A power grid safety supervision system and a safety supervision method

Abstract

The grid safety supervision system and safety supervision method of the present application belong to the technical field of power system safety monitoring and intelligent early warning, and comprise the following steps: S1, collecting and pre-processing multi-source heterogeneous monitoring data of the power grid to generate standardized data flow; S2, based on the standardized data flow, calculating the electromagnetic transient harmonic fingerprint fractal dimension, constructing the node-level safety event correlation degree tensor, calculating the whole-network safety situation topology entropy, and calculating the decision tree fractal complexity; S3, combining the electromagnetic transient harmonic fingerprint fractal dimension, the whole-network safety situation topology entropy and the decision tree fractal complexity, calculating the instantaneous risk factor; S4, based on the instantaneous risk factor and the preset safety threat propagation time lag kernel function, generating the normalized risk index through the normalized convolution operation, which overcomes the defects of the prior art that relies on isolated indicators and fixed thresholds and cannot identify potential threats early.

Description

Technical Field

[0001] This invention relates to the field of power system safety monitoring and intelligent early warning, specifically to a power grid safety supervision system and safety supervision method. Background Technology

[0002] In the safety monitoring of power systems, traditional monitoring methods mainly rely on monitoring isolated physical quantities and comparing them with fixed alarm thresholds. These methods are usually static and passive, with problems such as narrow coverage and single analytical dimensions. They cannot comprehensively and in real time reflect the true safety status of the entire power grid as a complex dynamic system. This situation makes it difficult for power grid managers to identify systemic risks caused by the coupling of multiple factors in the early stage, affecting the accuracy and timeliness of fault prevention measures.

[0003] The aforementioned shortcomings and deficiencies mainly stem from the limitations of traditional monitoring technologies in data analysis and decision-making mechanisms. First, monitoring data is often interpreted in isolation, lacking in-depth analysis of high-order correlations between events at different nodes and levels, failing to reveal complex dependencies hidden beneath the surface. Second, risk assessment is often instantaneous, ignoring the cumulative effect and memory characteristics of historical risks, and may underestimate potential threats due to oversimplification of the model. Most critically, it employs rigid, fixed alarm thresholds, unable to adaptively adjust according to the real-time overall risk level of the power grid. This leads to a large number of false alarms during stable operation due to normal fluctuations, while in high-risk critical states, it may be insufficiently sensitive to weak precursors to faults.

[0004] As a result, when power grid security is threatened, maintenance personnel often receive only simple alarm signals, lacking in-depth diagnostic information about the root cause of the risk, the scope of impact, and key nodes. This greatly delays the time for troubleshooting and developing accurate response strategies, limiting the overall effectiveness and intelligence level of power grid security supervision.

[0005] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a power grid safety supervision system and method to solve the problems mentioned in the background art.

[0007] The technical solution of the present invention includes the following steps:

[0008] S1. Collect and preprocess multi-source heterogeneous monitoring data of the power grid to generate a standardized data stream;

[0009] S2. Based on standardized data flow, calculate the fractal dimension of electromagnetic transient harmonic fingerprints; construct a node-level security event correlation tensor; calculate the topological entropy of the entire network security situation; and calculate the fractal complexity of the decision tree.

[0010] S3. Calculate the instantaneous risk factor by combining the fractal dimension of electromagnetic transient harmonic fingerprints, the topological entropy of the overall network security situation, and the fractal complexity of the decision tree.

[0011] S4. Based on the instantaneous risk factor and the preset security threat propagation delay kernel function, a normalized risk index is generated through normalized convolution operation;

[0012] S5. Based on the normalized risk index, calculate the adaptive adjustment factor used to adjust the supervision strategy;

[0013] S6. Based on the adaptive adjustment factor, adjust between the preset maximum alarm threshold and minimum alarm threshold to generate a dynamic alarm threshold;

[0014] S7. Compare the normalized risk index with the dynamic alarm threshold. If the normalized risk index is greater than the dynamic alarm threshold, generate an early warning message; if the normalized risk index is less than or equal to the dynamic alarm threshold, maintain the current supervision status.

[0015] Preferably, the multi-source heterogeneous monitoring data in S1 includes:

[0016] Synchronization vector data of the phasor measurement unit;

[0017] Low-frequency oscillation data from a wide-area measurement system;

[0018] High-frequency electromagnetic transient signals captured by a dedicated sensor.

[0019] Preferably, the preprocessing in S1 includes:

[0020] Perform timestamp alignment;

[0021] Perform outlier removal processing;

[0022] Perform signal normalization processing.

[0023] Preferably, the step of constructing the node-level security event correlation tensor in S2 specifically includes:

[0024] Based on conditional mutual information in multivariate information theory, we calculate the information correlation degree between events of two other nodes under the condition that an event occurs at a certain node, so as to construct a node-level security event correlation tensor.

[0025] Preferably, the security threat propagation delay kernel function preset in S4 is constructed using a stretching exponential function to characterize the memory effect of risk propagation.

[0026] Preferably, the warning information generated in S7 specifically includes:

[0027] By retrospectively analyzing the composition of the normalized risk index, the main sources of current risk are identified, and the power grid areas with the closest risk correlation are located using the node-level security event correlation tensor.

[0028] A power grid safety monitoring system, comprising:

[0029] The data acquisition and preprocessing module is used to acquire and preprocess multi-source heterogeneous monitoring data of the power grid to generate a standardized data stream;

[0030] The multi-dimensional feature extraction module is used to receive standardized data streams, calculate the fractal dimension of electromagnetic transient harmonic fingerprints, construct node-level security event correlation tensors, calculate the topological entropy of the entire network security situation, and calculate the fractal complexity of decision trees.

