A fault automatic diagnosis method for full-wave pulsating direct current power distribution system

Abstract

This application relates to the field of power distribution system fault diagnosis technology, specifically to an automatic fault diagnosis method for full-wave pulsating DC power distribution systems. The method includes: real-time acquisition of current signals on the pulsating DC bus in the full-wave pulsating DC power distribution system; defining the time interval between two adjacent troughs in the current signal as a period, and combining multiple consecutive periods into a detection window; acquiring the current waveform anomaly coefficient, current spectrum anomaly coefficient, and current comprehensive anomaly coefficient for a single detection window; predicting the current comprehensive anomaly coefficient for the next detection window and comparing it with a preset fault monitoring threshold to assess the fault status of the full-wave pulsating DC power distribution system. This application aims to improve the accuracy of fault diagnosis.

Description

Technical Field

[0001] This application relates to the field of power distribution system fault diagnosis technology, specifically to an automatic fault diagnosis method for full-wave pulsating DC power distribution systems. Background Technology

[0002] Full-wave pulsed DC power distribution systems, as an emerging power distribution solution, are attracting increasing attention in ship power systems, data centers, photovoltaic systems, and specific industrial scenarios. Under long-term electrical stress, thermal stress, and humid environments, the cables and insulation components in full-wave pulsed DC power distribution systems gradually age and deteriorate, potentially leading to short circuits or open circuits, affecting the normal operation of the system. Therefore, achieving rapid and accurate fault diagnosis is crucial for ensuring the safe and stable operation of full-wave pulsed DC power distribution systems.

[0003] However, the inherent AC ripple, switching frequency harmonics, and characteristic harmonics generated by nonlinear loads in full-wave pulsating DC distribution systems can mask or confuse transient characteristic signals generated by faults, causing conventional detection methods based on specific frequency bands or amplitude thresholds to fail. Furthermore, non-fault disturbances, such as dynamic load changes and electromagnetic interference, may occur during actual detection, further affecting the accurate extraction of fault characteristics and reducing the accuracy of fault diagnosis. Summary of the Invention

[0004] In view of the above, it is necessary to provide an automatic fault diagnosis method for full-wave pulsating DC power distribution systems, which improves the accuracy of fault diagnosis compared with traditional fault diagnosis methods for full-wave pulsating DC power distribution systems.

[0005] The automatic fault diagnosis method for full-wave pulsating DC power distribution systems proposed in this application adopts the following technical solution: One embodiment of this application provides an automatic fault diagnosis method for a full-wave pulsating DC power distribution system, the method comprising the following steps: Real-time acquisition of current signals on the pulsating DC bus in a full-wave pulsating DC power distribution system; The time interval between two adjacent troughs in the current signal is taken as a period, and multiple consecutive periods are combined into a detection window. For a single detection window, the current waveform anomaly coefficient of a single detection window is obtained by considering the significance of the current exceeding the effective value, the randomness of the peak occurrence in the current signal, and the similarity of the current signal waveform between any two adjacent periods. The current spectrum anomaly coefficient of a single detection window is obtained by considering the average level of the high-frequency band energy ratio of all periods and its increase over time, as well as the significance of the peak characteristics in a single detection window other than the basic power frequency and its preset integer multiples. This coefficient is then fused with the current waveform anomaly coefficient to obtain the comprehensive current anomaly coefficient of a single detection window. Based on the combined current anomaly coefficient of a single detection window and its neighboring detection windows, the combined current anomaly coefficient of the next detection window of the single detection window is predicted and compared with the preset fault monitoring threshold to assess the fault status of the full-wave pulsating DC power distribution system.

[0006] In one embodiment, the process of obtaining the current waveform anomaly coefficient is as follows: The mean of the ratios of all peak values ​​to effective values ​​of the current signal in a single detection window is denoted as the significant mean. Calculate the dispersion of the time interval between any two adjacent peaks of the current signal in a single detection window; Calculate the average of the maximum mutual information coefficients of the current signals between any two adjacent cycles in a single detection window; The current waveform anomaly coefficient is positively correlated with the significant mean and the dispersion, and negatively correlated with the average value.

[0007] In one embodiment, the calculation process for the current waveform anomaly coefficient is as follows: The dispersion and the significant mean are respectively subjected to positive normalization; the average value is subjected to inverse normalization. The current waveform anomaly coefficient is a weighted sum of the normalized value of the dispersion, the normalized value of the significant mean, and the normalized value of the average.

[0008] In one embodiment, the fundamental power frequency of the single detection window is the frequency with the largest amplitude in the spectrum of the current signal in the single detection window.

