WO2019182166A1 - Electrical load determination apparatus and method using modified mfcc - Google Patents

Electrical load determination apparatus and method using modified mfcc Download PDF

Info

Publication number
WO2019182166A1
WO2019182166A1 PCT/KR2018/003143 KR2018003143W WO2019182166A1 WO 2019182166 A1 WO2019182166 A1 WO 2019182166A1 KR 2018003143 W KR2018003143 W KR 2018003143W WO 2019182166 A1 WO2019182166 A1 WO 2019182166A1
Authority
WO
WIPO (PCT)
Prior art keywords
electric load
arc
current signal
feature vector
identified
Prior art date
Application number
PCT/KR2018/003143
Other languages
French (fr)
Korean (ko)
Inventor
최동환
임용배
김동우
이기연
문현욱
Original Assignee
한국 전기안전공사
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 한국 전기안전공사 filed Critical 한국 전기안전공사
Publication of WO2019182166A1 publication Critical patent/WO2019182166A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/25—Arrangements for measuring currents or voltages or for indicating presence or sign thereof using digital measurement techniques
    • G01R19/2506—Arrangements for conditioning or analysing measured signals, e.g. for indicating peak values ; Details concerning sampling, digitizing or waveform capturing
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R15/00—Details of measuring arrangements of the types provided for in groups G01R17/00 - G01R29/00, G01R33/00 - G01R33/26 or G01R35/00
    • G01R15/14—Adaptations providing voltage or current isolation, e.g. for high-voltage or high-current networks
    • G01R15/18—Adaptations providing voltage or current isolation, e.g. for high-voltage or high-current networks using inductive devices, e.g. transformers
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R15/00—Details of measuring arrangements of the types provided for in groups G01R17/00 - G01R29/00, G01R33/00 - G01R33/26 or G01R35/00
    • G01R15/14—Adaptations providing voltage or current isolation, e.g. for high-voltage or high-current networks
    • G01R15/18—Adaptations providing voltage or current isolation, e.g. for high-voltage or high-current networks using inductive devices, e.g. transformers
    • G01R15/186—Adaptations providing voltage or current isolation, e.g. for high-voltage or high-current networks using inductive devices, e.g. transformers using current transformers with a core consisting of two or more parts, e.g. clamp-on type
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/25—Arrangements for measuring currents or voltages or for indicating presence or sign thereof using digital measurement techniques
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
    • G01R23/16—Spectrum analysis; Fourier analysis
    • G01R23/165—Spectrum analysis; Fourier analysis using filters
    • G01R23/167—Spectrum analysis; Fourier analysis using filters with digital filters
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/141—Discrete Fourier transforms
    • G06F17/142—Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/02—Computing arrangements based on specific mathematical models using fuzzy logic
    • G06N7/04—Physical realisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/02—Computing arrangements based on specific mathematical models using fuzzy logic
    • G06N7/04—Physical realisation
    • G06N7/046—Implementation by means of a neural network
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S20/00—Management or operation of end-user stationary applications or the last stages of power distribution; Controlling, monitoring or operating thereof
    • Y04S20/30—Smart metering, e.g. specially adapted for remote reading

