An inverting full-compensation transformer core and clamp grounding current monitoring device

Abstract

The present invention provides a monitoring device for the grounding current of a transformer core and clamping parts with inverting full compensation, belonging to the technical field of transformer cores and clamping parts. The device includes a current sensor, an amplifying circuit, an inverting circuit, a VI conversion circuit, and a monitoring host that are connected in sequence. The current collected by the current sensor is amplified by the amplifying circuit to obtain an amplified current signal. The amplified current signal is transmitted to the inverting circuit, and after being inverted by the inverting circuit, an inverted signal is obtained. The inverted signal is output to the VI conversion circuit to obtain a target signal. A collection module is arranged in the monitoring host for real-time collection of the amplified signal, the inverted signal, and the target signal. A phase difference analysis module and a grounding current compensation module are arranged in the monitoring host for calculating the phase error between the amplified signal and the target signal based on the amplified signal, the inverted signal, and the target signal, and for performing compensation calculation on the current corresponding to the target signal based on the phase difference to obtain a more accurate monitored current and output it.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer cores and clamping parts, and more specifically, relates to a monitoring device for the grounding current of a transformer core and clamping parts with inverse full compensation. Background Art

[0002] A transformer is a key device in the power system, responsible for stepping up or stepping down the voltage for transmission. During operation, a certain amount of grounding current will be generated in the core and clamping parts of the transformer. This grounding current is neither a power frequency current nor a harmonic current, but is caused by the non-linear magnetization characteristics and eddy current losses of the transformer core and clamping parts. The existence of the grounding current will have an adverse impact on the transformer itself and its auxiliary equipment, such as accelerating insulation aging, interfering with protection devices, and damaging the grounding loop. Therefore, accurately monitoring the magnitude of the grounding current of the transformer core and clamping parts is crucial for the safe operation and fault diagnosis of the transformer.

[0003] Large and medium-sized transformers generally adopt the method of separately grounding the core and clamping parts and the clamping parts. By detecting the current in the grounding wire of the core and clamping parts, multi-point grounding faults of the core and clamping parts can be effectively detected. Generally, a clamp meter can be used to measure whether there is current in the lead wire on the external grounding wire of the transformer core and clamping parts. When the transformer is operating normally, since there is no current loop formed, the current in the grounding wire is very small, at the milliampere level, generally not exceeding 0.1 A; when there is multi-point grounding, it is equivalent to having a shorted turn around the main magnetic flux of the core and clamping parts, and the circulating current flowing through depends on the relative position between the fault point and the normal grounding point, that is, the amount of magnetic flux enclosed in the shorted turn, generally reaching dozens of amperes. By measuring whether the current in the grounding lead exceeds the standard, it is possible to accurately judge whether there are multi-point grounding faults in the core and clamping parts. The advantage of the electrical method is unparalleled by other methods. It can help operation and maintenance personnel detect early signs of faults in a timely manner, eliminate potential accident hazards in a timely manner, simplify the operation and maintenance costs, and improve the maintenance efficiency, thus providing sufficient guarantee for the normal operation of the transformer. However, due to the complex spatial electromagnetic field around the transformer oil tank wall, the leakage magnetic field of the transformer itself and the electromagnetic fields of surrounding equipment will cause large fluctuations in the measurement results of the current in the grounding wire of the core and clamping parts. The randomness of the spatial electromagnetic field has a great interference on the measurement of the grounding current of the core and clamping parts. It is very difficult to have the condition of shielding the spatial electromagnetic field during the on-site measurement process. Therefore, the current measurement of the grounding current lacks accuracy. Summary of the Invention

[0004] In view of this, the present invention provides a monitoring device for the grounding current of a transformer core and clamping parts with inverse full compensation, which can solve the technical problem that the measurement of the grounding current of the transformer core and clamping parts is greatly interfered by the randomness of the spatial electromagnetic field and lacks accuracy.

[0005] The present invention is implemented as follows: A ground current monitoring device for a transformer core and clamping parts with inverting full compensation includes a current sensor, an amplifying circuit, an inverting circuit, a VI conversion circuit, and a monitoring host connected in sequence. The current sensor is used to collect the ground current of the transformer core and clamping parts as an initial current signal. The current collected by the current sensor is amplified by the amplifying circuit to obtain an amplified current signal. The amplified current signal is transmitted to the inverting circuit and inverted by the inverting circuit to obtain an inverted signal. The inverted signal is output to the VI conversion circuit to obtain a target signal. Both the amplifying circuit and the inverting circuit are also connected to the monitoring host through a collection module. The output of the VI conversion circuit is electrically connected to the monitoring host. A collection module is provided in the monitoring host for real-time collection of the amplified signal, the inverted signal, and the target signal. A phase difference analysis module and a ground current compensation module are provided in the monitoring host for calculating the phase error between the amplified signal and the target signal based on the amplified signal, the inverted signal, and the target signal, and for performing compensation calculation on the current corresponding to the target signal based on the phase difference to obtain a more accurate monitoring current and output it.