[0031] The risk assessment module is used to combine the features calculated by the multi-dimensional feature extraction module to calculate the instantaneous risk factor, and generate a normalized risk index based on the instantaneous risk factor and the preset security threat propagation delay kernel function.

[0032] The adaptive decision-making module is used to calculate the adaptive adjustment factor for adjusting the supervision strategy based on the normalized risk index, and generate dynamic alarm thresholds.

[0033] The early warning generation module compares the normalized risk index with the dynamic alarm threshold. If the normalized risk index is greater than the dynamic alarm threshold, an early warning message is generated. If it is less than or equal to the threshold, the control system maintains the current supervision status.

[0034] Preferably, the data flow between modules is configured as follows:

[0035] The multi-dimensional feature extraction module receives a standardized data stream output from the data acquisition and preprocessing module;

[0036] The risk assessment module receives the fractal dimension of the electromagnetic transient harmonic fingerprint, the topological entropy of the overall network security situation, and the fractal complexity of the decision tree calculated by the multi-dimensional feature extraction module.

[0037] The adaptive decision-making module receives the normalized risk index generated by the risk assessment module.

[0038] The early warning generation module receives the normalized risk index from the risk assessment module and the dynamic alarm threshold from the adaptive decision-making module.

[0039] This invention provides an improved power grid safety supervision system and method, which, compared with the prior art, has the following improvements and advantages:

[0040] 1. This invention, by constructing a logically progressive data processing and analysis architecture, transforms power grid safety supervision from a traditional static and passive mode into a dynamic, forward-looking, and adaptive intelligent process. This method can deeply understand the systemic risks caused by the complex coupling of multiple sources and scales, and overcomes the shortcomings of existing technologies that rely on isolated indicators and fixed thresholds and cannot identify potential threats in the early stages.

[0041] 2. By collecting and refining the preprocessing of multi-source heterogeneous monitoring data of the power grid, a high-quality and highly consistent data foundation is provided for in-depth analysis. Compared with the neglect of data quality issues by traditional technologies, this invention ensures the effectiveness of subsequent correlation analysis, the robustness of the algorithm, and the fairness of multi-dimensional feature fusion through timestamp alignment, outlier removal, and signal normalization, thus guaranteeing the reliability and accuracy of the entire supervision process from the source. Attached Figure Description

[0042] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0043] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. Example 1

[0045] Please see Figure 1 This invention provides a power grid safety supervision method, comprising the following steps:

[0046] S1. Collect and preprocess multi-source heterogeneous monitoring data of the power grid to generate a standardized data stream;

[0047] S2. Based on standardized data flow, calculate the fractal dimension of electromagnetic transient harmonic fingerprints; construct a node-level security event correlation tensor; calculate the topological entropy of the entire network security situation; and calculate the fractal complexity of the decision tree.

[0048] S3. Calculate the instantaneous risk factor by combining the fractal dimension of electromagnetic transient harmonic fingerprints, the topological entropy of the overall network security situation, and the fractal complexity of the decision tree.

[0049] S4. Based on the instantaneous risk factor and the preset security threat propagation delay kernel function, a normalized risk index is generated through normalized convolution operation;

[0050] S5. Based on the normalized risk index, calculate the adaptive adjustment factor used to adjust the supervision strategy;

[0051] S6. Based on the adaptive adjustment factor, adjust between the preset maximum alarm threshold and minimum alarm threshold to generate a dynamic alarm threshold;

[0052] S7. Compare the normalized risk index with the dynamic alarm threshold. If the normalized risk index is greater than the dynamic alarm threshold, generate an early warning message. If the normalized risk index is less than or equal to the dynamic alarm threshold, maintain the current supervision status.

[0053] This invention provides a power grid safety supervision method. Through a series of logically progressive data processing and analysis steps, this method transforms power grid safety supervision from a static and passive mode into a dynamic, forward-looking, and adaptive intelligent process. The method as a whole constitutes a closed-loop adaptive supervision process, ensuring accurate and real-time assessment and response to the power grid safety situation.

[0054] S1. Collect and preprocess multi-source heterogeneous monitoring data of the power grid to generate a standardized data stream.

[0055] This step aims to provide a high-quality, uniformly formatted data foundation for subsequent in-depth analysis. In a specific embodiment, the data acquisition and preprocessing module is configured to continuously acquire multi-source heterogeneous monitoring data from various key monitoring points of the power grid. These data undergo a standardized preprocessing process to form a standardized data stream, which is then supplied to subsequent modules for unified analysis. This preprocessing is a crucial prerequisite for ensuring that subsequent complex algorithms can obtain reliable input and avoid distortion of analysis results due to data quality issues.

[0056] S2. Based on standardized data flow, calculate the fractal dimension of electromagnetic transient harmonic fingerprints; construct node-level security event correlation tensors; calculate the topological entropy of the entire network security situation; and calculate the fractal complexity of the decision tree.

[0057] The core task of this step is to receive a standardized data stream and calculate a series of key characteristic parameters that can characterize the nonlinear dynamic characteristics of the power grid from different dimensions and levels.

