A Smart Method for Power Security Situation Assessment Based on Multi-Source Sensor Networks

Abstract

This application relates to an intelligent method for power safety situation assessment based on a multi-source sensor network. It acquires multi-dimensional data from the power system through a multi-source sensor network to construct a raw dataset containing time-series features. The raw dataset is preprocessed and denoised to extract the dynamic change features of key parameters, resulting in a cleaned parameter change sequence. Based on the cleaned parameter change sequence, the interaction relationships between key parameters are analyzed to obtain spectral feature data. According to the spectral feature data and combined with a theoretical model of resonance phenomena, the existence of resonance risk in the system is determined, and a risk assessment report is generated. Based on the risk assessment report, unstable factors are identified and high-risk parameter combinations are extracted to generate unstable factor distribution feature data. Based on system stability assessment indicators, it is determined whether the power grid operating state is close to the instability boundary, and risk warning signals and parameter adjustment suggestions are generated.

Description

Technical Field

[0001] This application relates to the field of electrical variable technology, specifically to an intelligent method for power security situation assessment based on multi-source sensor networks. Background Technology

[0002] In the field of power system operation and safety management, traditional monitoring and assessment methods are often limited to a single data source and static analysis, making it difficult to fully capture the dynamic changes in system operation. Especially when faced with complex resonance phenomena, existing methods are insufficient in prediction and early warning capabilities, which can easily lead to the neglect of potential risks.

[0003] A core challenge in this field is how to accurately identify key factors that may cause instability in a system through multi-dimensional data collection and analysis. The primary problem is that the dynamic changes of certain key parameters in the system are difficult to capture accurately, such as the interaction between capacitors and inductors. This interaction directly affects the stability of the system. If the changing trends of these parameters cannot be grasped in time, it is difficult to further determine whether resonance will occur in the system. Once resonance occurs, it may cause abnormal fluctuations in voltage or current, thereby threatening the safe operation of the entire power grid. This logical chain from parameter monitoring to phenomenon prediction constitutes the current difficulty in achieving technological breakthroughs. Summary of the Invention

[0004] To address the problems existing in the prior art, this application aims to provide an intelligent method for power security situation assessment based on multi-source sensor networks.

[0005] The intelligent method for power security situation assessment based on multi-source sensor networks described in this application includes:

[0006] Step S101: By deploying a multi-source sensor network, multi-dimensional data is collected in real time from the power system, covering monitoring points of key parameters such as voltage, current, capacitance, and inductance values, and constructing a raw data set containing time series characteristics for subsequent dynamic change analysis.

[0007] Step S102: For the original data set obtained in step S101, signal processing technology is applied for preprocessing, filtering method is used to remove noise interference, and dynamic change characteristics of capacitance and inductance values ​​at different time points are extracted to obtain the cleaned parameter change sequence for subsequent time-frequency characteristic analysis.

[0008] Step S103: Based on the parameter change sequence obtained in step S102 after cleaning, the interaction between capacitance and inductance values ​​is analyzed using a time-frequency analysis model based on wavelet transform. The coupling characteristics of the two in the frequency domain are calculated, and spectral characteristic data reflecting the interaction strength are obtained for subsequent resonance risk assessment.

[0009] Step S104: Based on the spectral feature data obtained in step S103, and combined with the theoretical model of the resonance phenomenon, determine whether there are potential conditions for resonance to occur in the system. If the spectral feature data exceeds the preset resonance threshold based on historical data and theoretical calculation, it is marked as a high-risk state, and a resonance risk assessment report is generated for subsequent identification of unstable factors.

[0010] Step S105: Based on the resonance risk assessment report generated in step S104, the support vector machine classification algorithm is used to classify and identify the key factors that may cause instability in the system, extract the high-risk parameter combinations related to the resonance phenomenon, and generate the distribution characteristic data of instability factors for subsequent system simulation analysis.

[0011] Step S106: Based on the distribution characteristic data of unstable factors generated in step S105, construct a dynamic simulation model based on the Monte Carlo method to simulate the unstable evolution process under different parameter combinations in power grid operation, obtain voltage and current fluctuation data in the simulation scenario, and calculate system stability assessment index for subsequent risk warning judgment.

[0012] Step S107: Based on the system stability assessment index calculated in step S106 and combined with historical operating data, determine whether the power grid operation is on the unstable boundary. If the assessment index is lower than the safety threshold preset based on industry standards and historical experience, generate a risk warning signal and provide parameter adjustment suggestion data.

[0013] Preferably, in step S101, acquiring multi-dimensional data from the power system through a multi-source sensor network and constructing a raw data set containing time-series features includes:

[0014] Real-time collection of multi-dimensional data, including voltage, current, capacitance, and inductance values, from within the power system forms an initial dataset;

[0015] The initial dataset is segmented according to a time window to generate multiple subset datasets;

[0016] Anomaly detection is performed on the subset of data using a preset threshold to determine the location of abnormal data points;

[0017] If an outlier is detected, the data in the time window containing the outlier is smoothed by using a mean filter to correct the data and obtain smoothed time series data.

[0018] Using the smoothed time series data, feature vectors are constructed to obtain a quantitative representation of the changing trends of key parameters;

[0019] Based on the quantitative representation of the changing trend, a classification algorithm is applied to determine the potential system failure mode, and the classification result is obtained.

[0020] If the classification results show that a fault mode exists, then extract the key features of the data within the relevant time window to determine the scope of the fault's impact.

[0021] Preferably, in step S102, the preprocessing and denoising of the original dataset, extracting the dynamic change features of key parameters, and obtaining the cleaned parameter change sequence includes:

[0022] The original dataset is standardized using signal processing techniques to generate a pre-processed dataset.

[0023] The initially processed dataset is denoised by filtering to eliminate noise interference, resulting in a denoised dataset.

[0024] For the denoised dataset, feature information of capacitance and inductance values ​​is extracted, and dynamic change trends are analyzed in conjunction with time points to generate a change sequence after feature extraction.

[0025] The parameter sequence after feature extraction is decomposed using time-frequency analysis to obtain the distribution characteristics of different frequency components;

[0026] If there are abnormal frequency components in the distribution features, abnormal data are filtered out by a preset threshold to obtain the filtered frequency distribution data.

[0027] Based on the filtered frequency distribution data, a time-frequency characteristic model is constructed to generate a cleaned parameter change sequence.

[0028] Preferably, in step S103, the step of analyzing the interaction relationship between key parameters and obtaining spectral feature data based on the parameter change sequence after cleaning includes:

[0029] Parameter change data is obtained from the cleaned parameter change sequence, and a standardized parameter change set is generated using a standardization processing method.

[0030] For the standardized set of parameter variations, the time-frequency decomposition method is used to extract the frequency components of the capacitance and inductance values ​​at different time scales to determine the time-frequency distribution characteristics.