[0006] Among them, the phase difference analysis module is used to perform the following steps:

[0007] S11. Obtain the waveform data of the amplified current signal and the target signal;

[0008] S12. Preprocess the amplified current signal and the target signal;

[0009] S13. Use the cross-correlation function to calculate the phase difference between the two signals as the phase difference value.

[0010] Among them, the ground current compensation module is used to perform the following steps:

[0011] S21. Obtain the phase difference value output by the phase difference analysis module, the amplified current signal, and the target signal;

[0012] S22. Fine-tune the pre-trained current compensation model using the phase difference value, the waveform data of the amplified current signal, and the target signal;

[0013] S23. Use the fine-tuned current compensation model to input the phase difference value, the amplified current signal, and the target signal to obtain a compensation factor;

[0014] S24. Multiply the target signal by the compensation factor to obtain a compensated target signal;

[0015] S25. Extract the effective value of the ground current from the compensated target signal as the final monitoring current and output it.

[0016] Further, the step S11 specifically includes: reading the digital measurement values of the amplified current signal and the target signal from the acquisition module, where the digital measurement values are discrete data obtained after sampling and quantization; converting the discrete data into actual voltage values or current values to obtain sequences of the amplified current signal and the target signal; performing filtering processing on the sequences of the amplified current signal and the target signal to remove high-frequency noise and obtain smooth waveform data.

[0017] Further, the step S12 specifically includes: using wavelet transform to eliminate the DC component or low-frequency drift component existing in the waveform data; performing normalization processing on the waveform data after eliminating the DC component or low-frequency drift component to map the amplitude to the interval from -1 to 1 to obtain the standardized signal; detecting the period of the standardized signal and intercepting the standardized signal into one or more complete-period segments according to the detected period.

[0018] Further, the step S13 specifically includes: calculating the cross-correlation function between the standardized amplified current signal and the standardized target signal obtained in the step S12, where the cross-correlation function is a time-related function; finding the lag time of the maximum value point in the cross-correlation function; calculating the phase difference between the amplified current signal and the target signal according to the lag time corresponding to the maximum value point and the period of the signal.

[0019] Further, the step S22 specifically includes: using the phase difference analysis module to collect multiple groups of historical data, including phase difference, amplified current signal, and target signal, and combining them into a training data set. Among them, the amplified current signal and the target signal are used as network inputs, and the artificially labeled compensation factor is used as the label of the network output; using the training data set to fine-tune the pre-trained current compensation model.

[0020] Further, the step S23 specifically includes: inputting the phase difference, amplified current signal, and target signal into the fine-tuned current compensation model for forward propagation calculation to obtain the network output, that is, the compensation factor.

[0021] Further, the step S25 specifically includes: performing fast Fourier transform on the compensated target signal to obtain the frequency-domain signal; finding the amplitude of the fundamental wave component in the frequency-domain signal; calculating the effective current value corresponding to the fundamental wave component as the final monitored current according to the relationship between current and voltage.

[0022] Further, the specific steps of the training process of the current compensation model include: collecting the measured data of the transformer under different loads and different grounding current conditions, including the amplified current signal, the target signal, and the corresponding standard grounding current value; preprocessing the collected data to obtain a purified data set; extracting the amplified current signal and the target signal from the purified data set as inputs, and the compensation factor obtained by calculating the standard grounding current value through the compensation factor as the output label to form an input-output sample pair; inputting the input-output sample pair into the LSTM network in small batches for training, monitoring the loss value and performance indicators of the model on the validation set, and stopping training and saving the model parameters when the requirements are met.

[0023] Compared with the prior art, the beneficial effects of an inverting full-compensation transformer core and clamp grounding current monitoring device provided by the present invention are as follows:

[0024] 1. High measurement accuracy

[0025] Through neural network modeling, the mapping relationship between the current signal and the grounding current can be accurately learned, eliminating the errors in traditional estimation methods, thereby greatly improving the accuracy of grounding current measurement.

[0026] 2. Strong anti-interference ability

[0027] By adopting the inverting and compensation technology, the influence of external electromagnetic interference and harmonic components can be effectively suppressed, making the measurement results more reliable.

