CN100520425C - Post-wavelet analysis treating method and device for electric power transient signal - Google Patents

Abstract

The invention discloses a wavelet analysis post-processing method and device for a power transient signal. The method is to conduct the following one or more post-processing on the wavelet coefficients after the power transient signal is analyzed by wavelet, and the result is to send power Monitoring center processing: modulus maximum extraction, singularity detection, obtaining modulus maximum points and singularity; statistical processing or cluster analysis of wavelet coefficients; neural network classification, that is, wavelet coefficients or wavelet post-processed data as neural Network input, output neural network classification results; energy analysis, to obtain energy distribution coefficient and energy fluctuation coefficient; wavelet entropy calculation, to obtain wavelet energy spectrum entropy or wavelet time entropy or wavelet singular entropy or wavelet time-frequency entropy or wavelet average entropy or wavelet distance entropy. It can effectively extract the characteristics of power transient signals, which can be applied to power system transmission line traveling wave distance measurement and fault identification, power quality analysis and equipment fault diagnosis.

Description

Post-processing method and device for wavelet analysis of power transient signal

technical field

The invention relates to a wavelet analysis post-processing method and device for power transient signals.

Background technique

In power systems, transient signals can be generated due to faults, adverse power change events, or normal operation of power electronic switching devices. In order to grasp the characteristics of possible transient signals, their impact and hazards on the power system, and early diagnosis and prediction of the operation or failure of the power grid, transmission lines, and large-scale power equipment, carry out condition-based maintenance of power grids and power equipment, etc. , long-term monitoring of power transient signals is required, and the detected and recorded signals are processed and identified to provide users or power companies with an understanding of the power transients that occur; at the same time, large-scale electrical tests, such as high-voltage power equipment, Line switching, etc., also need to collect power transient signals during the test process; in addition, to carry out accurate transmission/distribution line fault location, line selection, distance measurement, etc., further research and analysis of power transient signals are required. It can be seen that the acquisition and analysis of transient signals is of great significance to the normal operation of various equipment in the power system.

Power transient signal is an unsteady signal with short duration and strong mutation. It has obvious and prominent characteristics different from normal steady state signal. The FFT transformation suitable for stationary signal analysis has been unable to meet the requirements of power transient signal. Require. At present, most of the transient signal analysis methods are focused on the wavelet transform, an emerging tool, which can perform localized fine analysis of the signal in the time domain and frequency domain at the same time, and is especially sensitive to changes in transient mutation signals or weak signals. Wavelet transform meets the analysis requirements of transient signal mutation characteristics.

Wavelet transform can detect the transient state, but there are many coefficients after transformation. If they are all used as features, the performance of the classifier will be seriously reduced, and it is not suitable for real-time applications. Dimensionality reduction is required; It is also quite difficult to extract or classify signal components. Only from these two aspects, the post-processing method of wavelet transform of transient signal is necessary. However, there is no complete and systematic analysis and processing method and its special device at present. The existing wavelet analysis instrument lacks post-processing analysis and has a single function, so it cannot be directly used in power system transient analysis. Therefore, there is an urgent need to develop a special device with wavelet analysis and post-processing functions for power transient signals.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the existing technology and provide a wavelet post-processing method for transient signals of power systems, which can not only perform wavelet analysis on transient signals of power systems, but also systematically analyze Post-processing can effectively extract the characteristics of power transient signals, which can be applied to power system transmission line traveling wave distance measurement and fault identification, power quality analysis and equipment fault diagnosis.

The present invention solves the technical problem, and the adopted technical solution is: a post-processing method of wavelet analysis of power transient signal, which is to perform one or more post-processing on wavelet coefficients after the power transient signal is analyzed by wavelet, The results are sent to the power monitoring center for processing; the post-processing is: (1) modulus maximum value extraction, singularity detection, to obtain the modulus maximum value point and singularity; (2) statistical processing or cluster analysis of wavelet coefficients; ( 3) Neural network classification, that is, the wavelet coefficient or the post-processed data of the wavelet coefficient is used as the input of the neural network, and the neural network classification result is output; (4) Energy analysis, to obtain the energy distribution coefficient and the energy fluctuation coefficient; (5) Wavelet entropy calculation , to obtain wavelet time-frequency entropy or wavelet average entropy or wavelet distance entropy, where the specific calculation methods of wavelet time-frequency entropy, wavelet average entropy and wavelet distance entropy are:

In the following methods (a)~(c), the wavelet transform coefficients of the signal at scale a and time t are expressed as WT _x (t, a);

(a) Wavelet time-frequency entropy

Denote the vector as z(r), r=1, 2,..., m _L , then the entropy value of the vector can be defined as ${\mathrm{\Σ}}_z=-\underset{r=1}{\overset{m_L}{\mathrm{\Σ}}}{P}_{z\left(r\right)}\ln P_{z\left(r\right)},$ in $P_{z\left(r\right)}={\left|z\left(r\right)\right|}^2/{\left|\right|z\left(r\right)\left|\right|}_2^2;$

