Harmonic suppression method based on virtual impedance cooperative control

Abstract

The invention discloses a harmonic suppression method based on virtual impedance cooperative control, relates to the technical field of power grid harmonic control, and is used for solving the technical problems of high harmonic time variation, high treatment energy consumption and parameter solidification in a high-permeability photovoltaic scene. The method comprises the following steps: acquiring an original electric signal and an impedance parameter of a power grid side and a filtering parameter of an inverter side; performing real-time phase correction according to the original electric signal to obtain a multi-band harmonic frequency; virtual impedance is constructed based on the multi-band harmonic frequency, and an impedance characteristic matrix is generated based on the impedance parameters and the filtering parameters; performing parameter optimization based on deep reinforcement learning on the virtual impedance by adopting the impedance characteristic matrix to obtain an optimal virtual impedance parameter; and changing the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameter so as to realize harmonic suppression.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid harmonic control, and particularly relates to a harmonic suppression method based on virtual impedance collaborative control, a harmonic suppression device based on virtual impedance collaborative control, an electronic device, and a storage medium. Background Art

[0002] Harmonics refer to the sinusoidal components in a circuit that have frequencies that are integer multiples of the fundamental frequency due to nonlinear loads or switching actions in the power supply or signal. In switched-mode power supplies (such as DC / AC converters (Direct Current to Alternating Current Converter), inverters), due to the high-frequency switching of switching elements (such as MOSFETs (Metal-Oxide-Semiconductor Field-Effect Transistors), IGBTs (Insulated-Gate Bipolar Transistors), etc.), harmonics will be introduced into the output waveform. In nonlinear loads (such as rectifiers, frequency converters, etc.), harmonics will also be generated due to the non-linear changes in current or voltage. In a power system, harmonics may be caused by the non-linear characteristics of transformers, cables, or other devices.

[0003] Harmonics are important parameters of the power quality of the power grid. For the power grid to operate stably, harmonics must be controlled within a certain level. The existence of harmonics will have a negative impact on the performance, efficiency, and stability of the circuit. Therefore, in order to reduce the impact of harmonics, harmonic suppression is a very important part of circuit design.

[0004] Currently, common harmonic suppression methods mainly include passive filtering, active filtering, etc. Passive filtering filters specific-frequency harmonics by adding passive components such as inductors and capacitors in the circuit to form an LC filter or other types of filtering networks. Although this method is easy to implement, its parameters are fixed, the filtering effect is limited, and it needs to be designed for specific frequencies, with poor adaptability, and is not suitable for the case where the harmonics have strong time-varying characteristics in a high-penetration photovoltaic scenario. Active filtering, on the other hand, uses devices such as active power filters (APFs) to detect harmonics in real time and inject equal amounts of reverse harmonics to cancel out the harmonics. Although this method has a good filtering effect and can handle harmonics of multiple frequencies, its cost is high, it requires complex control algorithms (such as APF algorithms), and the energy consumption for harmonic governance is high.

[0005] With the grid connection of high-penetration distributed photovoltaics, the switching of power electronic devices will generate more harmonics. Therefore, it is urgent to propose a more effective harmonic suppression method to solve the three major problems of strong harmonic time-variation, high governance energy consumption, and parameter solidification in high-penetration photovoltaic scenarios. Summary of the Invention

[0006] The present invention provides a harmonic suppression method based on virtual impedance collaborative control, a harmonic suppression device based on virtual impedance collaborative control, an electronic device, and a storage medium, which are used to solve or partially solve the technical problems of strong harmonic time-variation, high governance energy consumption, and parameter solidification in high-penetration photovoltaic scenarios.

[0007] The present invention provides a harmonic suppression method based on virtual impedance collaborative control, including:

[0008] Obtain the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side;

[0009] Perform real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies;

[0010] Construct a virtual impedance based on the multi-band harmonic frequencies, and generate an impedance characteristic matrix based on the impedance parameters and the filtering parameters;

[0011] Use the impedance characteristic matrix to perform parameter optimization of the virtual impedance based on deep reinforcement learning to obtain optimal virtual impedance parameters;

[0012] Change the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression.

[0013] Optionally, the performing real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies includes:

[0014] Decompose the original electrical signal to obtain the non-steady harmonic components;

[0015] Extract the instantaneous frequency according to the non-steady harmonic components, and perform real-time phase error correction on the instantaneous phase of the non-steady harmonic components based on Kalman filtering to obtain a phase correction error;

[0016] Optimize the instantaneous frequency according to the phase correction error to obtain an optimized instantaneous frequency;

[0017] Calculate the multi-band harmonic frequencies through integral calculation according to the optimized instantaneous frequency.