[0028] In summary, the technical solution of the present invention solves the technical problems that the measurement of the grounding current of the transformer core and clamp is greatly interfered by the randomness of the space electromagnetic field and lacks accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0030] Figure 1 Schematic diagram of the composition of an inverting full-compensation transformer core and clamp grounding current monitoring device;

[0031] Figure 2 Flowchart of the steps executed by the phase difference analysis module;

[0032] Figure 3 Flowchart of the steps executed by the grounding current compensation module;

[0033] Figure 4Circuit design diagram of an inverting circuit;

[0034] Figure 5 It is the design diagram of a VI conversion circuit. Specific implementation manner

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0036] As Figure 1 shown, it is a schematic diagram of the composition of a transformer core and clamp grounding current monitoring device with inverting full compensation provided by the present invention. This device includes a current sensor, an amplifying circuit, an inverting circuit, a VI conversion circuit, and a monitoring host connected in sequence. The current sensor is used to collect the grounding current of the transformer core and clamp as an initial current signal. The current collected by the current sensor is amplified by the amplifying circuit to obtain an amplified current signal. The amplified current signal is transmitted to the inverting circuit, and after being inverted by the inverting circuit, an inverted signal is obtained. The inverted signal is output to the VI conversion circuit to obtain a target signal. The amplifying circuit and the inverting circuit are also connected to the monitoring host through a collection module. The output of the VI conversion circuit is electrically connected to the monitoring host. A collection module is provided in the monitoring host for real-time collection of the amplified signal, the inverted signal, and the target signal. A phase difference analysis module and a grounding current compensation module are provided in the monitoring host for calculating the phase error between the amplified signal and the target signal based on the amplified signal, the inverted signal, and the target signal, and for performing compensation calculation on the current corresponding to the target signal based on the phase difference to obtain a more accurate monitoring current and output it.

[0037] As Figure 2 shown, the phase difference analysis module is used to perform the following steps:

[0038] S11. Obtain the waveform data of the amplified current signal and the target signal;

[0039] S12. Perform preprocessing on the amplified current signal and the target signal;

[0040] S13. Use the cross-correlation function to calculate the phase difference between the two signals as the phase difference value.

[0041] As Figure 3 shown, the grounding current compensation module is used to perform the following steps:

[0042] S21. Obtain the phase difference value output by the phase difference analysis module, as well as the amplified current signal and the target signal;

[0043] S22. Use the phase difference value, the waveform data of the amplified current signal, and the target signal to fine-tune a pre-trained current compensation model;

[0044] S23. Using the fine-tuned current compensation model, input the phase difference, the amplified current signal, and the target signal to obtain a compensation factor;

[0045] S24. Multiply the target signal by the compensation factor to obtain a compensated target signal;

[0046] S25. Extract the effective value of the grounding current from the compensated target signal as the final monitored current and output it.

[0047] The following is a detailed description of the specific implementation manners of the above steps:

[0048] First, the detailed descriptions of the amplifier circuit, the inverting circuit, and the VI conversion circuit are as follows:

[0049] The amplifier circuit consists of an operational amplifier, a feedback resistor, and an input resistor. The initial current signal collected by the current sensor is input to the inverting input terminal of the operational amplifier through the input resistor, and the operational amplifier amplifies it according to the ratio of the feedback resistor to the input resistor to obtain an amplified current signal. The amplifier circuit is used to amplify a relatively small initial current signal to an appropriate amplitude for subsequent circuit processing. A conventional amplifier circuit can be used for the amplifier circuit.

[0050] The inverting circuit consists of an operational amplifier, a feedback resistor, and an input resistor. The amplified current signal output by the amplifier circuit is input to the non-inverting input terminal of the operational amplifier through the input resistor, the inverting input terminal is grounded, and the operational amplifier inverts and amplifies the input signal according to the ratio of the feedback resistor to the input resistor and outputs an inverted signal. The function of the inverting circuit is to reverse the polarity of the amplified current signal to prepare for subsequent phase difference calculation. As Figure 4 shown, the inverting circuit module: According to the hardware design flow chart, after the signal output by the mutual inductor is sampled by the sampling circuit, it needs to be inverted. Design an inverting circuit; in the figure, Ref2 is the feedback resistor, Rf2 is the input resistor, and Rp2 is the balancing resistor. Select the value of Rf2 according to the matching requirements with the previous stage (amplifier circuit). When the value of Rf2 is relatively small, the influence of the bias current, offset current, and their drifts can be weakened. A larger closed-loop amplification factor Kf can weaken the influence of the offset voltage and its drift. However, the larger Kf is, the narrower the closed-loop frequency band is. According to actual needs, taking into account the requirements of drift error and closed-loop bandwidth, select the value of the closed-loop amplification factor Kf to be 0 to 100. Usually, the resistors of Rf2 and Ref2 are from 1KΩ to 1MΩ. And try to increase Kf by selecting a small value of Rf2 as much as possible. When the resistance value exceeds 1MΩ, it is difficult to ensure the stability of the resistance value, and the absolute error of the resistance value is relatively large. Therefore, determine the input resistor through design:

[0051] Rf 2 = Ref 2 = 11.8 KΩ; The balancing resistor RP2 = Ref 2 = Rf2 = 5.9 KΩ;

[0052] The VI conversion circuit consists of an operational amplifier, a feedback resistor, and an input resistor. The inverted signal output by the inverting circuit is input to the inverting input terminal of the operational amplifier through the input resistor, the non-inverting input terminal is grounded, and the operational amplifier amplifies the inverted signal according to the ratio of the feedback resistor and the input resistor to output the target signal. The function of the VI conversion circuit is to convert the inverted signal into a voltage signal for convenient acquisition and processing by the monitoring host. The VI conversion circuit can adopt a conventional VI conversion circuit or Figure 5 the designed VI conversion circuit.

