CN117411033B - An inertia damping control method and system based on radial neural network - Google Patents

Abstract

The invention provides an inertia damping control method based on a radial neural network, which adopts improved self-adaptive algorithm simulation to obtain initialization parameters of a synchronous motor; the initialization parameters comprise angular speed, angular speed offset rate, output inertia and output damping; screening the initialization parameters and classifying the initialization parameters into training set parameters and testing set parameters; performing artificial radial neural network training on the training set parameters and the testing set parameters to obtain the output inertia and output damping values; and carrying out error evaluation on the numerical value of the output inertia and the output damping. The invention can improve the control precision and response speed, optimize parameter design and adjustment, and realize excellent dynamic performance such as smoother power output, smaller overshoot, smaller frequency deviation, better power fluctuation suppression and the like.

Description

An inertial damping control method and system based on radial neural network

Technical field

The invention relates to the technical field of power grid control, and in particular to an inertia damping control method based on a radial neural network.

Background technique

As the proportion of renewable energy power generation increases, there will be less and less inertia and damping, seriously threatening the stability and reliability of power grid operations. Renewable energy power generation has volatility and uncertainty, which reduces the inertia and damping in the power grid, leading to excessive fluctuations in grid frequency and even large-scale power outages.

Grid frequency is an important indicator reflecting the balance of supply and demand in the grid, and its stability depends on the inertia and damping in the grid. Inertia can buffer power fluctuations in the power grid, and damping can suppress oscillations in the power grid.

Document "Self-adaptive virtual inertia control-based fuzzy logic to improve frequency stability of microgrid with high renewable penetration [J]. IEEE Access, 2019,7:76071-83. Penetration Microgrid Frequency Stability [J]. IEEE Access Magazine, 2019, 7:76071-83.)" discloses an adaptive virtual inertial control system using fuzzy logic, utilizing the input signal of real power injection and the system frequency Deviation, the virtual inertia constant is automatically adjusted to ensure frequency stability. This solution also uses intelligent control technology to optimize the parameters of the virtual synchronous machine to achieve fast inertial response and frequency stability. However, the shortcomings of this solution are that the design of the fuzzy control algorithm is entirely based on experience, without a clear theoretical basis, there are problems with rough parameter design, and the influence of damping is not considered.

Contents of the invention

The present invention provides an inertial damping control method based on radial neural network, including:

Use the improved adaptive algorithm simulation to obtain the initialization parameters of the synchronous motor;

The initialization parameters include angular velocity, angular velocity deviation rate, output inertia and output damping;

Filter the initialization parameters and classify them into training set parameters and test set parameters;

Perform artificial radial neural network training on the training set parameters and the test set parameters to obtain the values of the output inertia and output damping;

An error evaluation is performed on the values of the output inertia and the output damping.

It should be noted that the artificial radial neural network is a radial basis function network (RBF), which has the following structure and connection method:

(1) The input layer consists of two neurons, which receive power grid frequency deviation and power fluctuation as input signals respectively;

(2) The hidden layer is composed of several neurons, each neuron corresponds to a Gaussian basis function, and the center value, standard deviation and weight of the basis function are adaptively adjusted according to the feature points of the input signal;

(3) The output layer consists of two neurons, which respectively output virtual inertia and damping coefficient as control signals.

It should be noted that the steps of training the artificial radial neural network on the training set parameters and the test set parameters include:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the The calculation formula is:/> ;

Among them, the Represents the nearest clustering base point, /> Represents distance,/> represents the overlap coefficient.

It should be noted that the overlap coefficient is a positive real number, used to control the width of the Gaussian basis function. Its value range is (0,1), and its determination method is:/> ,wherein,/> is the maximum distance between clustering base points.