[0018] Optionally, the constructing a virtual impedance based on the multi-band harmonic frequencies includes:

[0019] Taking the gain coefficient and damping ratio as virtual impedance parameters to be optimized, a frequency-band adaptive virtual impedance is constructed according to the multi-band harmonic frequencies to achieve harmonic resonance suppression in different frequency bands.

[0020] Optionally, the impedance parameters include line resistance and line inductance; the filtering parameters include filtering inductance, filtering capacitance and damping resistance; the impedance characteristic matrix includes a system matrix and an input matrix; generating the impedance characteristic matrix based on the impedance parameters and the filtering parameters includes:

[0021] Based on Kirchhoff's law, a differential equation system is constructed according to the line resistance, the line inductance, the filtering inductance, the filtering capacitance and the damping resistance;

[0022] The differential equation system is converted into the form of a standard state equation to obtain a system matrix and an input matrix.

[0023] Optionally, using the impedance characteristic matrix to perform parameter optimization of the virtual impedance based on deep reinforcement learning to obtain the optimal virtual impedance parameters includes:

[0024] Reconstructing the grid current component by combining the system matrix and the input matrix, and performing harmonic separation on the grid current component to obtain the total grid current;

[0025] Constructing an output matrix for the voltage at the point of common coupling based on the line resistance, and solving the voltage at the point of common coupling based on the output matrix;

[0026] Based on the multi-band harmonic frequencies, the total grid current and the voltage at the point of common coupling, and considering the harmonic distortion level, a state space is constructed;

[0027] Generating an action space with the virtual impedance parameters as update parameters based on the state space;

[0028] Constructing a reward function according to the virtual impedance, the multi-band harmonic frequencies and the harmonic distortion level;

[0029] Constructing an action-value function according to the reward function, and constructing a policy network with a double Q-network structure according to the action-value function;

[0030] Iteratively updating the policy network within the range of the action space, and minimizing the loss function during the iterative update process;

[0031] When the harmonic distortion level is less than the preset fluctuation threshold and the reward function reaches a steady state, stop the iteration, and output the virtual impedance parameters obtained after the last iteration as the optimal virtual impedance parameters.

[0032] Optionally, generating an action space with virtual impedance parameters as update parameters based on the state space includes:

[0033] Generating a candidate parameter adjustment amount of the virtual impedance parameter to be optimized according to the state space;

[0034] Constructing an action space for parameter adjustment based on the candidate parameter adjustment amount, and setting a parameter update strategy for the action space; the parameter update strategy is to use the sum of the virtual impedance parameter before update and the candidate parameter adjustment amount as the virtual impedance parameter after update.

[0035] Optionally, the method further includes:

[0036] Taking the optimal virtual impedance parameter as a constraint condition to calculate the optimized harmonic power;

[0037] Calculating the energy recovery efficiency according to the optimized harmonic power, and adjusting the reward function of the optimized deep reinforcement learning based on the energy recovery efficiency.

[0038] The present invention also provides a harmonic suppression device based on virtual impedance collaborative control, including:

[0039] A data acquisition unit for acquiring the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side;

[0040] A phase correction unit for performing real-time phase correction according to the original electrical signal to obtain multi-band harmonic frequencies;

[0041] A virtual impedance and characteristic matrix construction unit for constructing a virtual impedance based on the multi-band harmonic frequencies, and generating an impedance characteristic matrix based on the impedance parameters and the filtering parameters;

[0042] A parameter optimization unit for performing parameter optimization of the virtual impedance based on deep reinforcement learning by using the impedance characteristic matrix to obtain the optimal virtual impedance parameter;

[0043] A harmonic suppression unit for changing the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameter to achieve harmonic suppression.

[0044] The present invention also provides an electronic device, the device includes a processor and a memory:

[0045] The memory is used to store program codes and transmit the program codes to the processor;

[0046] The processor is used to execute the harmonic suppression method based on virtual impedance collaborative control as described in any one of the above according to the instructions in the program codes.

[0047] The present invention also provides a computer-readable storage medium for storing program codes for executing the harmonic suppression method based on virtual impedance cooperative control as described in any one of the above.

