A method for online monitoring of dielectric loss of power capacitor bank

Abstract

The present invention discloses an online monitoring method for dielectric loss of a power capacitor group, which specifically relates to the technical field of insulation status monitoring of power system equipment, and is used to solve the problems of insufficient signal synchronization accuracy and large dielectric loss calculation error in existing methods under complex electromagnetic interference; the method comprises: real-time acquisition of voltage and current signals and dynamic tracking of fundamental frequency; intercepting the main cycle waveform and constructing a phase space trajectory to analyze complexity, combining the symbol sequence transfer entropy to generate distortion characteristics; dynamically adjusting filtering parameters based on the distortion characteristics and Lyapunov exponent gradient; calculating high-precision phase difference through zero-crossing interpolation correction and sliding window filtering, and finally determining the dielectric loss value; and effectively suppressing noise interference through dynamic frequency tracking, collaborative analysis of nonlinear characteristics and adaptive filtering strategies, thereby improving phase synchronization accuracy and dielectric loss calculation reliability, and providing accurate and real-time monitoring support for insulation performance evaluation of power capacitor groups.

Description

Technical Field

[0001] The present invention relates to the technical field of insulation status monitoring of power system equipment, and more particularly to an online dielectric loss monitoring method for a power capacitor group. Background Art

[0002] Power capacitor banks are key equipment used for reactive power compensation and voltage regulation in power systems. Real-time monitoring of their insulation performance is crucial to ensuring the safe operation of the power grid. Online detection of the dielectric loss tangent can directly reflect the internal insulation state of the capacitor, which is usually achieved by synchronously collecting the voltage and current signals of the capacitor during operation and calculating the phase difference. In existing technologies, dielectric loss monitoring of power capacitor banks is usually implemented in complex electromagnetic environments such as substations and transmission lines. However, there are many strong electromagnetic interference sources in the field environment, which have a significant impact on the reliability of signal acquisition.

[0003] In the existing technology, the online dielectric loss monitoring method is difficult to ensure the phase synchronization accuracy of voltage and current signals under low signal-to-noise ratio conditions. Since electromagnetic interference in the operating environment of the power capacitor group can easily lead to signal waveform distortion and noise superposition, the traditional synchronization method is prone to deviations in the process of signal feature extraction and phase alignment, which directly affects the calculation accuracy of the dielectric loss value, and thus causes the risk of misjudgment of the insulation status, making it difficult to meet the needs of high-precision online monitoring. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an online monitoring method for dielectric loss of a power capacitor bank to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for online monitoring of dielectric loss of a power capacitor bank comprises the following steps:

[0007] S1. Acquire the voltage signal and current signal of the power capacitor bank in real time and record the signal acquisition time series;

[0008] S2. Dynamically track the fundamental frequency of the voltage signal and the current signal according to the distribution of the zero point and extreme point of the waveform of the voltage signal and the current signal;

[0009] S3, intercepting a voltage signal segment and a current signal segment of a preset length according to the fundamental frequency, and extracting the main cycle waveform corresponding to the fundamental frequency in the voltage signal segment and the current signal segment;

[0010] S4. Construct the phase space trajectory of the main cycle waveform and analyze its complexity. At the same time, convert the main cycle waveform into a symbol sequence and calculate the transfer entropy between the symbol sequences. Generate signal distortion characteristics based on the complexity and transfer entropy.

[0011] S5. Filtering the voltage signal segment and the current signal segment according to the signal distortion characteristics and the maximum Lyapunov exponent gradients of the voltage signal segment and the current signal segment;

[0012] S6. Calculate the phase difference between the voltage signal and the current signal according to the zero-crossing timing difference of the main cycle waveforms of the filtered voltage signal and the current signal to determine the dielectric loss tangent of the power capacitor bank.

[0013] In a preferred embodiment, the voltage signal and current signal of the power capacitor bank are acquired in real time, and the signal acquisition time series is recorded, including:

[0014] The synchronous acquisition of voltage and current signals is started by a synchronous trigger signal, ensuring that the acquisition time deviation of voltage and current signals is less than a preset time threshold;

[0015] Align the sampling timestamps of the voltage signal and the current signal to the same time base, and generate a voltage signal sequence and a current signal sequence containing the timestamps;

[0016] Amplitude mutation detection is performed on the voltage signal sequence and the current signal sequence, and abnormal signal segments whose amplitude mutation exceeds the preset amplitude threshold are eliminated.

[0017] In a preferred embodiment, dynamically tracking the fundamental frequency of the voltage signal and the current signal according to the distribution of the waveform zero points and extreme points of the voltage signal and the current signal includes:

[0018] Extracting the waveform zero points and extreme points of the voltage signal sequence and the current signal sequence within a preset time window;

[0019] Determine the fundamental wave period of the voltage signal sequence and the current signal sequence according to the time interval between adjacent waveform zero points;

[0020] The extreme point offset errors of the voltage signal sequence and the current signal sequence are corrected by the cubic spline interpolation method to obtain the corrected fundamental wave period;

[0021] The corrected inverse of the fundamental wave period is taken as the fundamental wave frequency.

[0022] In a preferred embodiment, according to the fundamental frequency, a voltage signal segment and a current signal segment of a preset duration are intercepted, and a main cycle waveform corresponding to the fundamental frequency in the voltage signal segment and the current signal segment is extracted, including:

[0023] Determine the preset time length as an integer multiple of the fundamental wave period according to the fundamental wave frequency, and intercept the voltage signal segment and current signal segment of the corresponding time length from the voltage signal sequence and the current signal sequence;

[0024] In the intercepted voltage signal segment and current signal segment, the starting zero-crossing point of the main cycle waveform is located using the cycle length corresponding to the fundamental frequency as a unit;

[0025] Starting from the initial zero-crossing point, the voltage signal segment and the current signal segment within a complete fundamental wave cycle are extracted as the main cycle waveform;

[0026] The extracted main cycle waveform is processed by mean filtering to eliminate random noise interference.

[0027] In a preferred embodiment, a phase space trajectory of a main cycle waveform is constructed and its complexity is analyzed. At the same time, the main cycle waveform is converted into a symbol sequence and the transfer entropy between the symbol sequences is calculated. The signal distortion feature is generated based on the complexity and transfer entropy, including:

[0028] The phase space trajectory of the main periodic waveform is constructed using the time delay embedding method;

[0029] Calculate the recursive quantitative analysis parameters of the phase space trajectory, the recursive quantitative analysis parameters include recursion rate and certainty, and generate the phase space complexity index according to the recursion rate and certainty;

[0030] When converting the main cycle waveform into a symbol sequence, the symbol intervals are divided according to the amplitude distribution of the main cycle waveform, and each symbol interval corresponds to a preset amplitude range;

[0031] The transition probabilities of adjacent symbols in a statistical symbol sequence are calculated, and the transition entropy between symbol sequences is calculated based on the transition probabilities;

[0032] The signal distortion feature is generated according to the product of the normalized phase space complexity index and the transfer entropy, or when the phase space complexity index exceeds a preset threshold set based on historical data statistics, the transfer entropy is directly used as the signal distortion feature.