[0053] Specific implementation manners of steps S11 - S13:

[0054] Step S11 acquires the waveform data of the amplified current signal and the target signal, including the following sub-steps:

[0055] 1) Read the digital measurement values of the amplified current signal and the target signal from the acquisition module. These digital measurement values are discrete data after sampling and quantization;

[0056] 2) Convert the read digital measurement values into actual voltage values or current values to obtain the time series of the amplified current signal i amp (t) and the target signal v tgt (t), where t represents the sampling moment;

[0057] 3) Perform filtering on i amp (t) and v tgt (t) to remove high-frequency noise and obtain smoother waveform data and

[0058] Step S12 preprocesses the amplified current signal and the target signal, including the following sub-steps:

[0059] 1) Use wavelet transform or other baseline drift removal algorithms to eliminate and the possible DC components or low-frequency drift components in;

[0060] 2) Perform normalization on and to map the signal amplitude to the interval [-1, 1] to obtain the normalized signals and

[0061] 3) Detect and The period. According to the detected period, the signal is intercepted into one or more complete period segments to facilitate subsequent calculations.

[0062] Step S13 calculates the phase difference between two signals as the phase difference value using the cross-correlation function, including the following sub-steps:

[0063] 1) Calculate and of the cross-correlation function where τ is the lag time;

[0064] 2) Find the lag time τ corresponding to the maximum point in the cross-correlation function r(τ), i.e., max , that is

[0065] 3) According to τ max and the period T of the signal, calculate the phase difference φ = 2πτ max / T.

[0066] φ is the phase difference value between the amplified current signal and the target signal, and φ is output to the ground current compensation module.

[0067] The specific implementation manners of steps S21 - S25 and the current compensation model:

[0068] Step S21 obtains the phase difference value φ output by the phase difference value analysis module, the amplified current signal and the target signal

[0069] Step S22 fine-tunes the pre-trained current compensation model using the waveform data of the phase difference value φ, the amplified current signal and the target signal The current compensation model adopts the structure of a long short-term memory network (LSTM) and can better process time series data. The input of the current compensation model includes the amplified current signal

[0070] the target signal the target signal and the phase difference value φ, and the output is the compensation factor k(t).

[0071] The specific implementation manner of this step is as follows:

[0072] 1) Combine and φ into a training data set, where and are used as the network input, and the compensation factor k(t) is used as the label of the network output;

[0073] 2) Fine-tune the pre-trained model using the training dataset, and through the backpropagation algorithm and gradient descent optimization, make the model achieve optimal performance under the current data conditions.

[0074] The process of obtaining the label k(t) of the compensation factor is as follows:

[0075] ① Calculate the phases of the fundamental wave components of and by fast Fourier transform (FFT), and set them as θ amp and θ tgt ;

[0076] ② Calculate the phase difference Δθ = θ amp - θ tgt ;

[0077] ③ Use the phase difference Δθ to correct the target signal to obtain the corrected target signal

[0078] ④ Calculate the effective value of the fundamental wave component of the corrected target signal

[0079] ⑤ Take the effective value I of the fundamental wave component of amp as the standard value, then the compensation factor

[0080] 3) Set the hyperparameters of the LSTM network, such as the time step, the number of hidden layer units, the learning rate, etc. The setting of these hyperparameters can refer to empirical values or be obtained through tuning;

[0081] 4) Input the training data in small batches, perform end-to-end training on the LSTM network, use loss functions such as mean squared error, and adopt optimization algorithms such as Adam for parameter update;

[0082] 5) Monitor the loss value and performance metrics of the model on the validation set. When the performance meets the requirements or the metrics on the validation set no longer improve significantly, stop training and save the model parameters.

[0083] In step S23, use the fine-tuned current compensation model to input the phase difference φ and the amplified current signal and the target signal to obtain the compensation factor k(t). The specific implementation method is as follows:

[0084] 1) Preprocess and φ according to the model input requirements, such as standardization, segmentation, etc.;

[0085] 2) Input the preprocessed data into the fine-tuned LSTM network for forward propagation calculation to obtain the network output, i.e. the compensation factor k(t).

