Three-phase ammeter mutual inductor access state diagnosis method, device, equipment and medium

Abstract

The invention belongs to the technical field of mutual inductor access state diagnosis, and particularly relates to a three-phase electric meter mutual inductor access state diagnosis method and device, equipment and a medium. The method comprises the following steps: injecting a non-power-frequency excitation signal into an A-phase loop of a to-be-diagnosed three-phase ammeter mutual inductor, and collecting a digital sample flow at an output side; generating a reference signal based on the non-power-frequency excitation signal, and generating a net characteristic measurement vector based on the current response signal and the reference signal; generating equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; acquiring real-time temperature and real-time load, inputting the real-time temperature and the real-time load together with the equivalent complex impedance into a digital twinborn model, and outputting a dynamic standard vector by the digital twinborn model; generating auxiliary features based on the instantaneous waveforms of the phase B and the phase C; and inputting the auxiliary features, the dynamic standard vector and the net feature measurement vector into a reverse diagnosis model, and outputting health management comprehensive evaluation on the overall access state.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of mutual inductor access state diagnosis, and particularly relates to a three-phase electric meter mutual inductor access state diagnosis method, device, equipment and medium. BACKGROUND

[0002] In the technical field of fault prediction and health management of three-phase electric meter mutual inductor access state, the existing technology mainly adopts a power domain diagnosis method based on an artificial power disturbance signal. For a system with controllable charge and discharge capacity, the method first artificially manufactures a controllable and instantaneous power change in the measured loop as a diagnosis signal through a preset M-time discharge and N-time charge step action sequence. Then, the diagnosis system accurately measures the total power change measured by the measuring electric meter, judges whether the direction and value of the power change are consistent with the preset disturbance signal, and combines a strict quantitative threshold to determine the positive connection or reverse connection, so as to realize the fault prediction and health management of the three-phase electric meter mutual inductor access state.

[0003] However, the power domain diagnosis method faces technical bottlenecks when facing modern complex power systems. The technical bottlenecks are mainly manifested as insufficient robustness and decreased diagnosis accuracy. The robustness of the power domain diagnosis method depends on a basic premise: the controllable power change (i.e. the diagnosis signal) must dominate the total power change of the system. However, in complex environments such as multi-inverter grid-connected systems, AC-coupled or DC-coupled energy storage reconstruction systems, there are a large number of uncontrollable, non-synchronous and non-diagnostic power fluctuations in the background of the load system. These fluctuations constitute strong background noise in the complex electromagnetic environment, which can mask or severely distort the controllable power change. Since the power domain diagnosis method is based on direct diagnosis of the total power change, when a large-amplitude, non-synchronous background disturbance occurs, the weak diagnosis signal will be submerged, making it impossible for the method to accurately measure the total power change. Therefore, even if a strict threshold is used for judgment, the diagnosis result is easily distorted, ultimately leading to incorrect connection or false fault prediction conclusions. In summary, the existing technology cannot effectively isolate or decouple the synchronous signal response with specific time domain characteristics from the non-synchronous and chaotic environmental disturbances in a physical manner, greatly limiting its practicality and reliability in complex power system application scenarios. SUMMARY

[0004] The purpose of the present application is to provide a three-phase electric meter mutual inductor access state diagnosis method, device, equipment and medium, which solves the problem of inaccurate three-phase electric meter mutual inductor access state diagnosis in the background art.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions: The first aspect of the application provides a three-phase electric meter mutual inductor access state diagnosis method, comprising: Inject a non-power frequency excitation signal into the A-phase loop of the three-phase electric meter mutual inductor to be diagnosed, and collect a digital sample stream on the output side; the digital sample stream includes a current response signal on the secondary side of the A-phase mutual inductor, an A-phase voltage instantaneous waveform, an A-phase current instantaneous waveform, and instantaneous waveforms of B-phase and C-phase; Generate a reference signal based on the non-power frequency excitation signal, and generate a net characteristic measurement vector based on the current response signal and the reference signal; Generate an equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; collect real-time temperature and real-time load, and input them together with the equivalent complex impedance into a digital twin model, and the digital twin model outputs a dynamic standard vector; Generate an auxiliary feature based on the instantaneous waveforms of B-phase and C-phase; input the auxiliary feature, the dynamic standard vector and the net characteristic measurement vector into a reverse diagnosis model, the reverse diagnosis model performs nonlinear mapping on the A-phase loop, reversely traces the source phase of the fault, and predicts the fault type, and outputs a comprehensive evaluation of the health management of the overall access state.

[0006] Preferably, the step of generating a reference signal based on the non-power frequency excitation signal and generating a net characteristic measurement vector comprises: Generate a reference signal based on the non-power frequency excitation signal, and perform virtual phase-locked demodulation and long integration on the current response signal and the reference signal to separate background power noise and extract in-phase components and quadrature components; Calculate the net amplitude and the relative net phase based on the in-phase components and the quadrature components, and generate a net characteristic measurement vector.

[0007] Preferably, the specific implementation steps of the in-phase components and the quadrature components extraction comprise: Track the non-power frequency excitation signal through a digital phase-locked loop, and real-time adjust the phase and the frequency to generate in-phase reference signals and quadrature reference signals at the same frequency as the non-power frequency excitation signal; Perform product calculation on the current response signal with the in-phase reference signals and the quadrature reference signals respectively through virtual phase-locked demodulation to generate in-phase intermediate product signals and quadrature intermediate product signals; The product calculation of the current response signal with the reference signal through virtual phase-locked demodulation can convert the target signal into a direct current component, and convert all other frequency components into high-frequency alternating current components; Define the integral point number based on the integral time, filter out high-frequency components and low-frequency noise different from the excitation frequency through integration in the preset integral time, and sum and average the in-phase intermediate product signals and the quadrature intermediate product signals at the integral points to generate in-phase components and quadrature components; The long integration is a low-pass filter, and the AC component converted from the background noise is offset by long-time accumulation summation and averaging, so as to eliminate the average value of the AC component in the numerical value.

[0008] Preferably, the step of generating the equivalent complex impedance based on the phase A voltage instantaneous waveform and the phase A current instantaneous waveform; collecting the real-time temperature and the real-time load, and inputting the equivalent complex impedance and the real-time load into the digital twin model, and the digital twin model outputting the dynamic standard vector, comprises: The phase A voltage instantaneous waveform and the phase A current instantaneous waveform are subjected to fast Fourier transform and calculation to generate the equivalent complex impedance. The equivalent complex impedance is input into the digital twin model, the digital twin model adjusts internal model parameters according to the equivalent complex impedance and the real-time working condition parameters, and constructs a transfer function based on the internal model parameters to output the dynamic standard vector.

[0009] Preferably, the specific generation step of the equivalent complex impedance comprises: A continuous discrete sampling data is intercepted based on the phase A voltage instantaneous waveform and the phase A current instantaneous waveform to generate a data window. The data window is operated by fast Fourier transform to convert the data window to a frequency domain to obtain a voltage complex sequence and a current complex sequence. A corresponding frequency domain index is calculated according to a pre-set non-power frequency test excitation frequency. The voltage complex sequence and the current complex sequence are extracted by the frequency domain index to obtain complex components at the frequency domain index points, i.e. complex voltage and complex current, which contain amplitude and phase information of the signals at the frequency domain index points. The complex voltage is divided by the complex current by complex division operation to obtain the equivalent complex impedance. The equivalent complex impedance contains an equivalent resistance and an equivalent reactance, and represents the real-time operating impedance characteristics of the transformer circuit at the excitation frequency.