[0033] In a preferred embodiment, the delay time of the time delay embedding method is set according to the sampling rate of the main periodic waveform, and the embedding dimension is set according to the spectrum characteristics of the main periodic waveform.

[0034] In a preferred embodiment, filtering the voltage signal segment and the current signal segment according to the signal distortion characteristics and the maximum Lyapunov exponent gradient of the voltage signal segment and the current signal segment includes:

[0035] Calculate the maximum Lyapunov exponent gradient of the voltage signal segment and the current signal segment. The maximum Lyapunov exponent gradient is calculated by the rate of change of the maximum Lyapunov exponent in adjacent time windows.

[0036] Adjusting the cutoff frequency or order of the filter based on the comparison result of the absolute value of the maximum Lyapunov exponent gradient and the preset gradient threshold;

[0037] If the signal distortion characteristic exceeds the preset distortion threshold and the absolute value of the maximum Lyapunov exponent gradient exceeds the preset gradient threshold, the voltage signal segment and the current signal segment are filtered using adaptive mean filtering;

[0038] If the signal distortion characteristic does not exceed the preset distortion threshold or the absolute value of the maximum Lyapunov exponent gradient does not exceed the preset gradient threshold, a low-pass filter with fixed parameters is used.

[0039] In a preferred embodiment, the filter window size is dynamically set according to the absolute value of the maximum Lyapunov exponent gradient.

[0040] In a preferred embodiment, the phase difference between the voltage signal and the current signal is calculated based on the zero-crossing timing difference of the main cycle waveforms of the filtered voltage signal and the current signal to determine the dielectric loss tangent of the power capacitor bank, including:

[0041] In the main cycle waveforms of the filtered voltage and current signals, the zero crossing points of the voltage and current waveforms are located respectively, and the zero crossing points are determined by sign change detection and interpolation correction;

[0042] Calculate the time difference between the zero-crossing point of the voltage waveform and the zero-crossing point of the current waveform as the zero-crossing timing difference;

[0043] Convert the zero-crossing timing difference into a phase difference value according to the fundamental frequency;

[0044] The dielectric loss tangent value is calculated based on the phase difference value, and the dielectric loss tangent value is the tangent function value of the phase difference value.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. Through dynamic tracking of fundamental frequency and coordinated processing with adaptive filtering, the signal synchronization accuracy in complex electromagnetic environments is effectively improved. The fundamental frequency tracking mechanism based on the distribution of waveform zero points and extreme points can perceive the frequency fluctuations of voltage and current signals in real time. Combining phase space trajectory complexity analysis with two-dimensional distortion feature extraction of symbol sequence transfer entropy, it accurately quantifies the interference intensity, breaking through the limitations of traditional methods that rely on fixed frequency bands or single parameters. Through the dynamic decision logic of the maximum Lyapunov exponent gradient and distortion characteristics, the filtering strategy is adaptively adjusted to suppress high-frequency noise and chaotic interference while retaining the details of the fundamental waveform, significantly improving signal purity and providing a highly reliable data foundation for subsequent phase difference calculation.

[0047] 2. A multi-stage error suppression system is adopted to form a closed-loop optimization from signal acquisition, feature extraction to filtering processing; through zero-crossing interpolation correction and sliding window mean filtering, the timing jitter error caused by noise is systematically eliminated, and the phase difference calculation benchmark is combined with the real-time fundamental frequency dynamic calibration to ensure high-precision output of the dielectric loss value; through nonlinear feature fusion and dynamic parameter adaptation, the anti-interference ability and computing efficiency are taken into account, avoiding the error accumulation problem caused by the time-varying characteristics of the environment in traditional methods, providing more stable and sensitive monitoring support for the insulation status assessment of power capacitor banks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flow chart of a method for online monitoring of dielectric loss of a power capacitor bank. DETAILED DESCRIPTION

[0049] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] Example: Figure 1 The present invention provides an online monitoring method for dielectric loss of a power capacitor bank, comprising the following steps:

[0051] S1. Acquire the voltage signal and current signal of the power capacitor bank in real time and record the signal acquisition time series;

[0052] S2. Dynamically track the fundamental frequency of the voltage signal and the current signal according to the distribution of the zero point and extreme point of the waveform of the voltage signal and the current signal;

[0053] S3, intercepting a voltage signal segment and a current signal segment of a preset length according to the fundamental frequency, and extracting the main cycle waveform corresponding to the fundamental frequency in the voltage signal segment and the current signal segment;

[0054] S4. Construct the phase space trajectory of the main cycle waveform and analyze its complexity. At the same time, convert the main cycle waveform into a symbol sequence and calculate the transfer entropy between the symbol sequences. Generate signal distortion characteristics based on the complexity and transfer entropy.

[0055] S5. Filtering the voltage signal segment and the current signal segment according to the signal distortion characteristics and the maximum Lyapunov exponent gradients of the voltage signal segment and the current signal segment;

[0056] S6. Calculate the phase difference between the voltage signal and the current signal according to the zero-crossing timing difference of the main cycle waveforms of the filtered voltage signal and the current signal to determine the dielectric loss tangent of the power capacitor bank.

[0057] S1. Real-time acquisition of voltage and current signals of the power capacitor bank and recording of signal acquisition time series can be specifically implemented as follows:

[0058] The synchronous acquisition of voltage and current signals is initiated via a synchronous trigger signal, ensuring that the acquisition time deviation of the voltage and current signals is less than a preset time threshold. The synchronous trigger signal is generated by receiving a synchronous pulse signal from the power system via an external trigger device, or by a periodic square wave signal generated by an internal high-precision clock. The preset time threshold is set to one thousandth of the power frequency period of the power capacitor bank. For example, when the power frequency is 50Hz, the power frequency period is 20ms, and the preset time threshold is set to 20μs. This threshold ensures that the phase difference calculation error is less than 0.01° through sampling precision redundancy design. The specific logic is: when the sampling interval is less than one thousandth of the fundamental wave period, the zero-crossing positioning time error is ≤0.02μs, and the corresponding phase difference error is ≤0.00036° (much less than 0.01°).

[0059] The sampling timestamps of the voltage and current signals are aligned to the same time base, generating a time-stamped voltage and current signal sequence. The time base is either the Coordinated Universal Time output by the GPS timing module or an internal high-precision clock locked to the same oscillator via a phase-locked loop circuit. The specific method for timestamp alignment is as follows: If the sampling timestamp deviation between the voltage and current signals is within a preset time threshold, the instantaneous value at the alignment time point is calculated using linear interpolation. For example, if the sampling value of the voltage signal at time point t1 is V(t1), and the sampling value of the current signal at time point t2 is I(t2), when the time deviation |t1-t2| ≤ 20μs, the alignment value V(t) is calculated as V(t1) + (V(t1+Δt)-V(t1)) × (t-t1) / Δt, based on the intermediate time point t = (t1+t2) / 2, where Δt is the sampling interval.

