CN116609612A - A method and system for multi-harmonic source identification of distribution network - Google Patents

Abstract

The invention provides a method for identifying multi-harmonic sources in a distribution network, which includes acquiring harmonic voltage signals generated by superposition of multiple harmonic sources at PCC nodes, and measuring the harmonic voltage values; separating the harmonic voltage signals from the fast The changing component and slowly changing component, and based on the fast changing component, estimate the harmonic current value injected by the PCC node; import the obtained harmonic current value and harmonic voltage value into the pre-trained mutual information deep learning model, and get The mutual information value between the harmonic current emitted by each harmonic source and the harmonic voltage of the PCC node, and according to the obtained mutual information values, the main harmonic source is identified. The implementation of the invention can solve the problem that the existing methods are difficult to effectively identify and locate multi-harmonic sources in complex distribution networks.

Description

A method and system for multi-harmonic source identification of distribution network

technical field

The invention relates to the technical field of power system detection, in particular to a method and system for identifying multi-harmonic sources of a distribution network.

Background technique

In recent years, the construction of new power systems has continued to advance, and the challenges faced by the power grid have become more complex and diverse. In particular, the "double high" feature has led to the superposition of power quality problems and more complex characteristics, which put forward higher requirements for power quality. With the high-density connection of distributed power generation and power electronic equipment to the grid, the number of harmonic sources has increased sharply, and the operating status has changed, making harmonic pollution increasingly serious. Therefore, the location and identification of harmonic sources plays an important role in clarifying the source of harmonic pollution, and is the prerequisite for the division of harmonic responsibilities and the resolution of economic disputes.

The traditional harmonic source identification mostly starts from the electrical point of view, based on the mechanism model, mainly using the equivalent circuit model, based on harmonic state estimation and harmonic impedance based methods, etc. However, these methods are limited by many factors, and it is difficult to effectively identify and locate multi-harmonic sources in complex distribution networks.

Therefore, there is an urgent need for a new multi-harmonic source identification method for distribution networks, which can solve the problem that existing methods are difficult to effectively identify and locate multi-harmonic sources in complex distribution networks.

Contents of the invention

The technical problem to be solved by the embodiments of the present invention is to provide a method and system for identifying multi-harmonic sources in a distribution network, which can solve the problem that existing methods are difficult to effectively identify and locate multi-harmonic sources in complex distribution networks.

In order to solve the above technical problems, an embodiment of the present invention provides a method for identifying a multi-harmonic source in a distribution network, the method includes the following steps:

Obtain the harmonic voltage signal generated by the superposition of multiple harmonic sources at the PCC node, and measure the harmonic voltage value;

separating fast-changing components and slow-changing components in the harmonic voltage signal, and estimating a harmonic current value injected into the PCC node based on the fast-changing components;

Import the obtained harmonic current value and the harmonic voltage value into the pre-trained mutual information deep learning model to obtain the mutual information value between the harmonic current emitted by each harmonic source and the harmonic voltage of the PCC node, And according to the obtained mutual information values, the main harmonic source is identified.

Wherein, the harmonic voltage signal is separated by a filter to separate the fast changing component and the slow changing component.

Wherein, the harmonic current value injected by the PCC node is realized by the FastICA algorithm; wherein,

The FastICA algorithm includes the steps of: data preprocessing and establishing an objective function for optimization.

Wherein, the data preprocessing includes decentralization processing and whitening processing; wherein,

The decentralization process refers to subtracting the mean value of all sampled signals to obtain a set of original data with a mean value of zero: Among them, X(t _i ) is the sampling signal;

The whitening process is a process of decorrelating the sampled data, so that the observed signal X has unit variance: X=ED ^-1/2 E ^T X; where, It consists of n eigenvalues; the eigenvector corresponding to d _i is c _i , E=[c ₁ ,c ₂ ,...c _n ]; the observed signal X=AS, A is a mixing matrix, and S is a PCC node Injected harmonic current source.

