CN111145044A - Power quality disturbance detection method for power distribution network based on EWT and MFDE - Google Patents

Abstract

The power quality disturbance detection method of the power distribution network based on EWT and MFDE extracts PQ disturbance signals in an active power distribution network system and carries out pretreatment; performing modal decomposition on the preprocessed PQ disturbance signal by adopting empirical wavelet decomposition (EWT) to obtain an intrinsic mode function BLIMF containing characteristic information; taking an intrinsic mode function BLIMF containing characteristic information as the input of a multi-scale oscillation spreading entropy MFDE algorithm, performing spreading entropy value calculation on the intrinsic mode function BLIMF obtained by mode decomposition by using the multi-scale oscillation spreading entropy MFDE algorithm, and calculating to obtain a multi-dimensional entropy value vector of the PQ disturbance signal under each intrinsic mode function BLIMF; performing PCA dimensionality reduction on the obtained entropy value vector to serve as an input quantity of an SVM algorithm; and identifying PQ disturbance signals of the active power distribution network system containing the distributed energy. The method can accurately detect and classify each disturbance in the composite power quality disturbances, and has accurate classification and certain noise resistance.

Description

Power quality disturbance detection method for power distribution network based on EWT and MFDE

Technical Field

The invention relates to the technical field of power quality disturbance signal detection, in particular to a power quality disturbance detection method for a power distribution network based on EWT and MFDE.

Background

As distributed power generation becomes more and more integrated in power systems, power signals become more and more complex. Due to the limitation of natural conditions such as environment and climate, the output of DGs has the characteristics of randomness, fluctuation and intermittency, and may cause oscillation or flicker. Furthermore, due to the low inertia characteristics of DGs, the system is more susceptible to various disturbances. Attention should be paid to the detection and classification of the complex disturbance of the high-permeability active power distribution network. Due to the diversity and complexity of power quality disturbance signals in active power distribution networks, there are still some disadvantages in the power signal detection method.

For example, Local Mean Decomposition (LMD), which is used to extract features of non-stationary power signals in the distribution network and decompose the signals into a series of modal functions according to the frequency drop in the iterative screening process. However, due to the disadvantages of "end-point effects" and "modal aliasing", the accuracy of the calculated results is greatly affected, and this method does not have the capability of adaptive decomposition, and additional decomposition amounts need to be considered for different signals.

Variational Mode Decomposition (VMD), which is also commonly used to extract signal features of the microgrid, is also known. Although this algorithm converts the signal decomposition from a recursive filtering mode to a non-recursive filtering mode, solving some of the problems of LMD, it still cannot adaptively decompose according to different signal complexities at the cost of higher computational complexity, and the parameters must be carefully selected to obtain correct results.

In view of the defects of the existing method, the invention provides a feature extraction method combining an empirical wavelet and a composite multi-scale oscillation dispersion entropy, and the defects are overcome.

Disclosure of Invention

The invention provides a power quality disturbance detection method of a power distribution network based on EWT and MFDE, which combines empirical wavelet decomposition EWT and multi-scale oscillation dispersion entropy MFDE and provides a transient characteristic extraction method of multi-scale oscillation dispersion entropy based on empirical wavelet decomposition. An active power distribution network based on an IEEE13 node is built to serve as a test system, and an optical transformer can be used for extracting and analyzing PQ disturbance signals. The method has simple steps and accurate classification, and can improve the reliability of the power distribution network.

The technical scheme adopted by the invention is as follows:

the power quality disturbance detection method of the power distribution network based on the EWT and the MFDE comprises the following steps:

step 1: the method comprises the steps that an optical voltage transformer based on an optical fiber sensing technology is adopted to extract a PQ disturbance signal in an active power distribution network system, the PQ disturbance signal is subjected to A/D and D/A conversion to extract an analog signal, and the analog signal is preprocessed to ensure that the result is accurate;

step 2: performing modal decomposition on the preprocessed PQ disturbance signal by adopting empirical wavelet decomposition (EWT) to obtain an intrinsic mode function BLIMF containing characteristic information;

and step 3: taking an intrinsic mode function BLIMF containing characteristic information as the input of a multi-scale oscillation spreading entropy MFDE algorithm, performing spreading entropy value calculation on the intrinsic mode function BLIMF obtained by mode decomposition by using the multi-scale oscillation spreading entropy MFDE algorithm, and calculating to obtain a multi-dimensional entropy value vector of the PQ disturbance signal under each intrinsic mode function BLIMF;

and 4, step 4: carrying out PCA dimensionality reduction on the entropy vector obtained in the step 3, and using the entropy vector as an input quantity of an SVM algorithm;

and 5: and identifying PQ disturbance signals of the active power distribution network system containing the distributed energy.

In the step 1, a high-permeability active power distribution network system comprising photovoltaic, wind energy and an electric automobile is set up and used as a test system for the effectiveness of the power quality PQ disturbance detection method;

the built test system is an IEEE-13 bus power distribution network and is connected to a power grid with the rated power of 5MVA and the operating voltages of 4.16kV and 0.48 kV; the sampling frequency is 3.2kHz, the sampling time length is 0.2s, and other parameters can be inquired in tables 1 and 2. The robustness of the algorithm is tested by adding 20dB noise to the output signal;

the permeability of the active distribution network is based on the following formula (1):

in the formula, P_i,DG-nonThe installed capacity of the ith distributed power supply in the system; n is the number of distributed power supplies in the system; p_L,sumThe total power of the load in the system; total installed capacity P of distributed power supply_i,DG-non2.5MW, total load power P in the system_L,sumThe permeability of the system is 64.66% and exceeds 60% calculated by formula (1), which is 3.866MW, and accords with the description of high permeability.

