CN118503843A - Substation equipment partial discharge monitoring interference assessment method, medium and system - Google Patents

Abstract

The invention provides a method, medium and system for evaluating partial discharge monitoring interference of power transformation equipment, belonging to the technical field of partial discharge of power transformation equipment, comprising the following steps: acquiring operation parameters of the power transformation equipment in real time, and respectively recording the operation parameters as a current curve, a voltage curve and a frequency curve; the method comprises the steps of collecting ultrasonic signals in the power transformation equipment, which are collected by an ultrasonic array sensor arranged in the power transformation equipment, in real time; calculating the occurrence position and occurrence time of the ultrasonic signal; extracting characteristics of ultrasonic signals to obtain ultrasonic characteristics; calculating the difference value between the characteristic root and the average characteristic root of each cluster center to be used as the partial discharge signal deviation degree; judging whether each clustering center corresponds to harmful partial discharge or harmless partial discharge by using a pre-trained harmful and harmless partial discharge judging model to obtain a partial discharge hazard class corresponding to the clustering center; and (3) inputting the partial discharge signal deviation degree and the partial discharge hazard category corresponding to each cluster center by adopting a pre-trained partial discharge monitoring interference evaluation model, and obtaining and outputting the partial discharge monitoring interference degree.

Description

A method, medium and system for partial discharge monitoring interference assessment of substation equipment

Technical Field

The present invention belongs to the technical field of partial discharge of substation equipment, and in particular, relates to a method, medium and system for monitoring interference assessment of partial discharge of substation equipment.

Background Art

As a key component of the power system, the operating status of substation equipment is directly related to the safety and stability of the entire power grid. During the operation of substation equipment, partial discharge is the most important insulation defect, which will gradually deteriorate the insulation performance of the equipment and even cause equipment failure and accidents. Therefore, real-time monitoring and analysis of partial discharge signals inside substation equipment is of great significance for preventing and reducing the risk of equipment failure and maintaining the safe operation of the power grid.

The existing partial discharge monitoring technology for substation equipment mainly adopts ultrasonic detection method: by installing a highly sensitive ultrasonic sensor inside the substation equipment, the ultrasonic signal generated inside the equipment is detected, thereby diagnosing partial discharge. The acoustic wave method has a high accuracy in determining the location of partial discharge, but it is sensitive to background noise, especially when other ultrasonic interference is generated inside the substation equipment, which will affect the accuracy of the ultrasonic detection method. Therefore, it is necessary to conduct interference assessment of the ultrasonic signal in advance to avoid other ultrasonic interference affecting the accuracy of the ultrasonic detection method. There is currently no relevant research on the interference assessment of ultrasonic signals in advance.

Summary of the invention

In view of this, the present invention provides a method, medium and system for evaluating interference in partial discharge monitoring of substation equipment, which can solve the technical problem that it is difficult to evaluate other ultrasonic interferences existing in the process of partial discharge ultrasonic monitoring of substation equipment in the prior art.

The present invention is achieved in that:

A first aspect of the present invention provides a method for evaluating partial discharge monitoring interference of substation equipment, which comprises the following steps:

S10, obtaining the operating parameters of the substation equipment in real time, including current, voltage, and frequency; and establishing a curve of each of the operating parameters changing over time, recorded as a current curve, a voltage curve, and a frequency curve respectively;

S20, collecting in real time the ultrasonic signal inside the substation equipment collected by the ultrasonic array sensor disposed inside the substation equipment;

S30, screening abnormal points of the current curve, voltage curve and frequency curve;

S40, calculating the occurrence position and occurrence time of the ultrasonic signal based on the ultrasonic signal collected by the ultrasonic array sensor;

S50, performing time alignment on the acquired ultrasonic signal and the abnormal point, retaining the ultrasonic signal segment at the same occurrence time as the abnormal point in the ultrasonic signal, extracting features of the ultrasonic signal, and obtaining ultrasonic features;

S60, clustering the ultrasonic features using a pre-trained deep clustering algorithm model to obtain a plurality of cluster centers, each cluster center representing a different type of partial discharge signal;

S70, calculating the characteristic root of each cluster center, averaging the characteristic root to obtain the average characteristic root, and calculating the difference between the characteristic root of each cluster center and the average characteristic root as the partial discharge signal deviation degree;

S80, using a pre-trained harmful or harmless partial discharge judgment model to judge whether each cluster center corresponds to harmful partial discharge or harmless partial discharge, and obtaining a partial discharge hazard category corresponding to the cluster center;

S90, using a pre-trained partial discharge monitoring interference assessment model, inputting the partial discharge signal deviation degree and partial discharge hazard category corresponding to each cluster center, obtaining the partial discharge monitoring interference degree of the substation equipment, and outputting it.

On the basis of the above technical solution, a method for evaluating partial discharge monitoring interference of substation equipment of the present invention can also be improved as follows:

The method for screening abnormal points is to use a curve analysis algorithm to identify abnormal fluctuation points, namely, abnormal points, in the current curve, voltage curve and frequency curve.