Wherein, the process of establishing an objective function for optimization is to find a direction to maximize the non-Gaussianity of the output w ^T X (y=w ^T X); wherein,

Non-Gaussianity is measured by the approximate value of negative entropy: N _g (Y)={E[g(Y)]-E[g(Y _Gauss )]} ² , that is, J _G (w)=[E{G( w ^T X)}] ² maximum values; wherein, w is an m-dimensional variable, representing a row of the unmixing matrix W;

The objective function is defined as: According to the Kunhn-Tucker condition, it is converted into an unconstrained optimization problem, so that the objective function is transformed into: F(w)=E[G(w ^T X)]+C(||w|| ^2-1 ); , the optimal solution of the objective function is obtained by the Newton iterative method: />

Wherein, the mutual information deep learning model is constructed based on a neural network; wherein,

The neural network includes an input layer, a multi-layer hidden layer and an output layer; wherein, the input layer receives samples of the harmonic current value Y and the harmonic voltage value X as input, and the hidden layer undergoes a series of nonlinear After transformation, the result Z of the output layer is obtained.

Wherein, the neural network is obtained by performing the following steps for training, including:

7.1 Initialize the neural network parameters;

7.2 Draw b minibatch samples from the joint distribution; wherein, the b minibatch samples drawn from the joint distribution are recorded as (x ⁽¹⁾ ,y ⁽¹⁾ ),...,(x ^(b) ,y ^{( b)} ) obey the joint probability distribution

7.3 Draw b samples from the Y marginal distribution; wherein, the b samples drawn from the Y marginal distribution are denoted as obey the joint probability distribution />

7.4 Evaluate the lower bound of mutual information; wherein, the formula for calculating the lower bound of mutual information is:

7.5 Use EMA to correct the deviation correction gradient; wherein, the EMA correction deviation correction gradient is expressed as:

7.6 update the parameters of the neural network; wherein, the parameters of the updated neural network are

7.7 Repeat steps 7.1 to 7.6 until the convergence condition is reached.

Wherein, the mutual information lower bound between X and Y is calculated using the KL divergence calculated using the dual form:

Among them, T is all functions defined on X×Y that make the two expectations finite; And I(X;Y)≥I _θ (X,Y);/> is the amount of neural information, and by referring to the gradient formula as descend to maximize;

Wherein, the mutual information deep learning model can pass the formula To represent, among them, /> Represents the empirical distribution of the distribution P when n independent and identically distributed samples are given.

The embodiment of the present invention also provides a distribution network multi-harmonic source identification system, including:

The harmonic voltage acquisition unit is used to acquire the harmonic voltage signals generated by the superposition of multiple harmonic sources at the PCC node, and measure the harmonic voltage value;

a harmonic current estimating unit, configured to separate fast-changing components and slowly-changing components in the harmonic voltage signal, and estimate the harmonic current value injected into the PCC node based on the fast-changing component;

The main harmonic source identification unit is used to import the obtained harmonic current value and the harmonic voltage value into the pre-trained mutual information deep learning model, and obtain the harmonic current emitted by each harmonic source and the PCC node harmonic The mutual information value between wave voltages, and according to the obtained mutual information values, identify the main harmonic source.

Implementing the embodiment of the present invention has the following beneficial effects:

1. The present invention only needs to obtain the PCC node harmonic voltage signal, use the FastICA algorithm to estimate the harmonic current, and perform mutual information estimation through the mutual information deep learning model to identify the main harmonic source, so that the operation is simple, and the system network parameters The position of the harmonic current source can be identified when it is unknown, which solves the problem that the existing methods are difficult to effectively identify and locate the multi-harmonic source in the complex distribution network;

2. The FastICA algorithm in the present invention estimates the harmonic current, which not only has fast convergence speed, good separation effect, stable iteration, and can separate non-Gaussian independent components, but also improves the accuracy and reliability of harmonic current estimation;

3. The mutual information deep learning model in the present invention makes the calculation simple, can better adapt to the complex nonlinear relationship in the power grid, and has stronger universality and reliability.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, obtaining other drawings based on these drawings still belongs to the scope of the present invention without any creative effort.

Fig. 1 is a flow chart of a method for identifying a multi-harmonic source in a distribution network provided by an embodiment of the present invention;

Fig. 2 is the flow chart of neural network training in a kind of distribution network multi-harmonic source identification method provided by the embodiment of the present invention;

Fig. 3 is a schematic structural diagram of a distribution network multi-harmonic source identification system provided by an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings.

