Fault monitoring system and method for reactive compensation capacitor

Abstract

The invention discloses a fault monitoring system and method for a reactive compensation capacitor, and the system comprises an order sub-system, a first offset sequence construction sub-system, a second offset sequence construction sub-system, an attention obtaining sub-system, and a classification sub-system. Dividing the plurality of obtained IMF signals into a low order, a middle order and a high order, and obtaining the detection dominant frequency of each order; frequency deviation is obtained through the difference value between the prevailing frequency and the stored prevailing frequency, and a frequency deviation sequence is constructed; calculating the nearest distance between the fundamental frequency and 50Hz in each IMF frequency spectrum, determining the fundamental wave energy ratio offset and constructing a sequence; and then attention sequences of the two sequences are obtained respectively, and finally, a double-path attention fusion BP network is utilized to weight the features by taking the attention sequences as channel attention weights so as to obtain a fault value. According to the method, the capacitor fault monitoring precision is effectively improved, and reliable guarantee is provided for stable operation of the reactive compensation capacitor.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of capacitor detection, in particular to a fault monitoring system and method for a reactive compensation capacitor. BACKGROUND

[0002] In the stable operation of power systems, especially intelligent power distribution systems, power electronic components play a crucial role. For example, reactive compensation capacitors can effectively improve the power factor of the power grid, reduce line loss, and improve voltage quality. However, due to long-term exposure to complex power grid environments, factors such as voltage fluctuations, harmonic interference, and temperature changes can cause the reactive compensation capacitor to easily develop faults such as insulation aging, partial discharge, and capacitor value attenuation. If not promptly monitored and addressed, these faults can lead to equipment damage and even power grid paralysis.

[0003] In the prior art, the most common method is the traditional Fourier transform-based reactive compensation capacitor monitoring technology. This technology converts time-domain signals to frequency-domain by performing Fourier transform on the voltage and current signals of the capacitor's phase, thereby analyzing the frequency components and amplitude information of the signals to determine whether the capacitor has a fault. For example, when a capacitor fails, the harmonic components of its voltage or current signals will change, and traditional Fourier transform can extract these harmonic features to achieve fault monitoring.

[0004] In actual power systems, the voltage signal of the reactive compensation capacitor is often non-stationary, which may be caused by sudden changes in load, transient processes of power grid faults, etc. Traditional Fourier transform is based on the assumption that the signal is stationary, and the frequency spectrum obtained is the average characteristic of the signal in the entire time interval. It cannot accurately reflect the frequency changes of non-stationary signals at different times, and cannot accurately obtain the dominant frequency shift of different order signals such as low-order, intermediate-order, and high-order signals, as well as the energy proportion shift related to the 50Hz fundamental wave. This makes it difficult for this technology to accurately extract fault features when faced with non-stationary signals, leading to misjudgment or missed judgment of capacitor faults, and making it difficult to meet the demand for high-precision monitoring.

[0005] Therefore, the present application is proposed. SUMMARY

[0006] The technical problem to be solved by the present application is that the existing capacitor fault monitoring has insufficient accuracy in extracting fault features when faced with non-stationary signals, leading to misjudgment or missed judgment of faults, and making it difficult to meet the demand for high-precision monitoring. The present application aims to provide a fault monitoring system and method for a reactive compensation capacitor, solving the problem of low accuracy in existing capacitor fault monitoring.

[0007] The present application is achieved by the following technical solutions:

[0008] In a first aspect, the present application provides a fault monitoring system for a reactive compensation capacitor, comprising:

[0009] a hierarchical subsystem for EMD decomposing each voltage sub-signal of the voltage signal of the reactive compensation capacitor to obtain a plurality of IMF spectrums; and dividing the plurality of IMF spectrums into low-order, intermediate-order and high-order to obtain detection dominant frequencies of the low-order, intermediate-order and high-order, respectively;

[0010] a first offset sequence construction subsystem for obtaining frequency offsets according to differences between the detection dominant frequencies of the low-order, intermediate-order and high-order and corresponding stored dominant frequencies, respectively; and constructing a frequency offset sequence according to the plurality of frequency offsets;

[0011] a second offset sequence construction subsystem for calculating distances between a fundamental frequency and a fundamental wave in each IMF spectrum, finding an IMF spectrum corresponding to a minimum distance, and calculating a fundamental wave energy proportion offset to construct a fundamental wave energy proportion offset sequence;

[0012] an attention degree acquisition subsystem for respectively acquiring corresponding attention degree sequences of the frequency offset sequence and the fundamental wave energy proportion offset sequence;

[0013] a classification subsystem for processing the frequency offset sequence and the fundamental wave energy proportion offset sequence by using a dual-path attention fusion BP network, taking the corresponding attention degree sequences as channel attention weights, weighting features of different channels, and obtaining a fault value of the reactive compensation capacitor.

[0014] Further, the hierarchical subsystem comprises:

[0015] a voltage signal division unit for dividing the voltage signal of the reactive compensation capacitor to obtain a plurality of voltage sub-signals; and an EMD decomposition unit for EMD decomposing each voltage sub-signal to obtain a plurality of IMF signals;

[0016] an IMF signal division unit for dividing the plurality of IMF signals into low-order, intermediate-order and high-order;

[0017] a Fourier transform unit for Fourier transforming each IMF signal to obtain each IMF spectrum;

[0018] a detection dominant frequency acquisition unit for extracting a fundamental frequency from each IMF spectrum and acquiring detection dominant frequencies of the low-order, intermediate-order and high-order according to the fundamental frequency.