The specific steps of S10 include: Step 1, real-time collection of key operating parameters such as current, voltage and frequency of the substation equipment; Step 2, for each operating parameter, establish a curve of its change over time, recorded as current curve, voltage curve and frequency curve respectively; Step 3, use the curve analysis algorithm to identify abnormal fluctuation points in the current curve, voltage curve and frequency curve, i.e. abnormal points. These abnormal points may indicate that the equipment has partial discharge or other abnormal conditions, which require further analysis. The specific steps of S20 include: Step 1, deploy an ultrasonic array sensor inside the substation equipment; Step 2, collect ultrasonic signals inside the equipment in real time.

The step S40 specifically comprises: using the collected ultrasonic signal, combined with the position information of the array sensor, and by the principle of the time difference of sound wave propagation, calculating the specific position and time of the occurrence of the partial discharge event.

The step S50 is specifically: aligning the ultrasonic signal moment calculated in step S40 with the abnormal point screened out in step S30, retaining the ultrasonic signal segment at the same moment as the abnormal point and performing feature extraction to obtain ultrasonic feature parameters reflecting the local discharge characteristics.

The deep clustering algorithm model adopts an unsupervised learning method and is pre-trained through a large amount of partial discharge feature data of substation equipment to automatically identify clustering centers of different types of partial discharge signals.

Specifically, the model uses an unsupervised learning method, pre-trained with a large amount of partial discharge feature data of substation equipment, and learns cluster centers that can automatically identify different types of partial discharge signals. Classic clustering algorithms such as k-means and DBSCAN are used during training, and high-order features are extracted by combining deep neural networks, ultimately forming a deep clustering model that can accurately classify partial discharge signal types.

The specific implementation process of S60 is: using a pre-trained deep clustering algorithm model to perform cluster analysis on the ultrasonic features extracted in step S50. The deep clustering algorithm model can automatically identify different types of partial discharge signals contained in the ultrasonic features and obtain multiple cluster centers, each cluster center representing a different type of partial discharge signal.

The specific steps of S70 include: step 1, for each cluster center, calculating its characteristic root; step 2, calculating the average characteristic root; step 3, calculating the difference between the characteristic root of each cluster center and the average characteristic root as the deviation degree of the partial discharge signal corresponding to the cluster center. The larger the deviation degree, the greater the deviation of the partial discharge signal of this type from the normal signal, and the greater the harm to the equipment.

The harmful and harmless partial discharge judgment model adopts a supervised learning method and is pre-trained through a large number of labeled harmful and harmless partial discharge signal samples to accurately judge the classification of the harmfulness of partial discharge signals.

Specifically, the model uses a supervised learning method, and is pre-trained with a large number of labeled harmful and harmless partial discharge signal samples to learn a classifier that can accurately judge the harmfulness of partial discharge signals. During training, classic classification algorithms such as support vector machines and random forests are used, and deep learning is combined to extract high-order discriminant features of partial discharge signals, ultimately forming a robust harmful and harmless partial discharge judgment model.

The specific implementation process of S80 is: using a pre-trained harmful or harmless partial discharge judgment model, inputting the partial discharge signal deviation calculated in step S70, and outputting whether the type of partial discharge signal belongs to the harmful category. This provides an important basis for subsequent interference evaluation.

The partial discharge monitoring interference assessment model adopts a supervised learning method and is pre-trained through a large amount of actual operation data of substation equipment to accurately assess the degree of partial discharge monitoring interference.

Specifically, the model uses a supervised learning method and is pre-trained with a large amount of actual operation data of substation equipment to learn a regressor that can accurately evaluate the degree of partial discharge monitoring interference. The random forest algorithm is mainly used during training, with input features such as partial discharge signal deviation and hazard category, and the overall partial discharge monitoring interference of substation equipment is output, which can provide a reference for equipment status diagnosis.

The specific steps of S90 include: step 1, using a pre-trained partial discharge monitoring interference assessment model, inputting the partial discharge signal deviation calculated in step S70 and the partial discharge hazard category determined in step S80; step 2, the model outputs the partial discharge monitoring interference degree of the substation equipment. The model is trained using a random forest algorithm.

Furthermore, the outlier screening adopts a t-SNE anomaly detection algorithm based on a Gaussian mixture model.

A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for partial discharge monitoring interference assessment of substation equipment.

A third aspect of the present invention provides a system for partial discharge monitoring and interference assessment of substation equipment, which includes the above-mentioned computer-readable storage medium.

Compared with the prior art, the beneficial effects of the method, medium and system for partial discharge monitoring interference assessment of substation equipment provided by the present invention are:

1. Multi-source monitoring data fusion significantly improves monitoring accuracy. This method not only collects the current, voltage, frequency and other operating parameters of the substation equipment in real time, but also collects the ultrasonic signals inside the equipment. Through the comprehensive analysis of these multi-source data, the occurrence and characteristics of partial discharge signals can be more accurately identified, which greatly improves the accuracy and reliability of monitoring compared to the existing single monitoring method.

2. Application of deep learning algorithms to achieve automatic classification and evaluation. This method uses a pre-trained deep clustering algorithm, which can automatically identify different types of partial discharge signals contained in ultrasonic signals, laying the foundation for subsequent hazard assessment. At the same time, the supervised learning method is also used to train a harmful and harmless partial discharge judgment model and a partial discharge monitoring interference assessment model, realizing an intelligent assessment of the degree of hazard of partial discharge signals and the overall impact on the equipment. Compared with the traditional empirical judgment method, this automatic evaluation method based on machine learning is more objective and accurate.