[0031] Based on the time-frequency distribution characteristics, the frequency domain coupling relationship is calculated to obtain the correlation strength data between the two in each frequency band;

[0032] For the aforementioned correlation intensity data, intensity features are extracted and quantized to generate a spectral data set;

[0033] If the intensity characteristics of certain frequency bands in the spectrum data set exceed a preset threshold, then potentially abnormal frequency band data are marked.

[0034] Based on the data of the potential abnormal frequency bands, the correlation with resonance risk is analyzed, and the risk level distribution is determined.

[0035] Preferably, in step S104, determining whether there is a resonance risk in the system based on the spectral characteristic data and the theoretical model of resonance phenomena, and generating a risk assessment report, includes:

[0036] The spectral feature data is cleaned and normalized to generate a processed spectral feature dataset.

[0037] Based on the processed spectral feature dataset, and combined with the pre-established theoretical model of the resonance phenomenon, we analyze whether there are any potential conditional features related to resonance.

[0038] For the aforementioned potential condition characteristics, the deviation between the value and the preset resonance threshold is calculated. If the deviation exceeds the preset range, it is marked as a data out-of-range state.

[0039] Based on the data exceeding the state, identify potential unstable factors in the system, and use a classification algorithm to generate a set of classified unstable factors.

[0040] If at least one factor in the set of unstable factors meets the high-risk criteria, the system state is marked as a high-risk state.

[0041] Based on the high-risk status markers, a comprehensive risk assessment level and risk warning information are generated.

[0042] Preferably, in step S105, identifying unstable factors and extracting high-risk parameter combinations based on the risk assessment report to generate unstable factor distribution characteristic data includes:

[0043] Initial data is obtained from the risk assessment report, and a classification algorithm is used to preliminarily classify the key factors that may cause instability, generating a set of classified factors.

[0044] Based on the classified set of factors, high-risk parameter combinations associated with the resonance phenomenon are extracted, and feature mapping is performed to generate a preliminary distribution feature dataset.

[0045] The preliminary distribution feature dataset is cleaned and standardized. If outliers exceed a preset threshold, smoothing is performed to generate the final distribution feature data.

[0046] For the final distribution feature data, a correlation analysis is performed. If the correlation index is lower than a preset threshold, irrelevant parameters are removed to obtain a simplified parameter feature set.

[0047] Based on the simplified parameter feature set, simulation input data is generated, multi-dimensional analysis is performed, and key parameter combinations are extracted.

[0048] Preferably, in step S106, the step of constructing a dynamic simulation model using the distribution characteristic data of the unstable factors to simulate fluctuations in power grid operation and calculate system stability evaluation indicators includes:

[0049] Based on the distribution characteristics of the unstable factors, a dynamic simulation model is constructed using a stochastic simulation method to generate initial simulation scenario data and obtain the distribution information of the unstable evolution trajectory.

[0050] Based on the distribution information, voltage fluctuation information and current fluctuation information are separated, and independent feature sets of the two types of fluctuation information are determined;

[0051] The independent feature sets are integrated. If the data integrity is lower than a preset threshold, interpolation is performed to generate an integrated fluctuation feature dataset.

[0052] Based on the integrated fluctuation feature dataset, a multi-dimensional analysis framework is constructed to obtain feature weights;

[0053] If the feature weight distribution does not meet the preset balance condition, the weights are adjusted and the system stability evaluation index is recalculated.

[0054] Outliers in the evaluation metrics are filtered and optimized to generate an optimized evaluation metric dataset.

[0055] Preferably, in step S107, determining whether the power grid operating state is close to the instability boundary based on the system stability assessment index, and generating a risk warning signal and parameter adjustment suggestions, includes:

[0056] Real-time system stability data is obtained from the power grid operation database and standardized by combining it with historical data to generate a preliminary organized operation status dataset.

[0057] Based on the aforementioned operational status dataset, calculate the system stability evaluation index, filter out outliers, and determine the final evaluation index results.

[0058] If the final evaluation index result is lower than the preset safety threshold, it is determined that the power grid operation is close to the unstable boundary, triggering the risk warning mechanism and generating a risk warning signal;

[0059] In response to the aforementioned risk warning signals, key parameters leading to the unstable state are determined by comparing and analyzing historical data and current operating status information.

[0060] Based on the key parameters, predict the direction of parameter adjustment and generate parameter adjustment suggestion data;

[0061] The optimized control strategy is calculated based on the parameter adjustment suggestions and then transmitted to the system control unit for parameter updates.

[0062] The intelligent method for power safety situation assessment based on multi-source sensor networks described in this application has the advantage of real-time acquisition of key parameters such as voltage, current, capacitance, and inductance values ​​by deploying a multi-source sensor network. Signal processing techniques and wavelet transforms are applied to analyze the interaction between capacitance and inductance values ​​to determine whether potential conditions for resonance occur in the system. Furthermore, this invention employs a support vector machine algorithm to identify key factors that may cause instability, constructs a dynamic simulation model based on the Monte Carlo method to simulate the instability evolution process, and calculates system stability assessment indicators. When the assessment indicators are below a safety threshold, this invention generates a risk warning signal and provides parameter adjustment suggestions, achieving early warning of power system resonance risks and optimized stability control, effectively improving the safety and reliability of power grid operation. Attached Figure Description

[0063] Figure 1 This application describes a process for an intelligent method for power security situation assessment based on multi-source sensor networks. Figure 1 ;

[0064] Figure 2 This application describes a process for an intelligent method for power security situation assessment based on multi-source sensor networks. Figure 2 . Detailed Implementation

[0065] like Figures 1-2 As shown, the intelligent method for power security situation assessment based on multi-source sensor networks described in this application includes the following steps:

[0066] like Figures 1-2 As shown, in step S101, multi-source sensor networks are deployed to collect multi-dimensional data from the power system in real time, covering monitoring points of key parameters such as voltage, current, capacitance and inductance, and constructing a raw data set containing time series characteristics for subsequent dynamic change analysis.

[0067] Furthermore, in step S101, by deploying a multi-source sensor network, multi-dimensional data is collected in real time from the power system, covering monitoring points of key parameters such as voltage, current, capacitance and inductance, and constructing a raw data set containing time series features to obtain a basic dataset for subsequent analysis.

[0068] Based on the constructed original dataset, the data is segmented according to the time series characteristics, and the data is divided into multiple subsets according to the preset time window, and the time boundaries of each subset are determined.

[0069] An outlier detection is performed on each subset of data using a preset threshold method to analyze whether data points exceed the normal range and to obtain the location information of outlier data points.

[0070] If an outlier is detected, the data within the time window containing the outlier is smoothed by using a mean filter to correct the original data and obtain smoothed time series data.

[0071] By analyzing the fluctuations of voltage and current parameters through smoothed time series data, feature vectors are constructed to obtain a quantitative representation of the dynamic change trend.

[0072] Based on the quantitative representation of dynamic change trends, the support vector machine algorithm is applied to classify the data, determine whether there are potential system failure modes, and obtain the classification results.

