Power safety accident early warning and tracing method based on multi-dimensional feature analysis

Abstract

The present application belongs to the technical field of power accident early warning, and particularly relates to a power safety accident early warning and tracing method based on multi-dimensional feature analysis, which comprises the following steps: obtaining active power sequence, voltage sequence and frequency sequence with unified time label on a high-voltage bus; then, calculating power mutation component, voltage drop component and frequency instability degree respectively, and combining the three to construct a multi-index coordination matrix, and then extracting the maximum eigenvalue and eigenvector; using the difference between the change proportion of the voltage sequence and the active power sequence and the preset system nominal correlation value to obtain the physical correlation factor; finally, fusing the physical correlation factor and the maximum eigenvalue to obtain the final risk index, and comparing it with the preset risk threshold to determine the power safety accident state. The present application reduces the false positives caused by sensor interference, and improves the accuracy and response efficiency of power accident tracing.

Description

Technical Field

[0001] This invention relates to the field of power accident early warning technology. More specifically, this invention relates to a method for early warning and tracing power safety accidents based on multidimensional feature analysis. Background Technology

[0002] In the field of power system risk management and business management decision-making technology, the safe operation of the power grid is the fundamental guarantee for the stable operation of social production and life. With the expansion of the power grid scale and the increase in the complexity of connected loads, real-time monitoring of key parameters such as power, voltage and frequency during the operation of the power grid, as well as early warning and tracing of potential accident risks, has become the core means to ensure the stability of the power decision-making system and reduce the losses from safety accidents.

[0003] To ensure the safe operation of the power grid, data-driven models are currently widely used for accident monitoring. For example, Chinese patent application CN120471428A discloses a method and device for early warning of power system operation safety based on hybrid XLSTM. This method involves acquiring a sequence of characteristic signals including voltage and power angle, filling in missing values, and then inputting the sequence into a hybrid neural network for deep feature extraction, ultimately outputting a classification result to determine the system risk status.

[0004] However, in the process of model analysis and risk prediction, there is a problem that the pure data-driven mode lacks underlying physical mechanism constraints. When the secondary circuit of the substation is affected by electromagnetic pulse interference and produces pseudo data jumps, the neural network is prone to misjudging non-physical noise as system transient risks, resulting in frequent false alarms in daily monitoring. At the same time, the final output of such black-box models is only a risk classification label, which cannot analyze the implicit nonlinear coupling relationship between multi-dimensional parameters, resulting in unclear guidance on the physical causes of accidents and difficulty in meeting the subsequent needs of tracing the source of faults. Summary of the Invention

[0005] To address the aforementioned technical problems of false alarms and difficulty in tracing power safety accidents, this invention provides a method for power safety accident early warning and tracing based on multi-dimensional feature analysis, including: The electrical signal fluctuations on the primary side of the high-voltage bus are collected and converted into small voltage and current signals on the secondary side. These small voltage and current signals are sampled at high speed to obtain an active power sequence, voltage sequence, and frequency sequence with a unified time label. A time-sequential window is set, and based on the instantaneous change and local fluctuation of the active power sequence, the power surge score of the high-voltage bus within the window is obtained. Based on the static deviation and dynamic rate of change of the voltage sequence, the voltage drop score of the high-voltage bus within the window is obtained. Based on the real-time offset and cumulative deviation of the frequency sequence, the frequency instability of the high-voltage bus within the window is obtained. A multi-index collaborative matrix is ​​constructed by combining the power surge score, voltage drop score, and frequency instability. Eigenvalue decomposition is performed on the multi-index collaborative matrix to extract the maximum eigenvalue and the corresponding eigenvector. A physical correlation factor is obtained based on the difference between the change ratio of the voltage sequence and the active power sequence and the preset system nominal correlation value. The physical correlation factor is combined with the maximum eigenvalue to obtain the final risk index. Based on the relationship between the final risk index and the preset risk threshold, the power safety accident status is determined.

[0006] This invention achieves in-depth assessment of the safety status and evolution trend of the power system by comprehensively extracting multi-dimensional operating characteristics such as power, voltage, and frequency of high-voltage buses and combining them with physical correlation factors. It utilizes a multi-index synergistic matrix to deeply capture the inherent coupling correlation between data sequences of different dimensions and performs logical verification of observed waveform data using physical correlation factors. This improves the sensitivity of transient risk identification while reducing false alarms induced by non-physical electromagnetic interference or equipment sampling errors. Furthermore, through the extraction and weight analysis of multi-dimensional feature vectors, the dominant physical dimensions leading to grid instability can be identified in real time. This provides crucial data guidance for fault source location and nature tracing after an accident, enabling the entire early warning process to not only promptly detect potential safety risks but also assist management personnel in identifying system weaknesses. This enhances the grid's comprehensive perception and dispatch efficiency in responding to sudden disturbances under complex operating conditions.