[0060] Amplitude mutation detection is performed on the voltage signal sequence and the current signal sequence, and abnormal signal segments with amplitude mutations exceeding the preset amplitude threshold are eliminated. Amplitude mutation detection is achieved by calculating the amplitude change rate of adjacent sampling points. If the change rate exceeds the preset amplitude threshold, it is determined to be a mutation point. The preset amplitude threshold sets the initial value based on 10% of the rated voltage or current value of the power capacitor group. For example, when the rated voltage is 10kV, the initial threshold is set to 1kV, and the amplitude standard deviation σ within the last 100 power frequency cycles is statistically calculated through a sliding window and dynamically adjusted to 3σ. The elimination range of abnormal signal segments is centered on the mutation point and extends half a power frequency cycle forward and backward. For example, an extension of 10ms corresponds to a power frequency of 50Hz.

[0061] The synchronization accuracy between the trigger signal and the time base is verified by measuring the delay of a standard signal in an interference-free environment. If the delay exceeds a preset time threshold, the clock synchronization circuit is adjusted or the trigger source is replaced. Linear interpolation error control is achieved by limiting the timestamp deviation to no more than half the sampling interval. If it exceeds this limit, the data is discarded and re-acquired. Dynamic adjustment of the amplitude mutation threshold is performed every 5 minutes to ensure adaptation to real-time operating conditions.

[0062] The external trigger device uses an optocoupler isolation interface to prevent interference from ground potential differences. The internal high-precision clock maintains frequency stability through constant temperature control, with temperature fluctuations controlled within ±0.1°C. If the synchronization error of the GPS timing module's pulse-per-second signal exceeds 1μs, an alarm is triggered and the system switches to the internal clock. Abnormal segments detected during amplitude mutation detection are marked as invalid data and skipped during subsequent processing.

[0063] S2. Dynamically tracking the fundamental frequency of the voltage signal and the current signal according to the distribution of the waveform zero points and extreme points of the voltage signal and the current signal can be specifically implemented as follows:

[0064] Extract the waveform zero points and extreme points of the voltage and current signal sequences within a preset time window. The length of the preset time window is set to an integer multiple of the power frequency cycle of the power capacitor bank. For example, when the power frequency is 50Hz, the power frequency cycle is 20ms, and the preset time window is set to 5 power frequency cycles, or 100ms. When extracting zero points, traverse the sampling points of the voltage and current signal sequences. If the signs of two adjacent sampling points are opposite (such as from positive to negative or from negative to positive), they are determined to be zero points. When extracting extreme points, traverse the sampling point sequence. If the amplitude of a sampling point is greater than the amplitude of the previous and next points, it is determined to be a maximum point; otherwise, it is a minimum point.

[0065] The fundamental period of the voltage and current signal sequences is determined based on the time intervals between adjacent zero points in the waveform. The fundamental period is calculated by counting the time intervals between all adjacent zero points within a preset time window, removing abnormal intervals that deviate by ±20% from the average, and taking the arithmetic mean of the remaining intervals as the current fundamental period. For example, if six zero points are detected within a 100ms time window, with adjacent intervals of 19.8ms, 20.1ms, 20.3ms, 19.5ms, and 20.5ms, respectively, after removing the interval of 19.5ms (which is less than 80% of the average value of 20ms), the average of the remaining intervals is 20.1ms, indicating a fundamental period of 20.1ms.

[0066] The extreme point offset errors of the voltage and current signal sequences are corrected using the cubic spline interpolation method to obtain the corrected fundamental period. The method for correcting the extreme point offset error includes taking the extreme point as the center, selecting two sampling points before and after it, and five points in total as interpolation nodes, reconstructing the waveform curve of the interval using the cubic spline interpolation method, and relocating the precise position of the extreme point. For example, the original position of a certain maximum point is sampling point n. Through interpolation, it is found that the actual extreme value is between points n and n+1, and the offset is 0.3 sampling intervals. The corrected extreme point position is then n+0.3. The corrected fundamental period is recalculated by updating the zero point interval. For example, the corrected zero point interval is adjusted from 20.1ms to 19.9ms.

[0067] The inverse of the corrected fundamental period is used as the real-time fundamental frequency. The fundamental frequency is calculated as: fundamental frequency = 1 / fundamental period. For example, when the fundamental period is 19.9 ms, the fundamental frequency is 1 / 0.0199 ≈ 50.25 Hz. The real-time fundamental frequency is updated every preset time window, for example, outputting a frequency value every 100 ms to ensure dynamic tracking of power frequency fluctuations of the power capacitor bank.

[0068] The preset time window, as an integer multiple, must cover at least three power frequency cycles to avoid zero-point detection errors caused by short-term interference. The threshold for rejecting abnormal intervals (±20%) is determined based on historical operating data. For example, when the capacitor bank is operating normally, the fundamental period fluctuation range does not exceed ±2%. Therefore, a ±20% threshold is set to eliminate the impact of sudden interference. The number of nodes in the cubic spline interpolation method is set to five to ensure a smooth interpolation curve and a manageable computational load.

[0069] The boundary conditions for the cubic spline interpolation method are set to natural spline conditions, meaning that the second derivative is zero at the endpoints to ensure smoothness at the boundaries of the interpolated curve. The interpolation interval is centered around the extreme point and extends two sampling points before and after it. For example, at a sampling rate of 10kHz, each sampling interval is 0.1ms, and the interpolation interval covers 0.5ms, ensuring that the waveform characteristics near the extreme point are included.

[0070] The dynamic fundamental frequency update logic includes the following: If the fundamental frequency calculated in two adjacent preset time windows deviates by more than ±0.5Hz, a frequency jump alarm is triggered, and the frequency value is smoothed using a sliding average method. For example, if the current window frequency is 50.25Hz and the previous window frequency is 49.8Hz, the deviation is 0.45Hz, which does not exceed the threshold. If the frequency in the next window suddenly changes to 51.0Hz, an alarm is triggered and the smoothed value (50.25 + 51.0) / 2 = 50.625Hz is output.

[0071] Zero and extreme point extraction is implemented using a sliding window in real-time signal processing. Each time a new sample point is received, the window data is updated and zero and extreme point detection is re-executed. The sliding window is updated every sampling interval. For example, at a sampling rate of 10kHz, the window data is updated every 0.1ms to ensure real-time performance.

[0072] S3, intercepting a voltage signal segment and a current signal segment of a preset length according to the fundamental frequency, and extracting the main cycle waveform corresponding to the fundamental frequency in the voltage signal segment and the current signal segment, can be specifically implemented as follows:

[0073] Based on the fundamental frequency, a preset duration is determined as an integer multiple of the fundamental period, and voltage and current signal segments of corresponding duration are extracted from the voltage and current signal sequences. The integer multiples of the fundamental period are set to include at least five complete fundamental periods. For example, when the fundamental frequency is 50 Hz, the fundamental period is 20 ms, and the preset duration is set to 5 fundamental periods, or 100 ms. When extracting the voltage and current signal segments, data for the preset duration is extracted forward from the current moment, ensuring that the extracted segments contain complete periodic waveforms. For example, if the current moment is t, the voltage and current signal sequences within the time range [t-100 ms, t] are extracted. The number of integer multiples of the fundamental period is dynamically adjusted based on the operating stability of the power capacitor bank. For example, when the frequency fluctuation exceeds ±0.5 Hz, the integer multiple is increased to 10 periods to improve the anti-interference capability of the extracted segments.

