CN119087139B - A fault diagnosis method for power grid fault waveform analysis based on YOLOv5 dual-branch - Google Patents

本发明涉及一种基于YOLOv5双分支的电网故障波形分析的故障诊断方法，包括构建神经网络架构，对行波信号数据进行去噪处理；将去噪后的行波信号数据转换为故障波形图片；利用YOLOv5模型的第一分支，检测行波信号数据中的行波头和故障点；当检测到行波头，立即将行波头时刻前后的故障波形图片和电气数据特征进行特征融合；将融合后的特征向量输入YOLOv5模型的第二分支，输出故障诊断结果。利用神经网络去噪技术，能够精确提取波形信号中的关键特征，大幅度减少噪声的影响，从而提高了故障诊断的准确性和稳定性。采用双分支YOLOv5模型进行故障时刻和故障类型的自动识别，不仅快速高效，还能够在复杂环境中准确捕捉多种故障特征，提升了系统故障诊断的智能化水平。

The present invention relates to a fault diagnosis method for power grid fault waveform analysis based on YOLOv5 double branches, including constructing a neural network architecture, denoising traveling wave signal data; converting the denoised traveling wave signal data into a fault waveform image; using the first branch of the YOLOv5 model to detect the traveling wave head and the fault point in the traveling wave signal data; when the traveling wave head is detected, immediately perform feature fusion on the fault waveform image and electrical data features before and after the traveling wave head moment; input the fused feature vector into the second branch of the YOLOv5 model, and output the fault diagnosis result. Using the neural network denoising technology, the key features in the waveform signal can be accurately extracted, and the influence of noise can be greatly reduced, thereby improving the accuracy and stability of fault diagnosis. The dual-branch YOLOv5 model is used to automatically identify the fault time and fault type, which is not only fast and efficient, but also can accurately capture a variety of fault features in a complex environment, thereby improving the intelligent level of system fault diagnosis.

Fault diagnosis method based on YOLOv double-branch power grid fault waveform analysis

Technical Field

The invention relates to the technical field of power grid fault diagnosis equipment, in particular to a power grid fault waveform analysis fault diagnosis method based on YOLOv double branches.

Background

In the field of distribution network fault diagnosis, the traditional method mainly depends on manual inspection and signal measurement and analysis based on sensors. The manual inspection requires a lot of manpower resources and time, is easily affected by human factors, and causes inefficiency. Although the conventional signal measurement and analysis method can detect basic power grid anomalies, the processing capacity and accuracy of the conventional signal measurement and analysis method are significantly limited in the face of complex fault types and various power grid structures.

The traditional fault locating method generally depends on a pre-defined threshold value and a rule, and is difficult to realize high-precision fault locating and analyzing due to insufficient processing capacity for complex waveforms and noise interference. In addition, expert systems and model-based predictive methods, while capable of utilizing a priori knowledge to aid decision making, have limited adaptability to new types of faults and complications, and fail to achieve automatic learning and optimization. Kalman filtering is a recursive algorithm widely used in signal processing and control systems, and is capable of estimating the state of the system in the presence of noise and uncertainty. However, while Kalman filtering overcomes some of the limitations of conventional approaches to some extent, its application in fault diagnosis still faces some challenges and limitations.

Major limitations of the prior art include, but are not limited to, reliance on manual intervention, poor diagnostic accuracy, low processing efficiency, and poor processing power for noise and complex waveforms. These problems limit the application of the traditional method in large-scale power grid systems, and cannot meet the requirements of intelligent and automatic operation and maintenance of the power grid.

Disclosure of Invention

The technical problem to be solved by the invention is to provide a fault diagnosis method based on YOLOv double-branch power grid fault waveform analysis, so as to overcome the defects in the prior art.