[0009] In one embodiment, the process for determining the current spectrum anomaly coefficient is as follows: The background noise increase of a single detection window is obtained by measuring the average level of the high-frequency band energy percentage across all cycles and the degree of increase over time. The significance of abnormal spectral peaks in a single detection window is obtained by considering the significance of peak features other than the fundamental power frequency and its preset integer multiples in a single detection window. The current spectrum anomaly coefficient is a weighted sum of the rise in background noise and the significance of the abnormal spectral peak.

[0010] In one embodiment, the process of obtaining the background noise rise rate is as follows: Calculate the sum of the amplitudes of all frequencies greater than the base power frequency in the spectrum of the current signal within each cycle; The ratio of the sum of amplitudes to the sum of amplitudes at all frequencies in the spectrum of the current signal in each period is denoted as the high-frequency band total energy ratio. The background noise rise of a single detection window is obtained by comparing the total energy ratio of the high-frequency band in the last cycle to the remaining cycles in a single detection window, and combining this with the average level.

[0011] In one embodiment, the calculation process for the background noise rise rate is as follows: Calculate the difference between the total high-frequency band energy ratio of the last cycle in a single detection window and the total high-frequency band energy ratio of all other cycles; map the difference to a positive number; calculate the arithmetic mean of the mapping results of all the differences corresponding to the last cycle in a single detection window; The average of the total energy ratio of the high-frequency band in all cycles within a single detection window is denoted as the average high-frequency energy ratio. The rise rate of the background noise is positively correlated with the arithmetic mean and the average proportion of high-frequency energy, respectively.

[0012] In one embodiment, the process of obtaining the significance of the abnormal spectral peaks is as follows: After removing the DC bias from the current signal in a single detection window, a Fourier transform is performed to obtain the spectrum. Peak values ​​located in the single detection window other than the fundamental power frequency and its preset integer multiples are extracted. The abnormal peak value threshold of the single detection window is obtained by analyzing the distribution of the peak values ​​extracted from the single detection window, so as to filter out abnormal peak values ​​from the peak values ​​extracted from the single detection window. The significance of abnormal spectral peaks in a single detection window is obtained by analyzing the distribution of all abnormal peaks within that window. In one embodiment, the significance of the abnormal spectral peaks is the ratio of the mean of all abnormal peaks in a single detection window to the amplitude of the fundamental power frequency of the single detection window.

[0013] In one embodiment, the current composite anomaly coefficient is a weighted sum of the current waveform anomaly coefficient and the current spectrum anomaly coefficient.

[0014] This application has at least the following beneficial effects: This application achieves adaptive period division through trough point detection and uses the period as the basic analysis unit. It can make full use of the inherent periodicity of the full-wave pulsating DC, so that the current waveform under normal conditions has a predictable repeating pattern, thereby providing a stable reference benchmark for the detection of deviations in abnormal features. The detection window containing multiple periods can ensure sufficient samples to suppress random noise. Furthermore, by integrating the three dimensions of peak significance, peak time interval randomness, and periodic waveform similarity, the current waveform anomaly coefficient is obtained. The three work together to give the current waveform anomaly coefficient a comprehensive ability to characterize time-domain fault features, significantly improving the fault detection's ability to distinguish non-fault disturbances and reducing the false alarm rate and missed alarm rate. Furthermore, by analyzing the average level of the high-frequency band energy proportion, the continuous high-frequency harmonic energy injected by the fault can be quantified, and fault types such as high-resistance grounding that do not change significantly in the time domain but have significant characteristics in the frequency domain can be identified; by the degree of increase of high-frequency band energy over time, the gradual trend of fault development can be captured; by extracting the abnormal peak value at the non-power frequency harmonic, the fault characteristic frequency can be accurately located, and the new harmonic components generated by the fault can be distinguished from the inherent switching frequency harmonics of the full-wave pulsating DC power distribution system. Furthermore, by integrating time-domain and frequency-domain anomaly characteristics for fault diagnosis, it can cover different types and stages of fault development, significantly enhancing the comprehensiveness of fault diagnosis and thus improving its accuracy. By predicting the trend of the comprehensive anomaly coefficients of multiple consecutive detection windows, it avoids misjudgments caused by single measurement anomalies, improves the accuracy of fault diagnosis, and enables early warning of faults. Alarms are issued before the fault reaches a destructive level or triggers protection tripping, thereby effectively ensuring the safe and stable operation of the full-wave pulsating DC power distribution system. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the steps of an automatic fault diagnosis method for a full-wave pulsating DC power distribution system provided in this application; Figure 2 This is a schematic diagram of a full-wave pulsating DC power distribution system; Figure 3 This is a schematic diagram illustrating the process of obtaining the anomaly coefficient of the current waveform. Figure 4 This is a schematic diagram illustrating the process of obtaining the comprehensive anomaly coefficient of current. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, details a specific scheme for an automatic fault diagnosis method for a full-wave pulsating DC power distribution system provided in this application.

