Relay protection setting value remote checking method and system based on multi-source information fusion

Abstract

The present application relates to the field of power system automation, and more particularly to a relay protection setting value remote checking method and system based on multi-source information fusion, comprising: collecting the operation setting value data of a target relay protection device, performing feature extraction on the operation setting value data of the target relay protection device, and constructing a panoramic feature vector; taking the maximization of a configuration rationality function representing the weighted balance of protection operation sensitivity and protection selectivity as the optimization target, defining multiple constraint conditions, and constructing a relay protection setting value checking model; inputting the panoramic feature vector into the relay protection setting value checking model, calculating and outputting a configuration rationality score and a risk contribution degree vector; determining the risk level of the current setting value configuration, and according to the risk contribution degree vector, tracing back to the key risk items whose configuration rationality score is lower than a set value, and generating a comprehensive checking report. The present application can significantly improve the intelligent level, accuracy and efficiency of setting value checking.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and more specifically, to a method and system for remote verification of relay protection settings based on multi-source information fusion. Background Technology

[0002] Relay protection is the first line of defense for ensuring the safe and stable operation of the power system, and its core lies in a precise and reasonable set of settings. These settings are derived by professional technicians through complex calculations and simulations based on the power grid's operating mode, equipment parameters, and protection principles. Once set, a standard setting sheet is generated, serving as the sole basis for configuring field device parameters. With the development of smart grid technology, many modern substations have achieved unmanned operation and centralized management. The management of relay protection setting modification, issuance, and verification is gradually shifting to remote control centers. In this model, after issuing setting modification instructions from the main station system, maintenance personnel must remotely read back the actual operating settings of the device and compare them with the standard setting sheet to ensure the correctness and consistency of the setting parameters. The verification process is a crucial closed-loop step in preventing incorrect setting transmission and ensuring power grid safety; its accuracy and reliability have a vital impact on the safe operation of the entire power grid. Currently, the technology for remote verification of setpoints mainly focuses on superficial and isolated numerical comparisons of setpoint data. It simply checks whether the read-back running setpoints match the values ​​on the standard setpoint sheet, failing to delve into the underlying logical rationality of the setpoint configuration. It is completely incapable of identifying configurations that, while numerically consistent with the setpoint sheet, contain flaws or errors in their logical coordination. Furthermore, when problems are found in the setpoint configuration, it is difficult to provide accurate guidance for tracing the source, and it cannot automatically locate the key risk factors leading to unreasonable configurations. This makes subsequent troubleshooting and setpoint optimization processes time-consuming, labor-intensive, and inefficient. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for remote verification of relay protection settings based on multi-source information fusion. This method enables in-depth verification of the internal logic of relay protection setting configuration and evaluates its overall rationality with a quantitative score. At the same time, it can accurately trace and locate key risk items, significantly improving the intelligence level, accuracy and efficiency of setting verification.

[0004] This invention is achieved through the following technical solution:

[0005] The method for remote verification of relay protection settings based on multi-source information fusion includes the following steps: Collect the operating setting data of the target relay protection device. The operating setting data includes the baseline setting data and the online operating setting data. Feature extraction is performed on the operating setpoint data of the target relay protection device to obtain the first feature, the second feature and the third feature. The first feature, the second feature and the third feature are combined and quantized to construct a panoramic feature vector. The optimization objective is to maximize the configuration rationality function that characterizes the weighted balance between protection operation sensitivity and protection selectivity. At the same time, multiple constraints are defined to construct a relay protection setting verification model. The panoramic feature vector is input into the relay protection setting verification model. The relay protection setting verification model solves the optimization objective under the constraints, calculates and outputs the configuration rationality score that quantifies the overall quality of the current setting configuration, and the risk contribution vector that characterizes the degree of negative impact of each input feature on the configuration rationality score. Based on the output configuration rationality score, the risk level of the current fixed configuration is determined, and the key risk items with configuration rationality scores lower than the set value are traced back according to the risk contribution vector, and a comprehensive verification report is generated.