The technical scheme for solving the technical problems is as follows, a fault diagnosis method for analyzing the power grid fault waveform based on YOLOv double branches comprises the following steps:

S01, capturing traveling wave signal data within 1 second;

step S02, constructing a neural network architecture, and denoising traveling wave signal data;

Step S03, converting the denoised traveling wave signal data into a fault waveform picture;

step S04, detecting a traveling wave head and a fault point in traveling wave signal data by using a first branch of YOLOv model;

Step S05, when the traveling wave head is detected in the step S04, the fault waveform picture and the electrical data characteristic before and after the traveling wave head moment are fused with each other;

S06, inputting the fused feature vector into a second branch of YOLOv model, and outputting a fault diagnosis result;

the step S03 specifically includes the following steps:

Step S31, recording a traveling wave signal x (T) as time sequence data x (T ₁),x(t₂),...,x(t_N), dividing the long-time traveling wave signal into time windows T with fixed lengths, wherein each time window T contains traveling wave signal data from T _i to T _i+T-1;

and S32, converting the traveling wave signals in each time window T into a fault waveform picture I.

The fault diagnosis method based on the traveling wave signal transfer picture and YOLOv double-branch target detection has the advantages of being obvious. Firstly, key features in waveform signals can be accurately extracted by utilizing a neural network denoising technology, so that the influence of noise is greatly reduced, and the accuracy and stability of fault diagnosis are improved. And secondly, the double-branch YOLOv model is adopted to automatically identify the fault moment and fault type, so that the method is rapid and efficient, various fault characteristics can be accurately captured in a complex environment, and the intelligent level of system fault diagnosis is improved. The method overcomes the limitations of the traditional technology in efficiency, precision and application range, and provides an innovative and reliable solution for the safe operation and maintenance of the power grid system.

On the basis of the technical scheme, the invention can be improved as follows.

Further, the step S02 specifically includes the following steps:

Step S21, collecting traveling wave signal data containing noise and corresponding noiseless or low-noise traveling wave signal data;

S22, marking the collected traveling wave signal data, wherein the marking comprises a noise signal and a target noiseless signal;

step S23, the neural network architecture comprises:

An input layer for receiving original traveling wave signal data x= { x ₁,x₂,...,x_T }, wherein x _t is the t-th data point in the time series data;

the hidden layer is used for extracting the characteristics in the traveling wave signals by using a convolutional neural network and a long-term and short-term memory network structure;

an output layer for generating denoised traveling wave signal data

And step S24, measuring the difference between the denoised traveling wave signal data and the target noiseless signal through a loss function.

Further, the formula of the loss function in step S24 is as follows:

Where MSE is the mean square error, y _i is the target noise-free signal, Is the denoised signal.

Further, the conversion in step S32 specifically includes the steps of:

Step S321, the amplitude of the traveling wave signal is mapped into gray values or color pixel intensities by using a normalization mode, and the formula is as follows:

I(x,y)=A(t_i)·255,

Wherein a (t _i) represents the amplitude value at time t _i and I (x, y) represents the pixel value in the fault waveform picture;

And S322, mapping the time axis information to the horizontal axis of the fault waveform picture.

Further, step S03 includes the steps of:

Step S33, formatting the fault waveform picture I, wherein the formatting steps are as follows:

step S331, adjusting the picture to a fixed size required by the model;

Step S332, adjusting the boundary of the fault waveform pictures according to the requirement so that all the input fault waveform pictures have the same size.

Further, the feature fusion in step S05 specifically includes the following steps:

step S51, extracting high-dimensional features from the fault waveform picture by using a convolutional neural network, wherein F _img is set to represent feature vectors extracted from the fault waveform picture, and the formula is as follows:

F_img＝CNN_img(waveform_image),

step S52, extracting electrical data characteristics from electrical data by using a full connection layer or other neural network structures, wherein F _elec is set to represent a characteristic vector extracted from the electrical data, and the formula is as follows:

F_elec＝FC_elec(electrical_data),

Step S53, performing fusion operation on the F _img and the F _elec to obtain a fused feature vector F _combined, wherein the formula is as follows:

F_combined＝concat(F_img,F_elec)。

Further, the electrical data features extracted in step S52 include voltage and current.

Further, the output failure diagnosis result in step S06 is specifically:

based on the fused feature vector, outputting fault types, fault positions and diagnostic reports by using a pre-trained and fine-tuned YOLOv model;

The YOLOv model structure includes:

an input layer for receiving the fused feature vector F _combined;

a convolution layer for extracting spatial features and patterns;

And the detection layer is used for outputting fault types, fault positions and diagnostic reports by using a detection mechanism of YOLO.