In step 1, the PQ disturbance in the active power distribution network system includes the following types:

W1-W3 are fan event groups: w1 is that the fan is connected with a public power grid in a grid mode, W2 is that the fan is disconnected with the public power grid, and the W3 fan runs in an isolated island mode after being disconnected with the public power grid;

S1-S3 are photovoltaic event groups: s1 shows that the photovoltaic cell panel is connected with a public power grid in a grid mode, S2 shows that the photovoltaic cell panel is disconnected with the public power grid, and an S3 fan runs in an isolated island mode after being disconnected with the public power grid;

E1-E2 are electric vehicle event groups: e1 is the large-scale access of the electric automobile to the power grid, E2 is the large-scale off-grid of the electric automobile;

H1-H3 are fan and photovoltaic hybrid operations: h1 is that the fan system and the photovoltaic system are simultaneously connected with a public power grid in a grid mode, H2 is that the fan system and the photovoltaic system are simultaneously disconnected with the public power grid in an interruption mode, and H3 is that the fan system and the photovoltaic system are operated in an isolated island mode after being simultaneously disconnected with the public power grid.

In the step 1, preprocessing is adopted, namely the frequency spectrum of the disturbance signal after FFT is divided, and the accuracy of the EWT is directly influenced by the division result, so that the influence of factors such as inter-harmonic waves, frequency spectrum leakage and the like on the decomposition result is reduced.

The analog signal is preprocessed by the following steps:

firstly, constructing a frequency domain sequence of fundamental waves and harmonic waves of an original disturbance signal

Wherein the minimum component amplitude must not beBelow 2% of the fundamental wave, the minimum frequency is not lower than 45Hz, and psi is calculated with this set of frequencies as the center frequency_iBandwidth BW of the band pass filter_i＝10+2γf_iHz。f_iFor the frequency domain sequence of fundamental and harmonic waves of the original disturbance signal

The parameter γ ensures the minimum overlap width of two consecutive transition regions, the choice of which is based on the calculated boundary value, and γ in the present invention takes the value of 0.01.

Harmonic spectrum X_H(ω) can be calculated from the following formula (6):

in the formula, X_HIs a harmonic spectrum, omega is the input signal frequency, i is the dominant frequency order, Λ₁Is a frequency domain sequence, phi_iIs an empirical wavelet function, psi_i(ω) is a function of the scale of the experiment;

subtracting the harmonic spectrum X from the original spectrum X (ω)_H(ω) obtaining a spectrum X containing only the inter-harmonic residual_R(omega), constructing an inter-harmonic sequence with the minimum component amplitude not lower than 2% of the fundamental wave and the minimum frequency not lower than 5Hz

Will be Λ₁、Λ₂Combined and arranged in ascending order to obtain

To a₃Within the sequence, adjacent components not exceeding ± 5Hz are grouped to ensure that the actual frequency is only considered for filter design, thereby extracting single-frequency components. Processed frequency sequence Λ ═ f_i}_i＝1,2,…N，N≤M₁+M₂Frequency components spaced more than 10Hz apart are accurately segmented for the final frequency domain sequence present in the signal. Wherein M is₁For the frequency domain sequence of fundamental and harmonic waves of the original disturbance signal

The maximum frequency order of (a) is also Λ₁Maximum ordinal number of (1), M₂And M₁Similarly, as a new harmonic sequence

The maximum frequency order of (1), is also Λ₂The final frequency sequence after processing N is the maximum order number in (a), N is { f ═ f_i}_i＝1,2,…NThe value of N should not be greater than M₁+M₂。

In step 2, the processed signal is subjected to EWT decomposition, where each eigenmode function obtained by the decomposition is: including different perturbation characteristics within such signals, the number of decompositions is determined by the complexity of the signal.

The eigenmode function BLIMF containing the characteristic information is calculated as follows:

(1): extracting a dominant frequency f ═ f of a signal using FFT_i}_i＝1,2,…NAnd i is the number of the main frequencies extracted by the FFT.

(2): determining a boundary value Ω ═ { Ω }_i}_i＝1,2,…NWhich adaptively partitions the continuous fourier spectrum into several parts. Initial boundary value omega₀Set to 0, and the rest

For two adjacent minima f in the Fourier spectrum of the signal_i、f_i+1The midpoint therebetween.

(3): the construction of the empirical wavelet adopts a construction method of Meyer wavelet, and the sampling frequency is 3.2 kHz; the method can realize self-adaptive decomposition without additionally setting parameters such as decomposition times and the like. Empirical wavelet function phi_iAnd empirical scale function psi_i(ω) can be defined as follows:

wherein β (gamma, omega)_i)＝β(1/2γΩ_i(|ω|-(1-γ)Ω_i) ω is the frequency of the input signal,

for two adjacent minima f in the Fourier spectrum of the signal_i、f_i+1Where γ is an important parameter for ensuring a minimum overlap area between two successive state intervals, the value of which is determined by the calculated boundary values.

(4): the filtered detail factors can be obtained by inverse fast fourier transform:

W_x(1,n)＝IFFT(X(ω)φ₁(ω)) (4)；

W_x(i,n)＝IFFT(X(ω)ψ_i(ω)) (5)；

in step 3, the distribution entropy vector corresponding to the EWT component containing the feature information corresponds to the feature vector of the original signal, and the feature vector is calculated as follows:

for a signal x of length n ═ x₁,x₂,…,x_NFdispen of } can be derived as follows:

1): first, x ═ x₁,x₂,…,x_NMaps to class c with integer index and is labeled from 1 to c. Most of x is present in the process_iMay be assigned to a small number of categories of problems, particularly when the maximum or minimum value is significantly greater or less than the mean or median of the signal. A Normal Cumulative Distribution Function (NCDF) can solve this problem. The whole mapping process is as follows: the NCDF maps x to y, all values between 0 and 1, as shown in equation (7).

Wherein, sigma and mu are respectively the standard deviation and the mean e of the time series x are natural logarithms, and x_jIs the jth element in the original signal. Then each y_iAre linearly assigned to integers from 1 to c. For each value of the mapping signal

Get the whole, wherein

Represents the element of the jth sorted time series, and increments or decrements the number to the next integer using the round operator (round).

2): according to

Define a time series

Each time series being related to the embedding dimension m-1 and the time delay d

Mapping to a scattered pattern with oscillations

Wherein,

and pi is a generic representation used to distinguish the final quantities and has no practical mathematical significance. Can be assigned to each time series

Has a number of scattering modes of all possible oscillations of (2c-1)^(m-1)It is all kinds of pi.

3): from each potential scattering pattern

The following relative probabilities can be obtained:

wherein, # is the base number, m is the embedding dimension, c is the integer class category to be mapped, d is the time delay,

for the kind of the distribution pattern, i is all the distribution types

The number of the types of (a) to (b),

for the mapped time series, N is the maximum number of the original time series.