[0010] Preferably, the specific generation process of the dynamic standard vector comprises: The equivalent voltage and current of the primary side of the transformer at the current frequency are calculated based on the equivalent complex impedance through the transformation ratio relationship. The dynamic magnetic flux and excitation inductance of the core at the current frequency are obtained based on the equivalent voltage and current by using a nonlinear B-H curve, which is an inherent characteristic of the core material and describes the nonlinear relationship between the magnetic flux density and the magnetic field strength and is stored in the digital twin model in the form of a function. The adjustment data of the structural parameters and the electromagnetic characteristics are reversely calculated based on the dynamic magnetic flux and the excitation inductance by using the ideal transformer equation and the Kirchhoff's law. The dynamic transfer function of the healthy current transformer loop under the current operating condition and the non-working frequency excitation signal frequency is calculated by using the structural parameters, electromagnetic characteristics and equivalent reflection impedance theory. The dynamic transfer function is calculated at the non-working frequency excitation frequency to generate a dynamic standard vector containing amplitude and phase information.

[0011] Preferably, the specific training steps of the reverse diagnosis model include: The reverse diagnosis model adopts a four-layer fully connected neural network structure, the first hidden layer contains 64 neurons, the second hidden layer contains 32 neurons, the third hidden layer contains 16 neurons, and the number of neurons contained in the output layer corresponds to the total number of all fault types and fault phase combinations; For each fault condition, the current response signal and three-phase instantaneous waveform of phase A are collected by the electromagnetic transient simulation model to generate a training feature vector of the model; The training feature vector is input into a multi-layer fully connected layer, and a ReLU nonlinear activation function is used in each layer to accelerate convergence; The Softmax activation function is used to output the fault source phase and the fault type, and a data set is generated; A joint loss function is used to minimize the error between the predicted output and the true label, and an Adam optimizer is used to update the parameters iteratively.

[0012] In the second aspect of the present application, a three-phase electric meter transformer access state diagnosis device is provided, which comprises: The acquisition module is used for injecting a non-working frequency excitation signal into the A-phase loop of the three-phase electric meter transformer to be diagnosed, and collecting a digital sample stream at the output side; the digital sample stream includes the current response signal of the A-phase transformer secondary side, the A-phase voltage instantaneous waveform, the A-phase current instantaneous waveform, and the instantaneous waveforms of the B-phase and the C-phase; The first vector module is used for generating a reference signal based on the non-working frequency excitation signal, and generating a net feature measurement vector based on the current response signal and the reference signal; The second vector module is used for generating an equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; real-time temperature and real-time load are collected and input together with the equivalent complex impedance into a digital twin model, and the digital twin model outputs a dynamic standard vector; The evaluation module is used for generating auxiliary features based on the instantaneous waveforms of the B-phase and the C-phase; the auxiliary features, the dynamic standard vector and the net feature measurement vector are input into a reverse diagnosis model, the reverse diagnosis model performs nonlinear mapping on the A-phase loop, reversely traces the source phase of the fault, and predicts the fault type, and outputs a comprehensive evaluation of the health management of the overall access state.

[0013] In a third aspect, the present application provides an electronic device comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the three-phase metering transformer access state diagnosis method.

[0014] In a fourth aspect, the present application provides a computer readable storage medium storing at least one instruction, wherein the at least one instruction is configured to be executed by a processor to implement the three-phase metering transformer access state diagnosis method.

[0015] Compared with the prior art, the present application has the following advantages: The present application injects a non-power frequency excitation signal, and generates a clean feature measurement vector isolated from background noise by using virtual phase-locked demodulation and long integration technology, which can effectively isolate the background power interference in the environment in a complex system, so that the diagnosis only relies on the net response of the measured circuit to the excitation to determine the wiring state, thereby overcoming the defect that the diagnosis accuracy of the prior art is masked or distorted due to uncontrollable power fluctuations in the background.

[0016] The present application generates a dynamic standard vector suitable for the actual working conditions such as current load and temperature by dynamically calculating the voltage instantaneous waveform and current instantaneous waveform through fast Fourier transform and digital twin model, which can adaptively adjust to the dynamic environment and working scene, thereby improving the anti-interference ability and working condition adaptability of the diagnosis, and solving the defect that the preset threshold value of the prior art fails to determine in a complex system due to different use scenarios.

[0017] The present application uses a deep learning reverse diagnosis model to combine the A-phase deviation vector and the working condition data of B-phase and C-phase for non-linear mapping to generate a final diagnosis report, which can effectively accurately locate the phase fault in a tightly coupled three-phase transformer system, thereby avoiding the "incorrect connection" or incorrect conclusion that may be drawn due to insufficient robustness of the prior art.

[0018] The present application isolates the interference of environmental noise by using a non-power frequency excitation signal and feature extraction, and realizes dynamic adaptive diagnosis of the current system by using a digital twin model and a reverse diagnosis model, thereby realizing the isolation of the synchronous signal response with specific characteristics and the non-synchronous and chaotic environmental interference, and overcoming the problem of insufficient robustness of the prior art when applied to a complex system. BRIEF DESCRIPTION OF DRAWINGS

[0019] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the description of the exemplary embodiments of the present application and the explanation thereof serve to explain the present application, and do not constitute improper limitations on the present application. In the drawings: Figure 1A three-phase electric meter mutual inductor access state diagnosis method flow chart for embodiment 1 of the present application; Figure 2 A three-phase electric meter mutual inductor access state diagnosis system structure schematic diagram for embodiment 1 of the present application; Figure 3 A net characteristic measurement vector generation flow chart for embodiment 1 of the present application; Figure 4 A dynamic standard vector generation flow chart for embodiment 1 of the present application; Figure 5 A reverse diagnosis model work flow chart for embodiment 1 of the present application.

[0020] Figure 6 A three-phase electric meter mutual inductor access state diagnosis device structure block diagram for embodiment 2 of the present application; Figure 7 A three-phase electric meter mutual inductor access state diagnosis device structure block diagram for embodiment 2 of the present application; DETAILED DESCRIPTION

[0021] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0022] The following detailed description is exemplary description and is intended to provide further detailed description of the present application. Unless otherwise specified, all technical terms used in the present application have the same meaning as understood by those skilled in the art. The terms used in the present application are only for the purpose of describing the specific embodiments and are not intended to limit the exemplary embodiments according to the present application.

[0023] Embodiment 1 As shown in Figures 1-5 , the present application provides a three-phase electric meter mutual inductor access state diagnosis method and system, and the technical solutions are as follows: A three-phase electric meter mutual inductor access state diagnosis method, with reference to Figure 1 , the present application proposes the specific implementation steps of the method, which include: S1, injecting a non-power frequency excitation signal into the A-phase loop of the three-phase electric meter mutual inductor to be diagnosed, and collecting a digital sample stream at the output side; the digital sample stream includes the current response signal of the A-phase mutual inductor secondary side, the A-phase voltage instantaneous waveform, the A-phase current instantaneous waveform, and the instantaneous waveforms of the B-phase and C-phase; S2, generating a reference signal based on the non-power frequency excitation signal, virtually phase-locked demodulating and long-integrating the current response signal and the reference signal, separating the background power noise, and extracting the in-phase component and the quadrature component; calculating the net amplitude and the relative net phase based on the in-phase component and the quadrature component, and generating a net characteristic measurement vector; S3. Perform a fast Fourier transform on the instantaneous waveforms of phase A voltage and phase A current and calculate the equivalent complex impedance; input the equivalent complex impedance into the digital twin model, the digital twin model adjusts the internal model parameters according to the equivalent complex impedance and real-time operating parameters, and constructs a transfer function based on the internal model parameters to output a dynamic standard vector. S4. Generate auxiliary features based on the instantaneous waveforms of phases B and C; input the auxiliary features, dynamic standard vector, and net feature measurement vector into the reverse diagnostic model, which performs nonlinear mapping on the phase A circuit, traces the source phase of the fault in reverse, predicts the fault type, and outputs a comprehensive health management assessment of the overall access status.