[0007] Preferably, the step of acquiring the electrical signal fluctuations on the primary side of the high-voltage bus and converting them into small voltage and current signals on the secondary side includes: Voltage transformers, current transformers, and synchronous phasor measurement terminals are deployed on the primary side of the high-voltage busbar and in the feeder circuits of key load nodes in the substation. The voltage transformers and current transformers collect the electrical signal fluctuations on the primary side of the high-voltage busbar and convert the electrical signal fluctuations into small voltage and current signals on the secondary side using the transformation ratio of the transformer windings.

[0008] Preferably, the high-speed sampling of the voltage and current small signals to obtain an active power sequence, voltage sequence, and frequency sequence with a unified time label includes: The synchronous phasor measurement terminal is connected to the secondary circuit of the current transformer. The small voltage and current signals received from the secondary side are sampled at high speed to obtain the high-speed sampled digital signal. Then, the digital signal is reversed and restored using the preset current transformer ratio coefficient. Through the instantaneous power algorithm and the zero-crossing detection algorithm, the active power sequence, voltage sequence and frequency sequence with a unified time label are obtained.

[0009] Preferably, the power mutation satisfies the following relationship: ; In the formula, Indicates the high-voltage busbar at the 1st Power mutations within the sampling window; Represents the first active power sequence. Active power at the sampling time; Represents the first active power sequence. Active power at the sampling time; Indicates the active power sequence in the th order. The arithmetic mean within the window of sampling time; Indicates the active power sequence in the th order. The standard deviation within the sampling window; Represents the absolute value symbol.

[0010] This invention, by evaluating the instantaneous changes and local fluctuations in the active power sequence, can accurately identify the impact characteristics of the grid load from a transient perspective of energy balance. By analyzing the power difference between adjacent sampling times and its deviation from the statistical distribution within the window, it can capture the discontinuous evolution of energy flow on the time axis, reflecting the impact intensity on the supply and demand balance within the system, and reducing the problem of untimely identification of low-probability, high-intensity power pulse signals under the traditional single power average monitoring mode.

[0011] Preferably, the voltage drop satisfies the following relationship: ; In the formula, Indicates the high-voltage busbar at the 1st Voltage drop within the sampling window; Indicates the system's rated voltage; Represents the voltage sequence of the first... Voltage at the sampling time; This represents the derivative of voltage with respect to time. This indicates the preset voltage change rate reference; Represents the absolute value symbol.

[0012] This invention achieves comprehensive monitoring of grid voltage quality in two dimensions: spatial displacement depth and temporal evolution rate, by evaluating the static deviation amplitude and dynamic change rate of voltage sequences. By synergistically analyzing the deviation of voltage from the nominal value and the rate of voltage change over time, it can identify voltage collapse risks. Thus, when facing different types of transient disturbances or remote short-circuit faults, it can provide more reliable evaluation results that are more in line with the actual operation of the power system, reducing the limitations of a single indicator in judging complex voltage fluctuation attributes.

[0013] Preferably, the frequency instability satisfies the following relationship: ; In the formula, Indicates the high-voltage busbar at the 1st Frequency instability within the sampling window; Represents the frequency sequence of the th Frequency value at the sampling time; Indicates the rated frequency; Indicates the preset frequency offset reference; Indicates the first Frequency value at the sampling time; Indicates the number of sampling times within the window; Represents the absolute value symbol.

[0014] Preferably, obtaining the maximum eigenvalue and the corresponding eigenvector includes: The mean values ​​of the active power sequence and voltage sequence within the window are calculated separately. The arithmetic mean of the products of the power deviation values ​​of the active power at all sampling times from the power mean value and the voltage deviation values ​​of the voltage at the corresponding sampling times is calculated to obtain the original covariance of the active power sequence and voltage sequence. The original covariance is then divided by the product of the preset reference power and reference voltage to obtain the dimensionless covariance of the active power sequence and voltage sequence. By combining the power surge, voltage drop, frequency instability and dimensionless covariance of the high-voltage bus within the window, a 3×3 symmetric multi-index cooperative matrix is ​​constructed. Eigenvalue decomposition is performed on the multi-index cooperative matrix to extract the maximum eigenvalue and the corresponding eigenvector.

[0015] Preferably, the acquisition of the physical correlation factor includes: Calculate the ratio of the normalized difference of voltage between the current sampling time and the previous sampling time to the normalized difference of active power between the current sampling time and the previous sampling time to obtain the voltage power change rate; calculate the reciprocal of the sum of 1 and the absolute values ​​of the difference between the voltage power change rate and the preset dimensionless system nominal correlation value to obtain the physical correlation factor.

