A method and system for generating a power plant gateway setting scheme

Abstract

This invention discloses a method and system for generating power plant gate setting schemes, relating to the field of power system energy metering and management technology. The method includes: obtaining wiring structure features based on the electrical wiring diagram of the target power plant; generating candidate gate setting schemes based on the wiring structure features and a gate setting rule base, wherein each candidate gate setting scheme includes at least one gate setting location; performing multi-scenario operating condition simulations on the candidate gate setting schemes based on a digital twin of the target power plant, and generating optimized gate setting schemes through reinforcement learning based on the simulation results; performing multi-objective comprehensive evaluation on the optimized gate setting schemes, and obtaining the optimal gate setting scheme based on the evaluation results. This application replaces manual experience-based point selection and subjective comparison by automating the generation of gate setting schemes, improving the objectivity and overall efficiency of the schemes, and solving the problems of poor standardization, low accuracy, and lack of quantitative evaluation standards in existing technologies where gate setting relies on manual experience.

Description

Technical Field

[0001] This application relates to the field of power system energy metering and management technology, specifically a method and system for generating power plant gate setting schemes. Background Technology

[0002] In the power industry's electricity metering management, power plant checkpoints are crucial nodes for electricity metering, trade settlement, and grid operation monitoring. The selection of power plant checkpoints requires a comprehensive assessment considering the power plant's main structure, primary electrical wiring method, equipment attributes, and actual operating conditions. Current power plant checkpoint selection schemes primarily rely on manual processes, including: reviewing historical checkpoint setup cases; manually determining the wiring pattern based on the power plant's primary wiring diagram; selecting checkpoint locations based on experience; manually analyzing current flow and power data under generator start-up and shutdown, switch start-up and shutdown conditions to demonstrate the rationality of the checkpoint location; manually developing methods for calculating the purchased and sold electricity at corresponding checkpoint locations; and finally, if multiple checkpoint setup schemes exist, only simple manual comparisons of some indicators can provide subjective selection suggestions. The lack of quantitative evaluation criteria for comparing multiple checkpoint schemes leads to a lack of scientific rigor and objectivity in the optimal selection. Therefore, there is an urgent need for a method to automatically identify wiring characteristics, intelligently recommend and verify checkpoint schemes, and quantitatively select the optimal scheme for power plant checkpoint setup. Summary of the Invention

[0003] The purpose of this application is to address the problems in existing technologies where gate setting relies on human experience, resulting in poor standardization, low accuracy, and a lack of quantitative evaluation criteria. A method and system for generating gate setting schemes in power plants is proposed. This method automatically generates candidate schemes based on wiring structure features and a rule base, combines digital twins and reinforcement learning for operational condition simulation and optimization, and outputs the optimal scheme through multi-objective comprehensive evaluation, achieving self-learning, self-evolution, and adaptive intelligent generation of gate setting schemes.

[0004] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a method for generating a power plant gate setting scheme, the method comprising: The wiring structure features are obtained based on the electrical wiring diagram of the target power plant; candidate gate schemes are generated based on the wiring structure features and the gate setting rule base, each candidate gate scheme containing at least one gate setting location; multi-scenario operating condition simulations are performed on the candidate gate schemes based on the digital twin of the target power plant, and an optimized gate scheme is generated through reinforcement learning based on the simulation results; the optimized gate scheme is evaluated comprehensively based on multiple objectives, and the optimal gate setting scheme is obtained based on the evaluation results.

[0005] This solution acquires the wiring structure characteristics of the target power plant, providing a standardized data foundation for subsequent solution generation. Candidate solutions are generated based on a gate setting rule base, avoiding the problems of poor standardization and strong subjectivity in manual point selection, ensuring that candidate solutions comply with basic power metering regulations and engineering specifications. Furthermore, a digital twin-based operating condition simulation and reinforcement learning mechanism is introduced. Candidate solutions are mapped to a digital twin model to construct a simulation environment, and the simulation results from multiple scenarios are used as feedback signals for reinforcement learning, driving iterative optimization of the candidate solutions. This solves the problems of traditional manual calculations failing to cover all operating conditions and lacking a closed-loop optimization mechanism, achieving a shift from static generation to dynamic evolution. Finally, multi-objective comprehensive evaluation of optimized gate solutions replaces traditional subjective manual comparison, achieving automated and intelligent generation of gate setting solutions, improving the objectivity, accuracy, and adaptability of the solutions.

[0006] Optionally, the step of obtaining wiring structure features based on the electrical wiring diagram of the target power plant includes: performing semantic recognition on the electrical wiring diagram of the target power plant to obtain power plant wiring data containing equipment type, connection relationship, wiring mode and node number; generating a topological adjacency matrix based on the power plant wiring data, which maps equipment nodes, electrical circuits and topological levels one-to-one; and obtaining wiring structure features based on the topological adjacency matrix, wherein the wiring structure features include at least the number of main components, power source type, number of generating units, number of busbars, number of outgoing lines and circuit connection features of the target power plant.

[0007] This solution uses semantic recognition to obtain power plant wiring data containing information such as equipment type and connection relationships from the wiring diagram. Based on this, a topology adjacency matrix is ​​generated, and key information such as the number of main components, power supply type, and circuit connection characteristics is extracted. This allows for accurate extraction of multi-dimensional wiring structure features, providing structured data support for subsequent rule matching and avoiding oversights and errors caused by manual judgment of wiring patterns. By establishing a one-to-one mapping relationship between equipment nodes, electrical circuits, and topology levels, the completeness and accuracy of wiring structure feature extraction are ensured. This provides accurate topological basis for generating candidate solutions based on a rule base, avoiding deviations in gate location settings caused by topology recognition errors.

[0008] Optionally, the step of generating candidate gate schemes based on the wiring structure features and the gate setting rule base includes: constructing a gate knowledge graph containing the mapping relationship between combinations of wiring structure features and gate setting attributes; converting the knowledge elements in the gate knowledge graph into executable rules after conflict verification and normalization to form a gate setting rule base; constructing a reinforcement learning space based on the wiring topology of the target power plant and the gate setting rule base; constructing a multi-objective reward function based on compliance, metering integrity, and construction cost; constructing a gate selection agent based on the reinforcement learning space and the multi-objective reward function; and generating candidate gate schemes through the gate selection agent based on the wiring structure features.