[0019] Further, the detection dominant frequency acquisition unit comprises:

[0020] a fundamental frequency extraction sub-unit for obtaining a maximum amplitude in each IMF spectrum and taking a frequency corresponding to the maximum amplitude as the fundamental frequency.

[0021] The first detection dominant frequency obtaining subunit is configured to obtain the detection dominant frequency of the low order according to the fundamental frequency of each IMF spectrum in the low order;

[0022] The second detection dominant frequency obtaining subunit is configured to obtain the detection dominant frequency of the intermediate order according to the fundamental frequency of each IMF spectrum in the intermediate order;

[0023] The third detection dominant frequency obtaining subunit is configured to obtain the detection dominant frequency of the high order according to the fundamental frequency of each IMF spectrum in the high order.

[0024] Further, the detection dominant frequency obtaining process is as follows:

[0025] In the same order, the fundamental frequency and the amplitude of the same IMF spectrum are multiplied to obtain a fundamental frequency-amplitude product;

[0026] The fundamental frequency-amplitude products in the same order are added to obtain a fundamental frequency-amplitude sum;

[0027] The fundamental frequency-amplitude sums in the same order are added to obtain an amplitude total sum;

[0028] The ratio of the fundamental frequency-amplitude sum to the amplitude total sum is taken as the detection dominant frequency.

[0029] Further, the first offset sequence construction subsystem comprises:

[0030] The first frequency shift coefficient unit is configured to subtract the detection dominant frequency of the low order from the stored dominant frequency of the low order to obtain a first subtraction result; take the absolute value of the first subtraction result to obtain a first dominant frequency gap; and take the ratio of the first dominant frequency gap to the stored dominant frequency of the low order as the first frequency shift coefficient;

[0031] The second frequency shift coefficient unit is configured to subtract the detection dominant frequency of the intermediate order from the stored dominant frequency of the intermediate order to obtain a second subtraction result; take the absolute value of the second subtraction result to obtain a second dominant frequency gap; and take the ratio of the second dominant frequency gap to the stored dominant frequency of the intermediate order as the second frequency shift coefficient;

[0032] The third frequency shift coefficient unit is configured to subtract the detection dominant frequency of the high order from the stored dominant frequency of the high order to obtain a third subtraction result; take the absolute value of the third subtraction result to obtain a third dominant frequency gap; and take the ratio of the third dominant frequency gap to the stored dominant frequency of the high order as the third frequency shift coefficient;

[0033] The frequency offset calculation unit is configured to add the first frequency shift coefficient, the second frequency shift coefficient and the third frequency shift coefficient to obtain a frequency offset;

[0034] A frequency offset sequence unit is configured to arrange a plurality of frequency offsets in a time sequence to obtain a frequency offset sequence.

[0035] Further, the second offset sequence construction subsystem comprises:

[0036] An IMF spectrum order calculation unit is configured to calculate a distance between a fundamental frequency and a 50Hz fundamental wave in each IMF spectrum, and select an IMF spectrum order corresponding to a minimum distance;

[0037] A fundamental wave energy proportion offset calculation unit is configured to calculate a fundamental wave energy proportion offset of the 50Hz fundamental wave according to the IMF spectrum in the IMF spectrum order;

[0038] A second offset sequence construction unit is configured to arrange a plurality of fundamental wave energy proportion offsets in a time sequence to obtain a fundamental wave energy proportion offset sequence.

[0039] Further, a formula for calculating the fundamental wave energy proportion offset of the 50Hz fundamental wave is:

[0040]

[0041] wherein γ is the fundamental wave energy proportion offset of the 50Hz fundamental wave, y k is a kth amplitude in the IMF spectrum order, f k is a frequency value corresponding to the kth amplitude in the IMF spectrum order, K is a quantity of amplitudes corresponding to all IMF spectra in the IMF spectrum order, f min is a fundamental frequency corresponding to the minimum distance, y min is an amplitude of the fundamental frequency corresponding to the minimum distance, and k is a number of the amplitude.

[0042] Further, the attention acquisition subsystem comprises:

[0043] A segmentation unit is configured to respectively segment the frequency offset sequence and the fundamental wave energy proportion offset sequence according to a fixed length M, wherein M is a positive integer greater than 1;

[0044] A first attention unit is configured to count a quantity of frequency offsets greater than a frequency offset threshold in each segment of the frequency offset sequence, and perform normalization processing to obtain an attention of each segment in the frequency offset sequence;

[0045] A second attention unit is configured to count a quantity of fundamental wave energy proportion offsets greater than a fundamental wave energy proportion offset threshold in each segment of the fundamental wave energy proportion offset sequence, and perform normalization processing to obtain an attention of each segment in the fundamental wave energy proportion offset sequence;

[0046] A first attention sequence unit is configured to form an attention sequence of the frequency offset sequence by using the attentions corresponding to the frequency offset sequence;

[0047] The second attention degree sequence unit is configured to form an attention degree sequence of the fundamental wave energy proportion offset sequence with each attention degree corresponding to the fundamental wave energy proportion offset sequence.

[0048] Further, the dual-path attention fusion BP network comprises a first sequence processing unit, a second sequence processing unit and a full connection layer.

[0049] The first sequence processing unit and the second sequence processing unit each comprise an input layer, a plurality of first hidden layers, a splicing layer, a multiplier and a second hidden layer.

[0050] The input layer is configured to input a sequence into the plurality of first hidden layers.

[0051] Each first hidden layer is configured to extract a feature value for each segment in the frequency offset sequence or the fundamental wave energy proportion offset sequence.

[0052] The splicing layer is configured to splice the feature values output by the plurality of first hidden layers to obtain a feature sequence.