[0006] Optionally, the online operation setting data of the target relay protection device may be preprocessed, specifically as follows: Multiple online running fixed value datasets collected in parallel using at least two different communication protocols will be compared item by item to identify and remove communication outliers. Using the aforementioned benchmark data as a reference, consistency checks are performed on the differences in data across multiple online running benchmark datasets, and the data with the highest confidence level is selected as the valid data. All valid data are normalized and mapped to a preset numerical range to complete the data preprocessing for online operation of fixed-value data.

[0007] Optionally, the feature extraction of the operating setting data of the target relay protection device specifically includes: The numerical value of the setting itself is extracted as the first feature from the preprocessed benchmark setting data and online running setting data; Extract the device model, firmware version, and topology and power flow information that characterizes the current operating conditions of the power grid as the second feature; Extract the ratio relationship between different preset values ​​as the third feature.

[0008] Optionally, the construction of the panoramic feature vector specifically involves: The first feature, the second feature, and the third feature are quantized and encoded respectively; The encoded features are weighted and combined according to preset weight coefficients to form a single vector that represents the overall state of the current fixed configuration, which is defined as the panoramic feature vector.

[0009] Optionally, the optimization objective is to maximize the configuration rationality function that characterizes the weighted balance between protection operation sensitivity and protection selectivity. The specific calculation formula is as follows:

[0010] in, To configure the rationality function, For panoramic feature vectors, , These are the global weighting coefficients for sensitivity and selectivity, respectively. , These are panoramic feature vectors. The i-th component related to sensitivity and the j-th component related to selectivity are given, where N and M are the total number of features related to sensitivity and selectivity, respectively. For use in evaluating features The S-shaped gain function contributing to protection sensitivity. The preset sensitivity reference value, The gain coefficient is used to control the steepness of the S-shaped gain function curve. For use in evaluating features An exponential penalty function that negatively impacts protection selectivity. The preset selective matching reference value, An exponential coefficient used to control the severity of punishment.

[0011] Optionally, the definition of multiple constraints specifically includes: time step difference constraints and sensitivity coefficient constraints; The time-level difference constraint is calculated using the following formula:

[0012] in, , These are the operating time functions of the main protection and backup protection, respectively, and their values ​​are determined by the characteristic components. and Decide; The sensitivity coefficient constraint is calculated using the following formula:

[0013] in, This is the minimum permissible sensitivity coefficient value. To protect the sensitivity coefficient function, its value is determined by the characteristic components. Decide.

[0014] Optionally, calculate and output the configuration rationality score and risk contribution vector, specifically as follows: The panoramic feature vector is input to the input layer of the relay protection setting verification model, and then passed forward layer by layer through the hidden layer in the relay protection setting verification model. Each layer performs a matrix weighting operation and a nonlinear activation function processing. The output layer integrates and generates a scalar value, which is represented as the configuration rationality score. Based on the gradient sensitivity analysis algorithm, with the configuration rationality score as the objective function, the gradient of the objective function with respect to each component of the panoramic feature vector is calculated in reverse along the network structure of the relay protection setting verification model to obtain the gradient vector. The gradient vector is output as the risk contribution vector, and each component of the risk contribution vector represents the degree and direction of influence of its corresponding input feature on the configuration rationality score.

[0015] Optionally, the risk level of the current configuration setting is determined, and the specific steps are as follows: At least two risk level thresholds are preset, including a first risk threshold and a second risk threshold, wherein the first risk threshold is less than the second risk threshold; The calculated configuration rationality score is compared with the first risk threshold and the second risk threshold. If the configuration rationality score is lower than the first risk threshold, the risk level is determined to be the first level. If the configuration rationality score is higher than the first risk threshold but lower than the second risk threshold, the risk level is determined to be the second level. If the configuration rationality score is higher than the second risk threshold, the risk level is determined to be the third level. The risk levels are ranked from highest to lowest as Level 1, Level 2, and Level 3.

