Inverter voltage control method and system

Abstract

The present invention discloses an inverter voltage control method and system, the method comprising: sampling the voltage and current data of an LC inverter; inputting the sampled data into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: on the basis of a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and an AMSGrad optimization algorithm is used to improve the single neuron PI controller. The inverter voltage control method provided by the present invention can reduce the voltage overshoot of the LC inverter and make the response speed fast, and at the same time has good anti-interference ability and steady-state performance, which significantly improves the dynamic quality of the power system.

Description

Technical Field

[0001] The invention relates to an inverter voltage control method and system, belonging to the technical field of inverter voltage control. Background Art

[0002] With the advantages of simple structure, strong robustness, and high steady-state accuracy, the voltage-current dual-loop PI control strategy is widely used in the voltage tracking control of LC inverters. However, with the increase in the penetration rate of new energy and the development of AC / DC microgrid technology, the safe and stable operation of the power system will become more dependent on the control method of the inverter, which puts higher requirements on the control quality of the inverter. The traditional voltage loop PI controller cannot take into account both overshoot and rapidity indicators, and its dynamic response performance is insufficient, making it difficult to achieve high-quality voltage control of the inverter.

[0003] To this end, experts and scholars have made many improvements to the traditional PI controller. The error and error integral are used as the input of the fuzzy controller, and the parameters of the PI controller are adjusted according to the historical error, which improves the control effect and adaptability of the electric spring; the structure of the traditional PI controller is optimized, and a proportional link is added to its feedback channel to improve the control performance of the permanent magnet synchronous motor speed control system; the control scheme based on variable parameter PI is adopted to effectively suppress the power mutation of the doubly fed wind turbine at the initial moment of speed recovery, and avoid the secondary drop of the system frequency. As a typical method of advanced control theory, single neuron PI has been widely used in direct drive pump systems, dimming power supply control, agricultural machinery navigation and other occasions, but there are few reports on its use in LC inverter voltage control. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide an inverter voltage control method that can reduce the voltage overshoot of the LC inverter and provide a fast response speed, while having good anti-interference ability and steady-state performance, and significantly improving the dynamic quality of the power system. To achieve the above purpose, the present invention is implemented by adopting the following technical solutions:

[0005] In a first aspect, the present invention provides an inverter voltage control method, comprising:

[0006] Sampling the voltage and current data of the LC inverter;

[0007] The sampled data is input into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: based on a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and the AMSGrad optimization algorithm is used to improve the single neuron PI controller.

[0008] In combination with the first aspect, further, the LC inverter adopts a control method of dual-loop control with feedforward compensation, namely, an inner loop of inductor current and an outer loop of capacitor voltage.

[0009] In combination with the first aspect, further, the obtaining of voltage and current data of the LC inverter includes:

[0010] According to the main circuit of the three-phase LC inverter and Kirchhoff's law, the voltage and current equations of the inverter in the d, q synchronous coordinate system are obtained as follows:

[0011]

[0012] In formula (1), L is the three-phase filter inductor, R is the parasitic resistance of the filter inductor L, C is the three-phase filter capacitor, i d 、i q They represent the d-axis and q-axis components of the inverter inductor current, respectively, and u d 、u q They represent the d-axis and q-axis components of the inverter output voltage, respectively. od 、u oq Respectively represent the d-axis and q-axis components of the three-phase capacitor voltage, i od 、i oq They represent the d-axis and q-axis components of the three-phase load current respectively, and ω represents the angular frequency.

[0013] In combination with the first aspect, further, the single neuron PI controller is composed of a single neuron algorithm, and the algorithm of the single neuron PI controller is expressed by the following formula:

[0014]

[0015] In formula (2), u(m) represents the output of the controller at the mth sampling time; x p represents the proportional state variable, x i represents the integral state variable and is expressed by the following formula:

[0016]

[0017] In formula (3), T represents the sampling period; e(m) represents the error of the controlled object at the mth sampling moment, which is expressed by the following formula:

[0018] e(m)=r(m)-y(m) (4)

[0019] In formula (4), r(m) represents the input command value of the controlled object at the mth sampling, and y(m) represents the actual output value of the controlled object at the mth sampling;

