Multi-frequency point non-integer period disturbance signal generation and control method and system

Abstract

According to the multi-frequency-point non-integer periodic disturbance signal generation and control method and system, reference frequencies are generated based on a multi-layer neural network, and a candidate frequency set is formed by the reference frequencies and integer multiples of the reference frequencies; the perceptron distributes the candidate frequency of which the state coefficient is 1 as the disturbance frequency; determining an amplitude coefficient in an amplitude range by adopting a single-layer neural network; grouping the disturbance frequencies according to the reference frequency; taking uniform distribution of the disturbance frequency on the logarithmic scale, minimum capacity required for injecting the signal at the disturbance frequency and uniform distribution of the disturbance frequency in each group as optimization objectives to obtain multiple groups of disturbance signals; determining dq-axis reference values of each group of disturbance signals injected into the power grid by adopting a reinforcement learning method by taking maximization of the direct-current voltage utilization rate and minimization of the capacity of disturbance injection equipment as targets; a discrete controller of disturbance signals is formed; the discrete controller is integrated in a control system of the disturbance injection device. Therefore, the capacity required by the injection disturbance is smaller, the DC voltage utilization rate is higher, and the disturbance signal is accurately controlled.

Description

Technical Field

[0001] This invention belongs to the field of power electronic power system impedance measurement technology, specifically, it relates to a method and system for generating and controlling multi-frequency non-integer periodic disturbance signals based on a logarithmic scale. Background Technology

[0002] With the acceleration of the global energy transition, the surge in new energy sources and their interface converters has led to a high proportion of new energy and power electronic equipment in power systems, posing a series of technical challenges and difficulties for the safe and stable operation of power systems. The interactions between converters and between converters and the grid are complex, making system stability issues increasingly prominent. Furthermore, new energy converters are often connected to weak grids, further increasing the risk of system instability. However, traditional impedance modeling methods are difficult to effectively apply to controlling complex and "black box" new energy equipment, making system stability analysis difficult. Therefore, researching efficient and accurate impedance measurement techniques has become one of the key factors in improving the safe and stable operation of power systems.

[0003] In existing technologies, impedance measurement of power electronic power systems is mainly achieved by injecting disturbance signals into the system under test. This requires an additional measuring device, thus increasing system costs. Another disturbance injection method is to multiplex existing devices in the system, i.e., by reusing the functions of existing devices. However, this requires the multiplexed device to have a large redundancy capacity; otherwise, it will lead to various problems such as insufficient impedance measurement resolution and low accuracy. During the measurement process, the measuring device can inject a disturbance signal at a single measurement frequency or inject disturbance signals at multiple frequencies simultaneously. However, the single-frequency disturbance signal injection method requires a long measurement time, while the multi-frequency disturbance signal injection method is complex to control. In addition, when traditional integer-cycle disturbance signals are applied to impedance measurement, on the one hand, the disturbance signal is prone to interacting with harmonics in the system, thus affecting the detection accuracy; on the other hand, to obtain sufficient measurement signal strength, a large capacity of disturbance signal needs to be injected. This not only increases the capacity requirements and cost of the equipment but also brings additional power quality problems to the power grid, which in turn adversely affects the normal operation of other electrical equipment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for generating and controlling multi-frequency non-integer periodic disturbance signals based on a logarithmic scale. It addresses issues such as high injection capacity requirements, poor power quality, and insufficient impedance frequency resolution in multiplexed impedance measurement devices by offering a novel disturbance signal generation scheme. This scheme reduces the required injection capacity, increases the DC voltage utilization of the multiplexed impedance measurement device, and improves the resolution and accuracy of impedance measurements. Furthermore, through appropriate control strategies, the designed disturbance signal can be precisely injected into the system.

[0005] The present invention adopts the following technical solution.

[0006] This invention proposes a method for generating and controlling multi-frequency non-integer periodic disturbance signals, comprising:

[0007] N initial frequencies are obtained within the frequency range of impedance measurement. Based on a multi-layer neural network, N reference frequencies are generated. A candidate frequency set is constructed using the reference frequencies and their integer multiples. A perceptron is used to assign state coefficients to all candidate frequencies, and the candidate frequencies with a state coefficient of 1 are used as perturbation frequencies.

[0008] Determine the amplitude range of the signal at the disturbance frequency according to the frequency band where the disturbance frequency is located; use a single-layer neural network to determine the amplitude coefficient within the amplitude range; the square of the amplitude coefficient is the required capacity of the signal at the disturbance frequency;

[0009] The disturbance frequencies are grouped according to the reference frequency corresponding to the disturbance frequency;

[0010] With the optimization objectives of ensuring that all disturbance frequencies are uniformly distributed on a logarithmic scale, minimizing the capacity required to inject signals at all disturbance frequencies, and ensuring that all disturbance frequencies are uniformly distributed within each group, the parameters of multilayer neural networks, perceptrons, and single-layer neural networks are optimized to obtain multiple groups of disturbance signals to be injected into the power grid.

[0011] With the goal of maximizing the DC voltage utilization and minimizing the capacity of the disturbance injection equipment, a reinforcement learning method is used to determine the dq axis reference values ​​of each set of disturbance signals injected into the power grid.

[0012] By using the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq axis reference values ​​of each group of disturbance signals, multiple delay elements are established using an interpolation method; the sum of multiple delay elements constitutes a discrete controller for the disturbance signal; the discrete controller is integrated into the control system of the disturbance injection device.

[0013] In the frequency range of wideband impedance measurement [f min ,f max N frequencies that are uniformly distributed on a logarithmic scale and have non-integer periods are selected as initial frequencies, satisfying the following relationship:

[0014]

[0015] In the formula, f k Let f' be the k-th initial frequency, k = 1, 2, ..., N, where N is the total number of initial frequencies and N is a positive integer. min f max These are the lower and upper bounds of the frequency range, respectively.