[0058] To quantify microscopic anomalies at the physical device level, a fractal dimension of electromagnetic transient harmonic fingerprints is introduced. This parameter aims to capture the self-similarity or complexity of power grid transient signals at different scales, thereby identifying harmonic structure anomalies caused by latent faults such as insulation degradation and loose connections. The calculation is based on box counting in fractal geometry, and the mathematical expression is defined as:

[0059]

[0060] In this expression, : Fractal dimension of electromagnetic transient harmonic fingerprints, which aims to capture the self-similarity or complexity of power grid transient signals at different scales; : Dimensionless scaling variables; This represents the side length required to completely cover all signal points on the time-frequency distribution map of the transient signal. The minimum number of square units is a dimensionless count value; : Natural logarithm function;

[0061] The time-frequency distribution map is specifically generated by performing a continuous wavelet transform on a high-frequency electromagnetic transient signal. The wavelet basis function can be a Morlet wavelet suitable for analyzing transient impacts. The time-frequency energy matrix generated thereby constitutes the time-frequency distribution map of this invention. Its horizontal axis is time, and its vertical axis is frequency or scale. The value of the matrix element represents the energy density of the signal at that time-frequency point.

[0062] It is a dimensionless scaling variable, and its value range is dynamically set according to the signal sampling rate to ensure the effectiveness of the analysis; the calculation of this parameter does not require adjustment of weights and can objectively reflect the inherent complexity of the signal.

[0063] To describe the high-order coupling relationships between security events at different nodes in a power grid, a node-level security event correlation tensor was constructed. This third-order tensor construction goes beyond traditional binary relation analysis, aiming to reveal common triangular couplings or complex dependencies caused by common causes in the power grid. One element of this tensor... Defined as:

[0064]

[0065] in, It is a random variable representing a security-related event occurring at node x; It is a scalar, representing a node. The strength of the ternary association or the degree of information coupling between them.

[0066] : These represent random variables that indicate security-related events occurring at nodes i, j, and k, respectively; these random variables can be quantified by monitoring the key operating parameters of the nodes.

[0067] random variable This can be quantified by monitoring key operating parameters of node x, such as the normalized amplitude of voltage or frequency deviation; for example, the root mean square deviation of a pre-processed key parameter from the normal operating baseline within a time window can be used as... The instantaneous value; this definition transforms continuous monitoring data into a quantified random variable that can reflect the safety status of nodes, thus providing a basis for calculating conditional mutual information;

[0068] It is conditional mutual information in information theory, used to calculate the degree of information correlation between events of node i and events of node j given that an event occurs at node k. Its source is multivariate information theory.

[0069] In practical applications, considering the large number of nodes in a large power grid, optimization strategies can be adopted to ensure the real-time performance of the calculation. For example, instead of performing a ternary combination traversal on all nodes, a subset of key nodes can be selected based on prior knowledge or preliminary analysis, such as the degree centrality and betweenness centrality of nodes. The correlation tensor can be constructed only within this subset. Alternatively, a correlation threshold can be set to remove node combinations with low correlation during the calculation process, thereby significantly reducing the computational complexity and making it feasible in engineering.

[0070] Based on the constructed correlation tensor, the topological entropy of the entire network security situation is calculated. The purpose of this parameter is to elevate the microscopic correlation between nodes to a quantitative assessment of the overall macroscopic stability of the network. Its calculation formula is derived from Shannon information entropy and is defined as follows:

[0071]

[0072] Here It is about the correlation tensor The normalized correlation strength distribution obtained by normalizing (representing ternary correlation strength); The value represents The relative weight or contribution of the association strength of a node group to the total association strength of the entire network; The topological entropy of the overall network security situation is used to quantify the uncertainty of the distribution of network-wide correlation strength. Node index; The logarithmic function with base 2; simultaneously, monitoring is required. The degree of polarization of the distribution. When the entropy value When the value is extremely low, approaching 0, it is necessary to assist in determining whether a certain value exists. It is much larger than other items; if it exists, it should be interpreted as a deterministic, strongly correlated risk rather than system stability; for example, a parallel concentration index, such as the Gini coefficient, can be added to assess network risk together with topological entropy.

[0073] The normalization process here is specifically accomplished by applying the Softmax function, which transforms the correlation tensor... The elements, with their conditional mutual information values ​​considered as the energy or weight of the event association, are converted into a probability distribution; the calculation formula is:

[0074]

[0075] in, Defined as a normalized distribution of association strength or association contribution; The elements of the correlation tensor, namely the conditional mutual information values, are regarded as the energy or weight of the event correlation. : Iterate through the indices of all elements in the tensor; Exponential function; The value represents The relative weight or contribution ratio of the association strength of this specific triple in the total association strength of all triples in the entire network. Due to the properties of Softmax, all The sum of is 1, so it is a mathematically valid probability distribution that can be used for subsequent entropy calculations.

[0076] The summation iterates through all elements in the tensor; this method maps the mutual information values ​​representing the correlation strength to an effective joint probability distribution with a sum of 1, thus enabling the legitimate calculation of topological entropy.

[0077] This formula is used to quantify the uncertainty or disorder of the entire power grid security event correlation network;

[0078] Decision tree fractal complexity The method for quantifying the complexity of power grid dispatching or fault handling rule sets is as follows:

[0079]

[0080] In this expression, It is the total number of nodes in the decision tree, including the root node, internal nodes, and leaf nodes. It is the arithmetic mean of the path lengths from the root node to all leaf nodes; this definition effectively reflects the size and depth of the decision tree. A higher value indicates a more complex rule set, resulting in a higher cognitive load for operations and maintenance personnel or a greater dependence on automated systems. In specific situations, such as incomplete information or system response delays, this may indirectly introduce potential risks. To avoid singularities in the calculation, when the average path length... When the denominator is equal to 1, it is zero. In this case, it can be determined according to the convention that... Defined as 1 or a constant representing the basic complexity; explicitly specified when hour, Defined as the default value of 1.0. When the value is close to 1, a numerical stabilization method is used.

