IoT-based real-time status diagnostic system for current transformers

Abstract

The application discloses a current transformer real-time state diagnosis system based on an internet of things, and relates to the technical field of power equipment state diagnosis.The system comprises a reference construction module, a disturbance excitation module, a modeling identification module, a drive unwinding module, a self-adaptive setting module and a closed-loop control module.The reference construction module establishes a nanosecond time synchronization reference and locks a phase reference baseline during the operation of the secondary winding loop of the current transformer, continuously collects current vector trajectories based on the time synchronization reference and the phase reference baseline, and constructs a discrete phase residual graph.The application can accurately identify and suppress the lag oscillation of the secondary loop of the current transformer through high-precision time synchronization, phase disturbance excitation, sparse modeling identification, drive unwinding and self-adaptive setting control, significantly improves the fault diagnosis accuracy and system stability, and breaks through the limitations of traditional methods in terms of identification sensitivity and response capability.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition diagnosis technology, and more specifically to a real-time condition diagnosis system for current transformers based on the Internet of Things. Background Technology

[0002] The "IoT-based Real-time Status Diagnosis System for Current Transformers" refers to an intelligent diagnostic system that deeply integrates the monitoring of current transformer operation status with IoT technology. This system uses multi-dimensional sensing units on the current transformer to collect real-time operating parameters such as current, voltage, temperature, humidity, partial discharge, and insulation status. Data is remotely transmitted and processed via wireless communication, edge computing, or a cloud platform. Based on this, data analysis and intelligent diagnostic algorithms are used to perform real-time assessment and anomaly detection of the current transformer's health status, thereby enabling early warning of potential defects and dynamic prediction of fault trends. This system not only overcomes the limitations of traditional manual inspections and offline testing, improving the full lifecycle management of current transformers, but also integrates with power operation and maintenance platforms to form a closed loop of online monitoring, intelligent diagnosis, risk warning, and maintenance decision-making, ensuring the operational safety and reliability of the power system.

[0003] Existing technologies have the following shortcomings: In existing technologies, the secondary winding circuit of current transformers typically relies on conventional filtering and threshold discrimination methods to identify current fluctuations during power system operation. However, in dynamic scenarios with frequent sudden increases and decreases in power load, the secondary winding circuit is subject to the coupling effect of inherent parasitic parameters such as parasitic capacitance, parasitic inductance, and stray resistance, occasionally triggering transient hysteresis. This type of oscillation has an extremely low probability of occurrence and a very short duration; its waveform characteristics often fall between normal fluctuations and true fault signals, making it difficult for traditional filtering and diagnostic methods to identify and eliminate it in a timely manner, leading to misjudgments as normal operating signals. Once such misjudgments occur, the protection device will execute actions based on the incorrect current input, causing the protection settings to deviate and triggering maloperation, thereby disconnecting critical transmission lines and disrupting the continuity and stability of the power grid operation.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time status diagnostic system for current transformers based on the Internet of Things (IoT) to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time status diagnosis system for current transformers based on the Internet of Things, comprising a reference construction module, a disturbance excitation module, a modeling and identification module, a drive unwrapping module, an adaptive tuning module, and a closed-loop control module:

[0007] The reference construction module establishes a nanosecond-level time synchronization reference and locks the phase reference baseline during the operation of the secondary winding circuit of the current transformer. Based on the time synchronization reference and the phase reference baseline, it continuously collects the current vector trajectory and constructs a discrete phase residual map.

[0008] The perturbation excitation module, based on the discrete phase residual map, injects micro-amplitude bipolar phase perturbations within a set timing safety window to induce the response behavior of parasitic capacitance, parasitic inductance and stray resistance, extracts amplitude and phase hysteresis features, and generates hysteresis fingerprint data.

[0009] The modeling and identification module constructs a parameterized circuit model based on the hysteresis fingerprint data and outputs dynamic estimates of parasitic capacitance, parasitic inductance and stray resistance based on the sparse identification method.

[0010] The driving unwrapping module processes the dynamic estimate and the discrete phase residual map together, and uses a time-frequency causal unwrapping algorithm to separate the actual running drive and the parasitic coupling drive, and extracts the oscillation root cause sequence.

[0011] The adaptive tuning module constructs a hazard reference model based on the oscillation root cause sequence and links it with the setting window of the protection device to dynamically adjust the convergence threshold and time constant, thereby achieving adaptive tuning of the protection strategy.

[0012] The closed-loop control module, based on adaptive tuning, implements anti-phase dissipation and topology reconstruction control. By projecting an anti-phase damped electric field, switching the sampling path phase, and triggering pulse freezing and delayed release, it completes differential homeomorphic adjustment and achieves closed-loop stable control of the measurement link.

[0013] Preferably, the steps for constructing the discrete phase residual map are as follows:

[0014] A temperature-compensated crystal oscillator with high stability is installed at the measurement node of the secondary winding circuit of the current transformer as a local clock source. A unified time scale between measurement nodes is achieved by using a precision time synchronization protocol and combining it with an external satellite time signal.

[0015] Based on the measurement nodes that have completed time synchronization, current signals are continuously acquired at high frequency, the sampled signals are converted into complex vector form and the corresponding instantaneous phase values ​​are calculated;

[0016] A fixed-length rolling window method is used to extract the vector phase angle change sequence, and a reference baseline reflecting the phase evolution trend under normal operating conditions is constructed by using the minimum error fitting method.

[0017] The measured phase is compared point by point with the reference baseline to obtain the phase residual sequence. After normalization, it is mapped into a two-dimensional map. A weight function based on frequency density is introduced to enhance the expressiveness of the map, forming a discrete phase residual map that supports perturbation identification.

[0018] The preferred method for generating delayed fingerprint data is as follows:

[0019] Based on the constructed discrete phase residual map, the time segment with small phase residual amplitude, stable phase derivative change and spectral principal components concentrated in the power frequency and its low harmonic range is selected as the phase perturbation injection window.

[0020] Within the selected injection window, a multi-frequency superimposed micro-amplitude bipolar phase disturbance signal is generated and injected into the secondary winding circuit through a coupling path with low impedance.