[0053] The multiplier is configured to multiply the feature sequence and a corresponding attention degree sequence element by element to obtain an enhanced feature sequence.

[0054] Each second hidden layer is configured to output the enhanced feature sequence.

[0055] The full connection layer is configured to obtain a reactive compensation capacitor fault value according to the outputs of the second hidden layers in the first sequence processing unit and the second sequence processing unit.

[0056] In a second aspect, the present application further provides a method for monitoring the fault of a reactive compensation capacitor, which comprises:

[0057] Performing EMD decomposition on each voltage sub-signal of the voltage signal of the reactive compensation capacitor to obtain a plurality of IMF frequency spectrums; performing low-order, intermediate-order and high-order division on the plurality of IMF frequency spectrums to respectively obtain detection dominant frequencies of the low-order, intermediate-order and high-order;

[0058] Obtaining a frequency offset according to the difference between the detection dominant frequencies of the low-order, intermediate-order and high-order and the corresponding stored dominant frequencies; and constructing a frequency offset sequence according to the plurality of frequency offsets;

[0059] Calculating the distance between the fundamental frequency and the fundamental wave in each IMF frequency spectrum, finding the order of the IMF frequency spectrum corresponding to the minimum distance, calculating the fundamental wave energy proportion offset, and constructing a fundamental wave energy proportion offset sequence;

[0060] Respectively obtaining a corresponding attention degree sequence for the frequency offset sequence and the fundamental wave energy proportion offset sequence;

[0061] The frequency offset sequence and the fundamental energy proportion offset sequence are processed by adopting a two-way attention fusion BP network, corresponding attention degree sequences are taken as channel attention weights, features of different channels are weighted, and the fault value of the reactive compensation capacitor is obtained.

[0062] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0063] 1. The fault monitoring system and method of the reactive compensation capacitor, first, the capacitor voltage signal is subjected to EMD decomposition by section, the obtained multiple IMF signals are divided into low-order, intermediate-order and high-order, and the dominant frequencies of each order are obtained; the frequency offset is obtained through the difference between the dominant frequencies and the stored dominant frequencies, and the frequency offset sequence is constructed; at the same time, the nearest distance between the fundamental frequency and 50Hz in the spectrum of each IMF is calculated to determine the fundamental energy proportion offset and construct the sequence; then the attention degree sequences of the two sequences are obtained respectively, and finally the two-way attention fusion BP network is used to weight the features by taking the attention degree sequences as the channel attention weights, and the fault value is obtained.

[0064] 2. The fault monitoring system and method of the reactive compensation capacitor, the present application adopts EMD decomposition technology, and the voltage signal is decomposed into multiple IMF signals by section, and divided into low-order, intermediate-order and high-order, which can accurately capture the dynamic change characteristics of signals of different orders. Compared with the "averaging" processing of non-stationary signals by the traditional Fourier transform, this method can extract the detection dominant frequencies of each order, and solves the problem of inaccurate feature extraction of non-stationary signals by the traditional technology.

[0065] 3. The fault monitoring system and method of the reactive compensation capacitor, the present application reflects the deviation of the dominant frequencies of signals of different orders from the normal state through the frequency offset sequence, can sensitively capture the frequency characteristic abnormality caused by the fault, focuses on the change of the 50Hz fundamental wave energy through the fundamental energy proportion offset sequence, directly relates to the stability of the power grid fundamental wave, can effectively identify the fluctuation of the fundamental energy proportion caused by the capacitor fault, generates the attention degree sequence through the attention degree acquisition subsystem, and integrates it into the two-way attention fusion BP network as the channel attention weight, can automatically highlight the more important features for fault judgment, suppress irrelevant or interference information, make the network more focused on the key fault signal when processing the features, and improve the capacitor fault monitoring precision. BRIEF DESCRIPTION OF DRAWINGS

[0066] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0067] Figure 1A structure block diagram of a fault monitoring system of a reactive compensation capacitor according to the present application;

[0068] Figure 2 A normal voltage signal of the reactive compensation capacitor according to the present application;

[0069] Figure 3 A fault voltage signal of the reactive compensation capacitor according to the present application;

[0070] Figure 4 A structure schematic diagram of the dual-path attention fusion BP network according to the present application;

[0071] Figure 5 A structure schematic diagram of the first hidden layer according to the present application;

[0072] Figure 6 A connection schematic diagram of the second hidden layer and the multiplier according to the present application;

[0073] Figure 7 A corresponding condition table of the fault value and the fault type of the reactive compensation capacitor according to the present application;

[0074] Figure 8 A flow chart of the fault monitoring method of the reactive compensation capacitor according to the present application. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below with reference to the embodiments and the accompanying drawings, the illustrative embodiments and the description thereof are only used to explain the present application, and do not limit the present application.