[0074] In the intercepted voltage and current signal segments, the starting zero-crossing point of the main cycle waveform is located using the cycle length corresponding to the fundamental frequency as a unit. The method for locating the starting zero-crossing point is as follows: starting from the starting point of the intercepted voltage and current signal segments, a sliding search window is moved in steps of the fundamental period, and the first zero-crossing point within each window is detected as the starting point of the main cycle waveform. For example, when the fundamental period is 20ms, each 20ms window is a search window, and the first zero point within the window where the voltage signal changes from negative to positive is the starting zero-crossing point. If no zero-crossing point is detected within the search window, the search range is expanded to adjacent windows until the zero-crossing point is successfully located. The specific operation of expanding the search range is to extend the window length to 1.5 times the fundamental period, for example, 30ms, and re-detect the zero-crossing point within the expanded window. If the zero-crossing point still cannot be located, the window is skipped and the search continues from the next window.

[0075] Starting from the starting zero-crossing point, extract the voltage and current signal segments within a complete fundamental cycle as the main cycle waveform. To extract a complete fundamental cycle, use the starting zero-crossing point as the starting point and extract data backwards for the length of one fundamental cycle. For example, if the fundamental cycle is 20 ms, extract the voltage and current signal segments from the starting point to 20 ms after the starting point. If the end of the extracted segment does not contain a complete cycle, adjust the starting point forward to ensure cycle integrity. For example, if the starting zero-crossing point is at t0, the extracted segment is [t0, t0+20 ms]. If t0+20 ms is outside the signal segment range, adjust the starting point to t0-Δt to complete the extracted segment. The Δt for the adjusted starting point is calculated as: Δt = fundamental cycle - (t_max-t0), where t_max is the maximum timestamp of the signal segment. For example, if t_max is t0+15 ms, Δt = 20-15 = 5 ms, resulting in the extracted segment being [t0-5 ms, t0+15 ms].

[0076] The extracted main cycle waveform is subjected to mean filtering to eliminate random noise interference. The specific method for mean filtering is: for each sampling point in the main cycle waveform, the arithmetic mean of its value and the two adjacent sampling points before and after is calculated, and the original sampling value is replaced. For example, if the value of a certain sampling point n is V(n), then the filtered value V'(n) = [V(n-1) + V(n) + V(n+1)] / 3. The filtering operation is performed separately on the voltage signal segment and the current signal segment to ensure noise suppression while preserving the fundamental waveform characteristics. The boundary processing method of mean filtering is: for the starting and ending points of the signal segment, the mirror extension method is used to generate virtual sampling points. For example, the value of the virtual preceding point n = -1 of the starting point n = 0 is V(-1) = V(0), and the value of the virtual following point n = N + 1 of the ending point n = N is V(N+1) = V(N).

[0077] The logic for adjusting the start point of the intercepted segment includes the following: If the end of the intercepted segment exceeds the signal segment range, the starting point is moved forward by the remaining time. For example, if the intercepted segment requires 20ms but the remaining time is only 15ms, the starting point is moved forward 5ms to ensure the interception of the entire cycle. The trigger condition for dynamically adjusting the number of integer multiples is: if the standard deviation of the fundamental frequency exceeds 0.3Hz within three consecutive preset time windows, the number of integer multiples will be automatically increased. For example, if the standard deviation of the fundamental frequency within three 100ms windows is 0.4Hz, the integer multiple will be increased from 5 to 10.

[0078] The extraction and filtering of the main periodic waveform are implemented in real-time using a circular buffer. Each time a new sampling point is received, the buffer data is updated, and the extraction, positioning, and filtering are re-executed. The buffer length is double the preset duration. For example, if the preset duration is 100ms, the buffer length is 200ms, ensuring that sufficient data is always available for extraction. The circular buffer update logic is as follows: when a new sampling point arrives, the oldest data point in the buffer is removed and the new point is added to the end of the buffer to maintain a constant buffer length. For example, the original buffer data is [t-200ms, t], and the new sampling point is updated to [t-199.9ms, t+0.1ms].

[0079] The number of sampling points for the mean filter is set to three to balance noise suppression effectiveness and computational complexity. The mirror extension method operates by copying the value of the first sampling point at the beginning of the signal segment as the virtual front point, and the value of the last sampling point at the end as the virtual back point. For example, if the signal segment is [V(0), V(1),..., V(N)], the extended signal becomes [V(0), V(0), V(1),..., V(N), V(N)]. The upper limit of the number of dynamically adjusted integer multiples of cycles is set to 20 cycles to avoid real-time performance degradation caused by excessive extension of the intercepted segment.

[0080] S4. Construct the phase space trajectory of the main cycle waveform and analyze its complexity. At the same time, convert the main cycle waveform into a symbol sequence and calculate the transfer entropy between the symbol sequences. Generate signal distortion characteristics based on the complexity and transfer entropy. The specific implementation can be as follows:

[0081] The time delay embedding method is used to construct the phase space trajectory of the primary cycle waveform. The delay time is set according to the sampling rate of the primary cycle waveform, and the embedding dimension is set according to the spectral characteristics of the primary cycle waveform. The delay time is set as one-quarter of the sampling interval of the primary cycle waveform. For example, when the sampling rate is 10kHz and the sampling interval is 0.1ms, the delay time is set to 0.025ms. The embedding dimension is set by performing a fast Fourier transform on the primary cycle waveform and extracting the spectral amplitude corresponding to the fundamental frequency. If the fundamental amplitude accounts for more than 90% of the total spectral energy, the embedding dimension is set to 3; otherwise, it is set to 5. For example, if the fundamental amplitude of a primary cycle waveform accounts for 85% of the total energy, the embedding dimension is set to 5 to capture the high-frequency components. The phase space trajectory construction process of the time delay embedding method is to combine each sampling point of the primary cycle waveform with the points of the subsequent delay time interval to form a trajectory point in high-dimensional space. For example, when the embedding dimension is 3, the trajectory point consists of three consecutive delayed points: V(t), V(t+τ), and V(t+2τ), where τ is the delay time.