[0048] As can be seen from the above technical solutions, the present invention has the following advantages:

[0049] A harmonic suppression method based on multi-modal virtual impedance cooperative control and deep reinforcement learning is provided. In the first step, the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side are obtained as basic data for subsequent harmonic suppression calculations. In the second step, real-time phase correction is performed on the original electrical signal to obtain multi-band harmonic frequencies, so as to perform digital noise suppression based on real-time phase correction, correct errors in real time, and achieve dynamic harmonic decomposition and adaptive tracking of the original electrical signal. In the third step, a virtual impedance is constructed based on the multi-band harmonic frequencies, and an impedance characteristic matrix is generated based on the impedance parameters and filtering parameters. Thus, by constructing a frequency-band adaptive virtual impedance network and combining the adaptive sub-band virtual impedances for cooperative control, and through independent sub-band adjustment, the resonance risk of the traditional LCL filter can be avoided, and sub-band harmonic resonance suppression can be achieved. In the fourth step, the impedance characteristic matrix is used to optimize the parameters of the virtual impedance based on deep reinforcement learning to obtain the optimal virtual impedance parameters. Thus, by optimizing the virtual impedance parameters through intelligent algorithms, a parameter combination with a larger suppression range of harmonic distortion level and better harmonic suppression effect can be obtained. In the fifth step, the equivalent output impedance of the inverter is changed through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression. Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0051] Figure 1 It is a flowchart of the steps of a harmonic suppression method based on virtual impedance cooperative control;

[0052] Figure 2 It is a schematic diagram of the overall process of a harmonic suppression method based on virtual impedance cooperative control;

[0053] Figure 3 It is a structural block diagram of a harmonic suppression device based on virtual impedance cooperative control. Detailed Embodiments

[0054] Embodiments of the present invention provide a harmonic suppression method based on virtual impedance collaborative control, a harmonic suppression device based on virtual impedance collaborative control, an electronic device, and a storage medium, which are used to solve or partially solve the technical problems of strong harmonic time-variation, high governance energy consumption, and parameter solidification in high-penetration photovoltaic scenarios.

[0055] To make the objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] As an example, harmonics are important parameters of the power grid power quality. For the power grid to operate stably, harmonics must be controlled within a certain level. The existence of harmonics will have a negative impact on the performance, efficiency, and stability of the circuit. Therefore, in order to reduce the impact of harmonics, harmonic suppression is a very important part of circuit design.

[0057] Currently, common harmonic suppression methods mainly include passive filtering, active filtering, etc. Passive filtering filters specific frequency harmonics by adding passive components such as inductors and capacitors in the circuit to form an LC filter or other types of filtering networks. Although this method is easy to implement, its parameters are solidified, the filtering effect is limited, and it needs to be designed for specific frequencies, with poor adaptability, and is not suitable for the case of strong harmonic time-variation in high-penetration photovoltaic scenarios. Active filtering, on the other hand, uses devices such as active power filters (APFs) to detect harmonics in real time and inject equal amounts of reverse harmonics to cancel out the harmonics. Although this method has a good filtering effect and can handle harmonics of multiple frequencies, its cost is high, it requires complex control algorithms (such as APF algorithms), and the harmonic governance energy consumption is relatively high.

[0058] With the grid connection of high-penetration distributed photovoltaics, the switching of power electronic devices will generate more harmonics. Therefore, there is an urgent need to propose a more effective harmonic suppression means to solve the three major problems of strong harmonic time-variation, high governance energy consumption, and parameter solidification in high-penetration photovoltaic scenarios.

[0059] Therefore, one of the core inventive points of the embodiments of the present invention is to provide a harmonic suppression method based on multi-modal virtual impedance collaborative control and deep reinforcement learning. The first step is to obtain the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side as basic data for subsequent harmonic suppression calculations. The second step is to perform real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies, thereby performing digital noise suppression based on the real-time phase correction, real-time correcting errors, and achieving dynamic harmonic decomposition and adaptive tracking of the original electrical signal. The third step is to construct a virtual impedance based on the multi-band harmonic frequencies, and generate an impedance characteristic matrix based on the impedance parameters and filtering parameters. Thus, by constructing a frequency-band adaptive virtual impedance network and combining the adaptive sub-band virtual impedance for collaborative control, the resonance risk of the traditional LCL filter can be avoided through sub-band independent adjustment, and sub-band harmonic resonance suppression can be achieved. The fourth step is to optimize the parameters of the virtual impedance based on the impedance characteristic matrix through deep reinforcement learning to obtain the optimal virtual impedance parameters. Therefore, by optimizing the virtual impedance parameters through an intelligent algorithm, a parameter combination with a larger suppression range of harmonic distortion level and better harmonic suppression effect can be obtained. The fifth step is to change the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression. Further, the optimal virtual impedance parameters can be used as constraint conditions to calculate and optimize the harmonic power, calculate the energy recovery efficiency based on the optimized harmonic power, and adjust and optimize the reward function of the deep reinforcement learning based on the energy recovery efficiency to achieve two-way harmonic energy recycling and reduce the power consumption required for harmonic suppression.