[0076] Embodiment 1

[0077] As shown in the drawings, the fault monitoring system of the reactive compensation capacitor according to the present application comprises: Figure 1 The hierarchical subsystem is used for performing EMD decomposition on each voltage sub-signal of the voltage signal of the reactive compensation capacitor to obtain a plurality of IMF frequency spectrums; and performing low-order, intermediate-order and high-order division on the plurality of IMF frequency spectrums to respectively obtain the detection dominant frequency of the low-order, the detection dominant frequency of the intermediate-order and the detection dominant frequency of the high-order, i.e. to obtain the detection dominant frequency of the low-order, the detection dominant frequency of the intermediate-order and the detection dominant frequency of the high-order;

[0078] The first offset sequence construction subsystem is used for obtaining the frequency offset according to the difference between the detection dominant frequency of the low-order, the detection dominant frequency of the intermediate-order and the detection dominant frequency of the high-order and the corresponding stored dominant frequency; and constructing the frequency offset sequence according to the plurality of frequency offsets;

[0079]

[0080] ​A second offset sequence construction subsystem for calculating the distance between the fundamental frequency and the 50Hz fundamental wave in each IMF spectrum, finding the order of the IMF spectrum where the minimum distance is located, calculating the 50Hz fundamental wave energy proportion offset, and constructing a fundamental wave energy proportion offset sequence;

[0081] An attention degree acquisition subsystem for respectively acquiring corresponding attention degree sequences for the frequency offset sequence and the fundamental wave energy proportion offset sequence;

[0082] A classification subsystem for processing the frequency offset sequence and the fundamental wave energy proportion offset sequence using a dual-path attention fusion BP network, taking the corresponding attention degree sequences as channel attention weights, weighting the features of different channels, and obtaining the reactive compensation capacitor fault value.

[0083] In specific implementation, the common connection modes of the reactive compensation capacitor in the three-phase power grid mainly include star connection (Y type) and delta connection (Δ type). The star connection is to connect one end of three capacitors together to form a common neutral point, and the other end is connected to the A, B and C phase lines of the three-phase power grid. The delta connection is to connect three capacitors end to end to form a closed delta loop, and the three connection points are connected to the A, B and C phase lines of the three-phase power grid.

[0084] The technical solution of the present application is to measure the fault condition of any reactive compensation capacitor in the three-phase power grid.

[0085] As shown in Figure 2 , when the reactive compensation capacitor is normal, the voltage changes periodically according to the sine law with time, and the waveform is smooth and regular. As shown in Figure 3 , when the reactive compensation capacitor is faulty, there is serious waveform distortion, deviating from the sinusoidal waveform, and multiple sharp peaks, depressions and irregular fluctuations, Figure 2 and Figure 3 A is the amplitude.

[0086] Specifically, the order division subsystem includes:

[0087] A voltage signal division unit for dividing the voltage signal of the reactive compensation capacitor to obtain a plurality of voltage sub-signals; and an EMD decomposition unit for EMD decomposing each voltage sub-signal to obtain a plurality of IMF signals.

[0088] An IMF signal division unit for dividing the plurality of IMF signals into low-order, intermediate-order and high-order;

[0089] A Fourier transform unit for performing Fourier transform on each IMF signal to obtain an IMF frequency spectrum;

[0090] The detection dominant frequency acquisition unit is configured to extract a fundamental frequency from each IMF spectrum and acquire a low-order, intermediate-order and high-order detection dominant frequency according to the fundamental frequency.

[0091] It should be noted that one IMF signal corresponds to one IMF spectrum.

[0092] In this embodiment, the period length of the normal voltage signal is 20 ms, and the voltage signal is divided into 10 periods in the application, and the length of the obtained voltage sub-signal is 200 ms. The length of the voltage signal is L, and a total of L / 200 ms voltage sub-signals are obtained. The signal division method described in this embodiment is not limited, and the division length can be selected according to the actual situation.

[0093] In this embodiment, each voltage sub-signal is subjected to multiple EMD decompositions, the IMF signals obtained by the multiple EMD decompositions are arranged in the order of acquisition, the first IMF signal and the second IMF signal are divided into the high order, the third IMF signal, the fourth IMF signal and the fifth IMF signal are divided into the intermediate order, and the remaining IMF signals are classified into the low order. The division method of the application is not limited, but the IMF signal with a frequency close to 50 Hz needs to be divided into the intermediate order.

[0094] The high order corresponds to the high frequency details of the signal (such as partial discharge pulses and high frequency harmonics), the intermediate order IMF corresponds to the fundamental dominant component of the signal (50 Hz and adjacent low order harmonics), and the low order IMF corresponds to the low frequency trend of the signal (such as voltage long-term drift and load slow change).

[0095] The application divides the multiple IMF signals corresponding to each voltage sub-signal into low order, intermediate order and high order, extracts the detection dominant frequency of each order by combining the Fourier transform, and facilitates the evaluation of the frequency deviation in the three aspects of low order, intermediate order and high order.

[0096] Specifically, the detection dominant frequency acquisition unit comprises:

[0097] The fundamental frequency extraction sub-unit is configured to acquire the maximum amplitude in each IMF spectrum and take the frequency corresponding to the maximum amplitude as the fundamental frequency.

[0098] The first detection dominant frequency acquisition sub-unit is configured to acquire the detection dominant frequency of the low order according to the fundamental frequencies of the IMF spectra in the low order.

[0099] The second detection dominant frequency acquisition sub-unit is configured to acquire the detection dominant frequency of the intermediate order according to the fundamental frequencies of the IMF spectra in the intermediate order.

[0100] The third detection dominant frequency acquisition sub-unit is configured to acquire the detection dominant frequency of the high order according to the fundamental frequencies of the IMF spectra in the high order.

[0101] In this embodiment, the acquisition process of the detection dominant frequency is as follows:

[0102] In the same order, the fundamental frequency and amplitude of the same IMF spectrum are multiplied to obtain the fundamental-amplitude product;

[0103] The fundamental-amplitude products in the same order are added to obtain the fundamental-amplitude sum;

[0104] The fundamental-amplitude sums in the same order are added to obtain the amplitude total sum;

[0105] The ratio of the fundamental-amplitude sum to the amplitude total sum is taken as the detection dominant frequency.