Definitions

  • the present invention relates to an apparatus and method for determining an electric load using a modified MFCC, and more particularly, to a technique for determining an electric load matching a given current signal waveform using a conventional MFCC algorithm.
  • Prior material provides an adaptive smart power measurement apparatus and method capable of identifying the device.
  • the adaptive smart power measurement method capable of identifying such a device collects at least one measured value of current, power, and power measured from a plurality of devices to be measured, and detects a change in the measured value and changes the measured value. If is detected, the unique value of the power or current matching the variation value of the power or current is searched and the phase is calculated according to the search to determine one device matching the unique value of the registered phase to determine the value of each unique value.
  • the eigenvalue is always applied to the latest value to increase the accuracy in identifying the device using the eigenvalue. have.
  • the above-described device identification apparatus and method merely present a specific method for signal waveform recognition for identifying the eigenvalue in determining the device by setting the eigenvalue of the device as a current or a phase shift value. Since there is no configuration for evaluating and predicting the safety of the device, there is a problem in that an accident of the electric device is prevented in advance and the cause of the accident is difficult to analyze.
  • the present applicant can perform an MFCC algorithm and a neural network on a given current signal waveform to identify an electric load and perform an arc analysis that matches the identified electric load to evaluate and predict the safety of the electric load in advance. Therefore, we will propose a way to prevent accidents by cutting off the power delivered to the electrical load in advance.
  • the present invention can improve the accuracy of electric load identification by performing the MFCC algorithm and neural network on a given current signal waveform, and minimize the time required to identify the electric load through the modified MFCC algorithm.
  • the present invention provides a method and apparatus for determining an electric load using a modified MFCC.
  • the present invention by performing the arc analysis through the stored arc analysis algorithm matched with the identified electric load when the arc generation for the identified electric load and by controlling the power supply to the identified electric load according to the arc analysis results, provides a method and apparatus for determining electric load using a modified MFCC that can be used for evaluating the safety of electrical installations and can improve energy saving and safety for the identified electric load.
  • the present invention is a.
  • a data acquisition unit sampling a current signal measured by the CT sensor at predetermined intervals to obtain a predetermined number of discrete signals
  • a feature vector derivation unit configured to derive a feature vector for a current signal waveform by performing a Mel-Frequency Cepstral Coefficients (MFCC) algorithm on the predetermined number of discrete signals;
  • the feature vector for the current signal waveform is characterized in that it comprises an electrical load identification unit for extracting the electrical load matching the feature vector using a neural network.
  • the feature vector derivation unit Preferably, the feature vector derivation unit,
  • a fast fourier transform for performing a fast Fourier transform on the predetermined number of discrete signals to derive a result value in a frequency domain
  • the filter bank may include a filter bank provided with a plurality of filters and deriving a feature vector for the current signal waveform by passing a result value of the frequency domain through a current signal waveform of a frequency band of each filter.
  • It may be set to a frequency band that does not overlap each other.
  • the electrical load identification unit Preferably the electrical load identification unit,
  • a learning model may be constructed based on the feature vector of the measured current signal for each electric load, and the electric load matching the feature vector for the input current signal may be extracted using the constructed learning model.
  • the device Preferably the device,
  • the arc analysis unit may further include an arc analysis unit that performs an arc analysis of the electric load identified through the arc analysis algorithm.
  • the device Preferably the device,
  • the apparatus may further include a power supply control unit configured to generate and transmit a relay driving signal to cut off power supply to the identified electric load based on the arc analysis result of the identified electric load.
  • a power supply control unit configured to generate and transmit a relay driving signal to cut off power supply to the identified electric load based on the arc analysis result of the identified electric load.
  • the present invention in the device for identifying the electric load by the current signal waveform measured from the CT sensor installed in the electrical line between the switchboard and the electrical load, the measured current signal waveform is sampled at a predetermined interval to a predetermined number of discrete signals Acquiring data; A feature vector derivation step of deriving a feature vector for the measured current signal waveform after a fast Fourier transform on the predetermined number of discrete signals and passing through a predetermined number of filters; And an electric load identification step of extracting an electric load matching the characteristic vector for the measured current signal waveform through a neural network with respect to the feature vector for the current signal waveform.
  • the method may further include performing an arc analysis step, and further comprising a power supply control step of controlling a power supply of the identified electric load in case of an arc failure of the identified electric load according to the arc analysis result of the identified electric load.
  • the present invention having the above-described configuration, by using a predetermined number of filters of a frequency band different from the frequency band used in the conventional MFCC algorithm, it is possible to efficiently extract the characteristic vector for each electric load by which the difference of the current signal waveform for each electric load is obvious
  • the processing time and size can be reduced by performing the modified MFCC algorithm.
  • the arc of the identified electric load is generated by using the feature vector for the measured current signal, an arc mode matching the identified electric load is selected, and the arc of the electric load identified through the arc analysis algorithm corresponding to the selected arc mode.
  • the analysis can be performed to improve the accuracy of the arc accident cause determination of the identified electrical loads and to perform a safety assessment for the identified electrical loads based on the arc analysis results.
  • FIG. 1 is a block diagram of an electric load determination device according to an embodiment of the present invention.
  • 2 to 5 are graphs showing the output signal of each part of the electric load determination device according to an embodiment of the present invention.
  • FIG. 6 is a flowchart illustrating a method of determining an electric load according to an embodiment of the present invention.
  • the accuracy of electric load discrimination is reduced because the high frequency component of the measured current signal waveform is alleviated when performing the preprocessing process to remove noise noise and noise with respect to the measured current signal through the general MFCC (Mel-Frequency Cepstral Coefficients) algorithm.
  • MFCC Mel-Frequency Cepstral Coefficients
  • the present invention identifies the electric load through a modified MFCC algorithm that performs fast Fourier transform on N discrete signals obtained from the measured current signals, thereby eliminating pre- and post-processing of the existing MFCC algorithm. It is provided to reduce the time and the size of the electric load determination device to derive the feature vector for the current signal waveform of the.
  • the present invention includes a filter bank provided with a predetermined number of filters of the frequency band that does not overlap each other, extracts the feature vector for the measured current signal and performs a neural network on the extracted feature vector of the current signal By identifying the load, it is configured to improve the accuracy of the electric load determination.
  • the present invention includes a plurality of arc modes and an arc analysis algorithm that performs each arc mode to select an arc mode corresponding to the identified electric load when an arc of the identified electric load is generated, and an arc analysis algorithm corresponding to the selected arc mode.
  • the accuracy of the arc accident cause determination of the identified electric load can be improved, and the safety evaluation of the identified electric load can be performed based on the arc analysis result, and the arc analysis result is determined by the arc failure.
  • By controlling the power supply delivered from the main distribution panel to the electrical load it is configured to improve energy saving and safety of the identified electrical load.
  • the electric load determination device 100 is a data acquisition unit 110, a feature vector derivation unit ( 120, and an electric load determination unit 130.
  • the data acquisition unit 110 samples a plurality of row data by sampling a current signal measured at a predetermined interval using a CT (Current Transformer) sensor 50 installed in an electric line connected between the distribution panel 10 and the electric load 30.
  • CT Current Transformer
  • the raw data is a discrete signal used for signal processing for the determination of the electrical load.
  • the feature vector derivation unit 120 is provided to extract a feature vector for the current signal measured from the plurality of discrete signals using the modified MFCC algorithm, and the feature vector derivation unit 120 includes a fast fourier transform. ) 121 and the filter bank 123.
  • the FFT 121 performs a Fast Fourier Transform (FFT) on the plurality of discrete signals on the time axis to derive a result value in the frequency domain. That is, the FFT 121 derives the Discrete Fourier Transform (DFT) spectrum of the frequency domain, and the obtained Discrete Fourier Transform (DFT) spectrum form discrete transform result is transmitted to the filter bank 123.
  • FFT Fast Fourier Transform
  • FIG. 2 shows a graph of the discrete conversion result of the FFT 121. That is, referring to FIG. 2, it is possible to check the discrete conversion result converted into the frequency domain for each of various electric loads.
  • the filter bank 123 is provided with at least one filter having a frequency band that does not overlap each other, the discrete conversion result is passed through the at least one filter to derive a feature vector for the current signal.
  • each electric load has a different result value in the range of 10 to 10000 Hz
  • the frequency band of each filter is set and / or changed so as not to overlap each other for each electric load.
  • the result values of the frequency domain pass through at least one filter of the filter bank 123 to extract M coefficient values per frame, and M coefficient values per frame are output as a feature vector for the current signal.
  • FIG. 3 is a graph showing a feature vector for a current signal for each electric load that has passed through at least one filter for the discrete conversion result for each electric load in the range of 10 to 10000 Hz shown in FIG. 2.
  • the filter bank 123 includes 26 filters having frequency bands that do not overlap each other, and includes an air conditioner, a computer, a hair dryer, lighting, a microwave oven, a refrigerator, and a television that have passed the frequency band of each filter.
  • the feature vector for the current signal for each electric load can be checked.
  • the feature vector for the current signal for each electric load is transmitted to the electric load identification unit 130.
  • the electric load identification unit 130 receives the feature vector of the measured current signal and learns through the neural network to identify the electric load.
  • the electric load identification unit 130 constructs a learning model by classifying the electric load according to the electric load pattern consisting of the characteristic vector of the measured current signal, and matches the feature vector for the given current signal through the constructed learning model. Extract the electrical load of the electrical load pattern. At this time, the learning for electric load determination is performed by the back propagation neural network.
  • the back propagation neural network transmits a feature vector input from the outside to the neural network, and the neural network calculates an error of the feature vector and adjusts a weight to reduce the error.
  • the error is derived using Means Square Error (MSE).
  • the back propagation neural network is performed in the following order: (1) initialization step, (2) forward forward feature vector, (3) back propagation and weight correction step of error signal, and (4) repetition step.
  • the number of input output neurons is set, the number of necessary hidden bits and the number of hidden neurons are set, the number of hidden layers is set to 1, and the weight and threshold are also set by the manufacturer. Initialized to a value.
  • feature vector forward progress step is to set the input and target output, and is normalized between -1 and 1 or between 0 and 1.
  • the input vector (X p -) and an output vector (D p) are: It can be expressed by Equation 1. That is, the sum Net pj of input values of neurons j of the hidden layer derived using the input vector X p ⁇ is represented by Equation 2.
  • the actual output (O pj ) of the neurons (H) of the hidden layer is derived from Equation 3.
  • the output O pj of the neurons (H) of the hidden layer is used as an input of the neurons of the output layer, and the sum Net pk of the values input to the k-th neuron of the output layer satisfies Equation 4 below.
  • the back propagation and weight correction step of the error signal derives the error signal ( ⁇ pk ) by using the error between the target output (d pk ) and the actual output (O pk ) of the output neuron and the derived error signal ( ⁇ ).
  • pk It is represented by following Formula 5.
  • the sum of output neurons (E) is accumulated in the following equation.
  • the error signal ( ⁇ pj ) for the p-th signal of the j-th hidden neuron by using the error signal ( ⁇ pk ) and the weight (w jk ) is derived from the following equation 7, the actual output (O pj ) and the error signal
  • the weight w jk between the hidden layer and the output side is derived using ( ⁇ pk ), and the derived weight w jk is expressed by the following equation (8).
  • the electric load identification unit 130 uses the backpropagation neural network to generate an electric load pattern and an electric load pattern based on the feature vector for the given current signal waveform. Extract the matching electrical load.
  • the present invention can efficiently extract a feature vector having a distinct difference in current signals for electric loads using at least one filter of a frequency band different from that used in the existing MFCC algorithm, and a modified MFCC algorithm. By doing this, the time and size of processing the feature vector for the measured current signal can be reduced.
  • the present invention determines whether an arc of an electric load identified by the feature vector for the measured current signal is generated, selects an arc mode matching the identified electric load when an arc is generated, and analyzes an arc corresponding to the selected arc mode.
  • An arc analysis unit 140 that performs arc analysis on the electric load identified by the algorithm and a power supply control unit 150 for controlling the power supply of the electric load when determining the arc failure of the identified electric load are added.
  • the arc analyzer 140 includes a plurality of arc modes and an arc analysis algorithm corresponding to each arc mode, and determines whether an arc of an electric load identified as a feature vector for a current signal supplied from the feature vector extractor 120 is generated. When the arc generation is determined, an arc mode matching the identified electric load is selected. The arc mode is selected from one of a number of arc modes including nonlinear strong load, nonlinear weak load, and preload.
  • the arc analyzer 140 performs an arc analysis on the electric load identified through the arc analysis algorithm corresponding to the selected arc mode. Accordingly, the present invention can improve the accuracy of the determination of the cause of the arc accident of the load by using the arc analysis results, it is possible to perform a safety evaluation for the identified electric load based on the arc analysis results.
  • the power supply control unit 150 generates a relay driving signal for controlling the power of the main distribution panel 10 to be supplied to the identified electric load 30 according to the arc analysis result. That is, it is inserted between the main distribution panel 10 and the plug of the electric load 30, and is operated by the relay drive signal of the arc analysis unit 140, therefore, the power supply of the main distribution panel 10 is the electrical load ( Supplied to 30 is controlled.
  • the present invention can improve the energy saving and safety of the identified electric load by performing power supply control through a relay installed in the plug of the identified electric load based on the safety evaluation result of the identified electric load.
  • FIG. 4 is a graph showing a current signal of each electric load detected from the CT sensor.
  • a current signal for each electric load such as an air conditioner, a computer, a hair dryer, a stand, a microwave oven, a refrigerator, and a television
  • the current signal measured for each electric load may be a feature vector derivation unit 120.
  • the feature vector derivation unit 120 performs a MFCC algorithm on the measured current signal to derive a feature vector for the current signal.
  • FIG. 5 is a graph comparing a feature vector extracted by a modified MFCC algorithm and a feature vector extracted by a conventional MFCC algorithm.
  • FIG. 5 is a diagram illustrating a conventional method performed using 128 and 512 row data. We can check the feature vector of the current signal for each electric load extracted by the MFCC algorithm, and the current signal for each electric load extracted by the modified MFCC algorithm performed using 26 raw data equal to the number of filters in the filter bank. We can check the feature vector for.
  • the accuracy of electric load discrimination is 97.14% when the raw data is 512 and 91.43% when the raw data is 128.
  • the number of raw data used in the existing MFCC algorithm is small, discrete signal processing and electric load pattern discrimination It has the advantage of shortening the time, but it can be seen that the accuracy is reduced.
  • the accuracy of the electric load discrimination is 100% based on the feature vectors extracted from the 26 row data. Therefore, in the modified MFCC algorithm, the number of row data is used in the existing MFCC algorithm. In spite of the smaller number, it is confirmed that the accuracy of the electric load discrimination is excellent.
  • FIG. 6 is an overall flow chart of the electric load determination method according to another embodiment of the present invention, the data acquisition step (S11, S12), feature vector derivation step (S21, S22), electric load identification step (S30), arc analysis Steps S41 to S43 and power supply cutoff steps S51 to S53 are included, and each step function corresponds to the above-described electric load determination device of 110 to 140, and thus redundant description of the functions will be omitted.
  • N discrete signals are obtained by sampling a current signal supplied from the distribution panel 10 to the electric load 30 at regular intervals.
  • the feature vector derivation step (S21, S22) performs fast Fourier transform on the N discrete signals, derives a discrete transform result, and at least one filter having a frequency band in which the derived discrete transform result does not overlap each other.
  • the feature vector for the measured current signal is derived by passing through the filter bank.
  • the electric load identification step S30 extracts an electric load having a feature vector for the measured current signal by performing a backpropagation neural network algorithm on the feature vector for the measured current signal.
  • the arc analysis steps S41 to S43 determine whether an arc of the electric load identified in the electric load identification step S30 is generated on the basis of the feature vector for the current signal measured in the feature vector derivation step S22, and as a result of the determination.
  • an arc mode matching the identified electric load is selected, and an arc analysis algorithm matching the selected arc mode is performed to perform arc analysis on the identified electric load.
  • Power supply control step (S51 ⁇ S53) generates a relay drive signal for turning off the relay 70 when the electric load identified as a result of the arc analysis is an arc failure, and the relay 70 is turned off by the relay drive signal to identify Shut off the power supplied to the electric load (30). If the arc generated from the identified electric load satisfies a predetermined condition, it is determined that the identified electric load is an arc failure, and the predetermined condition can be understood by those skilled in the art related to the present embodiment. .
  • CT sensor 70 relay
  • filter bank 130 electric load identification unit

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Data Mining & Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • General Engineering & Computer Science (AREA)
  • Power Engineering (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Automation & Control Theory (AREA)
  • Biomedical Technology (AREA)
  • Fuzzy Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Discrete Mathematics (AREA)
  • Remote Monitoring And Control Of Power-Distribution Networks (AREA)

Abstract

An electrical load determination apparatus and method using a modified MFCC are disclosed. According to a particular embodiment of the present invention, an electrical load-specific feature vector showing a clear difference in an electrical load-specific current signal waveform can be efficiently extracted by using a predetermined number of filters of frequency bands that differ from the frequency bands used in a conventional MFCC algorithm, and processing time and size can be reduced by performing a modified MFCC algorithm. In addition, the accuracy of arc accident cause determination of an electrical load identified using a feature vector of a measured current signal can also be improved by selecting an arc mode matched with the identified electrical load when the arc of the identified electrical load occurs and performing an arc analysis of the identified electrical load through an arc analysis algorithm corresponding to the selected arc mode, and the safety of the identified electrical load can be evaluated on the basis of the arc analysis result. Therefore, power supply control is performed through a relay provided to a plug of the identified electrical load according to the safety evaluation result of the identified electrical load such that the energy saving and the safety of the identified electrical load can be improved.

