Power protection constant value multi-objective collaborative optimization method based on hierarchical optimization strategy

Abstract

The invention relates to the technical field of intelligent power grid relay protection, in particular to an electric power protection constant value multi-target collaborative optimization method based on a hierarchical optimization strategy, which comprises the following steps of: performing traceability acquisition on parameters of an electrical system of a power plant by utilizing a parameter traceability method to obtain a dynamic parameter set; performing association mapping on the topological information and the dynamic parameter set of the power plant electrical system to obtain a digital twinborn model; performing mixed semantic analysis on the protection setting value guide rule to obtain a rule constraint set; performing hierarchical optimization on the digital twinborn model based on the rule constraint set to obtain a local constant value solution set; performing consistency coordination on the local constant value solution set to obtain a setting value; adjusting the value of a setting value of a relay protection module of the digital twin model to the value of the setting value to obtain an electrical system model, and performing risk assessment on the electrical system model to obtain an assessment result; generating a constant value calculation book based on the evaluation result; and the accuracy and efficiency of setting calculation are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of smart grid relay protection technology, and in particular to a multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy. Background Technology

[0002] Relay protection setting calculations in power plants are a core component in ensuring the safe and stable operation of the power system. Their core task is to set accurate operating settings for various protection devices, strictly adhering to the principles of selectivity, speed, sensitivity, and reliability (referred to as the "four principles"). With the expansion of smart grids, their increasing complexity, and the large-scale integration of new energy sources, traditional relay protection setting calculation methods face severe challenges. (1) Existing methods typically use offline or static power grid models, which are difficult to integrate multi-source heterogeneous data from SCADA systems, fault recordings, equipment nameplates, etc. The multi-source heterogeneous data has spatiotemporal anisotropy and is affected by factors such as temperature drift, resulting in fluctuations. This leads to an error of up to 8% to 10% in the electrical topology model established based on multi-source heterogeneous data. (2) The current relay protection setting procedures and anti-accident measures are mostly described in natural language, which has semantic ambiguity. The computer has difficulty in identifying key information, resulting in low accuracy in converting the protection setting procedures and anti-accident measures into mathematical constraints. (3) There is an essential conflict between the selectivity, speed, sensitivity and reliability in the four principles. For example, improving speed will weaken selectivity. Traditional single-objective optimization methods cannot effectively coordinate the conflict, resulting in high complexity of the corresponding tuning calculation. (4) The coordination relationship of protection settings at the junction of power plants and power grids is complex. Improper setting of settings can easily lead to protection failure or false operation, expanding the scope of accidents. Existing technologies lack effective quantitative assessment methods for the robustness of setting schemes in the face of system uncertainties (such as changes in operating mode and parameter fluctuations), and cannot proactively identify and avoid potential risks.

[0003] Chinese Patent Publication No. CN118508383B discloses a multi-objective-based method for optimizing distribution network relay protection settings. This method obtains distribution network operation data and topology diagrams by real-time monitoring of the distribution network's operating status. It then uses the distribution network topology diagram to determine the protection type and protection level of each protection switch, and calls the corresponding deep learning model based on the protection type and protection level. The method calculates the setting values ​​of each protection switch based on the real-time operation data of the distribution network and updates the setting values ​​synchronously. However, this relay protection setting scheme does not consider the impact of multi-source heterogeneous data on establishing the electrical topology model, the difficulty in converting relay protection setting procedures and anti-accident measures, the inherent conflict of the four principles (property, safety, and reliability), and the identification of potential risks. This results in low accuracy, low computational efficiency, and poor security in the corresponding setting calculations. Summary of the Invention

[0004] To address this, the present invention provides a multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy. This method overcomes the problems of low accuracy and low efficiency in existing technologies, which arise from the failure to consider the impact of multi-source heterogeneous data on the establishment of electrical topology models, the difficulty in converting relay protection setting procedures and anti-accident measures, the essential conflicts of the four principles, and the identification of potential risks.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy, comprising the following sub-steps: Obtain the topology information of the power plant's electrical system and obtain the protection setting guidelines; A dynamic parameter set is obtained by tracing and collecting parameters of the power plant's electrical system using the parameter tracing method; A digital twin model is obtained by associating and mapping the topology information and dynamic parameter set of the power plant's electrical system. Hybrid semantic parsing is used to obtain the rule constraint set from the protection setting guidelines; A local constant value solution set is obtained by hierarchically optimizing the digital twin model based on the rule constraint set; The setpoint is obtained by unifying the local constant value solution set. The setting values ​​of the relay protection module in the digital twin model are adjusted to the value of the setting to obtain the electrical system model, and a risk assessment is performed on the electrical system model to obtain the assessment results. A fixed value calculation sheet is generated based on the evaluation results.

[0006] Furthermore, the specific process for obtaining the assessment results by performing a risk assessment on the electrical system model is as follows: Fault simulation is performed on the electrical system model to obtain the simulation dataset; The number of correct actions and the number of incorrect actions are obtained from the statistical simulation dataset, and the risk entropy Qp of the electrical system model is calculated based on the number of correct actions and the number of incorrect actions. The risk entropy Qp is compared with the preset risk entropy Qp0. The risk level of the setting values ​​of the electrical system model is judged based on the comparison results, and the setting values ​​are corrected based on the judgment results, wherein: When Qp≤Qp0, the setting value is determined to be low risk; When Qp > Qp0, the setting value is determined to be high risk. The setting value is then corrected based on the fault data in the simulation dataset to obtain the corrected setting value. By compiling the low-risk setpoints and the corrected setpoints, the risk assessment results are obtained.