[0016] Optionally, based on the risk contribution vector, key risk items with a reasonableness score lower than a set value can be configured, specifically as follows: Take the absolute value of each component of the risk contribution vector and sort them in descending order; Select the K largest components from the sorted vector, where K is a preset integer; The input feature items corresponding to the K components are identified as key risk items whose configuration reasonableness scores are lower than the set values.

[0017] A remote verification system for relay protection settings based on multi-source information fusion includes: The parallel acquisition unit acquires the operating setting data of the target relay protection device. The operating setting data includes the baseline setting data and the online operating setting data. The feature extraction unit extracts features from the operating setpoint data of the target relay protection device to obtain a first feature, a second feature, and a third feature. The first feature, the second feature, and the third feature are combined and quantized to construct a panoramic feature vector. The model building unit takes the maximization of the configuration rationality function, which characterizes the weighted balance between protection operation sensitivity and protection selectivity, as the optimization objective, and defines multiple constraints to build a relay protection setting verification model. The model calculation unit inputs the panoramic feature vector into the relay protection setting verification model. The relay protection setting verification model solves the optimization objective under the constraints, calculates and outputs the configuration rationality score that quantifies the overall quality of the current setting configuration, and the risk contribution vector that characterizes the degree of negative impact of each input feature on the configuration rationality score. The report generation unit determines the risk level of the current setpoint configuration based on the output configuration rationality score, and traces the key risk items whose configuration rationality scores are lower than the set value according to the risk contribution vector, and generates a comprehensive verification report.

[0018] The technical solution of the present invention has at least the following advantages and beneficial effects: This invention achieves a leap from numerical verification to logical validation by constructing a high-dimensional panoramic feature vector that comprehensively represents the status of setpoint configurations and using a deep learning model for analysis. It no longer views each setpoint in isolation but comprehensively evaluates the inherent logical connections and coordination between setpoints and between setpoints and power grid operating conditions, thereby uncovering deep-seated configuration risks that traditional methods cannot identify. Furthermore, this invention introduces a quantitative evaluation mechanism, providing an objective and intuitive assessment of the overall quality of the current setpoint configuration through a configuration rationality score. This completely changes the previous reliance on vague judgments based on human experience, making setpoint quality management standardized and measurable. In addition, this invention uniquely proposes the concept of a risk contribution vector, which, while verifying the existence of configuration risks, can accurately trace back and locate the key feature items that have the most significant negative impact on the overall score. This provides clear guidance for maintenance personnel to quickly and accurately optimize setpoint configurations, greatly improving the intelligence level and efficiency of setpoint management. Attached Figure Description

[0019] Figure 1 A flowchart illustrating the remote verification method for relay protection settings based on multi-source information fusion provided by this invention; Figure 2 A logical diagram illustrating the calculation and output of the configuration rationality score and risk contribution vector provided by this invention; Figure 3 This is a schematic diagram of the principle of the remote verification system for relay protection settings based on multi-source information fusion provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Reference Figure 1 As shown, this invention provides one embodiment: a remote verification method and system for relay protection settings based on multi-source information fusion. Its core idea is to overcome the limitation of setting verification being solely based on numerical comparison. By introducing artificial intelligence technology, it creates a new verification paradigm capable of deeply understanding the inherent logic of setting configuration, quantitatively assessing its overall rationality, and accurately tracing the sources of potential risks. This embodiment aims to elevate relay protection setting management from a human experience-driven model to a data-driven intelligent decision-making model.

[0022] This embodiment takes the deployment in the power grid dispatch automation master station system as an example to illustrate the specific execution process of the method of the present invention in detail.

[0023] Step 1: Collect the operating setting data of the target relay protection device. The operating setting data includes the baseline setting data and the online operating setting data.

[0024] In this embodiment, the baseline setting data originates from approved official setting sheets stored in higher-level systems such as setting management systems and PMS. The online operating setting data specifically refers to the settings retrieved in real-time from field relay protection devices via a communication network.