Description

변형된 MFCC를 이용한 전기부하 판별 장치 및 방법Apparatus and method for determining electric load using modified MFCC
본 발명은 변형된 MFCC를 이용한 전기부하 판별 장치 및 방법에 관한 것으로서, 더욱 상세하게는 기존의 MFCC 알고리즘을 이용하여 주어진 전류 신호 파형과 매칭되는 전기부하를 판별할 수 있도록 하는 기술에 관한 것이다.The present invention relates to an apparatus and method for determining an electric load using a modified MFCC, and more particularly, to a technique for determining an electric load matching a given current signal waveform using a conventional MFCC algorithm.
선행 자료(한국등록특허 제1757056호)는 기기 식별이 가능한 적응형 스마트 전력 측정 장치 및 방법이 제공된다. 이러한 기기 식별이 가능한 적응형 스마트 전력 측정 방법은 측정 대상이 되는 복수의 기기로부터 측정된 전류, 전력, 전력량 중 적어도 하나 이상의 측정 값을 수집하여 상기 측정 값의 변화를 감지하고, 상기 측정 값의 변화가 감지되는 경우 전력 또는 전류의 변이 값과 매칭되는 전력 또는 전류의 고유 값을 검색하고 상기 검색에 따라 위상을 계산하여 등록된 위상의 상기 고유 값과 매칭되는 하나의 기기를 판별하여 각 고유 값의 시간별/일별/월별 이동평균을 측정 및 저장하여 기기의 수명을 진단하는데 판단자료로 사용하고 이동평균값을 이용하여 고유 값을 항상 최근의 값으로 적용함으로써 고유 값을 이용한 기기의 식별에 정확도를 높일 수 있다.Prior material (Korean Patent No. 1757056) provides an adaptive smart power measurement apparatus and method capable of identifying the device. The adaptive smart power measurement method capable of identifying such a device collects at least one measured value of current, power, and power measured from a plurality of devices to be measured, and detects a change in the measured value and changes the measured value. If is detected, the unique value of the power or current matching the variation value of the power or current is searched and the phase is calculated according to the search to determine one device matching the unique value of the registered phase to determine the value of each unique value. By measuring and storing the moving average hourly, daily and monthly, it is used as a judgment data for diagnosing the life of the device and by using the moving average value, the eigenvalue is always applied to the latest value to increase the accuracy in identifying the device using the eigenvalue. have.
그러나, 상술한 기기 식별 장치 및 방법은 기기의 고유값을 전류나 위상 변이값으로 설정하여 기기를 판별함에 있어 상기 고유값 판별을 위한 신호 파형 인식에 대한 구체적인 방법을 제시하고 있을 뿐, 식별된 전기 기기에 대한 안전도를 평가 및 예측하는 구성이 없기 때문에 전기 기기의 사고 발생을 미연에 방지하고 사고 원인 분석이 어려운 문제점이 있었다.However, the above-described device identification apparatus and method merely present a specific method for signal waveform recognition for identifying the eigenvalue in determining the device by setting the eigenvalue of the device as a current or a phase shift value. Since there is no configuration for evaluating and predicting the safety of the device, there is a problem in that an accident of the electric device is prevented in advance and the cause of the accident is difficult to analyze.
이에 본 출원인은 주어진 전류 신호 파형에 대해 MFCC 알고리즘 및 신경회로망을 수행하여 전기부하를 식별하고 식별된 전기부하와 매칭되는 아크 분석을 수행하여 전기부하에 대한 안전도를 미리 평가 및 예측할 수 있고 예측 결과에 따라 전기부하에 전달되는 전원을 미리 차단하여 사고를 미연에 방지할 수 있는 방안을 제안하고자 한다.In this regard, the present applicant can perform an MFCC algorithm and a neural network on a given current signal waveform to identify an electric load and perform an arc analysis that matches the identified electric load to evaluate and predict the safety of the electric load in advance. Therefore, we will propose a way to prevent accidents by cutting off the power delivered to the electrical load in advance.
본 발명은 주어진 전류 신호 파형에 대해 MFCC 알고리즘 및 신경회로망을 수행하여 전기부하를 식별함으로써, 전기부하 식별의 정확도를 향상시킬 수 있고 변형된 MFCC 알고리즘을 통해 전기부하 식별하는데 드는 시간을 최소로 단축할 수 있는 변형된 MFCC를 이용한 전기부하 판별 방법 및 장치를 제공한다.The present invention can improve the accuracy of electric load identification by performing the MFCC algorithm and neural network on a given current signal waveform, and minimize the time required to identify the electric load through the modified MFCC algorithm. The present invention provides a method and apparatus for determining an electric load using a modified MFCC.
또한, 본 발명은, 식별된 전기부하에 대한 아크 발생 시 식별된 전기부하와 매칭되어 저장된 아크분석 알고리즘을 통해 아크 분석을 수행하고 아크 분석 결과에 따라 식별된 전기부하에 전원 공급을 제어함으로써, 아크 분석 결과를 전기설비 안전도 평가에 활용할 수 있고 식별된 전기부하에 대한 에너지 절감 및 안전성을 향상할 수 있는 변형된 MFCC를 이용한 전기부하 판별 방법 및 장치를 제공한다.In addition, the present invention, by performing the arc analysis through the stored arc analysis algorithm matched with the identified electric load when the arc generation for the identified electric load and by controlling the power supply to the identified electric load according to the arc analysis results, The present invention provides a method and apparatus for determining electric load using a modified MFCC that can be used for evaluating the safety of electrical installations and can improve energy saving and safety for the identified electric load.
본 발명은 The present invention
CT 센서에 의거 측정된 전류신호를 소정 간격으로 샘플링하여 소정 수의 이산 신호를 획득하는 데이터 획득부;A data acquisition unit sampling a current signal measured by the CT sensor at predetermined intervals to obtain a predetermined number of discrete signals;
상기 소정 수의 이산 신호에 대해 MFCC(Mel-Frequency Cepstral Coefficients) 일고리즘을 수행하여 전류신호 파형에 대한 특징 벡터를 도출하는 특징벡터 도출부; 및A feature vector derivation unit configured to derive a feature vector for a current signal waveform by performing a Mel-Frequency Cepstral Coefficients (MFCC) algorithm on the predetermined number of discrete signals; And
상기 전류신호 파형에 대한 특징 벡터를 신경회로망을 이용하여 특징 벡터와 매칭되는 전기부하를 추출하는 전기부하 식별부를 포함하는 것을 일 특징으로 한다. The feature vector for the current signal waveform is characterized in that it comprises an electrical load identification unit for extracting the electrical load matching the feature vector using a neural network.
바람직하게 상기 특징벡터 도출부는,Preferably, the feature vector derivation unit,
상기 소정 수의 이산 신호에 대해 고속 푸리에 변환을 수행하여 주파수 영역의 결과값을 도출하는 FFT(Fast Fourier Transform); 및A fast fourier transform (FFT) for performing a fast Fourier transform on the predetermined number of discrete signals to derive a result value in a frequency domain; And
복수의 필터로 구비되고 상기 주파수 영역의 결과값을 각 필터의 주파수 대역의 전류신호 파형을 통과시켜 전류신호 파형에 대한 특징 벡터를 도출하는 필터뱅크를 포함할 수 있다.The filter bank may include a filter bank provided with a plurality of filters and deriving a feature vector for the current signal waveform by passing a result value of the frequency domain through a current signal waveform of a frequency band of each filter.
바람직하게 상기 주파수 영역의 결과값은,Preferably the result value of the frequency domain,
다수의 전기부하 마다 서로 다른 값으로 도출될 수 있다.Different electrical loads can be derived with different values.
상기 복수의 필터는,The plurality of filters,
상호 중첩되지 아니한 주파수 대역으로 설정될 수 있다.It may be set to a frequency band that does not overlap each other.
바람직하게 상기 전기부하 식별부는,Preferably the electrical load identification unit,
전기부하 별 측정된 전류신호에 대한 특징 벡터를 토대로 학습 모델을 구축하고, 구축된 학습 모델을 이용하여 입력된 전류신호에 대한 특징 벡터와 매칭되는 전기부하를 추출하도록 구비될 수 있다. A learning model may be constructed based on the feature vector of the measured current signal for each electric load, and the electric load matching the feature vector for the input current signal may be extracted using the constructed learning model.
바람직하게 상기 장치는,Preferably the device,
상기 전류신호 파형에 대한 특징 벡터를 토대로 상기 식별된 전기부하의 아크 발생 여부를 판단하고, 상기 식별된 전기부하의 아크 발생 시 식별된 전기부하와 매칭되는 아크 모드를 선택하고 선택된 아크 모드에 해당하는 아크분석 알고리즘을 통해 식별된 전기부하의 아크 분석을 수행하는 아크 분석부를 더 포함할 수 있다.It is determined whether an arc of the identified electric load is generated based on the feature vector of the current signal waveform, selects an arc mode matching the identified electric load when an arc of the identified electric load occurs, and corresponds to the selected arc mode. The arc analysis unit may further include an arc analysis unit that performs an arc analysis of the electric load identified through the arc analysis algorithm.
바람직하게 상기 장치는,Preferably the device,
상기 식별된 전기부하의 아크 분석 결과를 토대로 아크 고장 시 식별된 전기부하에 전원 공급을 차단하는 릴레이 구동신호를 생성하여 전달하는 전원공급 제어부를 더 포함할 수 있다.The apparatus may further include a power supply control unit configured to generate and transmit a relay driving signal to cut off power supply to the identified electric load based on the arc analysis result of the identified electric load.
또한 본 발명은, 배전반과 전기부하 사이의 전기선에 설치된 CT 센서로부터 측정된 전류신호 파형으로 전기부하를 식별하는 장치에 있어서, 상기 측정된 전류신호 파형을 소정 간격으로 샘플링하여 소정 수의 이산 신호를 획득하는 데이터 획득단계; 상기 소정 수의 이산 신호에 대해 고속 푸리에 변환 후 소정 수의 필터를 통과하여 상기 측정된 전류신호 파형에 대한 특징 벡터를 도출하는 특징벡터 도출단계; 및 상기 전류신호 파형에 대한 특징 벡터에 대해 신경회로망을 통해 상기 측정된 전류신호 파형에 대한 특징 벡터와 매칭되는 전기부하를 추출하는 전기부하 식별단계를 포함하는 것을 다른 특징으로 한다. In addition, the present invention, in the device for identifying the electric load by the current signal waveform measured from the CT sensor installed in the electrical line between the switchboard and the electrical load, the measured current signal waveform is sampled at a predetermined interval to a predetermined number of discrete signals Acquiring data; A feature vector derivation step of deriving a feature vector for the measured current signal waveform after a fast Fourier transform on the predetermined number of discrete signals and passing through a predetermined number of filters; And an electric load identification step of extracting an electric load matching the characteristic vector for the measured current signal waveform through a neural network with respect to the feature vector for the current signal waveform.
바람직하게 상기 측정된 전류신호 파형에 대한 특징 벡터를 토대로 식별된 전기부하에 대한 아크 발생 여부를 판단하고, 아크 발생 판단 시 식별된 전기부하와 매칭되는 아크 모드에 해당하는 아크 분석 알고리즘으로 아크 분석을 수행하는 아크 분석단계를 더 포함할 수 있고, 식별된 전기부하에 대한 아크 분석 결과에 따라 식별된 전기부하의 아크 고장 시 식별된 전기부하의 전원 공급을 제어하는 전원공급 제어단계를 더 포함할 수 있다.Preferably, it is determined whether an arc is generated with respect to the identified electric load based on the feature vector of the measured current signal waveform, and the arc analysis is performed using an arc analysis algorithm corresponding to an arc mode matching the identified electric load when determining the arc occurrence. The method may further include performing an arc analysis step, and further comprising a power supply control step of controlling a power supply of the identified electric load in case of an arc failure of the identified electric load according to the arc analysis result of the identified electric load. have.