It should be noted that the step of training the artificial radial neural network on the training set parameters and the test set parameters also includes:

Determine the error of k and k-1 training set parameters;

Compare the error of the training set parameters with the set parameter values;

If the error of the training set parameters is less than the set parameter value, proceed:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the step of training the artificial radial neural network on the training set parameters and the test set parameters also includes:

Determine the error of k and k-1 training set parameters;

It should be further explained that k means the number of parameters in the current training set, and k-1 means the number of parameters in the previous training set. Therefore, this step can also be expressed as: judging the error between the number of parameters k in the current training set and the number of parameters k-1 in the previous training set.

Compare the error of the training set parameters with the set parameter values;

If the error of the training set parameters is greater than the set parameter value, adjust the values of the i-th clustering base point and weight;

Further, the step of "adjusting the numerical size of the i-th clustering base point and weight" includes:

Calculate error function ,wherein,/> is the number of neurons in the output layer,/> is the expected output value,/> is the actual output value;

Calculate the weight and error function of the i-th clustering base point pair partial derivative of ;

Update the weight and i-th clustering base point according to the partial derivative;

Repeat the above steps until the error function reaches the minimum value or meets the stopping condition;

When the error of the training set parameters is less than the set parameter value, and the error of the training set parameter is less than the set parameter value, proceed:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the steps of error evaluation of the values of the output inertia and the output damping include:

Perform statistics on the error value of the output inertia;

Perform statistics on the error value of the output damping;

Determine whether the error rate of the output inertia is less than one thousandth;

Determine whether the error rate of the output damping is less than one thousandth.

An inertial damping control system based on radial neural network, including:

The sampling module is used to simulate using the improved adaptive algorithm to obtain the initialization parameters of the synchronous motor;

The initialization parameters include angular velocity, angular velocity deviation rate, output inertia and output damping;

A signal processing module, used to filter the initialization parameters and classify them into training set parameters and test set parameters;

A radial neural network control module, used to perform artificial radial neural network training on the training set parameters and the test set parameters to obtain the values of the output inertia and output damping;

An error evaluation module is used to perform error evaluation on the values of the output inertia and the output damping.

It should be noted that the radial neural network control module includes:

The judgment unit is used to judge the error of k and k-1 training set parameters;

A comparison unit, used to compare the error of the training set parameter with the set parameter value;

The operation unit is used to perform the following steps if the error of the training set parameters is less than the set parameter value:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the radial neural network control module includes:

The judgment unit compares the error of the training set parameter with the set parameter value;

Comparing unit, if the error of the training set parameter is greater than the set parameter value, the value of the i-th clustering base point and weight is adjusted;

The adjustment unit adjusts the error of the training set parameters to be less than the set parameter value;

The computing unit runs the following formula:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the error evaluation module includes:

A statistical unit, used to make statistics on the error value of the output inertia; used to make statistics on the error value of the output damping;

An error judgment unit is used to judge whether the error rate of the output inertia is less than one thousandth; and to judge whether the error rate of the output damping is less than one thousandth.

Compared with the existing technology, the inertia damping control method based on the radial neural network selected by the present invention uses an improved adaptive algorithm simulation to obtain the initialization parameters of the synchronous motor; the initialization parameters include angular velocity, angular velocity deviation rate, Output inertia and output damping; filter the initialization parameters and classify them into training set parameters and test set parameters; perform artificial radial neural network training on the training set parameters and test set parameters to obtain the output inertia and output The value of damping; perform error evaluation on the output inertia and the value of output damping. The present invention uses radial neural network control technology to simultaneously consider the effects of inertia and damping, optimize external characteristics, and enhance the stability and reliability of the power grid. However, the existing technology does not consider the impact of damping; it can improve control accuracy and response speed, and optimize Parameter design and adjustment, while the existing technology has large overshoot and adjustment time, and poor control effect. The present invention uses radial neural network control technology to achieve smoother power output, smaller overshoot, smaller frequency deviation, better power fluctuation suppression and other excellent dynamic performances; it can also achieve rapid convergence, high precision and low complexity. , strong generalization ability and other characteristics, and optimize the network structure and parameters by adjusting the clustering base points and weights.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts, among which:

Figure 1 is a schematic flow chart of an inertia damping control method based on radial neural networks provided by the present invention;

Figure 2 is a schematic flow chart of an embodiment of step S3 shown in Figure 1;

Figure 3 is a schematic flow chart of another embodiment of step S3 shown in Figure 1;

Figure 4 is a schematic flow chart of step S4 shown in Figure 1;

Figure 5 is a schematic diagram of an inertia damping control system based on a radial neural network provided by the present invention;

Figure 6 is a schematic diagram of an embodiment of the radial neural network control module provided in Figure 5;

Figure 7 is a schematic diagram of another embodiment of the radial neural network control module provided in Figure 5;

FIG. 8 is a schematic diagram of the error evaluation module provided in FIG. 5 .

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Please refer to Figure 1, which is a schematic flow chart of an inertia damping control method based on a radial neural network provided by the present invention; including:

S1. Use the improved adaptive algorithm simulation to obtain the initialization parameters of the synchronous motor;

S2. Filter the initialization parameters and classify them into training set parameters and test set parameters;

S3. Perform artificial radial neural network training on the training set parameters and the test set parameters to obtain the values of the output inertia and output damping;

S4. Perform error evaluation on the values of the output inertia and the output damping.

It should be noted that the steps of training the artificial radial neural network on the training set parameters and the test set parameters include:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the The calculation formula is:/> ;

Among them, the Represents the nearest clustering base point, /> Represents distance,/> represents the overlap coefficient.

It should be noted that the overlap coefficient is a positive real number, used to control the width of the Gaussian basis function. Its value range is (0,1), and its determination method is:/> , where,/> is the maximum distance between clustering base points.

Please refer to Figure 2, which is a schematic flow chart of an embodiment of step S3 shown in Figure 1; including:

S31. Determine the error of k and k-1 training set parameters;

S32. Compare the error of the training set parameter with the set parameter value;

S33. The error of the training set parameters is less than the set parameter value;

S34. Perform the calculation of the following formula:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be further explained that k means the number of parameters in the current training set, and k-1 means the number of parameters in the previous training set. Therefore, this step can also be expressed as: judging the error between the number of parameters k in the current training set and the number of parameters k-1 in the previous training set.

Please refer to Figure 3, which is a schematic flow chart of another embodiment of step S3 shown in Figure 1; including:

S31. Determine the error of k and k-1 training set parameters;

S32. Compare the error of the training set parameter with the set parameter value;

S303. If the error of the training set parameter is greater than the set parameter value, adjust the value of the i-th clustering base point and weight;

S304. Adjust the error of the training set parameters until the error is less than the set parameter value;

S34. Perform the calculation of the following formula:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

Further, the S33 includes:

S331. Calculate error function ;

in, is the number of neurons in the output layer,/> is the expected output value,/> is the actual output value;

S332. Calculate the weight and error function of the i-th clustering base point pair partial derivative of ;

S333. Update the weight and the i-th clustering base point according to the partial derivative;

S334. Repeat the above steps until the error function The minimum value is reached or the stopping condition is met.

As shown in Figure 4, Figure 4 is a schematic flow chart of step S4 shown in Figure 1;

include:

S41. Calculate the error value of the output inertia;

S42. Calculate the error value of the output damping;

S43. Determine whether the error rate of the output inertia is less than one thousandth;

S44. Determine whether the error rate of the output damping is less than one thousandth.

Please refer to Figure 5, which is a schematic diagram of an inertia damping control system based on a radial neural network provided by the present invention; including:

Sampling module 1 is used to simulate using the improved adaptive algorithm to obtain the initialization parameters of the synchronous motor;

The initialization parameters include angular velocity, angular velocity deviation rate, output inertia and output damping;

Signal processing module 2, used to filter the initialization parameters and classify them into training set parameters and test set parameters;

Radial neural network control module 3 is used to perform artificial radial neural network training on the training set parameters and the test set parameters to obtain the values of the output inertia and output damping;

Error evaluation module 4 is used to perform error evaluation on the values of the output inertia and the output damping.