4): finally, the Fdispen value calculation is as shown in equation (9) according to Shannon's entropy definition:

wherein x is the time sequence, m is the embedding dimension, c is the integer class to be mapped, d is the time delay,

for the kind of scatter pattern, p can be calculated in equation (8) above, and ln is the base e logarithm.

In the step 4, the classification standard is a feature vector after PCA dimension reduction, and the feature vector is input into an SVM classifier to complete automatic classification. The ideal result can be obtained under the condition of not specially configuring the SVM, the OAO method is selected as the SVM classification method, and the polynomial kernel function is selected as the kernel function.

In the step 5, the fault state of the power distribution network containing the distributed energy sources comprises voltage sag, voltage rise, voltage interruption, harmonic waves, flicker, oscillation, depression or spikes;

the PQ disturbance signal comprises an amplitude disturbance or frequency disturbance composite signal, and the amplitude disturbance comprises voltage temporary rise, voltage temporary drop and voltage interruption; while harmonics, flicker, oscillations, pits or spikes are among the frequency disturbances.

The invention discloses a power quality disturbance detection method for a power distribution network based on EWT and MFDE, which has the following technical effects:

1) and constructing a power distribution network containing photovoltaic, wind energy and electric vehicles as a test system for testing the effectiveness of the power quality disturbance detection classification algorithm.

2) And an optical voltage transformer is adopted to extract the 11 types of disturbance signals, an EWT is used to detect the analog signals, and effective information such as disturbance amplitude, frequency and disturbance duration is accurately detected.

The optical voltage type mutual inductor can overcome the defects of the traditional mutual inductor, the principle of the optical mutual inductor lies in the Pockels effect, and when polarized light irradiates the surface of a BGO crystal, the polarized light is split into two beams of light beams with mutually vertical vibration directions. The phase difference is proportional to the applied voltage, and the structure and material are different, and the phase difference is also different. Therefore, the voltage of the distribution network can be extracted.

The traditional electromagnetic voltage transformer and the capacitive voltage transformer have the defects of complex insulation structure, magnetic saturation, ferromagnetic resonance and the like, and the optical voltage transformer based on the optical fiber sensing technology can effectively overcome the inherent defects of the traditional voltage transformer. The high-voltage power supply has the advantages of good insulating property, strong anti-electromagnetic interference capability, small volume, light weight, wide frequency response, large dynamic range, high safety, digital output and the like. It is very suitable for extracting voltage signals of a power distribution network.

3) And taking the eigenmode function containing the characteristic information as the input of the MFDE algorithm to obtain multidimensional characteristic vectors of different disturbances, wherein each characteristic vector represents disturbance information.

4) And after the feature vectors are subjected to PCA (principal component analysis) dimensionality reduction, inputting the feature vectors into the SVM (support vector machine) to finish disturbance classification.

5) Accurate detection and classification of the power quality problem are the premise of reliable operation of the power distribution network. The method can accurately detect and classify each disturbance in the composite power quality disturbances, and has accurate classification and certain noise resistance.

Drawings

FIG. 1 is a schematic diagram of an optical voltage type transformer;

wherein: the device comprises a light source 1, a first collimating lens 2, a polarizer 3, a wave plate 4-1/4, an optical crystal 5, a BGO 6, an analyzer 7, a second collimating lens and a signal processing module 8.

Fig. 2 is a schematic structural diagram of a high permeability active power distribution network.

FIG. 3(1) is a diagram of the EWT mode function;

FIG. 3(2) is a time-frequency diagram of the mode function.

FIG. 4(1) is an MFDE characteristic diagram of a disturbance signal C1;

FIG. 4(2) is an MFDE characteristic diagram of the disturbance signal C2;

FIG. 4(3) is an MFDE characteristic diagram of the disturbance signal C3;

FIG. 4(4) is an MFDE characteristic diagram of the disturbance signal C4;

FIG. 4(5) is an MFDE characteristic diagram of the disturbance signal C5;

FIG. 4(6) is an MFDE characteristic diagram of the disturbance signal C6;

FIG. 4(7) is an MFDE characteristic diagram of the disturbance signal C7;

FIG. 4(8) is an MFDE characteristic diagram of the disturbance signal C8;

FIG. 4(9) is an MFDE characteristic diagram of the disturbance signal C9;

FIG. 4(10) is an MFDE characteristic diagram of the disturbance signal C10;

fig. 4(11) is an MFDE characteristic diagram of the disturbance signal C11.

Fig. 5 is a confusion matrix diagram.

Detailed Description

And constructing a high-permeability active power distribution network system containing photovoltaic power, wind power and electric vehicles as a test system for the effectiveness of the detection method. The built test system is an IEEE-13 bus power distribution network and is connected to a power grid with the rated power of 5MVA and the operating voltages of 4.16kV and 0.48 kV; the sampling frequency is 3.2kHz, the sampling time length is 0.2s, and the robustness of the algorithm is tested by adding noise with the size of 20dB to an output signal.

In the active power distribution network system, a wind turbine set adopts a double-fed motor and is directly connected with the grid through a transformer Tr3, and the total capacity is 1.5 MW; photovoltaic power generation is connected to the grid through a three-phase voltage type PWM converter, and the total capacity is 1 MW; the electric automobile adopts a three-phase bridge type rectifying charger to be connected with a power grid, and the total load of the electric automobile is 0.5 MW. The electric automobile comprises three types of quick charging, mechanical charging and conventional charging, a quick charging machine for quick charging has high power, and the problem of electric energy quality is more serious.

The permeability of the active power distribution network is based on the following formula (10):

in the formula, P_i,DG-nonThe installed capacity of the ith distributed power supply in the system; n is the number of distributed power supplies in the system; p_L,sumThe total power of the load in the system. Total installed capacity P of distributed power supply_i,DG-non2.5MW, total load power P in the system_L,sumThe permeability of the system is 64.66% and exceeds 60% calculated by formula (10), which is 3.866MW, and accords with the description of high permeability.

