KR102389439B1 - RNN based partial discharge detection apparatus and method in gas-insulated switchgear - Google Patents

Abstract

A deep learning-based partial discharge diagnosis apparatus and method are disclosed. The partial discharge diagnosis method includes the steps of: receiving a partial discharge signal and generating a plurality of virtual partial discharge signals in consideration of frequency response characteristics according to distance; and applying the partial discharge signal and a plurality of virtual partial discharge signals to a deep learning model to extract and learn partial discharge characteristics.

Description

Deep learning-based partial discharge detection apparatus and method in gas-insulated switchgear

The present invention relates to a deep learning-based partial discharge diagnosis apparatus and method for a gas insulated switchgear.

Power systems are rapidly increasing in popularity as demand for power increases, and the reliability of the power grid is important for reliable power system operation. The gas insulated switchgear (GIS) applied to the substation is the main protection device of the power facility. Such GIS can protect the power system by quickly shutting off excessive current in case of failure as well as normal opening and closing. In the case of GIS, when a fault occurs and an overcurrent occurs, it has a large-scale effect, takes a long time to recover from the accident, and also increases the power outage time. Various anomalies that cause dielectric breakdown in GIS cause partial discharge before breakdown. Therefore, it is important to detect partial discharge (PD) in GIS to avoid failure and ensure high reliability and safety. Various electrical, mechanical, and chemical methods have been used to detect PD in GIS.

측정을 분석한다. PD 펄스 수, 최대 진폭 또는 각 위상의 평균 진폭을 분석하여 오류 유형을 식별한다. To investigate the characteristics of PD in GIS, time-resolved PD (TRPD) and phase-resolved PD (PRPD) analyzes were studied. The TRPD-based method is a method of analyzing time-domain characteristics of PD pulses, frequency-domain characteristics of PD pulses, and time-domain and frequency-domain characteristics of PD pulses. The PRPD-based diagnostic method is the number of phase amplitudes

Analyze the measurements. The type of error is identified by analyzing the number of PD pulses, the maximum amplitude, or the average amplitude of each phase.

In the conventional case, defect types were classified using knowledge-based fuzzy logic analysis, K-means cluster analysis, and machine learning techniques such as artificial neural networks (ANNs). The prior art has a problem in that a feature extractor using data and a classifier for fault diagnosis are individually configured.

The present invention provides a deep learning-based partial discharge diagnostic device and method for gas insulated switchgear that can extract accurate features and improve diagnostic performance by simultaneously performing signal feature extraction and classification for fault diagnosis using deep learning techniques is to provide

According to one aspect of the present invention, deep learning-based partial discharge for gas insulated switchgear that can extract accurate features and improve diagnostic performance by simultaneously performing signal feature extraction and classification for fault diagnosis using deep learning techniques A diagnostic device is provided.

According to an embodiment of the present invention, there is provided a device comprising: an acquisition unit configured to acquire a phase resolved partial discharge (PRPD) signal within each power cycle; And a deep learning-based part for gas insulated switchgear including a partial discharge diagnosis unit that applies the phase decomposition partial discharge signal to the learned deep learning model to extract signal feature values and determine whether there is a failure through classification for failure diagnosis A discharge diagnostic device may be provided.

The deep learning model is formed in a many-to-one RNN structure using Long Short-Term Memory (LSTM) cells composed of multiple layers, but each LSTM cell is erased. Four hidden gates, consisting of a forget gate, an input gate, an external output gate, and an output gate, can determine whether to update, retain or overwrite the sequence of the previous steps.

The deep learning model is trained to minimize the following cost function,

는 집합의 크기를 나타내며, G는 학습 데이터를 나타내며,

는 g번째 학습 데이터의 손실값을 나타낸다. here,

denotes the size of the set, G denotes the training data,

represents the loss value of the g-th training data.

The partial discharge diagnostic unit may generate training data by slicing and overlapping the phase-decomposed partial discharge signal of each power cycle to a predetermined size, and then train the deep learning model using the training data.