[0021] During the disturbance injection process, the current vector signal is acquired at a high frequency, and the phase delay, amplitude attenuation, response recovery time and frequency drift characteristics contained in multiple frequency components are extracted.

[0022] The feature data extracted in each perturbation cycle are used to construct a response vector in a uniform format, which is then combined to form a hysteresis fingerprint vector set, serving as the input data set for the subsequent modeling and recognition stages.

[0023] Preferably, the three-frequency superimposed micro-amplitude bipolar phase disturbance signal is composed of a fundamental frequency signal and multiple harmonic signals. The phase of the disturbance signal is bidirectionally modulated relative to the main current signal. The injection duration is one complete power frequency cycle, and the impedance of the injection path is less than the minimum equivalent impedance of the disturbed circuit.

[0024] Preferably, the dynamic estimate generation steps are as follows:

[0025] The maximum phase lag time, amplitude attenuation rate, phase recovery time constant, and frequency drift amplitude in the hysteresis response characteristics are constructed into a standardized feature matrix and arranged in timestamp order;

[0026] An equivalent circuit model containing series resistance, series inductance, and parallel capacitance is constructed based on a standardized feature matrix, and the initial and boundary values ​​of parasitic capacitance, inductance, and resistance are limited.

[0027] Adopting based on The norm-constrained least squares sparse regression algorithm is used to identify sparsity in the equivalent circuit model and removes high residual parameters through ten-fold cross-validation.

[0028] By appending corresponding timestamps to the values ​​of parasitic capacitance, parasitic inductance, and stray resistance identified in the output, a dynamic estimation sequence is constructed to describe the temporal evolution characteristics of the parasitic parameters.

[0029] Preferably, the steps for extracting the oscillatory root cause sequence are as follows:

[0030] The dynamic estimates of parasitic capacitance, parasitic inductance, and stray resistance are time-aligned with the phase response data of the corresponding perturbation period in the discrete phase residual diagram to construct a joint data matrix containing time index, phase value, phase derivative, parasitic capacitance, parasitic inductance, and stray resistance.

[0031] Wavelet transform and Granger causality analysis are performed on the joint data matrix to separate the actual driving component from the parasitic coupling driving component, and to determine whether the parasitic parameter is the dominant causal term of the phase response change.

[0032] Identify phase response fragments with abrupt change characteristics from parasitic driving responses, extract the core dynamic features of these fragments, and generate an oscillating root cause sequence containing nonlinear driving factors and evolutionary trends.

[0033] Preferably, the steps for dynamically adjusting the convergence threshold and time constant to achieve adaptive tuning of the protection strategy are as follows:

[0034] Based on the oscillation root cause sequence, a reference model for the harm of hysteresis is constructed. Historical root cause events are classified and graded according to the perturbation intensity and the dynamic change rate of parasitic parameters. An event template library is established to match the current perturbation event.

[0035] The system compares the real-time oscillation root cause information with the event template library for similarity, outputs a disturbance level score, and performs linkage adjustment on the protection action current threshold and response delay time based on the score results.

[0036] After completing the linkage adjustment, the updated results are written into the protection strategy control process, and the reference model is continuously trained based on historical disturbance samples to achieve adaptive closed-loop tuning and update of protection parameters.

[0037] Preferably, when the disturbance level score reaches the high-risk level, the protection action current threshold is increased, the response delay time is shortened, and after adjustment, the updated protection parameters are compared with the relay protection action curve to verify whether they are within the allowable range. If they exceed the deviation range, parameter correction is automatically performed.

[0038] Preferably, based on adaptive tuning, anti-phase dissipation and topology reconstruction control are implemented. By projecting an anti-phase damped electric field, switching the sampling path phase, and triggering pulse freeze and delayed release, the differential homeomorphism adjustment steps are completed as follows:

[0039] After completing the adaptive tuning parameter update, an anti-phase intervention signal is generated and injected through a low-coupling electrode to form an asymmetric damping electric field, so as to suppress the time-domain interference of the source of the hysteresis signal.

[0040] The measurement path reconstruction operation is performed. The main interference path is determined based on the oscillation amplitude and phase response characteristics of the three-channel sampling data, and the path switching is achieved by switching channels with low latency hardware.

[0041] After path switching, a pulse freeze mechanism is triggered. When the change in the sampling derivative exceeds a set threshold, data flow is paused, and an interpolation algorithm is used to restore data continuity after the freeze ends.

[0042] Perform differential homeomorphic structure adjustment, judge the path switching result by comparing the trend of sampled derivatives before and after the switching, and trigger path rollback and re-intervention injection when the deviation exceeds the limit;

[0043] The operation results are integrated into a closed-loop control process, and the oscillation suppression effect is evaluated and fed back to optimize the matching accuracy and response stability of the control parameters in the next round.

[0044] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0045] This invention significantly improves sampling accuracy and identification sensitivity by constructing nanosecond-level time synchronization and phase reference baselines; it actively induces and quantifies parasitic coupling effects by injecting micro-amplitude bipolar phase disturbances and extracting hysteresis features; it outputs dynamic parameters based on sparse identification methods to enhance the fitting ability and interpretability of circuit models; it separates coupled disturbances from actual operating conditions using time-frequency causal untangling to clarify the root cause of oscillations; and it achieves adaptive adjustment of convergence threshold and time constant through linkage with the setting window of protection devices, ultimately forming a stable closed loop with the aid of anti-phase dissipation and measurement path topology reconstruction mechanisms. This technical solution significantly improves the anomaly identification rate and anti-interference capability of current transformers under extreme operating conditions, ensuring the accuracy of protection actions and the continuity and safety of power system operation. It has comprehensive benefits such as strong engineering applicability, low deployment threshold, and high diagnostic accuracy, breaking through the technical bottlenecks of traditional methods in terms of timeliness, accuracy, and adaptability. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0047] Figure 1 This is a schematic diagram of the module of the Internet of Things-based real-time status diagnosis system for current transformers according to the present invention. Detailed Implementation

[0048] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0049] This invention provides, for example Figure 1 The IoT-based real-time status diagnostic system for current transformers shown includes a reference construction module, a disturbance excitation module, a modeling and identification module, a drive unwrapping module, an adaptive tuning module, and a closed-loop control module.