[0081] S3. Calculate the instantaneous risk factor by combining the fractal dimension of electromagnetic transient harmonic fingerprints, the topological entropy of the overall network security situation, and the fractal complexity of the decision tree.

[0082] This step aims to integrate the multiple isolated, multi-dimensional feature parameters extracted in the previous steps into a single indicator that comprehensively reflects the current risk level of the system; to this end, an instantaneous risk factor is defined. It is designed to be a dimensionless quantity; in the embodiment, the calculation formula is:

[0083] In this expression, the parameters are defined as follows: It is the fractal complexity of the decision tree at time t; It is a safe baseline value for the fractal complexity of decision trees, which is a dimensionless constant calibrated using historical data; This is the network-wide security situation topology entropy at time t. It needs to be normalized before use, for example, by dividing it by an entropy baseline value obtained from historical data statistics. Or theoretical maximum entropy This makes it a dimensionless index, thus ensuring the consistency of dimensions on both sides of the formula. This refers to the normalized topological entropy of the overall network security situation; Exponential function;

[0084] It is the fractal dimension of the electromagnetic transient harmonic fingerprint at time t; It is the normal reference value of the harmonic fractal dimension, which is a dimensionless constant calibrated through historical data; , , All are dimensionless weighting coefficients, whose values ​​are determined by regression analysis of historical failure data or by expert system calibration, in order to balance the contribution of different risk sources; this formula organically combines macro-management risk and micro-physical risk.

[0085] It should be noted that the model constructed in this embodiment mainly focuses on the endogenous risks caused by the internal physical state, topology, and established rules of the power grid. In future improvements, external influencing factors, such as severe weather warnings and cybersecurity threat intelligence, can be quantified and introduced as additional terms into the instantaneous risk factors. In the calculation, or as a weighting adjustment coefficient, it affects The value of is used to construct a more comprehensive and open risk assessment model;

[0086] This study employs a multiplicative and exponential function to model the nonlinear coupling and amplification effects between risk factors in a computationally efficient manner. The multiplicative term reflects the increased risk when management complexity and network uncertainty coexist, while the exponential term emphasizes the sensitivity to the degradation of physical equipment conditions. In practical deployment, this function form can be verified and optimized based on a large amount of simulation and experimental data, or it can be replaced with other nonlinear functions that can more accurately fit the specific power grid dose-effect relationship, such as using piecewise functions or machine learning-based models.

[0087] S4. Based on the instantaneous risk factor and the preset security threat propagation delay kernel function, a normalized risk index is generated through normalized convolution operation.

[0088] The purpose of this step is to introduce a time dimension into the instantaneous risk assessment, in order to normalize the overall risk of the power grid in a way that reflects the accumulation and decay memory effect of historical risks. For this purpose, a pre-defined security threat propagation time delay kernel function is required. This function characterizes the non-Markovian process of risk propagation in the power grid, namely the memory effect; it generates a normalized risk index through normalized convolution operations. ;

[0089] Based on this, the normalized risk index It is defined as the following normalized convolution form:

[0090]

[0091] The physical meaning of this formula lies in the risk index at the current time t. It is all historical moments Instantaneous risk factors The weighted average, with weights determined by the time-delay kernel function. Decision, among which It is a momentary risk factor in historical moments. It is a time-delay kernel function defined subsequently; normalization is achieved by dividing by the integral of the kernel function itself, ensuring... It is always a dimensionless, bounded value, suitable for making stable threshold judgments; this design reflects the cumulative effect of historical risks while avoiding the problem of the risk index growing indefinitely. Historical moment; the initial moment of system startup. When, define Equal to instantaneous risk factor Or a preset safety initial value;

[0092] S5. Based on the normalized risk index, calculate the adaptive adjustment factor used to adjust the supervision strategy.

[0093] This step aims to transform the quantified risk level into a standardized control signal that can be used to directly regulate subsequent strategies; to this end, an adaptive adjustment factor was designed. This factor is defined as a non-linear function based on a risk index; in a specific embodiment, it takes the form of a sigmoid function.

[0094]

[0095] In this expression, It is the dimensionless normalized risk index calculated in the preceding steps; It is a dimensionless preset risk response threshold, such as 0.5, which represents the critical point at which the risk level undergoes a qualitative change. The value is determined through simulation experiments or expert experience. It is a dimensionless positive number, called the response sensitivity coefficient, which is used to control the steepness of the adjustment curve. The value is also calibrated through simulation experiments to achieve a balance between early warning sensitivity and strategy stability. The output value is limited to between 0 and 1, serving as a standardized adjustment factor;

[0096] For example, in a typical regional power grid model, risk response thresholds can be determined through retrospective analysis of historical high-risk events. A value of 0.5 represents the critical point at which risk transitions from the attention level to the alarm level; simultaneously, to achieve a balance between the sensitivity and stability of early warning systems, the response sensitivity coefficient... It can be calibrated to 5, at which point the adjustment factor curve is in achieve Significant but not overly drastic changes can occur when the voltage is near the grid; these values ​​are for illustrative purposes only and should be adjusted accordingly for different grid environments.

[0097] S6. Based on the adaptive adjustment factor, adjust between the preset maximum alarm threshold and minimum alarm threshold to generate a dynamic alarm threshold.