[0009] This solution constructs a knowledge graph of key points and transforms it into executable rules, converting discrete power regulations into computer-executable logical rules. This overcomes the limitations of systematically reusing human experience, enabling conflict verification and normalization of rules, and improving the compliance of generated solutions. By combining reinforcement learning agents with a multi-objective reward function based on compliance, metering integrity, and construction cost, and utilizing key point selection agents to generate candidate solutions, the traditional exhaustive trial-and-error approach is transformed into intelligent search. This adaptively matches different wiring structure features, improving the accuracy and efficiency of candidate solution generation, while ensuring that the generated solutions meet metering requirements while also considering economic efficiency and engineering feasibility.

[0010] Optionally, the digital twin based on the target power plant performs multi-scenario operational condition simulations on the candidate gate solutions, and generates optimized gate solutions through reinforcement learning based on the simulation results. This includes: constructing a digital twin of the target power plant based on its topological adjacency matrix; mapping the candidate gate solutions to the digital twin model to configure the simulation environment of the candidate gate solutions at corresponding topological nodes through virtual metering gates; constructing a corresponding operational condition simulation matrix based on the initial boundary conditions, simulation parameters, verification indicators, and pass standards of the test scenario; performing simulations based on the operational condition simulation matrix for each group of candidate gate solutions to obtain simulation results; and using the simulation results as reward / penalty signals for the gate selection agent to provide positive / negative rewards to the candidate gate solutions, thereby generating optimized gate solutions.

[0011] In this scheme, by constructing a digital twin and executing the operating condition simulation matrix, the simulation results are fed back to the reinforcement learning agent, forming a closed-loop iterative mechanism for simulation verification and strategy optimization. This mechanism can comprehensively verify the metering completeness of candidate schemes in multiple scenarios and autonomously optimize defects, improve the robustness and reliability of the gate scheme, and avoid power omissions or metering errors caused by inadequate consideration of operating conditions.

[0012] Optionally, the step of using the deduction results as a reward / penalty signal for the gate selection agent to provide positive / negative rewards to the candidate gate solutions and generate optimized gate solutions includes: classifying the candidate gate solutions by defects, assigning positive / negative reward values ​​to the candidate gate solutions according to the defect level, and selecting solutions with locally optimizable defects to be optimized; generating a defect feature vector based on the defect features corresponding to the solutions to be optimized, including electrical topology nodes, scene type, defect type, defect range, and quantitative value of measurement deviation, and determining optimization constraint boundaries; correcting the weight coefficients of each optimization objective in the multi-objective reward function based on the defect feature vectors, and using the optimization constraint boundaries as state space constraints for the gate selection agent; generating optimized gate solutions based on the gate selection agent updated with weight coefficients and state space constraints, and simultaneously selecting candidate gate solutions whose multi-objective reward comprehensive value exceeds a preset threshold after being given positive reward values ​​as optimized gate solutions.

[0013] This scheme employs defect classification and feature vector generation to finely adjust the reward function weights and state space constraints, enabling the reinforcement learning agent to specifically optimize local defects. This overcomes the convergence difficulty of general reinforcement learning algorithms under specific engineering constraints. By assigning positive / negative reward values ​​and updating the agent, the optimization process can focus on eliminating key defects (such as measurement blind spots). This allows for dynamic adjustment of the optimization direction while ensuring compliance, generating gate solutions that satisfy both physical constraints and optimal performance indicators, thus improving optimization efficiency and accuracy.

[0014] Optionally, the method further includes: for the candidate gate scheme, deduce the current flow direction inside the power plant under generator start-up and shutdown and different switch start-up and shutdown combinations, and simultaneously perform power metering analysis under the corresponding operating conditions; verify the rationality of the gate setting position in the candidate gate scheme under different operating conditions based on the current deduction and power metering analysis results; adjust the gate setting position and / or the number of gates based on the verification results to obtain an optimized gate scheme.

[0015] In this scheme, the rationality of the gate location at the physical operation level is directly verified by the current flow direction deduction and power metering analysis under specific operating conditions. The gate location or number is dynamically adjusted according to the verification results, ensuring that the generated optimization scheme can accurately measure the purchased and sold power throughout the entire life cycle. This effectively prevents metering blind spots or duplicate metering caused by operating condition switching and ensures the fairness of trade settlement.

[0016] Optionally, the electricity metering analysis includes: calculating the purchased electricity and / or sold electricity corresponding to each gate setting position based on the current topology derivation results; determining whether the electricity calculation formula is applicable under each scenario and whether there is any electricity omission.

[0017] This solution provides an objective standard for verifying the completeness of metering by quantitatively calculating the electricity purchased and sold and judging the availability and omissions of the formula. This avoids errors and subjectivity caused by the inability to confirm metering accuracy through logical judgment alone.

[0018] Optionally, the step of performing a multi-objective comprehensive evaluation of the optimized checkpoint schemes and obtaining the optimal checkpoint setting scheme based on the evaluation results includes: establishing a multi-dimensional evaluation index system including measurement completeness, data integrity, operational stability, and maintenance convenience; employing a multi-objective optimization algorithm, using each dimension in the multi-dimensional evaluation index system as a parallel optimization objective, quantitatively scoring each optimized checkpoint scheme under each dimension to obtain multiple sets of optimized checkpoint scheme solutions; and dynamically adjusting the preset weights of each optimization objective based on user needs to determine the optimal checkpoint setting scheme that meets user needs from the multiple sets of optimized checkpoint scheme solutions.

[0019] This solution constructs a multi-dimensional indicator system and uses a multi-objective optimization algorithm to output a solution set. By dynamically adjusting the weights based on user needs, it achieves personalized solution optimization, taking into account the requirements of measurement, operation, and maintenance, and improving the transparency and adaptability of solution selection.