[0086] Step S24: the target signal Multiply by the compensation factor k(t) to get the compensated target signal

[0087] Step S25: The target signal after compensation is The effective value of the ground current is extracted as the final monitoring current and output. The specific implementation is as follows:

[0088] 1) Yes Perform fast Fourier transform to obtain the frequency domain signal

[0089] 2) In Find the amplitude of the fundamental component in

[0090] 3) According to the relationship between current and voltage, calculate the effective value of the current corresponding to the fundamental component Where R is a known resistance or impedance value;

[0091] 4) As the final monitoring current output.

[0092] The training process of the current compensation model:

[0093] The training data set of the current compensation model needs to contain enough input and output sample pairs to cover the situations under different working conditions. The data set acquisition process is as follows:

[0094] 1) Collect the measured data of the transformer under different loads and different grounding current conditions, including the amplified current signal, the target signal, and the corresponding standard grounding current value;

[0095] 2) Preprocess the collected data, including removing outliers, interpolating missing values, smoothing filtering, etc., to obtain a purified data set;

[0096] 3) Extract input and output sample pairs under different working conditions from the purified data set, i.e., the amplified current signal and the target signal are used as input, and the compensation factor obtained by the standard ground current value through the above compensation factor calculation process is used as the output label;

[0097] 4) Input the extracted input and output sample pairs into the LSTM network in small batches for training, using the same training strategy as the above fine-tuning;

[0098] 5) Monitor the loss value and performance metrics of the model on the validation set. Stop training and save the model parameters when the requirements are met.

[0099] The structure of the above current compensation model is an LSTM network, which includes the following main components:

[0100] 1) Embedding layer: Encode the input amplified current signal, target signal, and phase difference, and embed them into a low-dimensional dense vector space.

[0101] 2) LSTM layer: The core layer composed of multiple LSTM units, which can effectively capture the long-term dependencies of the input sequence.

[0102] 3) Fully connected layer: Map the output of the LSTM layer to the desired output dimension to generate a sequence of compensation factors.

[0103] 4) Loss function layer: Use loss functions such as mean squared error to calculate the difference between the network output and the label.

[0104] 5) Optimizer: Such as the Adam optimizer, adjust the network parameters according to the feedback of the loss function, so that the model output gradually approaches the label.

[0105] During the training process, regularization strategies such as Dropout and BN layers can be adopted to avoid overfitting of the model. Different optimizers and learning rate scheduling strategies can also be tried to accelerate the training convergence. In addition, for some hyperparameters of LSTM, such as the time step and the number of hidden layer units, etc., they need to be tuned according to the characteristics of the specific task to obtain the best performance.

[0106] The following is the explanation of all variables, subscripts, and constants that appear in the specific implementation description:

[0107] φ - Phase difference, representing the phase difference (phase angle) between the amplified current signal and the target signal.

[0108] - Amplified current signal, which is a function of time t, representing the time series of the current signal amplified by the amplification circuit.

[0109] - Target signal, which is a function of time t, representing the time series of the voltage signal obtained after the grounding current to be monitored passes through the inverting circuit and the VI conversion circuit.

[0110] k(t) - Compensation factor, which is a function of time t and is used to compensate the target signal.

[0111] - Compensated target signal, which is a function of time t and is equal to

[0112] - The frequency-domain representation of the compensated target signal, which is a function of frequency f and is obtained by performing a fast Fourier transform.

[0113] - The amplitude of the fundamental component of the compensated target signal.

[0114] - The effective value of the final monitored current, which is calculated from the amplitude of the fundamental component and is equal to calculated as

[0115] R - A known resistance value or impedance value used to calculate the effective current value from the amplitude of the fundamental component.

[0116] Variable without subscript. i amp (t) - The time-series signal output by the amplifier circuit, i.e., the amplified current signal, corresponding to but without the modification of the tilde symbol "~".

[0117] v tgt (t) - The time-series signal output by the VI conversion circuit, i.e., the target signal, corresponding to but without the modification of the tilde symbol "~".

[0118] v inv (t) - The time-series signal output by the inverting circuit, i.e., the inverted signal, which is the inversion of i amp (t).

[0119] i in (t) - The time series of the initial current signal collected by the current sensor.

[0120] i gnd - The standard ground current value used as the label of the model output when training the current compensation model.

[0121] - The estimated effective value of the ground current obtained by model prediction, corresponding to the final monitored current output

[0122] In addition, some descriptive variables also appear in the claims:

[0123] φ - Phase difference

[0124] t - Time variable

[0125] f - Frequency variable

[0126] There are no variables with subscripts.

[0127] Constants include the resistance values used in the amplifier circuit, inverting circuit, and VI conversion circuit, such as feedback resistors, input resistors, etc. Usually represented by R, it may have subscripts to distinguish the resistance values in different circuits or positions.