[0098] The goal of this step is to create alarm thresholds that can adaptively change based on the system's real-time risk level, rather than fixed, rigid thresholds; therefore, dynamic alarm thresholds are introduced. The threshold is dynamically adjusted using an adaptive adjustment factor calculated in the preceding steps. In one specific embodiment, the calculation formula is as follows:

[0099]

[0100] in, and These are the preset dimensionless maximum and minimum alarm thresholds, whose values ​​are set based on operational experience and security procedures, defining the upper and lower limits for dynamic adjustment of the alarm thresholds; when the system is secure, Lower, leading to Approaching 0, at this point Approaching a higher The system has high tolerance; when the system is in danger, Higher, leading to Approaching 1, at this point Pulled down to The system becomes more sensitive; : Adaptive adjustment factor calculated in the preceding steps;

[0101] S7. Compare the normalized risk index with the dynamic alarm threshold. If the normalized risk index is greater than the dynamic alarm threshold, generate an early warning message; if the normalized risk index is less than or equal to the dynamic alarm threshold, maintain the current supervision status.

[0102] This is the final judgment and output stage of the entire method; the risk warning generation module is configured to compare two dimensionless quantities in real time: namely, the normalized risk index. With dynamic alarm threshold When satisfied When the conditions are met, the warning condition is triggered, and the system will immediately generate and issue detailed warning information; conversely, if the conditions are not met, the warning condition will not be triggered. If the system determines that the current state is within an acceptable range, it will maintain the existing supervision status and continue with the next round of monitoring and evaluation.

[0103] The method disclosed in this embodiment achieves dynamic, forward-looking, and intelligent supervision of power grid safety status by constructing a complete closed loop from multi-dimensional feature extraction, risk quantification, time accumulation to adaptive threshold adjustment. Compared with traditional monitoring methods that rely on fixed thresholds and isolated indicators, this method can identify systemic risks caused by the coupling of multiple factors earlier. Through an adaptive alarm mechanism, it can effectively reduce the false alarm rate caused by normal fluctuations while ensuring sensitivity to serious faults, thus significantly improving the overall efficiency and intelligence level of power grid safety supervision.

[0104] The multi-source heterogeneous monitoring data in S1 includes:

[0105] Synchronization vector data of the phasor measurement unit;

[0106] Low-frequency oscillation data from a wide-area measurement system;

[0107] High-frequency electromagnetic transient signals captured by a dedicated sensor;

[0108] In this embodiment, the composition of the multi-source heterogeneous monitoring data in step S1 is defined; to ensure the comprehensiveness and accuracy of subsequent feature extraction, the data specifically includes:

[0109] Synchronization vector data from the phasor measurement unit aims to provide high-time-resolution synchronous snapshots of grid voltage, current phasors, and frequency, which form the basis for analyzing grid power angle stability and dynamic behavior.

[0110] Low-frequency oscillation data from wide-area measurement systems: The purpose is to capture low-frequency power oscillation modes that may exist in the power grid and span a wide area. These oscillations are key indicators for measuring the overall stability of the system.

[0111] High-frequency electromagnetic transient signals captured by dedicated sensors: The purpose is to monitor electromagnetic transient processes with an extremely wide bandwidth caused by events such as switching operations, lightning strikes, or early insulation degradation of equipment. It is the core basis for equipment-level health status diagnosis and latent fault identification.

[0112] By clearly defining these three sources of data with complementary properties, this embodiment constructs a more comprehensive and complete data foundation. Synchronization vector data focuses on the power frequency dynamics of the system, low-frequency oscillation data focuses on the macroscopic stability of the system, and high-frequency transient signals delve into the microscopic health status of equipment. This combination ensures that subsequent analysis can cover multiple scales from microscopic equipment to macroscopic systems, thereby obtaining a more comprehensive and profound understanding of the power grid security situation, providing a fundamental guarantee for the accuracy of subsequent risk assessment, and bringing about a gain in technical effects.

[0113] Preprocessing in S1 includes:

[0114] Perform timestamp alignment;

[0115] Perform outlier removal processing;

[0116] Perform signal normalization processing;

[0117] In this embodiment, the preprocessing process in step S1 is optimized; to generate the standardized data stream required for subsequent analysis, the preprocessing steps specifically include:

[0118] Perform timestamp alignment: The purpose is to resolve the inconsistency in time reference of data from different monitoring devices; by aligning all data streams to a unified, high-precision time coordinate, the temporal relationship between different data is ensured to be accurate, which is the basis for any correlation analysis.

[0119] Outlier removal is performed to identify and remove obvious erroneous data points caused by sensor malfunctions, communication interference, etc. In this embodiment, statistical methods or filtering algorithms can be used to complete this step to prevent these outliers from seriously interfering with subsequent feature calculations.

[0120] Perform signal normalization processing: The purpose is to eliminate the influence of different physical units and numerical ranges on the algorithm; through methods such as maximum-minimum normalization or Z-score standardization, data of different units are mapped to a unified range to ensure that the weight allocation of different features is fair in the subsequent risk factor calculation and will not be biased due to differences in units.

[0121] By implementing the above-described specific preprocessing procedures, this embodiment significantly improves the quality and consistency of input data; timestamp alignment ensures the effectiveness of event causal analysis, outlier removal enhances the robustness of the algorithm, and signal normalization ensures the fairness of multi-dimensional feature fusion; these steps work together to form a solid data quality defense line, bringing enhanced technical effects to the reliability and accuracy of the entire supervision method.

[0122] This invention no longer relies on monitoring a single physical quantity, but instead calculates a set of key characteristic parameters that can characterize the nonlinear dynamics of the power grid from different dimensions and levels based on standardized data streams. It introduces the fractal dimension of electromagnetic transient harmonic fingerprints, which can keenly capture microstructural anomalies caused by latent equipment faults. Furthermore, calculating the fractal complexity of decision trees quantifies the potential risks at the management level, such as scheduling rules. Crucially, by constructing a node-level security event correlation tensor and calculating the topological entropy of the entire network's security situation, this invention surpasses the limitations of traditional binary correlation analysis. It utilizes conditional mutual information to reveal the high-order coupling relationships and complex dependency patterns prevalent in the power grid, thereby enabling accurate assessment of the disorder and uncertainty of the entire system from a macroscopic level of network topology.