[0016] Multilayer neural networks include: hierarchical recurrent neural networks; N initial frequencies f1′, f2′, ..., f N N reference frequencies f1, f2, ..., f are obtained through a hierarchical recurrent neural network. N N reference frequencies in the frequency range of wideband impedance measurement [f min ,f max Within, it is uniformly distributed on a logarithmic scale and has a non-integer period;

[0017] The set of candidate frequencies, {f}, is constructed using the k-th reference frequency and integer multiples of the k-th reference frequency. k ,2f k ,3f k ,…,Mf k k = 1, 2, ..., N, where M is a positive integer less than 10.

[0018] A perceptron with a step activation function is used to assign state coefficients to all candidate frequencies. The state coefficients are either 0 or 1, and candidate frequencies with a state coefficient of 0 are discarded.

[0019] When the disturbance frequency is in the low-frequency band, the amplitude range of the disturbance signal is determined based on the amplitude of the fundamental signal of the power grid; when the disturbance frequency is in the high-frequency band, the amplitude range of the disturbance signal in the high-frequency band is determined by using the exponential decay method based on the amplitude range of the disturbance signal in the low-frequency band; where the low-frequency band is [10Hz, 100Hz] and the high-frequency band is (100Hz, 1000Hz).

[0020] The upper limit of the amplitude of the disturbance signal in the low-frequency band is 10% of the amplitude of the fundamental frequency signal of the power grid, and the lower limit of the amplitude of the disturbance signal in the low-frequency band is 1% of the amplitude of the fundamental frequency signal of the power grid. The upper limit of the amplitude of the disturbance signal in the high-frequency band is W times the upper limit of the amplitude of the disturbance signal in the low-frequency band, and the lower limit of the amplitude of the disturbance signal in the high-frequency band is W times the lower limit of the amplitude of the disturbance signal in the low-frequency band, where the multiple W satisfies the following relationship:

[0021]

[0022] In the formula, f is the candidate frequency in the high-frequency band, and α is the attenuation coefficient.

[0023] The input data of a single-layer neural network is a random number within the amplitude range, and the output data is the amplitude coefficient within the amplitude range.

[0024] When classifying perturbation frequencies into groups, each perturbation frequency corresponds to a reference frequency. Therefore, the perturbation frequencies are divided into N groups, f k_j Let n(k) be the j-th perturbation frequency in the k-th group, where k = 1, 2, ..., N, j = 1, 2, ..., n(k), and n(k) is the number of perturbation frequencies in the k-th group. The perturbation frequency in the k-th group corresponds to the k-th reference frequency.

[0025] The optimization objective F satisfies the following relationship:

[0026] F = min((1-U) 2 +P+σ 2 )

[0027] In the formula, U represents the uniformity of the logarithmic distribution of all perturbation frequencies, P represents the required capacity to inject signals at all perturbation frequencies, and σ is the sum of the squares of the amplitude coefficients of the signals at all perturbation frequencies; 2 Let be the variance of an N-order array, where the N-order array is {n(1),n(2),…,n(N)};

[0028] When U=1, all perturbation frequencies are uniformly distributed on a logarithmic scale; when the sum of the squares of the amplitude coefficients of the signals at all perturbation frequencies is minimized, the required capacity for injecting signals at all perturbation frequencies is minimized; when the variance of the N-order array is zero, all perturbation frequencies are uniformly distributed within each group; therefore, when the sum of the three terms is minimized, the optimization objective reaches its optimal value.

[0029] The uniformity of the logarithmic distribution of all perturbation frequencies can be calculated using the following formula:

[0030]

[0031] In the formula, n represents the number of all disturbance frequencies. n(k) represents the number of perturbation frequencies in the k-th group, N represents the number of groups of perturbation frequencies, and f i Let i be the frequency of the perturbation.

[0032] With the goal of achieving optimal logarithmic distribution uniformity of 1, the hierarchical recurrent network and perceptron are jointly trained to obtain a well-trained hierarchical recurrent network and perceptron.

[0033] During impedance measurement, the same set of multi-frequency disturbance signals is injected into the power grid. The d-axis reference value of the k-th set of disturbance signals is taken from the first phase angle. The q-axis reference value of the k-th disturbance signal adopts the second phase angle. And it satisfies:

[0034]

[0035] Based on the k-th group of multi-frequency non-integer period disturbance signal I p,k_j cos(2πf k_j Using t) as the reference signal, the k-th set of disturbance signal dq-axis reference values ​​are generated using the first phase angle and the second phase angle, as follows:

[0036]

[0037] In the formula, These are the reference values ​​for the dq axis of the k-th group of disturbance current signals, I. p,k_j f is the amplitude coefficient of the j-th disturbance current signal in the k-th group. k_j Let n(k) be the j-th perturbation frequency in the k-th group, where k = 1, 2, ..., N, j = 1, 2, ..., n(k), and n(k) is the number of perturbation frequencies in the k-th group.

[0038] The state variables are constructed using the first phase angle and the dq-axis reference values ​​of each set of disturbance signals, with the adjustment amount of the first phase angle as the variable. As the action variable, a reward function is established to maximize the DC voltage utilization and minimize the capacity of the disturbance injection device. A deep reinforcement learning method based on the near-end policy optimization algorithm is used to determine the optimized first phase angle; and the optimized second phase angle is determined based on the optimized first phase angle.

[0039] The reward function R is:

[0040]

[0041] In the formula, These are the reference values ​​for the dq axes of the k-th group of disturbance current signals, where λ is the penalty coefficient and V is the reference value for the dq axis. d and V q The steady-state voltage dq-axis component of the device is used to inject disturbances.