[0106] The formula for detecting the dominant frequency is: Wherein, f c For detecting the dominant frequency, f o,i is the fundamental frequency of the i-th IMF spectrum in the low order, the intermediate order or the high order, y i is the amplitude of the fundamental frequency of the i-th IMF spectrum in the low order, the intermediate order or the high order, I is the number of IMF spectra in the low order, the intermediate order or the high order, and i is a positive integer.

[0107] The present application extracts the maximum amplitude corresponding to the fundamental frequency of each IMF spectrum, and then calculates the dominant frequency according to the fundamental frequency and amplitude in the same order, covering the full frequency band from high frequency noise (partial discharge) to fundamental wave (50Hz) to low frequency trend (voltage drift), which can evaluate the dominant frequency deviation of the reactive compensation capacitor from three aspects.

[0108] Specifically, the first deviation sequence construction subsystem comprises:

[0109] The first frequency shift coefficient unit is configured to subtract the detection dominant frequency of the low order from the stored dominant frequency of the low order to obtain a first subtraction result, take the absolute value of the first subtraction result to obtain a first dominant frequency difference, and take the ratio of the first dominant frequency difference to the stored dominant frequency of the low order as the first frequency shift coefficient;

[0110] The second frequency shift coefficient unit is configured to subtract the detection dominant frequency of the intermediate order from the stored dominant frequency of the intermediate order to obtain a second subtraction result, take the absolute value of the second subtraction result to obtain a second dominant frequency difference, and take the ratio of the second dominant frequency difference to the stored dominant frequency of the intermediate order as the second frequency shift coefficient;

[0111] The third frequency shift coefficient unit is configured to subtract the detection dominant frequency of the high order from the stored dominant frequency of the high order to obtain a third subtraction result, take the absolute value of the third subtraction result to obtain a third dominant frequency difference, and take the ratio of the third dominant frequency difference to the stored dominant frequency of the high order as the third frequency shift coefficient;

[0112] a frequency offset calculation unit, configured to add the first frequency shift coefficient, the second frequency shift coefficient and the third frequency shift coefficient to obtain a frequency offset;

[0113] a frequency offset sequence unit, configured to arrange the plurality of frequency offsets in a time sequence to obtain a frequency offset sequence.

[0114] It should be noted that each voltage sub-signal corresponds to a frequency offset, and therefore there are multiple frequency offsets.

[0115] The present application calculates the frequency shift coefficient for low-order, middle-order and high-order frequency differences respectively, and can simultaneously cover the frequency intervals corresponding to different working states of the equipment. For example, low-order is related to low-frequency trend (such as long-term voltage drift caused by equipment aging), and high-order corresponds to high-frequency details (such as partial discharge pulse).

[0116] In the present application, the low-order storage dominant frequency, the middle-order storage dominant frequency and the high-order storage dominant frequency are three-order dominant frequencies obtained based on a normal voltage signal of a normal reactive power compensation capacitor.

[0117] Specifically, the second offset sequence construction subsystem includes:

[0118] an IMF spectrum order calculation unit, configured to calculate the distance (the distance is the absolute value of the difference) between the fundamental frequency and the 50Hz fundamental wave in each IMF spectrum, and select the IMF spectrum order corresponding to the minimum distance;

[0119] a fundamental wave energy proportion offset calculation unit, configured to calculate the fundamental wave energy proportion offset of the 50Hz fundamental wave according to the IMF spectrum in the IMF spectrum order;

[0120] a second offset sequence construction unit, configured to arrange the plurality of fundamental wave energy proportion offsets in a time sequence to obtain a fundamental wave energy proportion offset sequence.

[0121] It should be noted that each voltage sub-signal corresponds to a fundamental wave energy proportion offset, and therefore there are multiple fundamental wave energy proportion offsets.

[0122] In the present embodiment, the formula for calculating the fundamental wave energy proportion offset of the 50Hz fundamental wave is:

[0123]

[0124] wherein γ is the fundamental wave energy proportion offset of the 50Hz fundamental wave, y k is the kth amplitude in the IMF spectrum order, f k is the frequency value corresponding to the kth amplitude in the IMF spectrum order, and K is the number of amplitudes corresponding to all IMF spectra in the IMF spectrum order, f miny is the fundamental frequency corresponding to the minimum distance min k is the number of the amplitude value of the fundamental frequency corresponding to the minimum distance.

[0125] In this embodiment, the IMF signal close to 50Hz is divided into the middle order, and the application calculates the amplitude and frequency of the IMF spectrum of all IMF signals in the middle order to obtain the total energy, and then obtains y according to the fundamental frequency closest to 50Hz and the amplitude corresponding to the frequency. min ·f min The capacitor core working condition is evaluated by comparing the "energy contribution of the nearest 50Hz fundamental wave" with the "total energy contribution of the order", and the larger the fundamental wave energy proportion offset is, the worse the capacitor working condition is.

[0126] In this embodiment, the formula for calculating the 50Hz fundamental wave energy proportion offset can also be: Only the amplitude energy of the nearest 50Hz fundamental wave is compared with all the energy in the same order to highlight the energy proportion of the nearest 50Hz fundamental wave.

[0127] Specifically, the attention obtaining subsystem includes:

[0128] The segmentation unit is configured to segment the frequency offset sequence and the fundamental wave energy proportion offset sequence respectively according to a fixed length M, where M is a positive integer greater than 1.

[0129] The first attention unit is configured to count the number of frequency offsets greater than a frequency offset threshold in each segment of the frequency offset sequence, and normalize the number to obtain the attention of each segment in the frequency offset sequence.

