A visual tuning calculation system and method thereof

Abstract

The present invention discloses a visual setting calculation system and method thereof, which relate to the technical field of power grid setting calculation, and include the following steps: obtaining the operating data of the target system, extracting feature data, and constructing a knowledge graph of entities and relationships, wherein the feature data includes control data; constructing a system structure diagram, and generating a semantic enhancement graph in combination with the semantic information in the knowledge graph; inputting the semantic enhancement graph into a pre-trained graph neural network model, combining the prior rules of the knowledge graph, performing node feature aggregation and edge weight learning, and outputting a sensitivity score of each control data node to a performance indicator; constructing boundary constraints based on the sensitivity score, and generating control data setting candidate solutions based on a CSP solver; performing parameter response simulation verification on the control data setting candidate solutions, and screening out the optimal control data setting candidate solutions; the present application can achieve accurate sensitivity scoring by integrating the knowledge graph, the semantic enhancement graph and the graph neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid setting calculation technology, and more particularly, to a visual setting calculation system and method thereof. Background Art

[0002] Power grid systems typically consist of multiple control layers and complex subsystems, involving a vast array of devices, control parameters, and measurement data. The operational data from these systems encompasses multiple sources, including system status, device behavior, control instructions, and sensor information. The relationships between these data are often highly complex. Therefore, efficiently processing and analyzing this multidimensional data to extract the key features that significantly impact system performance is a pressing challenge.

[0003] For example, the invention patent with announcement number CN116599118A discloses a grid setting calculation method and device, which relates to the field of power system technology. The method includes: obtaining at least one new energy power source corresponding to the grid, and determining the new energy setting calculation principle based on the power data of the new energy power source, wherein the power data includes the equipment parameters and configuration parameters of the new energy power source; constructing a new energy setting calculation model according to the new energy grid setting calculation principle; when the new energy power source is connected to the grid and the grid fails, the grid is calculated using the new energy setting calculation model to obtain the short-circuit current of the grid; thus, the new energy setting calculation model determined based on the power data of the new energy power source accurately calculates the short-circuit current when the grid fails, thereby realizing accurate setting calculation of the new energy power source connected to the grid, and improving the accuracy of the setting calculation of the new energy power source connected to the grid.

[0004] The above disclosed technical solutions have at least the following technical problems:

[0005] Traditional methods rely on manually established control models or system equations. When the model is inaccurate or the system is complex, it is easy to cause parameter adjustment failure. There is a lack of modeling capabilities for complex semantic relationships and control logic in the system structure. In addition, the response relationship between control parameters and system performance usually relies on experience or linear estimation, which makes it difficult to reflect the impact intensity between nonlinear or high-dimensional features.

[0006] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a visual tuning calculation system and method thereof, which realizes intelligent tuning of control parameters by integrating knowledge graphs, semantically enhanced graphs and graph neural networks, and solves the problem of inaccurate sensitivity scores.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A visual tuning calculation method includes: obtaining operating data of a target system and extracting feature data to construct a knowledge graph of entities and relationships, wherein the feature data includes control data; constructing a system structure diagram and generating a semantically enhanced graph by combining semantic information in the knowledge graph; inputting the semantically enhanced graph into a pre-trained graph neural network model, performing node feature aggregation and edge weight learning in combination with prior rules of the knowledge graph, and outputting a sensitivity score of each control data node to a performance indicator; constructing boundary constraints based on the sensitivity score, and generating candidate control data tuning solutions based on a CSP solver; and performing parameter response simulation verification on the candidate control data tuning solutions to screen out the optimal candidate control data tuning solution.

[0010] In a preferred embodiment, the operation data of the target system is obtained, and feature data is extracted to construct a knowledge graph of entities and relationships, specifically as follows: the original operation data is obtained from the data acquisition layer of the target system, and the collected original data is structured to extract key feature data; based on the extracted feature data, the entity nodes of the knowledge graph are constructed; based on the system flow chart, process logic and control rules, the semantic relationships between entities are identified and constructed to form connecting edges between entities; the entities and relationships are organized into a structured knowledge graph in the form of RDF triples.

