Nuclear power station medium-voltage power supply grounding intelligent line selection method

Abstract

The invention provides an intelligent grounding line selection method for a medium-voltage power supply of a nuclear power station. The method comprises the following steps: acquiring a three-phase current signal and a zero-sequence voltage signal of each line in real time; performing multi-scale wavelet decomposition on the three-phase current signal, and extracting a transient current feature vector; performing phase modulation on the transient current feature vector based on the zero-sequence voltage abrupt change moment; performing frequency spectrum demodulation on the modulated signal, and calculating a frequency spectrum characteristic value; constructing a multi-dimensional logic judgment matrix in combination with the zero-sequence voltage amplitude and the frequency spectrum characteristic value; calculating a grounding fault probability score of each line according to the weight distribution; and dynamically adjusting the judgment threshold and outputting a final line selection result. Accuracy and reliability of ground fault detection of the medium-voltage power supply of the nuclear power station can be improved, and adaptability of the system under complex working conditions is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of grounding fault detection for medium - voltage power supply systems in nuclear power plants. More specifically, the present invention relates to an intelligent grounding line selection method for medium - voltage power supply in nuclear power plants. Background Art

[0002] In the medium - voltage power supply system of a nuclear power plant, grounding fault detection is a key link to ensure the safe operation of the power system. Existing grounding fault line selection methods mainly rely on steady - state current and voltage characteristics, such as the zero - sequence current method, zero - sequence power direction method, etc. Although these methods can identify fault lines to a certain extent, their accuracy and reliability will be significantly reduced under complex working conditions, such as when there is harmonic interference in the system, the transient process is complex, or the grounding fault characteristics are not obvious. In addition, most traditional methods are based on fixed - threshold judgment, which is difficult to adapt to the changes in the system operation state, resulting in an increased risk of misjudgment or missed judgment.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: the existing methods do not extract transient characteristics sufficiently and cannot effectively cope with grounding faults in complex transient processes; at the same time, there is a lack of a dynamic adjustment mechanism, which is difficult to adapt to changes in the system operation state and environmental factors, resulting in insufficient line selection accuracy and adaptability. Summary of the Invention

[0004] The present invention provides an intelligent grounding line selection method for medium - voltage power supply in nuclear power plants, including:

[0005] S1. Real - time collect the three - phase current signals and zero - sequence voltage signals of each line in the medium - voltage power supply system of the nuclear power plant;

[0006] S2. Perform multi - scale wavelet decomposition on the collected three - phase current signals to extract the transient current feature vectors of each line;

[0007] S3. Based on the mutation moment of the zero - sequence voltage signal, perform phase modulation on the transient current feature vectors to generate modulated signals;

[0008] S4. Perform spectrum demodulation on the modulated signals to calculate the spectrum eigenvalue of each line;

[0009] S5. Combine the zero - sequence voltage amplitude and the spectrum eigenvalue to construct a multi - dimensional logical judgment matrix;

[0010] S6. According to the weight distribution of each dimension in the logical judgment matrix, calculate the grounding fault probability score of each line;

[0011] S7. Dynamically adjust the decision threshold of the grounding fault probability score and output the final line selection result.

[0012] Further, the step S1 includes:

[0013] S11. Synchronously obtain the instantaneous values I a (t), I b (t), I c (t) of the three-phase currents of each line through a current transformer;

[0014] S12. Collect the instantaneous zero-sequence voltage U0(t) of the bus through a voltage transformer;

[0015] S13. Determine the sampling frequency f s according to the preset sampling theorem, satisfying f s ≥ 10f max ;

[0016] where, I a (t), I b (t), I c (t) are the instantaneous values of the three-phase currents respectively, U0(t) is the instantaneous zero-sequence voltage, f s is the sampling frequency, and f max is the highest frequency component of the signal.

[0017] Further, the specific steps of the multi-scale wavelet decomposition in step S2 include:

[0018] S21. Select the Daubechies wavelet basis function to decompose the three-phase current signal into N layers to obtain the wavelet coefficients W j (k) at each scale;

[0019] S22. Perform threshold denoising on the wavelet coefficients of the j-th layer and reconstruct the transient current signals I′ a (t), I′ b (t), I′ c (t);

[0020] S23. Calculate the transient current feature vector as shown in the following formula:

[0021] F i = [ΔI max , T r , E d

[0022] where, j is the decomposition layer number, k is the coefficient serial number, Wj(k) is the k-th wavelet coefficient of the j-th layer, I′ a (t), I′ b (t), I′ c (t) are the reconstructed transient current signals, ΔI max is the transient current peak value, T r is the rise time, E d is the energy density, and F i ​It is the transient current feature vector.

[0023] Further, the phase modulation process in step S3 is as follows:

[0024] S31. Determine the modulation window time as [t0 - ΔT, t0 + ΔT] according to the zero-sequence voltage mutation moment t0;

[0025] S32. Perform phase encoding on the transient current feature vector F i to generate a modulation signal, as shown in the following formula:

[0026]

[0027] where t0 is the zero-sequence voltage mutation moment, ΔT is the half-width of the modulation window time, A is the amplitude of the modulation signal, f c is the carrier frequency, and is the phase modulation function based on the transient current feature vector F i of.

[0028] Further, the spectrum demodulation in step S4 includes:

[0029] S41. Perform a fast Fourier transform on the modulated signal to obtain the spectrogram P(f);

[0030] S42. Calculate the spectrum eigenvalue, as shown in the following formula:

[0031]

[0032] S43. Generate a binary feature flag bit F th according to the comparison result between the spectrum eigenvalue Q and the preset spectrum feature threshold Q q ;

[0033] where Q is the spectrum eigenvalue, Δf is the characteristic frequency band width, f c is the carrier frequency, P(f) is the signal power spectral density, f s is the sampling frequency, Q th is the preset spectrum feature threshold, and F q is the binary feature flag bit.