The inventor found that when the nonlinear load is connected to the distribution network, the harmonic current emitted by different harmonic sources and the harmonic voltage at the PCC node are not completely random and independent, but there is a certain relationship, which can be used Mutual Information (MutualInformation, MI) characterization. Therefore, in order to effectively improve the accuracy of mutual information estimation, the inventor proposes a mutual information deep learning model for multi-harmonic source identification in distribution networks, through which the deep learning model can better adapt to complex nonlinear relationships in power grids , has stronger universality and reliability, and is simple to calculate.

As shown in Figure 1, in the embodiment of the present invention, a method for identifying a multi-harmonic source in a distribution network is provided, and the method includes the following steps:

Step S1, obtaining the harmonic voltage signals generated by the superposition of multiple harmonic sources at the PCC node, and measuring the harmonic voltage value;

Step S2, separating the fast-changing component and the slow-changing component in the harmonic voltage signal, and estimating the harmonic current value injected into the PCC node based on the fast-changing component;

Step S3, import the obtained harmonic current value and the harmonic voltage value into the pre-trained mutual information deep learning model, and obtain the interaction between the harmonic current emitted by each harmonic source and the harmonic voltage of the PCC node. Information value, and according to the obtained mutual information value, identify the main harmonic source.

The specific process is, before step S1, construct and train a neural network-based mutual information deep learning model, and the specific steps are as follows:

First, determine the structure of the neural network, including an input layer, a multi-layer hidden layer, and an output layer; where the input layer receives samples of harmonic current value Y and harmonic voltage value X as input, and the hidden layer undergoes a series of nonlinear transformations Get the result Z of the output layer.

Secondly, as shown in Figure 2, the neural network is trained, specifically:

(1) Initialize the neural network parameters;

(2) Draw b minibatch samples from the joint distribution; where, the b minibatch samples drawn from the joint distribution are recorded as (x ⁽¹⁾ ,y ⁽¹⁾ ),...,(x ^(b) ,y ^{( b)} ) obey the joint probability distribution

(3) Draw b samples from the Y marginal distribution; where, the b samples drawn from the Y marginal distribution are denoted as obey the joint probability distribution />

(4) Evaluate the lower bound of mutual information; where the formula for calculating the lower bound of mutual information is:

(5) Use EMA to correct the deviation correction gradient; wherein, the EMA correction deviation correction gradient is expressed as:

(6) update the parameters of the neural network; wherein, the parameters of the updated neural network are

(7) Steps (1) to (6) are repeated until the convergence condition is reached, and a trained mutual information deep learning model is obtained.

It should be noted that in a neural network, the mutual information lower bound between X and Y is calculated using the KL divergence calculated using the dual form:

Among them, T is all functions defined on X×Y that make the two expectations finite; And I(X; Y)≥I _θ (X,Y); I(X; Y)=H(X)-H(X|Y), at this time the mutual information between X and Y is equivalent to the joint probability The KL divergence of the product of the distribution and the marginal probability distribution is /> H is information entropy, H(X|Y) is conditional information entropy,

Denoted as the amount of neural information, it can be maximized by gradient descent, where the gradient formula is />

At this time, the mutual information deep learning model can be obtained by the formula To represent, among them, /> Represents the empirical distribution of the distribution P when n independent and identically distributed samples are given.

In step S1, the harmonic voltage signals generated by the superposition of multiple harmonic sources at the PCC node are directly obtained, and only the harmonic voltage value X needs to be measured.

In step S2, firstly, a filter is used to separate out fast changing components and slow changing components in the harmonic voltage signal.

Secondly, based on the rapidly changing component, the harmonic current value Y injected by the PCC node is estimated by the FastICA algorithm, as follows:

(1) Set the observed signal X to represent the PCC node harmonic voltage; the mixing matrix A to represent the admittance matrix Z; the source signal S to represent the injected harmonic current source, and the estimation matrix Y to represent the estimated harmonic current value;

(2) Estimate the Y value through the FastICA algorithm, making it infinitely close to the harmonic current source, including data preprocessing and establishing an objective function for optimization;

Among them, data preprocessing includes decentralization processing and whitening processing; decentralization processing refers to subtracting all sampled signals from their mean value to obtain a set of original data with an average value of zero: Among them, X(t _i ) is the sampling signal; the whitening process is the process of decorrelating the sampling data, so that the observed signal X has unit variance: X=ED ^-1/2 E ^T X; where, /> It consists of n eigenvalues; the eigenvector corresponding to d _i is c _i , E=[c ₁ ,c ₂ ,...c _n ]; the observed signal X=AS, A is the mixing matrix, and S is the injection of the PCC node Harmonic current source;