[0011] In a preferred embodiment, the system structure diagram is constructed and combined with the semantic information in the knowledge graph to generate a semantic enhancement graph, specifically as follows: the physical topology information and control configuration of the target system are obtained to construct an initial system structure diagram; based on the physical connection, control logic and signal flow between devices, the edge type between nodes is added to the structure diagram to obtain a system structure diagram; and the obtained semantic information is multimodally processed to obtain a semantic vector representation, and a similarity analysis is performed between the nodes in the structure diagram to establish a mapping relationship between the structure diagram nodes and the semantic information; according to the mapping relationship, semantic nodes are embedded in the system structure diagram, and semantic connections are established between them and existing structural nodes to form a new semantic node set and semantic edge set to obtain a semantic structure diagram; feature fusion is performed on each node of the semantic structure diagram, and its original structural attributes, control parameter attributes and their corresponding semantic embedding vectors are uniformly encoded to form a composite node feature vector; the composite node feature vector is used as a node representation, combined with structural edges and semantic edges to construct a semantic enhancement graph.

[0012] In a preferred embodiment, the semantically enhanced graph is input into a pre-trained graph neural network model, and node feature aggregation and edge weight learning are performed in combination with the prior rules of the knowledge graph to output the sensitivity score of each control data node to the performance indicator, specifically as follows: structurally mappable rule triples are extracted from the semantically enhanced graph, a logical prior tensor is constructed, and the edge weights in the graph neural network are initialized to form an initial graph neural network; according to the composite features of each node in the semantically enhanced graph, the node embedding vector is initialized, and the semantic and structural information between nodes is aggregated through a multi-round propagation mechanism in the initial graph neural network to update the node embedding vector; the control data node is identified by the node type label and the position of the node in the structural graph; a regression mapping function is established through the performance data in the historical operation data and the control data node embedding vector, and the response intensity between the control data node feature and the performance data is output; for each control data node, a sensitivity score is output according to its node embedding vector and the regression mapping function.

[0013] In a preferred embodiment, the method extracts structurally mappable rule triples from the semantically enhanced graph and constructs a logical prior tensor, specifically as follows: rules related to the control system logic are extracted from the semantically enhanced graph, expressed in triple form, to form an initial rule set; a semantic embedding method is used to map each triple into a vector, and the rule strength is output; the triple set is screened based on whether the rule entity exists in the semantically enhanced graph node set, and further screened by checking the node pairs in the semantically enhanced graph; based on the rule triple set, the rules are projected into the edge space of the graph to construct a three-dimensional logical prior tensor.

[0014] In a preferred embodiment, the boundary constraint conditions are constructed based on the sensitivity score as follows: the sensitivity score of each control data node is obtained, and according to the sorting of the sensitivity scores, the control data nodes with sensitivity scores higher than the set threshold are screened out as high-priority adjustment objects; cluster analysis is performed on the screened control data nodes based on the clustering algorithm, and the nodes are divided into several groups according to the sensitivity score of each node; for each group of control data nodes, dynamic boundary constraint conditions are generated according to their sensitivity scores and historical data.

[0015] In a preferred embodiment, the control data tuning candidate solutions are generated based on the CSP solver as follows: the search space of the control data nodes is initialized by a backtracking algorithm, and the control data node combinations are traversed based on the boundary constraints of each control data node; at each step of the backtracking algorithm, a heuristic scoring function is combined to assign a heuristic score to the possible values ​​of each control data node; during the backtracking search process, control data node combinations with high potential are preferentially selected based on the heuristic score; and multiple candidate solutions are generated through the backtracking algorithm and heuristic search.