The schematic structure of the test system is shown in fig. 1. The electric energy quality disturbance caused by DGs comprises 4 categories including wind turbines, photovoltaics, electric vehicles and various DGs mixed operations, which are respectively as follows:

W1-W3 are fan event groups: w1 is the grid connection of the fan and the public power grid, W2 is the island operation after the fan is interrupted with the public power grid and the W3 fan is disconnected with the public power grid;

S1-S3 are photovoltaic event groups: s1 is that the photovoltaic cell panel is connected with a public power grid in a grid mode, S2 is that the photovoltaic cell panel is disconnected with the public power grid, and an S3 fan runs in an isolated island mode after being disconnected with the public power grid;

E1-E2 are electric vehicle event groups: e1 is the large-scale access of the electric automobile to the power grid, E2 is the large-scale off-grid of the electric automobile;

H1-H3 are fan and photovoltaic hybrid operations: h1 is that the fan system and the photovoltaic system are simultaneously connected with a public power grid in a grid mode, H2 is that the fan system and the photovoltaic system are simultaneously disconnected with the public power grid in an interruption mode, and H3 is that the fan system and the photovoltaic system are operated in an isolated island mode after being simultaneously disconnected with the public power grid.

The power quality disturbance detection method of the power distribution network based on the EWT and the MFDE comprises the following steps:

step 1: and an optical voltage type mutual inductor is used for extracting the disturbance signal of the high-permeability power distribution network, and the extracted analog signal is preprocessed, so that the accuracy of a result is ensured.

Step 2: and E, decomposing the EWT by adopting empirical wavelet, decomposing the PQ disturbance signal of the active power distribution network system, filtering the noise of the PQ disturbance signal, and decomposing to obtain an EWT component containing characteristic information. The EWT decomposition mode function is self-adaptive decomposition and does not need manual setting.

And step 3: taking BLIMF components containing characteristic information as input of a multi-scale oscillation spreading entropy MFDE algorithm, performing spreading entropy value calculation on eigenmode functions (BLIMF) obtained by mode decomposition by using the multi-scale oscillation spreading entropy MFDE algorithm, and calculating to obtain a multi-dimensional entropy value vector of a PQ disturbance signal under each eigenmode function (BLIMF);

for a signal x of length n ═ x₁,x₂,…,x_NFdispen of } can be derived as follows:

1): first, x ═ x₁,x₂,…,x_NMaps to class c with integer index and is labeled from 1 to c. Most of x is present in the process_iMay be assigned to a small number of categories of problems, particularly when the maximum or minimum value is significantly greater or less than the mean or median of the signal. A Normal Cumulative Distribution Function (NCDF) can solve this problem. The whole mapping process is as follows: the NCDF maps x to y, all values between 0 and 1, as shown in equation (11).

Where σ and μ are the standard deviation and mean e of the time series x, respectively, as natural logarithms, x_jIs the jth element in the original signal. Then each y_iAre linearly assigned to integers from 1 to c. For each value of the mapping signal

Get the whole, wherein

Represents the element of the jth sorted time series, and increments or decrements the number to the next integer using the round operator (round).

2): according to

Define a time series

Each time series being related to the embedding dimension m-1 and the time delay d

Mapping to a scattered pattern with oscillations

Wherein

And pi is a generic representation used to distinguish the final quantities and has no practical mathematical significance. Can be assigned to each time series

Has a number of scattering modes of all possible oscillations of (2c-1)^(m-1)It is all kinds of pi.

3): from each potential scattering pattern

The following relative probabilities can be obtained:

where # is the cardinality, m is the embedding dimension, c is the integer class to be mapped, d is the time delay,

for the kind of the distribution pattern, i is all the distribution types

The number of the types of (a) to (b),

for the mapped time series, N is the maximum number of the original time series.

4): finally, according to Shannon's entropy definition, Fdispen values are calculated as follows:

wherein x is the time sequence, m is the embedding dimension, c is the integer class to be mapped, d is the time delay,

for the kind of the scattering pattern, p is the above formula (12), and ln is the logarithm based on e.

And 4, step 4: carrying out PCA dimensionality reduction on the entropy vector obtained in the step 3, and using the entropy vector as an input quantity of an SVM algorithm;

and 5: and identifying PQ disturbance signals of the active power distribution network system containing the distributed energy. The fault state of the power distribution network containing the distributed energy sources refers to voltage sag, harmonic waves, flicker, oscillation, sunken spines and the like, and the composite disturbance.

Example (b):

the invention discloses an implementation mode of detecting disturbance signals by empirical wavelet decomposition (EWT) and MFDE, which comprises the following steps:

the method comprises the steps of extracting signals in an active power distribution network by using an optical voltage transformer, preprocessing analog signals, performing modal decomposition on 11 types of power quality disturbance signals C1-C11 of the power distribution network containing the distributed power supply by adopting empirical wavelet transformation, extracting eigenmode functions (BLIMF) containing characteristic information and containing the characteristic information, and using the eigenmode functions as input signals of multi-scale oscillation spreading entropy MFDE, thereby realizing detection and classification of power quality.

The signal extraction of the optical voltage type mutual inductor is based on the Pockels effect and mainly comprises a sensor unit on a high-voltage side, a photoelectric unit on a low-voltage side and an electro-optic crystal. The electro-optical crystal has various types, wherein the BGO crystal has no pyroelectric characteristic, no optical rotation, no natural birefringence and good theoretical stability. It is one of the most widely used electro-optical crystals in the current photoelectric sensor. When the polarized light irradiates the surface of the BGO crystal, the polarized light is split into two beams with vibration directions perpendicular to each other. The phase difference is proportional to the applied voltage, and the structure and material are different, and the phase difference is also different. Fig. 1 is a schematic diagram of an optical voltage transformer. The phase difference can be calculated by the following equation (14):

where λ is the wavelength of the incident light wave, n₀Is the refractive index, gamma, of BGO crystals₄₁Is the linear electro-optic coefficient of the BGO crystal, and U is the measured voltage. U shape₀ω is the voltage amplitude and angular frequency, respectively