[0024] Furthermore, a non-power frequency excitation signal is injected into the A-phase circuit of the energy meter, and the current response signal on the secondary side of the A-phase transformer, as well as the instantaneous voltage waveform and instantaneous current waveform of the three phases, are collected in real time; corresponding to step S1 above, the specific process is as follows: A pure sine wave is generated as a non-power frequency excitation signal using a 16-bit or higher high-resolution direct digital frequency synthesizer controlled by a field-programmable gate array (FPGA). The frequency of the non-power frequency excitation signal must be far away from the power frequency and its main harmonics to facilitate efficient separation via virtual phase-locked demodulation. The non-power frequency excitation signal should meet the following requirements: frequency accuracy at the part-in-a-million (ppm) level, and total harmonic distortion better than [value missing]. 80 dB. A high-bandwidth, high-linearity current feedback power amplifier drives a high-frequency coupled current transformer with a ferrite core to inject a non-power frequency excitation signal in series into the A-phase current loop; the typical turns ratio of the high-frequency coupled current transformer is preferably 1:50 to 1:200 to ensure that the load impedance of the common transformer secondary circuit (0.1Ω) is within acceptable limits. At 10Ω, its equivalent series insertion impedance is less than 0.1Ω. The core material can be manganese-zinc ferrite, such as the PC40 series, to ensure low loss and high permeability under high-frequency excitation. The injection of the non-power frequency excitation signal must be strictly synchronized with the sampling clock of the analog-to-digital converter and the system's master clock, and a stable clock reference is provided by a high-precision temperature-controlled crystal oscillator.

[0025] As a preferred embodiment, the non-power frequency excitation signal can be dynamically frequency optimized. This dynamic frequency optimization selects a frequency with the lowest real-time noise energy that is not a peak value of all power frequency harmonics or major environmental noise as the excitation frequency. This ensures that the SNR of the non-power frequency excitation signal is maximized under any complex electromagnetic environment. The specific implementation is as follows: Without injecting any non-power frequency excitation signal, the field-programmable gate array (FPGA) performs high-speed, synchronous sampling of the secondary current of mutual inductor A, acquiring pure environmental noise and power frequency components. The acquired data is then analyzed in real time. The point domain data is subjected to fast Fourier transform to generate a noise spectrum map of the current environment, representing the noise power density of each frequency point. The noise spectrum map is filtered by a digital signal processor to identify and completely exclude the frequency band around the power frequency and all its main harmonics, and a noise power threshold is set, which should be dynamically set based on the real-time monitored noise spectrum map. The threshold is set to 3 times the average noise power density of the noise spectrum map after excluding the power frequency and its harmonics (i.e. 3σ principle), to effectively exclude non-peak background noise while ensuring sensitivity. Identify all peak noise frequencies (usually from high-frequency interference sources such as frequency converters, switching power supplies, EV charging piles, etc.) whose power exceeds the noise power threshold, and exclude these frequency points and the surrounding protection bandwidth. Among all the candidate frequencies that have not been excluded, the digital signal processor looks for the one with the smallest noise power density P f , which is set as the optimal excitation frequency for this diagnosis.

[0026] Through dynamic frequency optimization, the system can perceive the most unfavorable noise frequency band in real time, ensuring that the current response signals collected in complex industrial or residential mixed power grid environments can all obtain the best signal-to-noise ratio, greatly improving the data accuracy and robustness of subsequent virtual phase-locked demodulation.

[0027] A four-channel synchronous analog-to-digital converter with high resolution (e.g. 16 bits) and high sampling rate (e.g. 10kSPS to 50kSPS) is selected for parallel sampling of the three signals, including the current response signal at the secondary side of the A-phase transformer, the A-phase voltage instantaneous waveform, the A-phase current instantaneous waveform, and the instantaneous waveforms of the B-phase and C-phase. At the input end of the analog-to-digital converter, an automatic gain adjustment of the programmable gain amplifier is implemented to ensure that the signal amplitude always occupies more than 80% of the dynamic range of the analog-to-digital converter. Subsequently, the signal passes through a high-order (e.g. 8th order) active low-pass filter as an anti-aliasing filter, whose cutoff frequency must be strictly lower than the Nyquist frequency. A low-jitter clock distribution network is used to synchronously distribute the unified master clock to all analog-to-digital converters without attenuation or distortion, ensuring absolute synchronization of the sampling time. Finally, a digital sample stream containing the current response signal at the secondary side of the A-phase transformer, the A-phase voltage instantaneous waveform, the A-phase current instantaneous waveform, and the instantaneous waveforms of the B-phase and C-phase is generated.

[0028] The digital sample stream is subjected to pre-processing of digital calibration and timing reference establishment, including DC bias elimination and physical value conversion, and the timing reference establishment includes synchronization alignment of the three signals and algorithm time base, and the specific implementation is as follows: ​​​The field programmable gate array constructs a double-port first-in-first-out buffer using random access memory, continuously writes data in the first-in-first-out buffer through an analog-to-digital converter, and reads in the form of data blocks by a digital signal processor, so as to realize decoupling of read and write clock domains and ensure real-time of data. The digital signal processor calculates the average value of N power frequency cycle (N≥200) samples, takes the average value of the N power frequency cycle samples as a direct current bias, performs direct current (DC) bias and low frequency drift elimination on the original digital samples based on the direct current bias, and generates alternating current digital signals. The digital signal processor calls pre-stored calibration coefficients, converts the analog-to-digital converter code of the alternating current digital signal into a physical value with a unit of volt / ampere through floating point operation based on the calibration coefficients, and realizes non-linear compensation. At the same time, a fractional delay compensation algorithm is implemented to eliminate the multi-channel group delay difference caused by the difference of analog links, ensure strict synchronization alignment of three signals, and generate calibration data stream with accurate physical value and strict time synchronization alignment . The field programmable gate array realizes zero-crossing detection through a digital phase-locked loop to generate a system-level timing reference power frequency cycle , determines the total number of fast Fourier transform data frame sampling points and the long integration time required for virtual phase-locked demodulation based on the reference power frequency cycle, wherein the total number of fast Fourier transform data frame sampling points must be an integer power of 2, and the total time length determined by the total number of sampling points and the sampling frequency must be an integer multiple of the power frequency cycle . The long integration time T required for virtual phase-locked demodulation is set to an integer multiple of the power frequency cycle T power frequency. In order to ensure that the signal-to-noise ratio is greater than 30 dB while maintaining dynamic adaptability to working condition changes, the value of the integration time T is limited to between 10 times and 100 times the power frequency cycle.

[0029] Through digital calibration and establishment of timing reference, a pre-processed pure data stream is generated, which ensures high precision, high synchronization and applicability of the algorithm, and provides accurate reference standard and time basis for subsequent virtual phase-locked demodulation and fast Fourier transform.

[0030] The frequency of the non-power frequency excitation signal is not the power frequency or its main harmonic, which naturally separates the response of the diagnostic signal from the main background power interference in the environment in frequency, creating a prerequisite for subsequent virtual phase-locked demodulation and long integration technology.

[0031] Further, a reference signal is generated based on the non-power frequency excitation signal, and the current response signal and the reference signal are subjected to virtual phase-locked demodulation and long integration to separate background power noise and extract in-phase and quadrature components; the net amplitude and relative net phase are calculated based on the in-phase and quadrature components to generate a net characteristic measurement vector; corresponding to the above step S2, the specific process is as follows: A reference signal is generated based on the non-power frequency excitation signal through direct digital frequency synthesis technology, and the reference signal contains an in-phase reference signal and quadrature reference signals , the specific generation formula is: ), -sin ) ; wherein represents the frequency of the non-power frequency excitation signal, represents the sampling rate of the data acquisition system, and n represents the index of the sampling point.