[0073] If the classification results show that a fault mode exists, then the multi-dimensional data within the relevant time window will be deeply mined to extract key parameter features associated with the fault mode and determine the scope of the fault's impact.

[0074] By using the characteristic data of the fault's impact range and combining it with the power system's operation logs, association rules are constructed, the dependencies between key parameters are analyzed, and a comprehensive evaluation result of the system's operating status is obtained.

[0075] Based on the comprehensive evaluation results, signal processing techniques were applied to analyze the time-frequency characteristics of the dynamic changes in capacitance and inductance values, resulting in a sequence of parameter changes after cleaning.

[0076] Specifically, in step S101, a multi-source sensor network is deployed to collect multi-dimensional data of voltage, current, capacitance and inductance values ​​from the power system in real time, with a sampling frequency of 1kHz, to construct a raw data set containing time series features and form a basic dataset.

[0077] Based on the original dataset, the data is divided into multiple subsets according to time windows, each with a length of 10 seconds, and the time boundaries of each subset are determined.

[0078] An outlier detection was performed on each subset of data using a preset threshold method. The normal range for voltage was set to 220V±5% and the normal range for current to be 10A±2%. The analysis was conducted to determine whether the data points exceeded the normal range and to obtain the location information of the outlier data points.

[0079] If an outlier is detected, the data within the time window containing the outlier is smoothed. The mean filtering method is used to correct the original data. The window size is 5 data points, resulting in smoothed time series data.

[0080] By analyzing the fluctuations of voltage and current parameters through smoothed time series data, feature vectors are constructed, and the mean and standard deviation are extracted as quantitative indicators to obtain a quantitative representation of the dynamic trend.

[0081] Based on the quantitative representation of dynamic change trends, the support vector machine algorithm is applied to classify the data. The radial basis function kernel function is used to determine whether there are potential system failure modes and obtain the classification results.

[0082] If the classification results show that there is a fault mode, then the multi-dimensional data within the relevant time window will be deeply mined to extract the abrupt change features of capacitance and inductance values ​​and determine the scope of the fault impact.

[0083] By using the characteristic data of the fault's impact range and combining it with the power system's operation logs, association rules are constructed, the correlation coefficients between voltage and current are analyzed, and a comprehensive evaluation result of the system's operating status is obtained.

[0084] Based on the comprehensive evaluation results, and considering the dynamic changes in capacitance and inductance values, a fast Fourier transform was applied to analyze the time-frequency characteristics, resulting in a sequence of parameter changes after cleaning.

[0085] In one embodiment, the specific implementation details of step S101 include:

[0086] In the deployment of multi-source sensor networks, sensor types and configurations are as follows: Multi-source sensor networks are deployed at key nodes of the power system, including substations, distribution cabinets, and transmission lines, including voltage sensors (accuracy ±0.2%), current sensors (accuracy ±0.5%), capacitance measurement modules (based on the LCR bridge principle), and inductance measurement modules (based on the resonant frequency method).

[0087] Sampling frequency setting: Set the sampling frequency to 1kHz to meet the requirements for capturing rapid dynamic changes (harmonics, transient processes) in the power system;

[0088] Data synchronization mechanism: GPS clock synchronization technology is used to ensure the consistency of timestamps of multi-source sensor data, with an error of ≤1ms;

[0089] In data segmentation and anomaly detection, time window division: the original data is segmented into 10-second time windows, with each subset containing 10,000 data points (1kHz × 10 seconds). Anomaly detection: a dynamic threshold is set based on statistical methods.

[0090] Normal voltage range: 220V±5% (i.e., 209V~231V);

[0091] Normal current range: 10A±2% (i.e., 9.8A~10.2A);

[0092] If a data point exceeds the threshold, it is marked as an anomaly, and its time window position is recorded;

[0093] In the mean filtering process for data smoothing and feature extraction, a moving mean filter with a window size of 5 data points is used for outlier window data. The formula is as follows: ,

[0094] Character meanings: xi: raw data point (instantaneous measurement of voltage, current, capacitance or inductance), xsmoothing: data point after smoothing, k: index of the current data point, i: local index of the data point within the sliding window (range: k−2 to k+2).

[0095] Feature vector construction involves extracting the mean, standard deviation, and peak value of voltage and current within each time window as quantitative indicators of dynamic changes to form a feature vector.

[0096] ,

[0097] Character meanings: μV: mean voltage (reflects the average voltage level), σV: standard deviation of voltage (reflects the degree of voltage fluctuation), Vpp: peak voltage (the difference between the maximum and minimum values, reflecting the voltage dynamic range), μI: mean current, σI: standard deviation of current, Ipp: peak current;

[0098] In the parameter settings of the Support Vector Machine (SVM) for fault mode classification, the kernel function is radial basis function (RBF), the parameter γ=0.1, the penalty coefficient C=1.0, the classification logic is: the training set contains historical normal data and fault data (short circuit, resonance), and the output classification result is "normal", "capacitance abnormal" or "inductance abnormal".

[0099] like Figures 1-2 As shown, in step S102, the original data set obtained in step S101 is preprocessed using signal processing techniques. A filtering method is used to remove noise interference, and the dynamic change characteristics of capacitance and inductance values ​​at different time points are extracted to obtain the cleaned parameter change sequence for subsequent time-frequency characteristic analysis.

[0100] Furthermore, in step S102, for the original data set obtained from the initial stage, signal processing technology is used for preliminary preprocessing, and noise interference is removed from the data by filtering method to obtain a pre-cleaned data sequence.

[0101] Based on the data sequence after preliminary cleaning, the dynamic change characteristics of capacitance and inductance values ​​at different time points are extracted. Time series analysis is used to record parameter fluctuations and obtain a set of dynamic change characteristics.

[0102] By using a dynamically changing feature set, standardization processing techniques are applied to normalize and adjust the fluctuation data of capacitance and inductance values, thereby obtaining a normalized parameter change sequence.

[0103] Based on the normalized parameter change sequence, wavelet transform is used to decompose the data into time and frequency components at different time scales to determine the time and frequency distribution characteristics.

[0104] By analyzing the distribution patterns of capacitance and inductance values ​​in each frequency band using time-frequency distribution characteristic data, and using spectrum analysis tools to quantize the frequency components, a set of spectrum features is obtained.

[0105] If the feature values ​​of certain frequency bands in the spectral feature set exceed the preset threshold range, then the frequency bands are marked as abnormal to obtain potential abnormal frequency band data.

[0106] Based on the data of potential abnormal frequency bands, and combined with the pre-constructed risk assessment model, a correlation analysis is performed on the abnormal frequency bands to determine the distribution of resonance risk levels.

[0107] By analyzing the resonance risk level distribution, a corresponding spectral feature mapping relationship is generated. Data visualization technology is used to structure the mapping relationship to obtain risk assessment basis data.

[0108] Based on the risk assessment data and combined with the time-frequency characteristic description model, the parameter change sequence after cleaning is verified to determine whether the final data characteristics meet the predetermined standards.