전술한 바와 같은 구성의 본 발명에 의하면 기존의 MFCC 알고리즘에서 사용된 주파수 대역과 다른 주파수 대역의 소정 수의 필터를 사용하여 전기부하 별 전류 신호 파형의 차이가 분명한 전기부하 별 특징 벡터를 효율적으로 추출할 수 있고, 변형된 MFCC 알고리즘의 수행으로 처리 시간 및 사이즈를 줄일 수 있다.According to the present invention having the above-described configuration, by using a predetermined number of filters of a frequency band different from the frequency band used in the conventional MFCC algorithm, it is possible to efficiently extract the characteristic vector for each electric load by which the difference of the current signal waveform for each electric load is obvious The processing time and size can be reduced by performing the modified MFCC algorithm.
또한, 측정된 전류 신호에 대한 특징 벡터를 이용하여 식별된 전기부하의 아크 발생 시 식별된 전기부하와 매칭되는 아크 모드를 선택하고 선택된 아크 모드에 해당하는 아크분석 알고리즘을 통해 식별된 전기부하의 아크 분석을 수행하여 식별된 전기 부하의 아크 사고 원인 판정의 정확도를 향상할 수 있고, 아크 분석 결과를 토대로 식별된 전기 부하에 대한 안전도 평가를 수행할 수 있다.In addition, when the arc of the identified electric load is generated by using the feature vector for the measured current signal, an arc mode matching the identified electric load is selected, and the arc of the electric load identified through the arc analysis algorithm corresponding to the selected arc mode. The analysis can be performed to improve the accuracy of the arc accident cause determination of the identified electrical loads and to perform a safety assessment for the identified electrical loads based on the arc analysis results.
따라서, 식별된 전기 부하의 안전도 평가 결과에 의거 식별된 전기 부하의 플러그에 설치된 릴레이를 통해 전원 공급 제어를 수행함으로써, 식별된 전기부하의 에너지 절감 및 안전성을 향상할 수 있다. Therefore, by performing power supply control through a relay installed in the plug of the identified electric load based on the safety evaluation result of the identified electric load, it is possible to improve energy saving and safety of the identified electric load.
본 명세서에서 첨부되는 다음의 도면들은 본 발명의 바람직한 실시 예를 예시하는 것이며, 후술하는 발명의 상세한 설명과 함께 본 발명의 기술사상을 더욱 이해시키는 역할을 하는 것이므로, 본 발명은 그러한 도면에 기재된 사항에만 한정되어 해석되어서는 아니된다.The following drawings attached in this specification are illustrative of the preferred embodiments of the present invention, and together with the detailed description of the invention to serve to further understand the technical spirit of the present invention, the present invention is a matter described in such drawings It should not be construed as limited to.
도 1은 본 발명의 실시 예에 따른 전기부하 판별 장치의 구성도이다.1 is a block diagram of an electric load determination device according to an embodiment of the present invention.
도 2 내지 제5도는 본 발명의 실시 예에 따른 전기부하 판별 장치의 각 부의 출력 신호를 보인 그래프들이다.2 to 5 are graphs showing the output signal of each part of the electric load determination device according to an embodiment of the present invention.
도 6은 본 발명의 실시 예에 따른 전기부하 판별 방법의 처리 흐름도이다.6 is a flowchart illustrating a method of determining an electric load according to an embodiment of the present invention.
아래에서는 첨부한 도면을 참고하여 본 발명의 실시 예에 대하여 본 발명이 속하는 기술 분야에서 통상의 지식을 가진 자가 용이하게 실시할 수 있도록 상세히 설명한다. 그리고 도면에서 본 발명을 명확하게 설명하기 위해서 설명과 관계없는 부분은 생략한다.Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention.
일반적인 MFCC(Mel-Frequency Cepstral Coefficients) 알고리즘을 통해 측정된 전류 신호에 대해 노이즈 소음 및 노이즈를 제거하는 전처리 과정의 수행 시 측정된 전류 신호 파형의 고주파 성분이 완화되므로 전기부하 판별의 정확도가 저하되었다.The accuracy of electric load discrimination is reduced because the high frequency component of the measured current signal waveform is alleviated when performing the preprocessing process to remove noise noise and noise with respect to the measured current signal through the general MFCC (Mel-Frequency Cepstral Coefficients) algorithm.
이에 본 발명은 측정된 전류 신호로부터 획득된 N개의 이산 신호에 대해 고속푸리에 변환을 수행하는 변형된 MFCC 알고리즘을 통해 전기부하 식별함으로써, 기존의 MFCC 알고리즘의 전처리 과정 및 후처리 과정이 생략되어 전기부하의 전류 신호 파형에 대한 특징 벡터를 도출하는 시간 및 전기부하 판별 장치의 사이즈를 줄일 수 있도록 구비된다. Accordingly, the present invention identifies the electric load through a modified MFCC algorithm that performs fast Fourier transform on N discrete signals obtained from the measured current signals, thereby eliminating pre- and post-processing of the existing MFCC algorithm. It is provided to reduce the time and the size of the electric load determination device to derive the feature vector for the current signal waveform of the.
또한, 본 발명은 서로 중첩되지 아니한 주파수 대역의 소정 수의 필터로 구비된 필터뱅크를 구비하여 측정된 전류신호에 대한 특징벡터를 추출하고 추출된 전류신호의 특징벡터에 대해 신경회로망을 수행하여 전기부하를 식별함으로써, 전기부하 판별의 정확도를 향상하도록 구성된다.In addition, the present invention includes a filter bank provided with a predetermined number of filters of the frequency band that does not overlap each other, extracts the feature vector for the measured current signal and performs a neural network on the extracted feature vector of the current signal By identifying the load, it is configured to improve the accuracy of the electric load determination.
그리고 본 발명은 복수의 아크 모드와 각 아크 모드를 수행하는 아크 분석 알고리즘으로 구비되어 식별된 전기부하의 아크 발생 시 식별된 전기부하에 해당하는 아크 모드를 선택하고 선택된 아크 모드에 해당하는 아크 분석 알고리즘으로 아크 분석을 수행함으로써, 식별된 전기 부하의 아크 사고 원인 판정의 정확도를 향상할 수 있고, 아크 분석 결과를 토대로 식별된 전기 부하에 대한 안전도 평가를 수행할 수 있고, 아크 분석 결과 아크 고장 판정 시 메인 분전반에서 전기부하로 전달되는 전원 공급을 제어함으로써, 식별된 전기부하의 에너지 절감 및 안전성을 향상하도록 구성된다.In addition, the present invention includes a plurality of arc modes and an arc analysis algorithm that performs each arc mode to select an arc mode corresponding to the identified electric load when an arc of the identified electric load is generated, and an arc analysis algorithm corresponding to the selected arc mode. By performing the arc analysis, the accuracy of the arc accident cause determination of the identified electric load can be improved, and the safety evaluation of the identified electric load can be performed based on the arc analysis result, and the arc analysis result is determined by the arc failure. By controlling the power supply delivered from the main distribution panel to the electrical load, it is configured to improve energy saving and safety of the identified electrical load.
도 1은 본 발명의 실시 예에 따른 변형된 MFCC를 이용한 전기부하 판별 장치에 대한 구성도로서, 도시된 바와 같이, 전기부하 판별 장치(100)는 데이터 획득부(110), 특징벡터 도출부(120), 및 전기부하 판별부(130)를 포함한다.1 is a block diagram of an electric load determination device using a modified MFCC according to an embodiment of the present invention, as shown, the electric load determination device 100 is a data acquisition unit 110, a feature vector derivation unit ( 120, and an electric load determination unit 130.
데이터 획득부(110)는 분전반(10)과 전기부하(30) 사이에 연결된 전기선에 설치된 CT(Current Transformer) 센서(50)를 이용하여 측정된 전류신호를 일정 간격으로 샘플링하여 복수개의 로우 데이터를 획득한다. 여기서, 로우 데이터는 전기부하의 판별을 위한 신호 처리에 사용되는 이산 신호이다. The data acquisition unit 110 samples a plurality of row data by sampling a current signal measured at a predetermined interval using a CT (Current Transformer) sensor 50 installed in an electric line connected between the distribution panel 10 and the electric load 30. Acquire. Here, the raw data is a discrete signal used for signal processing for the determination of the electrical load.
한편, 특징벡터 도출부(120)는 변형된 MFCC 알고리즘을 이용하여 복수개의 이산 신호로부터 측정된 전류신호에 대한 특징 벡터를 추출하도록 구비되고, 이에 특징벡터 도출부(120)는 FFT(Fast Fourier Transform)(121) 및 필터뱅크(123)를 포함한다. Meanwhile, the feature vector derivation unit 120 is provided to extract a feature vector for the current signal measured from the plurality of discrete signals using the modified MFCC algorithm, and the feature vector derivation unit 120 includes a fast fourier transform. ) 121 and the filter bank 123.
여기서, FFT(121)는 시간 축 상의 복수개의 이산 신호에 대해 고속푸리에 변환(FFT)을 수행하여 주파수 영역의 결과값을 도출한다. 즉, FFT(121)는 주파수 영역의 DFT(Discrete Fourier Transform) 스펙트럼을 도출하고, 획득된 DFT(Discrete Fourier Transform) 스펙트럼 형태 이산 변환 결과값은 필터뱅크(123)로 전달된다. Here, the FFT 121 performs a Fast Fourier Transform (FFT) on the plurality of discrete signals on the time axis to derive a result value in the frequency domain. That is, the FFT 121 derives the Discrete Fourier Transform (DFT) spectrum of the frequency domain, and the obtained Discrete Fourier Transform (DFT) spectrum form discrete transform result is transmitted to the filter bank 123.
도 2는 FFT(121)의 이산 변환 결과값에 대한 그래프를 보여준다. 즉, 도 2을 참조하면, 다양한 전기부하 각각에 대해 주파수 영역으로 변환된 이산 변환 결과값을 확인할 수 있다.2 shows a graph of the discrete conversion result of the FFT 121. That is, referring to FIG. 2, it is possible to check the discrete conversion result converted into the frequency domain for each of various electric loads.
한편, 필터뱅크(123)는 서로 중첩되지 아니한 주파수 대역을 가지는 적어도 하나 이상의 필터로 구비되고, 이산 변환 결과값은 적어도 하나 이상의 필터를 통과하여 전류 신호에 대한 특징 벡터를 도출한다. On the other hand, the filter bank 123 is provided with at least one filter having a frequency band that does not overlap each other, the discrete conversion result is passed through the at least one filter to derive a feature vector for the current signal.
여기서, 고속 푸리에 변환 결과에 따르면 각 전기부하는 10~10000 Hz 범위에서 서로 다른 결과값을 가지므로, 각 필터의 주파수 대역은 각 전기부하 별로 상호 중첩되지 않도록 설정 및/또는 변경된다.Here, according to the fast Fourier transform result, since each electric load has a different result value in the range of 10 to 10000 Hz, the frequency band of each filter is set and / or changed so as not to overlap each other for each electric load.