[0020] Secondly, embodiments of this application provide a power plant gate setting scheme generation system, comprising: a wiring diagram recognition module, used to obtain wiring structure features based on the electrical wiring diagram of a target power plant; a rule base matching module, used to generate candidate gate schemes based on the wiring structure features and a gate setting rule base, wherein the candidate gate schemes include at least one gate setting location; an operating condition simulation module, used to perform multi-scenario operating condition simulations on the candidate gate schemes based on a digital twin of the target power plant, and generate an optimized gate scheme through reinforcement learning based on the simulation results; and a scheme optimization module, used to perform multi-objective comprehensive evaluation on the optimized gate scheme, and obtain the optimal gate setting scheme based on the evaluation results.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0022] The beneficial effects of this application are: 1. By obtaining wiring structure features based on electrical wiring diagrams and generating candidate gate setting schemes in conjunction with a gate setting rule base, the gate setting method, which is dominated by manual experience, is replaced. This achieves standardized matching of different power plant entities, wiring methods and gate setting schemes, fundamentally solving the problem of non-standard and inconsistent gate setting in existing technologies, and improving the standardization and uniformity of power plant gate setting. 2. By performing multi-scenario operating condition simulations based on the digital twin of the target power plant, and generating optimized gate solutions through reinforcement learning based on the simulation results, the manual simulation and calculation method is replaced, forming a closed-loop iterative mechanism for simulation verification and strategy optimization. It can autonomously cope with non-standard wiring scenarios and extreme operating conditions, quickly verify the rationality of gate locations under various operating conditions, and improve the efficiency and accuracy of gate rationality demonstration. 3. By conducting multi-objective comprehensive evaluation of the optimized gate scheme and outputting the optimal solution, the method of selecting the scheme by human subjective judgment is replaced. Through quantitative evaluation indicators and multi-objective optimization algorithms, the method takes into account the requirements of measurement, operation, cost and other aspects, thereby improving the effectiveness and objectivity of gate scheme selection. Attached Figure Description

[0023] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0024] Figure 1 A flowchart illustrating a method for generating a power plant gate setting scheme, as provided in an embodiment of this application.

[0025] Figure 2 A schematic diagram of the junction wiring mode and its feature hierarchy provided in the embodiments of this application.

[0026] Figure 3 This is a schematic diagram of a power plant gate setting scheme generation system module provided in an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Example 1: As Figure 1 As shown, a method for generating a power plant gate setting scheme includes steps S1-S4, wherein: S1. Obtain wiring structure features based on the electrical wiring diagram of the target power plant.

[0029] In an optional embodiment, step S1 includes: Semantic recognition is performed on the electrical wiring diagram of the target power plant to obtain power plant wiring data containing equipment type, connection relationship, wiring mode and node number; A topological adjacency matrix is ​​generated based on power plant wiring data, which maps equipment nodes, electrical circuits, and topological levels one-to-one. The wiring structure features are obtained based on the topological adjacency matrix. The wiring structure features include at least the number of main components of the target power plant, the power source type, the number of generating units, the number of busbars, the number of outgoing lines, and the circuit connection features.

[0030] Specifically, electrical wiring diagrams are the core carriers reflecting the physical connections of a power plant, typically containing topological connection information for equipment such as generating units, main transformers, busbars, outgoing lines, and switches. The number of main components, power source types, generating unit configurations, and outgoing line architectures vary greatly among different power plants. Only by explicitly and structurally representing the implicit topological relationships and equipment attributes in the drawings can they be accurately matched with a pre-defined rule base. Electrical wiring diagrams usually exist in the form of CAD drawings or schematic diagrams, belonging to unstructured or semi-structured visual data. The semantic recognition process is not simply image extraction, but rather uses visual recognition algorithms such as deep learning semantic segmentation models to map the graphic symbols in the drawings into entity objects with engineering semantics. For example, the recognized objects include not only specific equipment types such as generators, main transformers, busbars, and circuit breakers, but also the connection relationships between equipment (e.g., a circuit breaker connected between the high-voltage side of main transformer No. 1 and the 220kV busbar) and node numbers (e.g., busbar segment numbers, outgoing line bay numbers).

[0031] Specifically, each element in the topological adjacency matrix not only represents whether there is a physical connection between two device nodes (i.e., the adjacency relationship is 1 or 0), but also maps the hierarchical affiliation of electrical circuits (such as high-voltage side circuits and low-voltage side circuits) and topological levels (such as main grid layer and plant power supply layer) through the hierarchical structure of the matrix or the additional attribute matrix. The matrix representation provides a computable mathematical basis for subsequent pattern determination, rather than remaining at the level of fuzzy graphical associations.

[0032] In some embodiments, combined with Figure 2 As shown, the wiring structure characteristics are the core label determining the wiring mode of a power plant. The number of entities indicates whether there are multiple independently accounted ownership or operating entities within the power plant (e.g., single or multiple entities); the power source type distinguishes different power generation forms such as thermal power, nuclear power, gas power, wind power, and photovoltaic power, because the operating characteristics and metering requirements of different power sources differ significantly; the number of generating units, buses, and outgoing lines directly determines the complexity of the topology; and the loop connection characteristics further refine the internal interconnection relationships and circulating current situation. For example, Figure 2The wiring pattern shown, characterized by "multiple entities," "individual metering of electricity for each entity," and "busbar breakpoints," is uniquely determined by a specific combination of the aforementioned feature dimensions. It should be understood that the dimensions of the wiring structure features are not limited to those listed above. In other embodiments, they may also include outgoing asset ownership features or energy storage configuration features to accommodate more complex classification logic.

[0033] Furthermore, S2. Generate candidate gate schemes based on the wiring structure features and the gate setting rule base, wherein the candidate gate schemes include at least one gate setting position.

[0034] In an optional embodiment, generating candidate gateway schemes based on the wiring structure features and the gateway setting rule base includes: A gateway knowledge graph is constructed, which includes the mapping relationship between the combination of wiring structure features and gateway setting attributes. After the knowledge elements in the gateway knowledge graph are subjected to conflict verification and normalization processing, they are transformed into executable rules to form a gateway setting rule base. A reinforcement learning space is constructed based on the wiring topology of the target power plant and the rule base for setting the gate. A multi-objective reward function is constructed based on compliance, metering integrity and construction cost. A gate selection agent is constructed based on the reinforcement learning space and the multi-objective reward function. Based on the wiring structure features, candidate gateway schemes are generated by the gateway selection agent.

[0035] Specifically, the mapping relationship includes the gate setting location, metering range, and equipment configuration requirements corresponding to different combinations of wiring structure features.

[0036] In some embodiments, the attributes and relationships of each dimension are determined through a six-dimensional ontology relationship: power plant main body type, electrical wiring mode, equipment attributes, gate setting rules, metering accounting logic, and compliance verification clauses. Natural language processing technology is used to extract key gate setting elements (such as location, metering range, equipment configuration, line loss calculation rules, etc.) corresponding to different power plant types and wiring modes from historical gate schemes, benchmark project cases, and regulatory documents. This is combined with wiring structure characteristics to obtain knowledge entities, attributes, and relationships. Furthermore, various wiring structure characteristics (such as single main body, single outgoing line) and gate setting attributes (such as outgoing line side gate, main transformer high-voltage side gate, metering range covering grid-connected electricity) are used as nodes, and the mapping relationship between feature combinations and attributes (such as single main body + single outgoing line pointing to outgoing line side gate) is used as edges to construct a knowledge graph.