Among them, the process of establishing the objective function for optimization is to find a direction to maximize the non-Gaussianity of the output w ^T X (y=w ^T X); at this time, the non-Gaussianity is measured by the approximate value of negative entropy: N _g (Y )＝{E[g(Y)]-E[g(Y _Gauss )]} ² , that is to find the maximum value of J _G (w)=[E{G(w ^T X)}] ² ; where, w is m Dimension variable, representing a row of the unmixing matrix W;

Define the objective function as: According to the Kunhn-Tucker condition, it is transformed into an unconstrained optimization problem, so that the objective function is transformed into F(w)=E[G(w ^T X)]+C(||w|| ² -1); among them, the objective function The optimal solution of is obtained by the Newton iteration method: />

In step S3, firstly, import the harmonic current value Y and the harmonic voltage value X into the trained mutual information deep learning model to obtain the relationship between the harmonic current emitted by each harmonic source and the harmonic voltage of the PCC node Mutual information value; secondly, according to the obtained mutual information values, identify the main harmonic source. In one example, the harmonic current corresponding to the maximum mutual information value is found, and based on the found harmonic current, the corresponding emitted harmonic source is determined, that is, the main harmonic source.

As shown in Fig. 3, it is a distribution network multi-harmonic source identification system provided in an embodiment of the present invention, including:

The harmonic voltage acquisition unit 110 is used to acquire the harmonic voltage signals generated by the superposition of multiple harmonic sources at the PCC node, and measure the harmonic voltage value;

The harmonic current estimating unit 120 is configured to separate the fast-changing component and the slow-changing component in the harmonic voltage signal, and estimate the harmonic current value injected into the PCC node based on the fast-changing component;

The main harmonic source identification unit 130 is used to import the obtained harmonic current value and the harmonic voltage value into the pre-trained mutual information deep learning model to obtain the harmonic current emitted by each harmonic source and the PCC node The mutual information value between the harmonic voltages, and according to the obtained mutual information values, the main harmonic source is identified.

Implementing the embodiment of the present invention has the following beneficial effects:

1. The present invention only needs to obtain the PCC node harmonic voltage signal, use the FastICA algorithm to estimate the harmonic current, and perform mutual information estimation through the mutual information deep learning model to identify the main harmonic source, so that the operation is simple, and the system network parameters The position of the harmonic current source can be identified when it is unknown, which solves the problem that the existing methods are difficult to effectively identify and locate the multi-harmonic source in the complex distribution network;

2. The FastICA algorithm in the present invention estimates the harmonic current, which not only has fast convergence speed, good separation effect, stable iteration, and can separate non-Gaussian independent components, but also improves the accuracy and reliability of harmonic current estimation;

3. The mutual information deep learning model in the present invention makes the calculation simple, can better adapt to the complex nonlinear relationship in the power grid, and has stronger universality and reliability.

It is worth noting that in the above system embodiments, each system unit included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific functions of each functional unit The names are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the present invention.

Those of ordinary skill in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage Media such as ROM/RAM, magnetic disk, optical disk, etc.

The above disclosures are only preferred embodiments of the present invention, and certainly cannot limit the scope of rights of the present invention. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope of the present invention.

Claims (9)