[0050] The reference construction module establishes a nanosecond-level time synchronization reference during the operation of the secondary winding circuit of the current transformer and locks the phase reference baseline of the secondary winding circuit simultaneously. Based on the time synchronization reference and the phase reference baseline, the current vector trajectory of the secondary winding circuit is continuously acquired to construct a discrete phase residual map that reflects the phase change offset characteristics.

[0051] To ensure accurate identification of minute abnormal disturbances in the secondary winding circuit of a current transformer, a high-time-resolution phase reference baseline needs to be established, and a discrete phase residual map reflecting the changes in the current vector needs to be constructed based on a unified time scale. This process includes the following specific steps:

[0052] A highly stable time synchronization device is installed at the measurement node of the secondary winding circuit of the current transformer. A temperature-compensated crystal oscillator (OCXO) is preferably used as the local clock source, and this oscillator has a frequency stability of no more than [missing information]. The system utilizes the precise time synchronization protocol IEEE 1588-2008 to achieve network-wide time alignment among measurement nodes. This protocol implementation must include synchronization message exchange between master and slave devices, round-trip delay calculation, path symmetry verification, and hardware timestamp functionality to ensure clock deviations between different measurement channels are controlled within ±10 nanoseconds. All sampling devices must support a hardware-level time synchronization interface and perform real-time comparison and offset correction using GPS satellite clock signals to avoid time drift caused by local oscillator aging or environmental changes, thereby ensuring that each sampling moment has a unified reference time scale across the entire network.

[0053] Based on the time-synchronized measurement points, raw waveform data of the current signal in the secondary winding circuit of the current transformer are continuously acquired. A sampling frequency of 200kHz or higher is recommended to fully cover power frequency and high-frequency disturbance components. In the data processing stage, the dominant frequency component of the signal is extracted using Fast Fourier Transform (FFT), and the sampled data is converted into a complex vector form in polar coordinates. The instantaneous amplitude and instantaneous phase of each sampling point are calculated. Subsequently, the vector sequence is segmented using a fixed rolling window (recommended to be 16 power cycles, 20ms per cycle). Within each window, the phase angle difference between all adjacent sampling points is extracted to form a phase change sequence. To reduce the phase jump effect caused by window boundaries, the minimum mean square error (MMSE) fitting method is used to establish a phase fitting baseline. This fitting function takes the vector phase sequence within the window as input and outputs a smooth baseline representing the phase evolution trend under the current normal operating condition. By comparing the deviation between the measured phase trajectory and the fitting result, abnormal phase fluctuations caused by parasitic parameters, electromagnetic interference, or mechanical vibration can be effectively identified, thereby obtaining a stable, continuous, and traceable phase reference curve.

[0054] After establishing the phase baseline, the difference between the measured phase value at each sampling point in the rolling window and the corresponding phase value of the fitted baseline at the same time point is calculated. This difference is used as the phase residual value for that point. All the residual values ​​from the sampling points form a continuous discrete phase residual sequence, constituting the phase residual matrix. To facilitate subsequent perturbation response analysis and pattern recognition, the residual matrix is ​​z-score normalized to a mean of 0 and a standard deviation of 1, eliminating the influence of signal amplitude differences under different operating conditions. Furthermore, second-order envelope analysis is introduced to extract the envelope curve from the normalized residual sequence, used to identify high-frequency spikes, low-frequency gradual changes, and phase jump characteristics near zero crossings. Within each window, the location index and amplitude threshold of the phase abrupt change point are also labeled as shielding conditions during subsequent perturbation injection, avoiding the application of meaningless perturbations to the stable segment.

[0055] To visualize the residual sequence, the normalized discrete phase residual matrix is ​​mapped into a two-dimensional graph. The horizontal axis represents the sampling time index, the vertical axis represents the residual value amplitude, and the graph's color level reflects the absolute magnitude of the phase shift. It is recommended to use the Jet color mapping scheme, highlighting high-residual regions and softening low-residual regions to allow for rapid visual identification of abnormal trend segments and areas of dense disturbance. After graph generation, a weighting function is introduced to weight the residual information from different time periods. This weighting function is set based on the disturbance frequency density function derived from FFT analysis, enhancing the representation of short-term strong disturbance information and suppressing repetitive data interference in long-term stable periods. The resulting phase residual graph serves as a benchmark template for subsequent disturbance injection and parasitic response analysis, possessing time alignment, phase traceability, and sampling consistency. This effectively supports the entire process of state diagnosis based on vector disturbances, providing necessary technical support for the accurate identification and control of current transformer operating states.

[0056] The disturbance excitation module, based on the phase change reference constructed in the discrete phase residual diagram, injects a micro-amplitude bipolar phase disturbance signal into the secondary winding circuit of the current transformer within a set time safety window, inducing the response behavior of the parasitic parameter set composed of parasitic capacitance, parasitic inductance and stray resistance, and collects the amplitude hysteresis characteristics and phase hysteresis characteristics in the response signal to generate hysteresis fingerprint data containing parasitic response characteristic information.

[0057] To ensure the active excitation and accurate capture of the dynamic response characteristics in the secondary winding circuit of a current transformer caused by the combined effects of parasitic capacitance, parasitic inductance, and stray resistance, a micro-amplitude bipolar phase perturbation signal is injected into the existing discrete phase residual map to induce quantifiable parasitic response behavior within a strictly selected timing window, and the data constituting the hysteresis fingerprint is extracted. The specific implementation method is as follows:

[0058] Based on the constructed discrete phase residual map, a continuous sampling segment with a phase residual amplitude within ±0.02 radians, a phase derivative rate of change not exceeding 0.05 radians per millisecond, and a sustained stable time of not less than 200 milliseconds was selected as a candidate window for phase disturbance injection. To ensure that this disturbance segment does not contain any external disturbances or abnormal fluctuation signals, it is necessary to further verify whether its spectral principal components are concentrated only in the power frequency and its low-order harmonic components (not higher than the seventh harmonic), and that more than 95% of the spectral energy should be concentrated in the range of 50Hz to 350Hz. The above verification was completed by Fast Fourier Transform and confirmed by superposition using an energy density ratio screening function based on a periodic root mean square sliding window. The final selected injection window length was set to one complete power cycle, i.e., 20 milliseconds (for a 50Hz power frequency system). The window start time was consistent with the absolute timestamp under the unified time scale in the previous steps to ensure strict consistency and comparability between the disturbance and the measurement data.