[0042] Based on the k-th group of disturbance signals, the d-axis reference value of the k-th group of disturbance signals is determined using the optimized first phase angle, and the q-axis reference value of the k-th group of disturbance signals is determined using the optimized second phase angle, as follows:

[0043]

[0044] In the formula, These are the dq-axis reference values ​​for the injected k-th group of disturbance signals, respectively. These are the optimized first phase angle and the optimized second phase angle for the k-th group, respectively.

[0045] The transfer function H of a discrete controller d (z) is:

[0046]

[0047] In the formula, N i G represents the integer part of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of disturbance signals, while G represents the decimal part of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of disturbance signals. For N iThe delay is of order, where a0, a1, a2, a3 and b0, b1, b2, b3 are the coefficients of the delay element.

[0048] When a discrete controller is integrated into the forward channel of the control system of a disturbance injection device, the transmission data in the forward channel of the control system is input to the discrete controller after passing through a gain coefficient. The output data of the discrete controller is transmitted after passing through a low-pass filter and is accumulated with the transmission data in the forward channel of the control system to obtain an optimized output. At the same time, it is fed back to the input of the discrete controller and accumulated with the transmission data after passing through a gain coefficient to update the input data of the discrete controller.

[0049] This invention also proposes a multi-frequency point non-integer period disturbance signal generation and control system, comprising:

[0050] The disturbance signal generation module is used to acquire N initial frequencies within the frequency range of impedance measurement, generate N reference frequencies based on a multilayer neural network, form a candidate frequency set using the reference frequencies and their integer multiples, assign state coefficients to all candidate frequencies using a perceptron, and use the candidate frequency with a state coefficient of 1 as the disturbance frequency, determine the amplitude range of the signal at the disturbance frequency according to the frequency band where the disturbance frequency is located, determine the amplitude coefficient within the amplitude range using a single-layer neural network, and the square of the amplitude coefficient is the required capacity of the signal at the disturbance frequency, group the disturbance frequencies according to the reference frequencies corresponding to the disturbance frequencies, and optimize the parameters of the multilayer neural network, perceptron, and single-layer neural network with the optimization objectives of uniform distribution of all disturbance frequencies on a logarithmic scale, minimizing the required capacity of the signal injected at all disturbance frequencies, and uniform distribution of all disturbance frequencies within each group, so as to obtain multiple sets of disturbance signals to be injected into the power grid.

[0051] The disturbance signal control module is used to determine the dq-axis reference values ​​of each set of disturbance signals injected into the power grid using reinforcement learning, with the goal of maximizing the DC voltage utilization and minimizing the capacity of the disturbance injection device. Multiple delay elements are established using an interpolation method based on the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each set of disturbance signals. The discrete controller for the disturbance signal is constructed by summing the multiple delay elements and is integrated into the control system of the disturbance injection device.

[0052] The present invention is also a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0053] The present invention is also a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0054] The beneficial effects of this invention, compared with the prior art, include at least the following: To reduce the capacity required for injecting disturbance signals using a multiplexing device, this invention proposes a method for generating and controlling minimal disturbance signals at multiple frequency points with non-integer period frequencies on a logarithmic scale. This method reduces the capacity required for injecting disturbance signals through frequency point optimization, amplitude optimization calculation, and phase optimization, thereby improving system power quality while ensuring measurement accuracy. Phase optimization improves the DC utilization rate of the disturbance injection device, further ensuring high-precision measurement with minimal redundancy. The uniform logarithmic distribution of frequency points coordinates the frequency of the disturbance signal with the disturbance signal control strategy in the multiplexing device, improving the accuracy of disturbance signal injection and thus enhancing the resolution and accuracy of impedance measurement. The disturbance signal control strategy employs a generalized interpolation approximation method for non-integer period signals to achieve precise control. Attached Figure Description

[0055] Figure 1 This is a flowchart of a method for generating and controlling multi-frequency non-integer periodic disturbance signals proposed in this invention;

[0056] Figure 2 This is a partial block diagram of the control strategy in the forward channel of the disturbance injection device control system, which integrates the discrete controller of the multi-frequency non-integer period disturbance signal into the system. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0058] Active impedance measurement requires injecting a disturbance signal into the system and calculating the system impedance based on the response signal. While disturbance signal injection can be achieved using an additional disturbance injection device, this incurs high measurement costs. Multiplexing existing devices within the system for disturbance injection requires significant redundancy; otherwise, it leads to poor impedance measurement resolution and low accuracy. Therefore, this invention proposes a method for generating and controlling multi-frequency non-integer period disturbance signals based on a logarithmic scale, such as… Figure 1 As shown, it includes:

[0059] Step 1: Obtain N initial frequencies within the frequency range of impedance measurement. Based on a multi-layer neural network, generate N reference frequencies for the disturbance signal using the initial frequencies. Construct a candidate frequency set using the reference frequencies and their integer multiples. Assign state coefficients to all candidate frequencies using a perceptron.

[0060] Specifically, step 1 includes:

[0061] Step 1.1, within the frequency range of wideband impedance measurement [f] min ,f max N frequencies that are uniformly distributed on a logarithmic scale and have non-integer periods are selected as initial frequencies, satisfying the following relationship:

[0062]

[0063] In the formula, f k Let f' be the k-th initial frequency, k = 1, 2, ..., N, where N is the total number of initial frequencies and N is a positive integer. min f max These are the lower and upper bounds of the frequency range, respectively.

[0064] In this embodiment, within the frequency range [10Hz, 1000Hz], the total number of frequencies N is 11. The frequencies calculated uniformly on a logarithmic scale and obtained through a hierarchical recurrent neural network include frequencies with non-integer periods, such as 12.6Hz, 15.8Hz, etc. The reference frequency of the disturbance signal is selected as a frequency with a non-integer period to significantly distinguish the disturbance signal from power grid harmonics. Furthermore, the disturbance frequency is uniformly distributed on a logarithmic scale, which makes the impedance spectrum balanced and ensures that the number of frequency points in the impedance spectrum is consistent within each tenth harmonic range, thereby improving the comprehensiveness of broadband impedance measurement.