[0017] This application provides an embodiment of an automatic fault diagnosis method for full-wave pulsating DC distribution systems. Specifically, it provides the following automatic fault diagnosis method for full-wave pulsating DC distribution systems. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps: Step 1: Real-time acquisition of current signals on the pulsating DC bus in the full-wave pulsating DC power distribution system.

[0018] Traditional DC power distribution technology cannot achieve compatibility with both AC and DC appliances, which leads to a significant increase in construction costs and a more complex understanding of fault characteristics. To reduce the construction cost and fault detection complexity of a full-wave pulsating DC power distribution system, this application adopts the following schematic diagram: Figure 2 As shown.

[0019] Figure 2 In this system, the full-wave pulsating DC power distribution system includes a single-phase rectifier bridge [1], a pulsating DC bus [2], multiple AC load output branches [4], multiple stable DC load output branches [5], and multiple pulsating DC load output branches [6]. Among them, the AC load output branch [4] consists of an AC circuit breaker, a DC blocking capacitor [7], an even harmonic filter circuit [8], and an autotransformer [9]. The DC voltage of the pulsating DC bus [2] is connected in series with the AC circuit breaker, then in series with the DC blocking capacitor [7], and then in parallel with the even harmonic filter circuit [8] before being connected to the primary side of the autotransformer [9]. The secondary side of the autotransformer [9] is connected to the load output terminal of the AC load output branch [4]. The stable DC load output branch [5] Composed of an AC circuit breaker, an anti-reverse diode

[10] , and a filter capacitor

[11] , the DC voltage of the pulsating DC bus [2] is connected in series with the AC circuit breaker, then in series with the anti-reverse diode

[10] , and then in parallel with the filter capacitor

[11] , and then connected to the load output terminal of the stable DC load output branch [5]. The pulsating DC load output branch [6] is composed of an AC circuit breaker, and the DC voltage of the pulsating DC bus [2] is connected in series with the AC circuit breaker, and then connected to the load output terminal of the pulsating DC load output branch [6]. Multiple AC load output branches [4] are connected to the power input terminal of conventional AC electrical equipment of the same voltage level, multiple stable DC load output branches [5] are connected to the power input terminal of conventional DC electrical equipment of the corresponding voltage, and multiple pulsating DC load output branches [6] are connected to the power input terminal of DC electrical equipment with wide voltage input. The DC blocking capacitor [7] is an electrolytic capacitor. The even harmonic filter circuit [8] consists of two sets of parallel LC filters with tuning times of the 2nd and 4th harmonics. The turns ratio of the autotransformer [9] is 1:1.2.

[0020] The single-phase rectifier bridge [1] converts the input single-phase AC power into pulsating DC power, while the pulsating DC bus [2] serves as the main energy distribution channel for the output of the single-phase rectifier bridge [1], carrying pulsating DC power with a stable power frequency. The fluctuation characteristics of its current directly reflect the operating status of the entire full-wave pulsating DC power distribution system. Therefore, this application installs a Hall current sensor on the pulsating DC bus [2] to collect the current signal on the pulsating DC bus [2] in real time.

[0021] In this embodiment, the current signal is sampled at a frequency of 10kHz. The value of the sampling frequency is preset by the user and can be set by the implementer according to the actual situation. This application does not impose any special restrictions.

[0022] Step 2: Take the time interval between two adjacent troughs in the current signal as a period, and combine multiple consecutive periods into a detection window; obtain the current waveform anomaly coefficient, current spectrum anomaly coefficient and current comprehensive anomaly coefficient of a single detection window.

[0023] Step 2.1: For a single detection window, the current waveform anomaly coefficient of the single detection window is obtained by considering the degree to which the current exceeds the effective value, the randomness of the peak occurrence in the current signal, and the similarity of the current signal waveform between any two adjacent cycles.

[0024] When the full-wave pulsating DC power distribution system is running without faults, the single-phase rectifier bridge [1] will flip the negative half-cycle of the input AC power to positive and then output it to the pulsating DC bus [2], so that the current on the pulsating DC bus [2] exhibits a periodic pulsating ripple characteristic with a fixed direction and relatively stable amplitude. Taking a 50Hz input AC power as an example, the frequency of the pulsating DC power is twice that of the input AC power, i.e., 100Hz. The current signal waveform in each cycle is close to a sine wave. When the load and grid voltage are stable, the current signal amplitude between different cycles is relatively consistent. When the full-wave pulsating DC power distribution system experiences a fault, such as an intermittent arc fault, the current of the pulsating DC bus [2] rises sharply in a short time, and the peak value can reach several times the rated current, far exceeding any normal load switching change. Irregular, randomly spaced current pulses or spikes appear on the current waveform, and the original stable periodic pulsating ripple characteristic is also masked by high-frequency transient components.