[0109] Specifically, in step S102, a Butterworth low-pass filter is used for signal preprocessing of the original dataset, with a cutoff frequency of 100Hz to remove high-frequency noise interference, resulting in a preliminary cleaned data sequence with a signal-to-noise ratio improved by 20dB.

[0110] Based on the cleaned data sequence, the dynamic change characteristics of capacitance and inductance values ​​were extracted using the sliding window method. The window width was set to 50ms, the sampling interval was 10ms, and the parameter fluctuation curves were recorded to form a dynamic feature set containing 1000 time points.

[0111] By using a dynamic feature set, the Z-score normalization algorithm is applied to normalize the capacitance and inductance fluctuation data, adjusting the mean to 0 and the standard deviation to 1, thereby generating a normalized parameter sequence.

[0112] Based on the normalized sequence, the Daubechies wavelet basis function is used to perform a 5-level time-frequency decomposition, extracting frequency components in three frequency bands: 0-50Hz, 50-100Hz, and 100-200Hz. The energy proportion of each scale is calculated to form a time-frequency distribution matrix.

[0113] Using the time-frequency matrix, the fast Fourier transform is used to analyze the amplitude correlation between capacitors and inductors in the 60Hz and 120Hz frequency bands, calculate the frequency bands with Pearson correlation coefficients greater than 0.8, and output the set of spectral features.

[0114] If the amplitude of the 120Hz band in the spectrum set exceeds the threshold of 0.5V, it is marked as an abnormal band, and the segment whose duration exceeds 200ms is recorded as potential abnormal data.

[0115] Based on the abnormal data, a pre-trained random forest risk assessment model is invoked, with three features input: frequency band amplitude, duration, and correlation coefficient, and the frequency band distribution with a high risk probability greater than 80% is output.

[0116] Based on the risk distribution, a two-dimensional mapping table of frequency and risk level is constructed, and a visualization matrix is ​​generated using a heatmap algorithm. The horizontal axis represents the 50-200Hz frequency band, and the vertical axis represents risk levels 1-5.

[0117] Based on the visualization matrix and combined with the frequency domain stability index of the time-frequency characteristic model, when the standard deviation of fluctuation in the 60Hz frequency band is less than 0.1 and the risk level of the 120Hz frequency band is ≤3, the data is deemed to comply with the IEC 61000-4-30 standard.

[0118] In one embodiment, the specific implementation details of step S102 include:

[0119] In filtering and denoising, the Butterworth low-pass filter is designed as a 4th-order filter with a cutoff frequency of 100Hz, passband ripple ≤0.1dB, stopband attenuation ≥40dB, and signal-to-noise ratio (SNR) improvement: through filtering, the SNR is improved by 20dB, effectively suppressing high-frequency noise (switching noise, electromagnetic interference).

[0120] In dynamic feature extraction, the sliding window method is used with a window width of 50ms and a sampling interval of 10ms to extract the instantaneous rate of change of capacitance and inductance values.

[0121] ,

[0122] Symbol explanation: C: Capacitance (unit: Farad, F), L: Inductance (unit: Henry, H), Δt: Sampling time interval (unit: second, s). , Instantaneous rate of change of capacitance and inductance values ​​(reflecting the dynamic characteristics of the parameters);

[0123] In data standardization, Z-score standardization is used:

[0124] ,

[0125] Symbol explanation: μC: mean capacitance value, σC: standard deviation of capacitance value, μL: mean inductance value, σL: standard deviation of inductance value, C norm L norm Standardized capacitance and inductance values ​​(dimensionless).

[0126] In wavelet time-frequency decomposition, the wavelet basis function is selected as Daubechies wavelet (db4), with 5 decomposition levels, covering the frequency band from 0.1Hz to 10kHz. Energy proportion calculation: The energy proportion of each frequency band is calculated using wavelet coefficients, using the following formula:

[0127] ,

[0128] Symbol explanation: Wk(i): coefficients after wavelet decomposition at level k (reflecting the energy distribution of the signal in a specific frequency band), Ek: total energy of wavelet coefficients at level k (used to quantify the importance of the frequency band), N: number of wavelet coefficients.

[0129] like Figures 1-2 As shown, in step S103, based on the parameter change sequence after cleaning obtained in step S102, the interaction relationship between capacitance and inductance is analyzed using a time-frequency analysis model based on wavelet transform. The coupling characteristics of the two in the frequency domain are calculated, and spectral characteristic data reflecting the interaction strength are obtained for subsequent resonance risk assessment.

[0130] Further, in step S103, based on the parameter change sequence after cleaning, a time-frequency analysis model based on wavelet transform is used to decompose the time series data of capacitance and inductance values ​​to obtain the frequency component distribution of the two at different time scales.

[0131] By analyzing the frequency component distribution, the dynamic change characteristics of capacitance and inductance values ​​at various time scales are determined, and their distribution patterns in the time-frequency domain are identified.

[0132] Based on the distribution pattern in the time and frequency domain, the coupling characteristics of capacitance and inductance values ​​in the frequency domain are calculated to obtain spectral characteristic data reflecting the strength of the interaction.

[0133] By using spectral feature data and employing spectral analysis tools, the coupling characteristics are quantified to obtain the correlation strength values ​​for each frequency band.

[0134] If the correlation strength value exceeds the preset threshold in certain frequency bands, the frequency band will be marked as abnormal to identify potentially high-risk frequency band data.

[0135] Based on high-risk frequency band data and a pre-established resonance risk assessment model, the correlation between spectral characteristics and resonance phenomena is analyzed to obtain the risk level distribution.

[0136] By analyzing the risk level distribution, a corresponding spectral feature mapping relationship is generated to determine the key frequency band characteristics related to resonance risk.

[0137] Based on the characteristics of key frequency bands and combined with dynamic simulation data of system operating parameters, the frequency band characteristics are verified in multiple dimensions to obtain the verified feature distribution results.

[0138] Based on the verified feature distribution results, the parameter combinations directly related to the resonance risk are extracted to generate the final risk assessment basis data.

[0139] Specifically, in step S103, based on the parameter change sequence after cleaning, the time-series data of capacitance and inductance values ​​are decomposed using a time-frequency analysis model based on wavelet transform, and the data are decomposed into multiple scales using the Daubechies wavelet basis function to obtain the frequency component distribution of the two values ​​in the range of 0.1Hz to 10kHz.

[0140] By analyzing the frequency component distribution, the dynamic change characteristics of capacitance and inductance values ​​at various time scales are analyzed. The energy distribution in the time-frequency domain is calculated using short-time Fourier transform (STFT) to determine the distribution law of the two values ​​in the time-frequency domain.

[0141] Based on the distribution pattern in the time and frequency domain, the coupling characteristics of capacitance and inductance in the frequency domain are calculated. The cross spectral density function is used to analyze the interaction strength between the two in the 1kHz to 5kHz frequency band, and spectral characteristic data reflecting the interaction strength are obtained.