[0128] Specifically, the principle of the present invention is as follows:

[0129] The principle of the solution of the present invention is mainly based on the following key technologies:

[0130] 1. Inverting full compensation principle

[0131] Through the inverting circuit and compensation algorithm, the interference components unrelated to the ground current can be eliminated to the greatest extent, and the pure ground current component can be extracted. Specifically:

[0132] (1) The initial current signal i in (t) is amplified by the amplifier circuit to obtain the amplified current signal i amp (t);

[0133] (2) i amp (t) passes through the inverting circuit to obtain the inverted signal v inv (t), which cancels periodic interferences such as harmonics;

[0134] (3) v inv (t) passes through the VI conversion circuit to obtain the target signal v tgt (t), which cancels non-periodic interferences such as DC components;

[0135] (4) In the time domain, the target signal v tgt (t) only contains components related to the ground current component;

[0136] (5) A compensation algorithm is introduced to perform phase and amplitude compensation on v tgt (t) to obtain the effective value of the ground current.

[0137] 2. Deep learning modeling principle

[0138] Since the mapping relationship between the ground current and the target signal is a complex non-linear function relationship, it is difficult to describe with an analytical model. The solution of the present invention adopts the deep learning method, uses recursive neural network models such as long short-term memory network (LSTM), and automatically learns this mapping relationship through a large amount of data training. Specifically, it includes:

[0139] (1) Collect a large number of input-output data pairs under different working conditions as the training set, where the input is the amplified current signal, the target signal, and the phase difference, and the output is the standard ground current;

[0140] (2) Use the training set to perform supervised training on the LSTM network, so that the network learns how to predict the effective value of the ground current from the input.

[0141] (3) During measurement, input the currently collected data into the trained LSTM network, and an accurate estimated value of the real-time grounding current can be obtained.

[0142] 3. Principle of Phase Difference Compensation

[0143] During the training and application processes, due to factors such as circuit delay, there is a certain phase difference between the amplified current signal and the target signal. This solution calculates the phase difference value through the cross-correlation function and uses it as an additional input to the LSTM network to explicitly compensate for this phase difference, thereby improving the accuracy and robustness of the modeling.

[0144] 4. Principle of Hardware Integration

[0145] The present invention also relates to the principle of hardware integration, including:

[0146] (1) Principle of Inverting Circuit

[0147] It is still an inverting amplifier circuit of an operational amplifier. By inverting the output V of the amplifier circuit o by -G times, that is, v inv (t) = -G·V o = -G 2 ·i in R in . The inverted signal v inv (t) cancels out the periodic interference component.

[0148] (2) Principle of VI Conversion Circuit

[0149] Amplify the inverted signal v inv (t) by G times again to obtain the final target signal v tgt (t) = G·v inv (t) = -G 3 ·i in R in . The target signal v tgt (t) is a voltage signal proportional to the initial current i in (t), and non-periodic interferences such as DC components are eliminated.

[0150] Through the above series of analog circuit processes, the collected original current signal is amplified, inverted, and voltage-converted, thereby obtaining a pure target voltage signal v tgt (t) proportional to the grounding current. Combining with the deep learning algorithm, the effective value of the grounding current can be accurately calculated.

[0151] To better understand the technical solution of the present invention, the following will be described in detail with specific embodiments.

[0152] This embodiment is based on a 250MVA main transformer, and the typical value of the grounding current of this transformer during operation is about 5A. In order to accurately monitor the grounding current of the transformer core and clamping parts, the monitoring method of the present invention is adopted.

[0153] 1. Hardware circuit part

[0154] The hardware part of the circuit consists of components such as a current sensor, an amplification circuit, an inverting circuit, a VI conversion circuit, an A / D converter, and a microcontroller. Among them:

[0155] (1) The current sensor is a closed-loop Hall current sensor with a turns ratio of 1:1000 and an output voltage of ±4V;

[0156] (2) The amplification circuit consists of an operational amplifier OPA180 to form a non-inverting amplifier with an amplification factor of 100 times;

[0157] (3) The inverting circuit consists of an OPA180 of the same model to form an inverting amplifier with an amplification factor of -100 times;

[0158] (4) The VI conversion circuit uses an OPA820 amplifier with an amplification factor of 100 times;

[0159] (5) The A / D converter is an AD9235 analog-to-digital conversion chip with a 12-bit accuracy and a sampling rate of 100kSPS;

[0160] (6) The microcontroller is an STM32H743 processor with a 32-bit ARM Cortex-M7 core;

[0161] (7) The liquid crystal display is a 4.3-inch TFT-LCD with a resolution of 320x480.

[0162] The power supply of the hardware circuit is DC 24V.