[0034] Further, the construction process of the multi-dimensional logical judgment matrix in step S5 is as follows:

[0035] S51. Divide the zero-sequence voltage amplitude U0 into three levels: U0 < U min is the normal state, U min ≤ U0 < U max is the warning state, and U0 ≥ U max is the fault state;

[0036] S52. Establish a three-dimensional judgment space, with dimensions including: voltage level V d , spectral feature flag bit F q , transient energy ratio R e = E d / E0;

[0037] S53. Train the weight coefficients α, β, and γ for each dimension based on historical fault data, satisfying α + β + γ = 1;

[0038] where, U min is the lowest voltage threshold, U max is the highest voltage threshold, V d is the voltage level quantization value, R e is the transient energy ratio, E0 is the steady-state energy, and α, β, and γ are the weight coefficients of voltage level, spectral feature, and transient energy ratio respectively.

[0039] Furthermore, the step S6 includes:

[0040] S61. Calculate the comprehensive score for each line, as shown in the following formula:

[0041] S = αV d + βF q + γR e

[0042] S62. Generate a set of candidate fault lines L = L k |≥ S avg + σ;

[0043] S63. If there are multiple lines in the set L, perform the threshold dynamic adjustment in step S7;

[0044] where, S is the grounding fault probability score, S avg is the average score of all lines in the current system, σ is the standard deviation of the scores, L is the set of candidate fault lines, and L k is the candidate line number.

[0045] Furthermore, the method of threshold dynamic adjustment in step S7 is:

[0046] S71. Update the threshold reference value in real time according to the system operation status, as shown in the following formula:

[0047] S base = μS hist +(1 - μ)S curr

[0048] S72. When the environmental humidity H > H th , adjust the judgment threshold to S th = S base×K h ;

[0049] Before outputting the final line selection result, verify the logical relationship between the scores of each line and the threshold to ensure that the strict condition of S k / h>K r is satisfied;

[0050] wherein, S base is the dynamic threshold reference value, μ is the forgetting factor, S hist is the historical threshold, S curr is the current threshold calculated value, H is the environmental humidity, H th is the humidity threshold, K h is the humidity correction coefficient, K r is the threshold redundancy coefficient.

[0051] Furthermore, in the step S23, the calculation formula of the energy density F d is as follows:

[0052]

[0053] wherein, t1 is the starting moment of the transient process, and t2 is the ending moment of the transient process.

[0054] Furthermore, the training method of the weight coefficient in the step S53 is as follows:

[0055] An improved simulated annealing algorithm is used to solve the optimal weight combination, and the objective function is shown in the following formula:

[0056] min∑|S predicted -S actual | 2 +λ(α 2 +β 2 +γ 2 )

[0057] wherein, S prdicted is the predicted score, S actual is the actual fault score, λ is the regularization coefficient, and the constraint condition is α≥0.4, γ≤0.3.

[0058] According to the above embodiments of the present invention, it has at least the following beneficial effects: The present invention extracts the transient current feature vector through multi-scale wavelet decomposition, and combines the zero-sequence voltage mutation moment for phase modulation and spectrum demodulation, which can effectively capture the transient characteristics when the grounding fault occurs, and improve the sensitivity and accuracy of fault recognition. At the same time, a multi-dimensional logical judgment matrix is constructed based on the zero-sequence voltage amplitude and spectrum eigenvalue, and the judgment threshold is dynamically adjusted, which can adapt to the changes in the system operation state and the influence of environmental factors, and enhance the robustness and adaptability of the line selection method.

[0059] In addition, the present invention uses an improved simulated annealing algorithm to optimize the weight coefficients and introduces multi-dimensional features such as transient energy ratio, which can further improve the recognition accuracy of faulty lines and reduce the possibility of misjudgment and missed judgment. By dynamically updating the threshold reference value and the humidity correction mechanism, the reliability of the line selection result can also be ensured, providing a strong guarantee for the safe operation of the medium-voltage power supply system in nuclear power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, wherein:

[0061] Figure 1 FIG. is a schematic flow chart of an intelligent line selection method for grounding of medium-voltage power supplies in nuclear power plants provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present invention, and not to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to fully convey the scope of the present invention to those skilled in the art.

[0063] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, apparatus, device, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0064] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0065] Reference is made below to Figure 1 , Figure 1 FIG. is a schematic flow chart of an intelligent line selection method for grounding of medium-voltage power supplies in nuclear power plants provided by an embodiment of the present invention. As Figure 1 shown, an intelligent line selection method 100 for grounding of medium-voltage power supplies in nuclear power plants includes:

[0066] S1. Real-time collect the three-phase current signals and zero-sequence voltage signals of each line in the medium-voltage power supply system of the nuclear power plant;

[0067] S2. Perform multi-scale wavelet decomposition on the collected three-phase current signals to extract the transient current feature vectors of each line;

[0068] S3. Based on the mutation moment of the zero-sequence voltage signal, perform phase modulation on the transient current eigenvector to generate a modulated signal;

[0069] S4. Perform spectral demodulation on the modulated signal and calculate the spectral eigenvalue of each line;

[0070] S5. Combine the zero-sequence voltage amplitude and the spectral eigenvalue to construct a multi-dimensional logical judgment matrix;

[0071] S6. According to the weight distribution of each dimension in the logical judgment matrix, calculate the grounding fault probability score of each line;

[0072] S7. Dynamically adjust the decision threshold of the grounding fault probability score and output the final line selection result.

[0073] It should be noted that an intelligent grounding line selection method for medium-voltage power supplies in nuclear power plants proposed by the present invention aims to accurately identify the faulty line through a series of signal processing and intelligent judgment steps. Among them, real-time acquisition of the three-phase current signals and zero-sequence voltage signals of each line in the medium-voltage power supply system of the nuclear power plant is the basis of the entire method. The three-phase current signal refers to the instantaneous values of the A, B, and C phase currents obtained through current transformers, which are used to reflect the current changes of the line under normal operation or fault conditions; the zero-sequence voltage signal is the instantaneous value of the bus zero-sequence voltage collected through a voltage transformer, which is used to detect the voltage change during a grounding fault. The acquisition of these signals is a prerequisite for subsequent feature extraction and fault judgment.