[0142] Using spectral feature data, the coupling characteristics are quantized using Fast Fourier Transform (FFT) to obtain the correlation strength values ​​for each frequency band. For example, the correlation strength in the 2kHz band is 0.85.

[0143] If the correlation strength value exceeds a preset threshold (such as 0.8) in certain frequency bands, then the frequency band is marked as abnormal to identify potentially high-risk frequency band data.

[0144] Based on high-risk frequency band data and a pre-established resonance risk assessment model, the support vector machine (SVM) algorithm is used to analyze the correlation between spectral characteristics and resonance phenomena to obtain the risk level distribution. For example, the risk level in the 3kHz frequency band is high.

[0145] By analyzing the risk level distribution, a corresponding spectral feature mapping relationship is generated to determine the key frequency band characteristics related to resonance risk. For example, the feature value in the 4kHz band is 0.92.

[0146] Based on the characteristics of key frequency bands and combined with the dynamic simulation data of system operating parameters, the frequency band characteristics were verified in multiple dimensions using MATLAB simulation tools to obtain the verified feature distribution results. For example, the verification result in the 5kHz frequency band was 0.89.

[0147] Based on the verified feature distribution results, parameter combinations directly related to resonance risk are extracted to generate the final risk assessment basis data. For example, the parameter combination in the 6kHz frequency band is a capacitance value of 10μF and an inductance value of 50mH.

[0148] In one embodiment, the specific implementation details of step S103 include:

[0149] In cross-spectral density calculation, the formula for the cross-spectral density (CSD) of capacitance and inductance values ​​is: ,

[0150] Symbol explanation: R CL (τ): Cross-correlation function of capacitance C and inductance L (reflecting their time-domain correlation), F: Fourier transform (converting a time-domain signal to a frequency-domain signal), S CL (f): Cross-spectral density of capacitor and inductor at frequency f (quantization of frequency domain coupling strength).

[0151] In correlation strength quantization, the Pearson correlation coefficient is calculated in the 1kHz~5kHz frequency band: ,

[0152] Symbol explanation: Cov(C(f),L(f)): Covariance of capacitance and inductance at frequency f (reflecting linear correlation), σC, σL: Standard deviation of capacitance and inductance values, ρ: Correlation coefficient (range: -1~1, the larger the absolute value, the stronger the correlation).

[0153] In the high-risk frequency band marking, a preset threshold is set: frequency bands with a correlation strength ≥ 0.8 or an amplitude ≥ 0.5V are marked as abnormal. For example, the correlation strength of the 2kHz frequency band is 0.85, so it is marked as high risk.

[0154] like Figures 1-2 As shown, in step S104, based on the spectral feature data obtained in step S103, and combined with the theoretical model of the resonance phenomenon, it is determined whether there are potential conditions for resonance to occur in the system. If the spectral feature data exceeds the preset resonance threshold based on historical data and theoretical calculation, it is marked as a high-risk state, and a resonance risk assessment report is generated for subsequent identification of unstable factors.

[0155] Further, in step S104, spectral feature data is extracted from the system, and the data is cleaned and normalized using a standardization processing method to obtain a standardized spectral feature dataset.

[0156] Based on the standardized spectral feature dataset and combined with the pre-constructed theoretical model of resonance phenomenon, we analyze whether there are potential condition features related to resonance in the data and determine the distribution of potential conditions.

[0157] Based on the identified potential condition characteristics and combined with the theoretical calculation results based on historical data, the deviation between the spectral feature dataset and the preset resonance threshold is calculated to obtain the deviation data set;

[0158] If the deviation value in the deviation data set exceeds the preset threshold range, it is marked as a data out-of-state condition, and a preliminary abnormal state record is generated.

[0159] Based on the records of data exceeding the state, and combined with system analysis methods, potential unstable factors in the system are extracted to obtain a preliminary set of unstable factors;

[0160] For the initial set of unstable factors, the support vector machine algorithm is used to classify the factors and determine the set of unstable factors after classification.

[0161] If at least one factor in the classified set of unstable factors meets the criteria for a high-risk state, the system state is marked as a high-risk state, and a high-risk state identifier is generated.

[0162] Based on the high-risk status indicator and combined with the resonance risk assessment logic, a comprehensive analysis of the potential resonance phenomena in the system is conducted to obtain the final risk assessment level.

[0163] Based on the final risk assessment level, a structured resonance risk assessment report is generated, and instability factors and potential condition data are correlated to output a complete risk analysis dataset.

[0164] Specifically, in step S104, spectral feature data is extracted from the system, and the data is cleaned and normalized using a standardization method, such as limiting the frequency range to 0Hz to 100Hz, and using the Z-score method to convert the data into a normalized spectral feature dataset with a mean of 0 and a standard deviation of 1.

[0165] Based on the standardized spectral feature dataset, combined with a pre-built theoretical model of resonance phenomenon, such as an RLC circuit model, we analyze whether there are potential condition features related to resonance in the data, and determine the distribution of potential conditions, such as a peak frequency around 50Hz.

[0166] For the identified potential condition characteristics, combined with theoretical calculation results based on historical data, such as calculating the deviation between the spectral feature dataset and the preset resonance threshold (e.g., the amplitude threshold is 0.8), a deviation data set is obtained;

[0167] If the deviation value in the deviation data set exceeds the preset threshold range, for example, if the amplitude deviation at a certain frequency point reaches 1.2, it is marked as a data out-of-state condition, and a preliminary abnormal state record is generated.

[0168] Based on the records of data exceeding the state, and combined with system analysis methods such as principal component analysis (PCA), potential unstable factors in the system are extracted to obtain a preliminary set of unstable factors, such as frequency fluctuations and amplitude anomalies.

[0169] For the initial set of unstable factors, the support vector machine algorithm is used to classify the factors, for example, the radial basis function (RBF) kernel function is used for classification to determine the set of unstable factors after classification, such as high-frequency fluctuations and low-frequency resonances;

[0170] If at least one factor in the set of unstable factors after classification meets the criteria for a high-risk state, such as a risk score of more than 90 points for high-frequency fluctuations, then the system state is marked as a high-risk state and a high-risk state identifier is generated.

[0171] Based on the high-risk status indicator and combined with the resonance risk assessment logic, such as the weighted scoring method, the potential resonance phenomena in the system are comprehensively analyzed to obtain the final risk assessment level, such as a high-risk level of 5.

[0172] Based on the final risk assessment level, a structured resonance risk assessment report is generated, and instability factors and potential condition data are associated, such as associating high-frequency fluctuations with a 50Hz frequency point, to output a complete risk analysis dataset.

[0173] In one embodiment, the specific implementation details of step S104 include:

[0174] In the RCL series resonance model, the resonant frequency is calculated using the following formula:

[0175] ,

[0176] Symbol explanation: L: inductance (H), C: capacitance (F), fr: resonant frequency (Hz), i.e., the critical frequency at which the system may resonate;

[0177] In principal component analysis (PCA), the dimensionality of the spectral feature data is reduced, and the principal components with a contribution rate >85% are retained to extract key unstable factors (such as frequency shift and harmonic distortion).