[0059] Within the selected disturbance injection time window, a set of micro-amplitude bipolar sinusoidal signals is generated using a precision arbitrary waveform function generator. The fundamental frequency of the disturbance waveform is 50Hz, and its harmonic components include two terms at 150Hz and 250Hz, making the total disturbance waveform a non-harmonic form of superposition of three frequency components. This is used to activate parasitic elements in the circuit with significant differences in frequency response characteristics. The amplitude of this composite signal is controlled within ±1% of the rated current on the secondary side of the current transformer, and the unit amplitude accuracy must reach 0.01A. The phase of the current injection waveform is modulated by ±5° relative to the current main current signal to ensure that, without triggering malfunctions of the protection device, measurable amplitude attenuation, phase drift, and dissipative hysteresis responses are generated for parasitic capacitance, inductance, and resistance, respectively. The disturbance signal is injected into the secondary winding circuit through a programmable current source and a non-inductive resistor in series. The injection path must meet the following requirements: its impedance is less than 1% of the minimum equivalent impedance of the disturbed circuit, ensuring that the disturbance signal can be effectively coupled without causing substantial changes to the circuit topology. The disturbance process requires a high-precision timing controller to control the start and stop signals, which must be perfectly aligned with the timestamp of the sampling device, and the injection duration must be strictly 20 milliseconds.

[0060] Simultaneously with the completion of the disturbance signal injection, the secondary winding circuit current signal at each stage before, during, and after the disturbance is sampled in real time at a sampling rate of no less than 500 kHz. The sampled signal is first transformed into polar coordinates to extract the amplitude and phase information of each sampling point. In the polar coordinate domain, the phase derivative and phase acceleration are calculated using the central difference method to further analyze the inertial hysteresis characteristics in the disturbance response signal. For the response of each frequency component, its main response signal is extracted through a bandpass filter, and the phase response curve is synchronously compared with the disturbance signal. The maximum phase response delay time (in microseconds), the maximum amplitude response attenuation rate (based on the injected amplitude), the time constant required for response recovery (using a first-order regression model to fit the recovery segment data), and the peak frequency deviation amplitude of the response signal (in Hertz) are recorded. All of the above characteristics need to be repeatedly sampled and averaged over 5 independent disturbance cycles to ensure data stability.

[0061] The four types of features extracted in each perturbation injection cycle are respectively constructed into response vectors of length 4. The response vectors of all perturbation cycles are then concatenated in chronological order to form a hysteresis fingerprint vector set. Each vector set element includes: (1) maximum phase delay value, in radians; (2) amplitude response attenuation ratio, in percentage; (3) phase recovery time constant, in milliseconds; (4) main frequency drift amplitude, in Hertz. All vector data are stored in a double-precision floating-point structure. Each type of parameter is presented in the form of average value and standard deviation, which serve as key parameter inputs to characterize the dynamic response capability of parasitic capacitance, parasitic inductance and stray resistance for subsequent modeling and identification stages.

[0062] The modeling and identification module constructs a parameterized circuit model containing parasitic parameters based on the hysteresis fingerprint data. It performs modeling and calculation on the parameterized circuit model based on the sparse identification method and outputs dynamic estimates of parasitic capacitance, parasitic inductance and stray resistance.

[0063] To model the dynamic characteristics of parasitic capacitance, parasitic inductance, and stray resistance in the secondary winding circuit of a current transformer, an equivalent circuit structure with physical constraints needs to be constructed after the hysteresis fingerprint data extraction is completed. Then, sparse identification methods are used for parameter identification and estimation extraction, ultimately obtaining a parameter evolution sequence with temporal continuity and physical traceability. Specifically, the following steps are included:

[0064] After constructing the hysteresis fingerprint, the obtained hysteresis response feature data is structured. This data includes: the maximum phase lag time after perturbation injection (in microseconds), the maximum attenuation rate of the amplitude response relative to the injected signal (expressed as a percentage), the time constant of the phase recovery process (in milliseconds), and the amplitude of the dominant frequency drift in the frequency domain response (in Hertz). For each perturbation cycle, these four indices are constructed into a one-dimensional feature vector with dimension 4. To ensure a uniform scale for the features, all data vectors are standardized, with their mean set to zero and variance normalized to 1. They are then arranged sequentially by sample timestamps to form a feature sequence matrix containing complete time signatures. This feature matrix exhibits good temporal continuity and numerical comparability, and can be used as the input feature driving term in subsequent equivalent circuit modeling.

[0065] Based on the principles of physical model construction, the response behavior of the secondary winding circuit of the current transformer under injected disturbance signals is abstracted as a series-parallel hybrid circuit. The specific structure is as follows: an independent controlled voltage source is used as the external excitation source. Its output terminal is connected in series with a resistor and an inductor, and then in parallel with a capacitor. These three elements represent the equivalent resistance introduced by the cable stray impedance, the equivalent inductance caused by conductor deformation and electromagnetic coupling, and the equivalent parasitic capacitance formed by the spacing between the winding insulation layer and the conductor, respectively. The excitation waveform of the voltage source completely replicates the known disturbance signal, maintaining consistent phase and amplitude characteristics. In the initial modeling stage, the initial value of parasitic capacitance is set to 200 picofarads, parasitic inductance to 2 microhenries, and stray resistance to 10 milliohms. To ensure fitting stability, boundary constraints are introduced: the parasitic capacitance is set between 10 picofarads and 1000 picofarads, the parasitic inductance between 0.1 microhenries and 20 microhenries, and the stray resistance between 1 milliohm and 200 milliohms. The equivalent circuit model will be used as the object of identification for dynamic parameter inversion.

