Automated measurement and analysis method and system for reactor impedance characteristics

Abstract

The present invention provides a method and system for automatically measuring and analyzing the impedance characteristics of a reactor, which relates to the technical field of reactors. The method comprises collecting voltage and current signals, calculating original impedance frequency characteristic data, decomposing the data in the frequency domain, and processing the data using a neural network compensation model to establish an impedance frequency characteristic curve. The method also continuously collects real-time monitoring data for comparison with theoretical characteristics, constructs a fault feature vector, and establishes a fault discrimination criterion using Euclidean distance and Mahalanobis distance to achieve automatic detection and accurate positioning of reactor faults, thereby effectively improving the operational reliability of the power system.

Description

Technical Field

[0001] The present invention relates to reactor technology, and in particular to a method and system for automatically measuring and analyzing reactor impedance characteristics. Background Art

[0002] Reactors are essential components in power systems, widely used in power transmission, power quality control, and harmonic suppression. The impedance characteristics of reactors are a key performance indicator, directly impacting the stability and reliability of power systems. As power grids expand in size and complexity, real-time monitoring of reactor operating conditions, accurate measurement of impedance characteristics, and timely diagnosis of faults are becoming increasingly important. Traditionally, reactor impedance characteristic measurements rely primarily on offline testing methods, which typically require the equipment to be shut down for specialized testing. This method is not only time-consuming and labor-intensive, but also impacts the normal operation of power systems.

[0003] Traditional impedance measurement methods lack real-time and dynamic capabilities, failing to meet the demands of online power system monitoring. Existing technologies often rely on static measurement methods, which struggle to capture the impedance variations of reactors under varying operating conditions. This is especially true when grid parameters fluctuate significantly, leading to significant measurement errors that affect the accuracy of fault diagnosis.

[0004] Existing impedance characteristic analysis methods require high signal processing accuracy but lack effective compensation mechanisms. During actual measurement, errors introduced by the sampling device, environmental noise interference, and the influence of harmonic components can cause distortion in the impedance frequency characteristic curve. Existing technologies have difficulty effectively eliminating these influencing factors, thus affecting the accuracy of impedance characteristic analysis.

[0005] Traditional reactor fault diagnosis methods rely primarily on empirical judgment or simple threshold comparisons, lacking systematic fault feature extraction and intelligent fault identification mechanisms. This approach struggles to accurately identify complex fault patterns, particularly early-stage faults and multiple faults. This results in inaccurate diagnostic results, hindering the scientific and timely nature of equipment maintenance decisions. Summary of the Invention

[0006] The embodiments of the present invention provide a method and system for automatically measuring and analyzing the impedance characteristics of a reactor, which can solve the problems in the prior art.

[0007] A first aspect of an embodiment of the present invention provides a method for automatically measuring and analyzing reactor impedance characteristics, comprising:

[0008] The sampling device of the reactor collects voltage signals and current signals, and calculates original impedance frequency characteristic data of the reactor according to the voltage signals and the current signals;

[0009] Decomposing the original impedance frequency characteristic data in the frequency domain to obtain a fundamental component and a harmonic component, and performing Fourier transform on the fundamental component and the harmonic component to obtain spectrum data; inputting the spectrum data into a preset neural network compensation model to obtain a compensation coefficient, and multiplying the compensation coefficient by the spectrum data to obtain compensated impedance frequency characteristic data, wherein the neural network compensation model uses a long short-term memory network structure to extract time series features and dynamically compensate the spectrum data;

[0010] establishing an impedance-frequency characteristic curve of the reactor based on the compensated impedance-frequency characteristic data, and calculating real-time characteristic data of the impedance-frequency characteristic curve at different frequency points; continuously collecting real-time monitoring data of the reactor under different operating conditions by the sampling device, calculating a real-time impedance-frequency characteristic based on the real-time monitoring data, and comparing the real-time impedance-frequency characteristic with the real-time characteristic data to obtain a characteristic deviation;

[0011] A fault feature vector is constructed according to the feature deviation, a fault discrimination criterion is established by calculating the Euclidean distance and the Mahalanobis distance of the fault feature vector, the fault type and the fault location are determined according to the fault discrimination criterion, and a fault diagnosis report is generated.

[0012] Decomposing the original impedance frequency characteristic data in the frequency domain to obtain fundamental components and harmonic components, and performing Fourier transform on the fundamental components and the harmonic components to obtain spectrum data include:

[0013] Performing empirical mode decomposition on the original impedance frequency characteristic data to obtain an eigenmode function and a residual term, identifying an eigenmode function index set within a fundamental frequency range and an eigenmode function index set within a harmonic frequency range based on the frequency characteristics of the eigenmode function, and superimposing the eigenmode function index set within the fundamental frequency range and the eigenmode function index set within the harmonic frequency range to obtain a fundamental component and a harmonic component;

[0014] Performing short-time Fourier transform on the fundamental component and the harmonic component to obtain a time-frequency spectrum, wherein the short-time Fourier transform adopts a variable-length Kaiser window function, and the window length and shape parameters of the variable-length Kaiser window function are adaptively adjusted according to the non-stationary characteristics of the signal;

[0015] The time-frequency spectrum is subjected to feature enhancement processing, and the feature enhancement processing includes spectrum smoothing and noise suppression. The time-frequency spectrum is convolved in the frequency dimension by a smoothing kernel function to obtain a smoothed spectrum. The smoothed spectrum is subjected to wavelet transformation and the noise component is removed by a soft threshold function to obtain enhanced spectrum data.

[0016] Establishing an impedance frequency characteristic curve of the reactor according to the compensated impedance frequency characteristic data, and calculating real-time characteristic data of the impedance frequency characteristic curve at different frequency points includes:

[0017] An immune network model is constructed based on the compensated impedance-frequency characteristic data, and an impedance characteristic curve is constructed according to the antibody memory unit of the immune network model. The impedance characteristic curve is amplified and mutated by the antibody memory unit using a clonal selection algorithm, wherein the clonal selection algorithm includes two sub-steps: affinity calculation and clonal amplification. The clonal selection algorithm is used to optimize the fitting accuracy of the impedance characteristic curve in each frequency range;

[0018] Calculating the first-order derivative and the second-order derivative of the impedance amplitude at different frequency points of the impedance characteristic curve, re-inputting the first-order derivative and the second-order derivative into the immune network model, and updating the impedance change rate information stored in the antibody memory unit;

[0019] The impedance distortion degree of each frequency interval is calculated based on the updated impedance change rate information. The impedance distortion degree is used to determine the nonlinear change of the impedance characteristic curve through the clone selection algorithm. When the impedance distortion degree exceeds a preset distortion threshold, the clone selection algorithm triggers a high-frequency mutation operation to optimize the real-time characteristic data of the corresponding frequency interval by generating a new antibody population.

[0020] Constructing an immune network model based on the compensated impedance-frequency characteristic data, constructing an impedance characteristic curve based on the antibody memory unit of the immune network model, and amplifying and mutating the antibody memory unit using a clonal selection algorithm on the impedance characteristic curve includes:

[0021] The immune network model includes multiple antibody units, each of which corresponds to impedance-frequency characteristic data at a frequency point. The affinity between the antibody units is calculated using a Gaussian kernel function, and the affinity represents the similarity between the impedance-frequency characteristic data at adjacent frequency points.

[0022] Calculating the network activation of the antibody unit based on the affinity, wherein the network activation is obtained by weighted accumulation of the affinity, and the weight coefficient is determined by the antibody memory factor, and the impedance-frequency characteristic data corresponding to the antibody unit having the network activation higher than a preset activation threshold is preferentially stored in the antibody memory unit;

[0023] Cloning and amplifying the impedance-frequency characteristic data stored in the antibody memory unit, wherein the number of clones is proportional to the network activation degree, and the new antibody units generated by cloning inherit the original impedance-frequency characteristic data. An affinity maturation operation is performed on the new antibody units, wherein the affinity maturation operation perturbs the impedance-frequency characteristic data through Gaussian random mutation, and the magnitude of the mutation is inversely proportional to the network activation degree;

[0024] The mutated new antibody unit is re-input into the immune network model to calculate the updated network activation degree, the antibody memory units are screened according to the updated network activation degree, and the antibody memory units whose network activation degree is higher than the preset activation degree threshold are retained to construct an impedance characteristic curve.