[0178] In the risk level classification, support vector machine (SVM) is used for classification, and the risk level is divided into 5 levels: Level 1 (safe) to Level 5 (urgent).

[0179] like Figures 1-2 As shown, in step S105, based on the resonance risk assessment report generated in step S104, the support vector machine classification algorithm is used to classify and identify the key factors that may cause instability in the system, extract the high-risk parameter combinations related to the resonance phenomenon, and generate the distribution characteristic data of instability factors for subsequent system simulation analysis.

[0180] Further, in step S105, based on the resonance risk assessment report generated in step S104, the system operating parameters and risk index data recorded therein are extracted, and the support vector machine classification algorithm is used to classify and identify the key factors that may cause instability, so as to obtain a set of classified unstable factors.

[0181] By classifying the unstable factors and combining them with the theoretical model of the resonance phenomenon, we analyze the correlation between each factor and the resonance risk, extract the high-risk parameter combinations related to the resonance phenomenon, and determine the high-risk parameter combination dataset.

[0182] Based on the high-risk parameter combination dataset, the parameter characteristics are mapped to generate preliminary unstable factor distribution characteristic data, and a preliminary distribution characteristic dataset is obtained.

[0183] The initial distribution feature dataset is used for data cleaning and standardization to remove noisy data and outliers. If the proportion of outliers exceeds a preset threshold, the outliers are smoothed to obtain the cleaned distribution feature dataset.

[0184] Using the cleaned distribution feature dataset, correlation analysis is performed on high-risk parameter combinations. If the correlation index is lower than a preset threshold, irrelevant parameters are removed, and a simplified parameter feature set is determined.

[0185] Based on the simplified parameter feature set, and combined with the system simulation environment, the unstable factors are dynamically mapped to generate the mapped simulation input data, and the integrity and consistency of the data are judged.

[0186] Using the mapped simulation input data and combined with the pre-established simulation model, a multi-dimensional analysis of the distribution characteristics of unstable factors is conducted to obtain the characteristic distribution results after simulation analysis.

[0187] Based on the characteristic distribution results after simulation analysis, the key parameter combinations directly related to the resonance risk are extracted to generate the final unstable factor distribution characteristic data and identify potential risk points in system operation.

[0188] Based on the final distribution characteristics of unstable factors, combined with risk assessment logic and historical data analysis results, structured risk analysis data is generated to obtain input datasets for subsequent system optimization.

[0189] Specifically, in step S105, system operating parameters and risk index data are extracted from the resonance risk assessment report, including voltage fluctuation range (e.g., ±5%), frequency deviation (e.g., ±0.2Hz) and harmonic distortion rate (e.g., THD>3%). The key factors are classified using a support vector machine classification algorithm (kernel function selected as RBF, penalty coefficient C=1.0) to divide the set of unstable factors such as capacitor bank switching and nonlinear loads.

[0190] By combining resonance theory models (such as the RLC series resonance formula), the correlation coefficients between each factor and resonance risk are calculated (such as Pearson coefficient > 0.7), and high-risk parameter combinations are screened out (such as the combination of capacitance C > 100μF and inductance L < 10mH).

[0191] Perform feature mapping on the parameter combinations (such as principal component analysis (PCA) to reduce the dimensionality to 3 dimensions) to generate preliminary distribution feature data containing statistics such as mean and variance;

[0192] The data was cleaned using the Z-score standardization method. If the proportion of outliers exceeded 5%, the sliding window averaging method was used to smooth the data, resulting in a cleaned dataset with a standard deviation of less than 1.

[0193] The support between parameters is calculated through correlation analysis (such as the Apriori algorithm). If the support is less than 0.3, irrelevant parameters (such as temperature parameters) are removed, and a simplified feature set (such as voltage, frequency, and harmonic content) is retained.

[0194] Input the feature set into the simulation environment (such as MATLAB / Simulink), dynamically map it to simulation input data (such as time-domain waveforms), and check the data integrity (missing rate <1%) and consistency (correlation coefficient >0.9).

[0195] Multi-dimensional analysis (such as frequency domain FFT and time domain step response) is performed based on simulation models (such as state-space models) to output the distribution results of resonant peak frequency (such as 250Hz) and amplitude (such as 120dB).

[0196] Extract key parameter combinations (such as capacitance value C=150μF and harmonic order h=5), generate final distribution characteristic data (such as Gaussian distribution μ=230Hz, σ=15Hz), and mark potential risk points (such as resonant frequency close to the system's natural frequency).

[0197] By combining historical data (such as the results of the past 100 simulations) and risk assessment logic (such as a risk matrix), the system outputs structured data (such as JSON format), including risk level (such as high risk), parameter thresholds, and optimization suggestions.

[0198] In one embodiment, the specific implementation details of step S105 include:

[0199] In the parameter combination screening, a high-risk combination example is: capacitance value C > 100μF and inductance value L < 10mH, verified by Pearson correlation coefficient (> 0.7);

[0200] In data dimensionality reduction and mapping, principal component analysis (PCA) reduces features to 3 dimensions while retaining components with a variance percentage greater than 90%.

[0201] In the simulation input generation, a state-space model is constructed in MATLAB / Simulink, and the input parameters include voltage, frequency, and harmonic content (THD).

[0202] like Figures 1-2 As shown, in step S106, based on the distribution characteristic data of unstable factors generated in step S105, a dynamic simulation model based on the Monte Carlo method is constructed to simulate the unstable evolution process under different parameter combinations in power grid operation, obtain voltage and current fluctuation data in the simulation scenario, and calculate system stability evaluation index for subsequent risk warning judgment.

[0203] Further, in step S106, the distribution characteristic data of unstable factors are obtained from step S105, and a dynamic simulation model is constructed using the Monte Carlo method to simulate the unstable evolution process under different parameter combinations in power grid operation, thereby obtaining preliminary voltage and current fluctuation data in the simulation scenario.

[0204] Based on the obtained preliminary voltage and current fluctuation data, data standardization processing is performed, and calibration is carried out in combination with historical operating status information to determine the distribution characteristics of the calibrated fluctuation data.

[0205] By analyzing the distribution characteristics of the calibrated fluctuation data, the system stability assessment index is calculated. The index value is then quantified using a preset assessment algorithm to obtain the quantified stability assessment result.

[0206] If the quantified stability assessment result is lower than the preset safety threshold, the data deep analysis mechanism is triggered to obtain historical unstable event data related to the current simulation scenario and determine the potential risk level.

[0207] Based on the acquired historical instability event data and combined with the fluctuation data distribution characteristics of the current simulation scenario, cluster analysis was used to classify risk factors and determine the main instability impact parameters.

[0208] By identifying the main instability parameters, a parameter sensitivity analysis model is constructed to simulate the impact of different parameter adjustments on system stability, and preliminary impact data of parameter adjustments are obtained.