Please refer to Figure 6, which is a schematic diagram of an embodiment of the radial neural network control module provided in Figure 5; the radial neural network control module includes:

The judgment unit 31 is used to judge the errors of k and k-1 training set parameters;

The comparison unit 32 is used to compare the error of the training set parameter with the set parameter value;

The computing unit 33 is used to perform: when the error of the training set parameters is less than the set parameter value:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

Please refer to Figure 7, which is a schematic diagram of another embodiment of the radial neural network control module provided in Figure 5;

The radial neural network control module includes:

The judgment unit 31 compares the error of the training set parameter with the set parameter value;

Comparing unit 32, if the error of the training set parameter is greater than the set parameter value, adjust the value of the i-th clustering base point and weight;

The adjustment unit 302 adjusts until the error of the training set parameters is less than the set parameter value;

The computing unit 33 runs the following formula:

;

in, Represents the i-th output inertia;/> represents the i-th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer;

in, is the angular velocity of the i-th,/> is the angular velocity deviation rate of the i-th,/> is the standard deviation of the i-th,/> is the central value of the i-th basis function.

It should be noted that the error evaluation module includes:

The statistics unit 41 is used to make statistics on the error value of the output inertia; used to make statistics on the error value of the output damping;

The error judgment unit 42 is used to judge whether the error rate of the output inertia is less than one thousandth; and to judge whether the error rate of the output damping is less than one thousandth.

Compared with the existing technology, the inertia damping control method based on the radial neural network selected by the present invention uses an improved adaptive algorithm simulation to obtain the initialization parameters of the synchronous motor; the initialization parameters include angular velocity, angular velocity deviation rate, Output inertia and output damping; filter the initialization parameters and classify them into training set parameters and test set parameters; perform artificial radial neural network training on the training set parameters and test set parameters to obtain the output inertia and output The value of damping; perform error evaluation on the output inertia and the value of output damping. The present invention uses radial neural network control technology to simultaneously consider the effects of inertia and damping, optimize external characteristics, and enhance the stability and reliability of the power grid, while the existing technology does not consider the effects of damping. It can improve control accuracy and response speed, and optimize parameter design and adjustment. However, the existing technology has large overshoot and adjustment time and poor control effect. The present invention uses radial neural network control technology to achieve smoother power output, smaller overshoot, smaller frequency deviation, better power fluctuation suppression and other excellent dynamic performances.

Claims (6)

