An adaptive high-voltage circuit breaker diagnostic system

Abstract

The present invention relates to the field of high-voltage circuit breaker fault diagnosis, and in particular to an adaptively adjusted high-voltage circuit breaker diagnostic system, comprising a perception layer, a network layer and an application layer that are interconnected and communicated with each other, the perception layer is used to collect multi-dimensional signal data of the high-voltage circuit breaker in real time and send it to the network layer, the network layer is used to receive the multi-dimensional signal data and forward it to the application layer, and the application layer has a built-in adaptive fault diagnosis algorithm, including the following steps: S1, adaptively denoising the multi-dimensional signal data collected by the perception layer; S2, combining the denoised multi-dimensional signal data into a multi-dimensional feature vector as sample data; S3, inputting the sample data into an adaptive fault diagnosis model trained in advance to obtain the final diagnosis result. The adaptively adjusted high-voltage circuit breaker diagnostic system provided by the present invention achieves the effect of improving the accuracy of fault diagnosis by setting adaptive threshold denoising and adaptive learning rate.

Description

Technical Field

[0001] The invention relates to the technical field of high-voltage circuit breaker fault diagnosis, and in particular to a self-adaptive high-voltage circuit breaker diagnosis system. Background Art

[0002] High-voltage circuit breakers are key components of the power system. They play a role in control (load switching) and protection (fault disconnection) in the power system. Their operating status is directly related to the safety and stability of the lines where they are located. On the one hand, according to the actual operating needs of the power system, high-voltage circuit breakers play a control role in the power system, implementing the switching on and off operations of specific lines. On the other hand, high-voltage circuit breakers play a protective role in the power system. When a short-circuit fault occurs in the system, the high-voltage circuit breaker will quickly cut off the huge short-circuit current to protect the entire power system from further damage. Once a high-voltage circuit breaker fails, it will directly endanger the safety and stability of the entire power system.

[0003] In recent years, fault diagnosis methods based on deep learning have received more and more attention, among which the most representative is the convolutional neural network. At present, most fault diagnosis studies based on convolutional neural networks only focus on changes in learning rate, and few make changes to both learning rate and momentum. On the other hand, wavelet thresholds are often used for signal denoising. Traditional hard threshold denoising and soft threshold denoising methods require setting a threshold and comparing it with the wavelet coefficients of the signal, which can effectively filter out noise signals in the signal. Therefore, the setting of the threshold is crucial. If it is too large, the noise cannot be completely filtered out, and if it is too small, the characteristic information in the signal will be partially filtered out. Commonly used The threshold selection lacks a certain degree of adaptive ability. Summary of the invention

[0004] The object of the present invention is to provide a self-adaptive high-voltage circuit breaker diagnostic system to solve the above problems in the prior art.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] A self-adjusting high-voltage circuit breaker diagnosis system comprises a perception layer, a network layer and an application layer which are communicatively connected to each other, the perception layer comprises a sensing unit and a data transmission unit, the perception layer is used to collect multi-dimensional signal data of the high-voltage circuit breaker in real time and send it to the network layer, the network layer comprises a gateway device, the network layer is used to receive the multi-dimensional signal data collected by the perception layer and forward it to the application layer, the application layer comprises a visualization platform, the application layer has a built-in self-adaptive fault diagnosis algorithm, and the self-adaptive fault diagnosis algorithm comprises the following steps:

[0007] S1, the multi-dimensional signal data collected by the perception layer is adaptively denoised using the CEEMDAN algorithm combined with the adaptive wavelet threshold denoising method;

[0008] S2, performing homogeneous processing on the denoised multidimensional signal data and combining them into a multidimensional feature vector as sample data;

[0009] S3, inputting the sample data combined in step S2 into a pre-trained adaptive fault diagnosis model based on adaptive learning rate and Nesterov momentum to obtain a final diagnosis result.

[0010] Furthermore, the multi-dimensional signal data includes a current signal of an opening and closing coil of a high-voltage circuit breaker, a vibration signal at an operating mechanism, and a travel curve signal of a moving contact.

[0011] Furthermore, the sensing unit includes a Hall current sensor installed at the opening and closing coil, an acceleration sensor installed at the crossbeam of the operating mechanism, and a displacement sensor installed on the transmission mechanism.