The signal x(t) discrete wavelet represents WT _x (kT, a=2 ^j ), where T is the discrete time sampling interval, and a vector sequence can be obtained along the variables k and j, then the wavelet time-frequency entropy:

${\mathrm{<span\; class="notranslate">\; WTFE</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; ;</span>$

in: ${\mathrm{E.}}_{{\mathrm{WT}}_x}\left(\mathrm{kT}\right)=-\underset{j}{\mathrm{\&Sigma;}}{P}_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)}\ln P_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)},$ ${\mathrm{E.}}_{{\mathrm{WT}}_x}(a=2^j)=-\underset{k}{\mathrm{\&Sigma;}}{P}_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)}\ln P_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)};$

(b) Wavelet mean entropy

The wavelet transform coefficient of signal x(t) at scale a and time t is WT _x (t, a), and the time-frequency plane (t, a) of signal wavelet transform is divided into N time-frequency window blocks scaled , the energy in each block is E _i (i=1, 2,...N), and the total energy of the entire time-frequency plane is $\mathrm{E.}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{E.}}_i,$ wavelet mean entropy $\mathrm{WAE}=-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}({\mathrm{E.}}_i/\mathrm{E.})\ln ({\mathrm{E.}}_i/\mathrm{E.});$

(c) Wavelet distance entropy

For the discrete wavelet representation of signal x(t) WT _x (t=kT, a=2 ^j ), where T is the discrete time sampling interval, the vector sequence WT _x (k) can be obtained along variable k; wavelet distance entropy:

$\mathrm{<span\; class="notranslate">\; WDE</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; l</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; ln</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>$

in, $d_{\mathrm{kl}}^{\&prime;}=d_{\mathrm{kl}}/\underset{k=l}{\overset{m}{\mathrm{\&Sigma;}}}\underset{l=1}{\overset{m}{\mathrm{\&Sigma;}}}{d}_{\mathrm{kl}},$ d _kl =∥ WT _x (kT)-WT _x (lT)∥, k, l=1, 2, . . . m.

Compared with the prior art, the beneficial effect of the present invention is: after performing wavelet analysis on the power transient signal to obtain the wavelet coefficient, the effective features of the power transient signal are extracted through various post-processing methods, so as to be applied to power system transmission Line traveling wave ranging and fault identification, power quality analysis and equipment fault diagnosis:

(1) Obtain the maximum value of the modulus, and the maximum value of the modulus is related to the singular point, and the points of the maximum value of the modulus on each scale converge into a singular point, so the power monitoring center can use the maximum value of the modulus to conduct Wave distance measurement, signal denoising (removing non-extreme points), and singularity more directly reflect voltage fluctuations and flickers;

(2) The coefficients after wavelet transformation are used for statistical classification analysis, which can distinguish the high and low frequencies of the collected voltage and current, and analyze the harmonic components;

(3) The neural network classification results can be used for fault diagnosis, fault line selection/classification, stability analysis, and power system dynamic safety evaluation, etc.;

(4) The energy distribution coefficient can be used to reflect the evolution rule of the transient signal wavelet transform across the characteristic subspace, and the fluctuation coefficient can describe the local fluctuation form of the transient signal wavelet transform;

(5) All kinds of entropy are related to information, information is related to control decision-making, and the state of power system is related to its development direction. Therefore, wavelet entropy is closely related to the research of power system. as a valid basis. The method of the present invention can obtain three kinds of new entropy, the specific characteristics and meanings of power system power supply quality reflected by these three kinds of entropy: wavelet time-frequency entropy

The measurement result of wavelet time-frequency entropy consists of two vectors (or sequences), the first vector has time ergodicity, and the second vector has frequency ergodicity. A large entropy value at time kT indicates that the wavelet coefficients with a wide distribution are flooded in the frequency domain space. On the contrary, a small entropy value indicates that the wavelet coefficients are concentrated in a few frequency points (or frequency bands), so the wavelet time-frequency entropy value can be Measuring the information characteristics of signals at any time and any frequency can classify different signals, and has good application potential in the field of fault detection and diagnosis.

wavelet mean entropy

The more uniform the energy distribution, the greater the wavelet average entropy, and vice versa. It is applied to power system fault detection. When the system (such as transmission line) is normal, the current and voltage mainly contain fundamental waves and harmonics, and the energy is distributed on a few frequency points, which is relatively uneven. Therefore, the average wavelet entropy is small. When When a fault occurs, the current and voltage contain high-frequency transients with a wide distribution in the frequency domain, so the entropy will increase.

wavelet distance entropy

The wavelet distance entropy reflects the difference of information at different scales and at different times, so the difference between the wavelet distance entropy of the transient signal and the steady state signal is large, and the entropy will increase when the transient state occurs.