[0016] In a preferred embodiment, the parameter response simulation verification of the control data tuning candidate solutions is performed to screen out the optimal control data tuning candidate solutions, specifically as follows: each candidate solution is input into a simulation environment or actual system test for verification, and each candidate solution is evaluated to obtain evaluation data; the evaluation data is compared and analyzed with historical data to generate the optimization strength of the candidate solution on the system performance, and initial tuning candidate solutions are screened out based on the optimization strength; the objective function values ​​of the initial tuning candidate solutions are extracted, and the objective function values ​​are sorted based on Pareto frontier analysis to screen out the best tuning candidate solutions.

[0017] A visual tuning calculation system includes a data acquisition module, a semantic enhancement graph construction module, a sensitivity scoring module, a tuning candidate solution generation module and a screening module, and there are connections between the modules; the data acquisition module is used to obtain the operating data of the target system, extract feature data, and construct a knowledge graph of entities and relationships, wherein the feature data includes control data; the semantic enhancement graph construction module is used to construct a system structure diagram and generate a semantic enhancement graph in combination with the semantic information in the knowledge graph; the sensitivity scoring module is used to input the semantic enhancement graph into a pre-trained graph neural network model, combine the prior rules of the knowledge graph, perform node feature aggregation and edge weight learning, and output the sensitivity score of each control data node to the performance indicator; the tuning candidate solution generation module is used to construct boundary constraints according to the sensitivity score and generate control data tuning candidate solutions based on a CSP solver; the screening module is used to perform parameter response simulation verification on the control data tuning candidate solutions and screen out the optimal control data tuning candidate solution.

[0018] The technical effects and advantages of the visual tuning calculation system and method of the present invention are as follows:

[0019] 1. This invention introduces a knowledge graph construction mechanism to organize the target system's control data, device information, and logical rules into RDF triples using structured entities and relationships, forming a unified knowledge representation framework. By integrating system topology, control parameters, and semantic information, a semantically augmented graph (SAG) consisting of both structural and semantic nodes is generated, significantly enhancing the model's ability to express system control logic, signal flow, and device interactions. This modeling approach provides a strong structural foundation for subsequent data aggregation and optimization.

[0020] 2. This invention combines pre-trained graph neural networks with logical rules in knowledge graphs, introduces logical prior tensors to initialize edge weights in the network, and constructs a graph neural model that emphasizes both structural perception and semantic cognition. Under the multi-round propagation mechanism of the graph neural network, it can fully aggregate the control logic, semantic similarity, and topological constraints between nodes to achieve a deep representation of control data nodes. Furthermore, the regression mapping function is trained through historical operating data to output the sensitivity score of each control data node to the system performance indicator, achieving a quantitative analysis of the degree of influence of the control data with high accuracy and interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The figure is a flow chart of a visual setting calculation method of the present invention.

[0022] Figure 2 This is a structural diagram of a visual tuning calculation system of the present invention. DETAILED DESCRIPTION

[0023] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0024] Example 1, Figure 1 The present invention provides a visual setting calculation method, including:

[0025] S1, obtain the operating data of the target system, extract feature data, and build a knowledge graph of entities and relationships. The feature data includes control data;

[0026] In this embodiment, the operating data of the target system is obtained, and feature data is extracted to construct a knowledge graph of entities and relationships, as follows:

[0027] Obtain raw operating data from the target system's data acquisition layer. The data sources include, but are not limited to: control data (such as process variables and control instructions recorded by PLC and DCS systems), sensor data (such as physical quantities such as temperature, pressure, and flow), equipment status data (such as start / stop status, current load, fault codes, etc.), and historical operation logs and alarm records. This data can be integrated and collected through industrial communication protocols (such as Modbus, OPC UA, Profibus, etc.) or IoT platforms.