[0012] Furthermore, the communication method of the high-voltage circuit breaker diagnostic system adopts LoRa wireless communication.

[0013] Furthermore, the adaptive denoising method specifically comprises the following steps:

[0014] A1. Use CEEMDAN algorithm to decompose the signal into a series of IMF components;

[0015] A2, using the continuous mean square error criterion to determine the high-frequency IMF component containing more noise;

[0016] A3, using the adaptive wavelet threshold denoising method to denoise the high-frequency IMF component containing more noise, wherein the adaptive wavelet threshold is used: Where: N is the signal length, r 1,2 is the correlation coefficient between the decomposition signal and the source signal, ω b,k is the wavelet coefficient, M is the empirical constant;

[0017] A4. Reconstruct the denoised IMF component and the undenoised IMF component to obtain a denoised signal using a joint denoising method.

[0018] Furthermore, the training method of the adaptive fault diagnosis model comprises the following steps:

[0019] B1. Artificially simulate various fault types of high-voltage circuit breakers, collect multi-dimensional signal data through the perception layer, and perform adaptive denoising;

[0020] B2. Perform homogeneous processing on the multi-dimensional signal data and combine them into multi-dimensional feature vectors, randomly select 90% of them as training sample data and the remaining 10% as test sample data;

[0021] B3. Construct a convolutional neural network (CNN) model and initialize the weights, biases, and convolution kernel size parameters. A normalization layer is set between the convolution layer and the pooling layer to speed up the training. The formula of the normalization layer is as follows: Where: and N i l j are the input and response output of the lth layer, m is the number of samples in batch normalization, and γ l and β l They are the scaling and translation parameters for each feature map;

[0022] B4, input the training sample data in step B2 into the CNN model in step B3, and train the CNN model using adaptive learning rate and Nesterov momentum method;

[0023] B5. Input the test sample data in step B2 into the CNN model trained in step B4 to verify the diagnostic performance.

[0024] Furthermore, the artificial simulation method at least includes: loosening the base bolts to simulate the loose base bolt fault state; removing the buffer spring to simulate the buffer spring fatigue fault state; shortening the length of the transmission rod to simulate the transmission mechanism fault state; reducing the prestress of the closing spring to simulate the closing spring fatigue fault state.

[0025] Furthermore, the adaptive learning rate is used to solve the overfitting phenomenon caused by the acceleration of network convergence speed. The specific rules of the adaptive learning rate are as follows: Error change rate Where err(q) is the error of the qth training; the learning rate of the qth training is Where ɑ is the threshold factor, ɑ∈(0,0.2), and δ is the penalty factor used to prevent the learning rate from changing too quickly, δ∈(0,0.1).

[0026] Furthermore, the Nesterov momentum method is used to adjust the momentum gradient descent speed of the stochastic gradient descent method, and its formula is as follows: Where, d i and d i-1 are the update directions for the i-th and i-1-th times respectively, β is the correction factor, β∈(0,1), η is the learning rate, and g(θ) represents the gradient of the objective function at θ.

[0027] In summary, the present invention has the following beneficial effects:

[0028] The adaptively adjusted high-voltage circuit breaker diagnosis system of the present invention achieves the effect of improving the fault diagnosis accuracy by adopting the CEEMDAN algorithm combined with adaptive wavelet threshold denoising and the Nesterov momentum method combined with adaptive learning rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart of the adaptive fault diagnosis algorithm of the present invention. DETAILED DESCRIPTION

[0030] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0031] The same parts are denoted by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to the attached Figure 1 In the description of this specification, the words "bottom" and "top", "inside" and "outside" refer to directions toward or away from a particular component geometry, respectively. In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this specification, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0032] Embodiment 1:

[0033] like Figure 1 As shown, the present invention proposes an adaptively adjusted high-voltage circuit breaker diagnosis system, including a perception layer, a network layer and an application layer that are communicatively connected to each other, the perception layer includes a sensor unit and a data transmission unit, the perception layer is used to collect multi-dimensional signal data of the high-voltage circuit breaker in real time and send it to the network layer, the network layer includes a gateway device, the network layer is used to receive the multi-dimensional signal data collected by the perception layer and forward it to the application layer, the application layer includes a visualization platform, the application layer has a built-in adaptive fault diagnosis algorithm, and the adaptive fault diagnosis algorithm includes the following steps:

[0034] S1, the multi-dimensional signal data collected by the perception layer is adaptively denoised using the CEEMDAN algorithm combined with the adaptive wavelet threshold denoising method;

[0035] S2, performing homogeneous processing on the denoised multidimensional signal data and combining them into a multidimensional feature vector as sample data;

[0036] S3, inputting the sample data combined in step S2 into a pre-trained adaptive fault diagnosis model based on adaptive learning rate and Nesterov momentum to obtain a final diagnosis result.