In a word, the method of the present invention can make the power monitoring center obtain a comprehensive and complete evaluation for the power supply quality, system stability and fault analysis of the power system by comprehensively performing the above analysis, and take corresponding measures accordingly to improve the power supply quality and system stability .

Another object of the present invention is to provide a special device for the post-processing method of the wavelet analysis of the power transient signal.

The special device for the post-processing method of wavelet analysis of power transient signal of the present invention is composed of: the background computer is connected with the front-end data acquisition processing unit through the USB interface, and is connected with the GPS synchronous clock through the general serial port; at the same time, the front-end data acquisition The processing unit is connected with the GPS synchronous clock through the serial port.

The front-end data acquisition and processing unit collects the power transient signal, and sends it to the background computer through the USB interface, and the background computer performs wavelet analysis and post-processing on the received data according to the method of the present invention, and the GPS synchronous clock is used for the front-end data acquisition and processing. Time synchronization between the unit and the background computer. Therefore, a special high-efficiency power transient signal wavelet analysis post-processing method device is provided for the power system.

The above-mentioned pre-data acquisition and processing unit is composed of: the input port of the analog shaping module is connected to the power grid transformer, the output port is connected to the input port of the voltage and current high-speed acquisition and processing module; the output port of the synchronous clock interface module is the serial port of the voltage and current high-speed acquisition and processing module Connected, the input port is connected to the synchronous clock.

The analog quantity shaping module converts the voltage and current signals of the power grid to a voltage range suitable for A/D acquisition, and the voltage and current high-speed acquisition and processing module can collect the signals sent by the analog quantity shaping module at high speed, which meets the requirements of the present invention for high-speed acquisition of power transient signals Require.

The core chip of the above-mentioned pre-data acquisition and processing unit is the TMS320F2812 digital signal processing chip. The use of the mature high main frequency dedicated data acquisition and processing chip ensures reliable performance and easy implementation of the device of the invention.

The USB interface chip of the above-mentioned pre-data acquisition and processing unit is CY7C68001. The interface chip is used to ensure that the interface between the front-end data acquisition and processing unit and the back-end computer complies with the USB2.0 standard, and the data transmission speed can reach 480Mbps, which meets the high transmission rate requirements of the front-end data acquisition and processing unit and the back-end computer.

Description of drawings

Fig. 1 is a block diagram of the hardware structure of the embodiment of the present invention.

Fig. 2 is a schematic diagram of the composition of the pre-data acquisition and processing unit of the embodiment of the invention.

Fig. 3 is a background PC wavelet analysis display interface of the embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

Example

A specific implementation of the present invention is: a post-processing method of wavelet analysis of power transient signal, after the power transient signal is analyzed by wavelet, and then one or more of the following post-processing is performed on the wavelet coefficients, the result is power transmission Monitoring center processing:

(1) Modulus maximum value extraction and singularity detection to obtain the modulus maximum point and singularity;

(2) Statistical processing or cluster analysis of wavelet coefficients;

(3) Neural network classification, the wavelet coefficient or the post-processed data of the wavelet coefficient are used as the input of the neural network, and the neural network classification result is output;

(4) energy analysis, obtain energy distribution coefficient and energy fluctuation coefficient;

(5) Calculate wavelet entropy to obtain wavelet energy entropy or wavelet time entropy or wavelet singular entropy or wavelet time-frequency entropy or wavelet average entropy or wavelet distance entropy.

The specific algorithm of the above post-processing method is:

(1) Modulus maximum value extraction and singularity detection

The first wavelet transformation: perform wavelet transformation on the collected current and voltage signals i(t), u(t), and obtain the transformed function W _f (s, t).

The second step of modulus maximum value extraction: if for any point t belonging to a certain neighborhood of t ₀ , |W _f (s ₀ , t)|≤|W _f (s ₀ , t ₀ )|, it is called (s ₀ , t ₀ ) is the modulus maximum point of wavelet transform.

为在t₀点的窗信号离散采样

的小波分解，使 $\left|d_k^j\right|\&le;{c2}^{-j(1/2+\mathrm{\&alpha;})},$ j＝1，2，…MThe third step is the singularity index of the window: that is, the problem of finding the best c and α. remember

Discretely sample the window signal at point t ₀

The wavelet decomposition of $\left|d_k^j\right|\&le;{c2}^{-j(1/2+\mathrm{\&alpha;})},$ j=1, 2,...M