[0025] Based on the above analysis, the time interval between two adjacent troughs in the current signal is considered as a period. N consecutive periods are then combined into a detection window.

[0026] In this embodiment, the value of N is 20. The value of N is preset by the user and can be set by the implementer according to the actual situation. This application does not impose any special restrictions.

[0027] When the normal load is switched on, the waveform of the current signal on the pulsating DC bus [2] of the full-wave pulsating DC distribution system may be disturbed, such as a very short spike or vibration. However, the disturbance generated by the normal load switching is transient, has a small degree of fluctuation and is not repeatable.

[0028] Based on the above analysis, for a single detection window, the current waveform anomaly coefficient is obtained by considering the degree to which the current exceeds the effective value, the randomness of the peak occurrence in the current signal, and the similarity of the current signal waveform between any two adjacent cycles. The specific process is as follows: The average of the ratios of all peak values ​​to effective values ​​of the current signal in a single detection window is recorded as the significant mean. The larger the significant mean is, the greater the degree to which the current of the pulsating DC bus [2] exceeds the effective value of the current during normal operation. Calculate the dispersion of the time interval between any two adjacent peaks of the current signal in a single detection window; the larger the calculated dispersion, the more random the time interval between the peaks on the current signal waveform. During normal operation, the amplitude of the current signal between different cycles is relatively consistent. However, when a fault occurs in the full-wave pulsating DC power distribution system, the resulting irregular changes will disrupt this consistency. Therefore, the average value of the maximum mutual information coefficient of the current signal between any two adjacent cycles in a single detection window is calculated. The smaller the calculated average value, the more unstable the waveform change of the current signal in a single detection window is. The dispersion and the significant mean are respectively subjected to positive normalization; the average value is subjected to inverse normalization. The current waveform anomaly coefficient is a weighted sum of the normalized value of the dispersion, the normalized value of the significant mean, and the normalized value of the average. The calculation of the effective value of the current signal and the maximum mutual information coefficient are well-known techniques and will not be elaborated upon in this application.

[0029] In this embodiment, an adaptive multi-scale peak detection algorithm is used to detect the peak value of the current signal in a single detection window to obtain each peak value. The adaptive multi-scale peak detection algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, based on the ability to obtain the peak value of the current signal in a single detection window, the implementer may use other existing feasible technologies, such as extreme point detection algorithms, etc. This application does not impose any special restrictions.

[0030] In this embodiment, the dispersion of the time interval is the coefficient of variation. The calculation of the coefficient of variation is a well-known technique and will not be described in detail here. As other implementation methods, implementers may adopt other existing feasible techniques based on the ability to measure the unevenness of the distribution of time intervals. This application does not impose any special restrictions.

[0031] In this embodiment, during the normal operation of the full-wave pulsating DC power distribution system, historical data of no less than M detection windows are collected, and the distribution of the significant mean, the dispersion, and the average value under normal operating conditions are statistically analyzed. The mean of all significant means, the mean of all dispersion, and the mean of all average values ​​in the M detection windows are used as the reference values ​​for normalizing the significant mean, the dispersion, and the average value in a single detection window. Through formula Forward normalization is performed on the significant mean and dispersion of each individual detection window. Taking the significant mean of a single detection window as an example, X is assigned the significant mean of the single detection window. If the value is assigned as the baseline value when normalizing the significant mean in a single detection window, then Y is the normalized value of the significant mean in a single detection window; Through formula The average value in a single detection window is inversely normalized, and X is assigned the average value of the single detection window. If the value is assigned as the reference value when normalizing the average value in a single detection window, then Y is the normalized value of the average value in a single detection window.

[0032] The value of M is 100. Implementers can appropriately increase the value of M according to the complexity of the operating conditions of the full-wave pulsating DC power distribution system. For example, for industrial scenarios with frequent load changes, the value of M can be 200.

[0033] In this embodiment, the weight of the normalized value of the dispersion is 0.35, the weight of the normalized value of the significant mean is 0.35, and the weight of the normalized value of the average is 0.3. The weights of the normalized values ​​of the dispersion, the significant mean, and the average are all calculated from experimental data.

[0034] It should be noted that: the significant mean and the discreteness quantify the random characteristics of the pulse intensity and pulse time interval of the current signal on the pulsating DC bus [2] caused by the fault, and the average value quantifies the unstable characteristics of the current signal waveform change; by combining the significant mean, the discreteness and the average value, the time-domain abnormal characteristics of the current signal on the pulsating DC bus [2] under the influence of the fault are comprehensively evaluated. The larger the calculated current waveform abnormality coefficient, the more obvious the pulse abnormality and waveform instability characteristics of the current signal in a single detection window. The schematic diagram of the current waveform abnormality coefficient acquisition process is shown below. Figure 3 As shown.