[0074] Specifically, when collecting signals, the sampling frequency needs to be determined according to the preset sampling theorem to ensure the integrity and accuracy of the signals. The sampling frequency should satisfy that the sampling frequency is greater than or equal to 10 times the highest frequency component of the signal. This setting is to avoid signal aliasing and ensure that the collected signals can truly reflect the operating state of the line. For example, if the highest frequency component in the signal is about 500 Hz, the sampling frequency should not be lower than 5000 Hz. In addition, multi-scale wavelet decomposition is a key step in extracting the transient current eigenvector. It performs multi-layer decomposition on the three-phase current signal by selecting the Daubechies wavelet basis function to obtain wavelet coefficients at different scales. These wavelet coefficients can reflect the characteristic changes of the signal at different time scales, thus providing rich information for subsequent feature extraction and fault judgment.

[0075] Preferably, when performing multi-scale wavelet decomposition, 4 or 5 layers of decomposition can be selected to balance the computational complexity and the fineness of feature extraction. For the processing of wavelet coefficients, the soft threshold denoising method can be adopted to remove noise interference by setting an appropriate threshold, thereby reconstructing a purer transient current signal. In addition, when calculating the transient current feature vector, parameters such as the peak value of the transient current, the rise time, and the energy density can be introduced. These parameters respectively reflect the amplitude change, time characteristics, and energy distribution of the current when a fault occurs, and can comprehensively characterize the fault features, improving the sensitivity and accuracy of the line selection method.

[0076] In some embodiments, the step S1 includes:

[0077] S11. Synchronously obtain the instantaneous values I a (t), I b (t), I c (t) of the three-phase currents of each line through a current transformer;

[0078] S12. Collect the instantaneous zero-sequence voltage U0(t) of the bus through a voltage transformer;

[0079] S13. Determine the sampling frequency f s according to the preset sampling theorem, satisfying f s ≥ 10f max ;

[0080] Wherein, I a (t), I b (t), I c (t) are respectively the instantaneous values of the three-phase currents, U0(t) is the instantaneous zero-sequence voltage, f s is the sampling frequency, and f max is the highest frequency component of the signal.

[0081] It should be noted that in the signal acquisition stage of the present invention, the instantaneous values of the three-phase currents of each line are synchronously obtained through a current transformer, the instantaneous zero-sequence voltage of the bus is collected through a voltage transformer, and the sampling frequency is determined according to the preset sampling theorem. This process is the basis for realizing ground fault detection, ensuring the integrity and accuracy of signal acquisition. Among them, the current transformer and the voltage transformer are common sensors in the power system, which are respectively used to measure current and voltage signals. The sampling theorem is an important theory in signal processing, which stipulates that the sampling frequency must meet certain conditions to avoid signal aliasing, thereby ensuring the integrity and usability of the signal.

[0082] Specifically, current transformers are used to measure three-phase current signals, namely the instantaneous values of the currents in phases A, B, and C. These current signals reflect the current variations of the line under normal operation or fault conditions. Voltage transformers are used to collect the instantaneous zero-sequence voltage of the busbar. The zero-sequence voltage is one of the important characteristics of grounding faults, and its variation can reflect the occurrence and severity of grounding faults. When determining the sampling frequency, it needs to be set according to the highest frequency component of the signal. The sampling frequency should meet the requirement of being greater than or equal to 10 times the highest frequency component of the signal. For example, if the highest frequency component in the signal is about 500 Hz, the sampling frequency should not be lower than 5000 Hz. This setting can effectively avoid signal aliasing and ensure that the collected signal can truly reflect the operating state of the line.

[0083] Preferably, in practical applications, the sampling frequency can be optimized according to specific scenarios. For example, for a medium-voltage power supply system with more high-frequency signal components, the sampling frequency can be appropriately increased to ensure the integrity of the signal. At the same time, in order to further improve the accuracy of signal acquisition, high-precision current transformers and voltage transformers can be used, and digital filtering technology can be combined to remove high-frequency noise in the signal. In addition, when collecting signals, a synchronous clock mechanism can be introduced to ensure the synchronous acquisition of three-phase current signals and zero-sequence voltage signals, thereby improving the accuracy of subsequent signal processing.

[0084] In some embodiments, the specific steps of the multi-scale wavelet decomposition in step S2 include:

[0085] S21. Select the Daubechies wavelet basis function to decompose the three-phase current signal into N layers to obtain the wavelet coefficients W j (k);

[0086] S22. Perform threshold denoising on the wavelet coefficients of the j-th layer and reconstruct the transient current signals I′ a (t), I′ b (t), I′ c (t);

[0087] S23. Calculate the transient current feature vector as shown in the following formula:

[0088] F i =[ΔI max , T r , E d

[0089] Where j is the decomposition layer number, k is the coefficient serial number, Wj(k) is the k-th wavelet coefficient of the j-th layer, I′ a (t), I′ b (t), I′ c (t) are the reconstructed transient current signals, ΔI max ​is the peak value of the transient current, T r is the rise time, E d is the energy density, F i is the transient current eigenvector.

[0090] It should be noted that the multi-scale wavelet decomposition mentioned in the present invention is the key step for extracting the transient current eigenvector. By selecting the Daubechies wavelet basis function to perform multi-layer decomposition on the three-phase current signal, the wavelet coefficients at each scale can be obtained. Wavelet decomposition is a method that can effectively process non-stationary signals and can decompose the signal into components of different frequencies and time scales. The Daubechies wavelet basis function is a commonly used orthogonal wavelet basis with good time-frequency localization characteristics and is suitable for the analysis of transient signals. The transient current eigenvector is calculated by reconstructing the denoised transient current signal, which contains characteristics such as the peak value, rise time, and energy density of the transient current, and these characteristics can reflect the characteristics of the grounding fault.