[0065] In the embodiment, the selected N initial frequencies f1′, f2′, ..., f N The N reference frequencies f1, f2, ..., f are obtained through iterative updates using a hierarchical recurrent neural network. N N initial frequencies are selected according to the above formula, and N reference frequencies are within the frequency range of wideband impedance measurement [f]. min ,f max The frequencies are uniformly distributed on a logarithmic scale and have non-integer periods, making the hierarchical recurrent neural network (RNN) more likely to converge. Furthermore, the hierarchical RNN ensures that the selected reference frequencies are both uniformly distributed on a logarithmic scale and limited to the frequency range of wideband impedance measurements. The N frequencies updated iteratively by the hierarchical RNN are used as the reference frequencies for the perturbation signal. It is worth noting that the use of a hierarchical RNN in this invention is a non-limiting but preferred choice; those skilled in the art can choose different neural networks according to actual needs.

[0066] Step 1.2, the set of the kth candidate frequencies, consisting of the kth reference frequency and integer multiples of the kth reference frequency, is {f}. k ,2f k ,3f k ,…,Mf kk = 1, 2, ..., N, where M is a positive integer less than 10;

[0067] In this embodiment, frequencies that satisfy the positive integer multiple relationship of the reference frequency are divided into the same candidate frequency set, and the reference frequency is used as the first element of the candidate frequency set. Optimization is performed on the candidate frequency set to achieve the best control effect.

[0068] Step 1.3: Using a perceptron with a step activation function, assign state coefficients to all candidate frequencies. The state coefficients are either 0 or 1. Discard all candidate frequencies with a state coefficient of 0 and use all candidate frequencies with a state coefficient of 1 as perturbation frequencies.

[0069] The previous section generated multiple candidate frequencies. Here, a perceptron is used to generate state coefficients for each candidate frequency. In the subsequent selection of perturbation frequencies based on the objective function, candidate frequencies with a state coefficient of 1 will be preferentially selected as perturbation frequencies, while a state coefficient of 0 indicates that the candidate frequency is discarded.

[0070] In this embodiment, the present invention introduces a perceptron to assign state coefficients to all candidate frequencies. In subsequent steps, the hierarchical recurrent neural network, the perceptron, and the single-layer neural network are jointly trained according to the objective function. The parameters of the perceptron, the hierarchical recurrent neural network, and the single-layer neural network are optimized together, so that the selected candidate frequencies are evenly distributed on the logarithmic scale and there is a multiple relationship between the frequencies.

[0071] Step 2: Determine the amplitude range of the signal at the disturbance frequency according to the frequency band where the disturbance frequency is located; use a single-layer neural network to determine the amplitude coefficient within the amplitude range; the square of the amplitude coefficient is the required capacity of the signal at the disturbance frequency.

[0072] Specifically, step 2 includes:

[0073] Step 2.1: When the disturbance frequency is in the low-frequency band, the amplitude range of the disturbance signal is determined based on the amplitude of the fundamental signal of the power grid; when the disturbance frequency is in the high-frequency band, the amplitude range of the disturbance signal in the high-frequency band is determined by using the exponential decay method based on the amplitude range of the disturbance signal in the low-frequency band; wherein, the low-frequency band is [10Hz, 100Hz], and the high-frequency band is (100Hz, 1000Hz).

[0074] In this embodiment, the amplitude selection range is set considering that the disturbance signal should not affect the steady-state operating point of the system, and the response signal should be accurately identified. Simultaneously, the amplitude of low-frequency disturbances is set higher, while the amplitude of high-frequency disturbances gradually decreases. Therefore, the upper limit of the amplitude of the disturbance signal in the low-frequency band is defined. The amplitude of the fundamental frequency signal A0 of the power grid is 10%, and the lower limit of the amplitude of disturbance signals in the low-frequency band is... It is 1% of the amplitude A0 of the fundamental frequency signal of the power grid; the upper limit of the amplitude of the disturbance signal in the high-frequency band. The upper limit of the amplitude of the disturbance signal in the low frequency band. W times, the lower limit of the amplitude of the disturbance signal in the high-frequency band. The lower limit of the amplitude of the disturbance signal in the low frequency band W times, where W times satisfies the following relationship:

[0075]

[0076] In the formula, f is the candidate frequency in the high-frequency band, and α is the attenuation coefficient, which is 0.5 in the embodiment.

[0077] In the high-frequency range, the upper and lower limits of the amplitude of the disturbance signal are related to the candidate frequency.

[0078] Step 2.2: Use a single-layer neural network to assign amplitude coefficients to all amplitude ranges.

[0079] Specifically, the input data of a single-layer neural network is a random number within the amplitude range, and the output data is the amplitude coefficient within the amplitude range;

[0080] Since this invention aims to optimize the capacity required for injecting the perturbation signal, amplitude coefficients are assigned to the amplitude range. In subsequent steps, the hierarchical recurrent neural network, perceptron, and single-layer neural network are jointly trained according to the objective function. The parameters of the perceptron, the hierarchical recurrent neural network, and the single-layer neural network are optimized together. Finally, by selecting a suitable amplitude, the capacity required for injecting the perturbation signal is reduced.