Drawings

FIG. 1 is an overall flow chart of the present invention;

fig. 2 is a waveform diagram of the identification traveling wave head of the present invention.

Detailed Description

The principles and features of the present invention are described below with reference to the drawings, the examples are illustrated for the purpose of illustrating the invention and are not to be construed as limiting the scope of the invention.

As shown in fig. 1-2, in embodiment 1, a fault diagnosis method based on YOLOv dual-branch power grid fault waveform analysis includes the following steps:

and step S01, capturing traveling wave signal data within 1 second by using high-precision equipment, and ensuring that a sufficiently detailed and accurate traveling wave signal can be obtained for subsequent analysis and processing. In specific implementation, the waveform data is directly obtained by calling a database API.

And step S02, constructing a neural network architecture, denoising traveling wave signal data, wherein compared with the traditional filtering operation, the neural network denoising technology can effectively reduce noise influence and improve the definition and accuracy of the data.

Step S03, converting the denoised traveling wave signal data into a fault waveform picture;

Step S04, detecting a traveling wave head and a fault point in traveling wave signal data by using a first branch of a YOLOv model, wherein the first branch of the YOLOv model can quickly and efficiently identify the position of the traveling wave head, and ensure timeliness of fault detection;

And step S05, when the traveling wave head is detected in the step S04, the fault waveform picture and the electrical data characteristic before and after the traveling wave head moment are subjected to characteristic fusion, the second branch is combined with various characteristics to automatically identify and classify fault types, and through the characteristic fusion, the model can accurately capture various fault characteristics in a complex environment, so that the intelligent level of fault diagnosis is improved.

And step S06, inputting the fused feature vector into a second branch of the YOLOv model, and outputting a fault diagnosis result. The method can be used for maintenance and management of a power grid system and provides a timely fault treatment scheme.

The fault diagnosis method based on the traveling wave signal transfer picture and YOLOv double-branch target detection has obvious advantages. Firstly, key features in waveform signals can be accurately extracted by utilizing a neural network denoising technology, so that the influence of noise is greatly reduced, and the accuracy and stability of fault diagnosis are improved. And secondly, the double-branch YOLOv model is adopted to automatically identify the fault moment and fault type, so that the method is rapid and efficient, various fault characteristics can be accurately captured in a complex environment, and the intelligent level of system fault diagnosis is improved. The method overcomes the limitations of the traditional technology in efficiency, precision and application range, and provides an innovative and reliable solution for the safe operation and maintenance of the power grid system.

Example 2, which is a further improvement over example 1, is as follows:

the step S02 specifically includes the following steps:

and S21, collecting traveling wave signal data containing noise and corresponding noiseless or low-noise traveling wave signal data, wherein the data can be generated through operation record or simulation of an actual power system.

And step S22, marking the collected traveling wave signal data, wherein the marking comprises a noise signal and a target noise-free signal, and the marking data is used as the basis of training the neural network.

Step S23, the neural network architecture comprises:

An input layer for receiving original traveling wave signal data x= { x ₁,x₂,...,x_T }, wherein x _t is the t-th data point in the time series data;

the hidden layer is used for extracting the characteristics in the traveling wave signal by using a Convolutional Neural Network (CNN) and a long-short-term memory network (LSTM) structure, wherein the convolutional neural network is good at processing local characteristics and is suitable for extracting the detailed characteristics of the waveform signal, the output characteristic diagram is F _cnn＝f_cnn (X), the long-short-term memory network is good at processing sequence data and is suitable for capturing the time correlation in the waveform signal, and the hidden state sequence is H _lstm＝f_lstm (X);

an output layer for generating denoised traveling wave signal data To ensure the quality of the output signal, a full connection layer can be usedWherein σ is the activation function and W _fc and b _fc are the weights and biases of the full connection layer, respectivelyIs the F _deconv deconvolution operation, and F _hid is a feature representation of the hidden layer.

And step S24, measuring the difference between the denoised traveling wave signal data and the target noiseless signal through a loss function.