[0025] In this embodiment, to ensure the reliability and accuracy of the online operating setpoint data, a preprocessing step is also included, which specifically includes: The system collects setpoints from the same device in parallel using at least two different communication protocols, forming multiple online running setpoint datasets. By comparing each online running setpoint dataset, communication anomalies caused by communication channel interference, abnormal device response, etc., can be effectively identified and eliminated. In this embodiment, a baseline setpoint data is used as a reference to perform consistency verification on the multiple online running setpoint datasets that have been eliminated through anomaly removal. For data items with discrepancies, the system comprehensively evaluates their confidence level based on factors such as the reliability of the data source and communication quality, and selects the data with the highest confidence level as the final valid data. Finally, this embodiment normalizes all collected numerical valid data, including baseline and online setpoints, used for model calculation. This normalizes the original data with different physical units and orders of magnitude to a preset numerical range, eliminating the influence of units on model training and calculation, and ensuring that the initial weights of different features in the model are fair.

[0026] Step 2: Extract features from the operating setpoint data of the target relay protection device to obtain the first feature, the second feature, and the third feature. Combine and quantize the first feature, the second feature, and the third feature to construct a panoramic feature vector.

[0027] In this embodiment, the feature extraction process mainly revolves around three types of features: The first feature is the basic setting feature: it is the numerical value of the setting itself extracted from the benchmark setting data and the online operation setting data, including: the current, impedance, and time setting of each section of the line protection, the differential starting current and ratio braking coefficient of the transformer protection, etc. After normalization, the above values ​​constitute the first feature of this embodiment.

[0028] The second feature is the environmental and state feature: The second feature aims to take into account the physical environment in which the protection device is located and the operating state of the power grid, specifically including: the static information of the device itself and the dynamic information describing the current operating conditions of the power grid, the power grid topology and power flow information.

[0029] The third feature is the relational feature: In this embodiment, the third feature is not directly taken from a single data point, but is obtained by calculating the ratio relationship between different preset settings. The ratio relationship usually includes the coordination rules of relay protection. For example, the ratio of the impedance setting of the first stage of a line protection to the impedance of the entire line can be extracted to reflect the margin of the protection range, or the ratio of the backup protection action time of a line to the main protection action time to reflect the coordination of the time difference. The relational feature enables the model to directly evaluate the coordination logic of the protection configuration.

[0030] Specifically, after extracting the first, second, and third features, the process proceeds to the panoramic feature vector construction stage, which includes: The extracted first, second, and third features are mathematically encoded. In application, the numerical first and third features are already numerical, so only normalization is needed. For the text category information contained in the second feature, one-hot encoding is used to convert it into a multi-dimensional binary vector.

[0031] The encoded first, second, and third features are weighted and combined according to preset weight coefficients. All processed feature components are then concatenated in a predefined order to form a high-dimensional vector, which is the panoramic feature vector. The panoramic feature vector describes the overall state of the current configuration and its operating environment. In this embodiment, the weight coefficients can be set based on expert experience to adjust the initial importance of different types of features in the early stages of model construction.

[0032] Step 3: Maximize the configuration rationality function, which represents the weighted balance between protection operation sensitivity and protection selectivity, as the optimization objective. Simultaneously define multiple constraints and construct a relay protection setting verification model. Input the panoramic feature vector into the relay protection setting verification model. The model solves the optimization objective under the constraints, calculates and outputs a configuration rationality score that quantifies the overall quality of the current setting configuration, and a risk contribution vector representing the degree of negative impact of each input feature on the configuration rationality score.

[0033] The relay protection setting verification model used in this embodiment is essentially a deeply optimized mathematical model that incorporates expert knowledge and industry regulations in the field of relay protection. This embodiment uses the MLP (Multilayer Perceptron) model as an example. The model construction process aims to develop the ability to accurately evaluate the merits of any setting configuration by learning from a large amount of sample data or expert rules.