[0035] Step 2.2: Obtain the current spectrum anomaly coefficient of a single detection window by the average level of the high-frequency band energy ratio of all cycles and the degree of increase over time, as well as the significant peak characteristics of a single detection window other than the fundamental power frequency and its preset integer multiples.

[0036] Furthermore, when a non-destructive but potentially dangerous fault such as a high-resistance grounding occurs in a full-wave pulsating DC distribution system, the change in the fault current in the time domain may not be obvious and is easily masked by normal load fluctuations or environmental electromagnetic interference. However, these fault processes inject new and specific high-frequency harmonic components into the full-wave pulsating DC distribution system, leading to an overall increase in the background noise level of the current signal spectrum and the appearance of discrete abnormal peaks in the current signal spectrum. Therefore, further analysis is needed on these abnormal spectral characteristics generated by the fault that are independent of the stable power frequency.

[0037] Based on the above analysis, for a single detection window, Fourier transform technology is used to obtain the spectrum of the current signal in each cycle. The frequency with the largest amplitude in the spectrum of the current signal in each cycle is taken as the fundamental power frequency of each cycle. After removing the DC bias from the current signal in a single detection window, a Fourier transform is performed to obtain the spectrum. The frequency with the largest amplitude in the spectrum of the current signal in a single detection window is taken as the fundamental power frequency of the single detection window. By analyzing the average level of the high-frequency band energy ratio in all cycles and its increase over time, as well as the significance of peak characteristics other than the fundamental power frequency and its preset integer multiples in a single detection window, the current spectrum anomaly coefficient of a single detection window is obtained. The specific process is as follows: The background noise increase of a single detection window is obtained by measuring the average level of the high-frequency band energy percentage across all cycles and the degree of increase over time. The significance of abnormal spectral peaks in a single detection window is obtained by considering the significance of peak features other than the fundamental power frequency and its preset integer multiples in a single detection window. The weighted sum of the rise in background noise and the significance of the abnormal spectral peaks is used as the current spectrum anomaly coefficient for a single detection window.

[0038] The process for obtaining the background noise rise rate of a single detection window is as follows: Calculate the sum of the amplitudes of all frequencies greater than the base power frequency in the spectrum of the current signal within each cycle; The ratio of the sum of amplitudes to the sum of amplitudes at all frequencies in the spectrum of the current signal in each period is denoted as the high-frequency band total energy ratio. When the full-wave pulsating DC distribution system is operating normally, the high-frequency energy is mainly composed of a small amount of environmental interference such as switching noise, which accounts for a relatively low and stable proportion. However, when the full-wave pulsating DC distribution system fails, high-frequency band energy will be continuously injected, resulting in a significant increase in the calculated high-frequency band total energy ratio. Calculate the difference between the total high-frequency band energy ratio of the last cycle in a single detection window and the total high-frequency band energy ratio of all other cycles; map the difference to a positive number; calculate the arithmetic mean of the mapping results of all the differences corresponding to the last cycle in a single detection window; the larger the arithmetic mean is, the more significant the overall upward trend of the background noise level of the current signal in a single detection window is. The average of the total energy ratio of the high-frequency band in all cycles within a single detection window is denoted as the average high-frequency energy ratio. The larger the calculated average high-frequency energy ratio, the higher the proportion of high-frequency energy in the current signal within a single detection window. The rise in background noise of a single detection window is positively correlated with the arithmetic mean and the average proportion of high-frequency energy, respectively.

[0039] The process of obtaining the significance of abnormal spectral peaks in a single detection window is as follows: Extract the peak values ​​in a single detection window that are outside the fundamental power frequency and its preset integer multiples of the single detection window; obtain the abnormal peak value threshold of the single detection window by the distribution of the peak values ​​extracted in the single detection window, so as to filter out abnormal peak values ​​from the peak values ​​extracted in the single detection window; wherein, the DC bias is the 0Hz component; The ratio of the mean of all abnormal peaks in a single detection window to the amplitude of the fundamental power frequency of that window is used as the abnormal peak significance of that single detection window. The larger the calculated abnormal peak significance, the more significant the abnormal peak characteristics of the current signal in the single detection window. It should be noted that if there are no abnormal peaks, the abnormal peak significance is assigned a value of 0.

[0040] It should be noted that positive correlation means that the variables change in the same direction; when one variable increases, the other variable also increases, and when one variable decreases, the other variable also decreases.

[0041] In this embodiment, the preset integer multiplier is set to a maximum of 10 times.

[0042] In this embodiment, the difference is used as an exponent with the natural constant as the base to map the difference to a positive number. The natural constant is only one embodiment of this application. Implementers can replace the natural constant with other values ​​greater than 1 according to the actual situation. This application does not impose any special restrictions.