[0032] The collected current response signals are multiplied by the in-phase reference signal and the quadrature reference signal respectively through multiplication mixing to generate in-phase component signals and quadrature component signals , the signals and both mainly contain a direct current component and a double frequency (2 ) high-frequency alternating current component, and noise transferred to other frequencies. The required weak signal is converted into a direct current component, and all other components (including noise) are converted into a high-frequency alternating current component. By low-pass filtering, the in-phase component signal and the quadrature component signal are integrated for a long time to generate in-phase component X and quadrature component Y, and the specific calculation formula is: , ; wherein N represents the total number of data points participating in one long integration, and (T is the long integration time, is the sampling rate of the data acquisition system), represents the integral operation corresponding to the continuous time domain. The double frequency high-frequency alternating current component and the noise transferred to other frequencies will tend to zero after summation because their waveforms cancel each other out in the period. Finally, only the stable direct current components X and Y are left.

[0033] Based on the in-phase component X and the quadrature component Y, the calculation of the net amplitude NM and the relative net phase NP generates the net feature measurement vector =( ) ; the specific calculation formula is: .

[0034] By virtual phase-locked demodulation and long integration, the required weak signal can be converted into a direct current component, and the alternating current noise component can be eliminated. This method can effectively isolate the background power interference such as load or photovoltaic fluctuation in the environment, so that the diagnosis only depends on the net response of the measured loop to the excitation.

[0035] Further, the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform are subjected to fast Fourier transform and calculation to generate an equivalent complex impedance; the equivalent complex impedance is input into the digital twin model, the digital twin model adjusts internal model parameters according to the equivalent complex impedance and real-time working condition parameters, and constructs a transfer function based on the internal model parameters to output a dynamic standard vector; corresponding to the above step S3, the specific process is as follows: According to the above power frequency cycle Determine the data window length , the voltage instantaneous waveform and the current instantaneous waveform time sequence are synchronously intercepted continuous data points to generate the voltage data window and the current data window .

[0036] The voltage data window and the current data window are multiplied by using a Hanning window function, to generate the voltage data window and the current data window , the weight sequence of the Hanning window function is specifically calculated as follows: ; Wherein represents the index of the discrete time sequence, an integer, represents the length of the data window.

[0037] The processed voltage data window and the current data window are subjected to fast Fourier transform operation respectively to generate the voltage complex sequence and the current complex sequence , and the specific calculation formula is as follows: ; ; Wherein K represents the integer index of the data window, represents the index of the discrete time sequence, j represents the imaginary unit, and satisfies = 1, represents the basis function of the Fourier transform, which is essentially a rotation vector with the origin of the complex plane as the center.

[0038] According to the excitation frequency , the corresponding index is calculated, and the specific calculation formula is as follows: ). The calculated index is used to obtain the voltage complex sequence and the current complex sequence The complex component on the frequency point is extracted to generate a complex voltage and a complex current , the complex voltage and the complex current contain the amplitude and phase information of the signal at the specific high frequency point. Using complex division operation, the complex voltage is divided by the complex current to obtain an equivalent complex impedance , which contains an equivalent resistance and an equivalent reactance.

[0039] Through accurate extraction of the equivalent complex impedance, the accuracy of the amplitude and phase information of the signal at the frequency is ensured, and a data basis is provided for subsequent adjustment of the parameters of the digital twin model, so that it can generate dynamic standards conforming to the current environment.

[0040] The digital twin model is a digital model established based on the structural parameters and electromagnetic characteristic parameters of the three-phase electric meter transformer; the structural parameters define the geometric and electrical layout of the transformer and its installation environment, including: turns ratio (primary side ): determines the basic transformation ratio relationship of the ideal transformer; winding geometric size: used to calculate the initial static values of leakage inductance (primary side ) and parasitic capacitance ; winding resistance (primary side ): determines the reference value of the resistance and the coefficient for dynamic adjustment with temperature; core geometric parameters: including effective cross-sectional area , effective magnetic path length ; electric energy meter load characteristics: the actual load impedance (secondary side ) connected to the transformer, including the sampling loop impedance of the electric energy meter itself, the connection wire impedance, etc.; mutual coupling characteristic parameters: describe the mutual magnetic coupling coefficient between three-phase transformers. The electromagnetic characteristic parameters include: magnetization characteristic curve (B H curve): this is the inherent characteristic of the core material, which describes the nonlinear relationship between magnetic flux density (B) and magnetic field strength (H), the nonlinear relationship of the magnetization characteristic curve (B-H curve) can be fitted by an exponential model, and the digital twin model uses the specific mathematical function after fitting to dynamically calculate the real-time dynamic permeability under the current magnetic flux state, and then correct the excitation inductance.

[0041] The real-time temperature of the winding is collected by the thermal resistance sensor , and the real-time load current is collected by the electric meter transformer ; the winding resistance (primary side ) changes with the real-time temperature , and its specific formula is: ; where is the corrected resistance value of the transformer winding at real-time temperature T; represents the rated resistance value (usually the factory test value) of the winding at reference temperature ; represents the standard temperature according to which the calibration is carried out; ; represents the physical constant of the material. For copper windings, its value is approximately 0.00393 / K.

[0042] excitation inductance is a function of the magnetic permeability of the core, which varies with the magnetic field strength H caused by the load current, and its specific formula is: ; where is the magnetic permeability of the core material at real-time magnetic field strength H, which is a nonlinear parameter, obtained by fitting the nonlinear B H curve.

[0043] The digital twin model dynamically adjusts parameters according to the real-time collected temperature and load current, ensuring that its output dynamic standard vector accurately reflects the ideal health status under the current working condition, thereby ensuring the data accuracy and standardization of the model when facing complex systems.

[0044] The equivalent complex impedance is input into the digital twin model, and the equivalent complex impedance is converted to the equivalent impedance of the primary side through the transformation ratio, and its formula is: , which is also based on the turns ratio of the transformer, the complex voltage and the complex current are converted to the equivalent voltage of the primary side and the equivalent current of the primary side. Based on the equivalent voltage and the equivalent current , the real-time magnetic field strength working point of the core on the nonlinear B-H curve is determined based on the real-time load current , based on the working point , and the B-H curve is differentiated to determine the small signal tangent permeability of the core at this working point, and the excitation inductance is calculated using the above formula, at the same time, the dynamic magnetic flux at this frequency is calculated based on the equivalent voltage of the primary side, and its specific formula is: , the dynamic magnetic flux The magnetic flux density of the non-working frequency excitation signal is only used to verify that the core is not saturated, and the value of the excitation inductance is determined by the formula of the above excitation inductance ; ; The generated excitation impedance and the secondary side impedance are calculated by the winding resistance and the excitation inductance ; ; , and the specific formula is: ; ; , wherein represents the iron loss equivalent resistance determined based on model calibration, and the value can be further corrected based on ; represents the frequency of the non-working frequency excitation signal, is the secondary winding resistance, is the primary number of turns, is the secondary side leakage inductance.

[0045] Using the complete T-type equivalent circuit model, the dynamic transfer function of the current transformer loop in a healthy state under the current operating condition and the non-working frequency excitation signal frequency is calculated , in a three-phase compact coupling system, the currents of phase B and phase C will also generate induced electromotive force in the winding of phase A. Therefore, when constructing the dynamic transfer function of phase A, the mutual inductance term generated by the currents of phase B and phase C needs to be introduced into the T-type equivalent circuit model. Specifically, when calculating the total voltage of the secondary side of phase A, in addition to the voltage of its own loop, the mutual inductance electromotive force also needs to be superimposed, wherein and are the mutual inductance coupling characteristic parameters, representing the mutual inductance coefficients between phase B and phase A and between phase C and phase A, and are the current response components of phase B and phase C at the excitation frequency. After correction, the equivalent impedance of the secondary side will be replaced by an equivalent network containing the mutual inductance term, and the dynamic transfer function reflecting the coupling effect is derived. The excitation impedance and the secondary side impedance are substituted into the transfer function formula, and the formula is: , the complex number is converted to polar form, and the dynamic standard vector = ( , ) is obtained, the dynamic standard vector contains the amplitude and the phase , wherein =| |, = angle (V1, V2) ).