3. Comprehensively evaluate the partial discharge monitoring interference and provide a reference for equipment status management. Existing partial discharge monitoring technology often only focuses on the partial discharge signal itself and cannot comprehensively evaluate its impact on the overall operation of the equipment. The method proposed in the present invention calculates the deviation and harm degree of the partial discharge signal, and finally obtains the overall partial discharge monitoring interference degree of the substation equipment, which provides an important basis for equipment status diagnosis and fault warning. This is of great significance for improving the operation and maintenance level of power equipment.

In general, the method for evaluating interference in partial discharge monitoring of substation equipment proposed in the present invention makes full use of multi-source monitoring data and combines it with advanced machine learning algorithms to achieve comprehensive analysis and evaluation of partial discharge signals, thus solving the technical problem that the prior art is difficult to evaluate other ultrasonic interferences that exist in the process of ultrasonic monitoring of partial discharge of substation equipment.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

FIG1 is a flow chart of a method provided by the present invention;

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

As shown in FIG1 , it is a flow chart of a method for evaluating partial discharge monitoring interference of substation equipment provided by the present invention. The method comprises the following steps:

S10, obtaining the operating parameters of the substation equipment in real time, including current, voltage, and frequency; and establishing a curve of each of the operating parameters changing over time, recorded as a current curve, a voltage curve, and a frequency curve respectively;

S20, collecting in real time the ultrasonic signal inside the substation equipment collected by the ultrasonic array sensor disposed inside the substation equipment;

S30, screening abnormal points of the current curve, voltage curve and frequency curve;

S40, calculating the occurrence position and occurrence time of the ultrasonic signal based on the ultrasonic signal collected by the ultrasonic array sensor;

S50, performing time alignment on the acquired ultrasonic signal and the abnormal point, retaining the ultrasonic signal segment at the same time as the abnormal point in the ultrasonic signal, extracting the features of the ultrasonic signal, and obtaining the ultrasonic features;

S60, clustering the ultrasonic features using a pre-trained deep clustering algorithm model to obtain a plurality of cluster centers, each cluster center representing a different type of partial discharge signal;

S70, calculating the characteristic root of each cluster center, averaging the characteristic root to obtain the average characteristic root, and calculating the difference between the characteristic root of each cluster center and the average characteristic root as the partial discharge signal deviation degree;

S80, using a pre-trained harmful or harmless partial discharge judgment model to judge whether each cluster center corresponds to harmful partial discharge or harmless partial discharge, and obtaining a partial discharge hazard category corresponding to the cluster center;

S90, using a pre-trained partial discharge monitoring interference assessment model, inputting the partial discharge signal deviation degree and partial discharge hazard category corresponding to each cluster center, obtaining the partial discharge monitoring interference degree of the substation equipment, and outputting it.

The specific implementation methods of the above steps are described in detail below:

Step S10: Real-time acquisition of the operating parameters of the substation. Specifically, this step first sets up a collection module, such as various sensors, to collect key operating parameters of the substation such as current, voltage and frequency in real time. Then, for each operating parameter, a curve of its change over time is established, which is recorded as a current curve, a voltage curve and a frequency curve. These curves can reflect the real-time working status of the substation and lay the foundation for the subsequent screening of abnormal points.

Step S20: Real-time collection of ultrasonic signals inside the substation equipment. In this step, an ultrasonic array sensor is deployed inside the substation equipment to collect ultrasonic signals inside the equipment in real time. Ultrasonic signals can reflect the partial discharge situation inside the equipment and provide data support for subsequent partial discharge analysis.

Step S30: Screening abnormal points of the operating parameter curves. By analyzing the current curve, voltage curve and frequency curve, the abnormal fluctuation points in the curves, i.e., abnormal points, are identified. These abnormal points may indicate partial discharge or other abnormal conditions in the equipment, which require further analysis.

Step S40: Calculate the location and time of the ultrasonic signal. By using the collected ultrasonic signal, combined with the position information of the array sensor, and through the principle of sound wave propagation time difference, the specific location and time of the partial discharge event can be calculated. This lays the foundation for the subsequent time alignment of the ultrasonic signal with the abnormal point.

Step S50: Extract the features of the ultrasonic signal. First, align the ultrasonic signal moment calculated in step S40 with the abnormal point screened out in step S30, and retain the ultrasonic signal segment at the same moment as the abnormal point. Then, extract the features of these ultrasonic signal segments to obtain ultrasonic feature parameters reflecting the characteristics of partial discharge. These feature parameters lay the foundation for the subsequent cluster analysis.

Step S60: clustering the ultrasonic features using a deep clustering algorithm. A pre-trained deep clustering algorithm model is used here, which can automatically identify different types of partial discharge signals contained in the ultrasonic features. Through cluster analysis, multiple cluster centers can be obtained, each cluster center represents a different type of partial discharge signal.

Step S70: Calculate the deviation of the partial discharge signal. For each cluster center, first calculate its characteristic root, then find the average characteristic root, and then calculate the difference between the characteristic root of each cluster center and the average characteristic root as the deviation of the partial discharge signal corresponding to the cluster center. The larger the deviation, the greater the deviation of the partial discharge signal of this type from the normal signal, and the greater the harm to the equipment.

Step S80: Determine the harmfulness category of the partial discharge signal. A pre-trained harmful or harmless partial discharge judgment model is used here. The partial discharge signal deviation calculated in step S70 is input and the output is whether the type of partial discharge signal belongs to the harmful category. This provides an important basis for subsequent interference assessment.