In the process of diagnosing a power grid fault, traveling wave signals often suffer from interference of various noises, including environmental noise, electromagnetic interference, equipment noise and the like. Conventional denoising methods such as filters, while capable of reducing noise to some extent, tend to result in signal distortion or loss of important features when processing complex waveform signals. For this reason, we have introduced neural network denoising techniques to achieve more accurate and efficient denoising of traveling wave signals.

Example 3, which is a further improvement over example 2, is specifically as follows:

The formula of the loss function in step S24 is as follows:

Where MSE is the mean square error, y _i is the target noise-free signal, Is the denoised signal. The mean square error can amplify larger errors, and is more suitable for scenes needing accurate denoising.

Example 4, this example is a further improvement over example 1, which is specifically as follows:

the step S03 specifically includes the following steps:

Step S31, recording a traveling wave signal x (T) as time sequence data x (T ₁),x(t₂),...,x(t_N), dividing the long-time traveling wave signal into time windows T with fixed lengths, wherein each time window T contains traveling wave signal data from T _i to T _i+T-1;

and S32, converting the traveling wave signals in each time window T into a fault waveform picture I.

Example 5, which is a further improvement over example 4, is specifically as follows:

the conversion in step S32 specifically includes the steps of:

Step S321, the amplitude of the traveling wave signal is mapped into gray values or color pixel intensities by using a normalization mode, and the formula is as follows:

I(x,y)=A(t_i)·255,

Wherein a (t _i) represents the amplitude value at time t _i and I (x, y) represents the pixel value in the fault waveform picture;

And S322, mapping the time axis information to the horizontal axis of the fault waveform picture. The sequence and the change of the time sequence data can be accurately reflected in the fault waveform picture. For example, if the time window length is t=100, the traveling wave signal in each time window will be mapped into a fault waveform picture of size 100×w, where W is the pixel width, which can be adjusted according to the specific situation.

Example 6, which is a further improvement over example 4, is specifically as follows:

Step S03 further includes the steps of:

step S33, formatting the fault waveform picture I to adapt to the input requirement of a neural network model and optimize the image processing effect, wherein the formatting steps are as follows:

Step S331, adjusting the picture to a fixed size required by a model, wherein H is the pixel height, and W is the pixel width of the rectangular fault waveform picture which is usually H×W;

step S332, adjusting the boundary of the fault waveform pictures according to the requirement so that all the input fault waveform pictures have the same size. Simplifying the processing of the neural network.

Example 7, which is a further improvement over example 1, is as follows:

The feature fusion in step S05 specifically includes the following steps:

Step S51, extracting high-dimensional features from the fault waveform picture by using a Convolutional Neural Network (CNN), wherein F _img is set to represent feature vectors extracted from the fault waveform picture, and the formula is as follows:

F_img＝CNN_img(waveform_image),

step S52, extracting electrical data characteristics from electrical data by using a full connection layer or other neural network structures, wherein F _elec is set to represent a characteristic vector extracted from the electrical data, and the formula is as follows:

F_elec＝FC_elec(electrical_data),

Step S53, performing fusion operation on the F _img and the F _elec to obtain a comprehensive feature vector F _combined, wherein the formula is as follows:

F_combined＝concat(F_img,F_elec)。

The process of feature fusion may be performed by combining data from different sources to form a comprehensive input feature vector.

Example 8, which is a further improvement over example 7, is specifically as follows:

the electrical data features extracted in step S52 include voltage and current.

Example 9, which is a further improvement over example 1, is as follows:

the output fault diagnosis result in step S06 is specifically:

on the basis of the fused feature vector, the pre-trained and fine-tuned YOLOv model is used for outputting fault types, fault positions and diagnostic reports, and the YOLOv model has the advantages of rapid and efficient target detection capability and capability of accurately identifying faults of different types in a complex environment.

The YOLOv model structure includes:

an input layer for receiving the fused feature vector F _combined;

a convolution layer for extracting spatial features and patterns;

And the detection layer is used for outputting fault types, fault positions and diagnostic reports by using a detection mechanism of YOLO.

While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.