Generally, the delta is converted into the change of output light intensity by adopting a polarized light interferometry for detection. The 1/4 wave plate is used to increase the phase difference between the two beams by 90 deg., and the total phase difference is delta + pi/2. The output light intensity can be expressed as:

wherein, I₀As to the intensity of the incident light,

called half-wave voltage, can measure the measured voltage by the relationship between the output light intensity and the voltage through photoelectric conversion and signal processing. The sensor unit and the photoelectric unit are connected through a polarization-maintaining optical fiber. The sensing unit comprises a high-voltage sensing device, a shielding and insulating device and a sensing head. The high-voltage induction device induces a high-voltage potential from a high-voltage wire to form a stable electric field in the shielding and insulating device, and the sensor head is arranged in the electric field. The photoelectric unit includes an optical path portion and a circuit portion provided in the secondary-side device. The light emitted by the light source is divided into two orthogonal linearly polarized light by the light path part and is transmitted to the sensing head by the polarization maintaining optical fiber. Under the action of the electric field, the sensing head generates a phase difference between the two polarized lights. Two beams of polarized light returned from the sensing head are transmitted back to the light path part through the polarization maintaining optical fiber to interfere, and the light intensity signal is detected and processed by the circuit part to form digital signal output. The shield is gas insulated with SF 6. The optical path of the photovoltage transformer mainly comprises a light source, a circulator, a y waveguide modulator, a polarization maintaining coupler and a photoelectric detector. The basic principle of the optical path part is as follows: light emitted by the light source enters the Y waveguide modulator through the circulator, is divided into two orthogonal linearly polarized light in the Y waveguide modulator, is coupled through the polarization maintaining coupler, then enters the polarization maintaining optical fiber, is transmitted to the sensing head along the fast and slow axes of the polarization maintaining optical fiber, and after the polarization maintaining optical fiber is subjected to polarization coupling, carries two beams of light of voltage information to be measured again along the polarization maintaining optical fiber. The coupler, the fiber returns to and interferes with the Y waveguide modulator. The interference light intensity signal enters the photoelectric detector through the circulator, is converted into an electric signal, and then enters the circuit part for signal processing. The circuit mainly comprises an A/D converter, a digital signal processing unit, a D/A converter and a corresponding driving circuit. The signal processing adopts a digital closed loop signal detection technology.

The signal preprocessing of the invention is particularly characterized in that the signal is subjected to proper Fourier spectrum division, the division result on a frequency domain directly influences the final decomposition quantity of the EWT, and different frequency spectrum parts correspond to modes with different specific supporting frequencies as centers. When a disturbing signal which may generate inter-harmonics near the fundamental wave occurs, the fundamental component may have a spectrum leakage phenomenon, and if the disturbing signal is not properly processed, the fourier spectrum division may be excessively decomposed, and finally the EWT component may be deviated. Generally, the distance between two consecutive harmonics or between two consecutive inter-harmonics is longer than the distance between one harmonic and one inter-harmonic, and therefore, in order to more accurately decompose the voltage and current signals, the initial frequency spectrum of the signals needs to be preprocessed.

Firstly, a Fourier transform is used for constructing a frequency domain sequence of fundamental waves and harmonic waves of an original disturbance signal

Wherein the minimum component amplitude is not less than 2% of the fundamental, the minimum frequency is not less than 45Hz, and components outside the range are discarded. The corrected set of frequencies is substituted as the center frequency into equation (16) to calculate ψ_iWherein Ω is_i＝f_i-5Hz，Ω_i+1＝f_i+5Hz, γ is 0.01. Bandwidth BW of the band pass filter_i＝10+2γf_iHz. Harmonic spectrum X_H(ω) can be calculated from the following formula (16):

in the formula, X_HIs a harmonic spectrum, omega is the input signal frequency, i is the dominant frequency order, Λ₁Is a frequency domain sequence, phi_iFor empirical wavelet functions and psi_iThe (omega) is a function of the scale of experiments, and the specific formula is given in the above formulas (2) and (3).

Subtracting the harmonic spectrum X from the original spectrum X (ω)_H(ω) obtaining a spectrum X containing only the inter-harmonic residual_R(omega), constructing an inter-harmonic sequence with the minimum component amplitude not lower than 2% of the fundamental wave and the minimum frequency not lower than 5Hz

Will be Λ₁、Λ₂Combined and arranged in ascending order to obtain

However, due to the spectrum leakage, some fundamental wave components and integer harmonic components of extremely unstable signals still have component residues, so that the harmonic components of the extremely unstable signals still have component residuesΛ₃Within the sequence, adjacent components not exceeding ± 5Hz are grouped to ensure that the actual frequency is only considered for filter design, thereby extracting single-frequency components. Processed frequency sequence Λ ═ f_i}_i＝1,2,…N，(N≤M₁+M₂) Frequency components spaced more than 10Hz apart are accurately segmented for the final frequency domain sequence present in the signal.

Empirical wavelet decomposition EWT is used to decompose a non-recursive real-valued signal f (t) into k band-limited implicit EWT components with some sparseness, as follows:

(1): extracting a dominant frequency f ═ f of a signal using FFT_i}_i＝1,2,…N. Wherein i is the number of main frequencies extracted by FFT.

(2): determining a boundary value Ω ═ { Ω }_i}_i＝1,2,…NWhich adaptively partitions the continuous fourier spectrum into several parts. Initial boundary value omega₀Set to 0, and the rest

For two adjacent minima f in the Fourier spectrum of the signal_i、f_i+1The midpoint therebetween.

(3): the construction of the empirical wavelet adopts a construction method of Meyer wavelet, and the sampling frequency is 3.2 kHz; the method can realize self-adaptive decomposition without additionally setting parameters such as decomposition times and the like. Empirical wavelet function phi_iAnd empirical scale function psi_i(ω) can be defined as follows:

wherein β (gamma, omega)_i)＝β(1/2γΩ_i(|ω|-(1-γ)Ω_i) ω is the frequency of the input signal,

for two adjacent minima f in the Fourier spectrum of the signal_i、f_i+1Where γ is an important parameter for ensuring a minimum overlap area between two successive state intervals, the value of which is determined by the calculated boundary values.

(4): the filtered detail factors can be obtained by an inverse fast fourier transform.