[0066] After defining the circuit structure, a sparse identification method is used to solve for the model parameters. The identification method is based on the core idea of ​​sparse optimization and selects a method based on... The norm-constrained least squares sparse regression algorithm LASSO (Least Absolute Shrinkage and Selection Operator) is used as the solution strategy. This algorithm first uses the hysteresis fingerprint feature sequence as the input feature vector, inputting it into the constructed circuit model to perform a forward calculation on the disturbance response, generating a simulated output waveform. Then, it calculates the point-by-point error between the simulated waveform and the actual acquired response waveform, using this error as the loss function. Next, a sparsity penalty term is introduced, adding the sum of the absolute values ​​of all parameters to be determined as a regularization term to the loss function, aiming to filter out redundant parameters while ensuring fitting accuracy. The parameters are updated iteratively, continuously optimizing the objective function, and finally obtaining a set of optimal parasitic capacitance, inductance, and resistance values. During the identification process, a ten-fold cross-validation mechanism is introduced, dividing the dataset into a training set and a validation set, performing parameter learning and error validation separately to avoid overfitting. If a set of parameters exhibits high residuals or low fit on the validation set, that set of parameters is discarded and reinitialized for the next iteration.

[0067] After identifying and obtaining the optimal parameter set, the parasitic capacitance, inductance, and resistance values ​​are output, and timestamp information for the corresponding perturbation period is appended to them, forming a dynamic parameter sequence. This sequence uses the parasitic parameter values ​​under each perturbation period as a set of time-series samples, forming a complete dynamic evolution trajectory, showing the trend of parasitic capacitance changing over time, the response inertia of parasitic inductance at different perturbation frequencies, and the transient change amplitude of stray resistance when the current changes drastically. All parameter results are stored in double-precision floating-point form, along with their fitting error (root mean square error RMSE) and parameter confidence intervals (95% confidence level). In addition, moving average processing and trend smoothing are performed on the parameter results for all perturbation periods to remove the influence of occasional perturbation errors on parameter estimates, resulting in stable dynamic parameter curves.

[0068] The driving unwrapping module jointly processes the dynamic estimate and the phase change information in the discrete phase residual map, uses the time-frequency causal unwrapping algorithm to separate the actual running drive and the parasitic coupling drive, and extracts the oscillation root cause sequence that characterizes the nonlinear oscillation cause from the drive separation result;

[0069] To identify the root cause of the nonlinear coupled oscillation mechanism composed of parasitic capacitance, parasitic inductance, and stray resistance in the secondary winding circuit of a current transformer, signal fusion and causal deconstruction are required based on dynamic estimates and discrete phase residual diagrams. An algorithm with time-domain and frequency-domain resolution is used to decouple the hybrid drive response and extract the oscillation root cause sequence. Specifically, the steps include:

[0070] Based on the dynamic estimation results of parasitic parameters, numerical sequences of parasitic capacitance, inductance, and stray resistance obtained under each disturbance cycle are extracted to construct a dynamic parameter set with timestamps. Each parameter set contains three sub-items: parasitic capacitance in picofarads, parasitic inductance in microhenries, and stray resistance in milliohms. Synchronously with this parameter set, phase change data within the corresponding disturbance cycle are selected from the previously constructed discrete phase residual map, and the actual operating phase response is extracted at 20 microsecond sampling intervals. The two data sequences are aligned to a time reference using linear interpolation to ensure that there are unique parasitic parameter estimates and measured phase response values ​​at the same time point. These two data sources are jointly input to construct a joint data matrix consisting of six dimensions: time index, phase value, phase derivative, parasitic capacitance, inductance, and resistance, for subsequent signal deconstruction calculations. This data matrix accurately characterizes the dynamic relationship between the current signal phase evolution and parasitic parameter changes at each time point, forming the mathematical description basis for parasitic coupling-driven behavior.

[0071] For the constructed joint data matrix, a method combining wavelet transform and Granger causality analysis is employed to separate the causal drivers of the phase response signal. First, a continuous wavelet transform is applied to the measured phase response signal, using the complex Morlet mother wavelet as the basis function to obtain the time-frequency distribution spectrum of the signal in the 0–1 kHz frequency range. The local energy density and phase change rate are calculated for each frequency component to identify the time-domain locations of high-frequency abrupt change regions and low-frequency steady-state segments. Next, correlation analysis is performed between these time-frequency features and the dynamically estimated parasitic parameter sequence, using the Pearson correlation coefficient to calculate the linear dependence between parasitic parameter changes and the phase response frequency distribution. Within regions of significant correlation, a Granger causal regression model is further employed to construct a regression equation with parasitic parameters as independent variables and phase response as the dependent variable. The regression window width is set to 2 milliseconds to satisfy physical time delay characteristics. The model output includes indicators such as causality coefficient, residual variance, and significance level to determine the existence of causal driving paths. If the statistical criterion of significance less than 0.05 is met, it is determined that there is a parasitic parameter-driven response within that time period, and the phase response of that period is extracted as the parasitic coupling driving component. The remaining response sequences that fail the causality test are classified as actual operation driving components.

[0072] After obtaining the main driving and parasitic driving components, the parasitic driving part is further structured to extract the core causes of oscillation behavior. First, all first and second-order abrupt changes in the phase derivative in the parasitic driving response are identified. These points typically correspond to nonlinear dynamic excitation caused by sudden charging of parasitic capacitance, reverse gain of parasitic inductance, or local instability of stray resistance. By defining thresholds for derivative abrupt changes, a jump of more than 0.1 radians per millisecond in the first-order phase derivative or more than 0.2 radians per millisecond squared in the absolute value of the second-order derivative is set as the abrupt change criterion, locating the start and end points of each abrupt change event. Within each abrupt change segment, the following parameters are extracted: phase change amplitude, phase response fallback time, oscillation period, peak amplitude, average frequency drift, and the maximum rate of change of parasitic parameters within that time period. These six types of information are arranged chronologically to form an oscillation root cause sequence, with each element being a structured data unit including a timestamp, abrupt change type, corresponding parasitic parameter item, derivative eigenvalue, and phase response index. The resulting root cause sequence can fully track the physical causes, response intensity, and evolution trend of each nonlinear oscillation, providing quantitative basis for the subsequent construction of hysteresis hazard models and the formulation of dynamic prevention and control strategies.