[0060] Referring to Figure 1 , a flowchart of the steps of a harmonic suppression method based on virtual impedance collaborative control provided by the embodiments of the present invention is shown, which may specifically include the following steps:

[0061] Step 101, obtain the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side;

[0062] Among them, the original electrical signal represents the original current or voltage signal (including harmonic components). The impedance parameters mainly may include the line resistance and line inductance. The filtering parameters mainly may include the filtering inductor, filtering capacitor, and damping resistor.

[0063] Step 102, perform real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies;

[0064] This step mainly realizes the dynamic harmonic decomposition and adaptive tracking of the original electrical signal. Its working principle is to use the improved Hilbert-Huang transform (HHT) to realize the decomposition of non-steady harmonic components, and combine the Kalman filter for real-time phase calibration.

[0065] In some embodiments, the implementation process of obtaining multi-band harmonic frequencies by performing real-time phase correction based on the original electrical signal may include the following sub-steps S01 to S04:

[0066] Step S01: Decompose the original electrical signal to obtain non-steady harmonic components;

[0067] The signal decomposition formula is as follows:

[0068]

[0069] Among them, represents the original current or voltage signal (including harmonic components); represents the Intrinsic Mode Function (IMF) after the k-th order empirical mode decomposition, that is, the locally symmetric oscillation component generated by HHT decomposition; is the residual term after HHT decomposition, used to reflect the signal trend component.

[0070] Step S02: Extract the instantaneous frequency according to the non-steady harmonic components, and perform real-time phase error correction on the instantaneous phase of the non-steady harmonic components based on Kalman filtering to obtain the phase correction error;

[0071] Specifically, the instantaneous frequency can be extracted by the following formula:

[0072]

[0073] Among them, represents the Hilbert transform result (constructing an analytic signal) of the k-th order IMF; represents the original instantaneous frequency (uncalibrated, including noise) of the k-th order IMF; P represents the Cauchy Principal Value; V represents the Measurement Noise Covariance Matrix.

[0074] Compared with the traditional FFT (Fast Fourier Transform), HHT can process non-stationary signals, and the decomposition accuracy is improved by more than 30%.

[0075] Then, the phase error can be calibrated based on Kalman filtering.

[0076] The state variable of Kalman filtering is the phase error , and the observation variable is the instantaneous phase of . Its state equation and observation equation are designed as follows:

[0077]

[0078] wherein, is process noise, which is Gaussian white noise used to describe phase drift); is measurement noise, i.e., Gaussian white noise from the HHT decomposition error; represents the phase error of the k-th order IMF (caused by noise and non-stationarity); represents the observed phase (phase measurement value including sensor error); represents the original instantaneous phase, which is directly calculated by calculated directly.

[0079] The phase error is corrected in real time through the Kalman gain matrix to obtain the phase correction error for subsequent instantaneous frequency optimization . The correction formula is as follows:

[0080]

[0081] wherein, represents the Kalman gain matrix, which is used to dynamically balance the prediction and observation confidence.

[0082] Step S03: Optimize the instantaneous frequency according to the phase correction error to obtain the optimized instantaneous frequency;

[0083] The optimized and corrected instantaneous frequency is:

[0084]

[0085] Thus, through Kalman filter correction, the phase error of the instantaneous frequency is eliminated.

[0086] By executing steps S02 to S03, the present invention realizes the embedding of Kalman filter into the instantaneous frequency calculation of HHT. Compared with the current noise suppression methods using external filters or digital filters, it can correct errors in real time, and has the advantages of dynamic adaptability, high precision, and low latency. And this method is realized through digitalization and does not increase the hardware consumption.

[0087] Step S04: Calculate the multi-band harmonic frequencies through integration according to the optimized instantaneous frequency.