[0043] In this embodiment, the weighted sum of the normalized value of the arithmetic mean and the average value of the high-frequency energy proportion is used as the background noise rise of a single detection window.

[0044] In this embodiment, the mean of all the arithmetic means in the M detection windows during the normal operation of the full-wave pulsating DC distribution system is used as the reference value for normalizing the arithmetic mean in a single detection window; using the formula... The arithmetic mean is normalized, and X is assigned the arithmetic mean of a single detection window. If the value is assigned as the baseline value when normalizing the arithmetic mean in a single detection window, then Y is the normalized value of the arithmetic mean in a single detection window.

[0045] In this embodiment, the method for obtaining the abnormal peak threshold of a single detection window is as follows: the 3Sigma anomaly detection algorithm is used to obtain the 3Sigma range of the peak values ​​extracted from a single detection window, and the upper limit of the 3Sigma range is used as the abnormal peak threshold of a single detection window. The 3Sigma anomaly detection algorithm is a well-known technology and will not be described in detail here.

[0046] In this embodiment, peak values ​​that are greater than the abnormal peak value threshold among the peak values ​​extracted in each period are taken as abnormal peak values.

[0047] In this embodiment, the weight of the rise in background noise and the weight of the significance of the abnormal spectral peak are both 0.5, and the weights of the rise in background noise and the significance of the abnormal spectral peak are calculated from experimental data.

[0048] In this embodiment, the weight of the normalized value of the arithmetic mean and the weight of the average high-frequency energy ratio are both 0.5, and the weights of the normalized value of the arithmetic mean and the average high-frequency energy ratio are calculated from experimental data.

[0049] It should be noted that: the background noise rise rate reflects the average level of high-frequency energy proportion and the overall rise trend characteristics in a single detection window; the abnormal peak significance reflects the significance of abnormal peaks generated by current signal faults in a single detection window; thus, the abnormal spectral characteristics caused by faults and unrelated to stable power frequency are quantitatively evaluated. The larger the calculated current spectrum anomaly coefficient, the more obvious the overall rise characteristics of high-frequency band energy and abnormal peak characteristics caused by faults in a single detection window.

[0050] Step 2.3: Obtain the comprehensive current anomaly coefficient of a single detection window by using the current waveform anomaly coefficient and the current spectrum anomaly coefficient of a single detection window.

[0051] Furthermore, the faults occurring in full-wave pulsating DC distribution systems are quite complex and susceptible to load changes and environmental interference, leading to reduced accuracy in fault diagnosis. Therefore, fusing time-domain waveform anomaly characteristics with frequency-domain spectral anomaly characteristics helps to achieve comprehensive fault detection with stronger anti-interference capabilities and wider coverage.

[0052] Based on the above analysis, the comprehensive current anomaly coefficient of a single detection window is obtained by combining the current waveform anomaly coefficient and the current spectrum anomaly coefficient of a single detection window. The expression is as follows: In the formula, T represents the comprehensive anomaly coefficient of the current of a single detection window; , All represent preset weights greater than 0; F represents the current waveform anomaly coefficient of a single detection window; S represents the current spectrum anomaly coefficient of a single detection window.

[0053] In this embodiment, in scenarios where high-resistance grounding faults occur frequently, the abnormal characteristics reflected in the frequency domain are more obvious, thus increasing... The value of , , The values ​​are 0.4 and 0.6, respectively, and implementers can adjust them according to the actual situation. , The value of is not subject to any special restrictions in this application.

[0054] It should be noted that the larger the calculated current comprehensive anomaly coefficient, the more obvious the abnormal characteristics of the current signal on the pulsating DC bus [2] in a single detection window. The schematic diagram of the current comprehensive anomaly coefficient acquisition process is shown below. Figure 4 As shown.

[0055] Step 3: Based on the current comprehensive anomaly coefficient of a single detection window and its neighboring detection windows, predict the current comprehensive anomaly coefficient of the next detection window of the single detection window, and compare it with the preset fault monitoring threshold to assess the fault status of the full-wave pulsating DC power distribution system.

[0056] Faults in full-wave pulsating DC distribution systems typically exhibit progressive characteristics. While fluctuations may occur at individual points in time due to random disturbances, trend analysis of the combined anomaly characteristics across multiple consecutive detection windows helps determine whether a fault in a full-wave pulsating DC distribution system shows a trend of rapid escalation. Detecting such a trend usually indicates continuous deterioration of insulation performance or a slow change in grounding resistance, thus enabling early warning and preventative protection before the fault reaches a destructive level or triggers a trip.