[0073] The adaptive setting module constructs a hysteresis hazard reference model based on the oscillation root cause sequence, and performs linkage analysis with the setting threshold window in the protection device to dynamically adjust the convergence threshold and time constant of the protection action, thereby realizing the adaptive setting update of the current transformer protection strategy.

[0074] To ensure that current transformers possess accurate response and dynamic self-adjustment capabilities when facing hysteretic oscillations, a hazard reference model with time-series identification functionality needs to be constructed based on the oscillation root cause sequence. This model must then be linked to the setting thresholds in existing protection devices to achieve adaptive updates of the thresholds and delay parameters. This process includes the following steps:

[0075] A reference model for the hazards of hysteresis is constructed based on the extracted root cause sequence of oscillations. Each data unit in the root cause sequence includes the occurrence time, corresponding disturbance type, disturbance intensity index, derivative mutation value, parasitic parameter change rate, frequency drift amplitude, phase shift range, and disturbance duration. To improve the model's recognition accuracy and adaptability, historical root cause events are first classified and archived, and an index system based on event levels is established. The classification criteria use a joint grading based on two dimensions: disturbance intensity and the dynamic change rate of parasitic parameters. Specifically: if the phase derivative change rate is greater than 0.25 radians per millisecond and the parasitic inductance change amplitude exceeds 1 microhenry per 10 milliseconds, it is classified as a Level 1 high-risk event; if only one is met, it is classified as a Level 2 medium-risk event; if neither reaches the threshold, it is classified as a Level 3 low-risk event. Each grading result corresponds to an event behavior template, and the disturbance feature fields, evolution trajectory trend, and waveform change pattern are stored in vector form. All templates are uniformly organized to form a reference library for the hazards of hysteresis. In actual operation, the real-time input of oscillation root cause information is compared with the templates in the database. Using both Euclidean distance and cosine similarity as dual metrics, the historical template that best matches the current event is selected as the reference benchmark, and a disturbance level score is output. This score guides the dynamic adjustment of subsequent protection parameters. The score ranges from 0 to 100, representing the disturbance severity from extremely low to extremely high.

[0076] After obtaining the disturbance level score, the score result is linked to the current protection device setting parameters of the current transformer for adjustment. Traditional protection parameter setting methods rely on manual experience for static settings, typically fixing the setting value at 1.25 times the rated current as the operating current threshold and 30 milliseconds as the delay response time, which cannot adapt to rapid changes in disturbance behavior. Therefore, in this embodiment, a segmented response logic for the disturbance score is set: when the score is between 0 and 40, the protection operating current threshold remains unchanged, and the delay time is increased to 35 milliseconds to accommodate minor disturbances; when the score is between 41 and 75, the current operating threshold is raised to 1.35 times the rated current, and the delay time is shortened to 25 milliseconds to prevent moderate disturbances from being misinterpreted as normal signals after resonance amplification; when the score is above 75, the operating current threshold is immediately raised to 1.45 times the rated current, and the response delay is set to 15 milliseconds to improve the ability to quickly identify strong disturbances. To ensure the logical integrity of protection action adjustments, a comparison analysis with the relay protection action curve is performed after each parameter adjustment to verify whether the modified parameters are still within the allowable range. If the allowable deviation is exceeded, a secondary correction is automatically performed to ensure that parameter modifications do not cause protection boundary violations. All adjustment processes are completed within 200 microseconds after disturbance identification, using a local computing controller to perform online adjustments and maintaining synchronization with the master control time.

[0077] After completing the parameter linkage adjustment, the dynamically updated results are written into the protection strategy control flow, forming an adaptive closed-loop tuning mechanism based on disturbance identification. Each actual disturbance event and its corresponding adjusted parameters will be used as a training sample and automatically entered into the historical disturbance response database. When similar disturbance waveform characteristics reappear in the future, the system will prioritize calling the adjustment scheme corresponding to the historical sample to achieve rapid response and accurate matching. If a protection malfunction occurs in a disturbance event, the system will mark the event as an "erroneous response" and analyze whether its root cause stems from template mismatch or parameter adjustment error. If the same type of "erroneous response" occurs more than twice, the system will trigger the retraining mechanism of the reference model. In the retraining process, erroneous samples and similar samples are first selected to reconstruct the disturbance behavior pattern vector. Then, the original template weights are adjusted based on the new samples, or new template categories are created to expand the original template library. At the same time, to address the problem of overprotection or slow response caused by misjudgment of disturbance level, the piecewise function boundary in the threshold adjustment function will be revised to give it a more refined adaptability to certain types of special disturbances. The model correction and parameter update process can be completed in real time without affecting normal protection operations, and it has the capabilities of online learning, self-evolution, and adaptive updating. Through this closed-loop strategy update method, not only can the current transformer have stable and reliable protection capabilities when facing complex disturbance scenarios, but it can also gradually enhance the intelligence of the response as actual operating experience accumulates.

[0078] The closed-loop control module, based on adaptive tuning and updating, combines the anti-phase dissipation control mechanism and the measurement path topology reconstruction mechanism to implement differential homeomorphic structure adjustment by projecting an anti-phase damping electric field, switching the phase of the multi-channel sampling path, and triggering pulse signal freezing and delayed release. This dynamically suppresses the hysteresis oscillation of the secondary winding circuit and achieves closed-loop stable control of the measurement link of the current transformer secondary winding circuit.

[0079] To achieve accurate suppression and stable measurement of the secondary winding circuit of the current transformer under lagging oscillation disturbance, after completing the adaptive setting parameter update, it is necessary to sequentially execute the following operations: anti-phase dissipation intervention, measurement path switching, pulse freeze release, and structural continuity adjustment. This dynamically controls the oscillation propagation path at the physical path level, specifically including the following steps:

[0080] After completing the adaptive tuning parameter update, the anti-phase dissipative intervention operation is immediately initiated. Based on the oscillation root cause characteristics extracted in the previous steps, this operation selects the time period and corresponding frequency band of the high-frequency nonlinear disturbance, generating an anti-phase intervention signal with the opposite phase, consistent frequency, and 25% amplitude of the original signal. This signal is generated by an independent voltage source, amplified by a high-frequency broadband amplifier, and then injected into the external magnetic field region of the current transformer's secondary winding circuit through a low-coupling electrode embedded in an insulating medium, constructing an asymmetric damped electric field with reversed voltage polarity and time-phase synchronization. Once formed, this damped electric field spatially superimposes with the target disturbance waveform in a time-symmetric structure, achieving local temporal interference and blocking its continuous propagation chain in the secondary circuit, thereby suppressing its amplification and cascading risks at the source. To ensure synchronization accuracy, the trigger delay of the anti-phase intervention signal must not exceed 20 microseconds, and the voltage deviation must be controlled within ±1%.