[0034] The essence of model construction lies in its optimization objective—maximizing the configuration rationality function that represents the weighted balance between protection operation sensitivity and protection selectivity. The specific formula for calculating the configuration rationality function is as follows:

[0035] in, To configure the rationality function, For panoramic feature vectors, , These are the global weighting coefficients for sensitivity and selectivity, respectively. , These are panoramic feature vectors. The i-th component related to sensitivity and the j-th component related to selectivity are given, where N and M are the total number of features related to sensitivity and selectivity, respectively. For use in evaluating features The S-shaped gain function contributing to protection sensitivity. The preset sensitivity reference value, The gain coefficient is used to control the steepness of the S-shaped gain function curve. For use in evaluating features An exponential penalty function that negatively impacts protection selectivity. The preset selective matching reference value, An exponential coefficient used to control the severity of punishment.

[0036] This embodiment fundamentally changes the verification paradigm of remote modification and verification of relay protection settings. Its logical process no longer relies on the traditional decentralized, clause-by-clause static Boolean logic judgment. Instead, it constructs a configuration rationality function that maps multi-dimensional input variables to a single continuous output scalar, thereby transforming the verification problem into a quantifiable and optimizable mathematical problem.

[0037] This embodiment first encodes the setting scheme to be verified and all key information affecting protection performance in its operating environment into a high-dimensional panoramic feature vector through feature engineering. This step ensures that all heterogeneous data, which would otherwise be difficult to compare directly, are placed within a unified mathematical framework.

[0038] Next, we move to the core of this logic—the calculation of the nonlinear comprehensive evaluation model. This model creatively deconstructs the evaluation system into two mutually constraining sub-functions: Sensitivity gain function: For all components in the panoramic feature vector positively correlated with sensitivity, the function uses a sigmoid function for mapping. The essence of this design lies in the fact that it does not perform simple linear accumulation, but rather accurately simulates the physical law of diminishing marginal returns in engineering practice through the saturation region of the sigmoid curve. That is, after exceeding a certain reasonable threshold, the increase in safety gains from further increases will become gradual, thus avoiding the model from giving disproportionately high scores to unnecessarily oversensitive data.

[0039] In the selective loss function section: for all components related to inter-protection coordination and selectivity in the panoramic feature vector, such as time difference and setting coordination coefficient, an exponential penalty function is used for mapping. The technical essence of this design lies in its mathematical expression of the engineering principle of high risk or even zero tolerance for selective errors. The exponential characteristic of the function curve means that even a small deviation between the actual setting value and the coordination benchmark value will result in a sharp amplification of the calculated loss value. This ensures that any minor coordination error that could potentially disrupt the safe and stable operation of the power grid will be highlighted with extremely high weight in the final score, forming a powerful veto mechanism against potential cascading tripping risks.

[0040] This embodiment calculates a unique configuration rationality score that represents the overall merits of the current setting configuration scheme by subtracting the weighted sum of the sensitivity gain functions from the weighted sum of the selectivity loss functions. Essentially, this embodiment transforms the core and abstract contradiction between sensitivity and selectivity in relay protection setting theory from a qualitative trade-off relying on expert experience into a deterministic, differentiable numerical optimization problem. Based on the application of this embodiment, the art of verification is transformed into a computable science, providing a solid and unprecedented technical core for achieving efficient, accurate, automated, and intelligent verification of remote setting modifications.

[0041] The solution to the aforementioned optimization objective is not unbounded, but must be performed under strict constraints derived from industry safety regulations. These constraints define red lines that cannot be crossed by any fixed configuration. This embodiment specifically includes: The time-separation constraint is calculated using the following formula:

[0042] in, , These are the operating time functions of the main protection and backup protection, respectively, and their values ​​are determined by the characteristic components. and Decide.

[0043] The sensitivity coefficient constraint is calculated using the following formula:

[0044] in, This is the minimum permissible sensitivity coefficient value. To protect the sensitivity coefficient function, its value is determined by the characteristic components. Decide.