Step S90: Evaluate the interference degree of partial discharge monitoring. A pre-trained partial discharge monitoring interference evaluation model is used here. The partial discharge signal deviation calculated in step S70 and the partial discharge hazard category determined in step S80 are input, and the partial discharge monitoring interference degree of the substation equipment is output. The model is trained using the random forest algorithm. Through this step, the overall partial discharge monitoring interference of the equipment can be obtained, providing a reference for further status monitoring and early warning.

In order to better understand this method, the specific implementation of this method is described in more detail below in combination with a specific formula:

Step S10: Obtaining the operating parameters of the substation equipment in real time

The purpose of this step is to collect the key operating parameters of the substation equipment in real time, including current, voltage and frequency, and to establish the curves of these parameters changing with time.

First, a group of sensor modules are set up to collect the operating parameters of the substation equipment in real time, such as current I(t), voltage U(t) and frequency f(t), where t represents time. For each operating parameter, a curve fitting method is used to establish its time-varying curve, which is recorded as current curve v(t), voltage curve U(t) and frequency curve f(t).

The current curve can be expressed as:

I(t)＝a ₁ +b ₁ t+c ₁ t ² +d ₁ t ³ +…+k ₁ t ⁿ

The voltage curve can be expressed as:

U(t)＝a ₂ +b ₂ t+c ₂ t ² +d ₂ t ³ +…+k ₂ t ⁿ

The frequency curve can be expressed as:

f(t)＝a ₃ +b ₃ t+c ₃ t ² +d ₃ t ³ +…+k ₃ t ⁿ

Wherein, a _i , b _i , c _i , ..., k _i (i=1, 2, 3) are coefficients of curve fitting, and n is the order of fitting, which is determined according to actual conditions.

These curves can reflect the real-time working status of substation equipment and lay the foundation for the subsequent screening of abnormal points.

Step S20: Real-time collection of ultrasonic signals inside the transformer

The purpose of this step is to collect ultrasonic signals inside the substation equipment in real time to provide data support for subsequent partial discharge analysis.

An ultrasonic array sensor is deployed inside the substation to collect the ultrasonic signal S(t, x, y, z) generated inside the equipment in real time, where t represents time, and x, y, and z represent the coordinate positions of the ultrasonic sensor in three-dimensional space.

The acquisition process of ultrasonic signals can be expressed as:

S(t,x,y,z)=f(I(t),U(t),f(t),t,x,y,z)+∈

Among them, f(·) is an unknown function that describes the relationship between the ultrasonic signal and the operating parameters of the substation equipment and the sensor location, and ∈ is the noise term.

By analyzing the collected ultrasonic signals, relevant information about partial discharge inside the substation equipment can be obtained, laying the foundation for subsequent abnormal point identification and location calculation.

Step S30: Screening outliers in the operating parameter curve

The purpose of this step is to identify abnormal fluctuation points in the current curve, voltage curve and frequency curve, providing a basis for subsequent time alignment and feature extraction.

The three curves are analyzed by an anomaly detection algorithm to identify the abnormal points in the curves. Specifically, an anomaly detection method based on a statistical model can be used, such as an anomaly detection algorithm based on a Gaussian mixture model (GMM).

The basic idea of the algorithm is as follows:

1. Use GMM to model the current curve I(t), voltage curve U(t) and frequency curve f(t) respectively, and obtain the Gaussian mixture probability density function of each curve:

Where K, L, and M are the number of components of the Gaussian mixture model of the current, voltage, and frequency curves, respectively; _πi , _πj , and _πk are the mixing coefficients of each component; and _μi , _μj , _μk , and are the mean and variance of each component respectively.

2. Calculate the log-likelihood of the curve value at each time point t:

l _I (t) = log p (I (t))

l _U (t) = log p (U (t))

l _f (t) = log p (f (t))

3. Set the abnormal thresholds τ _I , τ _U , τ _f of the log-likelihood value. If the log-likelihood value at a certain time t is less than the corresponding threshold, the curve value at that time is considered to be an abnormal point:

For abnormal points

For abnormal points

For abnormal points

Through this method, abnormal points in the current curve, voltage curve and frequency curve can be effectively screened out, providing an important basis for subsequent partial discharge analysis.

Step S40: Calculate the location and time of occurrence of the ultrasonic signal

The purpose of this step is to calculate the specific location and time of the partial discharge event based on the collected ultrasonic signal and the principle of sound wave propagation.

Using the signal S (t, x, y, z) collected by the ultrasonic array sensor and the known position information ( _xi , _yi , _zi ), i = 1, 2, ..., N, the position coordinates ( _xe , _ye , _ze ) of the partial discharge event and the time _te of occurrence are calculated through the principle of sound wave propagation time difference.

The specific steps are as follows:

1. For the ultrasonic signal S _i (t) = S (t, x _i , y _i , z _i ) received by the i-th sensor, the arrival time t _i of the signal can be extracted by a signal processing algorithm (such as peak detection).

2. Assuming that the partial discharge event occurs at position (x _e , _ye , _ze ), the propagation time of the signal received by the i-th sensor can be expressed as:

Where v is the propagation speed of ultrasound in the medium.