[0025] Calculating a real-time impedance frequency characteristic according to the real-time monitoring data, and comparing the real-time impedance frequency characteristic with the real-time characteristic data to obtain a characteristic deviation includes:

[0026] Performing recursive least squares estimation on the real-time monitoring data, including: constructing an observation equation including voltage observation values, current observation values, and initial impedance parameters, and calculating a prediction error based on the observation equation, wherein the prediction error is an estimated deviation between the real-time monitoring data and historical impedance parameters;

[0027] Calculating a gain matrix based on the prediction error, wherein the gain matrix is ​​used to adjust the parameter update step size, multiplying the prediction error by the gain matrix to obtain a parameter update amount, and adding the parameter update amount to the historical impedance parameter to obtain the impedance parameter at the current moment;

[0028] Performing Kalman filtering on the impedance parameter to calculate a state prediction value of the impedance parameter, and performing smoothing based on a deviation between the state prediction value and the real-time monitoring data to obtain a filtered impedance state estimation value;

[0029] The real-time impedance frequency characteristic is calculated based on the impedance state estimation value, and the real-time impedance frequency characteristic includes an amplitude characteristic, a phase characteristic, and a quality factor; the real-time impedance frequency characteristic is compared with the real-time characteristic data to obtain a final characteristic deviation.

[0030] Constructing a fault feature vector based on the feature deviation, establishing a fault discrimination criterion by calculating the Euclidean distance and the Mahalanobis distance of the fault feature vector, determining the fault type and the fault location based on the fault discrimination criterion, and generating a fault diagnosis report includes:

[0031] Fusing the fault feature vector with the expert experience feature vector, weighting the fault feature vector by a cognitive weight matrix to obtain an experience fusion coefficient; weighting the expert experience feature vector by the experience fusion coefficient to obtain a cognitive enhancement feature vector;

[0032] Calculating the Euclidean distance and Mahalanobis distance of the cognitive enhancement feature vector, wherein the Euclidean distance represents the direct distance between feature vectors and the Mahalanobis distance considers the covariance relationship of feature distribution, and constructing a fault discrimination criterion based on the Euclidean distance and the Mahalanobis distance;

[0033] performing cognitive weight optimization on the fault discrimination criterion, inputting the Euclidean distance and the Mahalanobis distance into a loss function, introducing cognitive rule constraints, and obtaining an optimized cognitive weight by minimizing a weighted sum of the loss function and the cognitive rule constraints;

[0034] Calculating a fault probability distribution based on the optimized cognitive weights, adjusting the discrimination degree of fault types by using a discriminant sensitivity parameter, and combining the fault probability distribution with a fault space distribution function to obtain a fault location probability map;

[0035] Expert experience knowledge is updated using a reinforcement learning value function, the expert experience knowledge is weighed between historical experience and the reinforcement learning value function through a learning rate, the fault location probability map is optimized based on the updated expert experience knowledge, and a fault diagnosis report is generated.

[0036] Calculating the fault probability distribution based on the optimized cognitive weight, adjusting the discrimination degree of the fault type by using a discriminant sensitivity parameter, and combining the fault probability distribution with the fault space distribution function to obtain a fault location probability map includes:

[0037] Calculating the fault probability distribution based on the optimized cognitive weights, and adjusting the discrimination degree of the fault types by using a discrimination sensitivity parameter, wherein the discrimination sensitivity parameter is adaptively adjusted according to the degree of difference in the fault characteristics;

[0038] The fault probability distribution and the fault spatial distribution function are weightedly combined to obtain a fault location probability map, wherein the fault spatial distribution function describes the spatial distribution characteristics of the fault type, and the fault location probability map reflects the spatial probability distribution of the fault location.

[0039] A second aspect of an embodiment of the present invention provides an automated measurement and analysis system for reactor impedance characteristics, comprising:

[0040] The first unit is configured to collect voltage signals and current signals through a sampling device of the reactor, and calculate original impedance frequency characteristic data of the reactor based on the voltage signals and the current signals;

[0041] The second unit is configured to perform frequency domain decomposition on the original impedance frequency characteristic data to obtain a fundamental component and a harmonic component, perform Fourier transform on the fundamental component and the harmonic component to obtain spectrum data; input the spectrum data into a preset neural network compensation model to obtain a compensation coefficient, multiply the compensation coefficient by the spectrum data to obtain compensated impedance frequency characteristic data, and the neural network compensation model uses a long short-term memory network structure to perform time series feature extraction and dynamic compensation on the spectrum data;

[0042] A third unit is configured to establish an impedance-frequency characteristic curve of the reactor based on the compensated impedance-frequency characteristic data, and calculate real-time characteristic data of the impedance-frequency characteristic curve at different frequency points; continuously collect real-time monitoring data of the reactor under different operating conditions through the sampling device, calculate a real-time impedance-frequency characteristic based on the real-time monitoring data, and compare the real-time impedance-frequency characteristic with the real-time characteristic data to obtain a characteristic deviation;

[0043] The fourth unit is used to construct a fault feature vector based on the feature deviation, establish a fault discrimination criterion by calculating the Euclidean distance and Mahalanobis distance of the fault feature vector, determine the fault type and fault location according to the fault discrimination criterion, and generate a fault diagnosis report.

[0044] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0045] processor;

[0046] a memory for storing processor-executable instructions;

[0047] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0048] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0049] The beneficial effects of this application are as follows:

[0050] The present invention obtains the voltage and current signals of the reactor through a sampling device, calculates the original impedance characteristics, performs frequency domain decomposition and Fourier transform, and uses a neural network compensation model with a long short-term memory network structure to dynamically compensate the spectrum data, effectively improving the accuracy and stability of the reactor impedance characteristic measurement, and solving the problem that traditional measurement methods are easily interfered with under complex working conditions.

[0051] The present invention establishes a complete mechanism for comparing and analyzing impedance-frequency characteristic curves and real-time characteristic data. By continuously monitoring the real-time impedance characteristic changes under different operating conditions and calculating the characteristic deviation, dynamic tracking of the reactor's operating status and early warning of abnormalities are achieved, thereby improving the system's real-time monitoring capabilities and fault prediction accuracy.

[0052] The present invention constructs a fault diagnosis method based on feature vectors. By calculating the Euclidean distance and Mahalanobis distance to establish fault discrimination criteria, it can accurately identify the fault type and locate the fault location, realize intelligent diagnosis and analysis of reactor faults, provide a scientific basis for the maintenance and management of power grid equipment, and has significant engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the process of the method for automatically measuring and analyzing the impedance characteristics of a reactor according to an embodiment of the present invention;

[0054] Figure 2 Schematic diagram comparing the time-frequency resolutions of different decomposition methods according to an embodiment of the present invention;

[0055] Figure 3 This is a flow chart of establishing an impedance frequency characteristic curve and calculating characteristic data according to an embodiment of the present invention;

[0056] Figure 4 Schematic diagram showing a comparison of estimated stability under different environmental interferences according to an embodiment of the present invention;

[0057] Figure 5 This is a bar chart comparing and analyzing the performance of the cognitive enhancement fault diagnosis method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0059] The technical solution of the present invention is described in detail below with reference to specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0060] Figure 1 FIG. 1 is a flow chart of a method for automatically measuring and analyzing the impedance characteristics of a reactor according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0061] The sampling device of the reactor collects voltage signals and current signals, and calculates original impedance frequency characteristic data of the reactor according to the voltage signals and the current signals;

[0062] Decomposing the original impedance frequency characteristic data in the frequency domain to obtain a fundamental component and a harmonic component, and performing Fourier transform on the fundamental component and the harmonic component to obtain spectrum data; inputting the spectrum data into a preset neural network compensation model to obtain a compensation coefficient, and multiplying the compensation coefficient by the spectrum data to obtain compensated impedance frequency characteristic data, wherein the neural network compensation model uses a long short-term memory network structure to extract time series features and dynamically compensate the spectrum data;

[0063] establishing an impedance-frequency characteristic curve of the reactor based on the compensated impedance-frequency characteristic data, and calculating real-time characteristic data of the impedance-frequency characteristic curve at different frequency points; continuously collecting real-time monitoring data of the reactor under different operating conditions by the sampling device, calculating a real-time impedance-frequency characteristic based on the real-time monitoring data, and comparing the real-time impedance-frequency characteristic with the real-time characteristic data to obtain a characteristic deviation;

[0064] A fault feature vector is constructed according to the feature deviation, a fault discrimination criterion is established by calculating the Euclidean distance and the Mahalanobis distance of the fault feature vector, the fault type and the fault location are determined according to the fault discrimination criterion, and a fault diagnosis report is generated.