[0163] 2. Deep learning model training

[0164] (1) Data acquisition

[0165] First, under laboratory conditions, by injecting standard current signals with different amplitudes and frequencies, the working state of the grounding current is simulated. The input current signal, the target voltage signal, the phase difference, and the true value of the corresponding standard grounding current are collected respectively to construct a training set. Specifically, 100 groups of data are collected, and each group contains:

[0166] - Input current signal sequence I amp (t), with a length of 1000 points;

[0167] - Output target voltage signal sequence v tgt (t), with a length of 1000 points;

[0168] - Phase difference, ranging from -180° to 180°;

[0169] - Actual value of the standard grounding current, ranging from 0.1 A to 20 A.

[0170] These data are used as the training set for subsequent model training.

[0171] (2) Normalization and segmentation

[0172] Normalize the input and output data to the range between -1 and 1, and then segment the time series into multiple samples with a step size of 50.

[0173] (3) Model structure

[0174] Adopt the Long Short-Term Memory Network (LSTM) as the deep learning model, which includes the following layer structures:

[0175] Input layer -> Embedding layer (embedding dimension 128) -> LSTM layer (256 hidden units, 2 layers) -> Fully connected layer (128 nodes) -> Dropout layer (probability 0.3) -> Fully connected layer (64 nodes) -> Dropout layer (probability 0.3) -> Fully connected layer (1 node, no activation)

[0176] The input dimension is 3 (amplified current, target voltage, phase difference), and the output dimension is 1 (grounding current value).

[0177] (4) Model training

[0178] - Loss function: Smooth L1 loss

[0179] - Optimizer: Adam, initial learning rate is 0.001

[0180] - Regularization: L2 weight decay coefficient is 1e-4

[0181] - Number of training epochs: 2000 epochs

[0182] - Batch size: 32

[0183] After 2000 epochs of training, the loss of the model on the validation set tends to be stable, and the root mean square error of the grounding current prediction reaches 0.12 A, meeting the actual requirements.

[0184] 3. On-site integrated application

[0185] During on-site use, the current sensor measures the current on the grounding lead and outputs it to the amplifier circuit. After amplification, inversion, and VI conversion, the target voltage signal v is obtained tgt(t). After the signal is A / D converted, it is digitally processed by the microcontroller. The phase difference from the target voltage signal is calculated every 50 points, and is input into the pre-loaded deep learning model together with the amplified current signal sequence and the phase difference.

[0186] The deep learning model predicts the effective value of the grounding current in real time according to the input data The final prediction result is sent to the liquid crystal display for display through serial communication, and can also be sent to the remote monitoring center by the Wi-Fi module.

[0187] 4. On-site testing and result analysis

[0188] On-site grounding current monitoring is carried out on the 250MVA main transformer in operation, and the following working conditions are investigated respectively:

[0189] (1) Rated condition

[0190] When the transformer is operating under rated voltage and current conditions, the grounding current measured by the monitoring system is 4.87A, which is basically the same as the typical value of 5A.

[0191] (2) Light load condition

[0192] When the transformer is under 50% light load condition, the monitored grounding current is 3.26A, which is different from the rated condition. This is consistent with the fact that there is a certain correlation between the grounding current and the load current in theory.

[0193] (3) Overload condition

[0194] When the transformer load current is 120% of the rated value, the grounding current given by the monitoring system is 6.35A, which is about 30% higher than that under the rated condition.

[0195] (4) Harmonic condition

[0196] In order to investigate the anti-interference ability of the proposed solution of the present invention, 10% of the 5th and 7th harmonic components are artificially injected during monitoring. The results show that even with large harmonic interference, the grounding current value output by the monitoring system is 4.91A, with only a 0.8% deviation from 4.87A without harmonics, indicating that the proposed solution can effectively suppress the influence of harmonics on monitoring.

[0197] (5) Temperature rise condition

[0198] During the process of monitoring the temperature rise of the transformer winding, when the temperature rises from 45°C to 75°C, it is monitored that the grounding current gradually increases from 4.62 A to 5.24 A. This is consistent with the theoretical analysis result that the grounding current will increase with the temperature rise. In the above embodiments, the iron core and clamping parts can be regarded as a whole or two independent individuals, which will not affect the implementation of the steps of the present invention. That is, in the steps of the embodiment, each step of the present invention is implemented for the individual iron core or individual clamping parts.

[0199] As described above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should be covered within the protection scope of the present invention.

