates a multi-dimensional vector representation containing each component node, obtains the node embedding matrix, perceives the connection relationship between component nodes, captures local and global information, and generates a graph-level representation vector containing global feature representation of the device.

[0049] Specifically, the input to the graph neural network encoding module 21 is the preprocessed standard electronic material equipment structure knowledge graph, i.e., the component feature matrix. Based on the component feature matrix, the component node information in the component feature matrix is ​​aggregated using multi-layer graph convolution to generate a node embedding matrix. The generated node embedding matrix contains multi-dimensional vector representations of component nodes. Based on the relationship-aware function, the connection relationships between component nodes are perceived, capturing local and global information to obtain a graph-level representation vector containing the global feature representation of the entire device, which is used to guide the subsequent generation of context information.

[0050] Furthermore, the rule constraint fusion module 22 dynamically integrates the preprocessed standard electronic equipment dismantling rules, dismantling targets, and the current state of the power equipment to be dismantled, outputting conditional constraint guidance, including: Iterate through each rule in the preprocessed standard electronic material and equipment dismantling rules, determine whether it matches the initial state and dismantling target of the power material and equipment to be dismantled, and add the matching rules to the conditional rule list. Based on the constraints of the corresponding rules in the conditional rule list, a constraint representation matrix is ​​established; The conditional rule list and the constraint representation matrix are output as constraint guidance for conditionalization.

[0051] Specifically, the input to the rule constraint fusion module 22 is the preprocessed standard electronic equipment dismantling rules, the dismantling target, and the current state of the power equipment to be dismantled, i.e., the initial state representation, the target vector, and the encoded safety rules, sequence rules, operation rules, tool rules, and quality rules and their constraints. For each rule in the preprocessed standard electronic equipment dismantling rules, it is determined whether it is compatible with the initial state and dismantling target of the power equipment to be dismantled, that is, whether it is applicable to the dismantling of the components of the power equipment to be dismantled, and whether it can achieve the dismantling target. Rules that meet the conditions are added to the conditional rule list. For each rule in the conditional rule list, its safety rule constraints, sequence rule constraints, operation rule constraints, tool rule constraints, and quality rule constraints are obtained, and a constraint representation matrix is ​​established. The conditional rule list and the constraint representation matrix are used as the conditional constraint guidance output.

[0052] Furthermore, the Transformer sequence decoding module, based on the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and conditional constraints, guides autoregression to generate the optimal dismantling scheme, including: Iteratively perform the following operations until the stopping condition is met: Based on the list of conditional rules in the conditional constraint guidance, calculate the set of executable operations; for each operation in the set of executable operations, calculate the comprehensive score based on the current state of the power equipment to be dismantled, the target vector, the node embedding matrix, the graph-level representation vector, the constraint representation matrix in the conditional constraint guidance, and the optimization target; and select the operation with the highest comprehensive score to output and update the current state of the power equipment to be dismantled. After the stopping condition is met, output the sequence of each disassembled component and its operation to obtain the optimal disassembly scheme.

[0053] Specifically, the input to the Transformer sequence decoding module is the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and the conditional constraint guidance. Initially, the initial state of the power equipment to be dismantled is the current state. Based on the list of conditional rules in the conditional constraint guidance, an executable operation set is calculated. For each operation in the executable operation set, a query vector is calculated based on the current state of the power equipment to be dismantled and the target vector. A key-value vector is calculated based on the node embedding matrix, the graph-level representation vector, and the constraint representation matrix. An attention score is calculated based on the query vector and the key-value vector. A multi-objective score is calculated by weighted summation based on the number of operation steps, the degree of completeness, and the number of tool switching times under the corresponding operation. The attention score and the weighted score are added together to obtain a comprehensive score. The operation with the highest comprehensive score is selected as the output, the current state of the power equipment to be dismantled is updated, and the next iteration begins.

[0054] When the disassembly target is reached, the iteration stops, and the sequence of each disassembled component and its operation are output to obtain the optimal disassembly scheme.