[0091] Specifically, the process of multi-scale wavelet decomposition includes the following steps: First, select the Daubechies wavelet basis function to perform N-layer decomposition on the three-phase current signal, and the value of N can be adjusted according to the complexity of the signal and computing resources. The wavelet coefficients obtained after decomposition reflect the characteristic changes of the signal at different scales. Then, perform threshold denoising processing on the wavelet coefficients at the j-th layer, and the setting of the threshold can be adjusted according to the noise level of the signal and the requirement for retaining features. The denoised transient current signal is obtained through reconstruction and is used to calculate the transient current eigenvector. The peak value in the transient current eigenvector refers to the maximum value of the transient current, which reflects the degree of current mutation when the fault occurs; the rise time refers to the time when the transient current rises from the starting value to the peak value, which reflects the rapidity of the fault occurrence; the energy density is the energy distribution of the transient current within a certain time, which reflects the intensity of the fault.

[0092] Preferably, in practical applications, the parameters of multi-scale wavelet decomposition can be optimized. For example, the decomposition layer number N can be selected as 3 to 5 layers to balance the accuracy of feature extraction and computational complexity. For the threshold denoising of wavelet coefficients, an adaptive threshold method can be adopted to dynamically adjust the threshold according to the statistical characteristics of the signal, so as to better remove noise and retain fault features. In addition, other characteristic parameters, such as the zero-crossing time or frequency distribution characteristics of the transient current, can be introduced into the calculation of the transient current eigenvector to further enrich the connotation of the eigenvector and improve the accuracy of fault identification.

[0093] In some embodiments, the phase modulation process of step S3 is as follows:

[0094] S31. According to the moment t0 of the zero-sequence voltage mutation, determine the modulation window time as [t0 - ΔT, t0 + ΔT];

[0095] S32. Phase-encode the transient current feature vector F i to generate a modulation signal as shown in the following formula:

[0096]

[0097] where t0 is the moment of zero-sequence voltage mutation, ΔT is the half-width of the modulation window time, A is the amplitude of the modulation signal, f c is the carrier frequency, is the phase modulation function based on the transient current feature vector F i of.

[0098] It should be noted that the phase modulation process mentioned in the present invention is a key step in processing the transient current feature vector based on the moment of zero-sequence voltage mutation. By determining the modulation window time and phase encoding, the information of the transient current feature vector can be embedded into the modulation signal. The modulation window time refers to a time interval around the moment of zero-sequence voltage mutation, which is used to limit the action range of phase modulation. The phase modulation function is designed according to the characteristics of the transient current feature vector and is used to convert the information of the feature vector into the phase change of the modulation signal, thereby providing a basis for subsequent spectral demodulation.

[0099] Specifically, the phase modulation process first needs to determine the moment of zero-sequence voltage mutation, which is the starting point of modulation. The half-width of the modulation window time is set according to the duration and noise level of the transient current signal, usually taking about half of the duration of the transient process. For example, if the duration of the transient process is 10 milliseconds, the half-width of the modulation window time can be set to 5 milliseconds. The amplitude and carrier frequency of the modulation signal are two important parameters of phase modulation. The amplitude can be adjusted according to the signal intensity and detection requirements, while the carrier frequency needs to select a high-frequency signal that does not conflict with the frequency range of the transient current signal to avoid signal aliasing. The phase modulation function is a function that maps the characteristics of the transient current feature vector to the phase change, and its design needs to consider each dimension of the feature vector, such as peak value, rise time, and energy density, to ensure that the modulation signal can effectively reflect the fault characteristics.

[0100] Preferably, in practical applications, the half-width of the modulation window time can be dynamically adjusted according to the characteristics of the transient current signal. For example, for high-frequency transient signals, the half-width of the modulation window time can be appropriately reduced to improve the modulation accuracy; while for low-frequency transient signals, it can be appropriately increased. The selection of the carrier frequency needs to consider the frequency response characteristics of the system, and a high-frequency signal between 10 kHz and 100 kHz can be selected to ensure the separation of the frequency range of the modulation signal from that of the original transient current signal. In addition, the phase modulation function can adopt a non-linear mapping method to enhance the anti-interference ability of the modulation signal. For example, a high-order polynomial or logarithmic function can be introduced to design the phase modulation function, so as to better reflect the complex characteristics of the transient current eigenvector.

[0101] In some embodiments, the spectrum demodulation of step S4 includes:

[0102] S41. Perform a fast Fourier transform on the modulated signal to obtain a spectrogram P(f);

[0103] S42. Calculate the spectral eigenvalue, as shown in the following formula:

[0104]

[0105] S43. Generate a binary feature flag F th according to the comparison result between the spectral eigenvalue Q and the preset spectral eigenvalue threshold Q q ;

[0106] where Q is the spectral eigenvalue, Δf is the characteristic frequency band width, f c is the carrier frequency, P(f) is the signal power spectral density, f s is the sampling frequency, Q th is the preset spectral eigenvalue threshold, and F q is the binary feature flag.

[0107] It should be noted that the spectrum demodulation process mentioned in the present invention is to perform a fast Fourier transform (FFT) on the modulated signal to obtain its spectrogram and calculate the spectral eigenvalue, so as to provide a quantitative basis for fault detection. The core of spectrum demodulation lies in identifying fault characteristics by analyzing the spectral characteristics of the signal. The fast Fourier transform is an efficient algorithm that can convert a time-domain signal into a frequency-domain signal and reveal the frequency components of the signal. The spectral eigenvalue is obtained by calculating the energy distribution in a specific frequency band of the spectrogram and is used to reflect the characteristics of the signal. The binary feature flag is generated according to the comparison result between the spectral eigenvalue and the preset spectral eigenvalue threshold and is used to convert the spectral eigenvalue into a logic signal that is easy to judge.