[0016] This invention calculates the ratio of voltage to power changes and compares it in real time with a preset system nominal correlation value. It uses the inherent physical constraints of the power system to reverse verify the authenticity of the monitoring data. By judging whether the observed dynamic change ratio conforms to the inherent energy mapping characteristics of the line, it can identify and filter out non-physical signals that do not conform to the operating rules. Thus, energy balance constraints are introduced into the accident judgment logic, reducing the risk of false early warnings caused by digital sampling anomalies and improving the credibility of risk judgment conclusions.

[0017] Preferably, determining the state of a power safety accident includes: When the final risk index is greater than or equal to the risk threshold, the current power grid is determined to be in a transient unstable state. The system triggers a risk warning command and sends it to the substation monitoring backend and the remote dispatch automation system. Through the communication interface, the relay protection device or load shedding controller is linked to execute the corresponding protection action. The maximum value of the absolute value of the three components of the feature vector corresponding to the multi-index collaborative matrix is ​​extracted. By identifying the component with the highest weight in the feature vector, the source of the fault can be located. When the final risk index is less than the risk threshold, the current power grid is determined to be in a safe operating state or a state of disturbance fluctuation.

[0018] This invention integrates physical correlation factors with the maximum eigenvalue of a multi-index collaborative matrix and uses eigenvectors to perform directional analysis of fault sources. When a risk index exceeds a safety threshold, it can locate the core cause of a power safety accident by analyzing the distribution characteristics of each physical dimension. This shortens the technical investigation and analysis cycle after an accident. Furthermore, in the final risk assessment stage, the filtering function of physical linkage logic enhances the robustness of the early warning system in complex electromagnetic environments.

[0019] Preferably, the acquisition of the symmetric multi-index synergy matrix includes: Power mutation, voltage drop, and frequency instability are respectively used as the power, voltage, and frequency dimensions on the main diagonal of the multi-index collaborative matrix. The dimensionless covariance is filled into the off-diagonal positions corresponding to the power and voltage dimensions in the matrix, and the remaining positions are filled with zeros to obtain a symmetric multi-index collaborative matrix.

[0020] The beneficial effects of this invention are as follows: By constructing a monitoring system with multi-dimensional parametric collaborative features, this invention optimizes the problems of low accuracy in identifying safety risks and frequent false alarms in power systems during complex transient processes. It integrates scattered physical indicators into unified global features using a multi-index collaborative matrix, and extracts the system's dominant evolutionary trends through eigenvalue decomposition. This not only captures minute jumps within each indicator but also reveals hidden nonlinear coupling relationships between parameters. Simultaneously, this invention introduces a physical correlation factor as an evaluation benchmark for signal authenticity. By determining whether the voltage and power disturbance ratio conforms to the inherent system energy balance constraints of the line, it provides underlying physical laws supporting the monitoring data, thereby reducing false warnings induced by non-physical noise or sampling anomalies. This results in accident warnings with not only high sensitivity but also the ability to trace the source of faults, enabling rapid identification of weak components while detecting risks, thus improving the power system's perception depth and operational safety assurance level in response to sudden faults. Attached Figure Description

[0021] Figure 1 This is a flowchart of the power safety accident early warning and tracing method based on multidimensional feature analysis in this invention; Figure 2 This is a schematic diagram of power surge monitoring within the window of the high-voltage busbar; Figure 3 This is a schematic diagram of voltage drop monitoring of the high-voltage busbar within the window; Figure 4 This is a schematic diagram illustrating real-time monitoring of the largest eigenvalue of the collaborative matrix; Figure 5 This is a diagram illustrating the real-time determination of the final risk index. Detailed Implementation

[0022] This invention discloses a method for early warning and tracing power safety accidents based on multidimensional feature analysis, referring to... Figure 1 This includes steps S100-S700: S100, high-speed sampling and multi-dimensional sequence acquisition of high-voltage bus signals.

[0023] It should be noted that since the high voltage and high current signals on the primary side of the power system cannot be directly read by semiconductor monitoring equipment, and the transient fluctuation process of the power grid contains a large number of microsecond-level high-frequency characteristics, considering that the high voltage and high current signals can be mapped to small signals on the secondary side by using the principle of electromagnetic induction, and using high-frequency sampling technology to realize the digital restoration of the original waveform details, this invention uses mutual inductance windings in conjunction with a synchronous phasor measurement terminal to convert the physical fluctuations of the high voltage bus into a digital sequence with a unified time reference, thereby providing a high-fidelity data source for reflecting the operating status of the power grid.