[0055] refer to Figure 3 In some embodiments, a power equipment dismantling control device is provided, comprising: The knowledge base establishment module 201 is used to collect basic information data of standard power materials and equipment and construct a knowledge graph of the structure of standard electronic materials and equipment to establish a knowledge base. The rule base establishment module 202 is used to construct standard electronic material and equipment disassembly rules based on knowledge in the field of power materials and equipment, and to establish a disassembly rule base; The acquisition module 203 is used to acquire the corresponding standard electronic equipment structure knowledge graph and standard electronic equipment dismantling rules from the knowledge base and dismantling rule base respectively, based on the identity information of the power equipment to be dismantled. Model building module 204 is used to build and train a prediction model based on graph neural coding and Transformer decoding. Analysis module 205 is used to input the dismantling target, the initial state of the power materials and equipment to be dismantled, the corresponding standard electronic materials and equipment structure knowledge graph, and the standard electronic materials and equipment dismantling rules into the trained prediction model, and output the optimal dismantling scheme.

[0056] Furthermore, the basic information data for standard power equipment and materials is CAD drawing data or BOM list; The knowledge base building module 201 constructs a standard electronic materials and equipment structure knowledge graph, including: Identification of components and assembly relationships is performed based on the aforementioned standard power equipment and materials basic information data; Extract the constraints in the assembly relationship and identify the assembly features; Using standard power equipment as the root node, add all component nodes, create relationship edges for each assembly relationship, connect the corresponding component nodes, add constraints and assembly features to each relationship edge, and obtain a knowledge graph of the structure of standard electronic equipment.

[0057] Furthermore, the rule base establishment module 202, based on knowledge of the power equipment and materials domain, constructs standard electronic equipment and materials disassembly rules, including: The dismantling operation steps are extracted from the text of knowledge in the field of power materials and equipment. Each dismantling operation step is taken as the operation root node. The safety rules, sequence rules, operation rules, tool rules and quality rules and their constraints under each operation root node are obtained, structured and added to the operation root node.

[0058] Furthermore, the prediction model includes an input layer, a processing layer, and an output layer; the processing layer includes a graph neural network encoding module, a rule constraint fusion module, and a Transformer sequence decoding module; The input layer is used to input the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules, and performs preprocessing. The graph neural network encoding module is used to perform structural encoding on the preprocessed standard electronic equipment structure knowledge graph and output a node embedding matrix and a graph-level representation vector. The rule constraint fusion module dynamically integrates the preprocessed standard electronic equipment dismantling rules, the dismantling target, and the initial state of the power equipment to be dismantled, and outputs conditional constraint guidance. The Transformer sequence decoding module generates the optimal dismantling scheme through autoregression based on the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and the conditional constraint guidance. The output layer is used to output the optimal dismantling scheme.

[0059] Furthermore, the input layer preprocesses the dismantling target, the initial state of the power equipment to be dismantled, the corresponding standard electronic equipment structure knowledge graph, and the standard electronic equipment dismantling rules, including: Based on the standard electronic materials and equipment structure knowledge graph, component node features and edge features are extracted to generate a component feature matrix; The initial state of the power materials and equipment to be dismantled is encoded to generate an initial state description. The decomposed target is analyzed and converted into vector form to obtain the target vector; Safety rules, sequence rules, operation rules, tool rules, and quality rules and their constraints are extracted from the standard electronic equipment dismantling rules.

[0060] Furthermore, the graph neural network encoding module performs structural encoding on the preprocessed standard electronic material equipment structure knowledge graph, outputting a node embedding matrix and a graph-level representation vector, including: The graph neural network encoding module aggregates the component node information in the component feature matrix based on multi-layer graph convolution, generates a multi-dimensional vector representation containing each component node, obtains the node embedding matrix, perceives the connection relationship between component nodes, captures local and global information, and generates a graph-level representation vector containing global feature representation of the device.

