The Monitoring Interface of Insulator's State Based On The Leakage Characteristics

You might also like

You are on page 1of 4

The Monitoring Interface of Insulator’s State

Based on the Leakage Characteristics


Shi Yan , Hu Gang and Zhu Jiafu
Chongqing University of Science & Technology
Shapingba District,Chongqing 401331,China.
e-mail:ceeshy@163.com

that the magnitude of the leakage current has a certain relation


Abstract—In order to effective monitoring the state of with the state of the insulator surface.
insulator’s surface, an insulator flashover risk monitoring
interface was designed in this paper. The pollution flashover risk Secondly, the method of frequency analysis was used to
monitoring interface was established based on the Matlab analyze the characteristics of leakage current, so as to establish
GUI ,which called the trained BP network to realize the real-time the relationship between the spectral characteristics and the
monitoring of insulator’s state. The method of BP artificial neural surface state of the insulator. Suda measured and extracted
network was used to establish the relationship between the input harmonic to analysis, proposed harmonic warning value to
vector and output vector. The input vector of the neural network prevent insulator flashover accident[7]. Based on the analysis of
is relative humidity, leakage current amplitude and the ratio of the distortion of leakage current waveform, Kumagai divides
the third harmonic of the leakage current to the amplitude of the the leakage current of the insulator into sinusoidal component,
fundamental wave. Based on the fuzziness of the state of the local arc and transition component, and evaluates the
insulator, the output of the neural network is the membership characteristics of the hydrophobicity of the insulator[8].
degree of the security state of the insulator to the fuzzy subset. Cherney and other researchers extracted different harmonic
According to the analysis of the test results and the actual components of leakage current, obtained the leakage current
situation, three fuzzy subsets are defined on the universe U to fundamental and low frequency harmonic components can
represent the security state of the insulting insulators: A = [safe effectively assess the surface state of the insulator[9]. It can be
state], B = [light alarm] and C = Serious alarm]. In order to realize seen that the majority of researchers believe that the low
the network model, we quantify the fuzzy outputs to 1,2,3, so as to frequency components of leakage current contain more
establish the correspondence between input and output in the BP
information, which provides a reference for the further study of
network through training.
the status of insulators based on the frequency characteristics of
Keywords—monitoring, insulator, pollution, interface leakage current.
I. INTRODUCTION Due to the diversity and randomness of the factors
influencing the leakage current of insulators, the selection and
The common methods of detecting insulator state are the characterization of the leakage current’s characteristics have
measurement of equivalent salt density, the measurement of the great dispersion. There is no uniform detection method and
conductivity of the polluted layer and leakage current standard. The results by different systems and methods are
measurement method[1-3].For these methods, the equivalent salt different. Moreover, the leakage current changes with the
density method is simple, but can not reflect the surface state of operating environment. The preliminary project has considered
the insulator in real time. The conductivity measurement of the the characteristic both in time domain and in frequency domain,
contamination layer can effectively reflect the insulation also the environmental humidity, so as to establish a
characteristics of the insulator, but cannot continuously detect comprehensive evaluation model of the surface state on the
insulators. In comparison, the leakage current measurement surface of insulator[10]. On the basis of this, if a graphical
method can continuously detect the insulator; it can timely and interface for the surface condition monitoring of insulators is
accurately reflect the impact of various dynamic parameters on designed, with the accumulation of working data, the status of
the insulator state. the insulators would be judged to guide the actual production.
The characteristics of leakage current have been researched Based on the previous research, a monitoring interface is used
in widely range. First of all, the researchers tried to extract to determine the state of the insulator, by means of the ambient
characteristics of the leakage current in time domain which is humidity and the characteristics of leakage current in the time
related with the characteristics of arc on insulator’s surface. domain and frequency domain. Based on a large number of
In[4], the relationship between the leakage current amplitude experimental data, BP artificial neural network was used to
with insulator surface state was established. Japanese establish the model between the relative humidity, the
researchers proposed that leakage current amplitude can magnitude of the leakage current, the ratio of the third harmonic
effectively characterize the state of the insulator surface to the amplitude of the fundamental wave with flashover risk,
changes[5]. Flashover process is divided into different stages then the model was used to predict the flashover risk of polluted
according to the amplitude of leakage current, the continuous insulator, to achieve real-time monitoring of the insulator state.
monitoring of leakage current amplitude maybe offer reference Using the graphical interface design GUI of Matlab, polluted
for reducing the insulator flashover accident[6].The results show flashover risk monitoring interface would be designed after call