[0037] In this embodiment, the multi-dimensional signal data includes the current signal of the opening and closing coil of the high-voltage circuit breaker, the vibration signal at the operating mechanism, and the travel curve signal of the moving contact.

[0038] In this embodiment, the sensing unit includes a Hall current sensor installed at the opening and closing coil, an acceleration sensor installed at the crossbeam of the operating mechanism, and a displacement sensor installed on the transmission mechanism.

[0039] In this embodiment, the communication method of the high-voltage circuit breaker diagnostic system adopts LoRa wireless communication.

[0040] In this embodiment, the adaptive denoising method specifically includes the following steps:

[0041] A1. Use CEEMDAN algorithm to decompose the signal into a series of IMF components;

[0042] A2, using the continuous mean square error criterion to determine the high-frequency IMF component containing more noise;

[0043] A3, using the adaptive wavelet threshold denoising method to denoise the high-frequency IMF component containing more noise, wherein the adaptive wavelet threshold is used: Where: N is the signal length, r 1,2 is the correlation coefficient between the decomposition signal and the source signal, ω b,k is the wavelet coefficient, M is the empirical constant;

[0044] A4. Reconstruct the denoised IMF component and the undenoised IMF component to obtain a denoised signal using a joint denoising method.

[0045] In this embodiment, the training method of the adaptive fault diagnosis model includes the following steps:

[0046] B1. Artificially simulate various fault types of high-voltage circuit breakers, collect multi-dimensional signal data through the perception layer, and perform adaptive denoising;

[0047] B2. Perform homogeneous processing on the multi-dimensional signal data and combine them into multi-dimensional feature vectors, randomly select 90% of them as training sample data and the remaining 10% as test sample data;

[0048] B3. Construct a convolutional neural network (CNN) model and initialize the weights, biases, and convolution kernel size parameters. A normalization layer is set between the convolution layer and the pooling layer to speed up the training. The formula of the normalization layer is as follows: Where: and are the input and response output of the lth layer, m is the number of samples in batch normalization, and γ l and β l They are the scaling and translation parameters for each feature map;

[0049] B4, input the training sample data in step B2 into the CNN model in step B3, and train the CNN model using adaptive learning rate and Nesterov momentum method;

[0050] B5. Input the test sample data in step B2 into the CNN model trained in step B4 to verify the diagnostic performance.

[0051] In this embodiment, the artificial simulation method at least includes: loosening the base bolts to simulate the loose base bolt fault state; removing the buffer spring to simulate the buffer spring fatigue fault state; shortening the length of the transmission rod to simulate the transmission mechanism fault state; reducing the prestress of the closing spring to simulate the closing spring fatigue fault state.

[0052] In this embodiment, the adaptive learning rate is used to solve the overfitting phenomenon caused by the acceleration of network convergence speed. The specific rules of the adaptive learning rate are as follows: Error change rate Where err(q) is the error of the qth training; the learning rate of the qth training is Where ɑ is the threshold factor, ɑ∈(0,0.2), and δ is the penalty factor used to prevent the learning rate from changing too quickly, δ∈(0,0.1).

[0053] In this embodiment, the Nesterov momentum method is used to adjust the momentum gradient descent speed of the stochastic gradient descent method, and its formula is as follows: Where, d i and d i-1 are the update directions for the i-th and i-1-th times respectively, β is the correction factor, β∈(0,1), η is the learning rate, and g(θ) represents the gradient of the objective function at θ.

[0054] In the embodiments disclosed in the present invention, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; "connected" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments disclosed in the present invention can be understood according to specific circumstances.