Claims

1. A transformer core and clamp grounding current monitoring device with reverse full compensation, characterized in that: It includes a current sensor, an amplifier circuit, an inverting circuit, a VI conversion circuit and a monitoring host connected in sequence, wherein the current sensor is used to collect the grounding current of the core and the clamp of the transformer as the initial current signal, the current collected by the current sensor is amplified by the amplifier circuit to obtain an amplified current signal, the amplified current signal is transmitted to the inverting circuit, and is inverted by the inverting circuit to obtain an inverted signal, and the inverted signal is output to the VI conversion circuit to obtain a target signal; the amplifier circuit and the inverting circuit are also connected to the monitoring host through an acquisition module, and the output of the VI conversion circuit is electrically connected to the monitoring host, the monitoring host is provided with an acquisition module for real-time acquisition of the amplified current signal, the inverted signal and the target signal, the monitoring host is provided with a phase difference analysis module and a grounding current compensation module, which are used to calculate the phase difference between the amplified current signal and the target signal based on the amplified current signal, the inverted signal and the target signal, and to perform compensation calculation on the current corresponding to the target signal based on the phase difference, so as to obtain a more accurate monitoring current and output it; The ground current compensation module is used to perform the following steps: S21, obtaining the phase difference value output by the phase difference analysis module and the amplified current signal and the target signal; S22, fine-tuning a pre-trained current compensation model using the phase difference value, the amplified current signal, and the waveform data of the target signal; S23, using the fine-tuned current compensation model, inputting the phase difference value and the amplified current signal and the target signal to obtain a compensation factor; S24, multiplying the target signal by the compensation factor to obtain a compensated target signal; S25. Extract the effective value of the ground current from the compensated target signal as the final monitoring current and output it.

2. The inverting fully compensated transformer core and clamp grounding current monitoring device according to claim 1, characterized in that: The phase difference analysis module is used to perform the following steps: S11, obtaining waveform data of the amplified current signal and the target signal; S12, preprocessing the amplified current signal and the target signal; S13. Calculate the phase difference between the two signals using a cross-correlation function as a phase difference value.

3. The transformer core and clamp grounding current monitoring device with reverse phase full compensation according to claim 2 is characterized in that: The step S11 specifically includes: reading digital measurement values ​​of the amplified current signal and the target signal from the acquisition module, the digital measurement values ​​being sampled and quantized to obtain discrete data; converting the discrete data into actual voltage values ​​or current values ​​to obtain a sequence of the amplified current signal and the target signal; filtering the sequence of the amplified current signal and the target signal to remove high-frequency noise and obtain smooth waveform data.

4. The transformer core and clamp grounding current monitoring device with reverse phase full compensation according to claim 2 is characterized in that: The step S12 specifically includes: using wavelet transform to eliminate the DC component or low-frequency drift component in the waveform data; normalizing the waveform data after eliminating the DC component or low-frequency drift component, mapping the amplitude to the interval of -1 to 1, and obtaining a standardized signal; detecting the period of the standardized signal, and according to the detected period, cutting the standardized signal into one or more complete period segments.

5. The transformer core and clamp grounding current monitoring device with reverse phase full compensation according to claim 4 is characterized in that: The step S13 specifically includes: calculating the cross-correlation function between the standardized amplified current signal obtained in step S12 and the standardized target signal, wherein the cross-correlation function is a time-related function; finding the lag time of the maximum point in the cross-correlation function; and calculating the phase difference between the amplified current signal and the target signal according to the lag time corresponding to the maximum point and the signal period.

6. The reverse phase fully compensated transformer core and clamp grounding current monitoring device according to claim 1, characterized in that: The step S22 specifically includes: using a phase difference analysis module to collect multiple groups of historical data, including phase differences, amplified current signals, and target signals, and combining them into a training data set, wherein the amplified current signal and the target signal are used as network inputs, and manually labeled compensation factors are used as labels for network outputs; and using the training data set to fine-tune the current compensation model.

7. The transformer core and clamp grounding current monitoring device with reverse phase full compensation according to claim 1 is characterized in that: The step S23 specifically includes: inputting the phase difference value, the amplified current signal, and the target signal into the fine-tuned current compensation model for forward propagation calculation to obtain a network output, namely a compensation factor.

8. The reverse phase fully compensated transformer core and clamp grounding current monitoring device according to claim 1, characterized in that: The step S25 specifically includes: performing a fast Fourier transform on the compensated target signal to obtain a frequency domain signal; finding the amplitude of the fundamental component in the frequency domain signal; and calculating the effective value of the current corresponding to the fundamental component as the final monitoring current based on the relationship between the current and the voltage.

9. The transformer core and clamp grounding current monitoring device with reverse phase full compensation according to claim 1, characterized in that: The specific steps of the training process of the current compensation model include: collecting measured data of the transformer under different loads and different grounding current conditions, including an amplified current signal, a target signal, and a corresponding standard grounding current value; preprocessing the collected data to obtain a purified data set; extracting the amplified current signal and the target signal from the purified data set as input, and the compensation factor obtained by calculating the standard grounding current value through the compensation factor as an output label to form an input-output sample pair; inputting the input-output sample pair into the LSTM network in a small batch form for training, monitoring the loss value and performance index of the model on the verification set, and stopping the training and saving the model parameters when the requirements are met.

Images