[0045] like Figure 2 As shown, once the relay protection setting verification model is built, it can be deployed for online verification. For the panoramic feature vector representing the current field operating setting, the model will perform one forward calculation and one backward calculation, outputting two core results.

[0046] Specifically, the steps of the forward computation are as follows: The panoramic feature vector generated by the setting value to be verified is used as input data and input into the input layer of the relay protection setting value verification model.

[0047] The input data is passed forward layer by layer through multiple hidden layers in the relay protection setting verification model. In each hidden layer, the input data is subjected to a matrix weighting operation with the weight matrix of that layer, and the result is then processed through a nonlinear activation function.

[0048] Finally, the input data is passed to the output layer, where it is integrated and a scalar value is generated. This scalar value is the configuration rationality score, which quantifies the overall quality of the current fixed-value configuration.

[0049] Specifically, the steps for reverse calculation are as follows: After obtaining the configuration reasonableness score, the system immediately starts the reverse calculation process based on the gradient sensitivity analysis algorithm.

[0050] Using the calculated configuration rationality score as the objective function, the gradient of the objective function with respect to each component of the panoramic feature vector is calculated in reverse along the network structure of the relay protection setting verification model. This process is mathematically equivalent to the backpropagation algorithm in neural network training, but its purpose is not to update the weights, but to analyze the impact of the input.

[0051] The calculated result is a gradient vector with the same dimension as the panoramic feature vector.

[0052] The system outputs this gradient vector as a risk contribution vector. Each component of the risk contribution vector precisely represents the degree and direction of influence of its corresponding input feature on the final configuration rationality score. The larger the absolute value of the component, the higher the degree of influence; a negative sign of the component indicates that the increase of that feature has a negative impact on the score and is a potential source of risk.

[0053] Step 4: Determine the risk level of the current configuration based on the output configuration rationality score, and trace the key risk items whose configuration rationality scores are lower than the set value according to the risk contribution vector, and generate a comprehensive verification report.

[0054] In this embodiment, the system determines the risk level of the current configuration based on the output configuration rationality score. The system presets at least two risk level thresholds: a first risk threshold and a second risk threshold, wherein the first risk threshold is lower than the second risk threshold.

[0055] The calculated configuration rationality score is compared with these two thresholds: If the score is lower than the first risk threshold, the risk level is determined to be Level 1, which corresponds to high risk.

[0056] If the score is higher than the first risk threshold but lower than the second risk threshold, the risk level is determined to be level two, which corresponds to medium risk.

[0057] If the score is higher than the second risk threshold, the risk level is determined to be the third level, which corresponds to low risk.

[0058] For configurations with a high or medium risk level, this embodiment uses the risk contribution vector to automatically trace the root cause of the problem, specifically as follows: Take the absolute value of each component of the risk contribution vector and sort them in descending order.

[0059] Select the K largest components from the sorted vector.

[0060] The input feature terms corresponding to the K components are identified, and the identified feature terms are determined to be the key risk terms that cause the current configuration rationality score to be low. It is understandable that if the key risk term traced back is the time difference with the downstream line protection, and its original gradient value in the risk contribution vector is negative, then the system can infer that the time difference setting may be too small, resulting in selective risk.

[0061] Finally, the system integrates all the above analysis results to generate a comprehensive verification report. The comprehensive verification report not only includes the final configuration rationality score and risk level, but more importantly, it clearly lists the key risk items obtained through tracing, and can include their current values, risk contribution, and preliminary optimization suggestions, providing unprecedentedly accurate guidance for value verification and optimization work.

[0062] like Figure 3 As shown, corresponding to the above method embodiments, the present invention also provides a remote verification system for relay protection settings based on multi-source information fusion. This system can be a software system deployed on a master station server, or a hardware device integrating related functions, including the following functional units: Parallel acquisition unit: Responsible for executing data acquisition tasks. It acquires baseline setpoint data and online running setpoint data in parallel through multiple communication protocol interfaces, and performs preliminary data cleaning and alignment.