[0028] Structural processing is performed on the collected raw data to extract key feature data for knowledge modeling. Feature screening and standardization are performed through statistical analysis (such as sliding window mean / variance, FFT spectrum, DTW similarity, etc.) or machine learning methods (such as feature selection algorithms). Specifically, the following are performed:

[0029] Control data extraction: extract control parameters (such as PID Kp, Ki, Kd), setpoint, control mode (automatic / manual), control variables, etc.;

[0030] Performance index extraction: such as response time, steady-state error, overshoot rate, energy consumption index, etc.;

[0031] Operational feature extraction: such as system operating condition labels, event timestamps, frequent change points, etc.;

[0032] Based on the extracted feature data, the entity nodes of the knowledge graph are constructed, and each entity is accompanied by corresponding attributes, such as current value, unit, timestamp, source module, etc. Entity types include but are not limited to: control parameter entities (such as Kp, Ki, etc.), equipment entities (such as fans, motors, etc.), measurement indicator entities (such as outlet temperature, system flow, system pressure), performance indicator entities (such as system stability, response time, energy efficiency) and operation behavior entities (such as valve opening operation, controller switching, etc.);

[0033] Based on the system flow chart, process logic, and control rules, identify and construct semantic relationships between entities, forming connecting edges between entities. These semantic relationship types include, but are not limited to, control action relationships (the effect of control parameters on equipment, such as "PID.Kp" → "cooling pump"), physical causal relationships (the impact of equipment on performance indicators, such as "cooling pump" → "outlet temperature"), and constraint relationships (parameters are subject to upper and lower limits or other parameter states).

[0034] Entities and relationships are organized into a structured knowledge graph in the form of RDF triples.

[0035] S2, constructs a system structure diagram and combines it with the semantic information in the knowledge graph to generate a semantic enhancement graph;

[0036] In this embodiment, a system structure diagram is constructed and combined with the semantic information in the knowledge graph to generate a semantic enhancement graph, as follows:

[0037] Obtain the physical topology information and control configuration of the target system, and construct an initial system structure diagram including control device nodes, control parameter nodes, and sensor point nodes;

[0038] Based on the physical connections, control logic, and signal flow between devices, edge types between nodes are added to the structure diagram to obtain a complete system structure diagram. The edge types include control edges from parameters to devices and feedback edges from sensors to controllers.

[0039] Acquire various semantic information of the target system, including system design text, operating procedures, fault records, expert operation logs, equipment images, and alarm video screenshots;

[0040] Perform multimodal processing on semantic information to obtain the corresponding semantic vector representation, including:

[0041] Semantically encode text data using a pre-trained language model (such as BERT) to obtain text embedding vectors related to control intent.

[0042] Use image-text models (e.g., CLIP, BLIP-2) to convert image data into image embedding vectors that are semantically aligned with the control parameters;

[0043] Based on the similarity analysis between the semantic vector and the nodes in the structure graph, a mapping relationship between the structure graph nodes and the semantic information is established to form node semantic enhancement mapping and edge semantic enhancement mapping to support subsequent semantic injection operations;

[0044] According to the mapping relationship, semantic nodes are introduced into the system structure diagram, and semantic connections are established between them and existing structure nodes to form a new semantic node set and semantic edge set, thus obtaining a semantic structure diagram.

[0045] The features of each node in the semantic structure graph are fused, and its original structural attributes, control parameter attributes and their corresponding semantic embedding vectors are uniformly encoded to form a composite node feature vector, thereby enhancing the information expression capability of the graph structure.

[0046] The composite node feature vector is used as the node representation, and the structural edge and semantic edge are combined to construct a semantic enhancement graph containing structural information and semantic information. ,in is a node set (including control devices, control parameters, semantic labels, etc.), is the edge set (including structural edges and semantic edges), is a composite node feature vector, and the structural edge is an edge of the system structure graph.