[0020] In formula (2), k p represents the proportional weight, k irepresents the integral weight, and adaptive adjustment is achieved through the learning process of the neural network. In the process of weight adjustment, the quadratic performance index function J(m) is introduced. When J(m) is the smallest, the optimal control effect is achieved. p and k i Changing along the negative gradient direction, it satisfies the adaptive law:

[0021]

[0022] In formula (5), η represents the proportional weight k p and integral weight k i The learning rate; the quadratic performance index function J(m) is expressed by the following formula:

[0023]

[0024] From (6), we can see that in formula (5):

[0025]

[0026] From (2), we can see but:

[0027]

[0028] In formula (5), It is the Jacobian information of the controlled object, which is identified by the RBF neural network identification algorithm.

[0029] In combination with the first aspect, preferably, the single neuron PI controller dynamically adjusts the single neuron PI controller parameters according to the operating state of the power system.

[0030] In combination with the first aspect, further, the RBF neural network of the RBF neural network identification algorithm includes 2 input nodes, 5 hidden layer nodes and 1 output node, then:

[0031] Initialize the neural network input vector X = [x1, x2] T , where x1 = u(k-1), x2 = y(k-1), hidden layer radial basis vector H = [h1,h2,h3,h4,h5] T , weight vector W = [w1,w2,w3,w4,w5] T , then the output of the RBF neural network is:

[0032]

[0033] According to the gradient descent method, the weight, node center and base width parameters of the jth neuron node in the hidden layer are adjusted as follows:

[0034]

[0035] In formula (10), λ represents the weight w of the jth neuron node in the hidden layer j 、Node center b j and base width parameter c ji The learning rate; α is the momentum factor, and E(m) is the performance index function of RBF neural network identification, which is expressed by the following formula:

[0036]

[0037] In formulas (10) and (11), e d (m) represents the identification error of the RBF neural network at the mth sampling moment;

[0038] Initial parameters of RBF neural network It is expressed by the following formula:

[0039]

[0040] In formula (12), y min ,y max They represent the minimum and maximum values ​​of the expected output of the control object, respectively. min 、u max It is the minimum and maximum value of the controller output;

[0041] Then the Jacobian information of the controlled object is:

[0042]

[0043] In combination with the first aspect, preferably, the hidden layer radial basis vector is a Gaussian basis function, which is expressed by the following formula:

[0044]

[0045] In formula (14), C j is the center vector of the jth neuron node in the hidden layer, C j =[c j1 ,c j2 ] T , b j is the basis width parameter of the jth neuron node in the hidden layer.

[0046] In combination with the first aspect, further, the AMSGrad optimization algorithm uses historical gradient information based on the momentum method to achieve adaptive adjustment of the learning rate.

[0047] In combination with the first aspect, further, the pseudo code in the AMSGrad optimization algorithm includes:

[0048] Initial weight θ0, initial learning rate γ0, initial first-order momentum factor ε0, second-order momentum factor ρ, gradient g of the mth iteration m , the weight θ of the mth iteration m , the learning rate γ for the mth iteration m , the first-order momentum factor s for the mth iteration m , the first-order momentum S of the mth iteration m , the second-order momentum a of the mth iteration m , the second-order momentum of the mth iteration Constant σ.

[0049] In combination with the first aspect, preferably, the initial first-order momentum factor ε0=0.9.

[0050] In combination with the first aspect, preferably, the second-order momentum factor ρ=0.999.

[0051] In combination with the first aspect, preferably, the constant σ=10 -8 .

[0052] In combination with the first aspect, preferably, the method further includes: comparing the tracking performances of a traditional PI controller, a single neuron PI controller and a pre-optimized voltage loop controller.

[0053] In a second aspect, the present invention provides an inverter voltage control system, comprising: a sampling module: sampling voltage and current data of an LC inverter;

[0054] Processing module: used for inputting sampling data into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: based on a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and an AMSGrad optimization algorithm is used to improve the single neuron PI controller.