[0046] The dynamic standard vector generated by the digital twin model is an ideal health state response that adapts to the actual working conditions such as current load and temperature in real time, which can effectively improve the anti-interference ability and working condition adaptability of diagnosis, thereby avoiding the failure of threshold judgment due to background interference in complex systems.

[0047] Further, the net characteristic measurement vector is compared with the dynamic standard vector, when there is a deviation, based on the mutual inductance coupling characteristics of the three-phase ammeter transformer, the A-phase loop is nonlinearly mapped through the reverse diagnosis model, the source phase of the fault is traced back, the fault type is predicted, and the health management comprehensive evaluation of the overall access state is output; corresponding to the above step S4, the specific process is: Calculate the net characteristic measurement vector and the dynamic standard vector generated by the digital twin model The normalized Euclidean distance between the two is calculated to generate a deviation measure D, which is decomposed into amplitude deviation and phase deviation . The deviation measure D calculation formula is: ; where and represent the normalization factor, usually taking and standard deviation.

[0048] The reverse diagnosis model realizes the diagnosis of the access state by inputting the deviation vector of phase A (main input feature) and the working condition parameter change rate of phases B and C (auxiliary input feature) which are not excited into the trained multi-layer fully connected neural network, the working condition parameter change rate of phases B and C includes impedance amplitude and phase change rate and voltage and current change rate, the specific implementation steps are: Based on the above S3 step, the equivalent complex impedance of phases B and C is calculated and , the health reference impedance and , the complex voltage and , the complex current and . Among them, the health reference impedance of phases B and C and generated by the digital twin model. The digital twin model dynamically adjusts the T-type equivalent circuit parameters representing the health state according to the real-time collected temperature, load and other working condition parameters of phase B and phase C. Although the non-working frequency excitation signal is not actively injected into phase B and phase C, due to system coupling, the non-working frequency excitation signal of phase A will produce a weak response in phase B and phase C. Based on this coupling effect and in combination with real-time working conditions, the model calculates the theoretical equivalent complex impedance that phase B and phase C should present in the healthy state at the excitation frequency. This theoretical equivalent complex impedance is the health reference impedance. The calculated health reference complex impedance and is compared with the equivalent complex impedance and to generate impedance amplitude and phase change rates. The change rates of the complex voltage and current of phase B and phase C relative to their health average values are calculated to generate voltage and current change rates. The determination method of the health state reference is as follows: first, in the system first deployment or calibration stage, the working condition parameters of at least 1000 power frequency cycles are continuously collected in the healthy state of correct wiring, the statistical average and standard deviation are calculated to form an initial static health reference. Secondly, during the normal operation of the system, an adaptive sliding window algorithm is used to maintain a dynamic health reference. This algorithm continuously caches the working condition data of the past 24 hours as a sliding window and updates it with a step of 1 hour. In order to exclude transient disturbance data in the window, the highest and lowest 5% of the values in the window data are removed by using the truncated mean method, and then the average value of the remaining data is calculated as the current dynamic health reference. When calculating the change rate, the system preferentially uses the dynamic health reference; only when the system is restarted or the dynamic reference is invalid due to long-time data anomaly, the initial static health reference can be temporarily used. The impedance amplitude and phase change rates and the voltage and current change rates are used as auxiliary features, the amplitude deviation and phase deviation are used as main features, the main features and all auxiliary features are combined to form a complete input feature vector . The feature vector contains 10 dimensions, and the specific composition is: A-phase amplitude deviation, A-phase phase deviation, B-phase impedance amplitude change rate, B-phase impedance phase change rate, B-phase power frequency voltage change rate, B-phase power frequency current change rate, C-phase impedance amplitude change rate, C-phase impedance phase change rate, C-phase power frequency voltage change rate and C-phase power frequency current change rate. The input feature vector is subjected to Min-Max normalization, and all feature values are scaled to the range of [0, 1] to eliminate dimension differences, accelerate model convergence, and generate a feature vector .

[0049] By establishing the main input features and auxiliary input features, the model can effectively utilize the mutual inductance coupling characteristics of the three-phase mutual inductor circuit, and learn and identify the unique feature combination of this cross-phase coupling, so as to effectively trace back the actual source phase of the fault and predict the fault type.

[0050] Based on each possible fault type, such as 1% 10% turn-to-turn short circuit, different degrees of virtual connection, different resistance value grounding fault, etc., a continuous severity gradient is designed, for example, the turn-to-turn short circuit increases by 1% from 1% to 10%; the virtual connection resistance value changes from 1Ω to 100Ω, the fault type is dynamically adjusted based on the standard requirements of the specific system, and the severity gradient should be realized by simulating the change of the grounding resistance, the range of the grounding resistance should cover from 50Ω low resistance grounding to 20kΩ high resistance grounding, and the data is sampled at 20 logarithmic intervals or linear steps, for example, 1000Ω steps, to ensure that the training data covers all fault degrees that may occur in actual application. Based on the severity gradient and the fault type, a simulation working condition is generated, and a real label Y is generated for each simulation working condition, the real label Y uses One-Hot encoding to clearly indicate the “fault type” and “fault phase” (for example, the label of turn-to-turn short circuit 5% in A phase may be a specific index in the [0, 0, 1, 0] vector). Using PSCAD electromagnetic transient simulation software, a high-precision model containing the nonlinear B H curve of mutual inductor, complete T equivalent circuit is constructed. Run the excitation and collection process of the diagnostic system in the model, that is, in the power frequency operating state, inject high-frequency diagnostic excitation, and collect: A-phase current response signal and three-phase instantaneous waveform for each fault condition. The collected data is processed by the above fast Fourier transform and the calculation of dynamic standard vector and deviation and rate of change, and finally a training feature vector X containing all 10 features is generated. All training feature vectors are packaged as a training data set. The training data set is input into the reverse diagnosis model, which adopts a 4-layer fully connected neural network structure. The input layer receives 10 feature inputs; the first hidden layer contains 64 neurons; the second hidden layer contains 32 neurons; the third hidden layer contains 16 neurons; and the output layer contains 15 neurons, corresponding to the total number of all possible fault type and fault phase combinations (for example: A-phase virtual connection, B-phase virtual connection, C-phase virtual connection, etc. 15 cases). The reverse diagnosis model inputs the received 10-feature training feature vector into the hidden layer. The number of neurons in each hidden layer decreases or increases to extract different levels of feature association (10→64→32→16→15), and each connection has a weight W and a bias b, which are the parameters that the model needs to learn. After the output vector Z=W*X+b in each hidden layer, the ReLU function is activated to generate the original output . Ensure that the model can learn and represent complex non-linear relationships. The output layer outputs neurons equal to the total number of all possible fault types and phase combinations, and the last layer uses the Softmax function to convert the model's raw output into a probability distribution vector P. The cross-entropy loss is calculated using the joint loss function, which measures the difference between the model's predicted probability distribution P and the true label Y (One-Hot encoding). The smaller the cross-entropy loss value, the closer the model's prediction is to the true fault label. The specific calculation of the cross-entropy loss is as follows: ; where represents the number of samples used to train the model, represents the average and negative sign, represents the sum of the loss for M training samples (i from 1 to M), represents the sum of the loss terms for C possible fault categories (j from 1 to C), represents whether the ith sample belongs to the jth category, represents the model's predicted probability that the ith sample belongs to the jth category.