[0123] The specific steps for constructing a node-level security event correlation tensor in S2 are as follows:

[0124] Based on conditional mutual information in multivariate information theory, we calculate the information correlation degree between events of two other nodes under the condition that an event occurs at a certain node, so as to construct a node-level security event correlation degree tensor.

[0125] In this embodiment, the step of constructing the node-level security event correlation tensor in step S2 is explained; the underlying logic of this step is as follows:

[0126] Based on conditional mutual information in multivariate information theory, complex dependencies between nodes are quantified. For any three nodes i, j, and k in a power grid, the system calculates the information correlation degree between events at node i and node j, given a security event occurring at node k. This correlation degree is represented by tensor elements. The value of ;; The calculation process traverses all possible three-node combinations in the network, and finally forms a complete third-order tensor; The technical motivation of this method is that risk propagation and fault association in the power grid are often not simple point-to-point relationships, but complex network effects; Conditional mutual information is the ideal mathematical tool to capture such association between the two in the context of a given third party;

[0127] By employing conditional mutual information to construct the correlation tensor, this embodiment can reveal higher-order coupling relationships that traditional binary correlation analysis cannot capture. This enables the system to more accurately characterize complex dependency patterns such as triangular coupling or common causes in the power grid. The resulting gain is that the system can more accurately identify critical paths and core node clusters of risk propagation, providing a deeper and more fundamental insight for subsequent risk positioning and decision support.

[0128] The security threat propagation delay kernel function preset in S4 is constructed using a stretching exponential function to characterize the memory effect of risk propagation.

[0129] In this embodiment, the construction method of the preset security threat propagation time delay kernel function in step S4 is limited; in order to more realistically simulate the memory effect of risk propagation in the power grid, this embodiment specifies that the kernel function is constructed using a stretching exponential function;

[0130] The preset function is defined as:

[0131]

[0132] In this expression, ( () represents a time delay; It is the characteristic decay rate, with the dimension of the reciprocal of time. Its physical meaning is the basic speed at which the influence of risk dissipates over time. It is a dimensionless stretching exponent between 0 and 1, and its physical meaning is to describe the memory strength of the propagation process; when When, the function degenerates into a standard exponential decay; when At that time, the decay process exhibits significant non-Markovian characteristics or a memory effect; parameters and The value is obtained by statistically fitting a large amount of historical cascading failure data of the power grid, and is an empirical parameter with a clear physical source and statistical significance.

[0133]

[0134] By employing a stretched exponential function as the time-delay kernel function, this embodiment can more accurately model the real physical process of risk propagation in the power grid; compared with simple exponential decay, this function better captures the long-term tail effect and cumulative characteristics of risk impact; the resulting gain technical effect is that the calculated normalized risk index... It can more accurately reflect the ongoing impact of historical events on the current security status, making the risk assessment results more consistent with physical reality and avoiding the underestimation of potential risks due to model simplification.

[0135] The specific warning information generated in S7 includes:

[0136] By retrospectively analyzing the composition of the normalized risk index, the main sources of current risk are identified, and the power grid areas with the closest risk correlation are located using the node-level security event correlation tensor.

[0137] In this embodiment, the content of the warning information generated in step S7 is specified and enriched; when the warning condition is triggered, the system generates not a simple alarm signal, but a detailed report containing in-depth diagnostic information; the process of generating this warning information specifically includes:

[0138] By retrospectively analyzing the normalized risk index The system analyzes the composition of instantaneous risk factors to determine the main sources of current risk. The contribution of each component; if The increase is mainly due to If the contribution is positive, the system determines that the current risk mainly originates from the microscopic state anomaly of the physical equipment; if the contribution mainly comes from... In this case, the risk mainly stems from the macroscopic complexity of the system's relationships or the excessive complexity of the decision-making rules;

[0139] Based on the above judgment, the node-level security event correlation tensor is used. Locate the power grid area with the closest risk correlation; if the risk is identified as mainly caused by topological entropy The increase indicates a potential problem; the system will conduct further analysis. Identify the elements with the highest values ​​and determine which nodes form triples that contribute the most to the uncertainty of the overall network, thereby pinpointing the risk to a specific power grid area or key node cluster.

[0140] The early warning information generated in this way is far more valuable than a simple alarm; it not only informs maintenance personnel of risks, but also clearly answers the two key questions of what the risks are and where they are. The resulting technical benefits are that the early warning information becomes highly specific and actionable, providing direct decision support for maintenance personnel to quickly diagnose problems and formulate precise response strategies, greatly shortening fault response time and improving the efficiency and accuracy of emergency response.

[0141] This invention organically integrates microscopic physical risks with macroscopic network risks to calculate a comprehensive instantaneous risk factor. Furthermore, to reflect the accumulation and memory effect of risks, this invention innovatively uses a stretched exponential function to construct a kernel function for the propagation time delay of security threats, and generates a normalized risk index through normalized convolution operations. This design can more realistically model the long-term tail effect of risk propagation in the power grid, enabling the risk assessment results to accurately reflect the continuous impact of historical events on the current security status, and avoiding the underestimation of potential risks due to oversimplification of the model. Example 2

[0142] A power grid safety monitoring system includes:

[0143] The data acquisition and preprocessing module is used to acquire and preprocess multi-source heterogeneous monitoring data of the power grid to generate a standardized data stream;

[0144] The multi-dimensional feature extraction module is used to receive standardized data streams, calculate the fractal dimension of electromagnetic transient harmonic fingerprints, construct node-level security event correlation tensors, calculate the topological entropy of the entire network security situation, and calculate the fractal complexity of decision trees.