According to another aspect of the present invention, deep learning-based partial discharge for gas insulated switchgear that can extract accurate features and improve diagnostic performance by simultaneously performing signal feature extraction and classification for fault diagnosis using deep learning techniques A diagnostic method and a recording medium for performing the method are provided.

According to an embodiment of the present invention, there is provided a method comprising: obtaining a phase resolved partial discharge (PRPD) signal within each power cycle; And Deep learning-based partial discharge diagnosis for gas insulated switchgear comprising the step of applying the phase-resolved partial discharge signal to the learned deep learning model to extract signal feature values and determining whether a failure occurs through classification for failure diagnosis A method may be provided.

By providing a deep learning-based partial discharge diagnosis apparatus and method for a gas insulated switchgear according to an embodiment of the present invention, an accurate characteristic is obtained by simultaneously performing a signal feature extraction and a classification for fault diagnosis using a deep learning technique Extract and improve diagnostic performance.

1 is a flowchart illustrating a deep learning-based partial discharge diagnosis method for a gas insulated switchgear according to an embodiment of the present invention.
Figure 2 is a diagram showing a deep learning model according to an embodiment of the present invention.
3 is a diagram showing the structure of an LSTM module according to an embodiment of the present invention.
4 is a block diagram schematically showing the internal configuration of a deep learning-based partial discharge diagnostic device for a gas insulated switchgear according to an embodiment of the present invention.
5 is a view showing a comparison result of classification performance of a deep learning-based partial discharge diagnosis method for a switch insulation switchgear according to an embodiment of the present invention with the conventional one.

As used herein, the singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "consisting of" or "comprising" should not be construed as necessarily including all of the various components or various steps described in the specification, some of which components or some steps are It should be construed that it may not include, or may further include additional components or steps. In addition, terms such as "...unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software. .

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a deep learning-based partial discharge diagnosis method for a gas insulated switchgear according to an embodiment of the present invention, and FIG. 2 is a view showing a deep learning model according to an embodiment of the present invention, FIG. 3 is a diagram illustrating the structure of an LSTM module according to an embodiment of the present invention.

In operation 110 , the partial discharge diagnostic apparatus 100 acquires a phase resolved partial discharge (PRPD) signal (hereinafter, referred to as PRPD).

For example, the partial discharge diagnosis apparatus 100 may acquire a PRPD signal by using a UHF sensor.

In step 115, the partial discharge diagnosis apparatus 100 applies the PRPD signal to the learned deep learning model to extract signal feature values, and determines whether there is a failure through classification for failure diagnosis.

This will be described in more detail.

The PRPD signal in the m-th power cycle may be defined as in Equation (1).

이고,

은 전력 주기내에서 데이터 포인트의 수를 나타낸다. here,

ego,

denotes the number of data points in a power cycle.

2 is a diagram illustrating a deep learning model according to an embodiment of the present invention. In an embodiment of the present invention, it is assumed that the deep learning module is an RNN-based module and will be mainly described.

In the case of a general RNN structure, it is difficult to minimize the cost function of the network because the gradient value becomes exponentially smaller due to long-term dependences when the gradient is calculated using the gradient method.

The RNN structure according to an embodiment of the present invention is a many-to-one model and can perform expression learning of deep learning based on LSTM.

The structure of the LSTM module is shown in FIG. 3 .

로 구성된다. 여기서,

는 이전 레이어(l-1)의 m번째 LSTM 모듈의 출력이며,

과

은 계층 l번째 계층의 (m-1)번째 LSTM 모듈의 출력이다. The input of the m-th LSTM module of the l-th layer is

is composed of here,

is the output of the mth LSTM module of the previous layer (l-1),

class

is the output of the (m-1)-th LSTM module of the l-th layer.

Hidden states of the m-th LSTM module of layer l may be expressed as Equations 2 to 7.