[0037] Furthermore, due to the different data sources of knowledge graphs, inconsistencies in expression or logical contradictions exist. For example, a historical case might place the gate under a specific outgoing line on the power plant side, while management regulations might prioritize the substation side, thus creating a conflict. Conflict verification and normalization are precisely designed to solve this problem: the verification logic eliminates non-compliant historical mappings based on the compliance priority of management regulations, while normalization unifies semantically different but essentially identical expressions into standard terminology. This step is necessary because unverified and normalized knowledge graphs contain noise and contradictions; if directly used for rule matching, the generated candidate solutions will have compliance risks or be self-contradictory. Transforming them into executable rules involves extracting the mapping edges in the graph into conditional judgment logic of "IF feature combination THEN gate attribute," enabling computing devices to directly perform matching instead of complex graph traversal and reasoning.

[0038] Furthermore, the state space in the reinforcement learning space consists of the topological adjacency matrix of the target power plant, equipment attributes, and currently selected gate locations. The action space is defined as the set of operations for adding, deleting, or moving gate locations on topological nodes. A multi-objective reward function is used to quantify engineering evaluation criteria into feedback signals perceptible to the agent. The compliance reward ensures that the selected location does not violate mandatory industry standards; the metering integrity reward drives the agent to find gate combinations that cover all operating conditions of electricity flow; and the construction cost reward constrains the agent from indiscriminately adding metering equipment. Through this weighted or parallel reward design across these three dimensions, the gate selection agent, when exploring the reinforcement learning space, no longer blindly tries and fails, but autonomously evolves along the direction of engineering optimization.

[0039] In some examples, the multi-objective reward function is represented as: ; in, Indicates compliance reward, Indicates a reward for measurement integrity. Indicates construction cost incentive. , and These represent the weights of the reward items. .

[0040] The compliance rate is quantified as the percentage of correct checkpoints / the number of recommended checkpoints (referring to the number of standard checkpoints in the rule base for this wiring mode). If each checkpoint meets the requirements of the corresponding wiring mode, a positive value is awarded; if there are missing or redundant checkpoints, a negative value is awarded as a penalty.

[0041] Quantified as electricity metering deviation, with a value range of [0,1], it is represented as follows: ; Where N is the total number of operating conditions evaluated in the current simulation, and k is the operating condition index. This represents the weighting coefficient for the k-th operating condition, reflecting its importance in the integrity assessment (e.g., 0.6 for steady-state conditions, 0.4 for transient conditions, and so on for all operating conditions). The sum is 1). This represents the total electricity consumption obtained under the k-th operating condition according to the current threshold configuration (e.g., obtained by accumulating the electricity consumption at each threshold). This represents the actual total power consumption obtained through power flow calculation or simulation under the k-th operating condition. It represents an extremely small positive number to avoid quantization errors when the actual battery level is zero.

[0042] The installation cost at each checkpoint is quantified, with a value ranging from [0,1], and is represented as follows: ; in, , Let i represent the total construction cost of the current plan, i be the index of the candidate checkpoint, and S be the set of checkpoint locations selected in the current plan. This represents the total cost (including meters, CT / PT, communication equipment, construction and installation, etc.) required to set up a checkpoint at the i-th position. This represents the cost ceiling (such as the owner's budget or the average cost ceiling for similar power plant construction). The lower the total cost, the higher the reward; when the total cost exceeds the ceiling, the reward is 0.

[0043] Specifically, the agent takes the wiring structure features as its initial state input and executes a sequence of actions in the reinforcement learning space. Each action receives immediate feedback from a multi-objective reward function, ultimately outputting a set of gate position combinations with superior reward values ​​under the current understanding, i.e., candidate gate solutions. Through the iteration of the gate selection agent, the generation of candidate solutions is no longer limited to exhaustive matching of static rules, but possesses the ability of self-evolution and adaptive search, enabling it to cope with non-standard wiring scenarios not covered by the rule base.

[0044] In this embodiment, by binding the underlying framework of feature extraction and rule base construction as described above, this embodiment establishes a transformation depth from unstructured drawings to executable rules and then to intelligent agents' autonomous search, ensuring that the technical solution not only remains at the level of functional conception, but also has a complete engineering-feasible logic and data flow closed loop.

[0045] S3. Based on the digital twin of the target power plant, perform multi-scenario operational condition simulations on the candidate gate solutions, and generate optimized gate solutions through reinforcement learning based on the simulation results.

[0046] In an optional embodiment, step S3 includes: A digital twin of the target power plant is constructed based on the topological adjacency matrix of the target power plant. The candidate gate schemes are mapped to the digital twin model so as to configure the simulation environment of the candidate gate schemes through virtual metering gates at the corresponding topological nodes. Construct the corresponding working condition deduction matrix based on the initial boundary conditions, simulation parameters, verification indicators and pass criteria of the test scenario; For each group of candidate checkpoint schemes, a deduction is performed based on the working condition deduction matrix to obtain the deduction results; The simulation results are used as reward and punishment signals for the checkpoint selection agent to provide positive / negative rewards to the candidate checkpoint schemes and generate optimized checkpoint schemes.

[0047] In this embodiment, each device node in the topological adjacency matrix (such as the high-voltage side node of the No. 1 main transformer and the 220kV bus section node) corresponds to a virtual node with electrical calculation attributes in the digital twin. The adjacency relationship represented in the matrix (i.e., the connection with an element value of 1) corresponds to the electrical branch between virtual nodes in the twin. This branch not only reproduces the topological path of the physical connection but also carries electrical attributes such as impedance parameters and admittance parameters. It should be understood that although this embodiment uses the topological adjacency matrix as the basis for constructing the twin, in other embodiments, the twin can also be expanded in dimension by combining the three-dimensional spatial coordinates of the power plant or the real-time operating status data of the equipment, as long as it can faithfully reflect the static topology and dynamic electrical behavior of the power plant. By constructing the twin through mathematical matrices, the complex connection relationships of the physical world are transformed into a computable graph structure, enabling subsequent operating condition simulations to directly perform current topology traversal and power distribution calculations on the graph network, avoiding the high cost and safety risks of physical trial and error.

[0048] In some embodiments, when the candidate gate designation specifies setting up the gate on the substation side opposite the outgoing line, a virtual topology node corresponding to the physical location is located in the digital twin, and a virtual metering gate module is inserted into this node. This module functions equivalent to an electrical energy meter in the physical world; it can collect current and voltage data flowing through the virtual node and calculate the accumulated electricity accordingly, simulating the data collection behavior of the gate meter in actual operation, thereby providing a data source for subsequent metering completeness verification.