[0061] Furthermore, the rule constraint fusion module dynamically integrates the preprocessed standard electronic equipment dismantling rules, dismantling targets, and the current state of the power equipment to be dismantled, outputting conditional constraint guidance, including: Iterate through each rule in the preprocessed standard electronic material and equipment dismantling rules, determine whether it matches the initial state and dismantling target of the power material and equipment to be dismantled, and add the matching rules to the conditional rule list. Based on the constraints of the corresponding rules in the conditional rule list, a constraint representation matrix is ​​established; The conditional rule list and the constraint representation matrix are output as constraint guidance for conditionalization.

[0062] Furthermore, the Transformer sequence decoding module, based on the preprocessed dismantling target, the initial state of the power equipment to be dismantled, the node embedding matrix, the graph-level representation vector, and conditional constraints, guides autoregression to generate the optimal dismantling scheme, including: Iteratively perform the following operations until the stopping condition is met: Based on the list of conditional rules in the conditional constraint guidance, calculate the set of executable operations; for each operation in the set of executable operations, calculate the comprehensive score based on the current state of the power equipment to be dismantled, the target vector, the node embedding matrix, the graph-level representation vector, the constraint representation matrix in the conditional constraint guidance, and the optimization target; and select the operation with the highest comprehensive score to output and update the current state of the power equipment to be dismantled. After the stopping condition is met, output the sequence of each disassembled component and its operation to obtain the optimal disassembly scheme.

[0063] Furthermore, the objective function of the prediction model includes minimizing the number of operation steps, maximizing component integrity, and minimizing the number of tool switching operations.

[0064] The power equipment dismantling control method and apparatus provided in the above embodiments have at least the following beneficial effects: (1) A prediction model based on graph neural network coding and Transformer decoding is adopted to minimize the number of operation steps, maximize the integrity of components and minimize the number of tool switching. The generated optimal disassembly scheme can ensure the accuracy of disassembly steps while effectively improving the production efficiency of electronic material equipment disassembly. (2) The prediction model based on graph neural coding and Transformer decoding can effectively capture the spatial relationship between components, understand the assembly topology, identify physical constraints and perform global optimization, thus ensuring the accuracy of the disassembly steps; (3) The construction of knowledge base and rule base is applicable to the dismantling of different power materials and equipment, and has stronger universality.

[0065] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A power asset equipment disassembly control method, characterized by, The method comprises the following steps: Collecting standard power material equipment basic information data and constructing a standard electronic material equipment structure knowledge graph, and establishing a knowledge base; Based on the knowledge of the field of power material equipment, construct standard electronic material equipment disassembly rules, and establish a disassembly rule base; According to the identity information of the power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and standard electronic material equipment disassembly rules are obtained from the knowledge base and disassembly rule base respectively; Construct a prediction model based on graph neural coding and Transformer decoding and train it; The initial state of the disassembly target, the power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and standard electronic material equipment disassembly rules are input into the trained prediction model, and the optimal disassembly scheme is output.

2. The method of claim 1, wherein, The standard power material equipment basic information data is CAD drawing data or BOM list; Construct a standard electronic material equipment structure knowledge graph, including: Identify the parts and assembly relationship based on the standard power material equipment basic information data; Extract the constraint conditions in the assembly relationship and identify the assembly features; Take the standard power material equipment as the root node, add all part nodes, and create relationship edges for each assembly relationship, connect the corresponding part nodes, and add constraint conditions and assembly features to each relationship edge to obtain the standard electronic material equipment structure knowledge graph.

3. The method of claim 1, wherein, Based on the knowledge of the field of power material equipment, construct standard electronic material equipment disassembly rules, including: Extract the disassembly operation steps from the text of the power material equipment field knowledge, and take each disassembly operation step as an operation root node, obtain the safety rules, sequence rules, operation rules, tool rules and quality rules and their constraints under each operation root node, and perform structured processing and add them to the operation root node.

4. The method of claim 1, wherein, The prediction model includes an input layer, a processing layer and an output layer; the processing layer includes a graph neural network coding module, a rule constraint fusion module and a Transformer sequence decoding module; The input layer is used for inputting the disassembly target, the initial state of the power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and standard electronic material equipment disassembly rules and performing preprocessing; the graph neural network coding module is used for structural coding of the preprocessed standard electronic material equipment structure knowledge graph, outputting node embedding matrix and graph-level representation vector; the rule constraint fusion module is based on the preprocessed standard electronic material equipment disassembly rules, disassembly target and initial state of the power material equipment to be disassembled for dynamic integration, and outputs conditional constraint guidance; The Transformer sequence decoding module generates the optimal disassembly scheme based on the preprocessed disassembly target, initial state of the power material equipment to be disassembled, node embedding matrix, graph-level representation vector and conditional constraint guidance, and the output layer is used for outputting the optimal disassembly scheme.