[0055] In a third aspect, the present invention provides a computer device, including a processor and a storage medium;

[0056] The storage medium is used to store instructions;

[0057] The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0058] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0059] Compared with the prior art, the inverter voltage control method and system provided by the embodiment of the present invention have the following beneficial effects:

[0060] The present invention samples the voltage and current data of the LC inverter; inputs the sampled data into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: based on a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and an AMSGrad optimization algorithm is used to improve the single neuron PI controller;

[0061] The present invention utilizes the advantages of neural network self-organization and self-learning to realize automatic adjustment of PI controller parameters, and adopts AMSGrad optimization gradient algorithm with adaptive learning rate to achieve better control performance of single neuron PI controller;

[0062] The inverter voltage control method provided by the present invention can reduce the voltage overshoot of the LC type inverter and make the response speed fast, and at the same time have good anti-interference ability and steady-state performance, which significantly improves the dynamic quality of the power system; the pre-optimized voltage loop controller provided by the present invention has the advantages of simple structure, good control effect, strong adaptability and convenient application, and is an effective way to achieve high-quality voltage control of the inverter in the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a flow chart of an inverter voltage control method in Embodiment 1 of the present invention;

[0064] Figure 2 It is an inverter topology structure of an inverter voltage control method in Embodiment 1 of the present invention;

[0065] Figure 3 It is a structural diagram of a pre-optimized voltage loop controller in an inverter voltage control method in Example 1 of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0067] Embodiment 1:

[0068] An inverter voltage control method, comprising:

[0069] Sampling the voltage and current data of the LC inverter;

[0070] The sampled data is input into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is as follows: Figure 3 As shown in the figure: On the basis of the single neuron PI controller, the RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and the AMSGrad optimization algorithm is used to improve the single neuron PI controller.

[0071] The LC inverter adopts a dual-loop control method with feedforward compensation, i.e., an inner loop of inductor current and an outer loop of capacitor voltage. Both loops use PI controllers to reduce the system steady-state error and improve voltage tracking capability.

[0072] Get the voltage and current data of LC inverter, including:

[0073] like Figure 2 The figure shows the inverter topology. According to the main circuit of the three-phase LC inverter and Kirchhoff's law, the voltage and current equations of the inverter in the d, q synchronous coordinate system are:

[0074]

[0075] In formula (1), L is the three-phase filter inductor, R is the parasitic resistance of the filter inductor L, C is the three-phase filter capacitor, i d 、i q They represent the d-axis and q-axis components of the inverter inductor current, respectively, and u d 、u q They represent the d-axis and q-axis components of the inverter output voltage, respectively. od 、u oq Respectively represent the d-axis and q-axis components of the three-phase capacitor voltage, i od 、i oq They represent the d-axis and q-axis components of the three-phase load current respectively, and ω represents the angular frequency.

[0076] Specifically, the single neuron PI controller can learn from environmental changes and dynamically adjust controller parameters according to the system operating status. It has strong robustness, can suppress overshoot, and improve the dynamic response capability of the voltage loop. The single neuron PI controller is designed based on the digital incremental PI algorithm using a neural network. Assume that the error of the controlled object at the mth sampling moment is:

[0077] e(m)=r(m)-y(m) (2)

[0078] In formula (2), r(m) represents the input command value of the controlled object at the mth sampling, and y(m) represents the actual output value of the controlled object at the mth sampling;

[0079] Set the neuron input to:

[0080]

[0081] In formula (3), x p represents the proportional state variable, x i represents the integral state variable, T represents the sampling period; then the algorithm of the single neuron PI controller is expressed by the following formula:

[0082]

[0083] In formula (4), u(m) represents the output of the controller at the mth sampling. The proportional state variable x p and the integrated state variable x i That is, the proportional and integral coefficients of the PI controller are adaptively adjusted through the learning process of the neural network.

[0084] In the process of weight adjustment, the quadratic performance index function J(m) is introduced. When J(m) is the smallest, the optimal control effect is achieved. p and k i Changing along the negative gradient direction, it satisfies the adaptive law:

[0085]

[0086] In formula (5), η represents the proportional weight k p and integral weight k i The learning rate; the quadratic performance index function J(m) is expressed by the following formula:

[0087]

[0088] From (6), we can see that in formula (5):

[0089]

[0090] From (2), we can see but:

[0091]

[0092] In formula (5), It is the Jacobian information of the controlled object, which is identified by the RBF neural network identification algorithm.