[0007] Furthermore, the specific process of correcting the setting value based on the fault data in the simulation dataset to obtain the corrected setting value is as follows: The tuning sensitivity Lm is calculated based on the fault data in the simulation dataset and the tuning value. The tuning sensitivity Lm is compared with the preset sensitivity threshold Lm0. The tuning value is adjusted according to the comparison result to obtain the corrected tuning value, wherein: When Lm≥Lm0, it is determined that the setting value does not need to be adjusted, the risk level of the setting value is marked as low risk, and the setting value is marked as the corrected setting value; When Lm < Lm0, the setpoint is determined to need adjustment. The setpoint corresponding to the setpoint sensitivity Lm is adjusted based on a preset adjustment coefficient, where the adjustment coefficient is set to a constant. The adjustment coefficient is added to the setpoint to obtain the adjusted setpoint. The adjusted sensitivity Lm1 is calculated based on the adjusted setpoint. The adjusted sensitivity Lm1 is then compared again with the preset sensitivity threshold Lm0, where: When Lm1≥Lm0, it is determined that the adjusted setpoint does not need to be adjusted, the risk level of the adjusted setpoint is marked as low risk, and the setpoint is marked as the corrected setpoint. When Lm < Lm0, it is determined that the adjusted value needs to be adjusted again. Based on the preset adjustment coefficient, the adjusted value corresponding to the adjusted sensitivity Lm is adjusted again. The risk level of the adjusted value obtained after the second adjustment is marked as low risk, and the adjusted value obtained after the second adjustment is marked as the corrected adjusted value.

[0008] Furthermore, when the adjusted sensitivity Lm1 is less than the preset sensitivity threshold Lm0, the preset sensitivity threshold Lm0 is corrected based on the electrical system model, specifically as follows: Obtain the variation characteristics of the power plant electrical system corresponding to the electrical system model, and adjust the topology of the electrical system model based on the variation characteristics to obtain a new topology; The impedance matrix is ​​calculated for the new topology to obtain a new impedance matrix, and the short-circuit current is calculated based on the new impedance matrix and the rated voltage of the electrical system model. The new sensitivity threshold is calculated based on the short-circuit current and the setting value of the electrical system model, and the new sensitivity threshold is marked as the preset sensitivity threshold Lm0.

[0009] Furthermore, the specific process of using the parameter tracing method to collect parameters of the power plant's electrical system to obtain a dynamic parameter set is as follows: An initialization sequence is randomly generated, and a state sequence is obtained by performing a random walk proposal on the initialization sequence based on a preset proposal distribution. Randomly generate observation sequences, and calculate the probability distribution of the observation sequences based on a preset posterior distribution; The acceptance probability Pa is calculated based on the state sequence and probability distribution. Pa is then compared with a preset acceptance threshold Pa0, where Pa0 = 0.15. The state sequence is assessed based on the comparison result, and updated accordingly to obtain a dynamic parameter set. When Pa < Pa0, the state sequence is determined to be unnecessary to update, and the state sequence is marked as a dynamic parameter set; When Pa≥Pa0, the state sequence is determined to need to be updated. Based on the preset selection threshold, data in the probability distribution is selected to obtain selected data. The mean of the selected data is calculated to obtain the updated value. Based on the time series, the updated value is used to update the data in the state sequence to obtain the dynamic parameter set.

[0010] Furthermore, the specific process of associating and mapping the topology information and dynamic parameter set of the power plant's electrical system to obtain a digital twin model is as follows: Extract the acquisition nodes and time points of the dynamic parameter set data, and extract the topology nodes and connection relationships from the topology information of the power plant's electrical system; A topology graph is drawn based on the topology nodes and their connections, and the topology nodes are compared with the data collection nodes to obtain the correspondence between the topology nodes and the data collection nodes. The dynamic parameter set is divided based on the correspondence between the topological nodes and the acquisition nodes to obtain a node subset; Sort the data in the node subset using time points to obtain the ordered data; By filling the topological nodes of the topological graph with the sequential data, a digital twin model is obtained.

[0011] Furthermore, the specific process of obtaining a local constant value solution set by hierarchical optimization of the digital twin model based on the rule constraint set is as follows: Improve the multi-objective optimization algorithm by using a set of rule constraints as the constraint conditions; Extract the nodes and node connections from the digital twin model, and determine the node level of the nodes based on the node connections to obtain the master node; The digital twin model is systematically divided based on the master node to obtain subsystems; The subsystem is optimized as the target individual of the improved multi-objective optimization algorithm to obtain the fixed-value solution of the subsystem. The local set of constant solutions is obtained by organizing the constant solutions of the subsystem.

[0012] Furthermore, the specific process of obtaining the rule constraint set by performing hybrid semantic parsing on the protection setting guide is as follows: Semantic rule constraints are obtained by semantic recognition of the protection setting guidelines using the pre-trained model BERT. The initial rule constraints are obtained by logically reasoning the protection setting guidelines through the rule engine. The first confidence level Cc1 is obtained by calculating the confidence level of the semantic rule constraints using a preset confidence level formula. The initial rule constraints are matched with confidence using a preset rule matching strategy to obtain the second confidence level Cc2. The first confidence level Cc1 is compared with the second confidence level Cc2. Based on the comparison result, it is determined whether the initial rule constraints are qualified. The qualified initial rule constraints are then organized to obtain the rule constraint set, where: When Cc1≤Cc2, the initial rule constraint is deemed qualified; When Cc1 > Cc2, the initial rule constraint is deemed unqualified, and the initial rule constraint is corrected by semantic rule constraint to obtain a qualified initial rule constraint.

[0013] Furthermore, the specific process of modifying the initial rule constraints through semantic rule constraints to obtain qualified initial rule constraints is as follows: The confidence difference Cc3 is obtained by subtracting the first confidence level Cc1 from the second confidence level Cc2. This confidence difference Cc3 is then compared with a preset difference threshold Cc4. Based on the comparison result, the initial rule constraints are corrected to obtain qualified initial rule constraints, where: When Cc3 > Cc4, it is determined that the initial rule constraint needs to be modified, the initial rule constraint is deleted, and the semantic rule constraint is marked as a qualified initial rule constraint. When Cc3≤Cc4, it is determined that the initial rule constraint does not need to be modified, and the initial rule constraint is marked as a qualified initial rule constraint; Once the initial rule constraints are corrected, the qualified initial rule constraints are organized to obtain the rule constraint set.