[0108] Specifically, the fast Fourier transform converts the modulated signal from the time domain to the frequency domain to obtain a spectrogram. The spectrogram reflects the energy distribution of the signal at different frequencies. The calculation of the spectral feature values usually involves integrating the power spectral density of the signal within a specific frequency band to quantify the energy distribution of the signal. For example, the characteristic frequency band width of the signal can be selected from 100 Hz to 1 kHz, and the carrier frequency is 5 kHz. The spectral feature value is obtained by calculating the integral of the power spectral density within this frequency band. The preset spectral feature threshold is set according to historical data and experience and is used to determine whether the signal contains fault features. If the spectral feature value exceeds the threshold, a binary feature flag bit of 1 is generated, indicating that a fault may exist; otherwise, it is 0, indicating normal.

[0109] Preferably, in practical applications, the characteristic frequency band width can be adjusted according to the characteristics of the signal. For example, for high-frequency transient signals, the characteristic frequency band width can be appropriately increased to cover more possible fault characteristic frequencies. The selection of the carrier frequency also needs to be optimized according to the spectral distribution of the signal to ensure its separation from the frequency range of the transient signal. In addition, the spectral feature threshold can be dynamically adjusted by an adaptive algorithm to adapt to different system operating states and noise levels. For example, a statistical-based adaptive threshold algorithm can be used to dynamically adjust the threshold according to the background noise level of the signal, thereby improving the accuracy and reliability of fault detection.

[0110] In some embodiments, the process of constructing the multi-dimensional logical judgment matrix in step S5 is as follows:

[0111] S51. Divide the zero-sequence voltage amplitude U0 into three levels: U0 < U min is the normal state, U min ≤ U0 < U max is the warning state, and U0 ≥ U max is the fault state;

[0112] S52. Establish a three-dimensional judgment space, and the dimensions include: voltage level V d 、spectral feature flag bit F q 、transient energy ratio R e = E d / E0;

[0113] S53. Train the weight coefficients α, β, γ of each dimension according to historical fault data, satisfying α + β + γ = 1;

[0114] where, U min is the lowest voltage threshold, R max is the highest voltage threshold, V d is the voltage level quantization value, R e is the transient energy ratio, E0 is the steady-state energy, and α, β, γ are the weight coefficients of the voltage level, spectral feature, and transient energy ratio respectively.

[0115] It should be noted that the construction process of the multi-dimensional logic judgment matrix mentioned in the present invention is a key step for comprehensive judgment based on multi-dimensional information such as zero-sequence voltage amplitude and spectral eigenvalue. By dividing the zero-sequence voltage amplitude into different states and combining the spectral feature flag bit and transient energy ratio, a three-dimensional judgment space can be constructed, thereby providing a comprehensive quantitative basis for the identification of faulty lines. The multi-dimensional logic judgment matrix is a decision-making tool that comprehensively considers multiple feature dimensions and can effectively improve the accuracy and reliability of fault judgment. Among them, the voltage level, spectral feature flag bit, and transient energy ratio are three main judgment dimensions, respectively reflecting the voltage characteristics, spectral characteristics, and energy characteristics of the fault.

[0116] Specifically, the zero-sequence voltage amplitude is divided into three levels: normal state, warning state, and fault state, corresponding to different voltage threshold ranges. For example, the lowest voltage threshold can be set to 10V, and the highest voltage threshold to 50V. When the zero-sequence voltage is less than 10V, it is in the normal state; when it is between 10V and 50V, it is in the warning state; when it is greater than 50V, it is in the fault state. The spectral feature flag bit is a binary signal generated based on the comparison result of the spectral eigenvalue and the preset threshold, used to judge whether the signal contains fault features. The transient energy ratio is the ratio of transient energy to steady-state energy, reflecting the degree of energy mutation during the occurrence of a fault. By training the weight coefficients of each dimension with historical fault data, a quantitative evaluation of different feature dimensions can be achieved. The sum of the weight coefficients is 1, ensuring the normalization of the judgment matrix.

[0117] Preferably, in practical applications, the division of voltage levels and the setting of weight coefficients can be optimized. For example, the voltage threshold range can be adjusted according to the specific operating characteristics of the system to make it more in line with the actual working conditions. For the training of weight coefficients, an improved simulated annealing algorithm or other optimization algorithms can be used to improve the accuracy of weight allocation. In addition, more feature dimensions can be introduced, such as fault duration or fault occurrence frequency, to further enrich the connotation of the judgment matrix. For example, the fault duration can be used as the fourth dimension and corresponding weights can be assigned according to its length, thereby improving the comprehensiveness and accuracy of fault identification.

[0118] In some embodiments, step S6 includes:

[0119] S61. Calculate the comprehensive score for each line as shown in the following formula:

[0120] S = αV d + βF q + γR e

[0121] S62. Generate a set of candidate faulty lines L = L k |≥Savg +σ;

[0122] S63. If there are multiple lines in the set L, perform threshold dynamic adjustment in step S7;

[0123] where S is the grounding fault probability score, S avg is the average value of the scores of all lines in the current system, σ is the standard deviation of the scores, L is the set of candidate fault lines, and L k is the candidate line number.

[0124] It should be noted that the calculation of the grounding fault probability score and its dynamic adjustment mentioned in the present invention are important links for accurately identifying fault lines. By comprehensively considering the information of each dimension in the multi-dimensional logic judgment matrix, the grounding fault probability score of each line is calculated, and a set of candidate fault lines is generated according to the score ranking. This process can not only quantify the fault possibility of each line, but also adapt to the changes in the system operating state by dynamically adjusting the judgment threshold, thereby improving the reliability and adaptability of the line selection method. The grounding fault probability score is the result of comprehensive evaluation based on multi-dimensional features, reflecting the likelihood of a line having a grounding fault; while the dynamic adjustment of the judgment threshold is to cope with changes in the system operating environment and noise interference to ensure the accuracy of the line selection result.