W_x(1,n)＝IFFT(X(ω)φ₁(ω)) (19)；

W_x(i,n)＝IFFT(X(ω)ψ_i(ω)) (20)；

Where ω is the input signal frequency, i is the dominant frequency, φ_iFor empirical wavelet functions and psi_i(ω) is a function of the empirical scale, the specific formula given in (17) (18) above,

the sampling frequency of the built experimental system is 3.2kHz, and each mode except the first mode after EWT decomposition represents different disturbance characteristics. The photovoltaic island operation (S3) is that the photovoltaic power generation system operates independently after the utility grid support is lost, and due to the randomness of the illumination intensity and the illumination time, the power disturbance features mainly a flicker component and a harmonic component, as shown in fig. 3(1), it can be seen that within 0.6 second to 1 second, the power quality disturbance is mainly a harmonic wave and is reflected in components 4, 5, and 6, and after 1 second, high-frequency flicker occurs and is reflected in components 7 and 8, the frequency change of the EWT frequency spectrum can be observed, and the duration of each disturbance can also be reflected, so that the flicker frequency occurring after 1 second is about 900Hz, and the amplitude fluctuation is small, as shown in fig. 3 (2). EWT accurately identifies both the duration and magnitude components of both disturbances.

FIG. 4(1) is an MFDE characteristic diagram of a disturbance signal C1;

FIG. 4(2) is an MFDE characteristic diagram of the disturbance signal C2;

FIG. 4(3) is an MFDE characteristic diagram of the disturbance signal C3;

FIG. 4(4) is an MFDE characteristic diagram of the disturbance signal C4;

FIG. 4(5) is an MFDE characteristic diagram of the disturbance signal C5;

FIG. 4(6) is an MFDE characteristic diagram of the disturbance signal C6;

FIG. 4(7) is an MFDE characteristic diagram of the disturbance signal C7;

FIG. 4(8) is an MFDE characteristic diagram of the disturbance signal C8;

FIG. 4(9) is an MFDE characteristic diagram of the disturbance signal C9;

FIG. 4(10) is an MFDE characteristic diagram of the disturbance signal C10;

fig. 4(11) is an MFDE characteristic diagram of the disturbance signal C11.

Fig. 4(1) to fig. 4(11) are all illustrations of the oscillation dispersion entropy in the present case, wherein the components in the entropy method correspond to the components obtained by EWT decomposition, the change of color represents the change of entropy, no matter what entropy method is adopted, the entropy sequence can not express any actual physical meaning, and even if the entropy values of the same perturbation signal have the same change rule, the perturbation type can not be asserted. But the method is very suitable for being used as a further classification algorithm to input the feature vector, namely the entropy value sequence is used for replacing an original signal sequence to be input into the classification algorithm, the classification effect after the entropy value method processing is far superior to the classification effect of direct input, and the final confusion matrix is arranged in a figure 5, so that the effectiveness of the method can be proved.

The method for realizing the classification of the disturbance signals through the MFDE algorithm comprises the following steps:

1. and decomposing the EWT to obtain a modal function containing main characteristics, and using the modal function as an input signal of the MFDE algorithm to extract characteristic vectors of different PQ disturbances.

The dispersion entropy is high in calculation speed and is slightly influenced by the sudden change signals, and the amplitude relation among the signals is considered. However, the original dispersion entropy cannot analyze local or global trends of the signal. When the fluctuation of the data is meaningful or the local trend of the time series is meaningless, the oscillation mode thereof is not different from the original dispersion entropy. Therefore, the invention introduces the diffusion entropy MFDE based on multi-scale oscillation, solves the problems and is more suitable for extracting the signals of the power distribution network.

The signal after EWT modal decomposition is complexity aligned to highlight the signal characteristics, and MFDE can be calculated as follows:

1) first, x ═ x₁,x₂,…,x_NMapping to class c with integer index, and doing from 1 to cAnd (6) labeling. Most of x is present in the process_iMay be assigned to a small number of categories of problems, particularly when the maximum or minimum value is significantly greater or less than the mean or median of the signal. A Normal Cumulative Distribution Function (NCDF) can solve this problem. The whole mapping process is as follows: first, the NCDF maps x to y, all values being between 0 and 1, as shown in equation (21)

Where σ and μ are the standard deviation and mean e of the time series x, respectively, as natural logarithms, x_jIs the jth element in the original signal. Then each y_iAre linearly assigned to integers from 1 to c. For each value of the mapping signal

Get the whole, wherein

Represents the element of the jth sorted time series, and increments or decrements the number to the next integer using the round operator (round).

2) According to

Define a time series

Each time series being related to the embedding dimension m-1 and the time delay d

Mapping to a scattered pattern with oscillations

Wherein

And pi is a generic representation used to distinguish the final quantities and has no practical mathematical significance. Can be assigned to each time series

Has a number of scattering modes of all possible oscillations of (2c-1)^(m-1)It is all kinds of pi.

3) From each potential scattering pattern

The following relative probabilities can be obtained:

where # is the cardinality, m is the embedding dimension, c is the integer class to be mapped, d is the time delay,

for the kind of the distribution pattern, i is all the distribution types

The number of the types of (a) to (b),

for the mapped time series, N is the maximum number of the original time series.

4) Finally, according to Shannon's entropy definition, Fdispen values are calculated as follows:

wherein x is the time sequence, m is the embedding dimension, c is the integer class to be mapped, d is the time delay,

p is the above formula (22) for the kind of the scattering mode) Ln is the logarithm to the base of e.

2. And inputting the feature vector subjected to PCA dimensionality reduction into the SVM to complete disturbance classification.

The SVM requires a large amount of data to train, which is difficult to obtain from a simulation system, and the present invention selects analog data given in the IEEE standard to train. The training classification object of the SVM of the invention on disturbance is BLIMF with characteristic quantity obtained by previous EWT decomposition, and the disturbance characteristics of the modal functions are consistent with the characteristics of the training signals, so that the training can be carried out by adopting a mathematical model. Each type of disturbance generates 400 signals, a specific model and a signal specific parameter range (disturbance time, frequency, amplitude, phase angle). As shown in the power quality mathematical model in section 3 below.

The fundamental frequency of the signal is 60Hz and varies within a range of + -0.5 Hz. All signals are subjected to 200ms as value duration according to IEC standard 61000-4-7, and noise with a signal-to-noise ratio of 20dB to 50dB is added randomly in order to improve the accuracy of identifying noise interference signals. The SVM classification method adopts an OAO method, and the kernel function adopts a polynomial kernel function. The confusion matrix after SVM training is shown in FIG. 5, which shows that all kinds of signals can be effectively distinguished.