[0145] The risk assessment module is used to combine the features calculated by the multi-dimensional feature extraction module to calculate the instantaneous risk factor, and generate a normalized risk index based on the instantaneous risk factor and the preset security threat propagation delay kernel function.

[0146] The adaptive decision-making module is used to calculate the adaptive adjustment factor for adjusting the supervision strategy based on the normalized risk index, and generate dynamic alarm thresholds.

[0147] The early warning generation module is used to compare the normalized risk index with the dynamic alarm threshold. If the normalized risk index is greater than the dynamic alarm threshold, an early warning message is generated. If it is less than or equal to the threshold, the control system maintains the current supervision status.

[0148] This invention also provides a power grid safety supervision system, designed to apply the aforementioned power grid safety supervision method; the system includes an interconnected modular structure, with clear data flow and logical dependencies between modules to achieve a complete processing link from bottom-level data to top-level decision-making; including:

[0149] The data acquisition and preprocessing module executes step S1, which is configured to acquire multi-source heterogeneous monitoring data from the power grid and form a standardized data stream through standardized preprocessing steps.

[0150] The multi-dimensional feature extraction module performs step S2, which is used to receive a standardized data stream, calculate the fractal dimension of electromagnetic transient harmonic fingerprints, construct the node-level security event correlation tensor, calculate the topological entropy of the entire network security situation, and calculate the fractal complexity of the decision tree.

[0151] The risk assessment module, which executes steps S3 and S4, is configured to fuse multiple features calculated by the multi-dimensional feature extraction module to calculate an instantaneous risk factor, and based on this factor and a preset security threat propagation delay kernel function, generate a normalized risk index through normalized convolution operation.

[0152] The adaptive decision-making module executes steps S5 and S6 to receive the normalized risk index, calculate the adaptive adjustment factor, and further generate a dynamic alarm threshold based on the adjustment factor.

[0153] The early warning generation module executes step S7, which is configured to receive the normalized risk index and dynamic alarm threshold in real time, and continuously compare the two. Once the early warning conditions are met, an early warning message is generated and issued; otherwise, the current supervision status is maintained.

[0154] This embodiment provides a physical system capable of fully executing the aforementioned innovative methods; the modular structure makes the development, maintenance and upgrading of the system more convenient; each module has a clear function, together forming an efficient and reliable automated safety supervision platform, which solidifies complex algorithms and supervision logic into a deployable and runnable technical solution, thereby enabling the technical advantages of the method to be implemented in actual engineering, and has strong practical value.

[0155] The data flow between modules is configured as follows:

[0156] The multi-dimensional feature extraction module receives a standardized data stream output from the data acquisition and preprocessing module;

[0157] The risk assessment module receives the fractal dimension of the electromagnetic transient harmonic fingerprint, the topological entropy of the overall network security situation, and the fractal complexity of the decision tree calculated by the multi-dimensional feature extraction module.

[0158] The adaptive decision-making module receives the normalized risk index generated by the risk assessment module.

[0159] The early warning generation module receives the normalized risk index from the risk assessment module and the dynamic alarm threshold from the adaptive decision-making module.

[0160] In this embodiment, the data flow between its internal modules is explicitly configured to ensure that the entire system can operate efficiently and orderly according to the preset logic; the specific configuration of the data flow is as follows:

[0161] The multi-dimensional feature extraction module is configured to receive a standardized data stream output by the data acquisition and preprocessing module;

[0162] The risk assessment module is configured to receive the fractal dimension of the electromagnetic transient harmonic fingerprint, the topological entropy of the overall network security situation, and the fractal complexity of the decision tree calculated by the multi-dimensional feature extraction module, and to read preset risk model parameters, such as baseline values, from the system configuration library. and weighting coefficients wait;

[0163] The adaptive decision-making module is configured to receive a normalized risk index generated by the risk assessment module;

[0164] The early warning generation module is configured to simultaneously receive the normalized risk index from the risk assessment module and the dynamic alarm threshold from the adaptive decision-making module.

[0165] Through such a clear and rigorous data flow configuration, this embodiment ensures the unidirectional, timely, and accurate transmission of information within the system. This clear logical chain avoids the confusion and calculation errors that may be caused by cross-referencing of data, ensuring that every step of the calculation from the original data to the final warning is based on the correct preceding results. The resulting technical benefit is that the entire system operates with rigorous logic and the processing flow is efficient and stable, providing structural protection for achieving reliable and real-time power grid safety supervision, and enhancing the overall robustness and credibility of the system.

[0166] The most significant practical advantage of this invention lies in its adaptive decision-making and early warning mechanism. It abandons the rigid, fixed alarm thresholds of traditional technologies, generating dynamic alarm thresholds through an adaptive adjustment factor linked to the normalized risk index. This threshold automatically adjusts according to the real-time risk level of the system: when the system is safe, the alarm threshold is raised, enhancing tolerance to normal fluctuations; when the system is dangerous, the alarm threshold is lowered, increasing sensitivity to minor anomalies. This mechanism ensures a sensitive response to serious faults while effectively suppressing false alarm rates, significantly improving the efficiency of supervision.

[0167] When an early warning is triggered, this invention provides not just a simple alarm signal, but a detailed report containing in-depth diagnostic information. The system can accurately determine whether the current risk originates from the physical equipment level or the system topology level by tracing back the composition of the risk index, and quickly locate the power grid area or key node cluster with the closest risk correlation using the correlation tensor. This makes the early warning information highly specific and operable, providing clear decision support for operation and maintenance personnel and greatly improving the efficiency and accuracy of emergency response.

