A CNN-BiLSTM-based fault diagnosis method for high-voltage shunt reactor winding looseness

Abstract

The present application belongs to the field of electric power technology, and particularly relates to a high-voltage shunt reactor winding looseness fault diagnosis method based on CNN-BiLSTM. The method of the present application uses a vibration measuring device to collect vibration signals of a high-voltage shunt reactor, selects the surface of the reactor oil tank as the measurement position, pre-processes the collected signals, analyzes the pre-processed vibration signals, uses the high-voltage shunt reactor body vibration signals to judge and classify the winding state based on the algorithm combining CNN and Bi-LSTM, and realizes high-voltage shunt reactor winding fault diagnosis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field related to electric power, and particularly relates to a high-voltage shunt reactor winding looseness fault diagnosis method based on CNN-BiLSTM. BACKGROUND

[0002] The high-voltage shunt reactor is an important reactive power compensation device in the power system, and plays an important role in the safe and stable operation of the power system. With the increasing number of high-voltage shunt reactors put into operation in recent years, the fault defect diagnosis of the high-voltage shunt reactor has been paid more and more attention. The high-voltage shunt reactor will produce a lot of vibration and noise during operation, and the component looseness and damage caused by vibration is one of the main reasons for the defects and faults of the reactor, which can cause serious failure of the equipment in extreme cases and threaten the stability of the power system.

[0003] The current commonly used state evaluation methods such as electrical quantity monitoring and oil chromatography monitoring show high accuracy in defect diagnosis of the high-voltage shunt reactor in the later stage of the fault (insulation defects), but it is difficult to diagnose the early mechanical faults such as component looseness in time. The surface vibration signal of the power equipment is often closely related to its mechanical state, and the mechanical defect diagnosis method based on the vibration signal has been widely studied. The vibration method uses the surface vibration signal of the high-voltage shunt reactor to detect its mechanical state, has the advantages of online, non-invasive and sensitive to early mechanical faults, and related scientific research institutions and enterprises have developed various high-voltage shunt reactor online monitoring instruments based on vibration acoustic signals, realizing real-time observation and recording of the surface vibration signal of the high-voltage shunt reactor oil tank. However, the research on the later processing means of the vibration signal is not deep enough, and it is difficult to judge whether the winding of the high-voltage shunt reactor has mechanical faults by analyzing the surface vibration signal of the high-voltage shunt reactor, which brings certain difficulties to the fault diagnosis of the winding looseness of the high-voltage shunt reactor.

[0004] Therefore, it is of great significance to invent a high-voltage shunt reactor winding looseness fault diagnosis method based on the surface vibration signal of the high-voltage shunt reactor, which provides protection for the safe and stable operation of the power system, and the method has a solid engineering background and broad application prospect. SUMMARY

[0005] In order to solve the above problems, the application provides a high-voltage shunt reactor winding looseness fault diagnosis method based on CNN-BiLSTM, which adopts a vibration measuring device to collect the surface vibration signal of the high-voltage shunt reactor oil tank, and classifies the signal through the algorithm combined by CNN and Bi-LSTM, so as to judge whether the winding of the high-voltage shunt reactor has mechanical faults. The specific technical scheme is as follows:

[0006] A CNN-BiLSTM-based fault diagnosis method for high-voltage shunt reactor winding looseness, comprising the following steps:

[0007] S1: multi-channel acquisition of high-voltage shunt reactor oil tank surface vibration signals, obtaining vibration signals at each measuring point; S2: performing fast Fourier transform and wavelet denoising on the collected signals, extracting key data features according to frequency spectrum analysis, and selecting power spectrum as a feature for analysis;

[0008] S3: training the algorithm constructed by combining CNN and Bi-LSTM with the processed data;

[0009] S4: outputting the Softmax function to determine whether the input signal has winding looseness fault;

[0010] S5: when the Softmax function output value is 0, the high-voltage shunt reactor winding is normal; when the Softmax function output value is 1, the high-voltage shunt reactor has winding looseness fault.

[0011] Preferably, in step S1, the vibration measurement device is used to collect the vibration signals on the surface of the high-voltage shunt reactor oil tank.