[0081] After the inverse dissipation process is initiated, a structural reconstruction operation of the measurement path is performed based on the response differences of the disturbance waveform along the propagation path in different measurement channels. The original measurement architecture uses three equally spaced sampling channels, distributed around different sampling contacts around the secondary winding structure. By analyzing the differences in oscillation amplitude and phase response time difference among the three channels within the same time period, the main propagation path is determined. If a channel experiences more than three consecutive amplitude jumps within 0.5 milliseconds and the amplitude change rate exceeds 20% of the original signal, this path is identified as the main interference path. This path is then disconnected via a low-latency hardware switch, and the sampling task is transferred to the channel with the smallest amplitude and the most stable phase response. To avoid signal misalignment during switching, a transition buffer is performed between two channels for each switch. The buffer time is 200 sampling points for linear weighted fusion to ensure signal sequence continuity and structural smoothness.

[0082] After the path switching is completed, a pulse freeze control operation is performed based on the degree of abrupt change in the current channel's sampled waveform. Specifically, after the new path is connected, the rate of change of the first derivative of the sampled waveform is monitored in real time. When the absolute value of the derivative exceeds the threshold of 0.3 radians per millisecond, the current sampling buffer is immediately frozen, pausing data flow to the subsequent judgment loop. The freeze time is set to 40 microseconds. In the frozen state, the sampled values ​​are stored in the high-speed cache, and a sampling point retention mechanism is activated to ensure that this data segment can still be used for delayed reconstruction after the freeze is lifted. After the freeze phase ends, a delayed release operation is initiated, using a third-order interpolation algorithm to reconstruct the frozen data in the time domain, filling in the breakpoints during the freeze period and preventing omissions or false triggers in action recognition due to data loss.

[0083] After the freeze-release process is complete, a differential homeomorphic structure adjustment operation is performed. This operation is used to confirm the continuity of the signal structure and the maintenance of the physical trend after the path switch. First, the first and second derivatives of one second of continuous data from the new path sampling signal are calculated to generate a derivative change curve, which is then compared with the derivative trend sequence of the channel data before the switch. If the absolute deviation of the trend line difference is within 0.1, the path switch is considered successful; if it exceeds the threshold, the system triggers path rollback, restoring the sampling path to the previous channel and re-injecting an inverse dissipative signal. This structure adjustment operation ensures the signal continuity, waveform stability, and physical interpretability of the sampling path during the adjustment process, avoiding data distortion or judgment errors caused by signal interruption or erroneous transition.

[0084] The results of the above intervention operations are integrated through a unified data structure to form a stable measurement and control closed-loop mechanism. This mechanism includes operational indicators such as the number of reverse voltage injections, switching path statistics, the number of freeze windows, and the success rate of structural adjustment. After each round of operation, all operational parameters are matched with the root cause information of disturbances to extract the oscillation suppression effect of this round. The change rate of the standard deviation of signal amplitude, the number of phase jumps, and the delay response indicators before and after the operation are compared to generate a stability score, which is used for the next round of parameter adjustment. If the score continues to decline, the system will automatically optimize the injection amplitude, switching threshold, freeze time, and homomorphic tolerance range to ensure that the control parameters maintain an optimal matching state in long-term operation, ensuring that the current transformer measurement link is always in a controlled, safe, and high-precision stable operating state.

[0085] This invention significantly improves sampling accuracy and identification sensitivity by constructing nanosecond-level time synchronization and phase reference baselines; it actively induces and quantifies parasitic coupling effects by injecting micro-amplitude bipolar phase disturbances and extracting hysteresis features; it outputs dynamic parameters based on sparse identification methods to enhance the fitting ability and interpretability of circuit models; it separates coupled disturbances from actual operating conditions using time-frequency causal untangling to clarify the root cause of oscillations; and it achieves adaptive adjustment of convergence threshold and time constant through linkage with the setting window of protection devices, ultimately forming a stable closed loop with the aid of anti-phase dissipation and measurement path topology reconstruction mechanisms. This technical solution significantly improves the anomaly identification rate and anti-interference capability of current transformers under extreme operating conditions, ensuring the accuracy of protection actions and the continuity and safety of power system operation. It has comprehensive benefits such as strong engineering applicability, low deployment threshold, and high diagnostic accuracy, breaking through the technical bottlenecks of traditional methods in terms of timeliness, accuracy, and adaptability.

[0086] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A current transformer real-time state diagnosis system based on the Internet of Things, characterized in that, The method comprises a reference construction module, a disturbance excitation module, a modeling identification module, a drive disentanglement module, a self-adaptive setting module and a closed-loop control module. The reference construction module establishes a nanosecond time synchronization reference and locks a phase reference baseline during the operation of the current transformer secondary winding loop, continuously acquires current vector trajectories based on the time synchronization reference and the phase reference baseline, and constructs a discrete phase residual graph. The disturbance excitation module injects a micro-amplitude bipolar phase disturbance into a set time sequence safety window based on the discrete phase residual graph, induces the response behavior of the parasitic capacitance, parasitic inductance and stray resistance, extracts amplitude and phase lag characteristics, and generates lag fingerprint data. The modeling identification module constructs a parameterized circuit model according to the lag fingerprint data, and outputs dynamic estimated values of the parasitic capacitance, parasitic inductance and stray resistance based on a sparse identification method. The drive disentanglement module jointly processes the dynamic estimated values and the discrete phase residual graph, separates the actual operation drive and the parasitic coupling drive by using a time-frequency causal disentanglement algorithm, and extracts a shock root cause sequence. The self-adaptive setting module constructs a hazard reference model based on the shock root cause sequence, links a protection device setting window, dynamically adjusts a convergence threshold and a time constant, and realizes adaptive setting of a protection strategy. The closed-loop control module implements inverse phase dissipation and topology reconstruction control on the basis of adaptive setting, completes differential homeomorphism adjustment by projecting an inverse phase damping electric field, switching a sampling path phase and triggering pulse freezing and delayed release, and realizes closed-loop stable control of a measurement link.

2. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 1 wherein, The discrete phase residual graph is constructed as follows: A temperature-compensated crystal oscillator with high stability is installed as a local clock source at a measurement node of the current transformer secondary winding loop, a unified time scale is realized among the measurement nodes by using a precise time synchronization protocol in combination with an external satellite time signal; Based on the measurement nodes with completed time synchronization, current signals are continuously acquired at a high frequency, the sampling signals are converted into complex vector form and the corresponding instantaneous phase values are calculated; A fixed-length rolling window method is used to extract a vector phase angle change sequence, and a reference baseline reflecting the phase evolution trend under normal conditions is constructed by using an error minimum fitting method; The measured phase is compared with the reference baseline point by point to obtain a phase residual sequence, which is mapped into a two-dimensional graph after normalization, and a weight function based on frequency density is introduced to enhance the graph performance, forming a discrete phase residual graph supporting disturbance identification.

3. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 2 wherein, The lag fingerprint data is generated as follows: Based on the constructed discrete phase residual graph, a time sequence section with small phase residual amplitude, stable phase derivative change and main spectral components concentrated in the power frequency and its low-order harmonic range is selected as a phase disturbance injection window; In the selected injection window, a micro-amplitude bipolar phase disturbance signal superimposed with multiple frequencies is generated and injected into the secondary winding loop through a coupling path with low impedance; During the disturbance injection process, current vector signals are acquired at a high frequency, and the phase delay, amplitude attenuation, response recovery time and frequency drift characteristics contained in multiple frequency components are extracted; The feature data extracted in each perturbation cycle is constructed into a response vector of a unified format, and combined to form a set of lagging fingerprint vectors as input data for a subsequent modeling and identification stage.

4. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 3 wherein, The three-frequency superimposed micro-amplitude bipolar phase disturbance signal is composed of a fundamental frequency signal and a plurality of harmonic signals, the phase of the disturbance signal is bidirectionally modulated relative to the main current signal, the injection duration is one complete power frequency cycle, and the impedance of the injection path is less than the minimum equivalent impedance of the disturbed circuit.

5. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 3 wherein, The dynamic estimation value generation step is as follows: The maximum phase lag time, amplitude decay rate, phase recovery time constant, and frequency drift amplitude in the lagging response feature are constructed into a standardized feature matrix, and arranged in timestamp order; Based on the standardized feature matrix, an equivalent circuit model containing series resistance, series inductance, and parallel capacitance is constructed, and the initial values and boundary value ranges of the parasitic capacitance, inductance, and resistance are limited; Adopting based on The norm-constrained least squares sparse regression algorithm is used to identify sparsity in the equivalent circuit model and removes high residual parameters through ten-fold cross-validation. The values of the identified parasitic capacitance, parasitic inductance, and stray resistance are appended with corresponding timestamps to construct a dynamic estimation value sequence, which is used to describe the time sequence evolution characteristics of the parasitic parameters.

6. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 5 wherein, The shock root cause sequence extraction step is as follows: The dynamic estimation values of the parasitic capacitance, parasitic inductance, and stray resistance are time-aligned with the phase response data of the corresponding perturbation cycle in the discrete phase residual map to construct a joint data matrix containing time index, phase value, phase derivative, parasitic capacitance, parasitic inductance, and stray resistance; Wavelet transform and Granger causality analysis are performed on the joint data matrix to separate the actual operation driving component and the parasitic coupling driving component, and to determine whether the parasitic parameter is the dominant causal item of the phase response change; The phase response segment with mutation characteristics is identified from the parasitic driving response, and the core dynamic characteristics of the segment are extracted to generate a shock root cause sequence containing nonlinear driving incentives and evolution trends.

7. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 6 wherein, The dynamic adjustment of the convergence threshold and time constant to realize adaptive setting of the protection strategy is as follows: A lagging shock hazard reference model is constructed based on the shock root cause sequence, historical root cause events are classified according to the disturbance intensity and the dynamic change rate of the parasitic parameters, and an event template library is established for matching the current disturbance event; The real-time input shock root cause information is compared with the event template library for similarity, the disturbance level score is output, and the protection action current threshold and response delay time are adjusted based on the score result; After the linkage adjustment is completed, the updated results are written into the protection strategy control process, and the reference model is continuously trained based on historical disturbance samples to realize adaptive closed-loop setting and updating of protection parameters.

8. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 7 wherein, When the disturbance level score reaches the high-risk level, the protection action current threshold is increased, the response delay time is shortened, and after adjustment, the updated protection parameters are compared with the relay protection action curve to verify whether they are within the allowed range, and if they exceed the deviation range, parameter correction is automatically performed.

9. The IoT-based real-time condition diagnostic system for current transformers as claimed in claim 7 wherein, On the basis of adaptive setting, the steps of differential homeomorphism adjustment are as follows: After completing the adaptive parameter setting update, a counter-coherent intervention signal is generated and injected through a low-coupling electrode to form an asymmetric damping electric field, thereby achieving time-domain interference suppression of the source of the lagging oscillation signal; Performing a measurement path reconstruction operation, judging the main interference path according to the oscillation amplitude and phase response characteristics of the three-channel sampling data, and realizing path switching through low-delay hardware switching channels; After path switching, the pulse freezing mechanism is triggered, the data flow is paused when the sampling derivative changes exceed the set threshold, and the data continuity is restored after the freezing ends using an interpolation algorithm; Performing differential homeomorphism structure adjustment, judging the path switching result by comparing the sampling derivative trends before and after switching, and triggering path rollback and re-intervention when the deviation exceeds the limit; Integrating the operation results to form a closed-loop control process, and evaluating and feeding back the oscillation suppression effect for optimizing the matching accuracy and response stability of the next round of control parameters.

Images