[0168] The complete power grid safety monitoring system provided by this invention, through modular structural design and clear data flow configuration, solidifies the above-mentioned innovative methods into a logically rigorous and highly efficient physical solution. The system ensures that every step of the process, from raw data acquisition to final early warning generation, is accurate and error-free, providing a solid platform guarantee for the practical application of technological advantages in engineering projects, and has extremely strong practical value.

[0169] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method of grid safety oversight, the method comprising: The method comprises the following steps: S1, collecting and preprocessing multi-source heterogeneous monitoring data of the power grid to generate standardized data flow; S2, based on the standardized data flow, calculating electromagnetic transient harmonic fingerprint fractal dimension, constructing node-level security event correlation degree tensor, calculating global security situation topology entropy, and calculating decision tree fractal complexity; S3, combining electromagnetic transient harmonic fingerprint fractal dimension, global security situation topology entropy and decision tree fractal complexity to calculate instantaneous risk factor; S4, based on the instantaneous risk factor and the preset security threat propagation time lag kernel function, generating a normalized risk index through normalized convolution operation; S5, according to the normalized risk index, calculating an adaptive adjustment factor for adjusting the supervision strategy; S6, according to the adaptive adjustment factor, adjusting between the preset maximum alarm threshold and the minimum alarm threshold to generate a dynamic alarm threshold; S7, comparing the normalized risk index with the dynamic alarm threshold, if the normalized risk index is greater than the dynamic alarm threshold, generating an early warning information, if the normalized risk index is less than or equal to the dynamic alarm threshold, maintaining the current supervision state; The step of constructing the node-level security event correlation degree tensor in S2 is specifically: Based on the conditional mutual information in the multivariate information theory, the information correlation degree between the events of the other two nodes under the condition that an event occurs at a certain node is calculated to construct the node-level security event correlation degree tensor; The preset security threat propagation time lag kernel function in S4 is constructed by using a stretching exponential function to represent the memory effect of risk propagation; One element of this tensor is defined as: wherein, is a random variable representing that a security-related event occurs at node x; is a scalar representing the triadic relational strength or information coupling degree between nodes ; : respectively denote the random variables of the occurrence of a safety-related event at node i, node j, node k; Instantaneous risk factor The formula for calculating the instantaneous risk factor is: In this expression, the definition of each parameter is as follows: is the decision tree fractal complexity at time t; is the security benchmark value of the decision tree fractal complexity, which is a dimensionless constant calibrated by historical data; is the network security situation topology entropy at time t, which needs to be normalized by dividing by an entropy benchmark value based on historical data or the theoretical maximum entropy value to make it a dimensionless index, thereby ensuring the dimensional consistency of both sides of the formula; is the normalized network security situation topology entropy; is the exponential function; is the electromagnetic transient harmonic fingerprint fractal dimension at time t; is the normal reference value of harmonic fractal dimension, which is a dimensionless constant calibrated by historical data; , , are dimensionless weight coefficients, whose values are calibrated by regression analysis on historical failure data or according to an expert system, to balance the contribution of different risk sources; this formula organically combines macro-management risks and micro-physical risks together.

2. A power grid safety supervisory method according to claim 1, characterized in that, The multi-source heterogeneous monitoring data in S1 comprises: Synchronization vector data of a phasor measurement unit; Low-frequency oscillation data of a wide area measurement system; High-frequency electromagnetic transient signals captured by a dedicated sensor.

3. The method of claim 1, wherein, The preprocessing in S1 comprises: Performing timestamp alignment processing; Performing outlier rejection processing; Performing signal normalization processing.

4. The method of claim 1, wherein, The early warning information generated in S7 specifically comprises: By tracing the composition of the normalized risk index, the main source of the current risk is determined, and the node-level security event correlation degree tensor is used to locate the power grid area with the closest risk correlation.

5. A power grid safety supervision system applied to the power grid safety supervision method of any one of claims 1 to 4, characterized in that, Comprise: A data acquisition and preprocessing module for collecting and preprocessing multi-source heterogeneous monitoring data of the power grid to generate standardized data flow; A multi-dimensional feature extraction module for receiving the standardized data flow, calculating electromagnetic transient harmonic fingerprint fractal dimension, constructing node-level security event correlation degree tensor, calculating global security situation topology entropy, and calculating decision tree fractal complexity; A risk assessment module for calculating instantaneous risk factor in combination with the features calculated by the multi-dimensional feature extraction module, and generating a normalized risk index based on the instantaneous risk factor and the preset security threat propagation time lag kernel function; An adaptive decision module for calculating an adaptive adjustment factor for adjusting the supervision strategy according to the normalized risk index, and generating a dynamic alarm threshold; An early warning generation module for comparing the normalized risk index with the dynamic alarm threshold, if the normalized risk index is greater than the dynamic alarm threshold, generating an early warning information, if less than or equal to, the system maintains the current supervision state.

6. A grid security oversight system according to claim 5, wherein, The data flow between the modules is configured as follows: a multi-dimensional feature extraction module, which receives the standardized data stream output by the data acquisition and preprocessing module; a risk assessment module, which receives the electromagnetic transient harmonic fingerprint fractal dimension, the whole network security situation topological entropy, and the decision tree fractal complexity calculated by the multi-dimensional feature extraction module; an adaptive decision module, which receives the normalized risk index generated by the risk assessment module; an early warning generation module, which receives the normalized risk index from the risk assessment module and the dynamic alarm threshold from the adaptive decision module.

Images