이에 상기 주파수 영역의 결과값들은 필터뱅크(123)의 적어도 하나의 필터를 통과하여 한 프레임 당 M개의 계수값들을 추출하고, 한 프레임 당 M개의 계수값은 전류 신호에 대한 특징 벡터로 출력된다. Accordingly, the result values of the frequency domain pass through at least one filter of the filter bank 123 to extract M coefficient values per frame, and M coefficient values per frame are output as a feature vector for the current signal.
도 3은 도 2에 도시된 10~10000 Hz 범위의 각 전기부하 별 이산 변환 결과값에 대해 적어도 하나의 필터를 통과한 각 전기부하 별 전류신호에 대한 특징 벡터를 보인 그래프이다. 도 3을 참조하면, 필터뱅크(123)는 서로 중첩되지 아니한 주파수 대역을 가지는 26개의 필터로 구비되고, 각 필터의 주파수 대역을 통과된 에어컨, 컴퓨터, 헤어드라이어, 조명, 전자레인지, 냉장고, 텔레비전 등 각 전기부하 별 전류신호에 대한 특징 벡터를 확인할 수 있다. 그리고, 각 전기부하 별 전류신호에 대한 특징 벡터는 전기부하 식별부(130)에 전달된다. FIG. 3 is a graph showing a feature vector for a current signal for each electric load that has passed through at least one filter for the discrete conversion result for each electric load in the range of 10 to 10000 Hz shown in FIG. 2. Referring to FIG. 3, the filter bank 123 includes 26 filters having frequency bands that do not overlap each other, and includes an air conditioner, a computer, a hair dryer, lighting, a microwave oven, a refrigerator, and a television that have passed the frequency band of each filter. The feature vector for the current signal for each electric load can be checked. In addition, the feature vector for the current signal for each electric load is transmitted to the electric load identification unit 130.
전기부하 식별부(130)는 측정된 전류신호에 대한 특징 벡터를 제공받아 신경회로망을 통해 학습 수행하여 전기부하를 식별한다. The electric load identification unit 130 receives the feature vector of the measured current signal and learns through the neural network to identify the electric load.
즉, 전기부하 식별부(130)는 측정된 전류 신호에 대한 특징 벡터로 이루어진 전기부하 패턴에 따라 전기부하를 분류하여 학습 모델을 구축하고, 구축된 학습 모델를 통해 주어진 전류신호에 대한 특징 벡터와 매칭되는 전기부하 패턴의 전기부하를 추출한다. 이때 전기부하 판별을 위한 학습은 역전파 신경회로망으로 수행된다. That is, the electric load identification unit 130 constructs a learning model by classifying the electric load according to the electric load pattern consisting of the characteristic vector of the measured current signal, and matches the feature vector for the given current signal through the constructed learning model. Extract the electrical load of the electrical load pattern. At this time, the learning for electric load determination is performed by the back propagation neural network.
여기서, 역전파 신경회로망은 외부에서 입력되는 특징 벡터를 신경망에 전달하고 신경망은 특징 벡터의 오차를 연산하며, 오차가 줄이기 위한 가중치를 조정하는 알고리즘이다. 여기서 오차는 평균제곱오차(MSE: Means Square Error)을 사용하여 도출된다.Here, the back propagation neural network transmits a feature vector input from the outside to the neural network, and the neural network calculates an error of the feature vector and adjusts a weight to reduce the error. The error is derived using Means Square Error (MSE).
즉, 역전파 신경회로망은 (1) 초기화 단계, (2) 특징벡터 전방향 진행 단계, (3) 오차 신호의 역전파 및 가중치 수정 단계 및 (4) 반복 수행단계 순으로 수행된다.That is, the back propagation neural network is performed in the following order: (1) initialization step, (2) forward forward feature vector, (3) back propagation and weight correction step of error signal, and (4) repetition step.
즉, (1) 초기화 단계에서, 입력출력 뉴론의 수를 설정하고 필요한 은닉 츠의 수와 은닉 뉴론의 수를 설정하고, 은닉 층의 수는 1로 설정되고 가중치 및 임계치 역시 제조자에 의해 설정된 임의의 값으로 초기화된다.That is, in the (1) initialization step, the number of input output neurons is set, the number of necessary hidden bits and the number of hidden neurons are set, the number of hidden layers is set to 1, and the weight and threshold are also set by the manufacturer. Initialized to a value.
(2) 특징벡터 전방향 진행 단계는 입력과 목표 출력을 설정하고, 필요에 따라 -1과 1 사이 또는 0과 1 사이로 정규화되며, 입력 벡터(Xp -)와 출력 벡터(Dp)는 다음 식 1로 표현될 수 있다. 즉, 입력 벡터(Xp -)를 이용하여 도출된 은닉 층의 뉴론 j의 입력 값의 합(Netpj)은 식 2로 나타낸다. 즉, 시그모이드(sigmoid) 함수를 뉴론의 활성함수로 사용하여 은닉 층의 뉴론(H개)의 실제 출력(Opj)는 식 3으로 도출된다.2, feature vector forward progress step is to set the input and target output, and is normalized between -1 and 1 or between 0 and 1. If desired, the input vector (X p -) and an output vector (D p) are: It can be expressed by Equation 1. That is, the sum Net pj of input values of neurons j of the hidden layer derived using the input vector X p − is represented by Equation 2. In other words, using the sigmoid (sigmoid) function as the active function of the neuron, the actual output (O pj ) of the neurons (H) of the hidden layer is derived from Equation 3.
Figure PCTKR2018003143-appb-I000001
… 식 1
Figure PCTKR2018003143-appb-I000001
… Equation 1
Figure PCTKR2018003143-appb-I000002
… 식 2
Figure PCTKR2018003143-appb-I000002
… Equation 2
Figure PCTKR2018003143-appb-I000003
… 식 3
Figure PCTKR2018003143-appb-I000003
… Expression 3
그리고, 은닉 층의 뉴론(H개)의 출력(Opj)는 출력 층의 뉴론의 입력으로 사용하고 출력 층의 k번째 뉴론으로 입력되는 값들의 총합(Netpk)는 다음 식 4를 만족한다.The output O pj of the neurons (H) of the hidden layer is used as an input of the neurons of the output layer, and the sum Net pk of the values input to the k-th neuron of the output layer satisfies Equation 4 below.
Figure PCTKR2018003143-appb-I000004
… 식 4
Figure PCTKR2018003143-appb-I000004
… Equation 4
(3) 오차 신호의 역전파 및 가중치 수정 단계는 출력 뉴론의 목표 출력(dpk)및 실제 출력(Opk) 사이의 오차를 이용하여 오차 신호(δpk)를 도출하고 도출된 오차 신호(δpk) 다음 식 5로 나타낸다. 이때 출력 뉴론의 합(E)는 다음 식 6으로누적된다.(3) The back propagation and weight correction step of the error signal derives the error signal (δ pk ) by using the error between the target output (d pk ) and the actual output (O pk ) of the output neuron and the derived error signal (δ). pk ) It is represented by following Formula 5. At this time, the sum of output neurons (E) is accumulated in the following equation.
Figure PCTKR2018003143-appb-I000005
… 식 5
Figure PCTKR2018003143-appb-I000005
… Equation 5
Figure PCTKR2018003143-appb-I000006
Figure PCTKR2018003143-appb-I000006
Figure PCTKR2018003143-appb-I000007
… 식 6
Figure PCTKR2018003143-appb-I000007
… Equation 6
한편, 오차 신호(δpk)와 가중치(wjk)를 이용하여 j번째 은닉 뉴런의 p 번째 신호에 대한 오차 신호(δpj)는 다음 식 7로부터 도출되고, 실제 출력(Opj)과 오차 신호(δpk)를 이용하여 은닉층과 출력측 사이의 가중치(wjk)가 도출되고, 도출된 가중치(wjk)는 다음 식 8로 표현된다.On the other hand, the error signal (δ pj ) for the p-th signal of the j-th hidden neuron by using the error signal (δ pk ) and the weight (w jk ) is derived from the following equation 7, the actual output (O pj ) and the error signal The weight w jk between the hidden layer and the output side is derived using (δ pk ), and the derived weight w jk is expressed by the following equation (8).
Figure PCTKR2018003143-appb-I000008
… 식 7
Figure PCTKR2018003143-appb-I000008
… Equation 7
Figure PCTKR2018003143-appb-I000009
… 식 8
Figure PCTKR2018003143-appb-I000009
… Equation 8
이러한 가중치(wjk)에 모멘트(momentum)항이 추가되면 가중치(wjk)의 수렴 속도는 증가된다. 또한 입력층과 은닉층 사이의 가중치도 동일한 방법으로 연산된다. 여기서, 모멘트(momentum)항이 추가된 가중치(wjk)는 다음 식 9로 표현된다.When adding such a weight moment (momentum) to the term (w jk) increases the convergence rate of the weights (w jk). The weight between the input layer and the hidden layer is also calculated in the same way. Here, the weight w jk to which the moment term is added is expressed by the following equation (9).
Figure PCTKR2018003143-appb-I000010
… 식 9
Figure PCTKR2018003143-appb-I000010
… Equation 9
(4) 반복 수행 단계는, 모든 신호에 대해 학습이 수행될 수 있도록 p+1 번째 신호에서 입력신호 전방향 진행 단계로 반복 진행되고, 모든 신호를 학습한 후에는 누적 오차(E)가 미리 정해진 기준 오차 이하인 지를 확인하고 누적 오차가 기준 오차 이하인 경우 학습을 종료하고 누적 오차가 기준 오차 이하가 아닌 경우 이미 학습된 신호를 다시 학습하도록 p=1로 설정하여 입력 신호 전방향 진행 단계로 진행한다. (4) The repetition step is to proceed repeatedly from the p + 1 th signal to the input signal omnidirectional step so that learning can be performed on all signals, and after learning all the signals, the cumulative error (E) is predetermined. If the reference error is less than the reference error, and if the cumulative error is less than the reference error, the learning is terminated, and if the cumulative error is not less than the reference error by setting p = 1 to re-learn the already learned signal proceeds to the input signal omnidirectional progression step.
이러한 임의의 전류신호 파형에 대한 특징 벡터가 전기부하 식별부(130)에 입력되면 전기부하 식별부(130)는 역전파 신경회로망을 이용하여 주어진 전류신호 파형에 대한 특징 벡터를 토대로 전기부하 패턴과 매칭되는 전기부하를 추출한다. When the feature vector for the arbitrary current signal waveform is input to the electric load identification unit 130, the electric load identification unit 130 uses the backpropagation neural network to generate an electric load pattern and an electric load pattern based on the feature vector for the given current signal waveform. Extract the matching electrical load.
이에 따라, 본 발명은 기존의 MFCC 알고리즘에서 사용된 주파수 대역과 다른 주파수 대역의 적어도 하나의 필터를 사용하여 전기부하 별 전류신호의 차이가 분명한 특징 벡터를 효율적으로 추출할 수 있고, 변형된 MFCC 알고리즘의 수행으로 측정된 전류신호에 대한 특징 벡터를 처리하는 시간 및 사이즈를 줄일 수 있다.Accordingly, the present invention can efficiently extract a feature vector having a distinct difference in current signals for electric loads using at least one filter of a frequency band different from that used in the existing MFCC algorithm, and a modified MFCC algorithm. By doing this, the time and size of processing the feature vector for the measured current signal can be reduced.
한편, 본 발명은 측정된 전류신호에 대한 특징 벡터로 식별된 전기부하의 아크 발생 여부를 판단하고, 아크 발생 시 식별된 전기부하와 매칭되는 아크 모드를 선택하며, 선택된 아크 모드에 해당하는 아크 분석 알고리즘으로 식별된 전기부하에 대한 아크 분석을 수행하는 아크 분석부(140)와 식별된 전기부하의 아크고장 판정 시 전기부하의 전원 공급을 단속하는 전원공급 제어부(150)를 추가한다. Meanwhile, the present invention determines whether an arc of an electric load identified by the feature vector for the measured current signal is generated, selects an arc mode matching the identified electric load when an arc is generated, and analyzes an arc corresponding to the selected arc mode. An arc analysis unit 140 that performs arc analysis on the electric load identified by the algorithm and a power supply control unit 150 for controlling the power supply of the electric load when determining the arc failure of the identified electric load are added.