[0014] Furthermore, the qualified initial rule constraints are verified. When the verification of the initial rule constraints fails, the difference threshold Cc4 is adjusted, and the adjusted difference threshold Cc4 is compared with the confidence difference Cc3 again. Specifically, this includes: The logic of the qualified initial rule constraints is analyzed using preset logical criteria to determine logical contradiction points Md. These contradiction points Md are then compared with a preset contradiction point threshold Md0. Based on the comparison result, the adjustment requirement for the difference threshold is determined, and the difference threshold Cc4 is adjusted accordingly. When Md≤Md0, it is determined that the difference threshold Cc4 does not need to be adjusted, and the difference threshold Cc4 is marked as the adjusted difference threshold Cc4; When Md > Md0, the difference threshold Cc4 is determined to need adjustment. The difference threshold Cc4 is added to the preset threshold adjustment amount to obtain the adjusted difference threshold Cc4.

[0015] Compared with existing technologies, the multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy described in this invention has the following advantages: This invention utilizes a parameter tracing method to collect dynamic parameter sets from the parameters of a power plant's electrical system. This integrates parameters from multiple devices within the power plant's electrical system, including the SCADA system, fault recording devices, and equipment nameplates, effectively eliminating spatiotemporal differences and fluctuations in these parameters, such as impedance fluctuations caused by temperature drift. This significantly improves the accuracy of the dynamic parameter set. Furthermore, a digital twin model is obtained by mapping the topology information and the dynamic parameter set together, effectively enhancing the accuracy of the digital twin model and laying the foundation for subsequent setting calculations. Secondly, this invention employs a hybrid semantic parsing framework to perform hybrid semantic parsing on the protection setting guidelines corresponding to the power plant's electrical system, obtaining a set of rule constraints. The protection setting guidelines, written in natural language, are transformed into machine-executable mathematical constraints. During this transformation, the mathematical constraints undergo multi-dimensional comparison, i.e., correction is made based on the comparison results of the first confidence level Cc1 and the second confidence level Cc2. This effectively avoids logical recognition errors caused by semantic ambiguity in the protection setting guidelines, greatly improving the accuracy of the mathematical constraints and ensuring the correctness of the rule constraint set. However, due to the… The complex structure of power plant electrical systems leads to a complex digital twin model. This complexity results in high computational resource requirements and long calculation times for setting value calculation. To reduce computational resource requirements and improve calculation speed, the digital twin model is divided into multiple subsystems according to the logical hierarchy of the power plant electrical system. A set of rules and constraints is used as constraints for a multi-objective optimization algorithm. The subsystems are input into the multi-objective optimization algorithm to obtain the setting solutions, i.e., local setting solutions. Simultaneously, consistency coordination is performed on the local setting solutions to obtain the setting values. This allows the local setting solutions to fully track the decomposition process of the digital twin model, eliminating boundary protection conflicts between different subsystems and effectively improving the matching degree between the setting values ​​and the digital twin model. The setting values ​​of the relay protection modules in the digital twin model are adjusted to match the setting values ​​to obtain the electrical system model. A risk assessment is then performed on the electrical system model to obtain the assessment results, realizing the setting scheme—a quantitative assessment of the robustness of the setting values ​​in the face of electrical system uncertainties—and thus enabling effective prediction of potential risks. Finally, a setting value calculation report is generated based on the assessment results. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy, as described in this embodiment. Detailed Implementation

[0017] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0018] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0020] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] like Figure 1 As shown, this embodiment provides a multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy, including the following sub-steps: S1. Obtain the topology information of the power plant's electrical system and obtain the protection setting guidelines; S2. Use the parameter tracing method to collect dynamic parameter sets from the parameters of the power plant's electrical system. S21. Randomly generate an initialization sequence, and perform a random walk proposal on the initialization sequence based on a preset proposal distribution to obtain a state sequence; S22. Randomly generate observation sequences and calculate the probability distribution of the observation sequences based on the preset posterior distribution; S23. Calculate the acceptance probability Pa based on the state sequence and probability distribution. Compare the acceptance probability Pa with the preset acceptance threshold Pa0, where Pa0 = 0.15. Determine the state sequence status based on the comparison result and update the state sequence accordingly to obtain a dynamic parameter set, where: When Pa < Pa0, the state sequence is determined to be unnecessary to update, and the state sequence is marked as a dynamic parameter set; When Pa ≥ Pa0, the state sequence is determined to need updating. Data from the probability distribution is selected based on a preset selection threshold. The mean of the selected data is calculated to obtain the update value. The update value is then used to update the data in the state sequence based on the time series to obtain a dynamic parameter set. The preset selection threshold is essentially the probability of successful updates in the historical state sequence. It is obtained by statistically analyzing the number of updated data in the historical state sequence and dividing that number by the total number of data in the historical state sequence. Simultaneously, the probability in the probability distribution is compared with the selection threshold. When the probability in the probability distribution... When selecting a threshold, the observed values ​​in the observation sequence corresponding to the probability in the probability distribution are marked as selected data, and all selected data are input into the mean formula to calculate the average value of the selected data. The average value is marked as the updated value. In this embodiment, the mean formula is not specifically limited, such as the arithmetic mean formula. After the updated value is obtained by calculating the updated value through the mean, the time series of the state sequence is matched with the time series of the observation sequence to determine the observed values ​​marked as selected data and the state values ​​at the same time points as the observed values. The state values ​​are the data in the state sequence, and the observed values ​​are the data in the observation sequence. The state values ​​are replaced with the updated values ​​to update the state sequence and obtain a dynamic parameter set.