[0082] Calculate the recursive quantitative analysis parameters of the phase space trajectory. These parameters include the recursion rate and determinism. A phase space complexity index is generated based on the recursion rate and determinism. The recursion rate is calculated by constructing a recursion graph of the phase space trajectory and counting the proportion of all recursion points in the graph as the recursion rate. The recursion graph is constructed by calculating the Euclidean distance between trajectory points. If the distance is less than a preset radius (e.g., 10% of the average distance between trajectory points), the point is marked as a recursion point. For example, if there are 1000 trajectory points and the average distance is 0.5V, the preset radius is 0.05V. The number of recursion points that meet the condition, N_rec, is counted, and the recursion rate = N_rec / 1000². The determinism is calculated by counting the proportion of diagonal structures in the recursion graph. A diagonal structure is defined as a line segment formed by two consecutive recursion points with a length ≥ 2. For example, if there are 200 valid diagonals in the recursion graph and the total number of recursion points is 500, then the determinism = 200 / 500 = 0.4. The phase space complexity index is derived from the ratio of the recursion rate to the determinism. For example, if the recursion rate is 0.6 and the determinism is 0.4, the complexity index is 0.6 / 0.4 = 1.5. If the determinism is zero, the complexity index is set to a preset maximum value (e.g., 10) to avoid division by zero errors.

[0083] When converting the main cycle waveform into a symbol sequence, symbol intervals are divided based on the amplitude distribution of the main cycle waveform. Each symbol interval corresponds to a preset amplitude range. The symbol intervals are divided by calculating the mean and standard deviation of the main cycle waveform amplitude, dividing the amplitude range into three intervals: the low amplitude interval (below the mean - 1 standard deviation), the medium amplitude interval (mean ± 1 standard deviation), and the high amplitude interval (above the mean + 1 standard deviation), corresponding to the symbols "0," "1," and "2," respectively. For example, if the main cycle waveform amplitude has a mean of 5V and a standard deviation of 1V, the symbol intervals are: low amplitude (<4V), medium amplitude (4V-6V), and high amplitude (>6V). If the waveform amplitude exceeds the historical statistical range, the interval boundaries are dynamically adjusted to cover the current amplitude. This dynamic adjustment is achieved by recalculating the mean and standard deviation of the current main cycle waveform and updating the symbol intervals based on the new parameters. For example, if a main cycle waveform has an amplitude of 8V (the original high amplitude range is >6V), the recalculated mean is 5.5V, the standard deviation is 1.2V, and the updated high amplitude range is >6.7V.

[0084] The transition probabilities of adjacent symbols in a symbol sequence are counted, and the transition entropy between symbol sequences is calculated based on these transition probabilities. The transition probability is calculated by traversing the symbol sequence, counting the number of occurrences of each pair of adjacent symbols (e.g., "0→1" and "1→2"), and dividing this by the total number of transitions to obtain the probability. For example, if the symbol sequence length is 1000, the total number of symbol transitions is 999, and "0→1" occurs 300 times, the probability is 300 / 999≈0.3. The transition entropy is calculated by taking the logarithm of the transition probabilities of each pair of adjacent symbols and weighted summing them. The formula is the negative weighted sum. For example, if the probability of "0→1" in a symbol sequence is 0.3 and the probability of "1→2" is 0.2, the transition entropy is -(0.3×log20.3+0.2×log20.2)≈0.521+0.464=0.985. If a pair of symbols does not appear (e.g., “2→0”), its probability is set to the minimum value (e.g., 1e-6) to avoid logarithmic calculation errors.

[0085] Signal distortion features are generated by multiplying the normalized phase space complexity index and the transfer entropy. Alternatively, when the phase space complexity index exceeds a preset threshold based on historical data statistics, the transfer entropy is directly used as the signal distortion feature. The normalization method involves scaling the phase space complexity index and the transfer entropy to the range of 0-1. The complexity index normalization formula is: Normalized value = Original value / Historical maximum value. For example, if the historical maximum complexity is 10 and the current value is 8, the normalized value is 0.8. The transfer entropy normalization formula is: Normalized value = Original value / Theoretical maximum value. The theoretical maximum entropy is determined by the number of symbols (e.g., the theoretical maximum entropy for 3 symbols is log23≈1.585). When the current entropy value is 1.0, the normalized value is ≈0.63. The preset threshold is set by statistically analyzing the mean and standard deviation of the phase space complexity index in the historical normal operation data of the power capacitor bank and setting the threshold as the mean plus twice the standard deviation. For example, if the historical mean is 1.2 and the standard deviation is 0.3, the threshold is 1.2 + 2 × 0.3 = 1.8. When the real-time complexity index exceeds 1.8, the transfer entropy is used directly as the distortion feature; otherwise, the normalized product is used. For example, if the normalized complexity is 0.7 (original value 7) and the transfer entropy is 0.4 (original value 2), the product is 0.7 × 0.4 = 0.28. If the threshold is 0.8 (corresponding to the original value 8), the transfer entropy of 0.4 is used directly.

[0086] The default radius adjustment method for the recursion graph is: if the recursion rate is lower than 5% or higher than 95%, the radius is automatically adjusted to 5% or 20% of the average distance of the current trajectory points to balance the recursion point density. For example, if the recursion rate is 3% with an initial radius of 0.05V, the radius is adjusted to 0.025V (5% of the average distance). The dynamic adjustment trigger condition for the symbol interval is: the amplitude of three consecutive main cycle waveforms exceeds the current interval range, triggering the recalculation of the mean and standard deviation. For example, if the high amplitude of three consecutive waveforms is 7V, 7.2V, and 7.5V (the original high amplitude range is >6V), the interval update is triggered.

[0087] The maximum value of the phase space complexity index is set to twice the maximum ratio observed in the historical data. For example, if the historical maximum ratio is 5, the preset maximum value is 10. The logic for processing the minimum value of transfer entropy is as follows: if all transition probabilities are minimum (for example, all are 1e-6), the signal is considered to be completely random noise, and the transfer entropy is set to 0. The logical order for executing the normalized product and threshold judgment is as follows: first check whether the complexity index exceeds the threshold; if not, calculate the product. For example, if the normalized complexity is 0.7 (original value 7) and the transfer entropy is 0.4 (original value 2), the product is 0.7 × 0.4 = 0.28; if the threshold is 0.8 (corresponding to the original value of 8), the transfer entropy of 0.4 is directly used.

[0088] Calculating the diagonal structure length in recursive quantitative analysis involves calculating the longest sequence length of consecutive recursive points. For example, if a diagonal line contains five consecutive recursive points, its length is 5. When dividing the symbol interval, if the amplitude distribution of the main cycle waveform deviates significantly from a normal distribution (e.g., skewness > 2), the equal interval method is used instead of the standard deviation method. For example, if the amplitude range is 0-10V, it is divided into three equal intervals: 0-3.3V, 3.3-6.6V, and 6.6-10V. The dynamically adjusted symbol interval must cover at least 95% of the amplitude points of the current main cycle waveform; otherwise, the interval boundaries will be further expanded.