[0051] The backpropagation algorithm uses the chain rule of calculus to calculate the gradient of the loss function with respect to each layer's parameters, starting from the output layer and working backward (against the direction of data flow). This generates a vector g containing the gradients of all parameters in the network. The direction of the gradient g points to the direction in which the loss function increases the fastest. The Adam optimizer calculates the first moment (gradient mean m) and the second moment (gradient square mean v) of the gradient g, and uses the estimates of the first and second moments to perform bias correction, thereby adaptively adjusting the learning rate a. The learning rate a is used to calculate the update term, and the weights W are updated. The loop is iterated until the cross-entropy loss drops to a very small preset value (e.g. <0.01); the accuracy or loss of the model on data not used for training is monitored in real time. If the accuracy or loss does not improve for consecutive multiple loop iterations, training can also be stopped. The weights W and biases b at this time are considered optimal parameters, which are fixed and integrated into the online diagnostic system for receiving real-time feature vectors input and making forward probability predictions of faults.

[0052] By training the reverse diagnosis model and updating the weights W and biases b, the model's adaptive update is achieved, which can effectively ensure the accuracy of subsequent diagnostic reports.

[0053] The feature vector The input reverse diagnosis model is used to generate a probability distribution P based on the above operation, and threshold judgment is performed on the output probability P. The item with the highest probability is selected as the preliminary diagnosis result, and it is judged whether it exceeds the preset confidence threshold, which is preferably set to 0.9. When the highest probability of the probability distribution P is greater than the confidence threshold, it is determined that the diagnosis result is reliable, and the fault type and phase can be directly determined. Otherwise, it is considered that the model is 'uncertain'. The system should output the results of'suspected fault, need further monitoring' or 'no obvious fault', and may trigger more stringent online monitoring. After confirming the diagnosis result, the specific fault type and the phase of the fault are determined by consulting the fault-index mapping table. The amplitude deviation and the phase deviation are compared with the preset severity classification threshold to generate a severity classification judgment of mild, moderate and severe, and the severity classification threshold is set as follows: mild: normalized Euclidean distance D≤0.05; moderate: normalized Euclidean distance 0.05<D≤0.2; severe: normalized Euclidean distance D>0.2. This classification threshold is set based on simulation data and historical experience. Finally, an evaluation report containing the diagnosis result, the affected phase and the severity is generated.

[0054] Using the reverse diagnosis model, the A-phase deviation vector and the working condition data of B and C phases are combined for nonlinear mapping. The actual source phase causing the A-phase anomaly can be traced back, and the fault type can be predicted. The 'incorrect connection' or incorrect conclusion caused by insufficient robustness is avoided.

[0055] The present application isolates the interference of environmental noise by non-power frequency excitation signal and feature extraction, and realizes dynamic adaptive diagnosis of the current system through digital twin model and reverse diagnosis model, constructs a complete isolation interference and adaptive three-phase ammeter mutual inductor access state diagnosis method, and solves the problem of insufficient dynamic robustness when applied to complex systems.

[0056] As a preferred embodiment, the three-phase ammeter mutual inductor access state diagnosis system comprises: an excitation and data acquisition module, a double-path feature extraction and standard generation module, and an intelligent diagnosis and health evaluation module.

[0057] Further, the excitation and data acquisition module comprises a non-power frequency excitation signal unit, a data acquisition unit and a signal processing unit. The non-power frequency excitation signal unit generates a single-frequency pure sine wave as a non-power frequency excitation signal through a high-resolution direct digital frequency synthesizer, and injects it into the A-phase current loop in series. The data acquisition unit comprises a four-channel synchronous analog-to-digital converter with high resolution and high sampling rate, which is used to acquire A-phase current response signals, three-phase voltage instantaneous waveforms and current instantaneous waveforms in real time. The signal processing unit synchronously distributes a unified master clock to all ADCs through a clock distribution network, and is responsible for direct current bias elimination and fractional delay compensation, ensuring strict synchronous alignment of the three signals.

[0058] Further, the dual-path feature extraction and standard generation module comprises a measurement vector generation unit, an equivalent complex impedance calculation unit, a digital twin model and a dynamic standard vector generation unit. The measurement vector generation unit works by performing steps such as "virtual phase-locked demodulation" and "long integration", generates in-phase and quadrature reference signals synchronized with the non-power frequency excitation signal through a digital phase-locked loop, separates power frequency and background noise by multiplication mixing and long integration, and outputs a net feature measurement vector. The equivalent complex impedance calculation unit calculates the equivalent complex impedance by fast Fourier transform operation and accurately extracts complex voltage and complex current according to the frequency domain index corresponding to the excitation frequency, and then calculates the equivalent complex impedance by complex division. The digital twin model is a digital model established based on the structural parameters and electromagnetic characteristic parameters of the transformer, which is used to calculate the reflection impedance of the equivalent complex impedance and inversely calculate the current magnetic flux state. The dynamic standard vector generation unit is based on the digital twin model to calculate the dynamic transfer function and frequency response that the healthy state transformer loop should have under the current operating condition and non-power frequency excitation signal frequency, and generate a dynamic standard vector.

[0059] Further, the intelligent diagnosis and health assessment module comprises a deviation metric and feature generation unit, an auxiliary feature calculation unit, an inverse diagnosis model, and a health management comprehensive assessment unit. The deviation metric and feature generation unit calculates the normalized Euclidean distance between the net feature measurement vector and the dynamic standard vector, and generates a deviation metric. The deviation metric is decomposed into amplitude deviation and phase deviation as the main input features. The auxiliary feature calculation unit calculates the amplitude and phase of the equivalent complex impedance of phases B and C, and the change rate of the power frequency current and voltage of phases B and C with respect to the health state, and takes these as auxiliary input features, which are normalized together with the main features to generate an input feature vector. The inverse diagnosis model is a trained multi-layer fully connected network, which uses the Softmax function to output the probability distribution of the fault source phase and the fault type. The source phase of the fault is traced back through nonlinear mapping, and the fault type is predicted. The health management comprehensive assessment unit performs threshold judgment on the output probability, combines the fault type, the deviation degree, and the traced-back source phase, and finally outputs a health management comprehensive assessment report.

[0060] As a preferred embodiment, in the light peak period of photovoltaic power generation, due to the rapid movement of clouds, the output power of the inverter fluctuates frequently and greatly, and the fault signal is completely overwhelmed by background noise. Through virtual phase-locked demodulation and long integration, the power fluctuation of the photovoltaic system is isolated, and reference is made to Figure 3 A flowchart for generating the net feature measurement vector is shown.

[0061] The system injects a non-power frequency pure sine wave non-power frequency excitation signal with a frequency of 12.8 kHz into the A-phase loop through the non-power frequency excitation signal generation unit, and synchronously collects the current response signal at a rate of 51.2 kHz. The net feature measurement vector generation unit starts virtual phase-locked demodulation, digitally generates in-phase and quadrature reference signals synchronized with 12.8 kHz, and multiplies the collected current response signal with them respectively. This mixing operation is the key to signal noise reduction, which accurately converts the target 0.8 mA 12.8 kHz response signal into a direct current component, while the 100 mA 50 Hz noise is shifted to the 12.8 kHz ± 50 Hz alternating current region.

[0062] Subsequently, the system performs long integration (summation average) on the mixed signal for 0.5 s. Since 0.5 s is much larger than the period of the power frequency 20 ms, the integration operation can make all the power frequency and background noise AC components tend to zero due to positive and negative cancellation. Finally, through high-depth integration filtering, the background noise is suppressed to below the level of 0.005 mA.