[0024] Specifically, voltage transformers, current transformers, and synchronous phasor measurement terminals are deployed on the primary side of the high-voltage busbar and in the feeder circuits of key load nodes in the substation. The voltage transformers and current transformers collect the electrical signal fluctuations on the primary side of the high-voltage busbar and convert the electrical signal fluctuations into small voltage and current signals on the secondary side using the transformation ratio of the transformer windings.

[0025] A synchronous phasor measurement terminal is connected to the secondary circuit of the current transformer. The received small voltage and current signals from the secondary side are sampled at high speed to obtain a high-speed sampled digital signal. Subsequently, the digital signal is reverse-engineered using a preset current transformer ratio coefficient. Then, through an instantaneous power algorithm and a zero-crossing detection algorithm, an active power sequence, voltage sequence, and frequency sequence with a unified time label are obtained. For example, the high-speed sampling frequency is 1000Hz to capture detailed information during transient fluctuations. It should be noted that the instantaneous power algorithm and zero-crossing detection algorithm are existing technologies and will not be elaborated upon here.

[0026] Thus, the active power sequence, voltage sequence, and frequency sequence were obtained.

[0027] S200, Set the window and calculate the power surge of the high-voltage bus.

[0028] It should be noted that, since the switching of power electronic equipment can cause discontinuous jumps in active power in a short period of time, and considering that the instantaneous difference of the power sequence can characterize the severity of the load shock, this invention constructs a power jump component that includes a difference normalization term and a coefficient of variation weight term, and uses the statistical mean within the window to scale the instantaneous jump, thereby characterizing the degree of deviation of the load fluctuation relative to the local energy background.

[0029] Specifically, based on the instantaneous fluctuation increment and local dispersion of the active power sequence, the power mutation component of the high-voltage bus within the window is calculated, including: With the first Using the sampling time as the endpoint, a sequence is set up, including the data collected in chronological order. A window of sampling times. For example, the number of sampling times... Set according to sampling frequency and analysis requirements. .

[0030] Calculate the arithmetic mean and standard deviation of the active power series within the window.

[0031] It should be further explained that this invention, based on the concept of power system per-unit scaling, divides the instantaneous difference and standard deviation by the arithmetic mean, aiming to eliminate the dimensional differences caused by different load bases, transforming absolute fluctuations into relative proportions, and obtaining the degree of deviation representing the intensity of instantaneous fluctuations. On this basis, the degree of deviation is multiplied by the coefficient of variation, which reflects the overall degree of disorder. Essentially, this utilizes the local statistical background to perform confidence weighting on transient impacts, ultimately yielding a power mutation score. This power mutation score suppresses isolated high-frequency noise and amplifies the true instability characteristics.

[0032] The power mutation satisfies the following relationship: ; In the formula, Indicates the high-voltage busbar at the 1st Power mutations within the sampling window; Represents the first active power sequence. Active power at the sampling time; Represents the first active power sequence. Active power at the sampling time; Indicates the active power sequence in the th order. The arithmetic mean within the window of sampling time; Indicates the active power sequence in the th order. The standard deviation within the sampling window; Represents the absolute value symbol.

[0033] In this relation, This indicates the degree of deviation of the instantaneous fluctuation intensity of the power grid load from the local average energy level. The larger the value, the more likely it is to be the first. If the power grid experiences a load surge or equipment switching at the sampling time, the instantaneous power jump is abnormally drastic and can easily cause power grid frequency fluctuations; conversely, if the power grid is not affected, it indicates that the power grid is operating smoothly and the load changes are in a slow and continuous evolution state. This indicates the overall degree of disorder in the distribution of energy in the power grid. The larger the value, the more likely the power grid is in a state of high-frequency, large-amplitude irregular oscillation within the window, indicating poor load consistency and suggesting the risk of supply-demand imbalance or continuous process disturbance within the system; conversely, the smaller the value, the more likely the energy output within the window is dense and regular.

[0034] For example, Figure 2 This is a schematic diagram of power surge monitoring within a window on a high-voltage bus. The horizontal axis represents the window number, and the vertical axis represents the power surge score. The diagram shows a jump in the score at a certain window number, reflecting a drastic change in the instantaneous change and local fluctuation of the active power sequence. This can capture energy distribution disturbances caused by grid load shocks or equipment switching.

[0035] Thus, the power mutation of the high-voltage bus within the window was obtained.

[0036] S300, Calculate the voltage drop of the high-voltage bus.