[0065] In an optional embodiment, performing frequency domain decomposition on the original impedance frequency characteristic data to obtain a fundamental component and a harmonic component, and performing Fourier transform on the fundamental component and the harmonic component to obtain spectrum data includes:

[0066] Performing empirical mode decomposition on the original impedance frequency characteristic data to obtain an eigenmode function and a residual term, identifying an eigenmode function index set within a fundamental frequency range and an eigenmode function index set within a harmonic frequency range based on the frequency characteristics of the eigenmode function, and superimposing the eigenmode function index set within the fundamental frequency range and the eigenmode function index set within the harmonic frequency range to obtain a fundamental component and a harmonic component;

[0067] Performing short-time Fourier transform on the fundamental component and the harmonic component to obtain a time-frequency spectrum, wherein the short-time Fourier transform adopts a variable-length Kaiser window function, and the window length and shape parameters of the variable-length Kaiser window function are adaptively adjusted according to the non-stationary characteristics of the signal;

[0068] The time-frequency spectrum is subjected to feature enhancement processing, and the feature enhancement processing includes spectrum smoothing and noise suppression. The time-frequency spectrum is convolved in the frequency dimension by a smoothing kernel function to obtain a smoothed spectrum. The smoothed spectrum is subjected to wavelet transformation and the noise component is removed by a soft threshold function to obtain enhanced spectrum data.

[0069] For the frequency domain decomposition of the original impedance frequency characteristic data, this embodiment adopts the empirical mode decomposition (EMD) method. Assume that there is a set of collected original impedance frequency characteristic data with 10,000 data points and a sampling frequency of 1 kHz. The EMD method decomposes the signal into multiple intrinsic mode functions (IMFs) and a residual term. In actual operation, the EMD process forms an upper envelope and a lower envelope respectively by finding the local extreme points of the signal, connecting all the maximum points and minimum points. Through iterative calculation, a series of IMF components are obtained. For the 10,000-point original impedance data, 10 IMF components and 1 residual term are obtained after EMD decomposition.

[0070] For the obtained IMF components, it is necessary to identify which ones belong to the fundamental frequency range and which ones belong to the harmonic frequency range. This step is achieved by analyzing the frequency characteristics of each IMF component. Specifically, the main frequency component of each IMF is calculated. If the main frequency falls within the pre-defined fundamental frequency range (for example, 45-55Hz), the index of the IMF is added to the eigenmode function index set of the fundamental frequency range; if the main frequency falls within the harmonic frequency range (for example, 95-105Hz, 145-155Hz, etc.), the index of the IMF is added to the eigenmode function index set of the harmonic frequency range.

[0071] By analyzing the dominant frequency characteristics of the ten IMFs, we determined that the dominant frequencies of IMF1, IMF2, and IMF3 are 150 Hz, 100 Hz, and 50 Hz, respectively. Therefore, the eigenmode function index set for the fundamental frequency range is {3} (corresponding to IMF3), and the eigenmode function index set for the harmonic frequency range is {1, 2} (corresponding to IMF1 and IMF2). The dominant frequencies of the remaining IMFs are outside the frequency range of interest and are therefore excluded from these two index sets.

[0072] The fundamental component is obtained by superimposing the IMFs within the fundamental frequency range (i.e., IMF3), and the harmonic components are obtained by superimposing the IMFs within the harmonic frequency range (i.e., IMF1 and IMF2). In this way, the original impedance frequency characteristic data is decomposed into the fundamental component and the harmonic component.

[0073] A short-time Fourier transform (STFT) is performed on the fundamental and harmonic components to obtain time-frequency spectrum information. In the STFT, a variable-length Kaiser window function is used to segment the signal. The Kaiser window function has excellent frequency domain characteristics and can reduce spectral leakage. The window length and shape parameters of the variable-length Kaiser window function are adaptively adjusted based on the non-stationary characteristics of the signal.

[0074] Specifically, the choice of window length depends on the local stationarity of the signal. For relatively stable signal segments, a longer window (e.g., 512 points) is used to obtain higher frequency resolution; for rapidly changing signal segments, a shorter window (e.g., 128 points) is used to obtain higher time resolution. The shape parameter β controls the degree of sidelobe attenuation of the window function. A larger β value results in less spectral leakage, but the mainlobe width increases and the frequency resolution decreases. In this example, for the fundamental component, β=6.0 is selected as the shape parameter of the Kaiser window; for the harmonic component, β=8.0 is selected to obtain better frequency resolution.

[0075] Through STFT processing, the time-spectrum of the fundamental and harmonic components is obtained. The time-spectrum is a three-dimensional data, with time on the horizontal axis and frequency on the vertical axis, and the color depth represents the energy level. In this example, the time-spectrum of the fundamental component shows a clear energy concentration around 50Hz, while the time-spectrum of the harmonic components shows energy distribution at 100Hz and 150Hz.

[0076] To improve the quality of the time-frequency spectrum, feature enhancement processing is performed, including two steps: spectrum smoothing and noise suppression. Spectral smoothing is achieved by convolution with a smoothing kernel function in the frequency dimension. In this embodiment, a Gaussian kernel function is selected as the smoothing kernel, and the kernel width is set to 5 frequency points to balance the smoothing effect and frequency resolution. Through the convolution operation, the original time-frequency spectrum is converted into a smoothed spectrum, eliminating local mutations and glitches in the spectrum.

[0077] Noise suppression is achieved using a wavelet transform and a soft thresholding function. The smoothed spectrum data is subjected to wavelet decomposition, using the Daubechies4 (db4) wavelet basis and a five-level decomposition. A soft thresholding function is applied to the wavelet coefficients, with the threshold set at three times the standard deviation of the wavelet coefficients. The soft thresholding function sets coefficients below the threshold to zero and shrinks coefficients above the threshold, effectively removing random noise from the spectrum while preserving the signal's key features.

[0078] Through inverse wavelet transform, the processed wavelet coefficients are resynthesized into enhanced spectrum data. After feature enhancement, the signal-to-noise ratio of the time-frequency spectrum is significantly improved, and the frequency characteristics of the fundamental and harmonic components are more clearly distinguishable, providing a high-quality data foundation for subsequent feature extraction and pattern recognition.

[0079] In practical applications, this method can effectively separate the fundamental and harmonic components in the power system through decomposition and reconstruction, reveal the temporal variation of impedance characteristics through time-frequency analysis, and further improve the data quality through feature enhancement processing, providing accurate and reliable technical support for power system impedance modeling and harmonic assessment.

[0080] Figure 2 Schematic diagram comparing the time-frequency resolution of different decomposition methods according to an embodiment of the present invention:

[0081] This figure compares the analysis accuracy of three different analysis methods in the frequency range of 0-500 Hz. The proposed method (circles) demonstrates excellent performance in the low-frequency range, achieving an accuracy of 0.92 at 0 Hz and maintaining a high accuracy of 0.93-0.94 in the 50-100 Hz range. Its accuracy slowly decreases with increasing frequency, but it still maintains a high accuracy of 0.77 at 500 Hz. The fixed-window STFT method (diamonds) offers intermediate overall performance, with accuracy of approximately 0.68-0.70 in the low-frequency range, but gradually decreasing to 0.55-0.57 with increasing frequency. The standard Fourier transform method (squares) offers the worst performance, with accuracy of only 0.39-0.42 in the low-frequency range. While this performance improves slightly to 0.53-0.55 in the 200-250 Hz range, it continues to decline to around 0.40 in the high-frequency range. Judging from the trend, this technical solution maintains a clear performance advantage in the entire frequency band, especially in the low-frequency band, which is an average of 20-40 percentage points higher than the other two methods, demonstrating stronger frequency analysis capabilities.

[0082] In an optional embodiment, establishing an impedance frequency characteristic curve of the reactor based on the compensated impedance frequency characteristic data, and calculating real-time characteristic data of the impedance frequency characteristic curve at different frequency points includes:

[0083] An immune network model is constructed based on the compensated impedance-frequency characteristic data, and an impedance characteristic curve is constructed according to the antibody memory unit of the immune network model. The impedance characteristic curve is amplified and mutated by the antibody memory unit using a clonal selection algorithm, wherein the clonal selection algorithm includes two sub-steps: affinity calculation and clonal amplification. The clonal selection algorithm is used to optimize the fitting accuracy of the impedance characteristic curve in each frequency range;

[0084] Calculating the first-order derivative and the second-order derivative of the impedance amplitude at different frequency points of the impedance characteristic curve, re-inputting the first-order derivative and the second-order derivative into the immune network model, and updating the impedance change rate information stored in the antibody memory unit;

[0085] The impedance distortion degree of each frequency interval is calculated based on the updated impedance change rate information. The impedance distortion degree is used to determine the nonlinear change of the impedance characteristic curve through the clone selection algorithm. When the impedance distortion degree exceeds a preset distortion threshold, the clone selection algorithm triggers a high-frequency mutation operation to optimize the real-time characteristic data of the corresponding frequency interval by generating a new antibody population.