[0055] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. An adaptively adjusted high-voltage circuit breaker diagnostic system, comprising a perception layer, a network layer and an application layer that are interconnected, characterized in that: The perception layer includes a sensing unit and a data transmission unit. The perception layer is used to collect multi-dimensional signal data of the high-voltage circuit breaker in real time and send it to the network layer. The network layer includes a gateway device. The network layer is used to receive the multi-dimensional signal data collected by the perception layer and forward it to the application layer. The application layer includes a visualization platform. The application layer has a built-in adaptive fault diagnosis algorithm. The adaptive fault diagnosis algorithm includes the following steps: S1, the multi-dimensional signal data collected by the perception layer is adaptively denoised using the CEEMDAN algorithm combined with the adaptive wavelet threshold denoising method; S2, performing homogeneous processing on the denoised multidimensional signal data and combining them into a multidimensional feature vector as sample data; S3, inputting the sample data combined in step S2 into the pre-trained adaptive fault diagnosis model based on adaptive learning rate and Nesterov momentum to obtain the final diagnosis result; The training method of the adaptive fault diagnosis model comprises the following steps: B1. Artificially simulate various fault types of high-voltage circuit breakers, collect multi-dimensional signal data through the perception layer, and perform adaptive denoising; B2. Perform homogeneous processing on the multi-dimensional signal data and combine them into a multi-dimensional feature vector, randomly select 90% of them as training sample data, and the remaining 10% as test sample data; B3. Construct a convolutional neural network (CNN) model and initialize the weights, biases, and convolution kernel size parameters. A normalization layer is set between the convolution layer and the pooling layer to speed up the training. The formula of the normalization layer is as follows: , where: and are the input and response output of the lth layer, m is the number of samples in batch normalization, and They are the scaling and translation parameters for each feature map; B4, input the training sample data in step B2 into the CNN model in step B3, and train the CNN model using adaptive learning rate and Nesterov momentum method; B5, input the test sample data in step B2 into the CNN model trained in step B4 to verify the diagnostic performance; In the B1 process, the various fault types of the high-voltage circuit breaker are artificially simulated, including at least: loosening the base bolts to simulate the loose base bolt fault state; removing the buffer spring to simulate the buffer spring fatigue fault state; shortening the length of the transmission rod to simulate the transmission mechanism fault state; reducing the prestress of the closing spring to simulate the closing spring fatigue fault state; The adaptive learning rate is used to solve the overfitting phenomenon caused by the acceleration of network convergence speed. The specific rules of the adaptive learning rate are as follows: Error change rate , where err(q) is the error of the qth training; the learning rate of the qth training , where ɑ is the threshold factor, ɑ∈(0,0.2), δ is the penalty factor used to prevent the learning rate from changing too quickly, δ∈(0,0.1); The Nesterov momentum method is used to adjust the momentum gradient descent speed of the stochastic gradient descent method, and its formula is as follows: , where d i and d i-1 are the update directions for the i-th and i-1-th times respectively, β is the correction factor, β∈(0,1), η is the learning rate, and g(θ) represents the gradient of the objective function at θ.

2. The self-adaptive high-voltage circuit breaker diagnostic system according to claim 1, characterized in that: The multi-dimensional signal data includes a current signal of a closing and opening coil of a high-voltage circuit breaker, a vibration signal at an operating mechanism, and a travel curve signal of a moving contact.

3. The self-adaptive high-voltage circuit breaker diagnostic system according to claim 2, characterized in that: The sensing unit comprises a Hall current sensor installed at the opening and closing coil, an acceleration sensor installed at the crossbeam of the operating mechanism, and a displacement sensor installed on the transmission mechanism.

4. The self-adaptive high-voltage circuit breaker diagnostic system according to claim 1, characterized in that: The communication mode of the high-voltage circuit breaker diagnostic system adopts LoRa wireless communication.

5. The self-adaptive high-voltage circuit breaker diagnostic system according to claim 1, characterized in that: The adaptive denoising method specifically comprises the following steps: A1. Use CEEMDAN algorithm to decompose the signal into a series of IMF components; A2, using the continuous mean square error criterion to determine the high-frequency IMF components with more noise; A3. Adopting adaptive wavelet threshold denoising method to denoise the high-frequency IMF components with more noise, wherein the adaptive wavelet threshold is adopted: , where: N is the signal length, r 1,2 is the correlation coefficient between the decomposition signal and the source signal, , is the wavelet coefficient, M is the empirical constant; A4. Reconstruct the denoised IMF components and the undenoised IMF components to obtain the denoised signal using the joint denoising method.

Images