[0125] Specifically, the calculation of the grounding fault probability score is obtained by weighted summation of the information of each dimension in the multi-dimensional logic judgment matrix. For example, the voltage level, spectral feature flag bit, and transient energy ratio correspond to different weight coefficients, which are obtained by training with historical fault data to ensure the scientificity and accuracy of the score. The score result is used to generate a set of candidate fault lines, and the scores of the lines in the set need to be higher than a certain threshold, which is usually set as the average value of the scores of all lines in the current system plus several times the standard deviation. For example, the threshold can be set as the average value plus 1.5 times the standard deviation to screen out the lines with higher fault possibility. If there are multiple lines in the set, it is necessary to further narrow the candidate range by dynamically adjusting the judgment threshold.

[0126] Preferably, in practical applications, the details of score calculation and threshold adjustment can be optimized. For example, the weight coefficients can be dynamically adjusted according to the real-time changes in the system operating state to better reflect the importance of current fault characteristics. For dynamic threshold adjustment, environmental parameters (such as humidity, temperature) can be introduced as correction factors. For example, when the environmental humidity exceeds a certain threshold, the judgment threshold is appropriately increased to reduce misjudgment. In addition, machine learning algorithms can be used to mine historical data to automatically optimize the score model and threshold adjustment strategy, thereby further improving the intelligence level and adaptability of the line selection method.

[0127] In some embodiments, the method for dynamically adjusting the threshold in step S7 is as follows:

[0128] S71. Update the threshold reference value in real time according to the system operating state, as shown in the following formula:

[0129] S base = μS hist +(1 - μ)S curr

[0130] S72. When the environmental humidity H > H th , adjust the decision threshold to S th = S base ×K h ;

[0131] S73. Before outputting the final line selection result, verify the logical relationship between the scores of each line and the threshold to ensure that the strict condition of S k / h > K r is satisfied;

[0132] Among them, S base is the dynamic threshold reference value, μ is the forgetting factor, S hist is the historical threshold, S curr is the current threshold calculation value, H is the environmental humidity, H th is the humidity threshold, K h is the humidity correction coefficient, and K r is the threshold redundancy coefficient.

[0133] It should be noted that the method for dynamically adjusting the threshold mentioned in the present invention is to solve the influence of changes in the system operating environment on the accuracy of fault judgment. By updating the threshold reference value in real time and adjusting the decision threshold according to factors such as environmental humidity, the adaptability and reliability of the line selection method can be effectively improved. The core of the dynamic threshold adjustment lies in dynamically correcting the decision threshold according to the system operating state and environmental conditions to ensure accurate judgment of the faulty line under different working conditions. Among them, the dynamic threshold reference value is calculated based on the historical threshold and the current system state and is used to reflect the normal level of the system under the current operating conditions; while the humidity correction coefficient is a factor for adjusting the threshold according to the environmental humidity and is used to compensate for the influence of humidity changes on the signal characteristics.

[0134] Specifically, the update of the dynamic threshold reference value is achieved by combining the historical threshold and the current system state. For example, a forgetting factor (such as 0.8 or 0.9) can be set to balance the weights of historical data and current data. The closer the forgetting factor is to 1, the higher the dependence on historical data; the closer it is to 0, the higher the dependence on current data. The threshold for environmental humidity can be set according to the actual operating environment. For example, when the humidity exceeds 80%, it is considered that the environmental humidity has a significant impact on the signal characteristics, and at this time, the judgment threshold needs to be adjusted. The humidity correction coefficient can be set according to experimental data or experience. For example, in a high-humidity environment, the threshold is increased by 10% - 20% to reduce the possibility of misjudgment.

[0135] Preferably, in practical applications, the strategy of dynamic threshold adjustment can be further optimized. For example, in addition to humidity correction, a temperature correction factor can be introduced because temperature changes also affect signal characteristics. The temperature correction coefficient can be set according to the relationship between temperature and signal characteristics. For example, the threshold is appropriately reduced in a high-temperature environment. In addition, the update of the dynamic threshold reference value can be combined with more system operating parameters, such as the load change rate or the grid frequency deviation, to more comprehensively reflect the system state. An adaptive algorithm can also be used to adjust the threshold in real time. For example, an online learning algorithm based on machine learning dynamically optimizes the threshold according to real-time monitoring data, thereby further improving the intelligence level and adaptability of the line selection method.

[0136] In some embodiments, the energy density E in step S23 d is calculated by the formula:

[0137]

[0138] where t1 is the starting moment of the transient process, and t2 is the ending moment of the transient process.

[0139] It should be noted that the energy density calculation mentioned in the present invention is an important parameter in the transient current feature vector, which is used to characterize the energy distribution of the transient current signal within a specific time interval. The calculation of energy density is based on the square integral of the transient current signal, reflecting the degree of energy accumulation during the transient process. By calculating the energy density, the current characteristics during the occurrence of a fault can be more comprehensively described, thereby providing richer information for the identification of the faulty line. The starting moment and ending moment of the transient process are determined according to the characteristics of the transient current signal, usually corresponding to the time points when the fault occurs and the fault characteristics disappear.

[0140] Specifically, the calculation of energy density involves the square integral of the transient current signal within a specific time interval. The start and end times of the transient process can be determined by analyzing the mutation points and recovery points of the current signal. For example, the start time of the transient process can be the time point when the current signal first exceeds a certain threshold, and the end time can be the time point when the current signal returns to the normal level. When calculating the energy density, it is necessary to integrate the sum of the squares of the three-phase transient current signals to obtain the energy distribution of the entire transient process. For example, if the duration of the transient process is 10 milliseconds, the integration interval is this 10-millisecond time period. The calculation result of the energy density can be an important part of the transient current feature vector and can be used for fault identification together with other characteristic parameters (such as peak value, rise time).