The solution to α is carried out in the following three steps:

a) Find the discrete small decomposition of the signal $d_k^j=\max \left|d_k^j\right|>0,$ The problem becomes to find the best c and α such that $d_j^{*}\&le;{c2}^{-j(1/2+\mathrm{\&alpha;})},j=\mathrm{1,2},\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;m$

b) seek $b_j^{*}=\log d_j^{*},$ And record b=logc, then have $j(1/2+\mathrm{\&alpha;})+b_j^{*}\&le;b,j=\mathrm{1,2},\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;m$

c) To solve the above inequalities, let $j(1/2+\mathrm{\&alpha;})+b_j^{*}+{\mathrm{\&beta;}}_j=b,$ but ${\mathrm{\&beta;}}_j=b-j(1/2+\mathrm{\&alpha;})-b_j^{*},$ The problem becomes to find α and b such that they satisfy $\min \underset{j}{\mathrm{\&Sigma;}}{\mathrm{\&beta;}}_j^2=\underset{j}{\mathrm{\&Sigma;}}{[b-j(1/2+\mathrm{\&alpha;})-b_j^{*}]}^2,j=\mathrm{1,2},\&Center\; Dot;\&CenterDot;\&Center\; Dot;m$

Using the method of least squares, we can find $\mathrm{\&alpha;}=\frac{\mathrm{\&Sigma;j\&Sigma;}{b}_j^{*}-\mathrm{M\&Sigma;}{\mathrm{jb}}_j^{*}}{\mathrm{M\&Sigma;}{j}^2-{\left(\mathrm{\&Sigma;j}\right)}^2}-\frac{1}{2},j=\mathrm{1,2},\&Center\; Dot;\&CenterDot;\&CenterDot;m$

(2) Statistical processing and cluster analysis of wavelet transform coefficients.

Refer to the commonly used methods in engineering to calculate some statistical characteristics of the wavelet coefficient D _j (k) at each scale, such as:

a) average value: $A_{{\mathrm{VG}}_j}=\underset{k}{\mathrm{\&Sigma;}}{\mathrm{D.}}_j\left(k\right)$

b) Absolute mean: $A_{{\mathrm{VG}}_j}=\underset{k}{\mathrm{\&Sigma;}}\left|{\mathrm{D.}}_j\left(k\right)\right|$

c) Variance: ${\mathrm{\&sigma;}}_j=\frac{1}{k}\underset{k}{\mathrm{\&Sigma;}}{({\mathrm{D.}}_j\left(k\right)-A_{{\mathrm{VG}}_j})}^2$

d) Willison amplitude number: $W_{{\mathrm{AMP}}_j}=\underset{k}{\mathrm{\&Sigma;}}\mathrm{sgn}\left(|{\mathrm{D.}}_j\left(k\right)-{\mathrm{D.}}_j(k+1)|\right)$

e) Number of zero crossings: $Z_{C_j}=\underset{k}{\mathrm{\&Sigma;}}\mathrm{sgn}\left\{[{-\mathrm{D.}}_j\left(k\right)-{\mathrm{\&theta;}}_j][{\mathrm{D.}}_j(k+1)-{\mathrm{\&theta;}}_j)\right]\}$

(3) Neural network classification

x _j is the input signal, representing the feature vector (usually wavelet coefficient or other post-processing results), s _i is the external input signal, u _i is the internal state of the neuron, θ _i is the threshold, the model can be described as

When the neuron has no internal state, y _i = u _i , h = f, where f is called the activation function.

(4) Energy analysis

For the collected current and voltage, the energy of each scale after wavelet transformation can be directly obtained by the square of its wavelet coefficient. Let d _{j, k} be the decomposition coefficient of the signal on the wavelet, and based on the energy of each scale of the orthogonal wavelet transform, the energy distribution coefficient and the energy fluctuation coefficient are defined as follows:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; [</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{\mathrm{<span\; class="notranslate">\; N</span>}}$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; [</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; [</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2,3</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; J</span>}$

(5) Wavelet entropy calculation: In the following wavelet entropy calculation methods (a)~(c), the wavelet transform coefficient of the signal at scale a and time t is expressed as WT _x (t, a); methods (d)~(f) , the signal x(t) is at scale j, and the discrete wavelet coefficients of multiresolution analysis are expressed as D={d _j (k), k=1,...n, j=1, 2...,m}, ${\mathrm{E.}}_j=\underset{k}{\mathrm{\&Sigma;}}{\left|d_j\left(k\right)\right|}^2$ is the wavelet energy spectrum on scale j.

(a) Wavelet time-frequency entropy

Denote the vector as z(r), r=1, 2...m _L , then the entropy value of the vector can be defined as ${\mathrm{E.}}_z=-\underset{r=1}{\overset{m_L}{\mathrm{\&Sigma;}}}{P}_{z\left(r\right)}\ln P_{z\left(r\right)},$ in $P_{z\left(r\right)}={{\left|z\left(r\right)\right|}^2/\left|\right|z\left(r\right)\left|\right|}_2^2.$

For the discrete wavelet representation of the aforementioned signal WT _x (kT, a=2 ^j ), where T is the discrete time sampling interval, it is essentially a two-dimensional matrix, and a vector sequence can be obtained along the variables k and j, so the wavelet time-frequency entropy can be defined The measure W _TFE (Wavelet Time-Frequencies Entropy) is:

${\mathrm{<span\; class="notranslate">\; WTFE</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>$

in ${\mathrm{E.}}_{{\mathrm{WT}}_x}\left(\mathrm{kT}\right)=-\underset{j}{\mathrm{\&Sigma;}}{P}_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)}\ln P_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)}$

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}\mathrm{<span\; class="notranslate">\; ln</span>}{\mathrm{<span\; class="notranslate">\; P</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}$

(b) Wavelet mean entropy

Let the wavelet transform coefficients of signal u(t), i(t) at scale a and time _t0 be WT(t, a), divide the time-frequency surface (t, a) of signal wavelet transform into N scale-scaled Time-frequency window block kT-aΔ _ψ , KT+aΔ _ψ ]×[ω ^* /a-Δ _ω /a, ω ^* /a+Δ _ω /a], (Δ _ψ , Δ _ω are base wavelet time domain radius and frequency domain radius, ω ^* is the frequency domain center of the base wavelet, T is the time discrete interval, k is 0, 1...N discrete sequence), and the energy in each regional block is E _i (i=1, 2, ....N), the total energy of the entire time-frequency plane is $\mathrm{E.}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{E.}}_i,$ Normalize the energy of each block to get P _i =E _i /E, so we have $\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{P}_i=1,$ It meets the normalization condition of information entropy, so the wavelet average entropy (WAE) is defined as:

${\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; AE</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; ln</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; )</span>$

(c) Wavelet distance entropy

For the power transient signal discrete wavelet representation WT _x (kT, a=2 ^j ), where T is the discrete time sampling interval, essentially a two-dimensional matrix, the vector sequence WT _x (k) can be obtained along the variable k, so refer to the correlation The definition of distance, the calculation method of introducing information, can define the wavelet distance entropy measure W _DE (Wavelet Distance Entropy) as:

$\mathrm{<span\; class="notranslate">\; WDE</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; l</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; ln</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}^{<span\; class="notranslate">\; \&prime;</span>}$

in, $d_{\mathrm{kl}}^{\&prime;}=d_{\mathrm{kl}}/\underset{k=l}{\overset{m}{\mathrm{\&Sigma;}}}\underset{l=1}{\overset{m}{\mathrm{\&Sigma;}}}{d}_{\mathrm{kl}},$ d _kl =∥ WT _x (kT)-WT _x (lT)∥, k, l=1, 2, . . . m.

(d) Wavelet spectral entropy

U(t) and i(t) are transformed by orthogonal wavelet to obtain wavelet energy spectrum E=E ₁ , E ₂ ,...,E _m on m scales. The total signal power E is equal to the sum of the component power E _j . Let P _j =E _j /E, then $\underset{j}{\mathrm{\&Sigma;}}{p}_j=1,$ in ${\mathrm{E.}}_j=\underset{k}{\mathrm{\&Sigma;}}{\left|{\mathrm{D.}}_j\left(k\right)\right|}^2,$ Then the wavelet entropy W _EE (WaveletEnergy Entropy) is

${\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; EE</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\mathrm{<span\; class="notranslate">\; log</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$

(e) Wavelet time entropy

The first step of wavelet transform: u(t) and i(t) are signals, and at scale j, the discrete wavelet transform coefficients of its multiresolution analysis are D={d(k), k=1, 2, ..., N }

The second step is to define the sliding window: define a sliding window on the wavelet coefficient, set the window width as ω∈N, and the sliding factor as δ∈N. Then the sliding window is W(m _t ; ω, δ)={d( k), k=1+m _t δ, ..., ω+m _t δ}, m _t =1, 2, ..., M.

The third step is to calculate the wavelet entropy: let p ^mt (Z _l ) represent the probability that the wavelet coefficient d(k)∈W(m _t ; ω, δ) falls in the interval Z _l , according to the classical probability theory, d(k)∈W (m _t ; ω, δ) is the value between the number of Z _l and the total number of wavelet coefficients ω in W(m _t ; ω, δ). So there is the following wavelet time WTE (Wavelet Time Entropy):

${W_{\mathrm{TE}}}^j\left(m_t\right)=-\underset{i=1}{\overset{\mathrm{\&omega;}}{\mathrm{\&Sigma;}}}{p}^{m_t}\left(Z_l\right)\log \left(p^{m_t}\left(Z_l\right)\right)$ m _t =1, 2, . . . M

Where M=(Nw)/δ∈N, for each scale, its WTE ^j (m _t ) can be calculated correspondingly, m _t =1, 2,...M _j .