[0089] This step S4 generates signal distortion characteristics by constructing a collaborative analysis of phase space trajectory and symbol sequence transfer entropy. The recursive quantization parameters (recursion rate and certainty) of the phase space trajectory can quantify the chaos of the waveform, and the symbol sequence transfer entropy reflects the degree of causal breakage in the timing sequence. The two capture interference characteristics from the perspectives of dynamic systems and information theory, respectively. Compared with the existing technology that only relies on frequency domain energy or fixed thresholds, the dynamic parameters (delay time, embedding dimension) and adaptive symbol interval division are used to solve the problem of insufficient sensitivity of traditional methods under non-stationary signals and burst interference. The fusion analysis of nonlinear characteristics and information entropy breaks through the limitations of traditional linear filtering or single parameter evaluation, significantly improves the distortion detection accuracy under low signal-to-noise ratio conditions, and meets the needs of high-precision dielectric loss monitoring.

[0090] S5. Filtering the voltage signal segment and the current signal segment according to the signal distortion characteristics and the maximum Lyapunov exponent gradients of the voltage signal segment and the current signal segment can be specifically implemented as follows:

[0091] The maximum Lyapunov exponent gradient for the voltage and current signal segments is calculated. The maximum Lyapunov exponent gradient is calculated by the rate of change of the maximum Lyapunov exponent within adjacent time windows. The maximum Lyapunov exponent is calculated by constructing phase space trajectories for each voltage and current signal segment and calculating the maximum Lyapunov exponent using the Wolf algorithm to characterize the signal's sensitivity to initial conditions. The Wolf algorithm's parameter settings include an initial perturbation radius of 1% of the average distance between phase space trajectory points and an evolution step of 1 / 10 of the fundamental period. For example, when the fundamental period is 20 ms, the evolution step is set to 2 ms; when the average distance between trajectory points is 0.5 V, the initial perturbation radius is set to 0.005 V. The specific operation of the gradient calculation is to take the difference between the maximum Lyapunov exponent of the current time window and the exponent value of the previous time window, and divide the result by the time window interval length. For example, if the time window interval is 100ms, the current window index is 0.5 / s, and the previous time window index is 0.3 / s, then the gradient is (0.5-0.3) / 0.1=2.0 / s.

[0092] The filter's cutoff frequency or order is adjusted based on the comparison of the absolute value of the maximum Lyapunov exponent gradient with the preset gradient threshold. The preset gradient threshold is set by calculating the standard deviation σ of the maximum Lyapunov exponent gradient in the historical normal operating data of the power capacitor bank and setting the threshold to 3σ. For example, if the historical standard deviation is 0.5 / s, the threshold is 1.5 / s. The gradient threshold is dynamically updated every 24 hours to adapt to equipment aging or environmental changes. If the absolute value of the gradient exceeds the threshold, the filter cutoff frequency is increased to twice the fundamental frequency to suppress high-frequency chaotic noise; for example, when the fundamental frequency is 50 Hz, the cutoff frequency is set to 100 Hz. If the absolute value of the gradient does not exceed the threshold, the cutoff frequency is maintained at 1.2 times the fundamental frequency (e.g., 60 Hz).

[0093] If the signal distortion exceeds the preset distortion threshold and the absolute value of the maximum Lyapunov exponent gradient exceeds the preset gradient threshold, adaptive mean filtering is applied to the voltage and current signal segments. The filter window size is dynamically set based on the absolute value of the maximum Lyapunov exponent gradient. The adaptive mean filtering window size is set as follows: for every 0.5 / s increase in the absolute value of the gradient, the window size decreases by one sampling point. For example, when the absolute value of the gradient is 2.0 / s, the window size is 3 sampling points; when the gradient is 1.0 / s, the window size is 5 sampling points. The filter window size has a lower limit of 3 sampling points and an upper limit of 9 sampling points to avoid inadequate smoothing due to a too small window or loss of detail due to an too large window. The preset distortion threshold is set based on the 90th percentile of the signal distortion in historical data. For example, if the historical 90th percentile is 0.8, the threshold is 0.8. If the signal distortion exceeds 0.8 and the gradient exceeds 1.5 / s, adaptive filtering is initiated.

[0094] If the signal distortion characteristic does not exceed the preset distortion threshold or the absolute value of the maximum Lyapunov exponent gradient does not exceed the preset gradient threshold, fixed-parameter low-pass filtering is applied. The cutoff frequency of the fixed-parameter low-pass filter is set to 1.2 times the fundamental frequency, and the filter order is set to 4th order. For example, when the fundamental frequency is 50 Hz, the cutoff frequency is 60 Hz, and a 4th-order Butterworth filter is used for filtering. The order is selected based on the fact that the 4th-order Butterworth filter has an attenuation rate of -50 dB / decade at the cutoff frequency, which effectively suppresses high-frequency noise. The low-pass filter coefficients are calculated using the bilinear transformation method to ensure the stability of the digital filter.

[0095] During the calculation of the maximum Lyapunov exponent, if the correlation dimension of the phase space trajectory is less than 2, the signal is considered to be strongly noisy, the gradient calculation is skipped, and fixed parameter filtering is directly applied. The correlation dimension is calculated by calculating the slope of the logarithmic distance distribution of the trajectory points using the Grassberger-Procaccia algorithm. For example, if the slope is less than 1.5, the correlation dimension is considered insufficient. The boundary processing method of the adaptive mean filter is to use the mirror extension method to generate virtual sampling points at the beginning or end of the signal segment. For example, the virtual preceding point of the starting point n = 0 takes the value n = 1, and the virtual following point of the ending point n = N takes the value n = N - 1.

[0096] The dynamic update logic for the gradient threshold involves recollecting historical data and calculating the standard deviation every 24 hours. If the new standard deviation deviates from the original threshold by more than 20%, the threshold is updated. For example, if the original threshold of 1.5 / s corresponds to a standard deviation of 0.5 / s, and the new standard deviation is 0.6 / s, the new threshold is 1.8 / s. The cutoff frequency of the fixed-parameter low-pass filter is dynamically adjusted to 1.2 times the current fundamental frequency when the fundamental frequency fluctuates by more than ±0.5Hz. For example, if the fundamental frequency changes to 51Hz, the cutoff frequency is adjusted to 61.2Hz. The threshold trigger logic for the signal distortion feature is as follows: If the distortion feature exceeds the threshold in three consecutive time windows, it is determined to be continuous interference and the adaptive filtering mode is locked until the distortion feature falls below the threshold.

[0097] Step S5 dynamically adjusts the filtering strategy by collaboratively analyzing the signal distortion characteristics and the maximum Lyapunov exponent gradient. The signal distortion characteristics reflect the degree of waveform distortion, while the maximum Lyapunov exponent gradient quantifies the rate of change of the system's chaotic characteristics. These two characterize the interference intensity from the perspectives of static distortion and dynamic chaos, respectively. Compared to existing technologies that rely solely on a single parameter or fixed filtering mode, this method addresses the lack of sensitivity and high misjudgment rate of traditional methods in scenarios where sudden interference and steady-state noise are mixed, through dual threshold judgment (distortion characteristics and gradient threshold) and dynamic adjustment of the adaptive filtering window. The collaborative decision-making mechanism of nonlinear characteristics and dynamic parameters breaks through the fixed parameter limitations of traditional filters, accurately distinguishing steady-state noise from transient interference, effectively suppressing high-frequency chaotic noise and preserving fundamental waveform details, ensuring the accuracy of subsequent phase difference calculations and meeting the needs of dielectric loss monitoring in complex electromagnetic environments.