[0063] Feature extraction unit: responsible for performing feature engineering tasks, performing deep processing on the collected data, extracting the first, second and third features, and completing encoding, weighting and stitching to finally construct a standardized panoramic feature vector.

[0064] Model building unit: Responsible for executing model building tasks. Based on the optimization objective function and constraints set by experts, it uses historical data or simulation data to train the model and finally generates a relay protection setting verification model file with solidified verification logic.

[0065] Model computation unit: Responsible for executing online solution tasks. It receives the panoramic feature vector from the feature extraction unit, and through complete forward and backward computation, efficiently outputs the configuration rationality score and risk contribution vector.

[0066] Report generation unit: Responsible for executing decision-making and presentation tasks. It receives the output of the model calculation unit, determines the risk level based on preset thresholds, and uses the risk contribution vector to conduct source analysis of key risk items. Finally, it integrates all results to generate a comprehensive verification report, which is then presented to the operations and maintenance personnel.

[0067] The aforementioned units work closely together through their internal data bus and control flow to form a complete, automated, and intelligent closed loop from data input to decision output, thus realizing all the technical solutions of this invention.

[0068] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for remote verification of relay protection settings based on multi-source information fusion, characterized in that, The steps of this method include: Collect the operating setting data of the target relay protection device. The operating setting data includes the baseline setting data and the online operating setting data. Feature extraction is performed on the operating setpoint data of the target relay protection device to obtain the first feature, the second feature and the third feature. The first feature, the second feature and the third feature are combined and quantized to construct a panoramic feature vector. The optimization objective is to maximize the configuration rationality function that characterizes the weighted balance between protection operation sensitivity and protection selectivity. At the same time, multiple constraints are defined to construct a relay protection setting verification model. The panoramic feature vector is input into the relay protection setting verification model. The relay protection setting verification model solves the optimization objective under the constraints, calculates and outputs the configuration rationality score that quantifies the overall quality of the current setting configuration, and the risk contribution vector that characterizes the degree of negative impact of each input feature on the configuration rationality score. Based on the output configuration rationality score, the risk level of the current fixed configuration is determined, and the key risk items with configuration rationality scores lower than the set value are traced back according to the risk contribution vector, and a comprehensive verification report is generated.

2. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 1, characterized in that, The online operation setting data of the target relay protection device is preprocessed, specifically as follows: Multiple online running fixed value datasets collected in parallel using at least two different communication protocols will be compared item by item to identify and remove communication outliers. Using the aforementioned benchmark data as a reference, consistency checks are performed on the differences in data across multiple online running benchmark datasets, and the data with the highest confidence level is selected as the valid data. All valid data are normalized and mapped to a preset numerical range to complete the data preprocessing for online operation of fixed-value data.

3. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 2, characterized in that, The feature extraction of the operating setpoint data of the target relay protection device specifically includes: The numerical value of the setting itself is extracted as the first feature from the preprocessed benchmark setting data and online running setting data; Extract the device model, firmware version, and topology and power flow information that characterizes the current operating conditions of the power grid as the second feature; Extract the ratio relationship between different preset values ​​as the third feature.

4. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 3, characterized in that, The construction of the panoramic feature vector is specifically as follows: The first feature, the second feature, and the third feature are quantized and encoded respectively; The encoded features are weighted and combined according to preset weight coefficients to form a single vector that represents the overall state of the current fixed configuration, which is defined as the panoramic feature vector.

5. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 4, characterized in that, The optimization objective is to maximize the configuration rationality function that characterizes the weighted balance between protection operation sensitivity and protection selectivity. The specific calculation formula is as follows: in, To configure the rationality function, For panoramic feature vectors, , These are the global weighting coefficients for sensitivity and selectivity, respectively. , These are panoramic feature vectors. The i-th component related to sensitivity and the j-th component related to selectivity are given, where N and M are the total number of features related to sensitivity and selectivity, respectively. For use in evaluating features The S-shaped gain function contributing to protection sensitivity. The preset sensitivity reference value, The gain coefficient is used to control the steepness of the S-shaped gain function curve. For use in evaluating features An exponential penalty function that negatively impacts protection selectivity. The preset selective matching reference value, An exponential coefficient used to control the severity of punishment.

6. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 5, characterized in that, The definition includes multiple constraints, specifically: time step difference constraints and sensitivity coefficient constraints; The time-level difference constraint is calculated using the following formula: in, , These are the operating time functions of the main protection and backup protection, respectively, and their values ​​are determined by the characteristic components. and Decide; The sensitivity coefficient constraint is calculated using the following formula: in, This is the minimum permissible sensitivity coefficient value. To protect the sensitivity coefficient function, its value is determined by the characteristic components. Decide.

7. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 6, characterized in that, Calculate and output the configuration rationality score and risk contribution vector, specifically as follows: The panoramic feature vector is input to the input layer of the relay protection setting verification model, and then passed forward layer by layer through the hidden layer in the relay protection setting verification model. Each layer performs a matrix weighting operation and a nonlinear activation function processing. The output layer integrates and generates a scalar value, which is represented as the configuration rationality score. Based on the gradient sensitivity analysis algorithm, with the configuration rationality score as the objective function, the gradient of the objective function with respect to each component of the panoramic feature vector is calculated in reverse along the network structure of the relay protection setting verification model to obtain the gradient vector. The gradient vector is output as the risk contribution vector, and each component of the risk contribution vector represents the degree and direction of influence of its corresponding input feature on the configuration rationality score.

8. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 7, characterized in that, The specific steps to determine the risk level of the current configuration are as follows: At least two risk level thresholds are preset, including a first risk threshold and a second risk threshold, wherein the first risk threshold is less than the second risk threshold; The calculated configuration rationality score is compared with the first risk threshold and the second risk threshold. If the configuration rationality score is lower than the first risk threshold, the risk level is determined to be the first level. If the configuration rationality score is higher than the first risk threshold but lower than the second risk threshold, the risk level is determined to be the second level. If the configuration rationality score is higher than the second risk threshold, the risk level is determined to be the third level. The risk levels are ranked from highest to lowest as Level 1, Level 2, and Level 3.

9. The method for remote verification of relay protection settings based on multi-source information fusion according to claim 8, characterized in that, Based on the risk contribution vector, key risk items with reasonableness scores lower than a set value are configured, specifically as follows: Take the absolute value of each component of the risk contribution vector and sort them in descending order; Select the K largest components from the sorted vector, where K is a preset integer; The input feature items corresponding to the K components are identified as key risk items whose configuration reasonableness scores are lower than the set values.

10. A remote verification system for relay protection settings based on multi-source information fusion, characterized in that, include: The parallel acquisition unit acquires the operating setting data of the target relay protection device. The operating setting data includes the baseline setting data and the online operating setting data. The feature extraction unit extracts features from the operating setpoint data of the target relay protection device to obtain a first feature, a second feature, and a third feature. The first feature, the second feature, and the third feature are combined and quantized to construct a panoramic feature vector. The model building unit takes the maximization of the configuration rationality function, which characterizes the weighted balance between protection operation sensitivity and protection selectivity, as the optimization objective, and defines multiple constraints to build a relay protection setting verification model. The model calculation unit inputs the panoramic feature vector into the relay protection setting verification model. The relay protection setting verification model solves the optimization objective under the constraints, calculates and outputs the configuration rationality score that quantifies the overall quality of the current setting configuration, and the risk contribution vector that characterizes the degree of negative impact of each input feature on the configuration rationality score. The report generation unit determines the risk level of the current setpoint configuration based on the output configuration rationality score, and traces the key risk items whose configuration rationality scores are lower than the set value according to the risk contribution vector, and generates a comprehensive verification report.

Images