[0086] like Figure 3 As shown, the method includes:

[0087] The original impedance-frequency characteristic measurement data of the reactor is obtained and then compensated to eliminate the influence of measurement errors and environmental factors. Compensation uses measurement system error correction technology to identify and correct systematic and random errors in the original data. For example, for a set of measured impedance data [120.5Ω, 125.3Ω, 132.1Ω, 140.8Ω, 152.4Ω], compensation processing yields the corrected values ​​[120.2Ω, 125.0Ω, 132.5Ω, 141.0Ω, 152.0Ω] at the frequencies [50Hz, 100Hz, 500Hz, 1kHz, 5kHz].

[0088] An immune network model is constructed based on the compensated impedance-frequency characteristic data. This model consists of an antigen recognition layer, an antibody network layer, and a memory unit layer. The antigen recognition layer receives the compensated impedance data as input. The antibody network layer contains multiple antibody units, each corresponding to the impedance characteristics of a frequency point. The memory unit layer stores the optimized impedance characteristic curve parameters. In actual implementation, for a dry-type air-core reactor with a rated voltage of 35kV, the number of antibody network layer units can be set to 50, each of which contains three parameters: impedance amplitude, phase angle, and impedance change rate.

[0089] The impedance characteristic curve is constructed based on the contents of the antibody memory cells of the immune network model. Specifically, the frequency range is divided into multiple intervals, for example, 50 Hz to 5 kHz is divided into 20 intervals. Five characteristic frequency points are selected from each interval, for a total of 100 frequency points forming the characteristic frequency set. For each frequency point, the corresponding impedance characteristic parameters are extracted from the antibody memory cells to form the initial impedance characteristic curve.

[0090] A clonal selection algorithm is used to amplify and mutate antibody memory cells to optimize the fitting accuracy of the impedance characteristic curve. The clonal selection algorithm includes two sub-steps: affinity calculation and clonal amplification. In the affinity calculation stage, an affinity function is defined between the antibody and the antigen to evaluate the degree of fit between the current impedance characteristic curve and the actual measured data. For example, at the frequency point f1 = 100 Hz, if the actual measured impedance value is 125.0 Ω and the current curve predicts a value of 124.8 Ω, the calculated affinity is 0.998. In the clonal amplification stage, antibodies with high affinity are cloned and amplified, and the number of clones is proportional to the affinity. For example, an antibody with an affinity of 0.998 can produce 20 clones, while an antibody with an affinity of 0.85 can produce only 5 clones.

[0091] The amplified antibody population is mutated, with the probability of mutation inversely proportional to affinity, to increase population diversity. Mutation methods include Gaussian and uniform mutations, with different mutation strategies selected for different frequency ranges. For the low-frequency range (50Hz-500Hz), a small-amplitude Gaussian mutation is used, with the variation range controlled within ±2% of the original value; for the high-frequency range (1kHz-5kHz), a larger-amplitude uniform mutation is used, with the variation range reaching ±5% of the original value. After mutation, affinity is recalculated, and the antibody population with the highest affinity is selected to update the memory unit.

[0092] For the optimized impedance characteristic curve, calculate the first and second derivatives of the impedance amplitude at each frequency point. The first derivative represents the rate of change of impedance with frequency, and the second derivative represents the trend of the rate of change. For example, in the frequency range from 100Hz to 500Hz, the first derivative can be calculated to be 0.025Ω / Hz, indicating that the impedance increases by 0.025Ω for every 1Hz increase; the second derivative is 0.00002Ω / Hz. 2 , indicating that the impedance change rate shows a slight acceleration trend.

[0093] The calculated first- and second-order derivatives are re-input into the immune network model to update the impedance rate of change information stored in the antibody memory unit. The update process uses an adaptive weighting method to weight the newly calculated rate of change with the existing data in the memory unit. The weight coefficient is automatically adjusted based on the credibility of the current data. For example, for measurement intervals with a high signal-to-noise ratio, the weight of the new data can be set to 0.8, while for intervals with higher noise, the weight can be reduced to 0.4.

[0094] Based on the updated impedance change rate information, the impedance distortion level is calculated for each frequency range. Impedance distortion is defined as the deviation of the impedance curve from the ideal linear model. The actual impedance curve is compared with the theoretical linear model, and the degree of distortion is quantified using the cumulative sum of squared errors. For example, in the 500Hz to 1kHz range, a calculated impedance distortion level of 0.07 indicates a slight nonlinear impedance change within this range.

[0095] When the impedance distortion level in a certain frequency range exceeds a preset distortion threshold (such as 0.15), the high-frequency mutation operation of the clonal selection algorithm is triggered. The high-frequency mutation operation generates a more diverse antibody population by increasing the probability and amplitude of mutation, and searches for a better combination of characteristic parameters. For example, for the 1kHz to 2kHz range where the distortion level reaches 0.18, the mutation probability can be increased from the standard 0.1 to 0.3, and the amplitude of mutation can be expanded from ±5% to ±10%. 100 new antibody candidate solutions are generated, and the 10 with the highest affinity are selected from these candidate solutions to update the memory units and optimize the real-time characteristic data of this frequency range.

[0096] Through the above method, the impedance frequency characteristic curve of the reactor can be accurately constructed, and the real-time characteristic data of different frequency points can be calculated, providing reliable data support for the condition monitoring and fault diagnosis of the reactor.

[0097] In an optional embodiment, constructing an immune network model based on the compensated impedance-frequency characteristic data, constructing an impedance characteristic curve based on the antibody memory unit of the immune network model, and amplifying and mutating the antibody memory unit using a clonal selection algorithm on the impedance characteristic curve includes:

[0098] The immune network model includes multiple antibody units, each of which corresponds to impedance-frequency characteristic data at a frequency point. The affinity between the antibody units is calculated using a Gaussian kernel function, and the affinity represents the similarity between the impedance-frequency characteristic data at adjacent frequency points.

[0099] Calculating the network activation of the antibody unit based on the affinity, wherein the network activation is obtained by weighted accumulation of the affinity, and the weight coefficient is determined by the antibody memory factor, and the impedance-frequency characteristic data corresponding to the antibody unit having the network activation higher than a preset activation threshold is preferentially stored in the antibody memory unit;

[0100] Cloning and amplifying the impedance-frequency characteristic data stored in the antibody memory unit, wherein the number of clones is proportional to the network activation degree, and the new antibody units generated by cloning inherit the original impedance-frequency characteristic data. An affinity maturation operation is performed on the new antibody units, wherein the affinity maturation operation perturbs the impedance-frequency characteristic data through Gaussian random mutation, and the magnitude of the mutation is inversely proportional to the network activation degree;

[0101] The mutated new antibody unit is re-input into the immune network model to calculate the updated network activation degree, the antibody memory units are screened according to the updated network activation degree, and the antibody memory units whose network activation degree is higher than the preset activation degree threshold are retained to construct an impedance characteristic curve.

[0102] When building an immune network model based on compensated impedance-frequency characteristic data, the system first represents the impedance data corresponding to each frequency point as an antibody unit. For example, for the impedance data measured at 100 frequency points, the system creates 100 antibody units, each of which contains the impedance amplitude, phase information, and frequency identifier of the corresponding frequency point. In practical applications, such as performing EIS testing on lithium batteries, impedance data for a total of 64 frequency points ranging from 1 Hz to 10 kHz is obtained, and each of these 64 measured values ​​forms an antibody unit.

[0103] The affinity between antibody units is calculated by Gaussian kernel function, which reflects the similarity of impedance data at adjacent frequency points. i and A j Calculate the Euclidean distance between them, including the difference in the real and imaginary impedance parts. Substitute this distance into a Gaussian kernel function to obtain the affinity value. The kernel function's bandwidth parameter is set to 0.75, ensuring that impedance data at close frequency points have a high affinity, while affinity rapidly decreases at distant points. In practice, for example, if the impedance data for antibody units at two frequencies, 1 Hz and 1.5 Hz, are similar, the calculated affinity can reach 0.92; however, the affinity for antibody units at 1 Hz and 100 Hz is only 0.13.

[0104] When calculating network activation, the system performs a weighted accumulation of the affinity of each antibody unit with all other units. The weight coefficient is determined by the antibody memory factor. Initially, the memory factor of all antibodies is set to 1.0. As the learning process progresses, the memory factor of important frequency points will increase. For example, the affinity of an antibody unit with the other 64 units is calculated separately, multiplied by the corresponding memory factor (all initially 1.0), and the sum is calculated to obtain the network activation of the unit. In the lithium battery test case, the antibody unit corresponding to the inflection point frequency of the impedance spectrum generally obtains a higher activation degree. For example, the activation degree at the 30Hz frequency point can reach 6.8, while the activation degree at points far from the characteristic frequency, such as 9kHz, is only 2.3.