978-1-7281-5511-1/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on June 14,2021 at 10:00:00 UTC from IEEE Xplore. Restrictions apply.
2

the trained BP network to realize the real-time monitoring of


the insulator state.
II. OVERALL DESIGN OF MONITORING INTERFACE
To design monitoring interface, a method of sub function
called was get to achieve the state of the insulator. The main
function is the design of the monitoring interface, the Matlab
graphical interface tool GUI was used as a platform for design.
The design of monitoring interface mainly includes two parts,
visual interface design and the control callback function.
Function mainly includes two parts, establish the model of
program and alarm procedures. The model is mainly based on
BP neural network which was trained by a large number of
experimental data, then the monitoring new amount was the
inputs of BP network, the output was the corresponding fuzzy
set. For example, when the insulator is good, the green is on, Fig.1. The visual interface
when the partial discharge occurs, the yellow light is on, when
the flashover occurs on insulator, the red light is on. The existing tools in the toolbar can be drag the visual area
to establishment of the various controls, which can also be
The data used in the training come from the artificial prepared to display the corresponding procedures. Here we
pollution experiment. Sketch maps of the tested circuit as adopt the former method, at the same time, in the generation of
shown in Fig.1.In the experiment, the method of constant the corresponding M file. On this basis, to achieve different
voltage is used, that is, the test voltage is applied to the polluted function’s controls, the corresponding callback function is
insulator, and the applied voltage is less than the flashover written in the appropriate space. The visual interface is shown
voltage of the insulator, to ensure that the measured leakage in Fig. 1.
current was under the adjusted humidity of the insulator surface.
The atmospheric pressure is 99.5 kPa, the temperature is about IV. ESTIMATION MODEL OF INSULATOR’S STATE BASED ON
26℃, and the change of the indoor temperature is not more than BP NEURAL NETWORK
2℃. When the temperature changes for 2 K, the leakage current BP network is also called error back propagation neural
of insulator surface has no obvious change ,so the influence of network. It adopts error back propagation learning algorithm
atmospheric pressure and temperature on the leakage current of (BP algorithm), and its structure model is shown in Fig.2. BP
insulator surface can not be considered in the test. The network consists of three neuron layers, the first layer is input
desiccant is used for dehumidification, the ultrasonic fogger is layer, the second layer is middle layer, also known as hidden
used for humidification, and desiccant or spray is added to the layer, and the last layer is output layer. A neural network
fog chamber or glass test tank according to the need of the test. system can have many hidden layers according to the needs.
During the test, the constant voltage method is used ,the voltage
was quickly applied as a predetermined voltage, the waveform
data sample of 10 period voltage and leakage current were
recorded rapidly by oscilloscope.

Fig.2 The structure of BP network

The neurons in each layer form a full interconnection


connection, and there is no connection between the neurons in
Fig. 1 Sketch maps of the tested circuit the same layer. In addition to the output layer, the output of
each node is the weighted sum of the output values of all nodes
III. MONITORING INTERFACE in the previous layer. BP network has the characteristics of
realizing highly complex nonlinear mapping through its own
The design of monitoring interface mainly includes visual
learning.
interface design and the control callback function. The visual
interface is designed based on the principle of simple Learning algorithm is the key problem to design BP
understanding and convenient operation. The visual interface network. The traditional BP learning algorithm is a supervised
mainly includes four kinds of controls, namely edit, text, learning algorithm. Its learning process includes forward
pushbutton and X Active. calculation of input information and error propagation. It uses
gradient search technology to reduce the error between the
actual output value and the expected output value by
calculating the derivative of error when each weight changes,