[0088] The output multi-band harmonic frequencies are as follows:

[0089]

[0090] wherein, T is the signal period, which is determined by the harmonic order h and the fundamental frequency (50 Hz). represents the h - th harmonic frequency, which represents the mean value within the time window.

[0091] Through the state equation, noise suppression and phase calibration are carried out, and a steady - state frequency that is a strict integer multiple of the fundamental frequency can be output ( ( ), corresponding to the harmonic frequencies of multiple frequency bands or sub - frequency bands. For example, in the 50Hz fundamental wave scenario, the 3rd harmonic .

[0092] Multiple frequency bands or sub - frequency bands refer to cutting the harmonics in the power grid into multiple independent intervals according to the frequency (such as the 3rd harmonic 150Hz, the 5th harmonic 250Hz...). Virtual impedance parameters are designed separately for each interval.

[0093] Step 103: Construct a virtual impedance based on the multi - band harmonic frequencies, and generate an impedance characteristic matrix based on the impedance parameters and the filtering parameters;

[0094] This step mainly realizes the collaborative control of multi - modal virtual impedance. By constructing a frequency - band adaptive virtual impedance network, harmonic resonance is suppressed in sub - frequency bands.

[0095] In some embodiments, constructing a virtual impedance based on multi - band harmonic frequencies can specifically be: taking the gain coefficient and damping ratio as the virtual impedance parameters to be optimized, and constructing a frequency - band adaptive virtual impedance according to the multi - band harmonic frequencies to achieve harmonic resonance suppression in sub - frequency bands.

[0096] The constructed virtual impedance is as follows:

[0097]

[0098] Among them, represents the multi - modal virtual impedance, that is, the equivalent impedance with frequency - band adaptive adjustment; is the virtual impedance gain coefficient corresponding to the h - th harmonic, which is used to control the impedance amplitude; is the virtual impedance damping ratio corresponding to the h - th harmonic, which is used to suppress the resonance peak; represents the angular frequency of the h - th harmonic; s represents the Laplace variable (complex - frequency domain operator).

[0099] Furthermore, the impedance characteristic matrix can include a system matrix and an input matrix. Then, generating an impedance characteristic matrix based on the impedance parameters and the filtering parameters can specifically be: based on Kirchhoff's law, constructing a differential equation system according to the line resistance, line inductance, filtering inductance, filtering capacitance, and damping resistance; converting the differential equation system into the standard state - equation form to obtain the system matrix and the input matrix.

[0100] Among them, the state space equation is as follows:

[0101]

[0102] In the formula, the coefficient matrix A and the input matrix B are dynamically generated by the grid impedance and the parameters of the photovoltaic inverter LCL filter (Inductor-Capacitor-Inductor Filter). C and D are output matrices (C maps the state to the output, and D directly passes the input to the output); x represents the state variable vector, including the inductor current , capacitor voltage and other dynamic quantities; u represents the control input vector, corresponding to the inverter PWM modulation (Pulse Width Modulation) signal or current command.

[0103] Combined with the virtual impedance of the adaptive frequency band for cooperative control, through independent adjustment of the frequency band, the resonance risk of the traditional LCL filter can be avoided.

[0104] (the common coupling point voltage, the interaction voltage between the grid and the inverter) is the result of the interaction between the virtual impedance and the grid impedance. Its calculation is determined by the output of the state space equation, that is:

[0105]

[0106] In the subsequent processing process, the virtual impedance can suppress the distortion of the harmonic current at by changing the equivalent output impedance of the inverter.

[0107] The impedance characteristic matrices A and B generated in this step, as the dynamic model parameters of the grid impedance and the photovoltaic inverter LCL filter, are directly transmitted to the deep reinforcement learning in the following step for defining the grid state space.

[0108] The real-time measurement value of is used as a part of the state variable

[0109] of the deep reinforcement learning, directly affecting the reward function R of the reinforcement learning for evaluating the harmonic suppression effect.

[0110] From the perspective of the power grid topology, the impedance parameters on the power grid side include line resistance , inductance . For example, typical values are: , .

[0111] The filtering parameters of the inverter LCL filter include filtering inductance , capacitance , damping resistance . For example, typical values are: ).

[0112] The state variables are defined as . Among them, represents the inductor current on the power grid side (corresponding to ); represents the capacitor voltage of the filter (corresponding to ); represents the inductor current on the inverter side (corresponding to ).