[0049] In some embodiments, the initial boundary conditions in the operating condition simulation matrix define the starting state of the simulation, such as setting the generator output to 600MW or 0MW (shutdown state), the switch state to closed or open, and the plant power load to a specific power value. These conditions directly determine the initial distribution of the current topology. The simulation parameters define the electrical calculation rules in the simulation process, such as the value of the line loss rate, the transformer ratio and loss parameters, and the convergence accuracy of the power flow calculation, to ensure that the simulation results conform to physical laws. The verification index defines the dimensions of the observation data that need to be extracted from the virtual metering gate, such as the current flow direction data of each gate node, the cumulative value of input power, the cumulative value of output power, and the power metering deviation data between different gates. The qualification standard defines the threshold for the verification index to pass the simulation verification. For example, if the deviation rate between the total input power and the total output power under any operating condition does not exceed one percent, it is considered that there is no omission. If the data nodes required in the power calculation formula of each gate can be collected by the virtual gate, it is considered that the formula is usable. By using the working condition simulation matrix, the originally scattered working condition settings and evaluation logic are structured into matrix operation instructions that can be executed in batches. This enables the system to automatically and in parallel complete the simulation and evaluation of massive working conditions, completely replacing the inefficiency and subjectivity of manual item-by-item calculation.

[0050] Specifically, the simulation results include, but are not limited to, full-scenario simulation datasets (such as current flow data), compliance verification results, electricity metering statistics, and defect information (i.e., for metering blind spots, compliance violations, and logical anomalies that occur in the simulation, the simulation is automatically paused and the defect location, defect type, and root cause are marked to obtain defect information).

[0051] In an optional embodiment, the step of using the deduction results as a reward / penalty signal for the checkpoint selection agent to provide positive / negative rewards to the candidate checkpoint solutions and generate optimized checkpoint solutions includes: The candidate gate solutions are classified into defects, and positive / negative reward values ​​are assigned to the candidate gate solutions according to the defect level. Solutions with locally optimizable defects are then selected for optimization. Based on the defect features corresponding to the scheme to be optimized, a defect feature vector is generated, including electrical topology nodes, scene type, defect type, defect range and measurement deviation quantification value, and the optimization constraint boundary is determined. The weight coefficients of each optimization objective in the multi-objective reward function are corrected based on the defect feature vector, and the optimization constraint boundary is used as the state space constraint condition of the gate selection agent. Based on the updated weight coefficients and state space constraints, the gate selection agent generates optimized gate schemes. Simultaneously, the candidate gate schemes whose multi-objective reward comprehensive value exceeds a preset threshold after being given a positive reward value are selected as optimized gate schemes.

[0052] Specifically, based on the defect information in the simulation results, candidate gate solutions are classified into defect levels, including: severe defects, which refer to large-scale power omissions or complete unavailability of key calculation formulas under core operating conditions, such as the inability to measure plant power consumption under single-unit operation conditions. Such defects directly render the candidate solution unusable and require elimination, assigning a very large negative reward value (e.g., -100) to force the selected agent to stay away from such configuration areas; moderate defects, which refer to local omissions or deviation rates exceeding the acceptable standard under specific non-core operating conditions, such as line loss allocation calculation errors reaching a preset percentage. Such defects are assigned a moderate negative reward value (e.g., -50), but are marked as locally optimizable, meaning they can be repaired by adding auxiliary metering points or fine-tuning their positions; and minor defects, which refer to data completeness or maintenance convenience slightly lower than expected but still within the tolerable range, assigning a small negative reward value (e.g., -10). Solutions that fully meet the acceptable standards and have no defects are assigned a positive reward value (e.g., +100).

[0053] For severely flawed schemes that need to be eliminated, a preset maximum negative reward value is assigned, and their major flaw features are simultaneously included in the action disallowing library of the reinforcement learning model as a hard constraint to prevent the model from generating similar flawed schemes.

[0054] It should be noted that the logic of screening solutions with locally optimizable defects is that solutions with fatal defects usually require reconstruction of the topological logic, which is too costly to optimize. On the other hand, solutions with serious and minor defects have value for local repair. Screening them as solutions to be optimized can significantly reduce the search space of reinforcement learning and accelerate convergence.

[0055] Specifically, the defect feature information and optimization constraint boundary of the scheme to be optimized are determined according to the defect level; based on the defect feature information (including defect feature name and feature quantity), defect feature vectors are extracted according to the full-scenario simulation dataset and defect information. The defect features include the electrical topology node corresponding to the defect, the working scenario that triggered the defect, the defect type (insufficient metering accuracy, missing line loss allocation logic, local working condition omission, operation and maintenance boundary conflict, etc.), the defect impact range, and the quantified value of metering deviation.

[0056] Furthermore, based on the defect risk level, defect impact range, and measurement deviation quantification value, the negative reward value is quantitatively calculated. The higher the defect risk level, the more affected operating conditions, and the greater the measurement deviation, the higher the absolute value of the negative reward. For the dimension of full-condition adaptation without anomalies in the optimization scheme, positive reward values ​​are assigned based on the optimization objectives of measurement accuracy, construction cost, and compliance.

[0057] Furthermore, based on the user's preset needs and extracted defect features, the weight coefficients of each optimization objective in the multi-objective optimization reward function are modified, and the optimization constraint boundary is simultaneously transformed into the state space constraint condition of the reinforcement learning model, locking the compliance boundary in the model iteration process; the action space of the gate selection agent is restricted by the state space constraint parameters and the action disabling library, prohibiting the gate selection agent from outputting gate configuration actions with identified severe defects.

[0058] In some embodiments, the quantified negative and positive reward values ​​are input into the experience replay pool of the gate selection agent. A temporal difference learning algorithm is used to iteratively update the parameters of the agent's policy network and value network, focusing on optimizing the working conditions that trigger defects and the gate selection decision logic corresponding to topological nodes to reduce the recurrence probability of similar defects. After parameter updates, the agent undergoes multiple rounds of policy evaluation, using a 100% compliance rate for gate selection in defect scenarios as the convergence threshold. If the threshold is not reached, the iterative update process is repeated until the agent's gate selection schemes meet compliance requirements in all defect scenarios. Finally, the iteratively optimized model parameters are locked, and the agent autonomously explores and generates new gate candidate schemes based on the optimized parameters, forming an optimized gate scheme set.