[0037] It should be noted that, considering that voltage sequences can directly reflect the current reactive power balance of the power grid, and that the characteristics of voltage changes over time can reflect the severity of the evolution of disturbance events, this invention can characterize the current reactive power support gap of the power grid by calculating the difference between the voltage and the rated voltage, which is beneficial for assessing the static depth of equipment under voltage threat; by calculating the derivative of voltage with respect to time, it can characterize the steepness of voltage fluctuations, which is beneficial for capturing transient impacts caused by instantaneous faults or large-capacity load switching. Thus, by combining the two parts, a dual-dimensional characterization of power grid voltage quality in terms of spatial displacement and time can be achieved.

[0038] It should be further explained that the present invention divides the static voltage difference by the rated voltage and uses the physical law that voltage and energy are positively correlated to nonlinearly amplify the severe static deviation; on this basis, a constant penalty term containing the dynamic rate of change is multiplied by the static deviation, which is essentially using the voltage derivative with respect to time to dynamically weight the spatial drop depth, and finally obtains the dimensionless voltage drop score.

[0039] Specifically, based on the static deviation and dynamic rate of change of the voltage sequence, the voltage sag of the high-voltage bus within the window is calculated, and the voltage sag satisfies the following relationship: ; In the formula, Indicates the high-voltage busbar at the 1st Voltage drop within the sampling window; This indicates the system's rated voltage, which is set according to the design nominal voltage level of the high-voltage busbar. Represents the voltage sequence of the first... Voltage at the sampling time; This represents the derivative of voltage with respect to time. This indicates the preset voltage change rate benchmark, set to 500V / s, which is set according to the maximum voltage fluctuation slope allowed by the power system under normal operating conditions and the action delay requirements of the protection device. Represents the absolute value symbol.

[0040] In this relation, Indicates the high-voltage busbar at the 1st The static voltage difference between the measured operating voltage and the rated voltage at the sampling time indicates that the power grid may be under heavy load or a remote short circuit fault has occurred, and the equipment faces the risk of undervoltage damage; conversely, it indicates that the voltage regulation performance of the power grid is good and the power supply quality is in a stable state. This represents the absolute value of the voltage derivative with respect to time. The larger the voltage, the more likely it is that there has been a sudden drop or rise in voltage, and there may have just been a momentary fault in the power grid, or the start-up or disconnection of a large-capacity inductive load, which may easily trigger the protection device to malfunction or cause voltage collapse; conversely, the voltage change process is gradual and the operation process is highly controllable.

[0041] For example, Figure 3 This is a schematic diagram of voltage sag monitoring of the high-voltage bus within a window. The horizontal axis represents the window number, and the vertical axis represents the voltage sag score. The curve in the figure shows a trend of first rising and then gradually recovering, reflecting the evolution characteristics of the static deviation and dynamic rate of change of the voltage sequence in the transient process. It can identify the voltage drop and the attenuation slope of the evolution towards instability.

[0042] Thus, the voltage drop of the high-voltage bus within the window was obtained.

[0043] S400, Calculate the frequency instability of the high-voltage bus.

[0044] It should be noted that, due to the inertial characteristics and time-cumulative effects of frequency fluctuations, instantaneous frequency deviations are often accompanied by a continuous loss or surplus of the system's rotational kinetic energy. Considering that accumulation calculations can be used to calculate the energy accumulation of the frequency deviation state over time, which is beneficial for distinguishing between sporadic pulse interference and persistent frequency modulation failures, this invention calculates the frequency instability degree and uses a preset deviation benchmark to standardize and measure the frequency fluctuation trajectory, thereby extracting the cumulative instability characteristics of the frequency sequence from an instantaneous point to a time period within a window.

[0045] Specifically, based on the real-time offset of the frequency sequence and the historical cumulative deviation, the frequency instability of the high-voltage bus within the window is calculated, and the frequency instability satisfies the following relationship: ; In the formula, Indicates the high-voltage busbar at the 1st Frequency instability within the sampling window; Represents the frequency sequence of the th Frequency value at the sampling time; This indicates the rated frequency, set to 50Hz. This indicates the preset frequency offset reference, set to 0.2Hz, which is set according to the allowable frequency deviation limit in the normal operation procedures of the power system. Indicates the first Frequency value at the sampling time; Indicates the number of sampling times within the window; Represents the absolute value symbol.

[0046] In this relation, Indicates the first The absolute deviation of the frequency value at the sampling time from the rated frequency. The larger the value, the more severe the frequency fluctuation pressure the system is under. If intervention is not timely, it will lead to increased unit vibration or activation of the low-frequency automatic load shedding device. Conversely, it indicates that the instantaneous energy balance of the power grid is good. The cumulative deviation energy represents the frequency sequence within the window. The larger the cumulative deviation energy, the more likely the power grid frequency cannot recover to the rated value for a long time within the window, or is in a state of continuous large oscillation, indicating that irreversible process drift has occurred; conversely, it indicates that the control system has good recovery and damping characteristics.