아크 분석부(140)는 다수의 아크 모드과 각 아크 모드에 해당하는 아크 분석 알고리즘을 구비하고, 특징벡터 추출부(120)로부터 공급되는 전류신호에 대한 특징 벡터로 식별된 전기부하의 아크발생 여부를 판단하고, 아크 발생 판단 시 식별된 전기부하와 매칭된 아크 모드를 선택한다. 아크 모드는 비선형 강부하, 비선형 약부하, 및 선행부하를 포함하는 다수의 아크 모드 중 하나로 선택된다. The arc analyzer 140 includes a plurality of arc modes and an arc analysis algorithm corresponding to each arc mode, and determines whether an arc of an electric load identified as a feature vector for a current signal supplied from the feature vector extractor 120 is generated. When the arc generation is determined, an arc mode matching the identified electric load is selected. The arc mode is selected from one of a number of arc modes including nonlinear strong load, nonlinear weak load, and preload.
그리고 아크 분석부(140)는 선택된 아크 모드에 해당하는 아크 분석 알고리즘을 통해 식별된 전기부하에 대한 아크분석을 수행한다. 이에 본 발명은 아크 분석 결과를 이용하여 부하의 아크 사고 원인 판정의 정확도를 향상할 수 있고, 아크 분석 결과를 토대로 식별된 전기 부하에 대한 안전도 평가를 수행할 수 있다.The arc analyzer 140 performs an arc analysis on the electric load identified through the arc analysis algorithm corresponding to the selected arc mode. Accordingly, the present invention can improve the accuracy of the determination of the cause of the arc accident of the load by using the arc analysis results, it is possible to perform a safety evaluation for the identified electric load based on the arc analysis results.
이어 전원공급 제어부(150)는 아크 분석 결과에 따라 메인 분전반(10)의 전원이 식별된 전기부하(30)에 공급되는 것을 제어하는 릴레이 구동 신호를 생성한다. 즉, 메인 분전반(10)과 해당 전기 부하(30)의 플러그 사이에 삽입 설치되고, 아크 분석부(140)의 릴레이 구동신호에 의해 작동되며, 따라서, 메인 분전반(10)의 전원이 전기부하(30)에 공급되는 것이 제어된다.Subsequently, the power supply control unit 150 generates a relay driving signal for controlling the power of the main distribution panel 10 to be supplied to the identified electric load 30 according to the arc analysis result. That is, it is inserted between the main distribution panel 10 and the plug of the electric load 30, and is operated by the relay drive signal of the arc analysis unit 140, therefore, the power supply of the main distribution panel 10 is the electrical load ( Supplied to 30 is controlled.
이에 본 발명은 식별된 전기 부하의 안전도 평가 결과에 의거 식별된 전기 부하의 플러그에 설치된 릴레이를 통해 전원 공급 제어를 수행함으로써, 식별된 전기부하의 에너지 절감 및 안전성을 향상할 수 있다.Accordingly, the present invention can improve the energy saving and safety of the identified electric load by performing power supply control through a relay installed in the plug of the identified electric load based on the safety evaluation result of the identified electric load.
도 4는 CT센서로부터 검출된 각 전기부하의 전류신호를 보인 그래프이다. 도 4를 참조하면, 에어컨, 컴퓨터, 헤어드라이어, 스탠드, 전자레인지, 냉장고 및 텔레비전의 등 각 전기부하 별 전류신호를 확인할 수 있고, 각 전기부하 별 측정된 전류신호는 특징벡터 도출부(120)로 전달되고 특징벡터 도출부(120)는 측정된 전류신호에 대해 MFCC 알고리즘을 수행하여 전류신호에 대한 특징 벡터를 도출한다.4 is a graph showing a current signal of each electric load detected from the CT sensor. Referring to FIG. 4, a current signal for each electric load, such as an air conditioner, a computer, a hair dryer, a stand, a microwave oven, a refrigerator, and a television, may be checked, and the current signal measured for each electric load may be a feature vector derivation unit 120. The feature vector derivation unit 120 performs a MFCC algorithm on the measured current signal to derive a feature vector for the current signal.
도 5는 본 발명에서 변형된 MFCC 알고리즘으로 추출된 특징 벡터와 기존의 MFCC 알고리즘으로 추출된 특징 벡터를 비교한 그래프로서, 도 5를 참조하면, 128개와 512개의 로우 데이터를 사용하여 수행된 기존의 MFCC 알고리즘으로 추출된 각 전기 부하 별 전류신호에 대한 특징 벡터를 확인할 수 있고, 필터뱅크의 필터의 수와 동일한 26개의 로우 데이터를 사용하여 수행된 변형된 MFCC 알고리즘으로 추출된 각 전기부하 별 전류신호에 대한 특징 벡터를 확인할 수 있다.FIG. 5 is a graph comparing a feature vector extracted by a modified MFCC algorithm and a feature vector extracted by a conventional MFCC algorithm. Referring to FIG. 5, FIG. 5 is a diagram illustrating a conventional method performed using 128 and 512 row data. We can check the feature vector of the current signal for each electric load extracted by the MFCC algorithm, and the current signal for each electric load extracted by the modified MFCC algorithm performed using 26 raw data equal to the number of filters in the filter bank. We can check the feature vector for.
따라서, 기존의 MFCC 알고리즘의 경우 로우 데이터가 512일 때 전기부하의 판별 정확도는 97.14%이고 128일 때 91.43% 이며, 기존 MFCC 알고리즘에서 사용된 로우 데이터의 수가 작으면 이산 신호 처리 및 전기부하 패턴 판별 시간이 단축되는 장점이 있으나 정확도가 감소됨을 확인할 수 있다.Therefore, in the case of the conventional MFCC algorithm, the accuracy of electric load discrimination is 97.14% when the raw data is 512 and 91.43% when the raw data is 128. When the number of raw data used in the existing MFCC algorithm is small, discrete signal processing and electric load pattern discrimination It has the advantage of shortening the time, but it can be seen that the accuracy is reduced.
그러나, 변형된 MFCC 알고리즘의 경우 26개의 로우 데이터로부터 추출된 특징 벡터에 의거 전기부하 판별의 정확도는 100%이므로, 변형된 MFCC 알고리즘의 경우 로우 데이터의 수가 기존의 MFCC 알고리즘에서 사용되는 로우 데이터의 수보다 적음에도 불구하고 전기부하 판별에 대한 정확도가 우수함을 확인할 수 있다. However, in the modified MFCC algorithm, the accuracy of the electric load discrimination is 100% based on the feature vectors extracted from the 26 row data. Therefore, in the modified MFCC algorithm, the number of row data is used in the existing MFCC algorithm. In spite of the smaller number, it is confirmed that the accuracy of the electric load discrimination is excellent.
도 6은 본 발명의 다른 실시 예에 따른 전기부하 판별 방법에 대한 전체 흐름도로서, 데이터 획득단계(S11, S12), 특징벡터 도출단계(S21, S22), 전기부하 식별단계(S30), 아크 분석단계(S41~S43) 및 전원공급 차단단계(S51~S53)를 포함하며 각 단계별 기능은 상술한 110 내지 140의 전기부하 판별 장치에 대응되므로 이하에서는 기능에 대한 중복 설명은 생략한다. 6 is an overall flow chart of the electric load determination method according to another embodiment of the present invention, the data acquisition step (S11, S12), feature vector derivation step (S21, S22), electric load identification step (S30), arc analysis Steps S41 to S43 and power supply cutoff steps S51 to S53 are included, and each step function corresponds to the above-described electric load determination device of 110 to 140, and thus redundant description of the functions will be omitted.
데이터 획득단계(S11, S12)는 분전반(10)에서 전기부하(30)로 공급되는 전류신호를 일정 간격으로 샘플링하여 N개의 이산 신호를 획득한다.In the data acquisition steps S11 and S12, N discrete signals are obtained by sampling a current signal supplied from the distribution panel 10 to the electric load 30 at regular intervals.
특징벡터 도출단계(S21, S22)은 상기 N개의 이산 신호에 대해 고속 푸리에 변환을 수행하고 이산 변환 결과값을 도출하며, 도출된 이산 변환 결과값을 서로 중첩되지 아니한 주파수 대역을 가지는 적어도 하나의 필터로 구비된 필터뱅크를 통과하여 측정된 전류신호에 대한 특징 벡터를 도출한다.The feature vector derivation step (S21, S22) performs fast Fourier transform on the N discrete signals, derives a discrete transform result, and at least one filter having a frequency band in which the derived discrete transform result does not overlap each other. The feature vector for the measured current signal is derived by passing through the filter bank.
그리고 전기부하 식별단계(S30)는 상기 측정된 전류신호에 대한 특징 벡터에 대해 역전파 신경회로망 알고리즘을 수행하여 측정된 전류신호에 대한 특징 벡터를 가지는 전기부하를 추출한다. The electric load identification step S30 extracts an electric load having a feature vector for the measured current signal by performing a backpropagation neural network algorithm on the feature vector for the measured current signal.
아크 분석단계(S41~S43)는 특징벡터 도출단계(S22)에서 측정된 전류신호에 대한 특징 벡터를 토대로 전기부하 식별단계(S30)에서 식별된 전기부하의 아크 발생 여부를 판단하고, 판단 결과 상기 전기부하에 아크 발생 시 상기 식별된 전기부하와 매칭되는 아크 모드를 선택하며, 선택된 아크 모드에 매칭된 아크 분석 알고리즘을 수행하여 상기 식별된 전기부하에 대한 아크 분석을 수행한다.The arc analysis steps S41 to S43 determine whether an arc of the electric load identified in the electric load identification step S30 is generated on the basis of the feature vector for the current signal measured in the feature vector derivation step S22, and as a result of the determination. When an arc occurs in the electric load, an arc mode matching the identified electric load is selected, and an arc analysis algorithm matching the selected arc mode is performed to perform arc analysis on the identified electric load.
전원공급 제어단계(S51~S53)는 상기 아크 분석 결과 식별된 전기부하가 아크 고장인 경우 릴레이(70)를 오프시키기 위한 릴레이 구동신호를 생성하며 상기 릴레이 구동신호로 릴레이(70)가 오프되어 식별된 전기부하(30)로 공급되는 전원을 차단한다. 여기서 식별된 전기부하에서 발생한 아크가 기 정해진 조건을 만족하는 경우 식별된 전기부하가 아크 고장인 것으로 판단하며, 기 정해진 조건은 본 실시 예와 관련된 기술분야에서 통상의 지식을 가진 자라면 이해할 수 있다.Power supply control step (S51 ~ S53) generates a relay drive signal for turning off the relay 70 when the electric load identified as a result of the arc analysis is an arc failure, and the relay 70 is turned off by the relay drive signal to identify Shut off the power supplied to the electric load (30). If the arc generated from the identified electric load satisfies a predetermined condition, it is determined that the identified electric load is an arc failure, and the predetermined condition can be understood by those skilled in the art related to the present embodiment. .
이상에서는 본 발명의 바람직한 실시 예를 참조하여 설명하였지만, 해당 기술 분야의 숙련된 당업자라면 하기의 특허 청구범위에 기재된 본 발명의 사상 및 영역으로부터 벗어나지 않는 범위 내에서 본 발명을 다양하게 수정 및 변경시킬 수 있음을 이해할 수 있을 것이다.Although the above has been described with reference to a preferred embodiment of the present invention, those skilled in the art will be able to variously modify and change the present invention without departing from the spirit and scope of the invention described in the claims below. It will be appreciated.
[부호의 설명][Description of the code]
10 : 분전반 30 : 전기부하10: distribution panel 30: electric load
50 : CT 센서 70 : 릴레이50: CT sensor 70: relay
100 : 전기부하 판별 장치 110 : 데이터 획득부100: electric load determination device 110: data acquisition unit
120 : 특징벡터 도출부 121 : FFT120: feature vector derivation unit 121: FFT
123 : 필터뱅크 130 : 전기부하 식별부123: filter bank 130: electric load identification unit
140 : 아크 분석부 150 : 전원공급 제어부140: arc analysis unit 150: power supply control unit