[0105] The system sets an activation threshold of 4.0, prioritizing the storage of impedance-frequency characteristic data corresponding to antibody units with network activation levels above this threshold into the antibody memory unit. In actual cases, this threshold screening resulted in approximately 20 of the 64 frequency points being stored in the antibody memory unit. These points were primarily concentrated in the characteristic frequency regions of the impedance spectrum, such as the frequencies corresponding to SEI film impedance and charge transfer impedance during charge and discharge.

[0106] Impedance data stored in the antibody memory unit is cloned and amplified. The number of clones is positively correlated with the network activation level. For example, an antibody unit with a network activation level of 6.8 clones produces six new antibodies, while a unit with an activation level of 4.2 only produces four new antibodies. The cloned new antibody units inherit the characteristics of the original impedance data, but require affinity maturation.

[0107] The affinity maturation operation perturbs the impedance characteristic data through Gaussian random mutation. The magnitude of the mutation is inversely proportional to the network activation. Highly activated antibodies have smaller mutations, maintaining stability, while low-activated antibodies have larger mutations, increasing exploration. In practice, for an antibody unit with an activation of 6.8, the standard deviation of the mutation is set to 1.5% of the original impedance value; for a unit with an activation of 4.2, the standard deviation is set to 3.5%. This mechanism allows the system to fully explore marginal regions while maintaining the stability of important feature points.

[0108] The newly mutated antibody units are re-input into the immune network model, and the network activation is recalculated. For example, a point with an activation of 6.8 in the original antibody unit generates six new antibody units after clonal mutation, with new activations of 6.9, 6.7, 6.8, 6.5, 7.0, and 6.6, respectively. The system retains units with activations above the preset threshold of 4.0; in this case, all six new antibodies are retained.

[0109] After completing multiple iterations of clone selection (usually set to 5 rounds), the system integrates all retained antibody memory units to construct a complete impedance characteristic curve. The final result is an optimized impedance curve based on the original measurement data, highlighting the characteristics of key frequency points, while filling in data sparse areas through a mutation mechanism. In the case of battery impedance spectroscopy, the original 64 frequency points can be optimized through the immune network to obtain a smooth impedance curve containing more than 120 frequency points. This curve has higher point density and accuracy in key frequency regions (such as 0.5Hz to 10Hz), which can more accurately reflect the impedance characteristics of the electrochemical system.

[0110] In the system implementation, antibody memory cells are stored in a priority queue data structure, facilitating sorting by network activation. The clone selection process is executed in parallel on multi-core processors, improving computational efficiency. For large-scale impedance data containing 1,000 frequency points, the complete processing time is controlled within 2 seconds, meeting the requirements of real-time analysis.

[0111] In an optional embodiment, calculating the real-time impedance frequency characteristic based on the real-time monitoring data, and comparing the real-time impedance frequency characteristic with the real-time characteristic data to obtain a characteristic deviation includes:

[0112] Performing recursive least squares estimation on the real-time monitoring data, including: constructing an observation equation including voltage observation values, current observation values, and initial impedance parameters, and calculating a prediction error based on the observation equation, wherein the prediction error is an estimated deviation between the real-time monitoring data and historical impedance parameters;

[0113] Calculating a gain matrix based on the prediction error, wherein the gain matrix is ​​used to adjust the parameter update step size, multiplying the prediction error by the gain matrix to obtain a parameter update amount, and adding the parameter update amount to the historical impedance parameter to obtain the impedance parameter at the current moment;

[0114] Performing Kalman filtering on the impedance parameter to calculate a state prediction value of the impedance parameter, and performing smoothing based on a deviation between the state prediction value and the real-time monitoring data to obtain a filtered impedance state estimation value;

[0115] The real-time impedance frequency characteristic is calculated based on the impedance state estimation value, and the real-time impedance frequency characteristic includes an amplitude characteristic, a phase characteristic, and a quality factor; the real-time impedance frequency characteristic is compared with the real-time characteristic data to obtain a final characteristic deviation.

[0116] Recursive least squares estimation is performed on the collected real-time monitoring data of the power grid, followed by Kalman filtering. Finally, the real-time impedance frequency characteristics are calculated and compared with the standard characteristic data.

[0117] Real-time monitoring data for the power grid includes information such as sampling time, voltage amplitude, current amplitude, and phase angle. To implement recursive least squares estimation, the system first constructs an observation equation that integrates the voltage and current observations and initial impedance parameters. These initial impedance parameters can be determined based on historical operating data or theoretically calculated values. For example, at a 500kV power grid node, the initial resistance parameter can be set to 0.25 ohms, and the initial inductance parameter can be set to 0.0012 Henry.

[0118] After constructing the observation equation, the system calculates the prediction error—the difference between the real-time monitoring data and the predicted value based on historical impedance parameters. For example, if the system detects a voltage of 510 kV at a certain moment, but the predicted value based on historical impedance parameters is 505 kV, the prediction error is 5 kV. This prediction error reflects the difference between the current grid state and the historical model and is an important basis for updating the impedance parameters.

[0119] After the prediction error is calculated, the system calculates a gain matrix, which is used to adjust the parameter update step size. The gain matrix calculation takes into account the influence of historical observations on the current estimate. This gain matrix typically decreases as the amount of observations increases to ensure the stability of the estimation process. In practice, the weighting ratio of new and old data can be adjusted by setting a forgetting factor. For example, a forgetting factor of 0.98 indicates that new data is slightly more weighted than old data, helping impedance parameters adapt more quickly to grid state changes.

[0120] Multiplying the prediction error by the gain matrix yields the parameter update. For example, the resistance update at a given moment is 0.02 ohms, and the inductance update is 0.0001 henry. These updates are then added to the historical impedance parameters to obtain the current impedance parameter estimate. For example, the updated resistance is 0.27 ohms, and the inductance is 0.0013 henry.

[0121] To further improve the accuracy of impedance parameter estimation, the system performs Kalman filtering on the impedance parameters obtained by recursive least squares. Kalman filtering first calculates a state prediction value for the impedance parameter, which is derived based on the impedance state at the previous moment and the system model. In this embodiment, the state prediction model assumes that the impedance parameter changes little over a short period of time and uses the estimated value at the previous moment as the current prediction value. It also accounts for a certain amount of process noise, such as setting the standard deviation of the process noise for resistance parameters to 0.01 ohms and for inductance parameters to 0.0005 henry.

[0122] After the state prediction is complete, the system compares the deviation between the predicted value and the real-time monitoring data and performs smoothing based on this deviation. The key parameter in this smoothing process is the Kalman gain, which determines the system's confidence in the new measurement data. When measurement noise is low, the Kalman gain is large, and the system is more inclined to trust the new measurement value; when measurement noise is high, the Kalman gain is small, and the system is more inclined to trust the prior estimate. For example, when the standard deviation of the noise in the measured voltage is 0.5 kV and the standard deviation of the noise in the current is 20 A, the Kalman gain is approximately 0.7, indicating that the system has a high degree of confidence in the new measurement data.

[0123] Through Kalman filtering, the system generates filtered impedance state estimates that are smoother and more robust to interference than those obtained using recursive least squares alone. For example, the filtered resistance parameter is 0.265 ohms, and the inductance parameter is 0.00128 Henry, both of which have lower volatility than the original estimates.

[0124] Based on the filtered impedance state estimate, the system calculates real-time impedance frequency characteristics, including amplitude, phase, and quality factor at different frequencies. The amplitude characteristic indicates the magnitude of impedance changes with frequency, the phase characteristic indicates the phase angle of impedance changes with frequency, and the quality factor reflects the quality of the impedance characteristics. Under normal operating conditions, the standard impedance amplitude at a grid node at 50 Hz is 0.3 ohms, the phase is 25 degrees, and the quality factor is 10.5.

[0125] The calculated real-time impedance frequency characteristics are compared with the stored real-time characteristic data to obtain the final characteristic deviation. This characteristic deviation reflects the difference between the current grid impedance characteristics and the normal or expected state and is an important indicator for assessing grid health. For example, if the real-time calculated impedance amplitude at 50 Hz is 0.35 ohms, the phase is 28 degrees, and the quality factor is 9.8, then the amplitude deviation is 0.05 ohms, the phase deviation is 3 degrees, and the quality factor deviation is -0.7.