3. The mathematical model of the power quality disturbance signal is as follows:

①, normal signal, formula x (t) sin (2 π f)_ft-φ_f)，49.5≤f_f≤50.5,0≤φ_fLess than or equal to 180, wherein t is time, f_fIs the frequency phi_fIs the phase angle.

②, voltage interruption, x (t) ([ 1- α) (u (t-t))₁)-u(t-t₂))]sin(2πf_ft-φ_f) Wherein 0.9- α -1, T-T₂-t₁9T or less, wherein T is time, α is voltage interruption amplitude, u is original signal amplitude, T is original signal amplitude₁For interruption start time, t₂For the end time of interruption, f_fIs the frequency phi_fIs the phase angle, T is the period.

③, voltage sag x (t) ([ 1- α) (u (t-t))₁)-u(t-t₂))]sin(2πf_ft-φ_f) Wherein 0.1- α -0.9, T-T₂-t₁Less than or equal to 9T, wherein T is time, α is voltage sag amplitude, u is original signal amplitude, T₁To temporarily decrease the start time, t₂For temporarily decreasing the end time, f_fIs the frequency phi_fIs the phase angle, T is the period.

④, voltage ramp x (t) ([ 1+ α) (u (t-t))₁)-u(t-t₂))]sin(2πf_ft-φ_f) Wherein 0.1- α -0.8, T-T₂-t₁Less than or equal to 9T, wherein T is time, α is voltage temporary rising amplitude, u is original signal amplitude, T is original signal amplitude₁To suspend the start time, t₂For pausing the end time, f_fIs the frequency phi_fIs the phase angle, T is the period.

⑤ harmonic, x (t) sin (2 pi f)_ft-φ_f)+∑a_isin(2πf_it-φ_i) Wherein, a is more than or equal to 0.03_iLess than or equal to 0.25, wherein t is time, f_fIs the frequency phi_fIs a phase angle of_iIs the amplitude of the harmonic component, f_iIs the harmonic component frequency, phi_iIs the harmonic component phase angle.

⑥ flicker, x (t) ([ 1+ α sin (2 pi β t))]sin(2πf_ft-φ_f) Wherein 0.1- α -0.2, 5- β -25, α is flicker amplitude, β is flicker frequency, t is time, f is_fIs the frequency phi_fIs the phase angle.

⑦ oscillation, x (t) sin (2 pi f)_ft-φ_f)+α_texp(-(t-t₁)/τ)(u(t-t₁)-u(t-t₂))sin(2πf_tt)，0.1≤α_t≤0.8,0.5T≤t₂-t₁≤3T，300≤f_tNot more than 3500, not less than 8ms and not more than 40ms of tau, α_tIs the amplitude of oscillation, f_tExp is the natural logarithm based on e, τ is the oscillation interval, u is the original signal amplitude, t is the time, t is₁As oscillation start time, t₂As oscillation end time, f_fIs the frequency phi_fIs the phase angle.

⑧, recess:

wherein,

n is the number of notches and sign is a sign function, which is a function of taking a certain number of signs (positive or negative), u is the original signal amplitude, t is time, t is the sign function₁As oscillation start time, t₂As oscillation end time, f_fIs the frequency phi_fIs the phase angle.

⑨, spine:

wherein,

n is the number of sharp pricks, sign is a sign function, and the function is to take a certain number of signs (positive or negative), u is the original signal amplitude, t is time, t is₁As oscillation start time, t₂As oscillation end time, f_fIs the frequency phi_fIs the phase angle.

TABLE 1 load configuration Table

TABLE 2 Transformer configuration Table

Claims (8)

1. The power quality disturbance detection method of the power distribution network based on EWT and MFDE is characterized by comprising the following steps:

step 1: extracting a PQ disturbance signal in an active power distribution network system by adopting an optical voltage transformer, extracting an analog signal after the PQ disturbance signal is subjected to signal conversion, and preprocessing the analog signal;

step 2: performing modal decomposition on the preprocessed PQ disturbance signal by adopting empirical wavelet decomposition (EWT) to obtain an intrinsic mode function BLIMF containing characteristic information;

and step 3: taking an intrinsic mode function BLIMF containing characteristic information as the input of a multi-scale oscillation spreading entropy MFDE algorithm, performing spreading entropy value calculation on the intrinsic mode function BLIMF obtained by mode decomposition by using the multi-scale oscillation spreading entropy MFDE algorithm, and calculating to obtain a multi-dimensional entropy value vector of the PQ disturbance signal under each intrinsic mode function BLIMF;

and 4, step 4: carrying out PCA dimensionality reduction on the entropy vector obtained in the step 3, and using the entropy vector as an input quantity of an SVM algorithm;

and 5: and identifying PQ disturbance signals of the active power distribution network system containing the distributed energy.

2. The EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in the step 1, a high-permeability active power distribution network system comprising photovoltaic, wind energy and an electric automobile is set up and used as a test system for the effectiveness of the power quality PQ disturbance detection method;

the built test system is an IEEE-13 bus power distribution network and is connected to a power grid with the rated power of 5MVA and the operating voltages of 4.16kV and 0.48 kV; the sampling frequency is 3.2kHz, the sampling time length is 0.2s, and the robustness of the algorithm is tested by adding 20dB noise to an output signal;

the permeability of the active distribution network is based on the following formula (1):

in the formula, P_i,DG-nonThe installed capacity of the ith distributed power supply in the system; n is the number of distributed power supplies in the system; p_L,sumThe total power of the load in the system; the total installed capacity of the distributed power supply is P_i,DG-nonTotal power of load in system is P_L,sum。

3. The EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in step 1, the PQ disturbance in the active power distribution network system includes the following types:

W1-W3 are fan event groups: w1 is that the fan is connected with a public power grid in a grid mode, W2 is that the fan is disconnected with the public power grid, and the W3 fan runs in an isolated island mode after being disconnected with the public power grid;

S1-S3 are photovoltaic event groups: s1 shows that the photovoltaic cell panel is connected with a public power grid in a grid mode, S2 shows that the photovoltaic cell panel is disconnected with the public power grid, and an S3 fan runs in an isolated island mode after being disconnected with the public power grid;

E1-E2 are electric vehicle event groups: e1 is the large-scale access of the electric automobile to the power grid, E2 is the large-scale off-grid of the electric automobile;

H1-H3 are fan and photovoltaic hybrid operations: h1 is that the fan system and the photovoltaic system are simultaneously connected with a public power grid in a grid mode, H2 is that the fan system and the photovoltaic system are simultaneously disconnected with the public power grid in an interruption mode, and H3 is that the fan system and the photovoltaic system are operated in an isolated island mode after being simultaneously disconnected with the public power grid.