[0209] Based on the obtained parameters, the preliminary impact data is adjusted, and combined with real-time power grid operation status information, an optimization algorithm is used to calculate a control strategy suitable for the current scenario, and the optimized parameter adjustment scheme is determined.

[0210] Based on the determined optimized parameter adjustment scheme, a corresponding risk warning signal is generated and synchronously transmitted to the system control module to obtain the updated operation status data after feedback.

[0211] Based on the obtained operational status update data, the execution results of the adjustment plan are compared and analyzed with the historical database. Machine learning algorithms are used to update the distribution characteristics of unstable factors, resulting in improved risk prediction model data.

[0212] Specifically, in step S106, the distribution characteristic data of unstable factors are obtained from step S105, and a dynamic simulation model is constructed using the Monte Carlo method to simulate the unstable evolution process under different parameter combinations in the operation of the power grid. For example, under the condition that the voltage fluctuation range is ±5% and the current fluctuation range is ±3%, 1000 random simulation scenarios are generated to obtain preliminary voltage and current fluctuation data.

[0213] Based on the obtained preliminary voltage and current fluctuation data, data standardization processing is performed, and calibration is carried out in combination with historical operating status information. For example, the Z-score standardization method is used to convert the fluctuation data into a distribution with a mean of 0 and a standard deviation of 1, and the distribution characteristics of the calibrated fluctuation data are determined.

[0214] By analyzing the distribution characteristics of the calibrated fluctuation data, the system stability assessment index is calculated. The index value is then quantified using a preset assessment algorithm, such as the Lyapunov exponent method, to calculate the system stability and obtain the quantified stability assessment result.

[0215] If the quantified stability assessment result is lower than the preset safety threshold, such as the Lyapunov index being less than -0.5, the data deep analysis mechanism is triggered to obtain historical instability event data related to the current simulation scenario, such as 10 voltage collapse events that occurred in the past year, to determine the potential risk level.

[0216] Based on the acquired historical instability event data and combined with the fluctuation data distribution characteristics of the current simulation scenario, cluster analysis methods are used to classify risk factors. For example, the K-means algorithm is used to classify risk factors into three categories to determine the main instability impact parameters.

[0217] By identifying the main instability-affecting parameters, a parameter sensitivity analysis model is constructed to simulate the impact of different parameter adjustments on system stability. For example, adjusting the voltage fluctuation range to ±3% and the current fluctuation range to ±2% yields preliminary impact data on parameter adjustments.

[0218] Based on the obtained parameters, the preliminary impact data is adjusted, and combined with real-time power grid operation status information, an optimization algorithm is used to calculate a control strategy suitable for the current scenario. For example, a genetic algorithm is used to optimize the control parameters and determine the optimized parameter adjustment scheme.

[0219] Based on the determined optimized parameter adjustment scheme, a corresponding risk warning signal is generated and synchronously transmitted to the system control module to obtain the updated operation status data after feedback.

[0220] Based on the obtained operational status update data, the execution results of the adjustment plan are compared and analyzed with the historical database. Machine learning algorithms are used to update the distribution characteristics of unstable factors. For example, a new risk prediction model is trained using the random forest algorithm to obtain improved risk prediction model data.

[0221] In one embodiment, the specific implementation details of step S106 include:

[0222] In the random scene generation, Monte Carlo simulations were performed 1000 times, with parameter fluctuation ranges of: voltage ±5% and current ±3%.

[0223] In stability assessment metrics, the Lyapunov index is calculated using the following formula:

[0224] ,

[0225] Symbol explanation: δx(t): small initial deviation of the system state (e.g., small changes in voltage or current); λ: Lyapunov exponent (reflects the sensitivity of the system to initial conditions): λ<0: the system is stable; λ>0: the system is chaotic or unstable; the safety threshold is set to λ<-0.5.

[0226] In the parameter sensitivity analysis, a genetic algorithm is used to optimize the control parameters, with the objective function being to minimize voltage fluctuations.

[0227] like Figures 1-2 As shown, in step S107, based on the system stability assessment index calculated in step S106 and combined with historical operating data, it is determined whether the power grid operation is on the unstable boundary. If the assessment index is lower than the safety threshold preset based on industry standards and historical experience, a risk warning signal is generated, and parameter adjustment suggestion data is provided for subsequent system control optimization.

[0228] Further, in step S107, real-time system stability-related data are extracted from the power grid operation database, and combined with historical operation records for preliminary cleaning and standardization to obtain the sorted operation status dataset.

[0229] Based on the organized operational status dataset, a pre-built evaluation model is applied to perform calculations, outliers in the calculation process are filtered out, and preliminary results of the system stability evaluation indicators are determined.

[0230] If the preliminary results of the system stability assessment indicators are lower than the safety threshold preset based on industry standards and historical experience, the power grid operation is determined to be close to the unstable boundary, triggering the risk warning mechanism and generating an initial risk warning signal.

[0231] By using initial risk warning signals, historical operational data and current status information related to the unstable boundary are obtained, and comparative analysis is conducted to determine key parameters that may lead to unstable states.

[0232] Based on the identified key parameters, the support vector machine algorithm is used to predict the direction of parameter adjustment and generate preliminary parameter adjustment suggestion data.

[0233] Based on the preliminary parameter adjustment suggestions and combined with the real-time feedback information from the system control module, a dynamic simulation model based on the Monte Carlo method is constructed to simulate the unstable evolution process under different parameter combinations and obtain voltage and current fluctuation data in the simulated scenario.

[0234] By using voltage and current fluctuation data in simulated scenarios, the system stability assessment index is recalculated, and combined with historical operating data for comprehensive analysis, it is determined whether the power grid operation is on the unstable boundary.

[0235] If the recalculated system stability assessment index is lower than the preset safety threshold, a final risk warning signal will be generated, and parameter adjustment suggestion data will be updated for subsequent system control optimization reference.

[0236] The final risk warning signal and the updated parameter adjustment suggestions are transmitted to the system control unit for parameter updates, and the adjusted operating status information is recorded and stored in the historical database for subsequent analysis.

[0237] Specifically, in step S107, real-time voltage, frequency, and power data are extracted from the power grid operation database, combined with 10-minute sampling data from historical operation records, and processed using the Z-score standardization method to eliminate dimensional differences and form a standardized dataset.

[0238] Standardized data is input into a pre-trained LSTM evaluation model to calculate the system stability index. A safety threshold of 0.85 is set, and an early warning is triggered if the index falls below the threshold.

[0239] By retrieving similar abnormal data from the past 30 days using early warning signals, and employing Pearson correlation coefficient analysis, key parameters such as bus voltage deviation exceeding ±5% or frequency fluctuation greater than 0.2Hz were identified.

[0240] Based on key parameters, the kernel function is set to RBF using the SVM algorithm, with a penalty coefficient C=1.0. Suggestions for adjusting output parameters include increasing generator output by 3% or adjusting capacitor bank switching schemes.