[0081] Step 3: Group the disturbance frequencies according to the reference frequencies corresponding to the disturbance frequencies;

[0082] When classifying the disturbance frequencies, each disturbance frequency corresponds to a reference frequency. Therefore, the disturbance frequencies are divided into N groups, and the number of disturbance frequencies in groups 1 to N are represented by n(1), n(2)...n(N), respectively. The disturbance frequencies in each group satisfy the following relationship:

[0083] Group 1: f 1_1 ,f 1_2 ,…f 1_n(1)

[0084] Group 2: f 2_1 ,f 2_2 ,…f 2_n(2)

[0085] ······

[0086] Group N: f N_1 ,f N_2 ,…f N_n(N)

[0087] Among them, f k_j Let n(k) be the j-th perturbation frequency in the k-th group, where k = 1, 2, ..., N, j = 1, 2, ..., n(k), N is the number of groups of perturbation frequencies, and n(k) is the number of perturbation frequencies in the k-th group. The perturbation frequency in the k-th group corresponds to the k-th reference frequency.

[0088] Since impedance measurement requires injecting disturbance signals into the power grid multiple times, the disturbance signals need to be grouped. After grouping, each group has a disturbance frequency with multiple non-integer periods, thereby improving the efficiency of impedance measurement and making it easier to select the group containing the target disturbance frequency, thus achieving accurate injection of the disturbance signal.

[0089] Step 4: With the optimization objectives of uniform distribution of all disturbance frequencies on a logarithmic scale, minimum capacity required to inject signals at all disturbance frequencies, and uniform distribution of all disturbance frequencies within each group, the parameters of the multilayer neural network, the perceptron, and the single-layer neural network are optimized to obtain multiple groups of disturbance signals to be injected into the power grid.

[0090] The optimization objective F satisfies the following relationship:

[0091] F = min((1-U) 2 +P+σ 2 )

[0092] In the formula, U represents the uniformity of the logarithmic distribution of all perturbation frequencies, P represents the required capacity to inject signals at all perturbation frequencies, and σ is the sum of the squares of the amplitude coefficients of the signals at all perturbation frequencies; 2 Let be the variance of an N-order array, where the N-order array is {n(1),n(2),…,n(N)};

[0093] When U=1, all perturbation frequencies are uniformly distributed on a logarithmic scale; when the sum of the squares of the amplitude coefficients of the signals at all perturbation frequencies is minimized, the required capacity for injecting signals at all perturbation frequencies is minimized; when the variance of the N-order array is zero, all perturbation frequencies are uniformly distributed within each group; therefore, when the sum of the three terms is minimized, the optimization objective reaches its optimal value.

[0094] The uniformity of the logarithmic distribution of all disturbance frequencies is calculated using the following formula:

[0095]

[0096] In the formula, n represents the number of all disturbance frequencies. f i Let i be the frequency of the perturbation.

[0097] σ is considered in the optimization objective F 2This ensures that when disturbance signals are injected into the power grid in multiple stages, the disturbance frequency is evenly distributed across the groups of disturbance signals injected in each stage.

[0098] Based on the optimization objective, gradient descent is used to train a multilayer neural network, a perceptron, and a single-layer neural network. During training, the weights, state coefficients, and amplitude coefficients of neurons are gradually optimized. Using the trained multilayer neural network, perceptron, and single-layer neural network, the perturbation frequency and amplitude of the perturbation signal are determined. Specifically, the trained multilayer neural network outputs optimized candidate frequencies, and the trained perceptron assigns optimal state coefficients to all optimized candidate frequencies. The optimized candidate frequency with an optimal state coefficient of 1 is taken as the perturbation frequency. According to the frequency band of the perturbation frequency, the amplitude range of the signal at the perturbation frequency is determined. The trained single-layer neural network assigns optimal amplitude coefficients to all amplitude ranges. The optimal amplitude coefficient is taken as the amplitude of the signal at the perturbation frequency, ultimately obtaining a multi-frequency non-integer periodic perturbation signal. Steps 1 to 4 above implement a method for generating multi-frequency non-integer periodic perturbation signals. The obtained multi-frequency non-integer periodic perturbation signals are still divided into N groups.

[0099] Step 5: With the goal of maximizing the DC voltage utilization and minimizing the capacity of the disturbance injection device, a reinforcement learning method is used to determine the dq axis reference values ​​of each set of disturbance signals injected into the power grid.

[0100] Specifically, step 5 includes:

[0101] Step 5.1: During impedance measurement, inject the same set of multi-frequency disturbance signals into the power grid. The d-axis reference value of the k-th set of disturbance signals adopts the first phase angle. The q-axis reference value of the k-th disturbance signal adopts the second phase angle. And it satisfies:

[0102]

[0103] In this embodiment, the initial value of the first phase angle is set based on engineering experience, and the initial value of the second phase angle is set based on the first phase angle according to the above-mentioned relationship.

[0104] Step 5.2: Based on the multi-frequency non-integer period disturbance signal, generate the dq axis reference value of the disturbance signal using the first phase angle and the second phase angle;

[0105] Specifically, the k-th group of multi-frequency non-integer period disturbance signals I generated from steps 1 to 4 p,k_j cos(2πf k_j Using t) as the reference signal, the k-th set of disturbance signal dq-axis reference values ​​are generated using the first phase angle and the second phase angle, as follows:

[0106]

[0107] In the formula, These are the reference values ​​for the dq axis of the k-th group of disturbance current signals, I. p,k_j f is the amplitude coefficient of the j-th disturbance current signal in the k-th group. k_j Let n(k) be the j-th perturbation frequency in the k-th group, where k = 1, 2, ..., N, j = 1, 2, ..., n(k), and n(k) is the number of perturbation frequencies in the k-th group.

[0108] Step 5.3: Using the first phase angle and the dq-axis reference values ​​of each group of disturbance signals to construct state variables, and using the adjustment amount of the first phase angle... As the action variable, a reward function is established to maximize the DC voltage utilization and minimize the capacity of the disturbance injection device. A deep reinforcement learning method based on the near-end policy optimization algorithm is used to determine the optimized first phase angle; and the optimized second phase angle is determined based on the optimized first phase angle.