[0063] Through virtual phase-locked demodulation and long integration, the SNR of the system is improved to more than 40 dB, and the in-phase component and quadrature component of the 0.8 mA response signal caused by the fault are successfully separated and accurately extracted in the high background power noise environment.

[0064] The present application ensures the accuracy of the subsequent net feature measurement vector by virtual phase-locked demodulation and long integration, effectively avoiding the interference of background power in the multi-inverter system on the diagnosis.

[0065] As a preferred embodiment, the user side high-power load is connected, and the real-time winding temperature of the transformer is as high as 70℃ and the real-time load current reaches 90% of the rated value, which is a specific scenario. A dynamic standard vector is generated by a digital twin model to realize dynamic adaptation to system load changes. Referring to Figure 4 , a flowchart for generating a dynamic standard vector.

[0066] The digital twin model first receives temperature information of 70℃, and immediately calls the copper winding resistance temperature (α=0.00393 / ℃) formula to accurately correct the winding resistance to 1.39Ω. The model uses the corrected resistance value to fine-tune the excitation inductance in combination with the excitation characteristics under 90% load to reflect the magnetic saturation characteristics or stress influence of the core. The total impedance at the secondary side is calculated, and then the total impedance is reflected to the primary side through the turns ratio to generate a reflected impedance. Based on the equivalent T-circuit of the transformer and the reflected impedance theory, the ideal amplitude and phase that the healthy transformer should have under the current 70℃ working condition and at the excitation frequency of 12.8kHz are calculated. The ideal amplitude and phase constitute a dynamic standard vector.

[0067] Since the dynamic standard vector has dynamically included the resistance and impedance changes caused by the temperature rise of 70℃, the deviation between the measured net feature vector (including the actual influence at 70℃) and the dynamic standard vector is maintained within the healthy range of 0.1% or less.

[0068] The digital twin model successfully distinguishes the normal working condition deviation caused by temperature and load changes from the abnormal deviation caused by faults, eliminates false deviations caused by changes in operating conditions, avoids large-scale false positives, and significantly improves the working condition adaptability of the diagnosis.

[0069] As a preferred embodiment, in a tightly coupled three-phase transformer system, a moderate virtual connection fault occurs at the secondary side of phase C, which is a specific scenario. Through a reverse diagnosis model, the reverse tracing is performed, referring to Figure 5 , a workflow diagram for the reverse diagnosis model.

[0070] First, the deviation between the net feature measurement vector of phase A and the dynamic standard vector is calculated to generate =6.2% and = 1.8°. These data are considered as a symptom of A-phase malfunction, while the system calculates auxiliary input features for B-phase and C-phase. Since there is a 20Ω open fault in C-phase, its equivalent complex impedance is calculated compared with the healthy benchmark, and the impedance amplitude variation rate of C-phase is calculated as +18%, and the phase variation, power frequency voltage and current variation rate are also calculated with specific values. While B-phase remains healthy, its corresponding impedance, voltage and current variation rates are all calculated as small values close to zero (e.g. absolute values are less than 0.5%). The two main features of A-phase and the eight auxiliary features of B-phase and C-phase are combined to form a 10-dimensional input feature vector including , , four variation rates of B-phase (all <0.5%), four variation rates of C-phase (among which the amplitude variation rate is +18%). All ten features are Min Max normalized to form a complete input feature vector. The input feature vector is input into the trained multi-layer fully connected network, the network performs forward propagation, and through multi-layer hidden layer operations, it identifies the unique combination of A-phase 6.2% deviation intensity and C-phase +18% impedance increase, which mathematically points to the fault mode of "C-phase moderate open". The Softmax function of the output layer outputs the final probability distribution P. Due to the accurate guidance of the feature combination, the neuron of the "C-phase moderate open" class is activated to the highest probability of 0.96, while the probability of "A-phase moderate open" is suppressed to about 0.02. Finally, the health management comprehensive evaluation unit accurately generates the report according to the highest confidence of 96%: "Diagnosis result: open; affected phase: C".

[0071] The reverse diagnosis model realizes accurate tracing and positioning of cross-phase faults by reverse tracing. This capability avoids misjudgment and misoperation of the operation and maintenance personnel on the wrong phase, and significantly improves the accuracy of fault positioning.

[0072] Embodiment 2 As shown in Figure 6 , based on the same inventive concept as the above embodiment, the present application also provides a three-phase watt-hour meter transformer access state diagnosis device, comprising: An acquisition module is configured to inject a non-power frequency excitation signal into an A-phase circuit of a three-phase watt-hour meter transformer to be diagnosed, and to acquire a digital sample stream on the output side; the digital sample stream includes a current response signal of the secondary side of the A-phase transformer, an A-phase voltage instantaneous waveform, an A-phase current instantaneous waveform, and instantaneous waveforms of B-phase and C-phase; A first vector module is configured to generate a reference signal based on the non-power frequency excitation signal, and to generate a net feature measurement vector based on the current response signal and the reference signal; The second vector module is configured to generate an equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; collect real-time temperature and real-time load, and input the real-time temperature and the real-time load to a digital twin model together with the equivalent complex impedance, wherein the digital twin model outputs a dynamic standard vector; The evaluation module is configured to generate an auxiliary feature based on the instantaneous waveforms of the B phase and the C phase; input the auxiliary feature, the dynamic standard vector and a net feature measurement vector to a reverse diagnosis model, wherein the reverse diagnosis model performs nonlinear mapping on the A-phase circuit, reversely traces a source phase of the fault, and predicts a fault type, and outputs a comprehensive evaluation of health management of the overall access state.

[0073] Embodiment 3 As shown in Figure 7 The present application also provides an electronic device 100 for implementing the three-phase electric meter mutual inductor access state diagnosis method. The electronic device 100 comprises a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0074] The memory 101 can be used to store the computer program 103, and the processor 102 can realize the steps of the three-phase electric meter mutual inductor access state diagnosis method of embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0075] The memory 101 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 101 can include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device.

[0076] The at least one processor 102 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The processor 102 can be a microprocessor or can also be any conventional processor. The processor 102 is a control center of the electronic device 100, and is connected to various parts of the electronic device 100 through various interfaces and lines.

[0077] The memory 101 in the electronic device 100 stores a plurality of instructions to implement a three-phase electric meter mutual inductor access state diagnosis method, and the processor 102 can execute the plurality of instructions to implement: A non-power frequency excitation signal is injected into an A-phase loop of a three-phase electric meter mutual inductor to be diagnosed, and a digital sample stream on the output side is collected; the digital sample stream includes an A-phase mutual inductor secondary side current response signal, an A-phase voltage instantaneous waveform, an A-phase current instantaneous waveform, and B-phase and C-phase instantaneous waveforms; A reference signal is generated based on the non-power frequency excitation signal, and a net feature measurement vector is generated based on the current response signal and the reference signal; An equivalent complex impedance is generated based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; real-time temperature and real-time load are collected and input to a digital twin model together with the equivalent complex impedance, and the digital twin model outputs a dynamic standard vector; An auxiliary feature is generated based on the B-phase and C-phase instantaneous waveforms; the auxiliary feature, the dynamic standard vector and the net feature measurement vector are input to a reverse diagnosis model, the reverse diagnosis model performs nonlinear mapping on the A-phase loop, reversely traces a source phase of a fault, predicts a fault type, and outputs a health management comprehensive evaluation of an overall access state.

[0078] Embodiment 4 The modules / units integrated in the electronic device 100, if implemented in the form of software function units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, and read-only memory (ROM).