(f) Wavelet singular entropy

The wavelet decomposition of signals u(t) and i(t) at j (j=1, 2,..., m) scales is D _j (n), then the decomposition results at m scales can form an m×n Matrix D _m×n =U _m×1 A _l+l V _l×n . Wherein the main diagonal elements of the diagonal matrix Λ are formed by λ _i (i=1, 2, ..., l),

Arrange λ ₁ ≥λ ₂ ≥...λ _l ≥0 in descending order, which is the singular value of the wavelet transformation result matrix D _m× . So the j-th order incremental wavelet singular entropy can be calculated by the following formula:

${\mathrm{<span\; class="notranslate">\; \&Delta;p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; l</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; l</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

Then the wavelet singular entropy: $W_{{\mathrm{SE}}_k}=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&Delta;p}}_i.$

Fig. 1 shows, the special device of a kind of power transient signal wavelet analysis post-processing method of the present embodiment, it is composed of: background computer PC is connected with pre-data acquisition processing unit DSP through USB interface, and is connected with GPS through general-purpose serial port The synchronous clock GPS is connected, and at the same time, the pre-data acquisition processing unit DSP is connected with the GPS synchronous clock GPS through a serial port. The background computer PC of this embodiment is also connected with the printer P, so as to facilitate the printout of the analysis results.

This embodiment uses PC as the background machine, and the whole program can be divided into the following modules according to functions and applications: data acquisition program, display program, wavelet transformation program, wavelet post-processing program, application program, spectrum analysis and others Auxiliary program. Figure 3 shows the wavelet analyzer interface of the background PC, and the following main functions are introduced as examples:

a) Display function

This function can complete the display of the waveform according to the predetermined sampling rate, the total number of channels, the set number of data points to be read each time, and the unit data length; specifically, it can complete the following tasks: ① Press the selected The signal channel displays its waveform, ②Adjust the scrolling step of the waveform, ③Pause the scrolling of the waveform, ④When the display is paused, the waveform can be zoomed in and out for careful observation, ⑤Indicate the amplitude and time of a certain point by moving the red cursor . This series of functions also provides convenience for the selection of wavelet transform signals, the selection of data points and fault location in the future.

b) Wavelet transform function

The wavelet transform function can complete continuous wavelet transform, discrete wavelet transform and wavelet packet transform, and important parameters such as wavelet type and decomposition layers can be selected before transforming.

c) Post-processing function

The post-processing functions designed by the software include: modulus maximum extraction, singularity detection, statistical processing or cluster analysis of wavelet transform coefficients, wavelet neural network classification, energy distribution feature analysis based on wavelet analysis, wavelet entropy calculation and other modules.

d) Application analysis function

Based on the above conversion and post-processing, the background computer software is also designed with the following application analysis: traveling wave ranging, fault classification, fault line selection, power quality analysis, data compression, etc. Such as traveling wave distance measurement, by observing the above-mentioned data extracted by wavelet transform and modulus maximum value, displaying its waveform to analyze the initial traveling wave and the time when the reflected wave head at the opposite end reaches the measurement point, determine the wave velocity to measure the distance, and Displayed in the man-machine interface.

In addition, the software functions also include Fourier transform, spectrum analysis, saving, printing and help, etc.

Figure 2 shows that the front data acquisition and processing unit DSP of this embodiment is composed of: the input port of the analog shaping module A is connected to the transformer, the output port is connected to the input port of the voltage and current high-speed acquisition and processing module DAQ; the GPS synchronous clock interface The output port of the module GI is connected with the serial port of the voltage and current high-speed acquisition and processing module DAQ, and the input port is connected with the GPS synchronous clock.

The details of each module are as follows:

(1) Voltage and current high-speed acquisition and processing module DAQ

The acquisition processing module DAQ includes a core processing chip TMS320F2812, a current and voltage acquisition unit, and a communication unit with a background computer. The function of this module is to collect and record the current and voltage of power frequency steady state and fault transient state. The analog-to-digital conversion part of the module adopts the A/D embedded in the core processing chip TMS320F2812. The sampling frequency selection at power frequency steady state can be set according to actual needs, the sampling conversion time is 80ns, the A/D conversion accuracy of the analog-to-digital converter is 12 bits, and two channels of signals can be collected at the same time, and 16 channels of analog signals can be collected in time-sharing quantity.

The initial program is solidified in BOOT ROM, and the program running in the core processing chip TMS320F2812 describes the workflow of the front-end data acquisition and processing unit:

During normal operation, the working process of the system is as follows: the measured signal is conditioned by the analog shaping module A through linear gain amplification to input the signal into the AD converter between 0V and 3V; the AD converter is controlled by the AD clock and the internal timer of TMS320F2812 Convert the analog signal into a digital signal, write the data into the result buffer register inside the AD, and generate an interrupt, and notify the TMS320F2812 to read and process the data; TMS320F2812 calculates the effective value of the collected data or performs FFT transformation, according to the calculation The obtained result judges whether a fault has occurred, and whether to start the fault recording program.

When there is no fault, the A/D is continuously converted, but TMS320F2812 reads the sampled data, stores it in a circular storage area, and calculates the effective value of this circular storage area or performs FFT transformation, and judges whether to start Fault recording.