[0022] Specifically, the parameters of the power plant electrical system include at least the generator operating data (such as generator current, voltage, and frequency), transformer operating data (such as transformer current, voltage, and frequency), transmission line operating data (such as transmission line current, voltage, and frequency), SCADA system operating data (such as circuit breaker closing status, monitoring voltage and frequency of equipment in the power plant electrical system), fault recordings, equipment nameplates (such as rated voltage, rated current, and rated frequency of equipment in the power plant electrical system), transformer ratio data (such as the ratio of current transformers and voltage transformers), and line impedance (such as the impedance of transmission lines in the power plant electrical system). The fault recordings are essentially data obtained when a fault occurs in the power plant electrical system. The fault waveform data is obtained because when the power plant's electrical system changes, such as when the operating mode of a new energy power generation device switches, the parameters of the power plant's electrical system change accordingly. For example, when the operating mode of a new energy power generation device switches, the topology of the power plant's electrical system changes, leading to changes in the impedance of the transmission lines. This makes it difficult to collect the parameters of the power plant's electrical system. To improve the accuracy of parameter collection, this embodiment uses a parameter tracing method to collect the parameters of the power plant's electrical system. This embodiment does not limit the specific implementation of the parameter tracing method; those skilled in the art can freely choose according to actual needs, such as Markov chains. The Monte Carlo sampling method sets state variables based on historical parameters of the power plant's electrical system. These historical parameters are the operating parameters of the power plant's electrical system during the same historical period. The state variables are essentially the changes in these historical parameters during that same historical period. For example, if the current time is 1 PM on the 16th, the same historical period is 1 PM on the 15th. The parameters of the power plant's electrical system at 1 PM on the 15th are the historical parameters, and the changes in these historical parameters compared to the parameters at 12 PM on the 15th are the state variables. These state variables are used to adjust the observation model, posterior distribution, proposal distribution, and likelihood function in the Markov chain Monte Carlo sampling method, thereby improving the compatibility of the Markov chain Monte Carlo sampling method with the power plant's electrical system. To ensure the adaptability of the Markov chain Monte Carlo sampling method, which can collect parameters of the power plant's electrical system, the original initial state sequence of the Markov chain Monte Carlo sampling method is first randomly initialized to obtain an initialization sequence. Then, a random walk proposal is performed on the initialization sequence based on a proposal distribution to obtain a state sequence. This involves iteratively calculating the state sequence by inputting the initialization sequence into the proposal distribution. The iteration threshold for this calculation is set to a constant, such as 1000. This iteration threshold represents the maximum number of iterations set to avoid getting stuck in infinite iteration. An observation sequence is then randomly generated based on the observation model. This observation sequence is input into the posterior distribution to calculate the probability distribution. Finally, the acceptance probability Pa is calculated based on the state sequence and the probability distribution.The specific formula for calculating the acceptance probability Pa is: ; In the formula, In the observation model, the observed values Candidate state under the condition The posterior probability, For the observed values Under the condition of the current state The posterior probability, To transition from candidate state Transition to the current state The probability of a proposal, To start from the current state Transition to candidate state The proposal probability is calculated, and then the acceptance probability Pa is used to determine whether to update the state value of the state sequence in order to simulate the change process of dynamic parameters in the power plant electrical system and eliminate the influence of parameter fluctuations.

[0023] Secondly, the parameters of the power plant's electrical system exist in different physical units, such as line impedance, voltage, current, and power in generator operating data. If the raw data is used directly for calculation, the accuracy of the calculation will be low due to the dimensional issues caused by the different physical units. Therefore, this embodiment standardizes the parameters of the power plant's electrical system before performing the calculation to eliminate the dimensional influence between different parameters. The standard formula corresponding to the standardization process is as follows: ; In the formula, R represents the parameter used for standardization. For parameters In the The value at each moment, For parameters The minimum value, For parameters The maximum value.

[0024] S3. The topology information and dynamic parameter set of the power plant electrical system in S1 are correlated and mapped to obtain a digital twin model; S31. Extract the acquisition nodes and time points of the dynamic parameter set data, and extract the topology nodes and connection relationships in the topology information of the power plant's electrical system. S32. Draw a topology graph based on the topology nodes and their connections, and compare the topology nodes with the data acquisition nodes to obtain the correspondence between the topology nodes and the data acquisition nodes; S33. Based on the correspondence between the topology nodes and the acquisition nodes, the dynamic parameter set is divided to obtain a node subset; S34. Sort the data in the node subset using time points to obtain the ordered data; S35. Fill the topological nodes of the topological graph with the sequence data to obtain the digital twin model.

[0025] Specifically, using equipment in the power plant's electrical system as nodes and lines as connections, the topology information of the power plant's electrical system is transformed into a topology diagram, realizing the drawing conversion of the power plant's electrical system. The topology nodes in the topology diagram are compared with the data acquisition nodes of the collected parameters. According to the logical hierarchy of the corresponding power plant electrical system, the topology nodes are divided into nodes of different levels, obtaining node subsets. For example, according to the logical hierarchy of the power plant electrical system, the transmission lines of the power plant electrical system are divided into main lines and branch lines. The nodes corresponding to the main lines are marked as main nodes, and the nodes corresponding to the branch lines are marked as secondary nodes. The main nodes and secondary nodes are organized to obtain node subsets. Secondly, since the parameters of each node are a set of data collected based on time changes, the parameters in the dynamic parameter set are filled into the node subset according to the time series when the parameters of each node are collected, realizing the visualization of the power plant electrical system.

[0026] S4. Perform hybrid semantic parsing on the protection setting guidelines to obtain the rule constraint set; S41. Semantic rule constraints are obtained by semantically recognizing the protection setting guidelines using the pre-trained model BERT. S42. The initial rule constraints are obtained by logically reasoning the protection setting guidelines through the rule engine. S43. Calculate the confidence of the semantic rule constraints using a preset confidence formula to obtain the first confidence level Cc1; S44. The initial rule constraints are matched with confidence using a preset rule matching strategy to obtain the second confidence level Cc2. S45. Compare the first confidence level Cc1 with the second confidence level Cc2, determine whether the initial rule constraints are qualified based on the comparison results, and organize the qualified initial rule constraints to obtain the rule constraint set, where: When Cc1≤Cc2, the initial rule constraint is deemed qualified; When Cc1 > Cc2, the initial rule constraint is deemed unqualified, and the initial rule constraint is corrected by semantic rule constraint to obtain a qualified initial rule constraint. S451, the specific process of correcting the initial rule constraints through semantic rule constraints to obtain qualified initial rule constraints is as follows: The first confidence level Cc1 and the second confidence level Cc2 are subtracted to obtain a confidence difference value Cc3, and the confidence difference value Cc3 is compared with a preset difference threshold Cc4. Based on the comparison result, the initial rule constraints are corrected to obtain qualified initial rule constraints, wherein: When Cc3 > Cc4, it is determined that the initial rule constraint needs to be modified, the initial rule constraint is deleted, and the semantic rule constraint is marked as a qualified initial rule constraint. When Cc3≤Cc4, it is determined that the initial rule constraint does not need to be modified, and the initial rule constraint is marked as a qualified initial rule constraint; Once the initial rule constraints are corrected, the qualified initial rule constraints are organized to obtain the rule constraint set.