1. A multi-harmonic source identification method for distribution network, characterized in that, said method comprises the following steps: Obtain the harmonic voltage signal generated by the superposition of multiple harmonic sources at the PCC node, and measure the harmonic voltage value; separating fast-changing components and slow-changing components in the harmonic voltage signal, and estimating a harmonic current value injected into the PCC node based on the fast-changing components; Import the obtained harmonic current value and the harmonic voltage value into the pre-trained mutual information deep learning model to obtain the mutual information value between the harmonic current emitted by each harmonic source and the harmonic voltage of the PCC node, And according to the obtained mutual information values, the main harmonic source is identified. 2. The method for identifying multi-harmonic sources in a distribution network according to claim 1, wherein the harmonic voltage signal is separated from the fast-changing component and the slow-changing component by a filter. 3. the multi-harmonic source identification method of distribution network as claimed in claim 1, is characterized in that, the harmonic current value that described PCC node injects is realized by FastICA algorithm; Wherein, The FastICA algorithm includes the steps of: data preprocessing and establishing an objective function for optimization. 4. The multi-harmonic source identification method of distribution network as claimed in claim 3, wherein said data preprocessing comprises decentralization and whitening; wherein, The decentralization process refers to subtracting the mean value of all sampled signals to obtain a set of original data with a mean value of zero: Among them, X(t _i ) is the sampling signal; The whitening process is a process of decorrelating the sampled data, so that the observed signal X has unit variance: X=ED ^{-1/ 2} E ^T X; wherein, It consists of n eigenvalues; the eigenvector corresponding to d _i is c _i , E=[c ₁ ,c ₂ ,...c _n ]; the observed signal X=AS, A is a mixing matrix, and S is a PCC node Injected harmonic current source. 5. The multi-harmonic source identification method of distribution network as claimed in claim 3, characterized in that, the process of setting up the objective function for optimization is to find a direction so that output w ^T X (y=w ^T X) Non-Gaussian is the largest; among them, Non-Gaussianity is measured by the approximate value of negative entropy: N _g (Y)={E[g(Y)]-E[g(Y _Gauss )]} ² , that is, J _G (w)=[E{G( w ^T X)}] ² maximum values; wherein, w is an m-dimensional variable, representing a row of the unmixing matrix W; The objective function is defined as: According to the Kunhn-Tucker condition, it is converted into an unconstrained optimization problem, so that the objective function is transformed into: F(w)=E[G(w ^T X)]+C(||w|| ^2-1 ); , the optimal solution of the objective function is obtained by the Newton iterative method: /> 6. The multi-harmonic source identification method of distribution network as claimed in claim 5, wherein the mutual information deep learning model is constructed based on a neural network; wherein, The neural network includes an input layer, a multi-layer hidden layer and an output layer; wherein, the input layer receives samples of the harmonic current value Y and the harmonic voltage value X as input, and the hidden layer undergoes a series of nonlinear After transformation, the result Z of the output layer is obtained. 7. The multi-harmonic source identification method of distribution network as claimed in claim 6, wherein the neural network is trained by performing the following steps, specifically comprising: 7.1 Initialize the neural network parameters; 7.2 Draw b minibatch samples from the joint distribution; wherein, the b minibatch samples drawn from the joint distribution are recorded as (x ⁽¹⁾ ,y ⁽¹⁾ ),...,(x ^(b) ,y ^{( b)} ) obey the joint probability distribution 7.3 Draw b samples from the Y marginal distribution; wherein, the b samples drawn from the Y marginal distribution are denoted as obey the joint probability distribution /> 7.4 Evaluate the lower bound of mutual information; wherein, the formula for calculating the lower bound of mutual information is: 7.5 Use EMA to correct the deviation correction gradient; wherein, the EMA correction deviation correction gradient is expressed as: 7.6 update the parameters of the neural network; wherein, the parameters of the updated neural network are 7.7 Repeat steps 7.1 to 7.6 until the convergence condition is reached. 8. the multi-harmonic source identification method of distribution network as claimed in claim 7, is characterized in that, the mutual information lower bound between described X and Y utilizes the KL divergence that uses dual form to calculate: Among them, T is all functions defined on X×Y that make the two expectations finite; And I(X;Y)≥I _θ (X,Y);/> is the amount of neural information, and by referring to the gradient formula as descend to maximize; Wherein, the mutual information deep learning model can pass the formula To represent, among them, /> Represents the empirical distribution of the distribution P when n independent and identically distributed samples are given. 9. A distribution network multi-harmonic source identification system, characterized in that it comprises: The harmonic voltage acquisition unit is used to acquire the harmonic voltage signals generated by the superposition of multiple harmonic sources at the PCC node, and measure the harmonic voltage value; a harmonic current estimating unit, configured to separate fast-changing components and slowly-changing components in the harmonic voltage signal, and estimate the harmonic current value injected into the PCC node based on the fast-changing component; The main harmonic source identification unit is used to import the obtained harmonic current value and the harmonic voltage value into the pre-trained mutual information deep learning model, and obtain the harmonic current emitted by each harmonic source and the PCC node harmonic The mutual information value between wave voltages, and according to the obtained mutual information values, identify the main harmonic source.