[0057] Based on the above analysis, the comprehensive anomaly coefficient of the current in the next detection window of a single detection window is predicted according to the comprehensive anomaly coefficient of the current in a single detection window and its neighboring detection windows, and compared with a preset fault monitoring threshold to assess the fault status of the full-wave pulsating DC distribution system. The specific process is as follows: Predictive analysis is performed on the current comprehensive anomaly coefficient of a single detection window and the K detection windows preceding it to obtain the predicted value of the current comprehensive anomaly coefficient of the next detection window of the single detection window. Under normal operating conditions, the overall current comprehensive anomaly coefficient of a single detection window and the K detection windows preceding it is relatively small and stable. However, when a fault occurs, the current comprehensive anomaly coefficient of a single detection window and the K detection windows preceding it will exceed the average level under stable conditions. Therefore, a 3Sigma anomaly detection algorithm is used to obtain the 3Sigma range of the current comprehensive anomaly coefficient of a single detection window and the K detection windows preceding it. The maximum value between the upper limit of the 3Sigma range and the preset minimum absolute tolerance threshold is used as the preset fault monitoring threshold. If the predicted value of the current comprehensive anomaly coefficient of the next detection window of a single detection window is greater than the preset fault monitoring threshold, it is determined that there is a fault in the full-wave pulsating DC distribution system; otherwise, it is determined that the full-wave pulsating DC distribution system is operating normally.

[0058] In this embodiment, the value of K is 15. The value of K is preset by the user and the implementer can set it according to the actual situation. This application does not impose any special restrictions.

[0059] In this embodiment, the preset minimum absolute tolerance threshold is set to 0.2, and the value of the preset minimum absolute tolerance threshold is calculated from experimental data.

[0060] In this embodiment, the Holt linear trend prediction algorithm is used to predict and analyze the comprehensive anomaly coefficient of the current for a single detection window and the K preceding detection windows. The horizontal smoothing parameter is set to 0.3, and the trend smoothing parameter is set to 0.1. The comprehensive anomaly coefficients of the current for a single detection window and the K preceding detection windows are arranged chronologically. The first comprehensive anomaly coefficient in the arrangement is used as the initial horizontal smoothing value, and the difference between the second and first comprehensive anomaly coefficients is used as the initial trend smoothing value. The horizontal smoothing value and the trend smoothing value are iteratively updated using the iterative formula of the standard Holt linear trend prediction algorithm to obtain the predicted value of the comprehensive anomaly coefficient of the current for the next detection window. If the number of detection windows preceding a single detection window is less than K, the missing data is filled using edge padding. The standard Holt linear trend prediction algorithm and edge filling method are both well-known technologies, and will not be described in detail in this application. As other implementation methods, based on the ability to predict the current comprehensive anomaly coefficient of the next detection window of a single detection window according to the current comprehensive anomaly coefficient of a single detection window and the K detection windows before it, implementers may adopt other existing feasible technologies, and this application does not impose any special restrictions.

[0061] In summary, this application achieves adaptive period division through trough point detection and uses the period as the basic analysis unit. It can make full use of the inherent periodicity of full-wave pulsating DC, so that the current waveform under normal conditions has a predictable repetitive pattern, thereby providing a stable reference benchmark for the detection of deviations in abnormal features. The detection window containing multiple periods can ensure sufficient samples to suppress random noise. Furthermore, by integrating the three dimensions of peak significance, peak time interval randomness, and periodic waveform similarity, the current waveform anomaly coefficient is obtained. The three work together to give the current waveform anomaly coefficient a comprehensive ability to characterize time-domain fault features, significantly improving the fault detection's ability to distinguish non-fault disturbances and reducing the false alarm rate and missed alarm rate. Furthermore, by analyzing the average level of the high-frequency band energy proportion, the continuous high-frequency harmonic energy injected by the fault can be quantified, and fault types such as high-resistance grounding that do not change significantly in the time domain but have significant characteristics in the frequency domain can be identified; by the degree of increase of high-frequency band energy over time, the gradual trend of fault development can be captured; by extracting the abnormal peak value at the non-power frequency harmonic, the fault characteristic frequency can be accurately located, and the new harmonic components generated by the fault can be distinguished from the inherent switching frequency harmonics of the full-wave pulsating DC power distribution system. Furthermore, by integrating time-domain and frequency-domain anomaly characteristics for fault diagnosis, it can cover different types and stages of fault development, significantly enhancing the comprehensiveness of fault diagnosis and thus improving its accuracy. By predicting the trend of the comprehensive anomaly coefficients of multiple consecutive detection windows, it avoids misjudgments caused by single measurement anomalies, improves the accuracy of fault diagnosis, and enables early warning of faults. Alarms are issued before the fault reaches a destructive level or triggers protection tripping, thereby effectively ensuring the safe and stable operation of the full-wave pulsating DC power distribution system.