[0047] S3: Input the semantically enhanced graph into the pre-trained graph neural network model, combine it with the prior rules of the knowledge graph, perform node feature aggregation and edge weight learning, and output the sensitivity score of each control data node to the performance indicator;

[0048] In this embodiment, the semantic enhancement graph is input into the pre-trained graph neural network model. Combined with the prior rules of the knowledge graph, node feature aggregation and edge weight learning are performed to output the sensitivity score of each control data node to the performance indicator, as follows:

[0049] Extract structurally mappable regular triples from the semantically enhanced graph, construct a logical prior tensor, and use the logical prior tensor to initialize the edge weights in the graph neural network (for example, using the logical prior tensor as a bias term or initial value for the attention coefficient) to obtain the initial graph neural network.

[0050] Obtain the composite features of each node in the semantic enhancement graph, input the composite node feature vector into an embedding mapping layer (such as a multi-layer perceptron (MLP) or linear transformation layer), compress the high-dimensional features into a low-dimensional embedding vector, and use it as the initial embedding of the node to initialize the node embedding vector and use it as the input representation of the graph neural network. The composite features include structural attributes, control parameters, and semantic vectors.

[0051] In the initial graph neural network, the semantic and structural information between nodes is aggregated through a multi-round propagation mechanism, and the node embedding vector is dynamically updated;

[0052] Identify control data nodes by their node type label (e.g., "Control Parameter" label) or their position in the structure diagram (e.g., "Input" node);

[0053] By embedding the performance data (such as energy efficiency and system stability) in the historical operation data with the control data node embedding vector, a regression mapping function is established to output the response strength between the control data node characteristics and the performance data;

[0054] For each control data node, based on its node embedding vector and regression mapping function, the partial derivative of the node feature change to the performance data change is output as the sensitivity score.

[0055] The sensitivity scores are as follows:

[0056] ;

[0057] Where: Representative Node The sensitivity score of is the regression mapping function, is a node characteristic changes, is the regression mapping function Embed vector for control node The partial derivative of represents the impact of node feature changes on system performance.

[0058] The structure-mappable regular triples are extracted from the semantic enhancement graph to construct the logical prior tensor, as follows:

[0059] Extract rules related to control system logic from semantically enhanced graphs and express them in triple form ,in, Represents a rule antecedent entity (such as a control parameter node), Indicates logical relationships (such as "influence", "control", "dependence"), Represents rule consequence entities (such as device behavior, performance status, etc.), forming the initial rule set;

[0060] For each triple, a semantic embedding method (such as TransE, ComplEx or embedding technology based on pre-trained language models) is used to map the triple into a vector , and output the rule strength, which is used to measure the confidence of the causal relationship in the triple, where express dimensional real space, that is, the embedding representation of each entity and relationship is a -dimensional real vector;

[0061] According to the rule entity and Whether it exists in the semantic enhancement graph node set, filter out the set of triples that can apply the rule in the graph ;

[0062] By checking the node pairs in the semantic enhancement graph, a set of rule triples that match the current graph structure is screened out. That is, only the rules between the node pairs that exist in the semantic enhancement graph will be used.

[0063] For every pair of nodes in the graph If there is a rule Then record its regularity strength into the subsequent prior tensor;

[0064] According to the set of rule triples, the rules are projected into the edge space of the graph to construct a three-dimensional logical prior tensor ,in, is the number of nodes in the graph, The rule type number (for example, "control", "dependency", "impact").

[0065] The formula for calculating rule strength is as follows:

[0066] ;

[0067] Where: is the head entity, Represents the relationship between the head entity and the tail entity, such as measurement and control, is the tail entity, is the embedding vector of the head entity, is the embedding vector of the relation, is the embedding vector of the tail entity, is the rule strength.