3. Substituting the receiving time _ti of all sensors into the above formula, a nonlinear equation system can be established:

4. By solving the equations using a numerical iteration method (such as Newton's method), the coordinates of the location where the partial discharge event occurs (x _e , _ye , _ze ) can be obtained.

5. According to the receiving time _ti of any sensor and the corresponding propagation time _Δti , the time when the partial discharge event occurs can be calculated as _te = _ti - _Δti .

Through the above steps, the location and time of the local discharge event can be accurately calculated, providing basic data for subsequent time alignment and feature extraction.

Step S50: Extracting features of ultrasonic signals

The purpose of this step is to extract the characteristic parameters of the ultrasonic signal that reflect the characteristics of local discharge, laying the foundation for subsequent cluster analysis.

First, the local discharge event occurrence time _te calculated in step S40 is time-aligned with the abnormal point screened out in step S30, and the ultrasonic signal segment _Se (t)=S(t, _xe , _ye , _ze ) at the same time as the abnormal point is retained.

For these ultrasonic signal segments, the following characteristic parameters are extracted:

1. Time domain characteristics:

The mean value of the signal amplitude and variance

The peak value Se _,peak , the peak factor Se _,crest and the impulse factor Se _,impulse of the signal waveform

2. Frequency domain characteristics:

The signal's dominant frequency _{FE, peak} and bandwidth _ΔFE

Entropy of Power Spectral Density (PSD) and power concentration factor Pe _,conc

3. Time-frequency characteristics:

Energy distribution of wavelet transform

Hilbert-Huang transform instantaneous frequency fe _,inst (t) and instantaneous power Pe _,inst (t)

The above characteristic parameters can comprehensively reflect the local discharge characteristics contained in the ultrasonic signal and provide a basis for subsequent cluster analysis.

Step S60: Clustering the ultrasound features using a deep clustering algorithm

The purpose of this step is to use a deep learning clustering algorithm to automatically classify the ultrasonic signal features extracted in step S50 to obtain different types of local discharge signals.

A pre-trained deep clustering algorithm model is used. This model is based on the structure of a deep neural network and can automatically extract high-order representations of ultrasonic signal features and identify different types of partial discharge signals using an unsupervised clustering method.

The specific algorithm flow is as follows:

1. Take the ultrasonic signal feature vector _xi =[ _xi,1 , _xi,2 ,..., _xi,d ] ^T extracted in step S50 as input, where d is the feature dimension.

2. Construct a deep autoencoder (DAE) network with input x _i . After multiple layers of encoding and decoding, a low-dimensional feature representation z _i = [z _{i, 1} , z _{i, 2} , ... , z _{i, k} ] ^T is obtained, where k << d.

3. Input the low-dimensional feature representations Z = [z ₁ , z ₂ , ..., z _N ] ^T of all samples into a deep clustering network, which includes:

A deep neural network is used to further extract the high-order features of Z h = f(Z)

An end-to-end clustering layer uses soft clustering to output the probability that each sample belongs to C cluster centers _qi = [qi _{, 1} , qi _{, 2} , ..., qi _{, C} ] ^T

4. By alternately optimizing the deep neural network parameters and cluster centers, we finally converge to a stable clustering result and obtain C cluster centers.

This deep clustering algorithm can automatically mine different types of partial discharge signals contained in the ultrasonic signal characteristics, providing a basis for subsequent deviation calculation and hazard judgment.

Step S70: Calculate the deviation of the partial discharge signal

The purpose of this step is to calculate the degree of deviation between the partial discharge signal represented by each cluster center and the normal signal based on the clustering result obtained in step S60, as a basis for evaluating its harmfulness.

For the c-th cluster center μ _c =[μ _c,1 ,μ _c,2 , ...,μ _c,k ] ^T , firstly calculate its eigenvalues λ _c,1 ,λ _c,2 , ...,λ _c,k , representing the projection amplitude of the eigenvector in the direction of each principal component.

Then, find the average value of all cluster center characteristic roots As the characteristic root reference of normal partial discharge signal.

Finally, the difference between the characteristic root of each cluster center and the average characteristic root is calculated as the deviation degree D _c of the partial discharge signal corresponding to the cluster center:

The larger the deviation D _{c is} , the greater the deviation of this type of partial discharge signal from the normal signal is, and the greater the harm to the equipment may be. This indicator will provide an important basis for subsequent hazard judgment and interference assessment.

Step S80: Determine the hazard category of the partial discharge signal

The purpose of this step is to use the pre-trained harmful or harmless partial discharge judgment model to make a harmfulness judgment on the partial discharge signal represented by each cluster center calculated in step S70.

A harmful and harmless partial discharge judgment model based on supervised learning is adopted. The model is pre-trained with a large number of labeled harmful and harmless partial discharge signal samples to learn a classifier that can accurately judge the harmfulness of partial discharge signals.

Specifically, the input of the model is the partial discharge signal deviation D _c calculated in step S70, and the output is the probability of whether the type of partial discharge signal belongs to the harmful category. The training loss function of the model can be defined as:

Among them, θ is the model parameter, is the label of the PD signal corresponding to the c-th cluster center, which is 1 if it is a harmful category, otherwise it is 0.

Through training, the model can learn the mapping relationship between the partial discharge signal deviation and the degree of harmfulness, providing an important basis for subsequent interference assessment.