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on June 14,2021 at 10:00:00 UTC from IEEE Xplore. Restrictions apply.
3

adjusting the connection weight and neuron threshold between nonlinear multivariate function, the variable parameters of
neurons, so that the error gradually reaches the maximum Small network. Weight and threshold value, are independent variables,
or expected. and the error of network is dependent variable, this function is
called objective function. BP Algorithm adjusts the variable
The input vector of the neural network is relative humidity, parameters of the network by the first partial derivative of the
leakage current amplitude and the ratio of the third harmonic of variable parameters to the network error. For Nonlinear
the leakage current to the amplitude of the fundamental wave. multivariate functions, the downward direction of the network
As the input and output of the neural network should select the error is the fastest only in a small range of the direction of the
dimensionless vector, and in order to further response, Any first partial derivative, beyond which the network error may
input and output vectors should are normalized to the [-1,1] increase, this makes it difficult to choose the step size of BP
interval. Algorithm: The network is easy to oscillate when the step size
If more fuzzy subsets of the safety condition of polluted is too large, the convergence speed is too slow when the step
insulators are defined in universe of fuzzy set U, which will size is too small, and the network is difficult to jump out when it
result in more precise descriptions of the safety condition. But falls into the local minimum point. So the
as the number of output neurons of neural network system is Levenberg-Marquardt Algorithm based on second order partial
equal to the number of defined fuzzy subsets, it will be more derivative is used to train the multilayer feedforward neural
complex, thus the training time to neural network system also network. The neural network system was trained by sampled
greatly increases, which is more difficult to engineering data.
practice. So according to the analysis of discharge phenomenon,
three fuzzy subsets are defined on universe of fuzzy set U, three V. SIMULATED ACTUAL MONITORING
safety conditions are A= [safe], B= [light alarm] and C= On the basis of the BP network model, the main task of this
[serious alarm]. Where A refers to no discharge or only weak paper is to analyze and display the new monitoring value. In
discharge on surface of insulators, the insulators can run safely, Fig.3, the first is to select the type of insulator, and then enter
so the safety condition is safe. B refers to the fragmentary part the corresponding data, such as the relative humidity, leakage
of small electric arc on the surface of insulators, under this current amplitude, the ratio of leakage current three times
condition, would not affect safe operation of insulator, but if the harmonic and fundamental wave. And then click "OK" ,
contamination or humidity becomes heavier, close attentions degrees of three fuzzy subsets to safety condition of polluted
should be paid, so the safety condition is light alarm. C refers to insulators were calculated based on membership function,
intense main arc on surface of insulator, it is on the critical state, which were regarded as outputs of ANN. Then the result will
and in this case the protective device should act urgently, so the come to the corresponding insulator pollution level. As shown
safety condition is serious alarm. in Fig.3.
In order to predict the degrees of three fuzzy subsets about
safety condition of polluted insulators, a three-layer feed
forward neural network system is developed, which is input
layer, hidden layer band output layer. The neural network
system is adjusted the weights by a gradient descent method
such that the error decreases with each iteration and the neural
network system gets closer and closer to producing the desired
output.
The Output neurons were selected as the S transfer function,
which guarantees its output is in the range between [0,1], thus
the outputs of ANN satisfies both the non-linear and output
range of degree of membership function.
In order to realize the network model, we quantify the three
fuzzy outputs to 1,2,3, so as to establish the correspondence Fig.3 Simulated actual monitoring
between input and output in the BP network through training. In
the BP network hidden layer transfer function, output layer VI. CONCLUSION
transfer function, learning function and training function were
The method of BP artificial neural network, the input vector
selected tansig, purelin,trainlm and learngdm. The learning
of BP ANN is relative humidity, leakage current amplitude and
algorithm of trainlm is used to increase the learning speed, the
the ratio of the third harmonic of the leakage current to the
network can quickly achieve the predetermined accuracy.
amplitude of the fundamental. The output of the neural network
In the present study, a huge epoch number of 1000 and an is the membership degree of the security state of the insulator to
error goal or performance goal of 0.0001 were introduced to the the fuzzy subset. the relationship between the input vector and
network. The training procedure adjusted the weighting output vector was established after the BP network through
coefficients using the Levenberg–Marquardt algorithm for training.
local optimization. The Levenberg-Marquardt Algorithm Based on the model of evaluation insulators’ state, a
introduces the second order partial derivative information, condition monitoring interface of insulator was established on
which makes the convergence speed of the network have a the Matlab platform by graphical design. The simulation results
qualitative leap.In mathematics, the training of multilayer show that the monitoring interface has the advantages of simple
feedforward neural network is an optimization problem of operation, intuitive display and convenient. It can provide
reference, so as to the engineering staff to take corresponding