[0068] S4, constructs boundary constraints based on the sensitivity scores and generates candidate solutions for control data tuning based on the CSP solver;

[0069] In this embodiment, boundary constraints are constructed based on the sensitivity scores, and candidate solutions for control data tuning are generated based on the CSP solver, as follows:

[0070] Obtain the sensitivity score of each control data node, and based on the ranking of the sensitivity scores, select the control data nodes with a sensitivity score higher than the set threshold as high-priority adjustment targets;

[0071] Perform cluster analysis on the selected control data nodes based on clustering algorithms (such as K-Means or DBSCAN algorithms), and divide the nodes into several groups according to the sensitivity score of each node;

[0072] For each group of control data nodes, dynamic boundary constraints are generated based on their sensitivity scores and historical data. For highly sensitive control data nodes, the boundary constraint range is expanded to provide a larger adjustment space, thereby effectively optimizing system performance. For low-sensitivity control data nodes, the boundary constraint range is contracted to prevent excessive impact on system performance and improve stability.

[0073] In this embodiment, the control data tuning candidate solution is generated based on the CSP solver, as follows:

[0074] Initialize the search space of control data nodes through the backtracking algorithm, and based on the boundary constraints of each control data node, traverse all control data node combinations that meet the boundary constraints;

[0075] At each step of the backtracking algorithm, a heuristic scoring function is used to assign a heuristic score to each possible value of each control data node. The heuristic score is calculated based on the sensitivity score of the control data node, the potential optimization space (representing the adjustment space or control range of each control data node), and the optimization potential of the objective function (representing the strength of the control data node in optimizing the objective function, which includes energy efficiency, response speed, and stability during system operation). The heuristic score is calculated using a weighted average method.

[0076] During the backtracking search process, high-potential control data node combinations are prioritized based on the heuristic score to accelerate the search process and avoid traversing invalid solutions;

[0077] A plurality of candidate solutions are generated through backtracking algorithm and heuristic search, which satisfy the boundary constraints and have the potential to optimize the objective function.

[0078] The search space of the control data nodes is initialized by the backtracking algorithm, and based on the boundary constraints of each control data node, all combinations of control data nodes that meet the boundary constraints are traversed, as follows:

[0079] Starting from the initial node, the possible values ​​of each control data node are selected in turn. For each node, a possible value interval is generated, and the backtracking algorithm starts from the first node and gradually selects the value of the node;

[0080] After each selection, check whether the currently selected control parameters meet the existing constraints;

[0081] If the constraints are not met, backtrack immediately and try other possible values;

[0082] If the constraint is satisfied, continue searching downwards and select the value of the next control data node.

[0083] S5, performing parameter response simulation verification on the control data tuning candidate solutions to screen out the optimal control data tuning candidate solutions;

[0084] In this embodiment, parameter response simulation verification is performed on the control data tuning candidate solutions to screen out the optimal control data tuning candidate solutions, as follows:

[0085] Each candidate solution is input into a simulation environment or actual system test for verification, and each candidate solution is evaluated to obtain evaluation data. The evaluation data includes performance evaluation (whether the candidate solution can effectively improve system performance indicators such as energy efficiency, stability, and throughput) and constraint verification (whether the candidate solution satisfies all constraints, including boundary constraints and interaction constraints).

[0086] Compare and analyze the evaluation data with historical data to generate the optimization strength of candidate solutions on system performance, and select the initial tuning candidate solutions based on the optimization strength;

[0087] The objective function values ​​of the initial tuning candidate solutions are extracted and ranked based on Pareto front analysis to screen out tuning candidate solutions with better trade-off effects, wherein the objective function values ​​include minimizing energy consumption and maximizing stability.

[0088] It should be noted that the optimization results of all candidate solutions are displayed using cutting-edge Pareto visualization technology, using two-dimensional scatter plots or three-dimensional charts. Each dot represents a candidate solution, and its position is mapped according to the value of objective functions such as energy consumption and stability. Different types of solutions are distinguished by color coding or shape, making it easy to visually identify the best-performing solution.

[0089] Furthermore, the performance of each candidate solution under different objectives can be visualized, such as the trade-off between energy consumption and stability. The performance of multiple candidate solutions under different objectives can be displayed using scatter plots, radar charts, and other methods.