[0027] Specifically, the protection setting guidelines include clauses from DL / T 684 and GB / T 14285. These clauses are all described in natural language, leading to semantic ambiguity. To eliminate this ambiguity and transform all clauses into machine-recognizable mathematical constraints, this embodiment employs a hybrid semantic parsing framework of BERT + rule engine to transform the protection setting guidelines. Specifically, the protection setting guidelines are input into a BERT model to obtain semantic rule constraints output by the BERT model. While typical BERT models usually only recognize the semantics of the input data to obtain rule elements, the BERT model in this embodiment adds an output auxiliary module. This module processes the rule elements recognized by the BERT model to obtain semantic rule constraints. For example, when the input data to the BERT model is a maximum current of 10 amperes, the rule elements recognized by the BERT model are current, 10 amperes, and maximum. The output auxiliary module processes current, 10 amperes, and maximum to obtain current I. 10 amperes; Initial rule constraints are obtained by logically reasoning the protection setting guideline through a rule engine. It's important to note that both the semantic rule constraints and the initial rule constraints are transformations of the protection setting guideline. However, due to semantic ambiguity, the correctness of a single transformation result is difficult to guarantee. Therefore, a first confidence level is obtained by calculating the confidence level of the semantic rule constraints using a preset confidence formula, such as the Softmax formula. A second confidence level is obtained by matching the initial rule constraints with the confidence level using a preset rule matching strategy. This confidence matching involves comparing the logic of the initial rule constraints with the logic of the protection setting guideline. The degree of conflict between the logic of the initial rule constraints and the logic of the protection setting guidelines is examined. The degree of conflict is marked as the confidence level of the initial rule constraints, and a second confidence level is obtained. The first confidence level and the second confidence level are compared, and the initial rule constraints are corrected according to the comparison result. Specifically, when the first confidence level > the second confidence level, it is determined that the semantic rule constraints are closer to the original protection setting guidelines, and the semantic rule constraints are regarded as qualified initial rule constraints. When the first confidence level ≤ the second confidence level, it is determined that the initial rule constraints are closer to the original protection setting guidelines, and the initial rule constraints are regarded as qualified initial rule constraints, which effectively improves the accuracy of the initial rule constraints.

[0028] In addition, after obtaining qualified initial rule constraints, the qualified initial rule constraints are verified. If the verification of the initial rule constraints fails, the difference threshold Cc4 is adjusted, and the adjusted difference threshold Cc4 is compared with the confidence difference Cc3 again. Specifically, this includes: The logic of the qualified initial rule constraints is analyzed using preset logical criteria to determine logical contradiction points Md. These contradiction points Md are then compared with a preset contradiction point threshold Md0. Based on the comparison result, the adjustment requirement for the difference threshold is determined, and the difference threshold Cc4 is adjusted accordingly. The logical contradiction point Md is essentially the point where the logic of the qualified initial rule constraints contradicts the logic of the preset logical criteria; that is, the conflict point between the logic criteria and the logic of the initial rule constraints. When Md≤Md0, it is determined that the difference threshold Cc4 does not need to be adjusted, and the difference threshold Cc4 is marked as the adjusted difference threshold Cc4; When Md > Md0, the difference threshold Cc4 is determined to need adjustment. The difference threshold Cc4 is added to the preset threshold adjustment amount to obtain the adjusted difference threshold Cc4. The preset threshold adjustment amount is a constant and is set to 0.05.

[0029] Specifically, to further improve the accuracy of initial rule constraints, this implementation also verifies the qualified initial rule constraints. By using preset logical criteria, the logic of the qualified initial rule constraints is analyzed to determine logical contradiction points Md. These contradiction points essentially represent the degree of conflict between the logic of the initial rule constraints and the logic of the preset logical criteria. The preset logical criteria are essentially constraint logic set according to the operating logic of the power plant's electrical system, such as a maximum voltage of 360V. The difference threshold Cc4 is adjusted based on the comparison between the logical contradiction point Md and the preset contradiction point threshold Md0. When Md ≤ Md0, it proves that the degree of logical conflict between the initial rule constraints and the preset logical criteria is less than equal. At 90%, the initial rule constraint simultaneously satisfies the protection setting criterion and the requirements of the power plant electrical system. When Md > Md0, it proves that the logical conflict between the initial rule constraint and the preset logical criterion is greater than 90%, the initial rule constraint does not simultaneously satisfy the protection setting criterion and the requirements of the power plant electrical system, and the preset contradiction point threshold Md0 is set to 90%.

[0030] S5. Based on the set of rule constraints, perform hierarchical optimization on the digital twin model to obtain a local constant value solution set; S51. Improve the multi-objective optimization algorithm by using the rule constraint set as the constraint condition; S52. Extract the nodes and node connection relationships from the digital twin model, and determine the node level of the nodes based on the node connection relationships to obtain the master node; S53. Based on the master node, the digital twin model is systematically divided to obtain subsystems; S54. Optimize the subsystem as the target individual of the improved multi-objective optimization algorithm to obtain the fixed solution of the subsystem; S55. Organize the constant solutions of the subsystem to obtain a local constant solution set; S6. Consistency coordination is performed on the local fixed value solution set to obtain the setpoint.