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on June 14,2021 at 10:00:00 UTC from IEEE Xplore. Restrictions apply.
4

measures based on the results.


ACKNOWLEDGMENT
In this paper, the research was sponsored by Foundation of
Chongqing University of Science & Technology (Project No.
CK2014B17) and Foundation Chongqing Municipal Education
Committee(Project No. KJ1501338).
REFERENCES
[1] Montoya G, Ramirez I, Montoya J I. Correlation among ESDD, NSDD
and leakage current in distribution insulators [J]. IEEE Proceedings on
Generation Transmission and Distribution, 2004, Vol.151,No.3,pp
334-340.
[2] X.Jiang,J.Yuan, Z.zhang, J.Hu and C.Sun.“Study on ac
artificially-contaminated flashover performance of various types of
insulators”,IEEE Transactions on Power Delivery,
2007,Vol.22,No.4,pp.2567-2574.
[3] J.YUAN ,X.JIANG, L.SHU,Z.Zhang and Y.ZHANG. Influence of
Salt/Non-soluble Deposit Density on AC Artificially Pollution
Flashover Performance of Various Types Insulators [J]. Proceedings of
the CSEE ,2007,Vol.6,No.2,pp.96-101.
[4] G. Montoya, I. Ramirez and R. Hernandez. The leakage current as a
diagnostictool for outdoor insulation. IEEE/PES Transmission and
Distribution Conference and Exposition: 2008,pp.1~4.
[5] B. Marungsri, H. Shinokubo and R. Matsuoka. Effect of specimen
configurationon deterioration of silicone rubber for polymer insulators in
salt fog ageing test. IEEE Transactions on Dielectrics and Electrical
Insulation, 2006, Vol.13,No.1,pp.129~138.
[6] S.Kumagai, B. Marungsri, H. Shinokubo and R. Matsuoka.
Performances ofsilicone and ceramic materials used for outdoor
insulation in field test. International Symposium on Electrical Insulating
Materials, 2005,pp.336~339.
[7] T. Suda. Frequency characteristics of leakage current waveforms of an
artificially polluted suspension insulator. IEEE Transactions on
Dielectric and Electrical Insulation, 2001, Vol.8,No.4,pp.705~709.
[8] S.Kumagai and N. Yoshimura. Leakage current characterization for
estimating the conditions of ceramic and polymeric insulating surfaces.
IEEE Transactions on Dielectrics and Electrical Insulation,
2004,Vol.11,No. 4. pp.681~690.
[9] A. H. EI-Hag, S. H. Jayaram and E. A. Cherney. Effect of insulator
profile on aging performance of silicone rubber insulators in salt-fog.
IEEE Transactions on Dielectric and Electrical Insulation,
2007,Vol.14,No.2,pp. 352~359.
[10] Xingliang Jiang,Yan Shi,Zhijin Zhang and Caixin Sun.Evaluating the
Safety condition of Porcelain Insulators by the Time and Frequency
Characteristics of LC Based on Artificial Pollution Tests .IEEE
Transactions on Dielectrics and Electrical Insulation,Vol. 17, Issue 2, pp.
481-489, April 2010.

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on June 14,2021 at 10:00:00 UTC from IEEE Xplore. Restrictions apply.

You might also like