[0093] On the basis of the single neuron PI controller, the RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object. The specific steps include:

[0094] Step 1: Use an RBF neural network with 2 input nodes, 5 hidden layer nodes and 1 output node to identify the Jacobian information of the controlled object.

[0095] Step 2: Initialize the neural network input vector X = [x1, x2] T , where x1 = u(k-1), x2 = y(k-1), hidden layer radial basis vector H = [h1,h2,h3,h4,h5] T , weight vector W = [w1,w2,w3,w4,w5] T , where the hidden layer radial basis vector is a Gaussian basis function, which is expressed by the following formula:

[0096]

[0097] In formula (9), C j is the center vector of the jth neuron node in the hidden layer, C j =[c j1 ,c j2 ] T , b j is the basis width parameter of the jth neuron node in the hidden layer.

[0098] Then the output of the RBF neural network is:

[0099]

[0100] The identification error of the RBF neural network at the mth sampling moment is expressed by the following formula:

[0101] e d (m) = y(m) - y d (m) (11)

[0102] Step 3: Select the quadratic function as the performance index function of RBF neural network identification, which is expressed by the following formula:

[0103]

[0104] According to the gradient descent method, the weight, node center and base width parameters of the jth neuron node in the hidden layer are adjusted as follows:

[0105]

[0106] In formula (13), λ represents the weight w of the jth neuron node in the hidden layer j 、Node center b j and base width parameter c ji The learning rate; α is the momentum factor.

[0107] Then the Jacobian information of the controlled object is:

[0108]

[0109] Step 4: Determine the initial parameters of the RBF neural network, which are determined by the following formula:

[0110]

[0111] In formula (15), y min ,y max They represent the minimum and maximum values ​​of the expected output of the control object, respectively. min 、u max It is the minimum and maximum value of the controller output.

[0112] The AMSGrad optimization algorithm is used to improve the single neuron PI controller. Based on the momentum method, the historical gradient information is used to realize the adaptive adjustment of the learning rate, thereby improving the speed and stability of the algorithm.

[0113] The pseudo code of the AMSGrad optimization algorithm includes:

[0114] Initial weight θ0, initial learning rate γ0, initial first-order momentum factor ε0, second-order momentum factor ρ, gradient g of the mth iteration m , the weight θ of the mth iteration m , the learning rate γ for the mth iteration m , the first-order momentum factor s for the mth iteration m , the first-order momentum S of the mth iteration m , the second-order momentum a of the mth iteration m , the second-order momentum of the mth iteration Constant σ. In order to prevent the denominator from being 0, the initial first-order momentum factor ε0 = 0.9, the second-order momentum factor ρ = 0.999, and the constant σ = 10 -8 .

[0115] Also included is a comparison of the tracking performance of the traditional PI controller, the single neuron PI controller, and the pre-optimized voltage loop controller.

[0116] Embodiment 2:

[0117] This embodiment discloses an inverter voltage control system, including:

[0118] Sampling module: samples the voltage and current data of the LC inverter;

[0119] Processing module: used for inputting sampling data into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: based on a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and an AMSGrad optimization algorithm is used to improve the single neuron PI controller.

[0120] Embodiment three:

[0121] An embodiment of the present invention provides a computer device, including a processor and a storage medium;

[0122] The storage medium is used to store instructions;

[0123] The processor is used to operate according to the instructions to execute the steps of the method described in embodiment 1.

[0124] Embodiment 4:

[0125] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0127] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0130] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An inverter voltage control method, characterized in that: include: Sampling the voltage and current data of the LC inverter; The LC inverter adopts a dual-loop control method with feedforward compensation, i.e., an inner loop of inductor current and an outer loop of capacitor voltage; The step of sampling the voltage and current data of the LC inverter includes: According to the main circuit of the three-phase LC inverter and Kirchhoff's law, the voltage and current equations of the inverter in the d, q synchronous coordinate system are obtained as follows: In formula (1), L is the three-phase filter inductor, R is the parasitic resistance of the filter inductor L, C is the three-phase filter capacitor, i d 、i q They represent the d-axis and q-axis components of the inverter inductor current, respectively, and u d 、u q They represent the d-axis and q-axis components of the inverter output voltage, respectively. od 、u oq Respectively represent the d-axis and q-axis components of the three-phase capacitor voltage, i od 、i oq They represent the d-axis and q-axis components of the three-phase load current respectively, and ω represents the angular frequency; The sampled data is input into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: based on a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and an AMSGrad optimization algorithm is used to improve the single neuron PI controller; The single neuron PI controller is composed of a single neuron algorithm, and the algorithm of the single neuron PI controller is expressed by the following formula: In formula (2), u(m) represents the output of the controller at the mth sampling time; x p represents the proportional state variable, x i represents the integral state variable and is expressed by the following formula: In formula (3), T represents the sampling period; e(m) represents the error of the controlled object at the mth sampling moment, which is expressed by the following formula: e(m)=r(m)-y(m) (4) In formula (4), r(m) represents the input command value of the controlled object at the mth sampling, and y(m) represents the actual output value of the controlled object at the mth sampling; In formula (2), k p represents the proportional weight, k i represents the integral weight, and adaptive adjustment is achieved through the learning process of the neural network. In the process of weight adjustment, the quadratic performance index function J(m) is introduced. When J(m) is the smallest, the optimal control effect is achieved. p and k i Changing along the negative gradient direction, it satisfies the adaptive law: In formula (5), η represents the proportional weight k p and integral weight k i The learning rate; the quadratic performance indicator function J(m) is expressed by the following formula: From (6), we can see that in formula (5): From (2), we can see but: In formula (5), It is the Jacobian information of the controlled object, which is identified by the RBF neural network identification algorithm.

2. The inverter voltage control method according to claim 1, characterized in that: The RBF neural network of the RBF neural network identification algorithm includes 2 input nodes, 5 hidden layer nodes and 1 output node, then: Initialize the neural network input vector X = [x1, x2] T , where x1 = u(k-1), x2 = y(k-1), hidden layer radial basis vector H = [h1,h2,h3,h4,h5] T , weight vector W = [w1,w2,w3,w4,w5] T , then the output of the RBF neural network is: According to the gradient descent method, the weight, node center and base width parameters of the jth neuron node in the hidden layer are adjusted as follows: In formula (10), λ represents the weight w of the jth neuron node in the hidden layer j 、Node center b j and base width parameter c ji The learning rate; α is the momentum factor, and E(m) is the performance index function of RBF neural network identification, which is expressed by the following formula: In formulas (10) and (11), e d (m) represents the identification error of the RBF neural network at the mth sampling moment; Initial parameters of RBF neural network It is expressed by the following formula: In formula (12), y min ,y max They represent the minimum and maximum values ​​of the expected output of the control object, respectively. min 、u max It is the minimum and maximum value of the controller output; Then the Jacobian information of the controlled object is:

3. The inverter voltage control method according to claim 1, characterized in that: The AMSGrad optimization algorithm uses historical gradient information on the basis of the momentum method to achieve adaptive adjustment of the learning rate.

4. The inverter voltage control method according to claim 1, characterized in that: The pseudo code of the AMSGrad optimization algorithm includes: Initial weight θ0, initial learning rate γ0, initial first-order momentum factor ε0, second-order momentum factor ρ, gradient g of the mth iteration m , the weight θ of the mth iteration m , the learning rate γ for the mth iteration m , the first-order momentum factor s for the mth iteration m , the first-order momentum S of the mth iteration m , the second-order momentum a of the mth iteration m , the second-order momentum of the mth iteration Constant σ.

5. An inverter voltage control system based on the inverter voltage control method according to any one of claims 1 to 4, characterized in that: include: Sampling module: samples the voltage and current data of the LC inverter; Processing module: used for inputting sampling data into a pre-optimized voltage loop controller to control the voltage of the LC inverter in the power system; wherein the pre-optimized voltage loop controller is: based on a single neuron PI controller, an RBF neural network identification algorithm is introduced to identify the Jacobian information of the controlled object, and an AMSGrad optimization algorithm is used to improve the single neuron PI controller.

6. A computer device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images