[0047] Thus, the frequency instability of the high-voltage bus within the window was obtained.

[0048] S500: Construct a multi-index collaborative matrix and extract the maximum eigenvalue and eigenvector.

[0049] It should be noted that, considering that the covariance matrix can characterize the synchronous change trend of multiple variables within the same window, if the voltage also increases when the power increases, the covariance is positive, indicating that the two change trends are positively correlated; if the power increases while the voltage decreases, the covariance is negative, indicating that the two are negatively correlated; if one parameter fluctuates while another parameter remains stable, the covariance is close to 0, indicating that the two are unrelated. Therefore, this invention constructs a multi-index co-variance matrix that includes power mutation component, voltage drop component, frequency instability, and dimensionless covariance, and uses eigenvalue decomposition technology to extract the core components of system fluctuations, thereby transforming discrete operating parameters into tensor features that characterize the overall evolution trend of the system.

[0050] Specifically, the mean values ​​of the active power sequence and voltage sequence within the window are calculated separately. The arithmetic mean of the products of the active power deviation from the power mean at all sampling times and the voltage deviation from the voltage mean at the corresponding sampling times is calculated to obtain the original covariance of the active power sequence and voltage sequence. This original covariance is then divided by the product of a preset reference power and reference voltage to obtain the dimensionless covariance of the active power sequence and voltage sequence. For example, the reference power is set to 100W, based on the rated active load level of the high-voltage bus under normal operating conditions; the reference voltage is set to 10000V, corresponding to the design nominal voltage level of the high-voltage bus of the strain gauge substation, ensuring that the obtained dimensionless covariance can eliminate the difference in numerical magnitude between different physical dimensions.

[0051] Combining the power surge, voltage dip, frequency instability, and dimensionless covariance of the high-voltage bus within the window, a 3×3 symmetric multi-index cooperative matrix is ​​constructed. The power surge, voltage dip, and frequency instability are respectively used as the power, voltage, and frequency dimensions on the main diagonal of the multi-index cooperative matrix. The dimensionless covariance is filled into the off-diagonal positions corresponding to the power and voltage dimensions in the matrix, with the remaining positions padded with zeros. Eigenvalue decomposition is performed on the multi-index cooperative matrix to extract the largest eigenvalue and the corresponding eigenvector. For example, the eigenvalue decomposition uses the Jacobi iteration method, which is existing technology and will not be elaborated upon here. It should be noted that the eigenvector contains three components: power, voltage, and frequency.

[0052] For example, Figure 4 This is a schematic diagram of the monitoring of the maximum eigenvalue of the cooperative matrix. The horizontal axis represents the window number, and the vertical axis represents the intensity of the maximum eigenvalue. The diagram reflects the dominant trend after eigenvalue decomposition of the multi-index cooperative matrix. By fusing multi-dimensional features such as power, voltage, and frequency into a matrix, the nonlinear coupling relationship between different physical parameters is shown.

[0053] Thus, the largest eigenvalue and eigenvector are obtained.

[0054] S600, Calculate the physical correlation factor.

[0055] It should be noted that, because the secondary circuit of a substation is susceptible to interference from spatial electromagnetic pulses or sampling channel noise, resulting in pseudo-data jumps that are out of sync with the physical reality, and considering that voltage and power disturbances caused by actual load variations are usually constrained by system energy balance, their change ratios should conform to the voltage-power linkage law. Therefore, this invention constructs a physical correlation factor by the deviation between the voltage-power change rate and the system's nominal correlation value, and uses inverse proportional mapping logic to rapidly attenuate non-physical characteristics, thereby suppressing logical judgment misleading caused by non-physical abnormal signals.

[0056] It should be further explained that the present invention calculates the voltage-power change rate by dividing the normalized difference between voltage and power, and calculates the absolute value of the deviation vector of the voltage-power change rate from the preset nominal correlation value of the system. This is mainly to use the inherent energy mapping characteristics of the line as the underlying physical benchmark to characterize the degree of violation of objective physical laws by the measured data. Finally, a constant 1 is added and the reciprocal is taken to map the divergent deviation value into a convergent and smooth physical correlation factor, thereby attenuating abnormal fluctuations that do not conform to the physical linkage law.