When a fault signal is detected, TMS320F2812 records the current address. The data recorded before the address is steady-state data, 12500 words per channel, 100k words in total; the data recorded after the address is transient data. Each channel records 32,500 words, including 250k words. These wave recording data are temporarily stored in the extended off-chip storage space. Waiting for the background computer PC to send instructions to read these data, the front-end data acquisition and processing unit DSP and the background computer The transmission method between PCs can choose the general serial port or the USB2.0 method, and the transmission of a large amount of data chooses the method of using the USB2.0 interface, and its speed can reach 480Mbps.

In addition, the switch input and output information from the external circuit is connected to the general input and output interface (GPIO) of TMS320F2812 through the bus driver chip 74LS245, and is refreshed every 1 millisecond to detect the switch displacement and give the switch output information .

(2) GPS clock synchronization interface module (GI)

The function of the GPS clock synchronization interface module (GI) is to receive serial time information and precise second pulses from the GPS power system synchronous clock GPS to generate an accurate clock source. It is mainly composed of the central processing unit CPU and its connected data memory RAM, program memory EPROM, high-precision crystal oscillator and counter. Among them, the central processing unit CPU is the 8051XA series single-chip microcomputer produced by Philips, the model is PXAC37KFA; the data memory RAM is 628512ALP-7; the program memory EPROM is M27C64; the high-precision crystal oscillator is a 33Z MHz crystal oscillator; the counter is a 74LS161 circuit.

The working process of the program solidified in the program memory EPROM and running in the 8051XA microcontroller is as follows:

During normal operation, the CPU will receive accurate GPS clock information once a second through the serial port. The precise second pulse generated by the synchronous clock of the GPS power system is less than 1 microsecond, and the second pulse is frequency-divided into a microsecond pulse by the counter circuit.

When the detected equipment fails, the start signal will trigger the counter circuit to stop pulse counting and notify the CPU. At this time, CPU (8051XA) will detect the start signal and record the exact trigger moment. This is the exact microsecond time the failure occurred.

The CPU further combines the year, month, day, hour, minute, and second information received every second with the microsecond moment just now, and will obtain complete fault occurrence time information.

The CPU will also use the serial port to transmit this time information to the background computer for printout.

The core chip of the pre-data acquisition and processing unit (DSP) of this embodiment is a TMS320F2812 digital signal processing chip.

TMS320F2812 is a new product launched recently. It is a 32-bit processor with a high-performance static CMOS structure, which can provide 150 million instructions per second (MIPS), single-cycle 32*32-bit MAC function, and 128KW on-chip flash memory. 1K word OTP ROM, 4K word BOOT ROM, 18K word SARAM and on-chip 12-bit analog-to-digital converter (ADC). The off-chip memory is extended by 512k words of RAM.

The USB interface chip of the pre-data acquisition processing unit (DSP) of this embodiment is CY7C68001.

CY7C68001 is a USB2.0 interface chip of Cypress Company, which integrates a USB2.0 transceiver and a USB2.0 serial interface engine SIE; its transmission rate can reach up to 480Mbps, and there are 4K bytes of FIFO resources inside. The phase-locked loop also has synchronous and asynchronous FIFO interfaces.

The technical index of present embodiment device:

1. Main processor: TMS320F2812, main frequency 150MHz;

2. SRAM: On-chip: 18K×16-bit, 0 wait

    Off-chip: 512K×16-bit, 12ns wait;

3. FLASH: On-chip: 128K×16-bit, 36ns wait;

4. ROM: On-chip Boot ROM: 4K×16-bit

On-chip OTPROM: 1K×16—bit 36ns wait;

5. A/D: On-chip 2×8 channels, 12-bit, 80ns conversion time, 0～3V range

      Record the number of electrical quantities:

      ① Acquisition of 4-way voltage (three-phase voltage, zero-sequence voltage) data;

② Collect data of 4 channels of current (three-phase current, zero-sequence current);

③Sampling frequency can be adjusted at any time according to actual needs, up to 12.5Msps;

6. Asynchronous serial port: 2 channels, RS232/RS422/RS485 can be configured

Transmission rate: RS232: 1Mbaud; RS422/RS485: 9.375Mbaud;

7. USB bus: 1 channel, in line with USB2.0 specification, the highest transmission rate: 480Mbps;

8. The linearity of current transformer and voltage transformer is better than 0.1%, fully resin sealed, high isolation, strong impact resistance, small and light, and can be directly welded on the printed circuit board;

9. Voltage

Effective input of rated voltage: 220V, 50HZ;

Measuring range: 5% to 150% of the nominal value;

Load: less than or equal to 0.1VA;

Accuracy: ±0.1% within the measuring range;

Continuous overvoltage: 200% of nominal value;

Response frequency: the frequency response of 50HZ system is 45HZ to 1200HZ;

10. Current

Effective input of rated current: 5A 50HZ;

Measuring range: 2% to 150% of the nominal value;

Load: less than or equal to 0.01 ohms;

Accuracy: ±0.1% within the measuring range;

Continuous overload: 200% of nominal value;

Thermal overload: 2 times the nominal value continuously; 100 times the nominal value for 1 second;