[0098] S6. Calculate the phase difference between the voltage signal and the current signal according to the zero-crossing timing difference of the main cycle waveforms of the filtered voltage signal and the current signal to determine the dielectric loss tangent of the power capacitor bank. This can be specifically implemented as follows:

[0099] In the main cycle waveforms of the filtered voltage and current signals, the zero-crossing points of the voltage and current waveforms are located. These zero-crossing points are determined through sign change detection and interpolation correction. The sign change detection method traverses the sampling point sequence of the filtered voltage and current signals. If the sign (positive / negative) of a sampling point differs from the previous point, it is marked as a candidate zero-crossing point. The interpolation correction method selects the amplitude of the previous and current points near the candidate zero-crossing point and calculates the precise zero-crossing point location through linear interpolation. For example, if the voltage signal value at sampling point n is +0.1V and the value at point n+1 is -0.2V, the zero-crossing point location is n + 0.1 / (0.1+0.2) = n+0.33 sampling intervals. The interpolation interval is selected to be one sampling point before and after the candidate point. For example, if the sampling interval is 0.1ms, the interpolation interval is [t-0.1ms, t+0.1ms] to ensure calculation accuracy.

[0100] The time difference between the voltage waveform zero crossing and the current waveform zero crossing is calculated as the zero crossing timing difference. The time difference is calculated by taking the absolute value of the difference between the exact zero crossing timestamp of the voltage waveform and the exact zero crossing timestamp of the current waveform. For example, if the voltage zero crossing occurs at time t_v = 100.5 ms and the current zero crossing occurs at time t_i = 100.8 ms, the time difference is |100.5 - 100.8| = 0.3 ms. If multiple zero crossings exist within the same period, the time difference of the first zero crossing pair is used. The time difference screening logic includes the following: if the time interval between two adjacent zero crossings is less than 10% of the fundamental period, it is determined to be noise interference and is eliminated. For example, when the fundamental period is 20 ms, candidate points with an interval of less than 2 ms are considered invalid.

[0101] The zero-crossing timing difference is converted into a phase difference value based on the fundamental frequency. The phase difference value is the ratio of the time difference to the fundamental period length multiplied by 360 degrees. The fundamental period length is calculated based on the real-time fundamental frequency dynamically tracked in step S2. For example, when the fundamental frequency is 50Hz, the fundamental period is 20ms. If the time difference is 0.3ms, the phase difference value is (0.3 / 20)×360°=5.4°. The calculation of the phase difference value must retain at least two decimal places to improve accuracy. The calculated phase difference value needs to be processed by sliding window mean filtering with a window size of 5 cycles to suppress jumps caused by transient interference. For example, if the phase difference values ​​of 5 consecutive cycles are 5.4°, 5.6°, 5.2°, 5.5°, and 5.3°, the filtered value is (5.4+5.6+5.2+5.5+5.3) / 5=5.4°.

[0102] The dielectric loss tangent is calculated based on the phase difference. The dielectric loss tangent is the tangent function of the phase difference. This tangent function is calculated using a table lookup or numerical approximation. For example, when the phase difference is 5.4°, the dielectric loss tangent is tan(5.4°)≈0.0945. The calculation results are stored in the monitoring system database and trigger real-time alarm logic: if the tangent exceeds the preset alarm threshold (e.g., 0.1), an insulation abnormality alarm is generated. The alarm threshold is set based on the insulation level of the power capacitor bank and historical operating data. For example, the threshold for a capacitor bank rated for 10kV is set to 0.1, and for a 35kV bank it is set to 0.08. After the alarm signal is generated, historical data is automatically retrieved for comparison. If the tangent exceeds the threshold for three consecutive cycles, the alarm is confirmed and the fault type is recorded.

[0103] The candidate zero-crossing point screening logic for sign change detection includes the following: If no valid zero-crossing point is detected within the same cycle, the algorithm traces back to the nearest valid zero-crossing point. For example, if the voltage in the current cycle has no zero-crossing point, the calculation uses the timestamp of the last zero-crossing point in the previous cycle. The linear interpolation coefficient calculation for interpolation correction requires the denominator to be non-zero. If the denominator is zero, the candidate point is directly selected as the zero-crossing point. For example, if two adjacent points have the same amplitude (such as +0.1V and +0.1V), interpolation is skipped and the candidate point is marked as the final zero-crossing point.

[0104] The sliding window mean filter's window size is set to 5 periods, and the window duration is dynamically adjusted based on the real-time fundamental frequency. For example, when the fundamental frequency is 50 Hz, the window duration is 5 × 20 ms = 100 ms. The mean filter's boundary processing method is as follows: when insufficient data is available at the window start or end, virtual data points are supplemented using a mirroring extension method. For example, the virtual preceding point for the window start point n = 0 is set to n = 1, and the virtual following point for the window end point n = N is set to n = N - 1.

[0105] The calculated dielectric loss tangent value requires temperature compensation. The compensation coefficient is set based on the difference between the operating ambient temperature of the power capacitor bank and the standard temperature (25°C). For example, at an ambient temperature of 30°C, the compensation coefficient is 0.98, and the corrected tangent value is the original value × 0.98. The temperature compensation coefficient is obtained by fitting laboratory calibration data and is specifically a quadratic polynomial function. For example, the coefficient = a × ΔT² + b × ΔT + c, where ΔT is the temperature difference.

[0106] The real-time alarm logic's confirmation mechanism includes the following: If no manual confirmation is received within 10 minutes of an alarm being triggered, the backup capacitor bank is automatically activated and the faulty equipment is isolated. The historical data comparison rule only compares historical data under the same operating conditions (e.g., a load factor of 80%-100%), eliminating false positives caused by low loads. For example, if the current load factor is 90%, only historical data from the 85%-95% load factor range is compared.

[0107] Through cross-disciplinary technology integration and dynamic collaborative analysis, this approach breaks through the linear processing framework of traditional dielectric loss monitoring. By combining the phase space trajectory complexity of chaos theory with the symbol transfer entropy of information theory, this approach quantifies signal distortion from the dual dimensions of dynamic system stability and information flow disruption, replacing traditional frequency domain energy or single parameter threshold methods. Based on real-time fundamental frequency dynamic waveform capture, adaptive symbol interval partitioning, and gradient threshold-triggered filtering, this approach constructs an "interference assessment-dynamic filtering-precision calculation" closed loop to address the error accumulation problem of fixed parameters under time-varying interference. Through multi-stage error suppression using symbol change detection, interpolation correction, and sliding window filtering, this approach overcomes noise-induced zero-crossing jitter and improves phase difference calculation accuracy. Through the collaborative design of nonlinear characteristics and dynamic strategies, this approach integrates chaos analysis, entropy quantification, and adaptive filtering across disciplines to form an algorithmic closed-loop optimization system. This replaces traditional rigid solutions based on hardware synchronization or fixed-band filtering, significantly enhancing anti-interference capabilities and computational robustness in complex electromagnetic environments, and enabling high-precision dielectric loss monitoring.