[0057] Preferably, the physical correlation factor of the high-voltage bus within the window is calculated based on the ratio of voltage to power changes and their deviation from the system's nominal correlation value. This physical correlation factor satisfies the following relationship: ; In the formula, This indicates the physical correlation factor of the high-voltage busbar within the window; Indicates the first The normalized difference between the voltage at the sampling time and the previous sampling time; Indicates the first The normalized difference in active power between the sampling time and the previous sampling time; This represents a preset dimensionless system nominal correlation value, which is set based on the inherent energy mapping characteristic value of the high-voltage bus. Represents the absolute value symbol.

[0058] In this relation, Indicates the rate of change of voltage power. The absolute value of the deviation vector of the voltage power change rate relative to the nominal associated value of the system indicates that the current signal fluctuation may be a spurious signal caused by non-physical external interference, sampling channel failure or sensor distortion. Conversely, the smaller the absolute value of the deviation vector, the more likely it is that the current fluctuation is a physical event caused by real load switching, short circuit fault or system oscillation, and the data has high authenticity and reliability.

[0059] Thus, the physical correlation factor of the high-voltage busbar within the window was obtained.

[0060] S700: Calculate the final risk index and determine the status of power safety accidents.

[0061] It should be noted that, considering that physical correlation factors have physical switch attributes, can weight the credibility of mathematical analysis results, and that the directionality of feature vectors can map the physical dimension that contributes the most to the feature value, this invention compares the final risk index with the preset risk threshold and extracts the components of the feature vector, thereby reducing the risk of false alarms in the risk identification process and enhancing the directionality of the fault source.

[0062] Specifically, the physical correlation factor is multiplied by the maximum eigenvalue to obtain the final risk index. The final risk index is then compared with a preset risk threshold, including: When the final risk index is greater than or equal to the risk threshold, the current power grid is determined to be in a transient unstable state. The system triggers a risk warning command and sends it to the substation monitoring backend and the remote dispatch automation system. Through the communication interface, the relay protection device or load shedding controller is linked to execute the corresponding protection action. The maximum value of the absolute value of the three components of the feature vector corresponding to the multi-index collaborative matrix is ​​extracted. By identifying the component with the highest weight in the feature vector, the source of the fault can be located.

[0063] When the final risk index is less than the risk threshold, the current power grid is determined to be in a safe operating state or a state of disturbance fluctuation.

[0064] For example, the risk threshold is set to 1. It should be noted that, because the component indicators of each dimension have been dimensionless before constructing the multi-indicator synergy matrix, and the value range of the physical correlation factor is limited to 0 to 1, under normal operating conditions, the scores of each indicator and the maximum eigenvalue after synergy are at a low level, resulting in a final risk index much less than 1. When the system experiences substantial instability, the characteristic components of multiple dimensions will produce a nonlinear synergistic effect, driving the maximum eigenvalue to rapidly cross orders of magnitude. At this time, if the physical correlation factor is close to 1, the final risk index will exceed 1. If the risk threshold is set too high, such as 2, it may lead to a slow perception of mild to moderate transient disturbances, resulting in missed reports and delaying the handling of accidents. If the risk threshold is set too low, such as 0.5, it may lead to an overly sensitive system to normal capacitor switching or regular load fluctuations, potentially increasing the system's verification burden even with the interception of the physical correlation factor.

[0065] For example, Figure 5 This is a schematic diagram illustrating the real-time determination of the final risk index. The horizontal axis represents the window number, and the vertical axis represents the index value. The diagram includes the final risk index and a preset risk threshold. When the final risk index is greater than or equal to the risk threshold, the power grid is determined to be in a transient instability state, and the system triggers a risk warning command.

[0066] This completes the early warning and tracing of power safety accidents.