[0079] Those skilled in the art will appreciate that embodiments of the application can be supplied as a method, a system, or a computer program product. Therefore, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0080] The application is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0081] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0082] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 Figure 1

[0083] In the description of the present specification, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0084] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application rather than limit it, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand: the specific embodiments of the present application can still be modified or replaced by the equivalent, without departing from the spirit and scope of the present application, any modification or equivalent replacement, which should be covered in the protection scope of the claims of the present application.​​

Claims

1. A method of diagnosing the state of access of a three-phase electricity meter transformer, characterized by, The method comprises the following steps: Inject a non-power frequency excitation signal into the A-phase circuit of a three-phase electric meter transformer to be diagnosed, and collect a digital sample stream on the output side; the digital sample stream comprises an A-phase transformer secondary side current response signal, an A-phase voltage instantaneous waveform, an A-phase current instantaneous waveform, and B-phase and C-phase instantaneous waveforms; Generate a reference signal based on the non-power frequency excitation signal, and generate a net characteristic measurement vector based on the current response signal and the reference signal; Generate an equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; collect real-time temperature and real-time load, and input them together with the equivalent complex impedance into a digital twin model, which outputs a dynamic standard vector; Generate auxiliary characteristics based on the B-phase and C-phase instantaneous waveforms; input the auxiliary characteristics, the dynamic standard vector and the net characteristic measurement vector into a reverse diagnosis model, which performs nonlinear mapping on the A-phase circuit, reversely traces the source phase of the fault, predicts the fault type, and outputs a comprehensive evaluation of the health management of the overall access state.

2. The method of claim 1, wherein the method further comprises: The step of generating a reference signal based on the non-power frequency excitation signal and generating a net characteristic measurement vector comprises: Generate a reference signal based on the non-power frequency excitation signal, and perform virtual phase-locked demodulation and long integration on the current response signal and the reference signal to separate background power noise, extract in-phase components and quadrature components, and calculate net amplitude and relative net phase based on the in-phase components and the quadrature components to generate a net characteristic measurement vector. The specific implementation steps of the in-phase components and the quadrature components extraction comprise:

3. The method of claim 2, wherein the method further comprises: Track the non-power frequency excitation signal through a digital phase-locked loop to real-time adjust the phase and frequency, and generate in-phase reference signals and quadrature reference signals at the same frequency as the non-power frequency excitation signal; Perform product calculation on the current response signal with the in-phase reference signals and the quadrature reference signals respectively through virtual phase-locked demodulation to generate in-phase intermediate product signals and quadrature intermediate product signals; Virtual phase-locked demodulation can convert the target signal into a direct current component while converting all other frequency components into high-frequency alternating current components by performing product calculation on the current response signal and the reference signal; Define the integration point number based on the integration time, filter out high-frequency components and low-frequency noise different from the excitation frequency through integration in the preset integration time, and sum and average the in-phase intermediate product signals and the quadrature intermediate product signals at the integration points to generate in-phase components and quadrature components; Long integration is a low-pass filter that cancels out the average value of the alternating current components converted from background noise by long-time accumulation and average, thereby eliminating the average value of the alternating current components in numerical value. The step of generating an equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform, collecting real-time temperature and real-time load, and inputting them together with the equivalent complex impedance into a digital twin model, which outputs a dynamic standard vector, comprises:

4. The method of claim 1, wherein the method further comprises: Perform fast Fourier transform on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform to generate an equivalent complex impedance; Input the equivalent complex impedance into the digital twin model, and the digital twin model adjusts the internal model parameters according to the equivalent complex impedance and real-time working condition parameters, constructs a transfer function based on the internal model parameters, and outputs a dynamic standard vector. ​ 5. The method of claim 4, wherein the step of determining the status of the three-phase metering transformer access comprises the steps of: determining the status of the three-phase metering transformer access by comparing the three-phase metering transformer access status to the three-phase metering transformer access status stored in the memory. The specific generation steps of the equivalent complex impedance include: A continuous discrete sampling data is intercepted based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform, and a data window is generated; The data window is operated by fast Fourier transform to convert the data window to the frequency domain to obtain a voltage complex sequence and a current complex sequence; The corresponding frequency domain index is calculated according to a pre-set non-power frequency test excitation frequency; The voltage complex sequence and the current complex sequence are extracted through the frequency domain index to obtain complex components at the frequency domain index points, that is, complex voltage and complex current, which contain amplitude and phase information of the signals at the frequency domain index points; The complex voltage is divided by the complex current through complex division operation to obtain the equivalent complex impedance.

6. The method of claim 4, wherein the method further comprises: The specific generation process of the dynamic standard vector includes: The equivalent voltage and current of the transformer primary side at the current frequency are calculated based on the equivalent complex impedance through the transformation ratio relationship; Based on the equivalent voltage and current, the dynamic magnetic flux and excitation inductance of the core at the current frequency are obtained by using the nonlinear B-H curve, which is the inherent characteristic of the core material and describes the nonlinear relationship between the magnetic flux density and the magnetic field strength in the form of a function stored in the digital twin model; Based on the dynamic magnetic flux and excitation inductance, the adjustment data of the structural parameters and electromagnetic characteristics are calculated reversely through the ideal transformer equation and Kirchhoff's law; The dynamic transfer function of the current transformer circuit in the healthy state under the current operating condition and the non-power frequency excitation signal frequency is calculated by using the structural parameters, electromagnetic characteristics and equivalent reflection impedance theory; The dynamic standard vector containing amplitude and phase information is generated by calculating the dynamic transfer function at the non-power frequency excitation frequency.

7. The method of claim 1, wherein the method further comprises: The specific training steps of the reverse diagnosis model include: The reverse diagnosis model adopts a four-layer fully connected neural network structure, the first hidden layer contains 64 neurons, the second hidden layer contains 32 neurons, the third hidden layer contains 16 neurons, and the number of neurons in the output layer corresponds to the total number of all fault types and fault phase combinations; For each fault condition, the current response signal of phase A and the three-phase instantaneous waveform are collected by the electromagnetic transient simulation model to generate the training feature vector of the reverse diagnosis model; The training feature vector is input into the multi-layer fully connected layer, and the ReLU nonlinear activation function is used in each layer to accelerate convergence; The Softmax activation function is used to output the fault source phase and the fault type to generate a data set; The joint loss function is used to minimize the error between the predicted output and the real label, and the Adam optimizer is used for parameter iterative update.

8. A three-phase metering transformer connection state diagnostic device characterized by comprising: It includes: The acquisition module is used to inject a non-power frequency excitation signal into the A-phase circuit of the three-phase meter transformer to be diagnosed, and to collect the digital sample stream at the output side; the digital sample stream includes the current response signal of the A-phase transformer secondary side, the A-phase voltage instantaneous waveform, the A-phase current instantaneous waveform, and the instantaneous waveforms of the B-phase and C-phase; The first vector module is used to generate a reference signal based on the non-power frequency excitation signal, and to generate a net feature measurement vector based on the current response signal and the reference signal; A second vector module is configured to generate an equivalent complex impedance based on the A-phase voltage instantaneous waveform and the A-phase current instantaneous waveform; collect real-time temperature and real-time load, and input the equivalent complex impedance, the real-time temperature and the real-time load to a digital twin model, and the digital twin model outputs a dynamic standard vector; An evaluation module is configured to generate auxiliary features based on the instantaneous waveforms of the B-phase and the C-phase; input the auxiliary features, the dynamic standard vector and the net feature measurement vector to a reverse diagnosis model, and the reverse diagnosis model performs nonlinear mapping on the A-phase circuit, reversely traces a source phase of the fault, and predicts a fault type, and outputs a comprehensive evaluation of health management of an overall access state.

9. An electronic device, comprising: The processor is configured to execute a computer program stored in the memory to implement the three-phase electric meter mutual inductor access state diagnosis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction, and the at least one instruction is executed by the processor to implement the three-phase electric meter mutual inductor access state diagnosis method according to any one of claims 1 to 7.

Images