[0012] Preferably, the vibration measurement device comprises a vibration sensor, an AD conversion module and a microprocessor; the vibration sensor, the AD conversion module and the microprocessor are connected in sequence; the vibration sensor is provided with a plurality of vibration sensors, which are respectively arranged on each vibration measuring point; the vibration sensor is used to collect the vibration signals on the surface of the high-voltage shunt reactor oil tank and transmit the collected vibration signals to the AD conversion module; the AD conversion module is used to perform analog-digital conversion on the vibration signals on the surface of the high-voltage shunt reactor oil tank collected by the vibration sensor, and transmit the converted vibration signals on the surface of the high-voltage shunt reactor oil tank to the microprocessor; the microprocessor is used to process the digital vibration signals on the surface of the high-voltage shunt reactor oil tank converted by the AD conversion module.

[0013] Preferably, the measuring points are arranged in the following manner: the vibration measuring points are evenly distributed along the circumference of the oil tank in a counterclockwise direction at 1 / 4, 1 / 2 and 3 / 4 of the height of the mailbox on the surface of the high-voltage shunt reactor oil tank, and the vibration sensors are arranged on the vibration measuring points.

[0014] Preferably, step S3 specifically comprises the following steps:

[0015] S31: twice convolution of the input signal and the filter function;

[0016] S32: LeakyRelu linear activation unit is used in the convolution layer, and a LeakyRelu activation function is used after the convolution operation to increase the non-linear and sparse characteristics of the network structure;

[0017] S33: A max-pooling operation is performed, only the maximum value of the convolution output is selected, and the number of output neurons is reduced;

[0018] S34: Steps S31-S33 are repeated three times;

[0019] S35: After the last pooling, a Bi-LSTM recurrent neural network is entered, and the activation function is a tanh non-linear function;

[0020] S36: The data enters two fully connected layers, all activations of the layer and the previous layer are fully connected in the fully connected layer, and Dropout is used in the fully connected layer to reduce overfitting;

[0021] S37: The output is obtained;

[0022] Preferably, the convolution formula in step S31 is:

[0023]

[0024] where x n is the nth signal, h k-n is the k-n filter, N is the number of signals, the output vector is y, and the table k represents the kth element of the vector.

[0025] Preferably, the LeakyRelu function formula in step S32 is:

[0026] Preferably, the value of Dropout in step S36 is 0.5.

[0027] The application has the beneficial effects that the application provides a high-voltage shunt reactor winding looseness fault diagnosis method based on CNN-BiLSTM, multi-channel collection is performed on high-voltage shunt reactor oil tank surface vibration signals, and vibration signals of each measuring point are acquired; the collected signals are subjected to fast Fourier transform and wavelet denoising, key data features are extracted according to spectrum analysis, and power spectrum is selected as a feature for analysis; the processed data are used to train an algorithm constructed by combining CNN and Bi-LSTM; a Softmax function is used to output to judge whether the input signal exists winding looseness fault; when the Softmax function output value is 0, the high-voltage shunt reactor winding state is normal; and when the Softmax function output value is 1, the high-voltage shunt reactor occurs winding looseness fault. The method of the application uses a vibration measuring device to collect vibration signals of the high-voltage shunt reactor, selects the surface of the reactor oil tank as a measurement position, pre-processes the collected signals, analyzes the pre-processed vibration signals, uses an algorithm combined by CNN and Bi-LSTM, uses the high-voltage shunt reactor body vibration signals to judge and classify the winding state, and realizes high-voltage shunt reactor winding fault diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the specific embodiments or the prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual proportion.

[0029] Figure 1 The flowchart of the application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0031] It should be understood that, when used in the specification and the appended claims, the terms "comprise" and "include" indicate the existence of described features, integers, steps, operations, elements and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components and / or sets thereof.

[0032] It is also to be understood that the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in the description of the application and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0033] It should be further understood that the term "and / or" as used in the specification and in the claims, if any, means any one of the associated listed items, as well as all possible combinations of the items.