[0109] Specifically, the reward function R is:

[0110]

[0111] In the formula, Here, V represents the reference value of the dq axis for the k-th group of disturbance current signals, λ is the penalty coefficient used to reward the reduction of the capacity required to inject the disturbance signal, and V d and V q The steady-state voltage of the disturbance injection device is represented by the dq-axis component.

[0112] In the reward function, Since the DC voltage utilization rate of the disturbance injection device is inversely proportional to the DC voltage utilization rate, the training action variables maximize the reward function to obtain state variables that can improve DC voltage utilization and reduce device capacity.

[0113] Step 5.4: Based on the k-th group of disturbance signals, determine the d-axis reference value of the k-th group of disturbance signals using the optimized first phase angle, and determine the q-axis reference value of the k-th group of disturbance signals using the optimized second phase angle, as shown below:

[0114]

[0115] In the formula, These are the dq axis reference values ​​for the k-th group of disturbance signals, respectively. These are the optimized first phase angle and the optimized second phase angle for the k-th group, respectively.

[0116] In this embodiment, to improve training efficiency, the disturbance design unit uses a three-layer fully connected network to construct a peak prediction neural network. Based on the peak prediction neural network, the reward function value is directly output during the deep reinforcement learning training process. When the algorithm converges, the optimized first phase angle corresponding to each group of disturbance signals is obtained. Using the optimized first phase angle and second phase angle, the d-axis reference value and q-axis reference value of the disturbance signal injected into the power grid are determined respectively. The phase angle optimization method proposed in this invention reduces the peak value of the modulation wave of the synthesized disturbance signal, thereby improving the DC voltage utilization rate of the disturbance injection equipment and reducing the capacity required for disturbance signal injection. It significantly reduces the redundant capacity configuration of existing equipment when reusing existing equipment to inject disturbance signals. It can improve the DC voltage utilization rate and reduce the equipment capacity while ensuring that the disturbance signal is detectable, minimize its interference with the normal operation of the converter, and improve the accuracy and reliability of output impedance measurement.

[0117] Step 6: Using the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq axis reference values ​​of each group of disturbance signals, multiple delay elements are established using an interpolation method; the sum of the multiple delay elements constitutes a discrete controller for the disturbance signal; the discrete controller is integrated into the control system of the disturbance injection device.

[0118] To ensure the efficiency of impedance measurement, if multiple disturbance signals are injected simultaneously, a corresponding controller needs to be designed for each frequency, resulting in complex controller structures and difficult parameter design. Therefore, this invention proposes a method for designing controller parameters based on the disturbance frequency and Nyquist frequency to directly control multiple frequency disturbance signals. However, the control strategy has requirements on the frequency of the disturbance signals, specifically requiring a multiple relationship between the disturbance frequencies. Integrating a simplified controller into an existing device requires minimizing the capacity required for disturbance signal injection. Steps 1 to 5 have already achieved the multiple relationship between the disturbance frequencies and minimized the required capacity for the injected signal. Therefore, based on the disturbance signals obtained in steps 1 to 5, the special discrete controller proposed in this invention can control multi-frequency non-integer period disturbance signals. Furthermore, the discrete controller is integrated into the forward channel of the disturbance injection device's control system, thereby enabling the generation and control of multi-frequency non-integer period disturbance signals during the disturbance injection device control process.

[0119] In the embodiment, the transfer function H of the discrete controller d (z) is:

[0120]

[0121] In the formula, N iG represents the integer part of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of disturbance signals, while G represents the decimal part of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of disturbance signals. For N i The delay is of order, where a0, a1, a2, a3 and b0, b1, b2, b3 are the coefficients of the delay element;

[0122] The order of each delay element in the discrete controller proposed in this invention is directly related to the multi-frequency signal and is obtained based on the ratio between the Nyquist frequency and the injection frequency. Therefore, it can control multi-frequency disturbance signals simultaneously.

[0123] In this embodiment, the coefficients of the delay element are obtained by fitting using interpolation methods, including but not limited to spline interpolation, Runge-Kutta interpolation, and Lagrange interpolation.

[0124] like Figure 2 As shown, the integer and fractional parts of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of injected disturbance signals are used to establish the discrete controller H. d (z); When the discrete controller is integrated into the forward channel of the control system of the disturbance injection device, the transmitted data in the forward channel of the control system is transmitted through the gain coefficient k. r The data is then input to the discrete controller, and the output data of the discrete controller is filtered by a low-pass filter L. PF (z) The data is then transmitted and accumulated with the data transmitted in the forward channel of the control system to obtain the optimized output, which is simultaneously fed back to the input of the discrete controller. This optimized output is then combined with the transmitted data through a gain coefficient k. r The data is then accumulated to update the input data of the discrete controller. Through the above signal control, multi-frequency disturbance signals can be controlled simultaneously, supporting the requirements of multiple disturbance frequencies and minimizing the capacity of disturbance injection equipment. The disturbance signal control strategy adopts a generalized interpolation approximation method for non-integer period signals to achieve accurate and stable control of disturbance signals.

[0125] This invention also proposes a multi-frequency point non-integer period disturbance signal generation and control system, comprising:

[0126] The disturbance signal generation module is used to acquire N initial frequencies within the frequency range of impedance measurement, generate N reference frequencies based on a multilayer neural network, form a candidate frequency set using the reference frequencies and their integer multiples, assign state coefficients to all candidate frequencies using a perceptron, and use the candidate frequency with a state coefficient of 1 as the disturbance frequency, determine the amplitude range of the signal at the disturbance frequency according to the frequency band where the disturbance frequency is located, determine the amplitude coefficient within the amplitude range using a single-layer neural network, and the square of the amplitude coefficient is the required capacity of the signal at the disturbance frequency, group the disturbance frequencies according to the reference frequencies corresponding to the disturbance frequencies, and optimize the parameters of the multilayer neural network, perceptron, and single-layer neural network with the optimization objectives of uniform distribution of all disturbance frequencies on a logarithmic scale, minimizing the required capacity of the signal injected at all disturbance frequencies, and uniform distribution of all disturbance frequencies within each group, so as to obtain multiple sets of disturbance signals to be injected into the power grid.