[0108] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to actual conditions.

[0109] It should be noted that the present invention can be deployed on the device itself to implement embedded applications, and can also be run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0110] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0111] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0113] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0114] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0115] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0116] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

[0117] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for online monitoring of dielectric loss of a power capacitor bank, characterized in that: The steps include: S1. Acquire the voltage signal and current signal of the power capacitor bank in real time and record the signal acquisition time series; S2. Dynamically track the fundamental frequency of the voltage signal and the current signal according to the distribution of the zero point and extreme point of the waveform of the voltage signal and the current signal; S3, intercepting a voltage signal segment and a current signal segment of a preset length according to the fundamental frequency, and extracting the main cycle waveform corresponding to the fundamental frequency in the voltage signal segment and the current signal segment; S4. Construct the phase space trajectory of the main cycle waveform and analyze its complexity. At the same time, convert the main cycle waveform into a symbol sequence and calculate the transfer entropy between the symbol sequences. Generate signal distortion characteristics based on the complexity and transfer entropy. S5. Filtering the voltage signal segment and the current signal segment according to the signal distortion characteristics and the maximum Lyapunov exponent gradients of the voltage signal segment and the current signal segment; S6. Calculate the phase difference between the voltage signal and the current signal according to the zero-crossing timing difference of the main cycle waveforms of the filtered voltage signal and the current signal to determine the dielectric loss tangent of the power capacitor bank.

2. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 1, characterized in that: Acquire voltage and current signals of power capacitor banks in real time and record signal acquisition time series, including: The synchronous acquisition of voltage and current signals is started by a synchronous trigger signal, ensuring that the acquisition time deviation of voltage and current signals is less than a preset time threshold; Align the sampling timestamps of the voltage signal and the current signal to the same time base, and generate a voltage signal sequence and a current signal sequence containing the timestamps; Amplitude mutation detection is performed on the voltage signal sequence and the current signal sequence, and abnormal signal segments whose amplitude mutation exceeds the preset amplitude threshold are eliminated.

3. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 1, characterized in that: Dynamically track the fundamental frequency of the voltage and current signals based on the distribution of zero and extreme points of the voltage and current waveforms, including: Extracting the waveform zero points and extreme points of the voltage signal sequence and the current signal sequence within a preset time window; Determine the fundamental wave period of the voltage signal sequence and the current signal sequence according to the time interval between adjacent waveform zero points; The extreme point offset errors of the voltage signal sequence and the current signal sequence are corrected by the cubic spline interpolation method to obtain the corrected fundamental wave period; The corrected inverse of the fundamental wave period is taken as the fundamental wave frequency.

4. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 1, characterized in that: According to the fundamental frequency, a voltage signal segment and a current signal segment of a preset length are intercepted, and a main cycle waveform corresponding to the fundamental frequency in the voltage signal segment and the current signal segment is extracted, including: Determine the preset time length as an integer multiple of the fundamental wave period according to the fundamental wave frequency, and intercept the voltage signal segment and current signal segment of the corresponding time length from the voltage signal sequence and the current signal sequence; In the intercepted voltage signal segment and current signal segment, the starting zero-crossing point of the main cycle waveform is located using the cycle length corresponding to the fundamental frequency as a unit; Starting from the initial zero-crossing point, the voltage signal segment and the current signal segment within a complete fundamental wave cycle are extracted as the main cycle waveform; The extracted main cycle waveform is processed by mean filtering to eliminate random noise interference.

5. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 1, characterized in that: Construct the phase space trajectory of the main cycle waveform and analyze its complexity. At the same time, convert the main cycle waveform into a symbol sequence and calculate the transfer entropy between symbol sequences. Generate signal distortion features based on complexity and transfer entropy, including: The phase space trajectory of the main periodic waveform is constructed using the time delay embedding method; Calculate the recursive quantitative analysis parameters of the phase space trajectory, the recursive quantitative analysis parameters include recursion rate and certainty, and generate the phase space complexity index according to the recursion rate and certainty; When converting the main cycle waveform into a symbol sequence, the symbol intervals are divided according to the amplitude distribution of the main cycle waveform, and each symbol interval corresponds to a preset amplitude range; The transition probabilities of adjacent symbols in a statistical symbol sequence are calculated, and the transition entropy between symbol sequences is calculated based on the transition probabilities; The signal distortion feature is generated according to the product of the normalized phase space complexity index and the transfer entropy, or when the phase space complexity index exceeds a preset threshold set based on historical data statistics, the transfer entropy is directly used as the signal distortion feature.

6. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 5, characterized in that: The delay time of the time delay embedding method is set according to the sampling rate of the main cycle waveform, and the embedding dimension is set according to the spectral characteristics of the main cycle waveform.

7. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 1, characterized in that: According to the signal distortion characteristics and the maximum Lyapunov exponent gradient of the voltage signal segment and the current signal segment, the voltage signal segment and the current signal segment are filtered, including: Calculate the maximum Lyapunov exponent gradient of the voltage signal segment and the current signal segment. The maximum Lyapunov exponent gradient is calculated by the rate of change of the maximum Lyapunov exponent in adjacent time windows. Adjusting the cutoff frequency or order of the filter based on the comparison result of the absolute value of the maximum Lyapunov exponent gradient and the preset gradient threshold; If the signal distortion characteristic exceeds the preset distortion threshold and the absolute value of the maximum Lyapunov exponent gradient exceeds the preset gradient threshold, the voltage signal segment and the current signal segment are filtered using adaptive mean filtering; If the signal distortion characteristic does not exceed the preset distortion threshold or the absolute value of the maximum Lyapunov exponent gradient does not exceed the preset gradient threshold, a low-pass filter with fixed parameters is used.

8. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 7, characterized in that: The filter window size is set dynamically according to the absolute value of the maximum Lyapunov exponent gradient.

9. The method for online monitoring of dielectric loss of a power capacitor bank according to claim 1, characterized in that: According to the zero-crossing timing difference of the main cycle waveforms of the filtered voltage signal and current signal, the phase difference between the voltage signal and the current signal is calculated to determine the dielectric loss tangent value of the power capacitor bank, including: In the main cycle waveforms of the filtered voltage and current signals, the zero crossing points of the voltage and current waveforms are located respectively, and the zero crossing points are determined by sign change detection and interpolation correction; Calculate the time difference between the zero-crossing point of the voltage waveform and the zero-crossing point of the current waveform as the zero-crossing timing difference; Convert the zero-crossing timing difference into a phase difference value according to the fundamental frequency; The dielectric loss tangent value is calculated based on the phase difference value, and the dielectric loss tangent value is the tangent function value of the phase difference value.

Images