[0090] Furthermore, color coding or marking can be used to show which candidate solutions satisfy all constraints and which solutions may have problems with not satisfying constraints, thereby helping to screen out legal and effective candidate solutions.

[0091] Example 2, Figure 2 The present invention provides a visual tuning calculation system, which includes a data acquisition module, a semantic enhancement graph construction module, a sensitivity scoring module, a tuning candidate solution generation module and a screening module, and there are connections between the modules.

[0092] The data acquisition module is used to obtain the operating data of the target system, extract feature data, and build a knowledge graph of entities and relationships. The feature data includes control data.

[0093] The semantic enhancement graph construction module is used to construct a system structure diagram and generate a semantic enhancement graph by combining the semantic information in the knowledge graph;

[0094] The sensitivity scoring module is used to input the semantic enhancement graph into the pre-trained graph neural network model, combine the prior rules of the knowledge graph, perform node feature aggregation and edge weight learning, and output the sensitivity score of each control data node to the performance indicator;

[0095] The tuning candidate solution generation module is used to construct boundary constraints based on sensitivity scores and generate control data tuning candidate solutions based on the CSP solver;

[0096] The screening module is used to perform parameter response simulation verification on the control data tuning candidate solutions and screen out the optimal control data tuning candidate solutions.

[0097] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0098] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0099] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0100] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0101] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A visual setting calculation method, characterized in that: The steps include: Obtaining the target system's operational data, extracting feature data, and constructing a knowledge graph of entities and relationships, wherein the feature data includes control data; Build a system structure diagram and combine it with the semantic information in the knowledge graph to generate a semantic enhancement graph; The semantically enhanced graph is input into the pre-trained graph neural network model. Combined with the prior rules of the knowledge graph, node feature aggregation and edge weight learning are performed to output the sensitivity score of each control data node to the performance indicator. Based on the sensitivity scores, boundary constraints are constructed and candidate solutions for control data tuning are generated based on the CSP solver. Perform parameter response simulation verification on candidate control data tuning solutions to select the optimal candidate control data tuning solutions; The semantic enhancement graph is input into the pre-trained graph neural network model, and combined with the prior rules of the knowledge graph, node feature aggregation and edge weight learning are performed to output the sensitivity score of each control data node to the performance indicator, as follows: Extract structurally mappable regular triples from the semantically enhanced graph, construct a logical prior tensor, and initialize the edge weights in the graph neural network to form an initial graph neural network; Initialize the node embedding vector based on the composite features of each node in the semantic enhancement graph, and aggregate the semantic and structural information between nodes through a multi-round propagation mechanism in the initial graph neural network to update the node embedding vector; Identify control data nodes by their node type labels and their positions in the structure graph; By embedding the performance data and control data nodes in the historical operation data into vectors, a regression mapping function is established to output the response strength between the control data node features and the performance data. For each control data node, a sensitivity score is output based on its node embedding vector and regression mapping function; The structure-mappable regular triples are extracted from the semantic enhancement graph to construct the logical prior tensor, as follows: Extract rules related to control system logic from the semantic enhancement graph and express them in triple form to form the initial rule set; For each triple, a semantic embedding method is used to map the triple into a vector and output the rule strength. The set of triples is filtered based on whether the rule entity exists in the semantic enhancement graph node set, and further filtered by checking the node pairs in the semantic enhancement graph; According to the set of rule triples, the rules are projected into the edge space of the graph to construct a three-dimensional logical prior tensor.

2. A visual setting calculation method according to claim 1, characterized in that: The operation data of the target system is obtained, and feature data is extracted to construct a knowledge graph of entities and relationships, as follows: Obtain original operation data from the data acquisition layer of the target system, perform structured processing on the collected original data, and extract key feature data; Construct entity nodes of the knowledge graph based on the extracted feature data; According to the system flow chart, process logic and control rules, the semantic relationship between entities is identified and constructed to form the connection edges between entities; Organize entities and relationships into a structured knowledge graph in the form of RDF triples.