11. Active and reactive power

Rated current and effective voltage input: 0~5A, 0~220V, 50Hz;

Measuring range: active power: 500VA0～225%; reactive power 500VA0～225%;

Accuracy: ±0.2% within the measuring range;

12. Equipment working temperature: 0～70℃

13. The analysis software running on the background computer performs a series of wavelet transformation and post-processing on the collected data:

Functions: Not only has the main functions of wavelet transform, wavelet post-processing and application analysis, but also has various auxiliary functions such as Fourier transform, spectrum analysis, waveform display, data storage and printing;

Applicable conditions: pre-unit with data acquisition or saved existing data text files;

Applicable environment: windows2000 or above systems on ordinary computers;

Analysis object: mainly used for power system transient signals, and can also be used for processing and analysis of other engineering signals;

Installation and maintenance: CD-ROM installation, software can be updated and upgraded.

The present invention can be used for transient signal acquisition, wavelet analysis and post-processing in power plants and substation electrical equipment of various voltage levels such as high-voltage transmission lines, generators, large-scale power transformers, capacitors, and condensers, and can realize power system transmission lines Traveling wave ranging and fault identification, power quality analysis and equipment fault diagnosis.

Claims (1)

1. A post-processing method for wavelet analysis of power transient signals, which is to perform one or more post-processing on the wavelet coefficients after the power transient signals are analyzed by wavelet, and then send the results to the power monitoring center for processing; the post-processing is wherein : (1) modulus maxima extraction, singularity detection, obtain modulus maxima points and singularity; (2) statistical processing or cluster analysis of wavelet coefficients; (3) neural network classification, that is, wavelet coefficients or wavelet coefficients The post-processed data is used as the input of the neural network, and the classification result of the neural network is output; (4) Energy analysis, to obtain energy distribution coefficient and energy fluctuation coefficient; (5) Wavelet entropy calculation, to obtain wavelet time-frequency entropy or wavelet average entropy or wavelet Distance entropy, the specific calculation methods of wavelet time-frequency entropy, wavelet average entropy and wavelet distance entropy are: In the following methods (a)~(c), the wavelet transform coefficient of signal x(t) at scale a and time t is expressed as WT _x (t, a); (a) Wavelet time-frequency entropy Denote the vector as z(r), r=1, 2,..., m _L , then the entropy value of the vector can be defined as ${\mathrm{E.}}_z=-\underset{r=1}{\overset{m_L}{\mathrm{\&Sigma;}}}{P}_{z\left(r\right)}\ln P_{z\left(r\right)},$ in $P_{z\left(r\right)}={\left|z\left(r\right)\right|}^2/{\left|\right|z\left(r\right)\left|\right|}_2^2;$ For a signal x(t), the discrete wavelet transform coefficient represents WT _x (kT, a=2 ^j ), where T is the discrete time sampling interval, and a vector sequence can be obtained along the variables k and j, then the wavelet time-frequency entropy: ${\mathrm{<span\; class="notranslate">\; WTFE</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; kT</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; E.</span>}}_{{\mathrm{<span\; class="notranslate">\; WT</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; ;</span>$ in: ${\mathrm{E.}}_{{\mathrm{WT}}_x}\left(\mathrm{kT}\right)=-\underset{j}{\mathrm{\&Sigma;}}{P}_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)}\ln P_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)},$ ${\mathrm{E.}}_{{\mathrm{WT}}_x}(a=2^j)=-\underset{k}{\mathrm{\&Sigma;}}{P}_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)}\ln P_{{\mathrm{WT}}_x(\mathrm{kT},a=2^j)};$ (b) Wavelet mean entropy The wavelet transform coefficient of signal x(t) at scale a and time t is WT _x (t, a), and the time-frequency plane (t, a) of signal wavelet transform is divided into N time-frequency window blocks scaled , the energy in each block is E _i (i=1, 2,...N), and the total energy of the entire time-frequency plane is $\mathrm{E.}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{E.}}_i,$ wavelet mean entropy $\mathrm{WAE}=-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}({\mathrm{E.}}_i/\mathrm{E.})\ln ({\mathrm{E.}}_i/\mathrm{E.});$ (c) Wavelet distance entropy For the discrete wavelet transform coefficient of the signal x(t), WT _x (t=kT, a=2 ^j ), where T is the discrete time sampling interval, along the variable k, the vector sequence WT _x (k) can be obtained; wavelet distance entropy: $\mathrm{<span\; class="notranslate">\; WDE</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; l</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; ln</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>$ in, $d_{\mathrm{kl}}^{\&prime;}=d_{\mathrm{kl}}/\underset{k=l}{\overset{m}{\mathrm{\&Sigma;}}}\underset{l=1}{\overset{m}{\mathrm{\&Sigma;}}}{d}_{\mathrm{kl}},$ d _k =∥ WT _x (kT)-WT _x (lT)∥, k, l=1, 2, . . . m.

Images