[0141] Preferably, in practical applications, the calculation process of energy density can be optimized. For example, the edge effect of the transient current signal can be smoothed by introducing a window function, thereby improving the accuracy of energy density calculation. In addition, for complex transient current signals, the method of piecewise integration can be adopted, dividing the transient process into multiple sub-intervals, calculating the energy density of each sub-interval separately, and then performing weighted summation. This can more precisely reflect the energy change characteristics in the transient process. In addition, the calculation of energy density can also be combined with other characteristic parameters (such as frequency distribution) to form a more comprehensive feature vector to improve the sensitivity and accuracy of fault identification.

[0142] In some embodiments, the training method of the weight coefficient in step S53 is as follows:

[0143] An improved simulated annealing algorithm is used to solve the optimal weight combination, and the objective function is shown in the following formula:

[0144] min∑|S predicted -S actual | 2 +λ(α 2 +β 2 +γ 2 )

[0145] Where S prdicted is the predicted score, S actual is the actual fault score, λ is the regularization coefficient, and the constraint conditions are α≥0.4 and γ≤0.3.

[0146] It should be noted that the weight coefficient training method mentioned in the present invention solves for the optimal weight combination through an improved simulated annealing algorithm to optimize the weight distribution of each dimension in the multi-dimensional logical judgment matrix. The goal of this method is to ensure the rationality and stability of the weight distribution by minimizing the difference between the predicted score and the actual fault score and combining regularization constraints. The training of the weight coefficient is a key link in realizing the accurate identification of the fault line. By optimizing the weight combination, the accuracy and reliability of fault identification can be improved. Among them, the predicted score is the score calculated based on the current weight distribution, while the actual fault score is the true score determined according to historical fault data.

[0147] Specifically, the improved simulated annealing algorithm is a probability-based optimization algorithm that gradually approaches the optimal solution by simulating the temperature change in the physical annealing process. In the training of the weight coefficient, the objective function consists of two parts: one part is the mean square error between the predicted score and the actual fault score, which is used to measure the fitting degree of the model; the other part is the regularization term, which prevents overfitting by restricting the sum of the squares of the weight coefficients. The regularization coefficient is a hyperparameter used to balance the fitting error and the complexity of the weight coefficients. For example, the regularization coefficient can be set between 0.01 and 0.1 to ensure that the model has good generalization ability while fitting the data. In addition, the constraint conditions of the weight coefficients (such as the value range of the weight coefficients) can be set according to actual needs. For example, it is stipulated that the weight coefficient of the voltage level is not less than 0.4, while the weight coefficient of the spectral feature is not higher than 0.3, to ensure that the importance of each dimension in the decision-making process conforms to the actual working conditions.

[0148] Preferably, in practical applications, the weight coefficient training process can be further optimized. For example, a genetic algorithm or other heuristic algorithm can be combined with the simulated annealing algorithm to improve the optimization efficiency and global search ability. In addition, the regularization coefficient can be dynamically adjusted according to the results of cross-validation to better balance the fitting accuracy and complexity of the model. More historical fault data can also be introduced to expand the training samples, thereby improving the accuracy and robustness of the weight coefficient training. For example, by collecting fault data under different operating states, including fault cases under different loads and different environmental conditions, a more adaptable weight coefficient combination can be trained to further improve the performance of fault identification.

[0149] The above-mentioned various embodiments of the present invention have the following beneficial effects: By collecting three-phase current signals and zero-sequence voltage signals in real time and performing multi-scale wavelet decomposition on the three-phase current signals, the present invention can extract transient current feature vectors, thereby capturing the subtle features of grounding faults. Based on the zero-sequence voltage mutation moment, phase modulation and spectrum demodulation are performed, which can further enhance the significance of fault features and provide a basis for the accurate identification of fault lines. By constructing a multi-dimensional logical judgment matrix by combining the zero-sequence voltage amplitude and spectrum eigenvalue and calculating the grounding fault probability score according to weight distribution, the rapid screening and accurate judgment of fault lines can be realized. Dynamically adjusting the judgment threshold can adapt to the changes in the system operation state, ensure the reliability of the line selection result, and improve the anti-interference ability of the system. In addition, by using an improved simulated annealing algorithm to optimize the weight coefficient, the accuracy of fault line identification can be further improved, and the possibility of misjudgment and missed judgment can be reduced.

[0150] By setting the sampling frequency to meet the requirements of the highest frequency component of the signal, the present invention can ensure the integrity and accuracy of the collected signal and provide a reliable data basis for subsequent analysis. Threshold denoising of wavelet coefficients and reconstructing the transient current signal can effectively remove noise interference and improve the purity of feature extraction. When calculating the transient current feature vector, parameters such as peak value, rise time, and energy density are introduced, which can comprehensively reflect the fault features and improve the sensitivity of the line selection method. During the spectrum demodulation process, by performing fast Fourier transform and generating binary feature flags, the fault line and non-fault line can be quickly distinguished. Dividing the zero-sequence voltage amplitude into normal state, warning state, and fault state and establishing a three-dimensional judgment space can more intuitively reflect the system state and provide multi-dimensional basis for fault diagnosis.

[0151] Furthermore, the storage medium of the implementation manner of the present application stores program instructions capable of implementing all the above methods. Among them, the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various implementation manners of the present application. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.