3. A visual setting calculation method according to claim 2, characterized in that: The system structure diagram is constructed and combined with the semantic information in the knowledge graph to generate a semantic enhancement graph, as follows: Obtain the physical topology information and control configuration of the target system and build the initial system structure diagram; Based on the physical connections, control logic, and signal flow between devices, edge types between nodes are added to the structure diagram to obtain a system structure diagram; The obtained semantic information is processed multimodally to obtain a semantic vector representation, and similarity analysis is performed between the nodes in the structure diagram to establish a mapping relationship between the structure diagram nodes and the semantic information; According to the mapping relationship, semantic nodes are embedded in the system structure diagram, and semantic connections are established between them and existing structure nodes to form a new semantic node set and semantic edge set, thus obtaining a semantic structure diagram. The features of each node in the semantic structure graph are fused, and its original structural attributes, control parameter attributes and their corresponding semantic embedding vectors are uniformly encoded to form a composite node feature vector; The composite node feature vector is used as the node representation, and the structural edge and semantic edge are combined to construct a semantically enhanced graph.

4. A visual setting calculation method according to claim 3, characterized in that: According to the sensitivity score, the boundary constraints are constructed as follows: Obtain the sensitivity score of each control data node, and based on the ranking of the sensitivity scores, select the control data nodes with a sensitivity score higher than the set threshold as high-priority adjustment targets; Perform cluster analysis on the selected control data nodes based on the clustering algorithm, and divide the nodes into several groups according to the sensitivity score of each node; For each group of control data nodes, dynamic boundary constraints are generated based on their sensitivity scores and historical data.

5. A visual setting calculation method according to claim 4, characterized in that: The control data tuning candidate solution generated based on the CSP solver is as follows: Initialize the search space of control data nodes through the backtracking algorithm, and traverse the control data node combinations based on the boundary constraints of each control data node; At each step of the backtracking algorithm, a heuristic scoring function is combined to assign a heuristic score to the possible values ​​of each control data node; During the backtracking search process, high-potential control data node combinations are prioritized based on the heuristic score; Multiple candidate solutions are generated through backtracking algorithm and heuristic search.

6. A visual setting calculation method according to claim 5, characterized in that: The parameter response simulation verification of the control data tuning candidate solution is performed to screen out the optimal control data tuning candidate solution, as follows: Input each candidate solution into a simulation environment or actual system test for verification, and evaluate each candidate solution to obtain evaluation data; Compare and analyze the evaluation data with historical data to generate the optimization strength of candidate solutions on system performance, and select the initial tuning candidate solutions based on the optimization strength; The objective function values ​​of the initial tuning candidate solutions are extracted and sorted based on Pareto frontier analysis to select the best tuning candidate solutions.

7. A visual tuning calculation system according to any one of claims 1 to 6, characterized in that: It includes data acquisition module, semantic enhancement graph construction module, sensitivity scoring module, setting candidate solution generation module and screening module, and there are connections between modules; The data acquisition module is used to obtain the operating data of the target system, extract feature data, and build a knowledge graph of entities and relationships. The feature data includes control data. The semantic enhancement graph construction module is used to construct a system structure diagram and generate a semantic enhancement graph by combining the semantic information in the knowledge graph; The sensitivity scoring module is used to input the semantic enhancement graph into the pre-trained graph neural network model, combine the prior rules of the knowledge graph, perform node feature aggregation and edge weight learning, and output the sensitivity score of each control data node to the performance indicator; The tuning candidate solution generation module is used to construct boundary constraints based on sensitivity scores and generate control data tuning candidate solutions based on the CSP solver; The screening module is used to perform parameter response simulation verification on the control data tuning candidate solutions and screen out the optimal control data tuning candidate solutions.

Images