[0127] The disturbance signal control module is used to determine the dq-axis reference values ​​of each set of disturbance signals injected into the power grid using reinforcement learning, with the goal of maximizing the DC voltage utilization and minimizing the capacity of the disturbance injection device. Multiple delay elements are established using an interpolation method based on the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each set of disturbance signals. The discrete controller for the disturbance signal is constructed by summing the multiple delay elements and is integrated into the control system of the disturbance injection device.

[0128] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0129] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0130] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0131] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for generating and controlling multi-frequency non-integer periodic disturbance signals, characterized in that, include: N initial frequencies are obtained within the frequency range of impedance measurement, and N reference frequencies are generated based on a multi-layer neural network. A candidate frequency set is constructed using the reference frequency and its integer multiples; a perceptron is used to assign state coefficients to all candidate frequencies, and the candidate frequency with a state coefficient of 1 is used as the perturbation frequency. Determine the amplitude range of the signal at the disturbance frequency according to the frequency band where the disturbance frequency is located; use a single-layer neural network to determine the amplitude coefficient within the amplitude range; The disturbance frequencies are grouped according to the reference frequency corresponding to the disturbance frequency; With the optimization objectives of ensuring that all perturbation frequencies are uniformly distributed on a logarithmic scale, minimizing the capacity required to inject signals at all perturbation frequencies, and ensuring that all perturbation frequencies are uniformly distributed within each group, the parameters of multilayer neural networks, perceptrons, and single-layer neural networks are optimized to obtain multiple groups of perturbation signals. With the goal of maximizing the DC voltage utilization and minimizing the capacity of the disturbance injection equipment, a reinforcement learning method is used to determine the dq axis reference values ​​of each set of disturbance signals injected into the power grid. By using the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq axis reference values ​​of each group of disturbance signals, multiple delay elements are established using an interpolation method; the discrete controller of the disturbance signal is constructed by accumulating and summing the multiple delay elements. The discrete controller is integrated into the control system of the disturbance injection device.

2. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 1, characterized in that, In the frequency range of wideband impedance measurement [f min ,f max N frequencies that are uniformly distributed on a logarithmic scale and have non-integer periods are selected as initial frequencies, satisfying the following relationship: In the formula, f k Let f' be the k-th initial frequency, k = 1, 2, ..., N, where N is the total number of initial frequencies and N is a positive integer. min f max These represent the lower and upper bounds of the frequency range, respectively.

3. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 2, characterized in that, Multilayer neural networks include: hierarchical recurrent neural networks; N initial frequencies f1′, f2′, ..., f N N reference frequencies f1, f2, ..., f are obtained through a hierarchical recurrent neural network. N N reference frequencies in the frequency range of wideband impedance measurement [f min ,f max Within, it is uniformly distributed on a logarithmic scale and has a non-integer period; The set of candidate frequencies, {f}, is constructed using the k-th reference frequency and integer multiples of the k-th reference frequency. k ,2f k ,3f k ,…,Mf k k = 1, 2, ..., N, where M is a positive integer less than 10.

4. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 3, characterized in that, A perceptron with a step activation function is used to assign state coefficients to all candidate frequencies. The state coefficients are either 0 or 1, and candidate frequencies with a state coefficient of 0 are discarded.

5. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 4, characterized in that, When the disturbance frequency is in the low-frequency band, the amplitude range of the disturbance signal is determined based on the amplitude of the fundamental signal of the power grid; when the disturbance frequency is in the high-frequency band, the amplitude range of the disturbance signal in the high-frequency band is determined by using the exponential decay method based on the amplitude range of the disturbance signal in the low-frequency band; where the low-frequency band is [10Hz, 100Hz] and the high-frequency band is (100Hz, 1000Hz).

6. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 5, characterized in that, The upper limit of the amplitude of the disturbance signal in the low-frequency band is 10% of the amplitude of the fundamental frequency signal of the power grid, and the lower limit of the amplitude of the disturbance signal in the low-frequency band is 1% of the amplitude of the fundamental frequency signal of the power grid. The upper limit of the amplitude of the disturbance signal in the high-frequency band is W times the upper limit of the amplitude of the disturbance signal in the low-frequency band, and the lower limit of the amplitude of the disturbance signal in the high-frequency band is W times the lower limit of the amplitude of the disturbance signal in the low-frequency band, where the multiple W satisfies the following relationship: In the formula, f is the candidate frequency in the high-frequency band, and α is the attenuation coefficient.

7. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 6, characterized in that, The input data of a single-layer neural network is a random number within the amplitude range, and the output data is the amplitude coefficient within the amplitude range.

8. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 1, characterized in that, When classifying perturbation frequencies into groups, each perturbation frequency corresponds to a reference frequency. Therefore, the perturbation frequencies are divided into N groups, f k_j Let n(k) be the j-th perturbation frequency in the k-th group, where k = 1, 2, ..., N, j = 1, 2, ..., n(k), and n(k) is the number of perturbation frequencies in the k-th group. The perturbation frequency in the k-th group corresponds to the k-th reference frequency.

9. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 1, characterized in that, The optimization objective F satisfies the following relationship: F=min((1-U) 2 +P+s 2 ) In the formula, U represents the uniformity of the logarithmic distribution of all perturbation frequencies, P represents the required capacity to inject signals at all perturbation frequencies, and σ is the sum of the squares of the amplitude coefficients of the signals at all perturbation frequencies; 2 Let be the variance of an N-order array, where the N-order array is {n(1),n(2),…,n(N)}; When U=1, all perturbation frequencies are uniformly distributed on a logarithmic scale; when the sum of the squares of the amplitude coefficients of the signals at all perturbation frequencies is minimized, the required capacity to inject signals at all perturbation frequencies is minimized. When the variance of the N-order array is zero, all perturbation frequencies are uniformly distributed within each group; therefore, the optimization objective reaches its optimal value when the sum of the three terms is minimized.

10. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 9, characterized in that, The uniformity of the logarithmic distribution of all perturbation frequencies can be calculated using the following formula: In the formula, n represents the number of all disturbance frequencies. n(k) represents the number of perturbation frequencies in the k-th group, N represents the number of groups of perturbation frequencies, and f i Let i be the frequency of the perturbation. With the goal of achieving optimal logarithmic distribution uniformity of 1, the hierarchical recurrent network and perceptron are jointly trained to obtain a well-trained hierarchical recurrent network and perceptron.

11. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 1, characterized in that, During impedance measurement, the same set of multi-frequency disturbance signals is injected into the power grid. The d-axis reference value of the k-th set of disturbance signals is taken from the first phase angle. The q-axis reference value of the k-th disturbance signal adopts the second phase angle. And it satisfies: Based on the k-th group of multi-frequency non-integer period disturbance signal I p,k_j cos(2πf k_j Using t) as the reference signal, the k-th set of disturbance signal dq-axis reference values ​​are generated using the first phase angle and the second phase angle, as follows: In the formula, These are the reference values ​​for the dq axis of the k-th group of disturbance current signals, I. p,k_j f is the amplitude coefficient of the j-th disturbance current signal in the k-th group. k_j Let n(k) be the j-th perturbation frequency in the k-th group, where k = 1, 2, ..., N, j = 1, 2, ..., n(k), and n(k) is the number of perturbation frequencies in the k-th group.

12. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 11, characterized in that, The state variables are constructed using the first phase angle and the dq-axis reference values ​​of each set of disturbance signals, with the adjustment amount of the first phase angle as the variable. As the action variable, a reward function is established to maximize the DC voltage utilization and minimize the capacity of the disturbance injection device. A deep reinforcement learning method based on the near-end policy optimization algorithm is used to determine the optimized first phase angle; and the optimized second phase angle is determined based on the optimized first phase angle. The reward function R is: In the formula, These are the reference values ​​for the dq axes of the k-th group of disturbance current signals, where λ is the penalty coefficient and V is the reference value for the dq axis. d and V q The steady-state voltage dq-axis component of the device is used to inject disturbances.

13. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 12, characterized in that, Based on the k-th group of disturbance signals, the d-axis reference value of the k-th group of disturbance signals is determined using the optimized first phase angle, and the q-axis reference value of the k-th group of disturbance signals is determined using the optimized second phase angle, as follows: In the formula, These are the dq-axis reference values ​​for the injected k-th group of disturbance signals, respectively. These are the optimized first phase angle and the optimized second phase angle for the k-th group, respectively.

14. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 1, characterized in that, The transfer function H of a discrete controller d (z) is: In the formula, N i G represents the integer part of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of disturbance signals, while G represents the decimal part of the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each group of disturbance signals. For N i The delay is of order, where a0, a1, a2, a3 and b0, b1, b2, b3 are the coefficients of the delay element.

15. The method for generating and controlling multi-frequency non-integer periodic disturbance signals according to claim 14, characterized in that, When a discrete controller is integrated into the forward channel of the control system of a disturbance injection device, the transmission data in the forward channel of the control system is input to the discrete controller after passing through a gain coefficient. The output data of the discrete controller is transmitted after passing through a low-pass filter and is accumulated with the transmission data in the forward channel of the control system to obtain an optimized output. At the same time, it is fed back to the input of the discrete controller and accumulated with the transmission data after passing through a gain coefficient to update the input data of the discrete controller.

16. A multi-frequency point non-integer period disturbance signal generation and control system, used to implement the multi-frequency point non-integer period disturbance signal generation and control method according to any one of claims 1 to 15, characterized in that, include: The disturbance signal generation module is used to obtain N initial frequencies within the frequency range of impedance measurement and generate N reference frequencies based on a multi-layer neural network. A candidate frequency set is constructed using the reference frequency and its integer multiples. A perceptron is used to assign state coefficients to all candidate frequencies, and the candidate frequencies with a state coefficient of 1 are designated as perturbation frequencies. The amplitude range of the signal at the perturbation frequency is determined according to the frequency band in which the perturbation frequency is located. A single-layer neural network is used to determine the amplitude coefficients within the amplitude range. The perturbation frequencies are grouped according to the reference frequency corresponding to the perturbation frequency. The optimization objectives are to ensure that all perturbation frequencies are uniformly distributed on a logarithmic scale, that the capacity required to inject signals at all perturbation frequencies is minimized, and that all perturbation frequencies are uniformly distributed within each group. The parameters of the multilayer neural network, the perceptron, and the single-layer neural network are optimized to obtain multiple groups of perturbation signals. The disturbance signal control module is used to determine the dq-axis reference values ​​of each set of disturbance signals injected into the power grid with the goal of maximizing the DC voltage utilization and minimizing the capacity of the disturbance injection device. It uses reinforcement learning to determine the dq-axis reference values ​​of each set of disturbance signals injected into the power grid. It uses the ratio between the Nyquist frequency of the disturbance injection device and the reference frequency corresponding to the dq-axis reference values ​​of each set of disturbance signals to establish multiple delay elements using an interpolation method. The discrete controller of the disturbance signal is formed by accumulating and summing the multiple delay elements. The discrete controller is integrated into the control system of the disturbance injection device.

17. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-15.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-15.

Images