[0130] The second attention unit is configured to count the number of fundamental wave energy proportion offsets greater than a fundamental wave energy proportion offset threshold in each segment of the fundamental wave energy proportion offset sequence, and normalize the number to obtain the attention of each segment in the fundamental wave energy proportion offset sequence.

[0131] The first attention sequence unit is configured to construct an attention sequence of the frequency offset sequence from the attentions corresponding to the frequency offset sequence.

[0132] The second attention sequence unit is configured to construct an attention sequence of the fundamental wave energy proportion offset sequence from the attentions corresponding to the fundamental wave energy proportion offset sequence.

[0133] In this embodiment, M is set to 10.

[0134] The attention of each segment in the frequency offset sequence is the ratio of the number of frequency offsets greater than the frequency offset threshold to M.

[0135] The attention degree of each segment in the fundamental energy proportion offset sequence is: the ratio of the number of the fundamental energy proportion offset greater than the fundamental energy proportion offset threshold value to M.

[0136] The application first segments the frequency offset sequence and the fundamental energy proportion offset sequence with a fixed length M, and then respectively counts the number of abnormalities exceeding the threshold value in each segment, and converts the number of abnormalities into an attention degree by normalization, so that the segment with frequent faults and obvious characteristics obtains a higher weight.

[0137] In the embodiment, the frequency offset threshold value and the fundamental energy proportion offset threshold value are set according to experiments or experience.

[0138] Specifically, as shown in the figure, Figure 4 The double-channel attention fusion BP network includes a first channel sequence processing unit, a second channel sequence processing unit and a full connection layer.

[0139] The first channel sequence processing unit and the second channel sequence processing unit each include an input layer, a plurality of first hidden layers, a splicing layer, a multiplier and a second hidden layer.

[0140] The input layer is configured to input the sequence into the plurality of first hidden layers.

[0141] Each first hidden layer is configured to extract a feature value from each segment in the frequency offset sequence or the fundamental energy proportion offset sequence.

[0142] The splicing layer is configured to splice the feature values output by the plurality of first hidden layers to obtain a feature sequence.

[0143] The multiplier is configured to multiply the feature sequence and a corresponding attention degree sequence element by element to obtain an enhanced feature sequence.

[0144] Each second hidden layer is configured to output the enhanced feature sequence.

[0145] The full connection layer is configured to obtain the reactive compensation capacitor fault value according to the outputs of the second hidden layers in the first channel sequence processing unit and the second channel sequence processing unit.

[0146] The application independently processes the frequency offset sequence and the fundamental energy proportion offset sequence through the first channel sequence processing unit and the second channel sequence processing unit, cooperatively extracts multi-dimensional local features through the input layer and the plurality of first hidden layers, integrates the feature dimensions through the splicing layer, dynamically weights and enhances the features of the sensitive fault segment through the multiplier by introducing the attention degree sequence, focuses on the key fault information through the model, and finally fuses the double-channel enhanced features through the full connection layer to obtain the reactive compensation capacitor fault value.

[0147] The application inputs each 10 elements (consistent with M) of each sequence into a first hidden layer through the input layer to obtain a feature value, as shown in the figure. Figure 5The activation function of the first hidden layer is Sigmoid, and the number of nodes of the first hidden layer is 1.

[0148] As shown in the formula (1), the number of nodes of the second hidden layer is the same as the number of the multiplier output features, and the activation function of the second hidden layer is Sigmoid, Figure 6 Figure 6 Wherein, n is a positive integer, and N is the total number of nodes.

[0149] The length of the sequence obtained by splicing the layers is the same as that of the attention sequence.

[0150] When the dual-path attention fusion BP network of the application is used, training needs to be performed first, and the training samples are: the frequency offset sequence, the fundamental energy proportion offset sequence, the attention sequence of the frequency offset sequence and the attention sequence of the fundamental energy proportion offset sequence, and the training method adopts the existing gradient descent method for training.

[0151] In this embodiment, as shown in the formula (1), Figure 7 Figure 7 The fault value of the reactive power compensation capacitor and the corresponding fault type are shown in the formula (1), and the larger the value is, the more serious the fault is.

[0152] The application has the following advantages:

[0153] (1) The EMD decomposition technology is adopted in the application, the voltage signal is decomposed into multiple IMF signals according to segments, and is divided into low-order, intermediate-order and high-order, so that the dynamic change characteristics of different order signals can be accurately captured. Compared with the "averaging" processing of the traditional Fourier transform on the non-stationary signal, this method can extract the detection dominant frequency of each order, and solves the problem of inaccurate feature extraction of the traditional technology on the non-stationary signal.

[0154] (2) The frequency offset sequence reflects the deviation of the dominant frequency of different order signals from the normal state, and can sensitively capture the frequency characteristic abnormality caused by the fault. The fundamental energy proportion offset sequence focuses on the change of the 50Hz fundamental energy, directly relates to the stability of the power grid, can effectively identify the fluctuation of the fundamental energy proportion caused by the capacitor fault, generates the attention sequence through the attention obtaining subsystem, and integrates the attention sequence as the channel attention weight into the dual-path attention fusion BP network, can automatically highlight the more important features for fault judgment, suppress irrelevant or interference information, make the network more focused on the key fault signal when processing the features, and improve the capacitor fault monitoring precision.