4. The EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in the step 1, the analog signal is preprocessed:

firstly, constructing a frequency domain sequence of fundamental waves and harmonic waves of an original disturbance signal

Wherein the minimum component amplitude is not less than 2% of the fundamental wave, the minimum frequency is not less than 45Hz, and the set of frequencies is used as the center frequency to calculate psi_iBandwidth BW of the band pass filter_i＝10+2γf_iHz; harmonic spectrum X_H(ω) can be calculated from the following formula (6):

in the formula, X_HIs a harmonic spectrum, omega is the input signal frequency, i is the dominant frequency order, Λ₁Is a frequency domain sequence, phi_iIs an empirical wavelet function, psi_i(ω) is a function of the scale of the experiment;

subtracting the harmonic spectrum X from the original spectrum X (ω)_H(ω) obtaining a spectrum X containing only the inter-harmonic residual_R(omega), constructing an inter-harmonic sequence with the minimum component amplitude not lower than 2% of the fundamental wave and the minimum frequency not lower than 5Hz

Will be Λ₁、Λ₂Combined and arranged in ascending order to obtain

To a₃In the sequence, adjacent components not exceeding ± 5Hz are grouped to ensure that the actual frequency is only considered for filter design, thereby extracting single-frequency components; processed frequency sequence Λ ═ f_i}_i＝1,2,…N，，N≤M₁+M₂Frequency components spaced more than 10Hz apart are accurately segmented for the final frequency domain sequence present in the signal.

5. The EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in step 2, the eigenmode function BLIMF including the characteristic information is calculated as follows:

(1): extracting a dominant frequency f ═ f of a signal using FFT_i}_i＝1,2,…NWherein i is the number of the main frequencies extracted by the FFT;

(2): determining a boundary value Ω ═ { Ω }_i}_i＝1,2,…NAdaptively dividing a continuous Fourier spectrum into several parts; initial boundary value omega₀Set to 0, and the rest

For two adjacent minima f in the Fourier spectrum of the signal_i、f_i+1The midpoint therebetween;

(3): the empirical wavelet structure adopts Meyer wavelet structureThe manufacturing method, the sampling frequency is 3.2 kHz; empirical wavelet function phi_iAnd empirical scale function psi_i(ω) can be defined as follows:

wherein β (gamma, omega)_i)＝β(1/2γΩ_i(|ω|-(1-γ)Ω_i) ω is the frequency of the input signal,

for two adjacent minima f in the Fourier spectrum of the signal_i、f_i+1Where γ is an important parameter that ensures a minimum overlap area between two successive state intervals, the value of which is determined by the calculated boundary values;

(4): the filtered detail factors can be obtained by inverse fast fourier transform:

W_x(1,n)＝IFFT(X(ω)φ₁(ω)) (4)；

W_x(i,n)＝IFFT(X(ω)ψ_i(ω)) (5)。

6. the EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in step 3, the distribution entropy vector corresponding to the EWT component containing the feature information corresponds to the feature vector of the original signal, and the feature vector is calculated as follows:

for a signal x of length n ═ x₁,x₂,…,x_NFdispen of } can be derived as follows:

1): first, x ═ x₁,x₂,…,x_NMapping to c type with integer index, and labeling from 1 to c; most of x is present in the process_iProblems that may be assigned to a small number of categories, particularly when largest or largestSmall values are significantly greater or less than the mean or median of the signal; a Normal Cumulative Distribution Function (NCDF) can solve this problem; the whole mapping process is as follows: firstly mapping x to y in the NCDF, wherein the values are all between 0 and 1, and the formula (7) is shown as the following formula;

wherein, sigma and mu are respectively the standard deviation and the mean e of the time series x are natural logarithms, and x_jIs the jth element in the original signal; then each y_iIs linearly assigned to an integer from 1 to c; for each value of the mapping signal

Get the whole, wherein

Representing an element of the jth sorted time series and increasing or decreasing the number to the next integer using the round operator (round);

2): according to

Define a time series

Each time series being related to the embedding dimension m-1 and the time delay d

Mapping to a scattered pattern with oscillations

Wherein

And pi is a representative of a kind for distinguishing the final quantity, and has no practical mathematical significance; can be assigned to each time series

Has a number of scattering modes of all possible oscillations of (2c-1)^(m-1)All kinds of pi;

3): from each potential scattering pattern

The following relative probabilities can be obtained:

wherein, # is the base number, m is the embedding dimension, c is the integer class category to be mapped, d is the time delay,

for the kind of the distribution pattern, i is all the distribution types

The number of the types of (a) to (b),

the time sequence after mapping is N is the maximum number of the original time sequence;

4): finally, the Fdispen value calculation is as shown in equation (9) according to Shannon's entropy definition:

wherein x is the time sequence, m is the embedding dimension, c is the integer class to be mapped, d is the time delay,

for the kind of scatter pattern, p can be calculated in equation (8) above, and ln is the base e logarithm.

7. The EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in the step 4, the SVM classification method adopts an OAO method, and the kernel function adopts a polynomial kernel function.

8. The EWT and MFDE-based power quality disturbance detection method for the power distribution network according to claim 1, wherein: in the step 5, the fault state of the power distribution network containing the distributed energy sources comprises voltage sag, voltage rise, voltage interruption, harmonic waves, flicker, oscillation, depression or spikes;

the PQ disturbance signal comprises an amplitude disturbance or frequency disturbance composite signal, and the amplitude disturbance comprises voltage temporary rise, voltage temporary drop and voltage interruption; while harmonics, flicker, oscillations, pits or spikes are among the frequency disturbances.

Images