[0031] Specifically, the multi-objective optimization algorithm has the ability to resolve multiple conflicting objectives. However, in the calculation of the setpoint, the four principles—selectivity, speed, sensitivity, and reliability—are contradictory. Therefore, the four principles are used as the objectives of the multi-objective optimization algorithm, the rule constraint set is used as the constraint condition, and the parameter combination of the digital twin model is used as the input to run the multi-objective optimization algorithm to obtain the setpoint of the power plant electrical system corresponding to the digital twin model. However, the structure of the digital twin model is complex. If it is treated as a whole and the parameter combination of the digital twin model is input into the multi-objective optimization algorithm, the algorithm will require a large amount of computation. Resource constraints lead to low computational efficiency. To address this, the digital twin model is divided according to node hierarchy, with the master node as the boundary. The secondary nodes corresponding to the master node and their connections are considered subsystems, resulting in multiple relatively simple subsystems. A multi-objective optimization algorithm is used to calculate the parameter combinations of these subsystems, yielding their setpoints, or fixed-value solutions. These fixed-value solutions are essentially local fixed-value solutions of the digital twin model. Furthermore, a consistency coordination algorithm, such as the ADMM algorithm, is used to integrate the setpoints of the subsystems, eliminating boundary protection conflicts between different subsystems and ensuring the coordinated optimization of the four principles (property, performance, and stability) across the entire digital twin model, i.e., the global scope.

[0032] S7. Adjust the setting value of the relay protection module of the digital twin model to the value of the setting value to obtain the electrical system model, and perform a risk assessment on the electrical system model to obtain the assessment result. S71. Perform fault simulation on the electrical system model to obtain the simulation dataset; S72. The number of correct actions and the number of incorrect actions are obtained from the statistical simulation dataset. When a fault occurs in the power plant's electrical system, the fault recording data of the power plant's electrical system will show characteristic waveforms, such as voltage drops and current surges. The fault point is obtained from the characteristic waveforms of the fault recording data in the statistical simulation dataset. The fault data such as current and voltage at the fault point are compared with the setting value. If the fault data does not meet the requirements of the setting value, such as the current being greater than the maximum current value specified by the setting value, the protection action corresponding to the fault point is determined to be an incorrect action. Otherwise, the protection action corresponding to the fault point is determined to be a correct action. The number of incorrect actions and the number of correct actions are counted. The number of fault points is essentially the total number of protection actions. The risk entropy Qp of the electrical system model is calculated based on the number of correct actions and the number of incorrect actions. S73. Compare the risk entropy Qp with the preset risk entropy Qp0, determine the risk level of the electrical system model's setting value based on the comparison result, and correct the setting value based on the determination result, wherein: When Qp≤Qp0, the setting value is determined to be low risk; When Qp > Qp0, the setting value is determined to be high risk. The setting value is then corrected based on the fault data in the simulation dataset to obtain the corrected setting value. S731, the specific process of correcting the setting value based on the fault data in the simulation dataset is as follows: Calculate the setting sensitivity Lm based on the fault data in the simulation dataset and the setting value; compare the setting sensitivity Lm with the preset sensitivity threshold Lm0; adjust the setting value according to the comparison result to obtain the corrected setting value, wherein: When Lm≥Lm0, it is determined that the setting value does not need to be adjusted, the risk level of the setting value is marked as low risk, and the setting value is marked as the corrected setting value; When Lm < Lm0, the setpoint is determined to need adjustment. The setpoint corresponding to the setpoint sensitivity Lm is adjusted based on a preset adjustment coefficient, where the adjustment coefficient is set to a constant (e.g., 1). The adjustment coefficient is added to the setpoint to obtain the adjusted setpoint. The adjusted sensitivity Lm1 is calculated based on this adjusted setpoint. The adjusted sensitivity Lm1 is then compared again with the preset sensitivity threshold Lm0, where: When Lm1≥Lm0, it is determined that the adjusted setpoint does not need to be adjusted, the risk level of the adjusted setpoint is marked as low risk, and the setpoint is marked as the corrected setpoint. When Lm < Lm0, it is determined that the adjusted value needs to be adjusted again. Based on the preset adjustment coefficient, the adjusted value corresponding to the adjusted sensitivity Lm is adjusted again. The risk level of the adjusted value obtained after the second adjustment is marked as low risk, and the adjusted value obtained after the second adjustment is marked as the corrected adjusted value.

[0033] Specifically, a power system digital simulation platform, such as the EMTP-RV platform, is used to simulate the digital twin model in various operating modes, including maximum, minimum, and N-1 modes. During the simulation, fault states of the digital twin model are simulated, such as three-phase short circuits, two-phase grounding, and open circuits. The correctness of the protection actions of the digital twin model is determined based on the current and voltage readings during the simulation. For example, if a voltage drop or current surge occurs, indicating a fault in the digital twin model, the voltage and current readings of the digital twin model are extracted and compared with the set values. If the voltage exceeds the maximum voltage value specified in the setting, or the current exceeds the maximum current value specified in the setting, the protection action of the digital twin model is determined to be an erroneous action; otherwise, the protection action of the digital twin model is determined to be a correct action. The fault waveforms, protection action sequences, erroneous protection actions, correct protection actions, and measured electrical quantities such as current and voltage generated during the simulation operation are compiled into a simulation dataset. Then, the number of correct actions and the number of erroneous actions are statistically analyzed from the simulation dataset. Based on the number of correct actions and the number of erroneous actions, the probability of incorrect action of the corresponding protection device is calculated. The specific formula for calculating the probability of incorrect action is as follows: ; In the formula, TY is the probability of incorrect action, ty2 is the number of correct actions, and ty1 is the number of incorrect actions; the incorrect actions include failure to act and false action. When the digital twin model meets the action conditions but does not perform the protection action, it is determined that the digital twin model fails to act. When the digital twin model does not meet the action conditions but performs the protection action, it is determined that the digital twin model falsely acts. The action conditions are essentially the data thresholds specified by the setting value, such as the maximum current and the maximum voltage. The risk entropy is calculated by inputting the probability of the incorrect action into the risk entropy formula, which is as follows: ; In the formula, Risk entropy; When the risk entropy is less than or equal to 0.1, it proves that the corresponding tuning scheme is stable and robust under various disturbances. When the risk entropy is greater than 0.1, it proves that the corresponding tuning scheme is unstable and has poor robustness under various disturbances. Therefore, it is necessary to use the fault data in the simulation dataset to correct the tuning value to improve the robustness of the tuning scheme. The sensitivity Lm is calculated based on the fault data and the tuning value. The calculation formula for the sensitivity Lm is as follows: Ksen= ; In the formula, Ksen is the sensitivity Lm, Zd is the maximum voltage or current specified by the setting value, and Gz is the fault data, such as fault voltage and fault current. And compare the sensitivity Lm with the preset sensitivity threshold Lm0. When Lm When Lm = 0, it proves that the setting value has reached the optimal state and no adjustment is needed. When Lm = 0... When m0, it is proven that the setting value still has a certain adjustment range. The setting value corresponding to the setting sensitivity Lm is adjusted based on the preset adjustment coefficient. That is, the setting value is added to the preset adjustment coefficient to obtain the adjusted setting value. The adjusted setting value is then input into the sensitivity calculation formula to calculate the adjusted sensitivity Lm1. The preset adjustment coefficient is set to 0.05 to avoid fluctuations in the setting value during adjustment. The sensitivity threshold Lm0 is the short-circuit current of the digital twin model.