[0059] In some examples, the steps of adjusting the weight coefficients of each optimization objective in the multi-objective optimization reward function based on defect characteristics include: Based on defect characteristics, defect types and measurement deviation values ​​are extracted. If the defect type is omission or duplication and the deviation value is greater than a preset threshold (e.g., 0.1), the completeness weight is increased; if the defect type is compliance deficiency (e.g., missing essential checkpoints), the compliance weight is increased; if the current cost score is less than a preset threshold (e.g., 0.3) and there are no serious measurement defects, the cost weight is increased. The specific adjustment coefficients are expressed as follows: ; ; ; in, The importance enhancement factor for the compliance dimension is adjusted based on whether there are any compliance deficiencies. The enhancement factor for the integrity dimension is adjusted based on measurement deviation and defect range; This is an enhancement factor for the cost dimension, which is adjusted based on whether the current cost score is too low; This is an indicator function: it takes the value 1 if the condition is true, and 0 otherwise. The target weight is obtained based on the adjusted coefficients and the dynamic weight formula, which is expressed as follows: ; ; ; in, , and These represent the normalized weights of each reward item. , , These represent the original scores for compliance, integrity, and cost in the current round, respectively, derived from the above. , , Calculated using the following method; After normalizing the unnormalized weights, the corresponding target weights are obtained. , and The normalized representation is: Similarly, we can obtain and , .

[0060] In some possible real-time methods, to avoid drastic weight fluctuations, an exponential moving average can be used to update the actual target weights, expressed as: ; in, This represents the target weight actually used in round t (or time step). , and (any one of them) Indicates the weight of the previous round. This represents the target weight calculated based on the current defect characteristics. , or ), The smoothing coefficient is usually set to 0.9, which means that 90% of the historical weights are retained and 10% are from the newly calculated values.

[0061] In this embodiment, a parallel output mechanism for optimizing gate-level solutions is adopted: In the inference results, if a candidate solution does not obtain the highest positive reward, but its multi-objective reward comprehensive value (i.e., the weighted sum of scores and weight coefficients of each dimension) exceeds a preset high threshold (e.g., 90 points), it indicates that the solution, although not perfect, has reached the engineering good standard. Such solutions do not need to enter subsequent iterations for optimization and are directly output as optimized gate-level solutions. The parallel output mechanism prevents reinforcement learning from falling into an over-optimization loop, ensuring that a high-quality solution that meets the engineering bottom line can be locked and output in a timely manner, balancing optimization depth and output efficiency.

[0062] In an optional embodiment, the method further includes: For the candidate gate schemes, the current flow inside the power plant is deduced under generator start-up and shutdown and different switch start-up and shutdown combinations, and the power metering analysis under the corresponding operating conditions is carried out simultaneously. The rationality of the gate setting location in the candidate gate scheme under different operating conditions is verified based on the results of current extrapolation and power metering analysis. Adjust the location and / or number of checkpoints based on the verification results to obtain an optimized checkpoint scheme.

[0063] In an optional embodiment, the electricity metering analysis includes: calculating the purchased electricity and / or sold electricity corresponding to each gate setting position based on the current topology derivation results; determining whether the electricity calculation formula is applicable under each scenario and whether there is any electricity omission.

[0064] In this embodiment, the current flow deduction rule is a set of mapping logic based on Kirchhoff's current law and the physical properties of the equipment. For example, for generator units, when they are in the start-up state and the power generation is greater than zero, the deduction rule determines that the current flows out from the generator node and is transmitted to the bus or main transformer; when they are in the shutdown state, if the plant power still needs to be maintained, the deduction rule determines that the current flows into the generator node from the system side to supply power for the standby transformer. For switchgear, when it is in the closed state, the deduction rule determines that its branch is connected and the current can flow normally according to the topology path; when it is in the open state, the deduction rule determines that its branch is disconnected, and the current that originally flowed through the branch needs to find a new path or return to zero according to the new topology. For outgoing line equipment, the deduction rule determines the flow direction based on the comparison between the current overall power generation and the plant power load: when the power generation is greater than the power consumption, the outgoing line current flows to the output (power is sent to the external grid); when the power generation is less than the power consumption, the outgoing line current flows to the input (power is purchased from the external grid). It should be understood that the above-mentioned deduction rules are only typical examples. In other embodiments, the deduction rules may also cover the reversal of high and low voltage side flow direction under transformer step-up / step-down conditions, the switching of charging and discharging states of energy storage devices, and the bidirectional power flow of tie lines under different topologies, as long as they can faithfully reflect the mapping relationship between device state and physical current flow direction.

[0065] The simultaneous electricity metering analysis is based on the deduced current flow data, as shown in Table 1. Table 1. Electricity Measurement Analysis Table

[0066] As shown in Table 1, the electricity metering analysis is based on the deduced current flow data. By superimposing the voltage and current integral data collected by the virtual metering gate, the input and output electricity of each gate point are numerically calculated, thereby transforming the physical current distribution into comparable economic metering data.

[0067] Furthermore, the calculation of purchased and / or sold electricity volume is based on signed integral operations performed on the flow direction attributes determined by current topology deduction. When the deduction result determines that the current flow direction of a certain checkpoint node is output, the accumulated electrical energy at that checkpoint is marked as sold electricity volume; when the deduction result determines that the current flow direction of a certain checkpoint node is input, the accumulated electrical energy at that checkpoint is marked as purchased electricity volume. For scenarios involving complex calculations such as line loss allocation, the calculation of purchased and sold electricity volume also needs to be combined with the electricity volume difference between different checkpoint nodes and the line loss rate parameter for joint calculation to ensure that the calculation results reflect the actual trade settlement logic. When all metering point data involved in the calculation formula of purchased and / or sold electricity volume can be directly obtained by the virtual checkpoints in the current candidate scheme, it indicates that the electricity volume calculation formula is usable.

[0068] The detection method for determining whether there is a power leakage is based on the macroscopic comparison logic of the law of conservation of energy: For each simulation scenario, the system sums and compares the total purchased power measured by all input-side gates with the total sold power measured by all output-side gates and the power lost in the plant. If the deviation rate between the total input and the total output exceeds the preset physical loss tolerance threshold (e.g., 1%), it is determined that there is a power leakage that has not been captured by any gate under this operating condition, indicating that there is a hidden flow path in the current topology that is not covered by the gate.

[0069] Furthermore, the mechanism for verifying rationality lies in the fact that a qualified gate setting scheme must ensure that the flow of electrical energy can be captured by the set gate nodes under all possible physical operating conditions, and there are no metering blind spots caused by topology path switching. If a candidate scheme shows current flowing through a branch without a gate under a specific operating condition, or if the virtual gate cannot collect the required current data, then it is directly determined that the gate location is not rational under that operating condition.