Step S90: Evaluate the interference degree of partial discharge monitoring

The purpose of this step is to use the pre-trained partial discharge monitoring interference assessment model, comprehensively consider the partial discharge signal deviation calculated in step S70 and the harmfulness determined in step S80, and output the overall partial discharge monitoring interference degree of the substation equipment.

A partial discharge monitoring interference assessment model based on supervised learning is adopted. The model is pre-trained with a large amount of actual operation data of substation equipment to learn a regressor that can accurately assess the degree of partial discharge monitoring interference.

Specifically, the inputs of the model include:

1. Deviation of partial discharge signals at each cluster center D _c , c = 1, 2, ..., C

2. The probability of harmfulness corresponding to each cluster center c＝1，2，...，C

The output of the model is the overall partial discharge monitoring interference level I _mon of the substation equipment, which can be defined as:

Among them, f(·) is a regression model trained based on the random forest algorithm, and θ is the model parameter.

The training loss function of this model can be defined as the mean squared error:

Where N is the number of training samples, is the actual PD monitoring interference degree of the i-th sample, is the predicted value of the model.

Through training, the model can accurately assess the overall partial discharge monitoring interference level of substation equipment, providing an important reference for equipment status diagnosis and fault warning.

The explanation of relevant variables is shown in the following table:

A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for partial discharge monitoring interference assessment of substation equipment.

A third aspect of the present invention provides a system for partial discharge monitoring and interference assessment of substation equipment, which includes the above-mentioned computer-readable storage medium.

Specifically, the principle of the present invention is: the method first collects the key operating parameters of the substation equipment such as current, voltage, frequency, etc. in real time, and establishes a curve of its change over time. At the same time, the ultrasonic signal inside the equipment is also collected in real time. By screening the abnormal points of these operating parameter curves, the abnormal moment when partial discharge may exist can be found.

Then, using the principle of sound wave propagation, the specific location and time of the partial discharge event are calculated based on the ultrasonic signal. Next, these ultrasonic signals are time-aligned with the abnormal points to extract parameters reflecting the characteristics of partial discharge. Using a pre-trained deep clustering algorithm, these characteristic parameters are automatically classified to obtain different types of partial discharge signals.

For each type of partial discharge signal, the deviation between its characteristic root and the normal signal is calculated as the basis for evaluating its degree of harm. At the same time, the pre-trained harmful and harmless partial discharge judgment model is used to determine whether the type of partial discharge signal belongs to the harmful category. Finally, considering the deviation and harm degree of various types of partial discharge signals, the pre-trained partial discharge monitoring interference assessment model is used to obtain the overall partial discharge monitoring interference degree of the substation equipment.

This evaluation method based on multi-source monitoring data fusion and deep learning algorithm application has the following advantages over existing technologies:

1. Comprehensive monitoring data and high accuracy in identifying partial discharge signals. By synchronously collecting equipment operating parameters and internal ultrasonic signals, combined with the screening and positioning of abnormal points, the occurrence of partial discharge can be more comprehensively and accurately perceived, laying a good foundation for subsequent classification and evaluation.

2. Automatic classification and evaluation greatly improve efficiency. The deep clustering algorithm and supervised learning model are used to realize the automatic identification of partial discharge signal types and the intelligent evaluation of their degree of harm and monitoring interference, without relying on experience judgment, which greatly improves the analysis efficiency.

3. Comprehensively evaluate the impact of partial discharge and provide a reference for status management. The existing technology usually only focuses on the partial discharge signal itself, while the method of the present invention comprehensively evaluates the impact of partial discharge on equipment operation by calculating the overall partial discharge monitoring interference of the equipment, providing a more valuable basis for subsequent status diagnosis and fault warning.

The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

The following is a specific embodiment of the present invention, in which a 110kV substation was selected as the test object, and real-time monitoring and analysis was performed on it for a period of 3 months. The main transformer capacity of the substation is 100MVA. It has been in operation for about 10 years and the equipment is generally in good condition. However, two faults caused by partial discharge have occurred in the past year.

Step S10: Obtaining operating parameters in real time

Current, voltage and frequency sensors are deployed inside the substation to collect key operating parameters such as the three-phase current I _a (t), three-phase voltage U _a (t) and frequency f (t) of the main transformer in real time. The sampling frequency is 1kHz and the collection time is 3 months.

For the current signal I _a (t), Gaussian process regression is used to perform curve fitting, and the fitting curve is obtained. From the fitting results, we can see It can better reflect the changing trend of the actual current signal. Similarly, the voltage signal U _a (t) and the frequency signal f (t) are also curve fitted to obtain and

Step S20: Collecting ultrasonic signals

Eight PVDF piezoelectric ceramic ultrasonic sensors are arranged at key locations inside the substation main transformer, such as windings and bushings, to collect the ultrasonic signals S (t, x, y, z) inside the equipment in real time. These sensors have the characteristics of wide bandwidth and high sensitivity, and can fully capture the weak sound wave signals generated by partial discharge. The sampling frequency is also 1kHz, and the time length is 3 months.

Step S30: Screening for abnormal points in operating parameters

The current curve obtained by fitting Voltage curve and frequency curve Perform outlier detection. The t-SNE anomaly detection algorithm based on the Gaussian mixture model is used. This algorithm can adaptively identify abnormal fluctuation points in the curve and suppress interference caused by normal operation fluctuations.