[0241] By combining real-time SCADA feedback, a Monte Carlo model was built to simulate 1000 parameter combinations and record abnormal scenarios where the voltage drop was less than 0.9 pu or the current distortion rate exceeded 5%.

[0242] The transient stability margin index is recalculated using simulated data. When the index value remains below 0.82 for 2 consecutive seconds, a final warning is generated, and the optimization adjustment suggestion is to correct the generator droop coefficient to 0.05.

[0243] The adjustment command is sent to the PMU device via the IEC 61850 protocol, and the adjusted data is synchronously stored in the Oracle database and marked as an unstable boundary case.

[0244] In one embodiment, the specific implementation details of step S107 include:

[0245] In the generation of early warning signals, real-time data is compared with historical database (Oracle). If the Lyapunov index is less than -0.5, an early warning is triggered.

[0246] In the parameter adjustment recommendations: Example: Increase generator output by 3%, optimize capacitor bank switching scheme;

[0247] During the issuance of control commands, the adjustment commands are transmitted to the PMU (Synchronous Phasor Measurement Unit) via the IEC 61850 protocol.

[0248] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this application.

Claims

1. A smart method for power security situation assessment based on multi-source sensor networks, characterized in that, include: Multi-source sensor networks are used to collect multi-dimensional data on voltage, current, capacitance, and inductance values ​​in real time from the power system, and a raw data set containing time series characteristics is constructed. The original dataset is preprocessed and denoised. The sliding window method and Z-score normalization are used to extract the dynamic change features of capacitance and inductance values ​​to obtain the cleaned parameter change sequence. The parameter variation sequence is decomposed into five levels based on the Daubechies wavelet basis function, and the cross spectral density of capacitance and inductance in the 1kHz–5kHz frequency band is calculated to obtain spectral feature data. The risk level of the spectral feature data is classified by support vector machine, and the resonance phenomenon theoretical model is combined to determine whether there is resonance risk in the system and generate a risk assessment report. A dynamic simulation model is constructed based on the Monte Carlo method to simulate voltage and current fluctuations in power grid operation, and the Lyapunov index is used to calculate the system stability evaluation index. If the evaluation index is lower than the safety threshold, parameter adjustment suggestions are generated through a genetic algorithm and sent to the control unit via the IEC 61850 protocol.

2. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 1, characterized in that, The deployment of the multi-source sensor network includes: Voltage sensors, current sensors, capacitance measurement modules, and inductance measurement modules are deployed at key nodes of power system substations, distribution cabinets, and transmission lines. The voltage sensor has an accuracy of ±0.2%, the current sensor has an accuracy of ±0.5%, the capacitance measurement module is based on the LCR bridge principle, and the inductance measurement module is based on the resonant frequency method. The sampling frequency of the multi-source sensor network is set to 1kHz, and the consistency of data timestamps is ensured by GPS clock synchronization technology, with an error of ≤1ms. The original dataset is divided into multiple subsets by a 10-second time window, and each subset contains 10,000 data points; The dynamic thresholds for outlier detection are: 220V ± 5% for normal voltage and 10A ± 2% for normal current. For abnormal window data, a moving average filter with a window size of 5 data points is used for processing; The mean, standard deviation, and peak value of voltage and current within each time window are extracted to construct a feature vector; The data is classified using the support vector machine algorithm, with the radial basis function (RBF) as the kernel function, parameter γ=0.1, and penalty coefficient C=1.

0. The classification results include normal, abnormal capacitance, or abnormal inductance.

3. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 1, characterized in that, The construction and stability evaluation of the dynamic simulation model include: The Monte Carlo method was used to generate 1000 random simulation scenarios to simulate the operation of the power grid under voltage fluctuation range of ±5% and current fluctuation range of ±3%. The system stability evaluation index was calculated by the Lyapunov exponent method. When the Lyapunov index is less than -0.5, a deep analysis mechanism is triggered. Combining historical unstable event data, the risk factors are divided into 3 categories using the K-means clustering algorithm. A parameter sensitivity analysis model was constructed, and the control parameters were optimized using a genetic algorithm. The objective function was to minimize voltage fluctuations. The optimized parameter adjustment scheme was transmitted to the synchronous phasor measurement unit via the IEC61850 protocol, and the adjusted data was recorded in the historical database.

4. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 1, characterized in that, The preprocessing and noise reduction include: A fourth-order Butterworth low-pass filter was used to denoise the original dataset, with a cutoff frequency of 100Hz, passband ripple ≤0.1dB, and stopband attenuation ≥40dB. The instantaneous rate of change of capacitance and inductance values ​​was extracted using the sliding window method, with a window width of 50ms and a sampling interval of 10ms. The capacitance and inductance values ​​were normalized using the Z-score normalization method.

5. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 1, characterized in that, The acquisition of the spectral feature data includes: The Daubechies wavelet basis function db4 was used to perform a 5-level time-frequency decomposition on the parameter variation sequence, covering the frequency band from 0.1Hz to 10kHz; The interaction strength between capacitors and inductors in the 1kHz to 5kHz frequency band was analyzed using the cross-spectral density function. If the frequency band correlation strength is ≥0.8 or the amplitude is ≥0.5V, it is marked as an abnormal frequency band; The support vector machine algorithm is used to analyze the correlation between spectral characteristics and resonance risk. The risk level is divided into 5 levels, and a high-risk warning is generated when the risk level is ≥4.

6. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 1, characterized in that, The risk assessment report generation includes: Calculate the resonant frequency based on the RLC series resonance model; Principal component analysis (PCA) was used to reduce the dimensionality of the spectral feature data, retaining principal components with a variance contribution rate greater than 85%. When the high-frequency volatility risk score is greater than 90, the system is marked as high-risk and a comprehensive report including frequency offset and harmonic distortion rate is generated.

7. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 1, characterized in that, The generation of the parameter adjustment suggestions includes: The stability index of the real-time system was calculated using a Long Short-Term Memory (LSTM) network model, with a safety threshold set at 0.

85. When the bus voltage deviation exceeds ±5% or the frequency fluctuation is >0.2Hz, the generator output or capacitor bank switching scheme is optimized through a genetic algorithm. The adjusted parameters are sent to the control unit via the IEC 61850 protocol and marked as an unstable boundary case in the Oracle database.

8. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 2, characterized in that, The training set of the support vector machine algorithm includes historical normal data and fault data, and the fault types include short circuit and resonance. When the classification result is normal, capacitance abnormal, or inductance abnormal, deep data mining is triggered for the corresponding time window.

9. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 3, characterized in that, The number of iterations of the genetic algorithm is set to 1000, the crossover probability is 0.8, and the mutation probability is 0.01; the optimized parameter combination includes adjusting the voltage fluctuation range to ±3% and the current fluctuation range to ±2%.

10. The intelligent method for power security situation assessment based on multi-source sensor networks according to claim 5, characterized in that, When the duration of a high-risk frequency band is greater than 200ms, the risk probability is calculated. If the probability is greater than 80%, an emergency control command is generated and synchronized to the power grid dispatch center.

Images