Claims (10)

  1. CT 센서에 의거 측정된 전류신호를 소정 간격으로 샘플링하여 소정 수의 이산 신호를 획득하는 데이터 획득부;A data acquisition unit sampling a current signal measured by the CT sensor at predetermined intervals to obtain a predetermined number of discrete signals;
    상기 소정 수의 이산 신호에 대해 MFCC(Mel-Frequency Cepstral Coefficients) 일고리즘을 수행하여 전류신호 파형에 대한 특징 벡터를 도출하는 특징벡터 도출부; 및A feature vector derivation unit configured to derive a feature vector for a current signal waveform by performing a Mel-Frequency Cepstral Coefficients (MFCC) algorithm on the predetermined number of discrete signals; And
    상기 전류신호 파형에 대한 특징 벡터를 신경회로망을 이용하여 특징 벡터와 매칭되는 전기부하를 추출하는 전기부하 식별부를 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.And an electric load identification unit for extracting an electric load matching the feature vector using a neural network as a feature vector for the current signal waveform.
  2. 제1항에 있어서, 상기 특징벡터 도출부는,The method of claim 1, wherein the feature vector derivation unit,
    상기 소정 수의 이산 신호에 대해 고속 푸리에 변환을 수행하여 주파수 영역의 결과값을 도출하는 FFT(Fast Fourier Transform); 및A fast fourier transform (FFT) for performing a fast Fourier transform on the predetermined number of discrete signals to derive a result value in a frequency domain; And
    복수의 필터로 구비되고 상기 주파수 영역의 결과값을 각 필터의 주파수 대역의 전류신호 파형을 통과시켜 전류신호 파형에 대한 특징 벡터를 도출하는 필터뱅크를 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.Electricity using a modified MFCC comprising a filter bank which is provided with a plurality of filters and passes through the current signal waveform of the frequency band of each filter to derive a feature vector for the current signal waveform Load determination device.
  3. 제2항에 있어서, 상기 주파수 영역의 결과값은,The method of claim 2, wherein the result value of the frequency domain is
    다수의 전기부하 마다 서로 다른 값으로 도출되는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.Electrical load determination device using a modified MFCC, characterized in that derived by different values for each of the plurality of electrical loads.
  4. 제3항에 있어서, 상기 복수의 필터는,The method of claim 3, wherein the plurality of filters,
    상호 중첩되지 아니한 주파수 대역으로 설정하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.Electrical load discrimination apparatus using a modified MFCC, characterized in that the frequency band is set not to overlap each other.
  5. 제1항에 있어서, 상기 전기부하 식별부는,The method of claim 1, wherein the electrical load identification unit,
    전기부하 별 측정된 전류신호에 대한 특징 벡터를 토대로 학습 모델을 구축하고, 구축된 학습 모델을 이용하여 입력된 전류신호에 대한 특징 벡터와 매칭되는 전기부하를 추출하도록 구비되는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.The trained model is constructed based on the feature vector of the measured current signal for each electric load, and the extracted electric load is matched to the feature vector for the input current signal using the constructed learning model. Electric load discrimination apparatus using MFCC.
  6. 제1항에 있어서, 상기 전기부하 판별장치는,The apparatus of claim 1, wherein the electric load determination device is
    상기 전류신호 파형에 대한 특징 벡터를 토대로 상기 식별된 전기부하의 아크 발생 여부를 판단하고, 상기 식별된 전기부하의 아크 발생 시 식별된 전기부하와 매칭되는 아크 모드를 선택하고 선택된 아크 모드에 해당하는 아크분석 알고리즘을 통해 식별된 전기부하의 아크 분석을 수행하는 아크 분석부를 더 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.It is determined whether an arc of the identified electric load is generated based on the feature vector of the current signal waveform, selects an arc mode matching the identified electric load when an arc of the identified electric load occurs, and corresponds to the selected arc mode. An apparatus for determining electric load using a modified MFCC, further comprising an arc analyzer configured to perform arc analysis of the electric load identified through the arc analysis algorithm.
  7. 제6항에 있어서, 상기 전기부하 판별장치는,The apparatus of claim 6, wherein the electric load determination device is
    상기 식별된 전기부하의 아크 분석 결과를 토대로 아크 고장 시 식별된 전기부하에 전원 공급을 차단하는 릴레이 구동신호를 생성하여 전달하는 전원공급 제어부를 더 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 장치.An electric load using the modified MFCC, further comprising a power supply controller configured to generate and transmit a relay driving signal to cut off power supply to the identified electric load based on the arc analysis result of the identified electric load; Discrimination device.
  8. 배전반과 전기부하 사이의 전기선에 설치된 CT 센서로부터 측정된 전류신호 파형으로 전기부하를 판별하는 장치에 있어서, In the apparatus for determining the electrical load by the current signal waveform measured from the CT sensor installed in the electrical line between the switchboard and the electrical load,
    상기 측정된 전류신호 파형을 소정 간격으로 샘플링하여 소정 수의 이산 신호를 획득하는 데이터 획득단계; Acquiring a predetermined number of discrete signals by sampling the measured current signal waveform at predetermined intervals;
    상기 소정 수의 이산 신호에 대해 고속 푸리에 변환 후 소정 수의 필터를 통과하여 상기 측정된 전류신호 파형에 대한 특징 벡터를 도출하는 특징벡터 도출단계; A feature vector derivation step of deriving a feature vector for the measured current signal waveform after a fast Fourier transform on the predetermined number of discrete signals and passing through a predetermined number of filters;
    및 And
    상기 전류신호 파형에 대한 특징 벡터에 대해 신경회로망을 통해 상기 측정된 전류신호 파형에 대한 특징 벡터와 매칭되는 전기부하를 추출하는 전기부하 식별단계를 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 방법.An electric load identification step of extracting an electric load matching the characteristic vector of the measured current signal waveform through a neural network with respect to the feature vector of the current signal waveform; How to determine.
  9. 제8항에 있어서, The method of claim 8,
    상기 전기부하 식별단계 이후에 After the electric load identification step
    상기 측정된 전류신호 파형에 대한 특징 벡터를 토대로 식별된 전기부하에 대한 아크 발생 여부를 판단하고, 아크 발생 판단 시 식별된 전기부하와 매칭되는 아크 모드에 해당하는 아크 분석 알고리즘으로 아크 분석을 수행하는 아크 분석단계를 더 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 방법.Determining whether an arc occurs with respect to the identified electric load based on the feature vector of the measured current signal waveform, and performing arc analysis with an arc analysis algorithm corresponding to the identified arc load when determining the arc generation. Electric load determination method using the modified MFCC, characterized in that it further comprises an arc analysis step.
  10. 제9항에 있어서, The method of claim 9,
    상기 아크 분석단계 이후에 After the arc analysis step
    식별된 전기부하에 대한 아크 분석 결과에 따라 식별된 전기부하의 아크 고장 시 식별된 전기부하의 전원 공급을 제어하는 전원공급 제어단계를 더 포함하는 것을 특징으로 하는 변형된 MFCC를 이용한 전기부하 판별 방법.And a power supply control step of controlling a power supply of the identified electric load in the event of an arc failure of the identified electric load according to the arc analysis result of the identified electric load. .
PCT/KR2018/003143 2018-03-19 2018-03-19 Electrical load determination apparatus and method using modified mfcc WO2019182166A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR1020180031308A KR102014751B1 (en) 2018-03-19 2018-03-19 Apparatus and method for detecting electric load using modified mfcc
KR10-2018-0031308 2018-03-19