[0152] The above description is only some preferred embodiments of the present invention and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for intelligent line selection of medium-voltage power grounding in a nuclear power plant, characterized in that: The following steps are involved: S1. Real-time acquisition of three-phase current signals and zero-sequence voltage signals of each line in the medium-voltage power supply system of the nuclear power plant; S2. Perform multi-scale wavelet decomposition on the collected three-phase current signal to extract the transient current feature vector of each line; S3. Based on the mutation moment of the zero-sequence voltage signal, phase modulate the transient current characteristic vector to generate a modulated signal; S4. Perform spectrum demodulation on the modulated signal and calculate the spectrum characteristic value of each line; S5. Combine the zero-sequence voltage amplitude and the spectrum characteristic value to construct a multi-dimensional logic judgment matrix; S6. Calculate the ground fault probability score of each line according to the weight distribution of each dimension in the logic judgment matrix; S7. Dynamically adjust the judgment threshold of the ground fault probability score and output the final line selection result.

2. The method for intelligent line selection of medium voltage power supply grounding in a nuclear power plant according to claim 1, characterized in that: The step S1 comprises: S11. Synchronously obtain the instantaneous value I of the three-phase current of each line through the current transformer a (t), I b (t), I c (t); S12. Collect the instantaneous value of bus zero-sequence voltage U0(t) through the voltage transformer; S13. Determine the sampling frequency f according to the preset sampling theorem s , satisfying f s ≥10f max ; Among them, f s is the sampling frequency, f max is the highest frequency component of the signal.

3. The method for intelligent line selection of medium voltage power supply grounding in a nuclear power plant according to claim 2, characterized in that: The multi-scale wavelet decomposition in step S2 includes: S21. Select Daubechies wavelet basis function to decompose the three-phase current signal into N layers, and obtain the wavelet coefficients W of each scale. j (k); S22. Perform threshold denoising on the j-th layer wavelet coefficients to reconstruct the transient current signal I′ a (t), I′ b (t), I′ c (t); S23. Calculate the transient current characteristic vector, as shown in the following formula: F i =[ΔI max ,T r ,E d ] Among them, ΔI max is the transient current peak value, T r is the rise time, E d is the energy density, F i is the transient current eigenvector.

4. The method for intelligent line selection for medium voltage power grounding in a nuclear power plant according to claim 3, characterized in that: The phase modulation of step S3 includes: S31. According to the zero-sequence voltage mutation time t0, the modulation window time is determined as: [t0-ΔT, t0+ΔT]; where t0 is the moment of zero-sequence voltage mutation, and ΔT is the half-width of the modulation window; S32. For transient current characteristic vector F i Phase encoding is performed to generate a modulated signal, as shown in the following formula: Where A is the amplitude of the modulation signal, f c is the carrier frequency, Based on the transient current characteristic vector F i The phase modulation function.

5. The method for intelligent line selection of medium voltage power supply grounding in a nuclear power plant according to claim 4, characterized in that: The spectrum demodulation of step S4 includes: S41. Perform fast Fourier transform on the modulated signal to obtain a spectrum graph P(f); S42. Calculate the spectrum characteristic value, as shown in the following formula: Among them, Q is the spectral characteristic value, Δf is the characteristic frequency bandwidth, and f c is the carrier frequency, P(f) is the signal power spectrum density, f s is the sampling frequency; S43. Based on the spectrum characteristic value Q and the preset spectrum characteristic threshold Q th The comparison result generates the binary feature flag F q .

6. The method for intelligent line selection for medium voltage power grounding in a nuclear power plant according to claim 5, characterized in that: The construction process of the multidimensional logic judgment matrix in step S5 is: S51. Divide the zero-sequence voltage amplitude U0 into three levels: When U0 min is normal,​ When U min ≤U0 max For early warning,​ When U0≥U max It is a fault state; Among them, U min is the minimum voltage threshold, U max is the highest voltage threshold; S52. Establish a three-dimensional judgment space, the dimensions include: voltage level V d , spectrum feature flag F q , transient energy ratio R e ; S53. Train the weight coefficients α, β, and γ of each dimension according to the historical fault data to satisfy α+β+γ=1; Among them, α, β, and γ are the weight coefficients of voltage level, spectrum characteristics, and transient energy ratio, respectively.

7. The method for intelligent line selection for medium voltage power grounding in a nuclear power plant according to claim 6, characterized in that: The step S6 comprises: S61. Calculate a comprehensive score for each route, as shown in the following formula: S=αV d +βF q +γR e S62. Generate a candidate fault line set L=L according to the score sorting k |≥S avg +σ; Among them, S avg is the average score of all lines in the current system, σ is the standard deviation of the score, L is the set of candidate fault lines, L k Number the candidate routes; S63. If there are multiple lines in the set L, execute step S7 to dynamically adjust the determination threshold of the ground fault probability score.

8. The method for intelligent line selection for medium voltage power grounding in a nuclear power plant according to claim 1, characterized in that: The dynamically adjusting the determination threshold of the ground fault probability score in step S7 includes: S71. Update the threshold reference value in real time according to the system operation status, as shown in the following formula: S base =μS hist +(1-μ)S curr Among them, S base is the dynamic threshold reference value, μ is the forgetting factor, S hist is the historical threshold, S curr Calculate the value for the current threshold; S72. When the ambient humidity H> humidity threshold H th When the judgment threshold is adjusted to S th =S base ×K h ; Among them, K h is the humidity correction factor, K r is the threshold redundancy coefficient.

9. The method for intelligent line selection for medium voltage power grounding in a nuclear power plant according to claim 8, characterized in that: The energy density E in step S23 d The calculation formula is: Among them, t1 is the starting time of the transient process, and t2 is the ending time of the transient process.

10. The method for intelligent line selection for medium voltage power grounding in a nuclear power plant according to claim 9, characterized in that: The training method of the weight coefficient in step S53 is: The improved simulated annealing algorithm is used to solve the optimal weight combination. The objective function is shown in the following formula: min∑|S predicted -S actual | 2 +λ(a 2 +b 2 +g 2 ) Among them, S predicted S is the prediction score, actual is the actual fault score, and λ is the regularization coefficient.

Images