[0070] In this embodiment, by directly binding the simulation verification and power calculation to the physical electrical operation rules, a closed loop from virtual simulation data to engineering rationality proof is realized, ensuring the robustness and comprehensive coverage of the optimization scheme in actual physical operation.

[0071] S4. Perform a multi-objective comprehensive evaluation on the optimized checkpoint scheme, and obtain the optimal checkpoint setting scheme based on the evaluation results.

[0072] In an optional embodiment, step S4 includes: Establish a multi-dimensional evaluation index system that includes measurement completeness, data integrity, operational stability, and maintenance convenience; A multi-objective optimization algorithm is adopted, with each dimension in the multi-dimensional evaluation index system as the parallel optimization objective, to quantify and score each optimization threshold scheme under each dimension, and obtain multiple sets of optimization threshold scheme solutions; The preset weights of each optimization objective are dynamically adjusted based on user needs, so as to determine the optimal gate setting scheme that meets user needs from the solution set of the multiple optimization gate schemes.

[0073] Understandably, while the gate design optimized through reinforcement learning achieves metering completeness across multiple operating scenarios, its implementation in actual engineering projects still requires consideration of comprehensive needs across multiple dimensions, including data integrity, operational stability, maintenance convenience, and construction costs. The reason for needing a multi-objective comprehensive evaluation to conclude the iteration and output the final solution is that a single metering completeness indicator cannot comprehensively measure the engineering merits of a solution, and different power plant users have varying priorities for each indicator. Multi-objective comprehensive evaluation establishes a multi-dimensional quantitative evaluation index system to score optimized solutions in parallel and dynamically adjusts the weight preferences of each dimension based on user needs, thereby selecting the optimal gate setting solution that best suits the current user requirements from multiple sets of optimized solution solutions. It should be understood that the specific algorithm for multi-objective comprehensive evaluation is not limited to simple weighted scoring; it can also use a multi-objective optimization algorithm to output a Pareto front solution set for user interaction and selection, as long as it can achieve multi-dimensional parallel evaluation and adaptive weight adjustment.

[0074] Specifically, metering completeness refers to the ability of the checkpoint scheme to capture electricity flow without blind spots under all simulated operating conditions. Its quantitative scoring method can be based on the proportion of missed electricity in the simulated results relative to the total flowing electricity. For example, a missed proportion of 0% earns a perfect score of 100 points, while a missed proportion exceeding 5% earns 0 points, with intermediate scores calculated using linear interpolation. This dimension directly reflects the core trade settlement requirements of the checkpoint setup. Data integrity refers to the recoverability and resilience of checkpoint-collected data in the event of communication link interruptions or equipment failures. Its quantitative scoring method can be based on the coverage rate of redundant metering points or historical communication interruption rate statistics. For example, points are added if a key metering node has a backup data acquisition path; otherwise, points are deducted. Operational stability refers to the ability of the checkpoint location to maintain effective metering during physical electrical topology changes or equipment maintenance. Its quantitative scoring method can be based on the frequency of switch opening and closing affecting the branch where the checkpoint is located. The lower the frequency of impact, the higher the score, ensuring the robustness of the checkpoint under complex operating conditions. Maintenance convenience refers to the physical accessibility and operational safety of the meter at the actual site during installation, inspection, and replacement. Its quantitative scoring method can be based on assessing the inspection path length or power outage range according to the spatial coordinates of the meter's location. For example, meter locations located at high altitudes or requiring a large-scale power outage for access will receive lower scores. The specific formulas for quantifying the above-mentioned indicators are not limited to linear interpolation; step functions or exponential decay functions can also be used to adapt to the fault-tolerance characteristics of different evaluation dimensions.

[0075] Furthermore, this embodiment employs a multi-objective optimization algorithm, such as a non-dominated sorting genetic algorithm, with four parallel optimization objectives: metric completeness, data integrity, operational stability, and maintenance convenience. Non-dominated sorting is used to find the optimal solution set from the reinforcement learning-output optimization threshold population. Non-dominated sorting means that if solution A scores no less than solution B in all dimensions and is strictly higher than solution B in at least one dimension, then solution B is dominated by solution A and eliminated. If solution A is better than solution B in metric completeness, but solution B is better than solution A in maintenance convenience, then they do not dominate each other and are both retained. This parallel optimization mechanism ultimately outputs not a single solution, but multiple sets of non-dominated optimization threshold solutions.

[0076] Furthermore, when dynamically adjusting indicator weights, the preset weight coefficients of each dimension indicator are adjusted based on user demand tags. For example, when the user demand tag is "metering priority" or the power plant involves complex multi-entity trade settlement, the weight coefficients of metering completeness and data integrity are increased to a high range (e.g., adjusted to 0.4 and 0.3 respectively), while the weights of maintenance convenience and operational stability are reduced (e.g., adjusted to 0.1 and 0.2). The optimal solution is then selected from the solution set of the optimization threshold solution, based on the solutions with extremely high scores in the dimensions of metering completeness and data integrity.

[0077] In this embodiment, dynamic weight adjustment achieves a smooth transition from objective algorithm optimization to subjective user decision-making. It utilizes multi-objective optimization algorithms to avoid subjective blind spots and respects the differentiated preferences of different power plants in engineering implementation. This makes the final output solution not only theoretically self-consistent but also highly adaptable and satisfying to users in practical applications.

[0078] Based on the same inventive concept, this application also provides a system for generating a power plant gate setting scheme, corresponding to a method for generating a power plant gate setting scheme, such as... Figure 3 As shown, it includes: The wiring diagram recognition module is used to obtain wiring structure features based on the electrical wiring diagram of the target power plant; The rule base matching module is used to generate candidate gate schemes based on the wiring structure features and the gate setting rule base, wherein the candidate gate schemes include at least one gate setting position; The operating condition simulation module is used to perform multi-scenario operating condition simulations on the candidate gate schemes based on the digital twin of the target power plant, and generate optimized gate schemes through reinforcement learning based on the simulation results. The scheme optimization module is used to perform multi-objective comprehensive evaluation of the optimized checkpoint scheme and obtain the optimal checkpoint setting scheme based on the evaluation results.

[0079] In this embodiment, through the coordinated operation of various modules, the entire process of gate setting scheme from feature recognition, candidate generation, deduction and optimization to comprehensive evaluation is automated, thereby improving the overall operating efficiency of the system and the quality of scheme output.