[0126] By leveraging these technologies, this method can accurately monitor changes in grid impedance parameters in real time, promptly identifying anomalies and providing strong support for safe and stable grid operation. Practice has demonstrated that this method maintains an impedance estimation accuracy exceeding 95% in complex grid environments, with an average characteristic deviation detection delay of less than 50 milliseconds, meeting the requirements for real-time grid monitoring.

[0127] Figure 4 This is a schematic diagram comparing the estimated stability under different environmental interferences according to an embodiment of the present invention:

[0128] This boxplot compares the measurement error distribution of three different methods under five operating conditions. Under normal conditions, this technical solution has the smallest measurement error, with a median of only 0.5%, while the median errors of the traditional method and the comparison method are 1.2% and 1.1%, respectively. As the complexity of the operating conditions increases, the measurement errors of all three methods show an upward trend: under temperature fluctuation conditions, the median errors of the three methods rise to 0.9%, 2.0%, and 1.5%, respectively; in electromagnetic interference environments, the errors further increase to 1.5%, 3.1%, and 2.5%; under load fluctuation conditions, the errors reach 2.0%, 3.7%, and 3.1%, respectively; and in the most complex harmonic interference conditions, the errors are the largest, at 2.5%, 4.7%, and 4.1%, respectively. In terms of the error distribution range, this technical solution exhibits a smaller error fluctuation range under various operating conditions, demonstrating better stability. Especially under complex operating conditions, the upper limit of the error of this technical solution is significantly lower than that of the other two methods, demonstrating stronger anti-interference ability and adaptability.

[0129] In an optional embodiment, constructing a fault feature vector based on the feature deviation, establishing a fault discrimination criterion by calculating the Euclidean distance and the Mahalanobis distance of the fault feature vector, determining the fault type and fault location based on the fault discrimination criterion, and generating a fault diagnosis report includes:

[0130] Fusing the fault feature vector with the expert experience feature vector, weighting the fault feature vector by a cognitive weight matrix to obtain an experience fusion coefficient; weighting the expert experience feature vector by the experience fusion coefficient to obtain a cognitive enhancement feature vector;

[0131] Calculating the Euclidean distance and Mahalanobis distance of the cognitive enhancement feature vector, wherein the Euclidean distance represents the direct distance between feature vectors and the Mahalanobis distance considers the covariance relationship of feature distribution, and constructing a fault discrimination criterion based on the Euclidean distance and the Mahalanobis distance;

[0132] performing cognitive weight optimization on the fault discrimination criterion, inputting the Euclidean distance and the Mahalanobis distance into a loss function, introducing cognitive rule constraints, and obtaining an optimized cognitive weight by minimizing a weighted sum of the loss function and the cognitive rule constraints;

[0133] Calculating a fault probability distribution based on the optimized cognitive weights, adjusting the discrimination degree of fault types by using a discriminant sensitivity parameter, and combining the fault probability distribution with a fault space distribution function to obtain a fault location probability map;

[0134] Expert experience knowledge is updated using a reinforcement learning value function, the expert experience knowledge is weighed between historical experience and the reinforcement learning value function through a learning rate, the fault location probability map is optimized based on the updated expert experience knowledge, and a fault diagnosis report is generated.

[0135] By constructing a fault feature vector and combining Euclidean and Mahalanobis distances to establish a fault discrimination criterion, the system accurately determines the fault type and location. The system first collects equipment operating status data, including key parameters such as vibration signals, temperature fluctuations, and current fluctuations. This data is preprocessed to eliminate noise and extract valid information. The system then calculates the characteristic deviation between the normal operating state and the current state to form an initial fault feature vector.

[0136] This method fuses the fault feature vector with the expert experience feature vector. Assume that the initial fault feature vector is [0.58, 0.23, 0.67, 0.41], representing the deviation values ​​of the four key feature parameters, and the expert experience feature vector is [0.60, 0.20, 0.65, 0.45]. The fault feature vector is weighted using a cognitive weight matrix. This cognitive weight matrix, derived from historical case analysis, can be set to [0.4, 0.2, 0.3, 0.1] for this type of fault, representing the importance of the four feature parameters. Multiplying the fault feature vector by the cognitive weight matrix yields an empirical fusion coefficient of 0.484. This coefficient is then used to weight the expert experience feature vector, resulting in a cognitive enhancement feature vector of [0.290, 0.097, 0.315, 0.218].

[0137] The Euclidean distance and Mahalanobis distance of the cognitive enhancement feature vectors were calculated to accurately determine the fault type. The Euclidean distance represents the direct distance between feature vectors and intuitively reflects the degree to which the fault deviates from the normal state. The Euclidean distance between the sample feature vector [0.290, 0.097, 0.315, 0.218] and the standard fault type A feature vector [0.300, 0.100, 0.320, 0.220] was calculated to be 0.013, indicating a close match. The Mahalanobis distance considers the covariance relationship of the feature distribution and can better handle the correlation between features. For the same data set, the calculated Mahalanobis distance was 0.867, which is lower than the fault discrimination threshold of 1.0, further confirming that the fault belongs to type A. Based on these two distance metrics, a comprehensive fault discrimination criterion was constructed: when the Euclidean distance is less than 0.05 and the Mahalanobis distance is less than 1.0, the fault type is determined to be the corresponding fault type.

[0138] Cognitive weight optimization is performed on the fault discrimination criteria to improve discrimination accuracy. Euclidean distance and Mahalanobis distance are used as input parameters, and a loss function is introduced to calculate the discrimination error. The loss function is designed as a weighted combination of Euclidean distance and Mahalanobis distance, with initial weights of 0.6 and 0.4, respectively. Cognitive rule constraints are also introduced, and the characteristic distribution patterns of different fault types are defined based on expert knowledge. An iterative optimization method is used to minimize the weighted sum of the loss function and cognitive rule constraints. After 50 iterations, the optimized cognitive weights are [0.45, 0.15, 0.32, 0.08], which more accurately reflect the actual contribution of each characteristic parameter to fault diagnosis than the initial weights.

[0139] Calculating the fault probability distribution based on the optimized cognitive weights helps determine the fault type and location. For the identified fault type A, the probability distribution of each fault subtype calculated based on the optimized cognitive weights is [0.65, 0.20, 0.10, 0.05], indicating that subtype A1 has the highest probability. The fault probability distribution is adjusted using the discriminant sensitivity parameter, which is set to 1.2. This improves the differentiation between fault types and makes the main fault characteristics more prominent. The fault probability distribution is combined with the equipment's fault spatial distribution function, which is predefined based on the equipment's structural characteristics and describes the distribution of faults in physical space. This combination results in a fault location probability map, indicating that the fault occurs in the equipment's rotor bearing with a probability of 0.78.

[0140] This method also uses a reinforcement learning value function to dynamically update the expert experience knowledge base. After each fault diagnosis, the diagnostic accuracy is calculated based on the actual fault confirmations, which serves as a reinforcement learning reward signal. For this diagnosis, the actual fault confirmations matched the diagnostic results 92% of the time, generating a positive reward of 0.92. Expert experience knowledge uses a learning rate of 0.3 to balance historical experience with the reinforcement learning value function, ensuring that the knowledge base maintains 70% historical stability while absorbing 30% of new experience. Based on this updated expert experience, the system fine-tunes the fault location probability mapping, increasing the bearing fault location probability to 0.82 and reducing the probabilities of other locations accordingly.

[0141] A fault diagnosis report was generated, including the fault type (bearing fault type A1), fault location (the front rotor bearing), fault probability (82%), fault cause analysis (due to insufficient bearing lubrication), and repair recommendations (replace the bearing and inspect the lubrication system). Maintenance personnel successfully corrected the fault according to the report's instructions, validating the effectiveness and accuracy of this method.

[0142] Figure 5 This is a bar chart showing the performance comparison and analysis of the cognitive enhancement fault diagnosis method according to an embodiment of the present invention:

[0143] The chart compares and analyzes the recognition accuracy performance of three different analysis methods in three test scenarios. Specifically, in terms of transformer internal fault identification, the traditional analysis method had an accuracy of 64.5%, the Euclidean distance method increased to 76.3%, and the cognitive enhancement method achieved the highest accuracy of 91.8%. In the coil group status assessment scenario, the accuracy of the traditional analysis method was 68.2%, the Euclidean distance method increased to 78.6%, and the cognitive enhancement method achieved an excellent performance of 89.5%. In the abnormal resonant frequency detection test, the accuracy of the traditional analysis method was 71.9%, the Euclidean distance method significantly increased to 82.2%, and the cognitive enhancement method achieved the best result of 96.2%. From the overall trend, the three methods showed an accuracy ranking of cognitive enhancement method > Euclidean distance method > traditional analysis method in all test scenarios. The cognitive enhancement method has a significant performance advantage over the other two methods, with an average improvement of more than 20 percentage points, demonstrating strong analytical capabilities.