[0067] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A power safety accident early warning and tracing method based on multi-dimensional feature analysis, characterized in that, include: The electrical signal fluctuations on the primary side of the high-voltage bus are collected and converted into small voltage and current signals on the secondary side. High-speed sampling of small voltage and current signals yields active power, voltage, and frequency sequences with a unified time label. A time-sequential window is set up. Based on the instantaneous change and local fluctuation of the active power sequence, the power mutation score of the high-voltage bus within the window is obtained; based on the static deviation and dynamic rate of change of the voltage sequence, the voltage drop score of the high-voltage bus within the window is obtained; based on the real-time offset and cumulative deviation of the frequency sequence, the frequency instability of the high-voltage bus within the window is obtained. By combining power mutation, voltage sag, and frequency instability, a multi-index coordination matrix is ​​constructed. This includes: calculating the mean of the active power sequence and voltage sequence within the window; multiplying the power deviation of the active power from the power mean at all sampling times with the voltage deviation of the voltage mean at the corresponding sampling times and calculating the arithmetic mean to obtain the original covariance of the active power sequence and voltage sequence; dividing the original covariance by the product of the preset reference power and reference voltage to obtain the dimensionless covariance of the active power sequence and voltage sequence; using the power mutation, voltage sag, and frequency instability as the power dimension, voltage dimension, and frequency dimension components on the main diagonal of the multi-index coordination matrix, respectively; filling the dimensionless covariance into the off-diagonal positions corresponding to the power dimension and voltage dimension in the matrix, and padding the remaining positions with zeros to obtain a 3×3 symmetric multi-index coordination matrix. The eigenvalues ​​of the multi-index collaborative matrix are decomposed to extract the largest eigenvalue and the corresponding eigenvector. Based on the difference between the change ratios of the voltage sequence and the active power sequence and the preset nominal system correlation value, the physical correlation factor is obtained. This includes: calculating the ratio of the normalized difference between the voltage at the current sampling time and the previous sampling time to the normalized difference between the active power at the current sampling time and the previous sampling time to obtain the voltage-power change rate; and calculating the reciprocal of the sum of 1 and the absolute values ​​of the differences between the voltage-power change rate and the preset dimensionless nominal system correlation value to obtain the physical correlation factor. The final risk index is obtained by combining the physical correlation factor with the maximum eigenvalue. The state of power safety accidents is determined based on the relationship between the final risk index and the preset risk threshold.

2. The method for early warning and tracing of power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The process of collecting electrical signal fluctuations on the primary side of the high-voltage bus and converting them into small voltage and current signals on the secondary side includes: Voltage transformers, current transformers, and synchronous phasor measurement terminals are deployed on the primary side of the high-voltage busbar and in the feeder circuits of key load nodes in the substation. The voltage transformers and current transformers collect the electrical signal fluctuations on the primary side of the high-voltage busbar and convert the electrical signal fluctuations into small voltage and current signals on the secondary side using the transformation ratio of the transformer windings.

3. The method for early warning and tracing power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The high-speed sampling of small voltage and current signals yields an active power sequence, a voltage sequence, and a frequency sequence with a unified time label, including: The synchronous phasor measurement terminal is connected to the secondary circuit of the current transformer. The small voltage and current signals received from the secondary side are sampled at high speed to obtain the high-speed sampled digital signal. Then, the digital signal is reversed and restored using the preset current transformer ratio coefficient. Through the instantaneous power algorithm and the zero-crossing detection algorithm, the active power sequence, voltage sequence and frequency sequence with a unified time label are obtained.

4. The method for early warning and tracing power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The power mutation satisfies the following relationship: ； In the formula, Indicates the high-voltage busbar at the 1st Power mutations within the sampling window; Represents the first active power sequence. Active power at the sampling time; Represents the first active power sequence. Active power at the sampling time; Indicates the active power sequence in the th order. The arithmetic mean within the window of sampling time; Indicates the active power sequence in the th order. The standard deviation within the sampling window; Represents the absolute value symbol.

5. The method for early warning and tracing power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The voltage drop satisfies the following relationship: ； In the formula, Indicates the high-voltage busbar at the 1st Voltage drop within the sampling window; Indicates the system's rated voltage; Represents the voltage sequence of the first... Voltage at the sampling time; This represents the derivative of voltage with respect to time. This indicates the preset voltage change rate reference; Represents the absolute value symbol.

6. The method for early warning and tracing of power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The frequency instability satisfies the following relationship: ； In the formula, Indicates the high-voltage busbar at the 1st Frequency instability within the sampling window; Represents the frequency sequence of the th Frequency value at the sampling time; Indicates the rated frequency; Indicates the preset frequency offset reference; Indicates the first Frequency value at the sampling time; Indicates the number of sampling times within the window; Represents the absolute value symbol.

7. The method for early warning and tracing power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The acquisition of the maximum eigenvalue and the corresponding eigenvector includes: The multi-index collaborative matrix is ​​subjected to eigenvalue decomposition to extract the largest eigenvalue and the corresponding eigenvector.

8. The method for early warning and tracing of power safety accidents based on multidimensional feature analysis according to claim 1, characterized in that, The determination of the state of a power safety accident includes: When the final risk index is greater than or equal to the risk threshold, the current power grid is determined to be in a transient unstable state. The system triggers a risk warning command and sends it to the substation monitoring backend and the remote dispatch automation system. Through the communication interface, the relay protection device or load shedding controller is linked to execute the corresponding protection action. The maximum value of the absolute value of the three components of the feature vector corresponding to the multi-index collaborative matrix is ​​extracted. By identifying the component with the highest weight in the feature vector, the source of the fault can be located. When the final risk index is less than the risk threshold, the current power grid is determined to be in a safe operating state or a state of disturbance fluctuation.

Images