Claims

1. An automatic fault diagnosis method for full-wave pulsating DC power distribution systems, characterized in that, The method includes the following steps: Real-time acquisition of current signals on the pulsating DC bus in a full-wave pulsating DC power distribution system; The time interval between two adjacent troughs in the current signal is taken as a period, and multiple consecutive periods are combined into a detection window. For a single detection window, the current waveform anomaly coefficient of a single detection window is obtained by considering the significance of the current exceeding the effective value, the randomness of the peak occurrence in the current signal, and the similarity of the current signal waveform between any two adjacent periods. The current spectrum anomaly coefficient of a single detection window is obtained by considering the average level of the high-frequency band energy ratio of all periods and its increase over time, as well as the significance of the peak characteristics in a single detection window other than the basic power frequency and its preset integer multiples. This coefficient is then fused with the current waveform anomaly coefficient to obtain the comprehensive current anomaly coefficient of a single detection window. Based on the combined current anomaly coefficient of a single detection window and its neighboring detection windows, the combined current anomaly coefficient of the next detection window of the single detection window is predicted and compared with the preset fault monitoring threshold to assess the fault status of the full-wave pulsating DC power distribution system.

2. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 1, characterized in that, The process for obtaining the current waveform anomaly coefficient is as follows: The mean of the ratios of all peak values ​​to effective values ​​of the current signal in a single detection window is denoted as the significant mean. Calculate the dispersion of the time interval between any two adjacent peaks of the current signal in a single detection window; Calculate the average of the maximum mutual information coefficients of the current signals between any two adjacent cycles in a single detection window; The current waveform anomaly coefficient is positively correlated with the significant mean and the dispersion, and negatively correlated with the average value.

3. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 2, characterized in that, The calculation process for the current waveform anomaly coefficient is as follows: The dispersion and the significant mean are respectively subjected to positive normalization; the average value is subjected to inverse normalization. The current waveform anomaly coefficient is a weighted sum of the normalized value of the dispersion, the normalized value of the significant mean, and the normalized value of the average.

4. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 1, characterized in that, The fundamental power frequency of a single detection window is the frequency with the largest amplitude in the spectrum of the current signal in that single detection window.

5. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 1, characterized in that, The process for determining the current spectrum anomaly coefficient is as follows: The background noise increase of a single detection window is obtained by measuring the average level of the high-frequency band energy percentage across all cycles and the degree of increase over time. The significance of abnormal spectral peaks in a single detection window is obtained by considering the significance of peak features other than the fundamental power frequency and its preset integer multiples in a single detection window. The current spectrum anomaly coefficient is a weighted sum of the rise in background noise and the significance of the abnormal spectral peak.

6. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 5, characterized in that, The process for obtaining the background noise rise rate is as follows: Calculate the sum of the amplitudes of all frequencies greater than the base frequency in the spectrum of the current signal within each cycle; The ratio of the sum of amplitudes to the sum of amplitudes at all frequencies in the spectrum of the current signal in each period is denoted as the high-frequency band total energy ratio. The background noise rise of a single detection window is obtained by comparing the total energy ratio of the high-frequency band in the last cycle to the remaining cycles in a single detection window, and combining this with the average level.

7. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 6, characterized in that, The calculation process for the rise rate of the background noise is as follows: Calculate the difference between the total high-frequency band energy ratio of the last cycle in a single detection window and the total high-frequency band energy ratio of all other cycles; map the difference to a positive number; Calculate the arithmetic mean of the mapping results of all the differences corresponding to the last period in a single detection window; The average of the total energy ratio of the high-frequency band in all cycles within a single detection window is denoted as the average high-frequency energy ratio. The rise rate of the background noise is positively correlated with the arithmetic mean and the average proportion of high-frequency energy, respectively.

8. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 5, characterized in that, The process for obtaining the significance of the abnormal spectral peaks is as follows: After removing the DC bias from the current signal in a single detection window, a Fourier transform is performed to obtain the spectrum. Peak values ​​located in the single detection window other than the fundamental power frequency and its preset integer multiples are extracted. The abnormal peak value threshold of the single detection window is obtained by analyzing the distribution of the peak values ​​extracted from the single detection window, so as to filter out abnormal peak values ​​from the peak values ​​extracted from the single detection window. The significance of abnormal spectral peaks in a single detection window is obtained by analyzing the distribution of all abnormal peaks within that window.

9. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 8, characterized in that, The significance of the abnormal spectral peaks is the ratio of the mean of all abnormal peak values ​​in a single detection window to the amplitude of the fundamental power frequency of the single detection window.

10. The automatic fault diagnosis method for a full-wave pulsating DC power distribution system as described in claim 1, characterized in that, The current composite anomaly coefficient is the weighted sum of the current waveform anomaly coefficient and the current spectrum anomaly coefficient.

Images