Publications (1)

Publication Number Publication Date
WO2019182166A1 true WO2019182166A1 (en) 2019-09-26

Family

ID=67775519

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2018/003143 WO2019182166A1 (en) 2018-03-19 2018-03-19 Electrical load determination apparatus and method using modified mfcc

Country Status (2)

Country Link
KR (1) KR102014751B1 (en)
WO (1) WO2019182166A1 (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113049922A (en) * 2020-04-22 2021-06-29 青岛鼎信通讯股份有限公司 Fault arc signal detection method adopting convolutional neural network
CN113285859A (en) * 2021-05-19 2021-08-20 珠海格力电器股份有限公司 Two-bus system, control method and device, sensor and access method
CN113970680A (en) * 2021-10-22 2022-01-25 段骁晗 Arc detection method and device

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102087926B1 (en) * 2019-11-06 2020-05-27 주식회사엔클라우드 Iot detection control apparatus
KR102225817B1 (en) * 2019-11-14 2021-03-11 고려대학교 산학협력단 Apparatus and method of testing semiconductor device by using frequency output feature of semiconductor device and machine learning
KR102396564B1 (en) * 2021-10-25 2022-05-12 (주) 엔츠 Method for disaggregating loads of electric devices and apparatus using the same
CN116223908B (en) * 2023-03-02 2023-11-28 广州致远仪器有限公司 Spectrum analysis circuit structure

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH063376A (en) * 1992-06-24 1994-01-11 Matsushita Electric Works Ltd Loaded condition discrimination circuit
KR20080063128A (en) * 2006-12-27 2008-07-03 제너럴 일렉트릭 캄파니 Methods and systems for detecting series arcs in electrical systems
KR20110035087A (en) * 2009-09-29 2011-04-06 한국전력공사 Power quality monitoring system and power quality measuring method
KR101299183B1 (en) * 2012-06-22 2013-08-22 성균관대학교산학협력단 Classification method for quality of power signal using speech processing techniques, monitoring apparatus for quality of power signal using speech processing techniques and monitoring system for quality of power signal using speech processing techniques
JP2014509396A (en) * 2011-02-28 2014-04-17 エスエムエー ソーラー テクノロジー アーゲー Method and system for detecting arc faults in power circuits

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100897555B1 (en) * 2007-02-21 2009-05-15 삼성전자주식회사 Apparatus and method of extracting speech feature vectors and speech recognition system and method employing the same

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH063376A (en) * 1992-06-24 1994-01-11 Matsushita Electric Works Ltd Loaded condition discrimination circuit
KR20080063128A (en) * 2006-12-27 2008-07-03 제너럴 일렉트릭 캄파니 Methods and systems for detecting series arcs in electrical systems
KR20110035087A (en) * 2009-09-29 2011-04-06 한국전력공사 Power quality monitoring system and power quality measuring method
JP2014509396A (en) * 2011-02-28 2014-04-17 エスエムエー ソーラー テクノロジー アーゲー Method and system for detecting arc faults in power circuits
KR101299183B1 (en) * 2012-06-22 2013-08-22 성균관대학교산학협력단 Classification method for quality of power signal using speech processing techniques, monitoring apparatus for quality of power signal using speech processing techniques and monitoring system for quality of power signal using speech processing techniques

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113049922A (en) * 2020-04-22 2021-06-29 青岛鼎信通讯股份有限公司 Fault arc signal detection method adopting convolutional neural network
US11860216B2 (en) 2020-04-22 2024-01-02 Qingdao Topscomm Communication Co., Ltd Fault arc signal detection method using convolutional neural network
CN113285859A (en) * 2021-05-19 2021-08-20 珠海格力电器股份有限公司 Two-bus system, control method and device, sensor and access method
CN113970680A (en) * 2021-10-22 2022-01-25 段骁晗 Arc detection method and device

Also Published As

Publication number Publication date
KR102014751B1 (en) 2019-08-28

Similar Documents

Publication Publication Date Title
WO2019182166A1 (en) Electrical load determination apparatus and method using modified mfcc
WO2021107342A1 (en) Device and method for monitoring state of vibration of rotating equipment using deep learning-based time series analysis
WO2014115988A1 (en) Transient normalization for appliance classification, disaggregation, and power estimation in non-intrusive load monitoring
Wijayakulasooriya et al. Electric power quality disturbance classification using self-adapting artificial neural networks
WO2023063587A1 (en) Battery diagnosis method and device therefor
WO2022103185A1 (en) Battery capacity measuring device and method, and battery control system comprising battery capacity measuring device
CN109409425B (en) Fault type identification method based on neighbor component analysis
WO2023043215A1 (en) Industry facility operation control device based on standard operation level evaluation, and operation method for same
WO2021107422A1 (en) Nonintrusive load monitoring method using energy consumption data
CN112651849B (en) Method and system for identifying abnormal voltage monitoring data based on unbalanced data set
CN108303624A (en) A kind of method for detection of partial discharge of switch cabinet based on voice signal analysis
WO2020076083A1 (en) Power plant early warning device and method employing multiple prediction model
WO2018044041A1 (en) Method for learning latest data considering external influences in early warning system and system for same
CN110702966A (en) Fault arc detection method, device and system based on probabilistic neural network
WO2021040236A1 (en) Ess battery state diagnosis and lifespan prediction device and method
WO2019035544A1 (en) Face recognition apparatus and method using learning
CN110290466B (en) Floor distinguishing method, device, equipment and computer storage medium
CN114999527A (en) Transformer anomaly detection model training and deploying method and device
JP2001519911A (en) How to determine the location of partial discharge
CN112257533B (en) Perimeter intrusion detection and identification method
CN113671363A (en) High-voltage circuit breaker state identification system and method
CN117310239A (en) High-voltage power supply control method of test system
Niazazari et al. Event cause analysis in distribution networks using synchro waveform measurements
CN115754790A (en) Power supply side fault diagnosis and automatic switching method for low-voltage dual-power system
CN109583323B (en) Subway vibration signal identification method based on door control circulation unit

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18911228

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18911228

Country of ref document: EP

Kind code of ref document: A1