5. The method of claim 4, wherein, The input layer preprocesses the disassembly target, the initial state of the power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and standard electronic material equipment disassembly rules, including: Based on the standard electronic material equipment structure knowledge graph, component node features and edge features are extracted, and a component feature matrix is generated. The initial state of the to-be-disassembled power material equipment is encoded to generate an initial state description. The disassembly target is parsed and converted into a vector form to obtain a target vector. From the standard electronic material equipment disassembly rules, safety rules, sequence rules, operation rules, tool rules, and quality rules, as well as their constraints, are extracted.

6. The method of claim 5, wherein, The graph neural network encoding module encodes the preprocessed standard electronic material equipment structure knowledge graph, outputs a node embedding matrix and a graph-level representation vector, including: The graph neural network encoding module aggregates component node information in the component feature matrix based on multi-layer graph convolution, generates a multi-dimensional vector representation of each component node, obtains the node embedding matrix, and perceives the connection relationship between component nodes, capturing local and global information to generate a graph-level representation vector containing the global feature representation of the equipment.

7. The method of claim 5, wherein, The rule constraint fusion module dynamically integrates based on the preprocessed standard electronic material equipment disassembly rules, the disassembly target, and the current state of the to-be-disassembled power material equipment, and outputs a conditional constraint guide, including: Each rule in the preprocessed standard electronic material equipment disassembly rules is traversed to determine whether it is compatible with the initial state and disassembly target of the to-be-disassembled power material equipment, and the compatible rules are added to the conditional rule list; Based on the constraints of the corresponding rules in the conditional rule list, a constraint representation matrix is established; The conditional rule list and the constraint representation matrix are output as a conditional constraint guide.

8. The method of claim 5, wherein, The Transformer sequence decoding module generates an optimal disassembly scheme based on the preprocessed disassembly target, the initial state of the to-be-disassembled power material equipment, the node embedding matrix, the graph-level representation vector, and the conditional constraint guide, including: Iteratively perform the following operations until the stopping condition is met: based on the conditional rule list in the conditional constraint guide, calculate the executable operation set, for each operation in the executable operation set, calculate the comprehensive score based on the current state of the to-be-disassembled power material equipment, the target vector, the node embedding matrix, the graph-level representation vector, the constraint representation matrix in the conditional constraint guide, and the optimization target, and select the operation with the highest comprehensive score as the output, and update the current state of the to-be-disassembled power material equipment; After the stopping condition is met, output each disassembly component sequence and its operation to obtain the optimal disassembly scheme.

9. The method of claim 1, wherein, The objective function of the prediction model includes minimizing the number of operation steps, maximizing component integrity, and minimizing the number of tool switches.

10. An electric power asset equipment disassembly control device characterized by comprising: It includes: A knowledge base establishment module for collecting standard power material equipment basic information data and constructing a standard electronic material equipment structure knowledge graph to establish a knowledge base; A rule base establishment module for constructing standard electronic material equipment disassembly rules based on power material equipment domain knowledge and establishing a disassembly rule base; An acquisition module is configured to acquire, according to the identity information of the power material equipment to be disassembled, a corresponding standard electronic material equipment structure knowledge graph and a standard electronic material equipment disassembly rule from the knowledge base and the disassembly rule base respectively; A model construction module is configured to construct a prediction model based on graph neural coding and based on Transformer decoding and to train the prediction model; An analysis module is configured to input the disassembly target, the initial state of the power material equipment to be disassembled, the corresponding standard electronic material equipment structure knowledge graph and the standard electronic material equipment disassembly rule into the trained prediction model, and to output an optimal disassembly scheme.

Images