After screening the abnormal points, a total of 24 abnormal points were detected. Representative current curve The identification results of its abnormal points. It can be seen that these abnormal points are likely to correspond to the occurrence of partial discharge events.

Step S40: Calculate the location and time of the partial discharge event

According to the collected ultrasonic signal S (t, x, y, z) and the known position information of the eight sensors, the specific position coordinates (x _e , _ye , _ze ) and time _te of the partial discharge event are calculated using the principle of sound wave propagation time difference.

In order to improve the accuracy and robustness of positioning, the nonlinear least square method based on optimization is used for solution. After calculation, the location and time of partial discharge events corresponding to the 24 abnormal points are shown in Table 1. It can be seen that most partial discharge events occur in the windings and bushings of the main transformer.

Table 1 Location and time of partial discharge events

Step S50: Extracting ultrasonic signal features

For the above 24 partial discharge events, the ultrasonic signal segments _Se (t) at the corresponding moments were extracted and their characteristics were analyzed. Specifically, they include:

1. Time domain characteristics:

Signal Amplitude Mean and variance

The peak value Se _,peak , the peak factor Se _,crest and the impulse factor Se _,impulse of the signal waveform

2. Frequency domain characteristics:

Main frequency _{fe, peak} and bandwidth _Δfe

Power Spectral Entropy and the power spectrum concentration factor Pe _,conc

3. Time-frequency characteristics:

Wavelet Packet Energy Entropy

Hilbert-Huang transform instantaneous frequency fe _,inst (t) and instantaneous power Pe _,inst (t)

By extracting these time domain, frequency domain and time-frequency domain features, the local discharge characteristics contained in the ultrasonic signal can be more comprehensively characterized.

Step S60: Classification using deep clustering algorithm

Based on the extracted ultrasonic signal features, a pre-trained deep clustering algorithm is used to automatically classify these features. Specifically, an end-to-end clustering model based on variational autoencoder (VAE) and K-means algorithm is used.

The model first uses VAE to extract high-order representations of ultrasonic signal features, and then inputs them into the K-means algorithm for cluster analysis. After training, the model finally identifies five different types of partial discharge signals. These five types of partial discharge signals have obvious differences in time-frequency characteristics.

Step S70: Calculate the partial discharge signal deviation

For the above five types of cluster centers, calculate their characteristic roots λ _c,j (c = 1, 2, ..., 5; j = 1, 2, ..., k) respectively, and find the average characteristic root Then, the Euclidean distance between the characteristic root and the average value of each cluster center is calculated as the deviation degree D _c of this type of partial discharge signal:

In order to better reflect the difference between the partial discharge signal and the normal signal, the time-frequency characteristics of the signal such as wavelet packet entropy and power spectrum entropy are also considered. After calculation, the deviation of the five types of partial discharge signals is shown in Table 2. It can be seen that the deviation of the fourth and fifth types of partial discharge signals is the largest, indicating that they are the most different from the normal signal and may cause the greatest harm to the equipment.

Table 2 Deviation of various partial discharge signals

Step S80: Determine the degree of harm caused by partial discharge signals

The pre-trained harmful and harmless partial discharge judgment model is used to input the deviation of the five types of partial discharge signals calculated in step S70, and the probability of whether they belong to the harmful category is output. The model uses support vector machine (SVM) as the base classifier and uses AdaBoost iterative algorithm for ensemble learning, which has high classification accuracy.

After model prediction, the 4th and 5th types of partial discharge signals were determined to be harmful. The values are 0.86 and 0.91 respectively, and the remaining 3 categories are judged as harmless. This is consistent with the deviation result calculated in step S70, further confirming the potential harm of the 4th and 5th category partial discharge signals to the operation of the equipment.

Step S90: Evaluate the interference level of partial discharge monitoring

Finally, the pre-trained partial discharge monitoring interference assessment model is used, and the five types of partial discharge signal deviations D _c calculated in step S70 and the probability of harmfulness determined in step S80 are input. Output the overall partial discharge monitoring interference degree I _mon of the main transformer of the substation.

The model uses GBDT as a regressor, which can effectively learn the complex nonlinear relationship between the characteristics of partial discharge signals and the monitoring interference degree. In order to enhance the ability to evaluate the overall status of the equipment, the model also considers the historical trend information of the main transformer operating parameters as an auxiliary input.

After model calculation, the partial discharge monitoring interference degree of the main transformer is obtained to be I _mon = 0.71. This indicator is at a relatively high level, indicating that the current main transformer has a serious partial discharge problem and has also caused great interference to the equipment status monitoring, which requires the operation and maintenance personnel to pay great attention to it.

Technical effect verification

Through the above embodiments, it can be seen that the method for evaluating partial discharge monitoring interference of substation equipment proposed in the present invention has obvious advantages over the prior art in terms of improving monitoring accuracy, realizing automated analysis, and comprehensively evaluating the impact of partial discharge:

1. Multi-source monitoring data fusion to accurately perceive partial discharge signals

In this embodiment, not only the operating parameters of the substation main transformer, such as current, voltage and frequency, are collected in real time, but also the ultrasonic signals inside the equipment are collected. Through the comprehensive analysis of these multi-source data, the occurrence of partial discharge inside the main transformer can be more comprehensively and accurately perceived, laying a good foundation for subsequent classification and evaluation. Compared with the existing single monitoring method, this multi-source data fusion method greatly improves the accuracy and reliability of monitoring.