[0144] In an optional embodiment, calculating a fault probability distribution based on the optimized cognitive weight, adjusting the discrimination degree of fault types by using a discriminant sensitivity parameter, and combining the fault probability distribution with a fault space distribution function to obtain a fault location probability map includes:

[0145] Calculating the fault probability distribution based on the optimized cognitive weights, and adjusting the discrimination degree of the fault types by using a discrimination sensitivity parameter, wherein the discrimination sensitivity parameter is adaptively adjusted according to the degree of difference in the fault characteristics;

[0146] The fault probability distribution and the fault spatial distribution function are weightedly combined to obtain a fault location probability map, wherein the fault spatial distribution function describes the spatial distribution characteristics of the fault type, and the fault location probability map reflects the spatial probability distribution of the fault location.

[0147] Obtain optimized cognitive weights. These weights are typically optimized based on historical fault data and expert experience, reflecting the importance of different fault features. For example, for a power equipment monitoring system, the cognitive weight for temperature anomalies is 0.35, the cognitive weight for vibration anomalies is 0.28, the cognitive weight for sound anomalies is 0.22, and the cognitive weight for current anomalies is 0.15. These weights reflect the relative importance of each feature for fault diagnosis.

[0148] Based on these optimized cognitive weights, the system calculates a fault probability distribution. This distribution represents the probability of various fault types occurring given given sample characteristics. Specifically, the system calculates a probability value, Pk, for each fault type, k. For example, for a set of collected equipment operating data, the system calculates a bearing failure probability of 0.62, a gear failure probability of 0.25, a lubrication system failure probability of 0.08, and other failure probabilities of 0.05. These probability values ​​are calculated by fully considering the cognitive weights of each fault characteristic.

[0149] When calculating the fault probability distribution, the system introduces a discriminant sensitivity parameter to adjust the degree of fault type differentiation. This parameter adaptively adjusts based on the degree of difference in fault characteristics, increasing the probability distribution gap between significantly different fault types, thereby improving fault diagnosis accuracy. The discriminant sensitivity parameter can be dynamically adjusted based on the variance or entropy of the input features. In practice, the system can set a base sensitivity value, such as 1.5, and then adjust it based on the degree of difference in fault characteristics. When the detected characteristics differ significantly, the sensitivity parameter is increased to 2.2; when the difference is small, the sensitivity parameter is reduced to 1.2.

[0150] Taking equipment vibration monitoring as an example, when the vibration spectra at different locations show significant differences (e.g., a frequency peak difference exceeding 30%), the system increases the sensitivity parameter to 2.0. When the vibration spectrum differences are small (e.g., a frequency peak difference less than 10%), the sensitivity parameter is reduced to 1.3. This adaptive adjustment ensures reasonable fault type differentiation regardless of the presence of varying fault characteristics.

[0151] With the adjusted fault probability distribution, the system needs to further consider the spatial distribution characteristics of faults. The system introduces a fault spatial distribution function, which describes the distribution characteristics of various fault types within the equipment space. For example, in a large industrial equipment, bearing failures often occur at the connection between rotating parts, gear failures are mainly concentrated in the transmission case, and lubrication system failures are distributed throughout the lubrication pipeline network.

[0152] The fault spatial distribution function can be derived from historical fault data statistics, representing the relationship between spatial location and fault probability. For example, for 10 key monitoring points on a piece of equipment, the system determines that the probability of a bearing fault occurring at location 1 is 0.35, the probability at location 2 is 0.28, and so on. This spatial distribution characteristic is crucial for accurately locating the fault location.

[0153] The fault probability distribution and the fault spatial distribution function are weighted together to generate a fault location probability map. In this step, the system assigns appropriate weights to the fault probability distribution and spatial distribution function, such as 0.6 and 0.4, and then obtains the final fault location probability map through weighted summation.

[0154] For example, let's assume the system calculates a bearing failure probability of 0.62 for a piece of industrial equipment, and the spatial distribution probability of this fault at location 3 is 0.45. The joint probability of a bearing failure at location 3 is 0.62 × 0.6 + 0.45 × 0.4 = 0.552. The system performs this calculation for all combinations of fault types and locations, ultimately generating a comprehensive fault location probability map.

[0155] This fault location probability map intuitively reflects the spatial probability distribution of fault locations, providing maintenance personnel with accurate fault location information. For example, the map shows that device location 3 has a 55.2% probability of failure, location 7 has a 23.8% probability of failure, and other locations have lower probability of failure. Maintenance personnel can use this information to prioritize high-risk locations, improving maintenance efficiency.

[0156] This method has been validated in a large manufacturing company's predictive maintenance system. This approach has increased the system's fault location accuracy from 78% to 91%, and reduced average fault diagnosis time from 4 hours to 1.5 hours, significantly improving maintenance efficiency and reducing equipment downtime.

[0157] A second aspect of an embodiment of the present invention provides an automated measurement and analysis system for reactor impedance characteristics, comprising:

[0158] The first unit is configured to collect voltage signals and current signals through a sampling device of the reactor, and calculate original impedance frequency characteristic data of the reactor based on the voltage signals and the current signals;

[0159] The second unit is configured to perform frequency domain decomposition on the original impedance frequency characteristic data to obtain a fundamental component and a harmonic component, perform Fourier transform on the fundamental component and the harmonic component to obtain spectrum data; input the spectrum data into a preset neural network compensation model to obtain a compensation coefficient, multiply the compensation coefficient by the spectrum data to obtain compensated impedance frequency characteristic data, and the neural network compensation model uses a long short-term memory network structure to perform time series feature extraction and dynamic compensation on the spectrum data;

[0160] A third unit is configured to establish an impedance-frequency characteristic curve of the reactor based on the compensated impedance-frequency characteristic data, and calculate real-time characteristic data of the impedance-frequency characteristic curve at different frequency points; continuously collect real-time monitoring data of the reactor under different operating conditions through the sampling device, calculate a real-time impedance-frequency characteristic based on the real-time monitoring data, and compare the real-time impedance-frequency characteristic with the real-time characteristic data to obtain a characteristic deviation;

[0161] The fourth unit is used to construct a fault feature vector based on the feature deviation, establish a fault discrimination criterion by calculating the Euclidean distance and Mahalanobis distance of the fault feature vector, determine the fault type and fault location according to the fault discrimination criterion, and generate a fault diagnosis report.

[0162] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0163] processor;

[0164] a memory for storing processor-executable instructions;

[0165] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0166] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0167] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The method for automatically measuring and analyzing the impedance characteristics of a reactor is characterized by: include: The sampling device of the reactor collects voltage signals and current signals, and calculates original impedance frequency characteristic data of the reactor according to the voltage signals and the current signals; Decomposing the original impedance frequency characteristic data in the frequency domain to obtain a fundamental component and a harmonic component, and performing Fourier transform on the fundamental component and the harmonic component to obtain spectrum data; inputting the spectrum data into a preset neural network compensation model to obtain a compensation coefficient, and multiplying the compensation coefficient by the spectrum data to obtain compensated impedance frequency characteristic data, wherein the neural network compensation model uses a long short-term memory network structure to extract time series features and dynamically compensate the spectrum data; establishing an impedance-frequency characteristic curve of the reactor based on the compensated impedance-frequency characteristic data, and calculating real-time characteristic data of the impedance-frequency characteristic curve at different frequency points; continuously collecting real-time monitoring data of the reactor under different operating conditions by the sampling device, calculating a real-time impedance-frequency characteristic based on the real-time monitoring data, and comparing the real-time impedance-frequency characteristic with the real-time characteristic data to obtain a characteristic deviation; A fault feature vector is constructed according to the feature deviation, a fault discrimination criterion is established by calculating the Euclidean distance and the Mahalanobis distance of the fault feature vector, the fault type and the fault location are determined according to the fault discrimination criterion, and a fault diagnosis report is generated.