는 N x N 가중치 행렬을 나타내고,

는 N x 1 바이어스 벡터를 나타내며,

는 각 요소별 곱을 나타내고,

는 시그모이드 활성화 함수를 나타내며,

는 하이퍼볼릭 탄젠트 활성화 함수를 나타내고,

는 합성(concatenation) 함수를 나타낸다. here,

denotes an N x N weight matrix,

represents the N x 1 bias vector,

represents the product of each element,

represents the sigmoid activation function,

denotes the hyperbolic tangent activation function,

denotes a concatenation function.

과 같은 PRPD 벡터이다. 여기서,

이다. LSTM 모델은 4개의 히든 레이어(

)를 사용하여 정보 흐름을 조정함으로써 롱-텀 의존성 문제를 피할 수 있다. 여기서, 각 게이트는 셀 상태(

) 내의 정보 흐름을 제어할 수 있다. For the first layer, the input of the LSTM block is

It is a PRPD vector such as here,

am. The LSTM model has four hidden layers (

) to avoid the long-term dependency problem by coordinating the information flow. Here, each gate is a cell state (

) can control the flow of information in

는 셀 상태로부터 불필요한 정보가 무언인지 결정할 수 있는 지움 게이트(forget gate)이다.

is a forget gate that can determine what unnecessary information is from the cell state.

는 셀 상태의 값이 갱신되어야 할지를 결정할 수 있는 입력 게이트이다.

is the input gate that can determine whether the value of the cell state should be updated.

는 외부 상태에 추가될 수 있는 새로운 후보값을 출력하는 출력 게이트이다.

is an output gate that outputs a new candidate value that can be added to an external state.

는 수학식 2와 같이 각 단계 사이에서 셀 상태를 수정하는데 사용될 수 있다.

can be used to modify the cell state between each step as shown in Equation (2).

는 출력 게이트로, 현재 셀 상태의 어느 부분이 출력되어야 할지를 결정하는 필터 역할을 한다.

is the output gate, which acts as a filter to determine which part of the current cell state should be output.

를 통해 입력되고, 수학식 3과 같이 현재 단계의 히든 상태(

)이 되도록

를 통해 필터될 수 있다. the cell state

is input through, and as shown in Equation 3, the hidden state of the current stage (

) to be

can be filtered through.

The output y for class K is related to z of the last LSTM layer as in Equation (8).

는 K x N 가중치 행렬이며,

는 K x 1 바이어스 벡터이고,

는 소프트맥스 함수를 나타낸다. here,

is a K x N weight matrix,

is the K x 1 bias vector,

represents the softmax function.

In Equation (8), the j-th element of the output y represents the probability that the defect is recognized as the j-th category in the K class, and can be defined as in Equation (9).

The parameters of the LSTM RNN according to an embodiment of the present invention may be learned through training data for minimizing a cost function as in Equation (10).

는 집합의 크기를 나타내고,

는 g번째 학습 데이터의 손실 값을 나타낸다. 수학식 10에서

는 본 발명의 일 실시예에 따른 LSTM RNN 모델이 얼마나 정확하게 학습 데이터의

라벨 벡터를 예측했는지를 측정한다. here,

represents the size of the set,

represents the loss value of the g-th training data. in Equation 10

is how accurately the LSTM RNN model according to an embodiment of the present invention

Measures whether the label vector was predicted.

To minimize the loss function, many modifications of gradient descent methods such as AdaGrad, AdaDelta and Adam optimizer have been studied. Such an optimizer can adaptively change the learning function to appropriately minimize the loss function. In an embodiment of the present invention, the Adam optimization algorithm may be used as a loss function for training a deep learning model.

4 is a block diagram schematically illustrating an internal configuration of a deep learning-based partial discharge diagnostic apparatus for a gas insulated switchgear according to an embodiment of the present invention.

Referring to FIG. 4 , a deep learning-based partial discharge diagnosis apparatus 100 for a gas insulated switchgear according to an embodiment of the present invention includes an acquisition unit 410 , a partial discharge diagnosis unit 420 , a memory 430 and and a processor 440 .

The acquisition unit 410 is a signal for acquiring a PRPD signal within each power cycle.