[0155] Embodiment 2

[0156] As shown in the formula (1), the number of nodes of the second hidden layer is the same as the number of the multiplier output features, and the activation function of the second hidden layer is Sigmoid, Figure 8 ​​As shown, the difference between the embodiment and embodiment 1 is that the embodiment provides a fault monitoring method of a reactive compensation capacitor, which corresponds to the fault monitoring system of a reactive compensation capacitor of embodiment 1; the method comprises:

[0157] Step 1, performing EMD decomposition on each voltage sub-signal of the voltage signal of the reactive compensation capacitor to obtain a plurality of IMF frequency spectrums; performing low-order, middle-order and high-order division on the plurality of IMF frequency spectrums to respectively obtain detection dominant frequencies of the low-order, middle-order and high-order;

[0158] Step 2, obtaining frequency offsets according to the difference between the detection dominant frequencies of the low-order, middle-order and high-order and the corresponding stored dominant frequencies; and constructing a frequency offset sequence according to the plurality of frequency offsets;

[0159] Step 3, calculating the distance between the fundamental frequency in each IMF frequency spectrum and the 50Hz fundamental wave, finding the IMF frequency spectrum corresponding to the minimum distance, calculating the 50Hz fundamental wave energy proportion offset, and constructing a fundamental wave energy proportion offset sequence;

[0160] Step 4, respectively obtaining corresponding attention degree sequences for the frequency offset sequence and the fundamental wave energy proportion offset sequence;

[0161] Step 5, processing the frequency offset sequence and the fundamental wave energy proportion offset sequence by using a dual-channel attention fusion BP network, taking the corresponding attention degree sequence as a channel attention weight, weighting the features of different channels, and obtaining a reactive compensation capacitor fault value.

[0162] Wherein, each step can be executed according to the execution process of each unit of the fault monitoring system of a reactive compensation capacitor of embodiment 1, which will not be described one by one in this embodiment.

[0163] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0164] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0165] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0166] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0167] The above description is only specific implementation of the present application, and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fault monitoring system for a reactive compensation capacitor, characterized by The system comprises: A sub-system for EMD decomposition of each voltage sub-signal of the voltage signal of the reactive compensation capacitor to obtain a plurality of IMF frequency spectrums; dividing the plurality of IMF frequency spectrums into low-order, intermediate-order and high-order to obtain detection dominant frequencies of the low-order, intermediate-order and high-order respectively; A first offset sequence construction sub-system for obtaining frequency offsets according to the differences between the detection dominant frequencies of the low-order, intermediate-order and high-order and the corresponding stored dominant frequencies respectively; and constructing a frequency offset sequence according to the plurality of frequency offsets; A second offset sequence construction sub-system for calculating the distance between the fundamental frequency and the fundamental wave in each IMF frequency spectrum, finding the order of the IMF frequency spectrum corresponding to the minimum distance, calculating the fundamental wave energy proportion offset, and constructing a fundamental wave energy proportion offset sequence; An attention acquisition sub-system for respectively acquiring corresponding attention sequences of the frequency offset sequence and the fundamental wave energy proportion offset sequence; A classification sub-system for processing the frequency offset sequence and the fundamental wave energy proportion offset sequence by using a dual-channel attention fusion BP network, taking the corresponding attention sequences as channel attention weights, weighting the features of different channels, and obtaining a reactive compensation capacitor fault value.

2. A system for monitoring the condition of a power compensation capacitor according to claim 1, wherein The sub-system for grading comprises: A voltage signal division unit for dividing the voltage signal of the reactive compensation capacitor to obtain a plurality of voltage sub-signals; An EMD decomposition unit for EMD decomposition of each voltage sub-signal to obtain a plurality of IMF signals; An IMF signal division unit for dividing the plurality of IMF signals into low-order, intermediate-order and high-order; A Fourier transform unit for Fourier transforming each IMF signal to obtain each IMF frequency spectrum; A detection dominant frequency acquisition unit for extracting the fundamental frequency from each IMF frequency spectrum and acquiring the detection dominant frequencies of the low-order, intermediate-order and high-order according to the fundamental frequency.

3. A system for monitoring the condition of a power compensation capacitor according to claim 2, wherein The detection dominant frequency acquisition unit comprises: A fundamental frequency extraction sub-unit for obtaining the maximum amplitude in each IMF frequency spectrum and taking the frequency corresponding to the maximum amplitude as the fundamental frequency; A first detection dominant frequency acquisition sub-unit for acquiring the detection dominant frequencies of the low-order according to the fundamental frequencies of the IMF frequency spectrums in the low-order; A second detection dominant frequency acquisition sub-unit for acquiring the detection dominant frequencies of the intermediate-order according to the fundamental frequencies of the IMF frequency spectrums in the intermediate-order; A third detection dominant frequency acquisition sub-unit for acquiring the detection dominant frequencies of the high-order according to the fundamental frequencies of the IMF frequency spectrums in the high-order.

4. A system for monitoring the condition of a power compensation capacitor according to claim 1 or 3, characterized in that The acquisition process of the detection dominant frequency is as follows: In the same order, the fundamental frequency and the amplitude of the same IMF frequency spectrum are multiplied to obtain a fundamental frequency-amplitude product; The fundamental frequency-amplitude products in the same order are added to obtain a fundamental frequency-amplitude sum; The fundamental frequency-amplitude sums in the same order are added to obtain a total amplitude sum; The ratio of the fundamental frequency-amplitude sum to the total amplitude sum is taken as the detection dominant frequency.