[0034] Furthermore, when the adjusted sensitivity Lm1 is less than the preset sensitivity threshold Lm0, it indicates that the topology of the power plant's electrical system has changed. The preset sensitivity threshold Lm0 is no longer applicable to the power plant's electrical system after the topology change, and the topology of the electrical system model corresponding to the power plant's electrical system has also changed. Therefore, the preset sensitivity threshold Lm0 is corrected based on the change in the topology of the electrical system model, specifically as follows: Obtain the variation characteristics of the power plant electrical system corresponding to the electrical system model, and adjust the topology of the electrical system model based on the variation characteristics to obtain a new topology; The impedance matrix is ​​calculated for the new topology to obtain a new impedance matrix, and the short-circuit current is calculated based on the new impedance matrix and the rated voltage of the electrical system model. The new sensitivity threshold is calculated based on the short-circuit current and the setting value of the electrical system model, and the new sensitivity threshold is marked as the preset sensitivity threshold Lm0.

[0035] Specifically, when the power plant electrical system corresponding to the digital twin model changes, such as a change in the operating mode of the new energy power generation device, the setting value of the digital twin model will also change. At this time, the sensitivity threshold Lm0 of the power plant electrical system also needs to be adjusted; otherwise, the accuracy of the sensitivity Lm determined based on the sensitivity threshold Lm0 will decrease. Therefore, based on the changed topology of the power plant electrical system, the changed impedance of the power plant electrical system is calculated to obtain a new impedance matrix. Ohm's law is then used to calculate the short-circuit current corresponding to the new impedance matrix. Based on the short-circuit current and the original setting value, a new short-circuit current is calculated. The specific calculation formula for the new short-circuit current is as follows: Inew=

[0036] In the formula, Inew is the new short-circuit current, Un is the rated voltage of the power plant electrical system, Iset is the rated current of the power plant electrical system, Znew is the new impedance matrix, and Z1 is the new impedance matrix of the branch corresponding to the new short-circuit current.

[0037] S74. Organize the low-risk setpoints and the corrected setpoints to obtain the risk assessment results; S8. Generate a fixed value calculation sheet based on the evaluation results.

[0038] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy, characterized in that, The method includes the following sub-steps: Obtain the topology information of the power plant's electrical system and obtain the protection setting guidelines; A dynamic parameter set is obtained by tracing and collecting parameters of the power plant's electrical system using the parameter tracing method; A digital twin model is obtained by associating and mapping the topology information and dynamic parameter set of the power plant's electrical system. Hybrid semantic parsing is used to obtain the rule constraint set from the protection setting guidelines; A local constant value solution set is obtained by hierarchically optimizing the digital twin model based on the rule constraint set; The setpoint is obtained by unifying the local constant value solution set. The setting values ​​of the relay protection module in the digital twin model are adjusted to the value of the setting to obtain the electrical system model, and a risk assessment is performed on the electrical system model to obtain the assessment results. A fixed value calculation sheet is generated based on the evaluation results.

2. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 1, characterized in that, The specific process for obtaining the assessment results by performing a risk assessment on the electrical system model is as follows: Fault simulation is performed on the electrical system model to obtain the simulation dataset; The number of correct actions and the number of incorrect actions are obtained from the statistical simulation dataset, and the risk entropy Qp of the electrical system model is calculated based on the number of correct actions and the number of incorrect actions. The risk entropy Qp is compared with the preset risk entropy Qp0. The risk level of the setting values ​​of the electrical system model is judged based on the comparison results, and the setting values ​​are corrected based on the judgment results, wherein: When Qp≤Qp0, the setting value is determined to be low risk; When Qp > Qp0, the setting value is determined to be high risk. The setting value is then corrected based on the fault data in the simulation dataset to obtain the corrected setting value. By compiling the low-risk setpoints and the corrected setpoints, the risk assessment results are obtained.

3. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 2, characterized in that, The specific process of correcting the setting value based on the fault data in the simulation dataset to obtain the corrected setting value is as follows: The tuning sensitivity Lm is calculated based on the fault data in the simulation dataset and the tuning value. The tuning sensitivity Lm is compared with the preset sensitivity threshold Lm0. The tuning value is adjusted according to the comparison result to obtain the corrected tuning value, wherein: When Lm≥Lm0, it is determined that the setting value does not need to be adjusted, the risk level of the setting value is marked as low risk, and the setting value is marked as the corrected setting value; When Lm < Lm0, the setpoint is determined to need adjustment. The setpoint corresponding to the setpoint sensitivity Lm is adjusted based on a preset adjustment coefficient, where the adjustment coefficient is set to a constant. The adjustment coefficient is added to the setpoint to obtain the adjusted setpoint. The adjusted sensitivity Lm1 is calculated based on the adjusted setpoint. The adjusted sensitivity Lm1 is then compared again with the preset sensitivity threshold Lm0, where: When Lm1≥Lm0, it is determined that the adjusted setpoint does not need to be adjusted, the risk level of the adjusted setpoint is marked as low risk, and the setpoint is marked as the corrected setpoint. When Lm < Lm0, it is determined that the adjusted value needs to be adjusted again. Based on the preset adjustment coefficient, the adjusted value corresponding to the adjusted sensitivity Lm is adjusted again. The risk level of the adjusted value obtained after the second adjustment is marked as low risk, and the adjusted value obtained after the second adjustment is marked as the corrected adjusted value.

4. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 3, characterized in that, When the adjusted sensitivity Lm1 is less than the preset sensitivity threshold Lm0, the preset sensitivity threshold Lm0 is corrected based on the electrical system model, specifically as follows: Obtain the variation characteristics of the power plant electrical system corresponding to the electrical system model, and adjust the topology of the electrical system model based on the variation characteristics to obtain a new topology; The impedance matrix is ​​calculated for the new topology to obtain a new impedance matrix, and the short-circuit current is calculated based on the new impedance matrix and the rated voltage of the electrical system model. The new sensitivity threshold is calculated based on the short-circuit current and the setting value of the electrical system model, and the new sensitivity threshold is marked as the preset sensitivity threshold Lm0.

5. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 1, characterized in that, The specific process of using the parameter tracing method to collect parameters of the power plant's electrical system to obtain a dynamic parameter set is as follows: An initialization sequence is randomly generated, and a state sequence is obtained by performing a random walk proposal on the initialization sequence based on a preset proposal distribution. Randomly generate observation sequences, and calculate the probability distribution of the observation sequences based on a preset posterior distribution; The acceptance probability Pa is calculated based on the state sequence and probability distribution. Pa is then compared with a preset acceptance threshold Pa0, where Pa0 = 0.

15. The state sequence is assessed based on the comparison result, and updated accordingly to obtain a dynamic parameter set. When Pa < Pa0, the state sequence is determined to be unnecessary to update, and the state sequence is marked as a dynamic parameter set; When Pa≥Pa0, the state sequence is determined to need to be updated. Based on the preset selection threshold, data in the probability distribution is selected to obtain selected data. The mean of the selected data is calculated to obtain the updated value. Based on the time series, the updated value is used to update the data in the state sequence to obtain the dynamic parameter set.

6. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 1, characterized in that, The specific process of associating and mapping the topology information and dynamic parameter set of the power plant's electrical system to obtain a digital twin model is as follows: Extract the acquisition nodes and time points of the dynamic parameter set data, and extract the topology nodes and connection relationships from the topology information of the power plant's electrical system; A topology graph is drawn based on the topology nodes and their connections, and the topology nodes are compared with the data collection nodes to obtain the correspondence between the topology nodes and the data collection nodes. The dynamic parameter set is divided based on the correspondence between the topological nodes and the acquisition nodes to obtain a node subset; Sort the data in the node subset using time points to obtain the ordered data; By filling the topological nodes of the topological graph with the sequential data, a digital twin model is obtained.

7. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 1, characterized in that, The specific process of obtaining a local constant value solution set by hierarchical optimization of the digital twin model based on the rule constraint set is as follows: Improve the multi-objective optimization algorithm by using a set of rule constraints as the constraint conditions; Extract the nodes and node connections from the digital twin model, and determine the node level of the nodes based on the node connections to obtain the master node; The digital twin model is systematically divided based on the master node to obtain subsystems; The subsystem is optimized as the target individual of the improved multi-objective optimization algorithm to obtain the fixed-value solution of the subsystem. The local set of constant solutions is obtained by organizing the constant solutions of the subsystem.

8. The multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 1, characterized in that, The specific process of obtaining the rule constraint set by performing hybrid semantic parsing on the protection setting guide is as follows: Semantic rule constraints are obtained by semantic recognition of the protection setting guidelines using the pre-trained model BERT. The initial rule constraints are obtained by logically reasoning the protection setting guidelines through the rule engine. The first confidence level Cc1 is obtained by calculating the confidence level of the semantic rule constraints using a preset confidence level formula. The initial rule constraints are matched with confidence using a preset rule matching strategy to obtain the second confidence level Cc2. The first confidence level Cc1 is compared with the second confidence level Cc2. Based on the comparison result, it is determined whether the initial rule constraints are qualified. The qualified initial rule constraints are then organized to obtain the rule constraint set, where: When Cc1≤Cc2, the initial rule constraint is deemed qualified; When Cc1 > Cc2, the initial rule constraint is deemed unqualified, and the initial rule constraint is corrected by semantic rule constraint to obtain a qualified initial rule constraint.

9. A multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy as described in claim 8, characterized in that, The specific process of modifying the initial rule constraints through semantic rule constraints to obtain qualified initial rule constraints is as follows: The confidence difference Cc3 is obtained by subtracting the first confidence level Cc1 from the second confidence level Cc2. This confidence difference Cc3 is then compared with a preset difference threshold Cc4. Based on the comparison result, the initial rule constraints are corrected to obtain qualified initial rule constraints, where: When Cc3 > Cc4, it is determined that the initial rule constraint needs to be modified, the initial rule constraint is deleted, and the semantic rule constraint is marked as a qualified initial rule constraint. When Cc3≤Cc4, it is determined that the initial rule constraint does not need to be modified, and the initial rule constraint is marked as a qualified initial rule constraint; Once the initial rule constraints are corrected, the qualified initial rule constraints are organized to obtain the rule constraint set.

10. A multi-objective collaborative optimization method for power protection settings based on a hierarchical optimization strategy according to claim 9, characterized in that, The qualified initial rule constraints are verified. When the verification of the initial rule constraints fails, the difference threshold Cc4 is adjusted, and the adjusted difference threshold Cc4 is compared with the confidence difference Cc3 again. Specifically, this includes: The logic of the qualified initial rule constraints is analyzed using preset logical criteria to determine logical contradiction points Md. These contradiction points Md are then compared with a preset contradiction point threshold Md0. Based on the comparison result, the adjustment requirement for the difference threshold is determined, and the difference threshold Cc4 is adjusted accordingly. When Md≤Md0, it is determined that the difference threshold Cc4 does not need to be adjusted, and the difference threshold Cc4 is marked as the adjusted difference threshold Cc4; When Md > Md0, the difference threshold Cc4 is determined to need adjustment. The difference threshold Cc4 is added to the preset threshold adjustment amount to obtain the adjusted difference threshold Cc4.

Images