[0034] As Figure 1 The specific embodiment of the application provides a CNN-BiLSTM-based high-voltage shunt reactor winding looseness fault diagnosis method. The vibration signal of the high-voltage shunt reactor is collected by a vibration measuring device, the surface of the reactor oil tank is selected as the measurement position, the collected signal is preprocessed, and the preprocessed vibration signal is analyzed to detect whether the reactor winding has looseness fault, including the following steps:

[0035] S1: Collecting the vibration signals on the surface of the high-voltage shunt reactor oil tank and obtaining the vibration signals at each measuring point; specifically, the vibration signals on the surface of the high-voltage shunt reactor oil tank are collected by a vibration measuring device. The vibration measuring device includes a vibration sensor, an AD conversion module, and a microprocessor; the vibration sensor, the AD conversion module, and the microprocessor are connected in sequence; the vibration sensor is provided with a plurality of vibration sensors, which are arranged on each vibration measuring point. The vibration sensor is used to collect the vibration signals on the surface of the high-voltage shunt reactor oil tank and transmit the collected vibration signals to the AD conversion module; the AD conversion module is used to convert the vibration signals on the surface of the high-voltage shunt reactor oil tank collected by the vibration sensor into digital signals and transmit the converted vibration signals on the surface of the high-voltage shunt reactor oil tank to the microprocessor; the microprocessor is used to process the digital vibration signals on the surface of the high-voltage shunt reactor oil tank converted by the AD conversion module.

[0036] The vibration measuring points are arranged in the following manner: the vibration measuring points are evenly distributed along the circumference of the oil tank at the height of 1 / 4, 1 / 2 and 3 / 4 of the mailbox on the surface of the high-voltage shunt reactor, and the vibration sensors are arranged on the vibration measuring points.

[0037] S2: The collected signal is subjected to Fast Fourier Transform (FFT) and wavelet denoising to achieve the pre-processing of smoothing filtering and denoising, which mainly suppresses the unnecessary components, extracts the key data features according to the frequency spectrum analysis, and selects the power spectrum as the characteristic for analysis.

[0038] S3: Train the algorithm constructed by combining CNN (Convolutional Neural Networks) and Bi-LSTM (Bi-directional Long Short-Term Memory) after processing the data; specifically including the following steps:

[0039] S31: Convolve the input signal with the filter function twice; the convolution formula is:

[0040]

[0041] where x n is the nth signal, h k-n is the k-n filter, N is the number of signals, the output vector is y, and the table k represents the kth element of the vector.

[0042] S32: Use LeakyRelu linear activation unit in the convolution layer, and use LeakyRelu activation function after convolution operation to increase the non-linear and sparse characteristic structure in the network; the LeakyRelu function formula is:

[0043] S33: Perform maximum pooling operation, only select the maximum value of the convolution output, and reduce the number of output neurons;

[0044] S34: Repeat steps S31-S33 three times;

[0045] S35: After the last pooling, enter the Bi-LSTM recurrent neural network, and the activation function is the tanh non-linear function;

[0046] S36: The data enters two fully connected layers, all activations of the layer and the previous layer are fully connected in the fully connected layer, and Dropout is used in the fully connected layer to reduce overfitting; the value is 0.5.

[0047] S37 obtains the output.

[0048] S4: The output uses the Softmax function to determine whether the input signal has a winding looseness fault;

[0049] S5: When the Softmax function output value is 0, the winding state of the high-voltage shunt reactor is normal; when the Softmax function output value is 1, the high-voltage shunt reactor has a winding looseness fault.

[0050] The method of the present application uses a vibration measuring device to collect vibration signals of the high-voltage shunt reactor, selects the surface of the reactor oil tank as the measurement position, pre-processes the collected signals, analyzes the pre-processed vibration signals, uses the high-voltage shunt reactor body vibration signals to judge and classify the winding state based on the algorithm combined by CNN and Bi-LSTM, and realizes the winding fault diagnosis of the high-voltage shunt reactor.

[0051] In use, as shown in Figure 1 The vibration measuring device is used to collect the vibration signals on the surface of the high-voltage shunt reactor, and records are saved once every interval, and the interval time is 1 hour. With the help of fast Fourier transform (FFT), the time domain data saved by the record is converted into frequency domain data, and then the low frequency energy, the medium frequency energy and the high frequency energy of each data are extracted as characteristics for analysis; the algorithm model constructed by using CNN and Bi-LSTM takes the characteristic quantity as input, convolves in the CNN algorithm convolution layer and selects the improved activation function; after the data input, the output uses the Softmax function (the function calculates the probability distribution of k output classes) to judge whether the input signal exists winding looseness fault, and the output value of p is between 0 and 1, and their sum is equal to 1;

[0052]

[0053] When the output value of p is 0, the winding state of the high-voltage shunt reactor is normal; when the output value of p is 1, the high-voltage shunt reactor occurs winding looseness fault.