5. A system for monitoring the condition of a power compensation capacitor according to claim 1, wherein The first offset sequence construction sub-system comprises: The first frequency shift coefficient unit is configured to subtract the low-order detected dominant frequency from the low-order stored dominant frequency to obtain a first subtraction result, take an absolute value of the first subtraction result to obtain a first dominant frequency gap, and take a ratio of the first dominant frequency gap to the low-order stored dominant frequency as a first frequency shift coefficient; The second frequency shift coefficient unit is configured to subtract the middle-order detected dominant frequency from the middle-order stored dominant frequency to obtain a second subtraction result, take an absolute value of the second subtraction result to obtain a second dominant frequency gap, and take a ratio of the second dominant frequency gap to the middle-order stored dominant frequency as a second frequency shift coefficient; The third frequency shift coefficient unit is configured to subtract the high-order detected dominant frequency from the high-order stored dominant frequency to obtain a third subtraction result, take an absolute value of the third subtraction result to obtain a third dominant frequency gap, and take a ratio of the third dominant frequency gap to the high-order stored dominant frequency as a third frequency shift coefficient; The frequency offset calculation unit is configured to add the first frequency shift coefficient, the second frequency shift coefficient, and the third frequency shift coefficient to obtain a frequency offset. The frequency offset sequence unit is configured to arrange the plurality of frequency offsets in a time sequence to obtain a frequency offset sequence.

6. A system for monitoring the condition of a power compensation capacitor according to claim 1, wherein The second offset sequence construction subsystem includes: The IMF spectrum order calculation unit is configured to calculate a distance between a fundamental frequency and a 50Hz fundamental wave in each IMF spectrum, and select an IMF spectrum order corresponding to a minimum distance. The fundamental wave energy proportion offset calculation unit is configured to calculate a fundamental wave energy proportion offset of the 50Hz fundamental wave according to the IMF spectrum in the IMF spectrum order. The second offset sequence construction unit is configured to arrange a plurality of fundamental wave energy proportion offsets in a time sequence to obtain a fundamental wave energy proportion offset sequence.

7. A system for monitoring the condition of a power compensation capacitor as defined in claim 1, wherein, The formula for calculating the fundamental wave energy proportion offset of the 50Hz fundamental wave is: Wherein, γ is the fundamental energy proportion offset of 50Hz fundamental wave, y k is the kth amplitude value in the order of IMF spectrum, f k is the kth amplitude value in the order of IMF spectrum corresponding to the frequency value, K is the number of amplitude values corresponding to all IMF spectrum in the order of IMF spectrum, f min is the fundamental frequency corresponding to the minimum distance, y min is the amplitude value of the fundamental frequency corresponding to the minimum distance, k is the number of amplitude values.

8. A system for monitoring the condition of a power compensation capacitor according to claim 1, wherein, The attention degree acquisition subsystem includes: The segmentation unit is configured to segment the frequency offset sequence and the fundamental wave energy proportion offset sequence into segments of a fixed length M, where M is a positive integer greater than 1. The first attention degree unit is configured to count a number of frequency offsets greater than a frequency offset threshold in each segment of the frequency offset sequence, and normalize the number to obtain an attention degree of each segment of the frequency offset sequence. The second attention degree unit is configured to count a number of fundamental wave energy proportion offsets greater than a fundamental wave energy proportion offset threshold in each segment of the fundamental wave energy proportion offset sequence, and normalize the number to obtain an attention degree of each segment of the fundamental wave energy proportion offset sequence. The first attention degree sequence unit is configured to construct an attention degree sequence of the frequency offset sequence from the attention degrees corresponding to the frequency offset sequence. The second attention degree sequence unit is configured to construct an attention degree sequence of the fundamental wave energy proportion offset sequence from the attention degrees corresponding to the fundamental wave energy proportion offset sequence.

9. A system for monitoring the condition of a power compensation capacitor according to claim 1 wherein, The dual-path attention fusion BP network includes a first path sequence processing unit, a second path sequence processing unit, and a fully connected layer. The first path sequence processing unit and the second path sequence processing unit each include an input layer, a plurality of first hidden layers, a concatenation layer, a multiplier, and a second hidden layer. The input layer is configured to input a sequence into the plurality of first hidden layers. Each first hidden layer is configured to extract a feature value for each segment in the frequency offset sequence or the fundamental energy proportion offset sequence; The splicing layer is configured to splice the feature values output by the multiple first hidden layers to obtain a feature sequence; The multiplier is configured to multiply the feature sequence and a corresponding attention degree sequence element by element to obtain an enhanced feature sequence; Each second hidden layer is configured to output the enhanced feature sequence; The full connection layer is configured to obtain the reactive compensation capacitor fault value according to the outputs of the second hidden layers in the first sequence processing unit and the second sequence processing unit.

10. A method of fault monitoring of a reactive compensation capacitor, characterized by, The method comprises: performing EMD decomposition on each voltage sub-signal of the voltage signal of the reactive compensation capacitor to obtain multiple IMF frequency spectrums; performing low-order, intermediate-order and high-order division on the multiple IMF frequency spectrums to respectively obtain detection dominant frequencies of the low-order, intermediate-order and high-order; obtaining frequency offsets according to the differences between the detection dominant frequencies of the low-order, intermediate-order and high-order and corresponding stored dominant frequencies, and constructing a frequency offset sequence according to the multiple frequency offsets; calculating the distances between the fundamental frequency and the fundamental wave in each IMF frequency spectrum, finding the order of the IMF frequency spectrum corresponding to the minimum distance, calculating the fundamental energy proportion offset, and constructing a fundamental energy proportion offset sequence; respectively obtaining corresponding attention degree sequences for the frequency offset sequence and the fundamental energy proportion offset sequence; processing the frequency offset sequence and the fundamental energy proportion offset sequence by using a two-path attention fusion BP network, taking the corresponding attention degree sequences as channel attention weights, weighting the features of different channels, and obtaining the reactive compensation capacitor fault value.

Images