2. The method according to claim 1, characterized in that Decomposing the original impedance frequency characteristic data in the frequency domain to obtain fundamental components and harmonic components, and performing Fourier transform on the fundamental components and the harmonic components to obtain spectrum data include: Performing empirical mode decomposition on the original impedance frequency characteristic data to obtain an eigenmode function and a residual term, identifying an eigenmode function index set within a fundamental frequency range and an eigenmode function index set within a harmonic frequency range based on the frequency characteristics of the eigenmode function, and superimposing the eigenmode function index set within the fundamental frequency range and the eigenmode function index set within the harmonic frequency range to obtain a fundamental component and a harmonic component; Performing short-time Fourier transform on the fundamental component and the harmonic component to obtain a time-frequency spectrum, wherein the short-time Fourier transform adopts a variable-length Kaiser window function, and the window length and shape parameters of the variable-length Kaiser window function are adaptively adjusted according to the non-stationary characteristics of the signal; The time-frequency spectrum is subjected to feature enhancement processing, and the feature enhancement processing includes spectrum smoothing and noise suppression. The time-frequency spectrum is convolved in the frequency dimension by a smoothing kernel function to obtain a smoothed spectrum. The smoothed spectrum is subjected to wavelet transformation and the noise component is removed by a soft threshold function to obtain enhanced spectrum data.

3. The method according to claim 1, characterized in that Establishing an impedance frequency characteristic curve of the reactor according to the compensated impedance frequency characteristic data, and calculating real-time characteristic data of the impedance frequency characteristic curve at different frequency points includes: An immune network model is constructed based on the compensated impedance-frequency characteristic data, and an impedance characteristic curve is constructed according to the antibody memory unit of the immune network model. The impedance characteristic curve is amplified and mutated by the antibody memory unit using a clonal selection algorithm, wherein the clonal selection algorithm includes two sub-steps: affinity calculation and clonal amplification. The clonal selection algorithm is used to optimize the fitting accuracy of the impedance characteristic curve in each frequency range; Calculating the first-order derivative and the second-order derivative of the impedance amplitude at different frequency points of the impedance characteristic curve, re-inputting the first-order derivative and the second-order derivative into the immune network model, and updating the impedance change rate information stored in the antibody memory unit; The impedance distortion degree of each frequency interval is calculated based on the updated impedance change rate information. The impedance distortion degree is used to determine the nonlinear change of the impedance characteristic curve through the clone selection algorithm. When the impedance distortion degree exceeds a preset distortion threshold, the clone selection algorithm triggers a high-frequency mutation operation to optimize the real-time characteristic data of the corresponding frequency interval by generating a new antibody population.

4. The method according to claim 3, characterized in that Constructing an immune network model based on the compensated impedance-frequency characteristic data, constructing an impedance characteristic curve based on the antibody memory unit of the immune network model, and amplifying and mutating the antibody memory unit using a clonal selection algorithm on the impedance characteristic curve includes: The immune network model includes multiple antibody units, each of which corresponds to impedance-frequency characteristic data at a frequency point. The affinity between the antibody units is calculated using a Gaussian kernel function, and the affinity represents the similarity between the impedance-frequency characteristic data at adjacent frequency points. Calculating the network activation of the antibody unit based on the affinity, wherein the network activation is obtained by weighted accumulation of the affinity, and the weight coefficient is determined by the antibody memory factor, and the impedance-frequency characteristic data corresponding to the antibody unit having the network activation higher than a preset activation threshold is preferentially stored in the antibody memory unit; Cloning and amplifying the impedance-frequency characteristic data stored in the antibody memory unit, wherein the number of clones is proportional to the network activation degree, and the new antibody units generated by cloning inherit the original impedance-frequency characteristic data. An affinity maturation operation is performed on the new antibody units, wherein the affinity maturation operation perturbs the impedance-frequency characteristic data through Gaussian random mutation, and the magnitude of the mutation is inversely proportional to the network activation degree; The mutated new antibody unit is re-input into the immune network model to calculate the updated network activation degree, the antibody memory units are screened according to the updated network activation degree, and the antibody memory units whose network activation degree is higher than the preset activation degree threshold are retained to construct an impedance characteristic curve.

5. The method according to claim 1, wherein Calculating a real-time impedance frequency characteristic according to the real-time monitoring data, and comparing the real-time impedance frequency characteristic with the real-time characteristic data to obtain a characteristic deviation includes: Performing recursive least squares estimation on the real-time monitoring data, including: constructing an observation equation including voltage observation values, current observation values, and initial impedance parameters, and calculating a prediction error based on the observation equation, wherein the prediction error is an estimated deviation between the real-time monitoring data and historical impedance parameters; Calculating a gain matrix based on the prediction error, wherein the gain matrix is ​​used to adjust the parameter update step size, multiplying the prediction error by the gain matrix to obtain a parameter update amount, and adding the parameter update amount to the historical impedance parameter to obtain the impedance parameter at the current moment; Performing Kalman filtering on the impedance parameter to calculate a state prediction value of the impedance parameter, and performing smoothing based on a deviation between the state prediction value and the real-time monitoring data to obtain a filtered impedance state estimation value; The real-time impedance frequency characteristic is calculated based on the impedance state estimation value, and the real-time impedance frequency characteristic includes an amplitude characteristic, a phase characteristic, and a quality factor; the real-time impedance frequency characteristic is compared with the real-time characteristic data to obtain a final characteristic deviation.

6. The method according to claim 1, characterized in that Constructing a fault feature vector based on the feature deviation, establishing a fault discrimination criterion by calculating the Euclidean distance and the Mahalanobis distance of the fault feature vector, determining the fault type and the fault location based on the fault discrimination criterion, and generating a fault diagnosis report includes: Fusing the fault feature vector with the expert experience feature vector, weighting the fault feature vector by a cognitive weight matrix to obtain an experience fusion coefficient; weighting the expert experience feature vector by the experience fusion coefficient to obtain a cognitive enhancement feature vector; Calculating the Euclidean distance and Mahalanobis distance of the cognitive enhancement feature vector, wherein the Euclidean distance represents the direct distance between feature vectors and the Mahalanobis distance considers the covariance relationship of feature distribution, and constructing a fault discrimination criterion based on the Euclidean distance and the Mahalanobis distance; performing cognitive weight optimization on the fault discrimination criterion, inputting the Euclidean distance and the Mahalanobis distance into a loss function, introducing cognitive rule constraints, and obtaining an optimized cognitive weight by minimizing a weighted sum of the loss function and the cognitive rule constraints; Calculating a fault probability distribution based on the optimized cognitive weights, adjusting the discrimination degree of fault types by using a discriminant sensitivity parameter, and combining the fault probability distribution with a fault space distribution function to obtain a fault location probability map; Expert experience knowledge is updated using a reinforcement learning value function, the expert experience knowledge is weighed between historical experience and the reinforcement learning value function through a learning rate, the fault location probability map is optimized based on the updated expert experience knowledge, and a fault diagnosis report is generated.

7. The method according to claim 6, characterized in that Calculating the fault probability distribution based on the optimized cognitive weight, adjusting the discrimination degree of the fault type by using a discriminant sensitivity parameter, and combining the fault probability distribution with the fault space distribution function to obtain a fault location probability map includes: Calculating the fault probability distribution based on the optimized cognitive weights, and adjusting the discrimination degree of the fault types by using a discrimination sensitivity parameter, wherein the discrimination sensitivity parameter is adaptively adjusted according to the degree of difference in the fault characteristics; The fault probability distribution and the fault spatial distribution function are weightedly combined to obtain a fault location probability map, wherein the fault spatial distribution function describes the spatial distribution characteristics of the fault type, and the fault location probability map reflects the spatial probability distribution of the fault location.

8. A system for automatically measuring and analyzing the impedance characteristics of a reactor, for implementing the method according to any one of claims 1 to 7, characterized in that: include: The first unit is configured to collect voltage signals and current signals through a sampling device of the reactor, and calculate original impedance frequency characteristic data of the reactor based on the voltage signals and the current signals; The second unit is configured to perform frequency domain decomposition on the original impedance frequency characteristic data to obtain a fundamental component and a harmonic component, and perform Fourier transform on the fundamental component and the harmonic component to obtain spectrum data; Inputting the spectrum data into a preset neural network compensation model to obtain a compensation coefficient, multiplying the compensation coefficient by the spectrum data to obtain compensated impedance frequency characteristic data, wherein the neural network compensation model uses a long short-term memory network structure to extract time series features and dynamically compensate the spectrum data; A third unit is configured to establish an impedance-frequency characteristic curve of the reactor based on the compensated impedance-frequency characteristic data, and calculate real-time characteristic data of the impedance-frequency characteristic curve at different frequency points; continuously collect real-time monitoring data of the reactor under different operating conditions through the sampling device, calculate a real-time impedance-frequency characteristic based on the real-time monitoring data, and compare the real-time impedance-frequency characteristic with the real-time characteristic data to obtain a characteristic deviation; The fourth unit is used to construct a fault feature vector based on the feature deviation, establish a fault discrimination criterion by calculating the Euclidean distance and Mahalanobis distance of the fault feature vector, determine the fault type and fault location according to the fault discrimination criterion, and generate a fault diagnosis report.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images