[0054] The method of the present application uses the algorithm combined by CNN and Bi-LSTM to judge and classify the winding state of the high-voltage shunt reactor, and realizes the winding fault diagnosis of the high-voltage shunt reactor.

[0055] Those skilled in the art can appreciate that the units of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both, and in order to clearly illustrate the interchangeability of hardware and software, the components of the examples have been described in the above description in general terms. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0056] In the embodiments provided in the present application, it should be understood that the division of units is only a logical functional division, and when actually implemented, there can be another division manner, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A CNN-BiLSTM-based fault diagnosis method for high-voltage shunt reactor winding looseness, characterized by: It comprises the following steps: S1: collecting the vibration signals on the surface of the high-voltage shunt reactor oil tank at multiple points to obtain the vibration signals at each measuring point; S2: performing fast Fourier transform and wavelet denoising on the collected signals, extracting key data features according to spectral analysis, and selecting power spectrum as a feature for analysis; S3: training the algorithm constructed by combining CNN and Bi-LSTM with the processed data; S4: outputting the judgment of the input signal by the Softmax function to determine whether there is a winding looseness fault; S5: when the output value of the Softmax function is 0, the winding state of the high-voltage shunt reactor is normal; and when the output value of the Softmax function is 1, the high-voltage shunt reactor has a winding looseness fault; The vibration signals on the surface of the high-voltage shunt reactor oil tank are collected by a vibration measuring device in step S1; The vibration measuring device comprises a vibration sensor, an AD conversion module and a microprocessor, which are connected in sequence; the vibration sensor is provided in several sets and arranged at each vibration measuring point; the vibration sensor is used to collect the vibration signals on the surface of the high-voltage shunt reactor oil tank and transmit the collected vibration signals to the AD conversion module; the AD conversion module is used to perform analog-digital conversion on the vibration signals on the surface of the high-voltage shunt reactor oil tank collected by the vibration sensor and transmit the converted vibration signals on the surface of the high-voltage shunt reactor oil tank to the microprocessor; and the microprocessor is used to process the digital vibration signals on the surface of the high-voltage shunt reactor oil tank converted by the AD conversion module; The measuring points are arranged in the following manner: the vibration measuring points are evenly distributed along the circumference of the oil tank at the height of 1 / 4, 1 / 2 and 3 / 4 of the mailbox on the surface of the high-voltage shunt reactor, and the vibration sensors are arranged on the vibration measuring points.

2. The CNN-BiLSTM-based fault diagnosis method for winding looseness of high-voltage shunt reactors according to claim 1, characterized in that: Step S3 specifically comprises the following steps: S31: performing twice convolution of the input signal and the filter function; S32: using a LeakyRelu linear activation unit in the convolution layer, and using a LeakyRelu activation function after the convolution operation to increase the non-linear and sparse characteristic structure in the network; S33: performing a max-pooling operation to select only the maximum value of the convolution output and reduce the number of output neurons; S34: repeating steps S31-S33 three times; S35: after the last pooling, entering a Bi-LSTM recurrent neural network, and the activation function is a tanh non-linear function; S36: the data enters two fully connected layers, all activations of the layer and the previous layer are fully connected in the fully connected layer, and Dropout is used in the fully connected layer to reduce overfitting; S37: obtaining the output.

3. The CNN-BiLSTM-based fault diagnosis method for winding looseness of high-voltage shunt reactors according to claim 2, characterized in that: The convolution formula in step S31 is: ； wherein is the nth signal, is the k-nth filter, N is the number of intermediate signals, and the output vector is y , the index k denotes the k element of the vector.

4. The CNN-BiLSTM-based fault diagnosis method for winding looseness of high-voltage shunt reactors according to claim 2, characterized in that: The LeakyRelu function formula in the step S32 is: .

5. The CNN-BiLSTM-based fault diagnosis method for winding looseness of high-voltage shunt reactors according to claim 2, characterized in that: The value of Dropout in step S36 is 0.5.

Images