For example, the acquisition unit 410 may acquire the PRPD signal by sampling the power signal within each power cycle. In this case, the acquisition unit 410 may perform 1/60 second sampling, thereby acquiring a PRPD signal having 128 samples.

The partial discharge diagnosis unit 420 may apply the PRPD signal to the learned deep learning model to extract signal feature values, and determine whether there is a failure through classification for failure diagnosis.

Since this is the same as that described in FIG. 1 , a redundant description thereof will be omitted.

The memory 430 is a means for storing a program code (or instructions) for performing a deep learning-based partial discharge diagnosis method for a switch isolation switchgear according to an embodiment of the present invention.

The processor 440 includes internal components (eg, an acquisition unit 410, a partial discharge diagnosis unit ( 420), the memory 430, etc.).

5 is a comparison result of classification performance of a deep learning-based partial discharge diagnostic method for a switch insulation switchgear according to an embodiment of the present invention with the conventional one.

It can be seen that the partial discharge diagnostic apparatus according to an embodiment of the present invention achieves the highest overall classification accuracy performance (96.23%) compared to the prior art.

The apparatus and method according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the computer readable medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - Includes magneto-optical media and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

So far, the present invention has been focused on the embodiments thereof. Those of ordinary skill in the art to which the present invention pertains will understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the scope equivalent thereto should be construed as being included in the present invention.

100: partial discharge diagnostic device
410: acquisition unit
420: partial discharge diagnostic unit
430: memory
440: processor

Claims (7)

an acquisition unit configured to acquire a phase resolved partial discharge (PRPD) signal within each power cycle; and
A partial discharge diagnosis unit that applies the phase decomposition partial discharge signal to the learned deep learning model to extract a signal characteristic value and determines whether a failure occurs through classification for failure diagnosis,
The deep learning model is a deep learning-based partial discharge diagnostic device for gas insulated switchgear, characterized in that it is trained to minimize the following cost function.

here,

denotes the size of the set, G denotes the training data,

represents the loss value of the g-th training data.
According to claim 1,
The deep learning model is formed in an RNN structure of a many-to-one structure using Long Short-Term Memory (LSTM) cells composed of multiple layers,
Gas isolation, characterized in that each LSTM cell determines whether to update, maintain or overwrite the sequence of the previous step by four hidden gates consisting of a forget gate, an input gate, an external output gate and an output gate. Deep learning-based partial discharge diagnostic device for switchgear.
delete According to claim 1,
The partial discharge diagnosis unit generates training data by slicing and overlapping the phase decomposition partial discharge signal of each power cycle to a predetermined size, and then uses the training data to train the deep learning model. Deep learning-based partial discharge diagnostic device for
obtaining a phase resolved partial discharge (PRPD) signal within each power cycle; and
Comprising the step of applying the phase decomposition partial discharge signal to the learned deep learning model, extracting signal feature values and determining whether there is a failure through classification for failure diagnosis,
The deep learning model is a deep learning-based partial discharge diagnostic method for a gas insulated switchgear, characterized in that it is trained to minimize the following cost function.

here,

denotes the size of the set, G denotes the training data,

represents the loss value of the g-th training data.
6. The method of claim 5,
The deep learning model is formed in an RNN structure of a many-to-one structure using Long Short-Term Memory (LSTM) cells composed of multiple layers,
Gas isolation, characterized in that each LSTM cell determines whether to update, maintain or overwrite the sequence of the previous step by four hidden gates consisting of a forget gate, an input gate, an external output gate and an output gate. Deep learning-based partial discharge diagnostic method for switchgear.
In a computer-readable recording medium recording a program code for performing a deep learning-based partial discharge diagnostic method for a gas insulated switchgear,
obtaining a phase resolved partial discharge (PRPD) signal within each power cycle; and
Performing the step of applying the phase decomposition partial discharge signal to the learned deep learning model to extract a signal feature value and determining whether a failure occurs through classification for failure diagnosis,
The deep learning model is a recording medium, characterized in that it is trained to minimize the following cost function.

here,

denotes the size of the set, G denotes the training data,

represents the loss value of the g-th training data.

Images