2. Deep learning algorithms enable automatic classification and evaluation

In this embodiment, a deep clustering algorithm based on variational autoencoder and K-means is used to automatically identify different types of partial discharge signals contained in ultrasonic signals. At the same time, the harmful and harmless partial discharge judgment model and the partial discharge monitoring interference assessment model trained by supervised learning are used to realize the intelligent assessment of the degree of harm of partial discharge signals and the impact on equipment. Compared with traditional empirical judgment, this automatic analysis method based on machine learning is more objective and accurate, and greatly improves the analysis efficiency.

3. Comprehensively evaluate the interference effect of partial discharge on equipment

Different from the existing partial discharge monitoring technology which only focuses on the partial discharge signal itself, this embodiment calculates the deviation and harm degree of the partial discharge signal, and finally obtains the overall partial discharge monitoring interference degree of the main transformer. This indicator not only reflects the severity of the partial discharge of the main transformer, but also considers its comprehensive impact on the equipment status monitoring. It is of great significance to improve the operation and maintenance management level of power equipment, and provides strong support for subsequent status diagnosis and fault warning.

Claims (10)

1. A method for evaluating partial discharge monitoring interference of substation equipment, characterized in that it comprises the following steps: S10, obtaining the operating parameters of the substation equipment in real time, including current, voltage, and frequency; and establishing a curve of each of the operating parameters changing over time, recorded as a current curve, a voltage curve, and a frequency curve respectively; S20, collecting in real time the ultrasonic signal inside the substation equipment collected by the ultrasonic array sensor disposed inside the substation equipment; S30, screening abnormal points of the current curve, voltage curve and frequency curve; S40, calculating the occurrence position and occurrence time of the ultrasonic signal based on the ultrasonic signal collected by the ultrasonic array sensor; S50, performing time alignment between the acquired ultrasonic signal and the abnormal point, retaining the ultrasonic signal segment at the same occurrence time as the abnormal point in the ultrasonic signal, extracting features of the ultrasonic signal, and obtaining ultrasonic features; S60, clustering the ultrasonic features using a pre-trained deep clustering algorithm model to obtain a plurality of cluster centers, each cluster center representing a different type of partial discharge signal; S70, calculating the characteristic root of each cluster center, averaging the characteristic root to obtain the average characteristic root, and calculating the difference between the characteristic root of each cluster center and the average characteristic root as the partial discharge signal deviation degree; S80, using a pre-trained harmful or harmless partial discharge judgment model to judge whether each cluster center corresponds to harmful partial discharge or harmless partial discharge, and obtaining a partial discharge hazard category corresponding to the cluster center; S90, using a pre-trained partial discharge monitoring interference assessment model, inputting the partial discharge signal deviation degree and partial discharge hazard category corresponding to each cluster center, obtaining the partial discharge monitoring interference degree of the substation equipment, and outputting it. 2. According to a method for evaluating partial discharge monitoring interference of substation equipment in claim 1, it is characterized in that the method for screening abnormal points is: using a curve analysis algorithm to identify abnormal fluctuation points, i.e., abnormal points, in the current curve, voltage curve and frequency curve. 3. A method for partial discharge monitoring interference assessment of substation equipment according to claim 1 is characterized in that the step S40 specifically comprises: using the collected ultrasonic signal, combined with the position information of the ultrasonic array sensor, and through the principle of sound wave propagation time difference, calculating the specific location and the time of occurrence of the partial discharge event. 4. A method for evaluating partial discharge monitoring interference of substation equipment according to claim 1, characterized in that the step S50 specifically comprises: time-aligning the occurrence time calculated in step S40 with the abnormal point screened out in step S30, retaining the ultrasonic signal segment at the same time as the abnormal point and performing feature extraction, and obtaining the parameters of the ultrasonic feature reflecting the partial discharge characteristics. 5. According to a method for partial discharge monitoring interference assessment of substation equipment according to claim 1, it is characterized in that the deep clustering algorithm model adopts an unsupervised learning method and is pre-trained by a large amount of partial discharge feature data of substation equipment to automatically identify the clustering centers of different types of partial discharge signals. 6. A method for evaluating partial discharge monitoring interference of substation equipment according to claim 1, characterized in that the harmful and harmless partial discharge judgment model adopts a supervised learning method and is pre-trained through a large number of labeled harmful and harmless partial discharge signal samples to accurately judge the classification of the harmfulness of partial discharge signals. 7. A method for evaluating partial discharge monitoring interference of substation equipment according to claim 1, characterized in that the partial discharge monitoring interference evaluation model adopts a supervised learning method and is pre-trained through a large amount of actual operation data of substation equipment to accurately evaluate the degree of partial discharge monitoring interference. 8. A method for evaluating partial discharge monitoring interference of substation equipment according to claim 2, characterized in that the abnormal point screening adopts a t-SNE anomaly detection algorithm based on a Gaussian mixture model. 9. A computer-readable storage medium, characterized in that program instructions are stored in the computer-readable storage medium, and when the program instructions are run, they are used to execute the method for partial discharge monitoring and interference assessment of substation equipment according to any one of claims 1 to 8. 10. A partial discharge monitoring interference assessment system for substation equipment, comprising the computer-readable storage medium according to claim 9.