1. An inertia damping control method based on radial neural network, characterized by including: Use the improved adaptive algorithm simulation to obtain the initialization parameters of the synchronous motor; The initialization parameters include angular velocity, angular velocity deviation rate, output inertia and output damping; Filter the initialization parameters and classify them into training set parameters and test set parameters; Perform artificial radial neural network training on the training set parameters and the test set parameters to obtain the values of the output inertia and output damping; Perform error evaluation on the values of the output inertia and the output damping; The steps of performing artificial radial neural network training on the training set parameters and the test set parameters include: ; in, Represents the i -th output inertia;/> represents the i -th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer; in, is the angular velocity of the i -th,/> is the angular velocity deviation rate of the i -th,/> is the standard deviation of the i -th,/> is the central value of the i -th basis function; described The calculation formula is:/> ; Among them, the Represents the nearest clustering base point, /> Represents distance,/> Represents the overlap coefficient; The steps of error evaluation of the values of the output inertia and the output damping include: Perform statistics on the error value of the output inertia; Perform statistics on the error value of the output damping; Determine whether the error rate of the output inertia is less than one thousandth; Determine whether the error rate of the output damping is less than one thousandth. 2. The inertia damping control method based on radial neural network as claimed in claim 1, characterized in that the step of performing artificial radial neural network training on the training set parameters and the test set parameters further includes: Determine the error of k and k- 1 training set parameters; Compare the error of the training set parameters with the set parameter values; If the error of the training set parameters is less than the set parameter value, proceed: ; in, Represents the i -th output inertia;/> represents the i -th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer; in, is the angular velocity of the i -th,/> is the angular velocity deviation rate of the i -th,/> is the standard deviation of the i -th,/> is the central value of the i -th basis function. 3. The inertia damping control method based on radial neural network as claimed in claim 1, characterized in that the step of performing artificial radial neural network training on the training set parameters and the test set parameters further includes: Determine the error between the k -th and k- 1th training set parameters; Compare the error of the training set parameters with the set parameter values; If the error of the training set parameters is greater than the set parameter value, adjust the values of the i-th clustering base point and weight; The error of the training set parameters is adjusted to be less than the set parameter value; Further, the step of adjusting until the error of the training set parameters is less than the set parameter value includes: Calculate error function ; in, is the number of neurons in the output layer,/> is the expected output value,/> is the actual output value; Calculate the weight and error function of the i-th clustering base point pair partial derivative of ; Update the weight and i-th clustering base point according to the partial derivative; Repeat the above steps until the error function reaches the minimum value or meets the stopping condition; When the error of the training set parameters is less than the set parameter value, proceed: ; in, Represents the i -th output inertia;/> represents the i -th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer; in, is the angular velocity of the i -th,/> is the angular velocity deviation rate of the i -th,/> is the standard deviation of the i -th,/> is the central value of the i -th basis function. 4. An inertia damping control system based on radial neural network, characterized by including: The sampling module is used to simulate using the improved adaptive algorithm to obtain the initialization parameters of the synchronous motor; The initialization parameters include angular velocity, angular velocity deviation rate, output inertia and output damping; A signal processing module, used to filter the initialization parameters and classify them into training set parameters and test set parameters; A radial neural network control module, used to perform artificial radial neural network training on the training set parameters and the test set parameters to obtain the values of the output inertia and output damping; Among them, the radial neural network control module runs the following operations: ; in, Represents the i -th output inertia;/> represents the i -th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer; in, is the angular velocity of the i -th,/> is the angular velocity deviation rate of the i -th,/> is the standard deviation of the i -th,/> is the central value of the i -th basis function; described The calculation formula is:/> ; Among them, the Represents the nearest clustering base point, /> Represents distance,/> Represents the overlap coefficient; An error evaluation module, used to perform error evaluation on the values of the output inertia and the output damping; The error evaluation module includes: A statistical unit, used to make statistics on the error value of the output inertia; used to make statistics on the error value of the output damping; An error judgment unit is used to judge whether the error rate of the output inertia is less than one thousandth; and to judge whether the error rate of the output damping is less than one thousandth. 5. The inertia damping control system based on radial neural network as claimed in claim 4, characterized in that the radial neural network control module includes: The judgment unit is used to judge the error of k and k- 1 training set parameters; A comparison unit, used to compare the error of the training set parameter with the set parameter value; The operation unit is used to perform the following steps if the error of the training set parameters is less than the set parameter value: ; in, Represents the i -th output inertia;/> represents the i -th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer; in, is the angular velocity of the i -th,/> is the angular velocity deviation rate of the i -th,/> is the standard deviation of the i -th,/> is the central value of the i -th basis function. 6. The inertia damping control system based on radial neural network as claimed in claim 4, characterized in that the radial neural network control module includes: The judgment unit compares the error of the training set parameter with the set parameter value; Comparing unit, if the error of the training set parameter is greater than the set parameter value, the value of the i-th clustering base point and weight is adjusted; The adjustment unit adjusts the error of the training set parameters to be less than the set parameter value; The computing unit runs the following formula: ; in, Represents the i -th output inertia;/> represents the i -th output damping, i represents the sequence number, and m represents the number of neurons in the current hidden layer; in, is the angular velocity of the i -th,/> is the angular velocity deviation rate of the i -th,/> is the standard deviation of the i -th,/> is the central value of the i -th basis function.