[0080] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for generating a power plant gate setting scheme.

[0081] The above-described embodiments are preferred embodiments of this application and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape, structure, and method of this application are within the protection scope of this application.

Claims

1. A method for generating a power plant gate setting scheme, the method comprising: Includes the following steps: ​ Obtain wiring structure features based on the electrical wiring diagram of the target power plant; Based on the wiring structure features and the gate setting rule base, candidate gate schemes are generated, and each candidate gate scheme includes at least one gate setting location. Based on the digital twin of the target power plant, multi-scenario operational condition simulations are performed on the candidate gate solutions, and optimized gate solutions are generated through reinforcement learning based on the simulation results. The optimized checkpoint scheme is evaluated using a multi-objective comprehensive assessment, and the optimal checkpoint setting scheme is obtained based on the assessment results.

2. The method of claim 1, wherein: The method of obtaining wiring structure features based on the electrical wiring diagram of the target power plant includes: Semantic recognition is performed on the electrical wiring diagram of the target power plant to obtain power plant wiring data containing equipment type, connection relationship, wiring mode and node number; A topological adjacency matrix is ​​generated based on power plant wiring data, which maps equipment nodes, electrical circuits, and topological levels one-to-one. The wiring structure features are obtained based on the topological adjacency matrix. The wiring structure features include at least the number of main components of the target power plant, the power source type, the number of generating units, the number of busbars, the number of outgoing lines, and the circuit connection features.

3. The method of claim 2, wherein: The generation of candidate gateway schemes based on the wiring structure features and the gateway setting rule base includes: A gateway knowledge graph is constructed, which includes the mapping relationship between the combination of wiring structure features and gateway setting attributes. After the knowledge elements in the gateway knowledge graph are subjected to conflict verification and normalization processing, they are transformed into executable rules to form a gateway setting rule base. A reinforcement learning space is constructed based on the wiring topology of the target power plant and the rule base for setting the gate. A multi-objective reward function is constructed based on compliance, metering integrity and construction cost. A gate selection agent is constructed based on the reinforcement learning space and the multi-objective reward function. Based on the wiring structure features, candidate gateway schemes are generated by the gateway selection agent.

4. The method of claim 1, wherein: The digital twin based on the target power plant performs multi-scenario operational condition simulations on the candidate gate solutions, and generates optimized gate solutions through reinforcement learning based on the simulation results, including: A digital twin of the target power plant is constructed based on the topological adjacency matrix of the target power plant. The candidate gate schemes are mapped to the digital twin model so as to configure the simulation environment of the candidate gate schemes through virtual metering gates at the corresponding topological nodes. Construct the corresponding working condition deduction matrix based on the initial boundary conditions, simulation parameters, verification indicators and pass criteria of the test scenario; For each group of candidate checkpoint schemes, a deduction is performed based on the working condition deduction matrix to obtain the deduction results; The simulation results are used as reward and punishment signals for the checkpoint selection agent to provide positive / negative rewards to the candidate checkpoint schemes and generate optimized checkpoint schemes.

5. The method for generating a power plant gate setting scheme according to claim 4, characterized in that: The step of using the deduction results as a reward / penalty signal for the checkpoint selection agent to provide positive / negative rewards to the candidate checkpoint solutions and generate optimized checkpoint solutions includes: The candidate gate solutions are classified into defects, and positive / negative reward values ​​are assigned to the candidate gate solutions according to the defect level. Solutions with locally optimizable defects are then selected for optimization. Based on the defect features corresponding to the scheme to be optimized, a defect feature vector is generated, including electrical topology nodes, scene type, defect type, defect range and measurement deviation quantification value, and the optimization constraint boundary is determined. The weight coefficients of each optimization objective in the multi-objective reward function are corrected based on the defect feature vector, and the optimization constraint boundary is used as the state space constraint condition of the gate selection agent. Based on the updated weight coefficients and state space constraints, the gate selection agent generates optimized gate schemes. Simultaneously, the candidate gate schemes whose multi-objective reward comprehensive value exceeds a preset threshold after being given a positive reward value are selected as optimized gate schemes.

6. The method for generating a power plant gate setting scheme according to claim 4, characterized in that: The method further includes: For the candidate gate schemes, the current flow inside the power plant is deduced under generator start-up and shutdown and different switch start-up and shutdown combinations, and the power metering analysis under the corresponding operating conditions is carried out simultaneously. The rationality of the gate setting location in the candidate gate scheme under different operating conditions is verified based on the results of current extrapolation and power metering analysis. Adjust the location and / or number of checkpoints based on the verification results to obtain an optimized checkpoint scheme.

7. The method for generating a power plant gate setting scheme according to claim 6, characterized in that: The electricity metering analysis includes: calculating the purchased electricity and / or sold electricity corresponding to each gate setting position based on the current topology derivation results; determining whether the electricity calculation formula is applicable under each scenario and whether there is any electricity omission.

8. The method for generating a power plant gate setting scheme according to claim 1, characterized in that: The step of performing a multi-objective comprehensive evaluation of the optimized checkpoint scheme and obtaining the optimal checkpoint setting scheme based on the evaluation results includes: Establish a multi-dimensional evaluation index system that includes measurement completeness, data integrity, operational stability, and maintenance convenience; A multi-objective optimization algorithm is adopted, with each dimension in the multi-dimensional evaluation index system as the parallel optimization objective, to quantify and score each optimization threshold scheme under each dimension, and obtain multiple sets of optimization threshold scheme solutions; The preset weights of each optimization objective are dynamically adjusted based on user needs, so as to determine the optimal gate setting scheme that meets user needs from the solution set of the multiple optimization gate schemes.

9. A system for generating a power plant gate setting scheme, characterized in that: A method for generating a power plant gate setting scheme as described in any one of claims 1-8, comprising: The wiring diagram recognition module is used to obtain wiring structure features based on the electrical wiring diagram of the target power plant; The rule base matching module is used to generate candidate gate schemes based on the wiring structure features and the gate setting rule base, wherein the candidate gate schemes include at least one gate setting position; The operating condition simulation module is used to perform multi-scenario operating condition simulations on the candidate gate schemes based on the digital twin of the target power plant, and generate optimized gate schemes through reinforcement learning based on the simulation results. The scheme optimization module is used to perform multi-objective comprehensive evaluation of the optimized checkpoint scheme and obtain the optimal checkpoint setting scheme based on the evaluation results.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for generating a power plant gate setting scheme as described in any one of claims 1-8.

Images