[0113] Matrix A (system matrix) is used to describe the dynamic characteristics of power grid resonance and energy exchange.

[0114] Based on Kirchhoff's voltage / current law, the following differential equations are established:

[0115]

[0116] After arranging it into the standard state equation form , we can get:

[0117]

[0118] Matrix B (input matrix) is used to represent the driving effect of the inverter output voltage on the state.

[0119] In the differential equation , is used as the control input. We can get:

[0120]

[0121] Matrix C (output matrix) defines the observed output quantity (corresponding to the voltage at the point of common coupling ).

[0122] The output equation of matrix C is , which can be transformed into: .

[0123] Matrix D (direct transmission matrix) is used for the direct transmission of the input to the output. Since there is no direct feedthrough scenario in the technical solution of the present invention, it can be simplified and set as .

[0124] Exemplarily, assume that the parameters of a certain photovoltaic power station are , , , , .

[0125] It can be obtained through calculation that:

[0126]

[0127] Furthermore, the influence of the virtual impedance can also be embedded in matrix A through the feedback loop to achieve dynamic expansion. Specifically, by modifying the equation: Convert the transfer function of into the state space form and expand the matrix dimension (equivalent to adding virtual impedance state variables).

[0128] Step 104: Use the impedance characteristic matrix to optimize the parameters of the virtual impedance based on deep reinforcement learning to obtain the optimal virtual impedance parameters;

[0129] This step mainly uses the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to dynamically optimize the virtual impedance parameters to achieve parameter optimization based on deep reinforcement learning.

[0130] In a specific implementation, the execution process of using the impedance characteristic matrix to optimize the parameters of the virtual impedance based on deep reinforcement learning to obtain the optimal virtual impedance parameters mainly includes the following sub-steps S11 to S18:

[0131] Step S11: Reconstruct the grid current component by combining the system matrix and the input matrix, and perform harmonic separation on the grid current component to obtain the total grid current;

[0132] The total grid current can be obtained with the assistance of a calculation model, that is, by inverting the state space equation. Specifically, first combine the state space model in step 103 , and reconstruct the unmeasurable grid current component through an observer (such as a Luenberger observer):

[0133]

[0134] where is the estimated value of the state variable.

[0135] Next, harmonic separation is performed on the grid current components. Specifically, the can be subjected to a Fast S Transform (FST) to separate the fundamental wave and the harmonic currents of each order . The vector sum of the fundamental wave and the harmonic currents of each order is the total grid current :

[0136]

[0137] In another feasible embodiment, the total grid current can also be obtained by deploying sensors to directly measure the source.

[0138] Specifically, a high-precision Hall current sensor is installed at the Point of Common Coupling (PCC) to collect the grid-side current waveform in real time , and the sampling frequency ≥ 10 kHz (meeting the measurement requirements of 50th harmonic). Then, through an anti-aliasing filter and an ADC (Analog-to-Digital Converter) conversion, the analog signal is converted into a discrete sequence to eliminate high-frequency noise interference.

[0139] Among them, when the grid environment is stable (such as no strong electromagnetic interference and good sensor calibration), direct measurement is preferred (high accuracy and strong real-time performance). When the sensor fails or is interfered (such as signal distortion caused by lightning strike), the high-frequency harmonics exceed the sensor bandwidth (such as components > 2 kHz) or predict the future state (such as predicting the current change before the sudden change of photovoltaic output), it is necessary to calculate the effective value of the total grid current in combination with the form of the calculation model.

[0140] For measurement or calculation errors, the allowable error for direct measurement is ±0.5% (for example, when the sensor range is 200 A, the error ≤ 1 A). The allowable error for model calculation is ±1.2%, and this error is mainly caused by the fluctuation of grid parameters.

[0141] Furthermore, when the error is relatively large (> 1.2%), different levels of response measures can also be set in combination with the specific error situation.

[0142] For example, when the error is between 1.5% and 3%, a first-level (primary) response is set correspondingly. At this time, it is possible to: switch to a redundant sensor or a backup model; trigger online calibration (such as injecting white noise to calibrate the sensor frequency response); limit the inverter output to a safe threshold (such as 80% of the rated power).

[0143] When the error is between 3% and 5%, set the secondary (intermediate) response accordingly. At this time, the following can be done: enable interpolation compensation of historical optimal parameters; switch to the "passive filtering" mode (in this case, safety is prioritized over efficiency); upload the fault code to the operation and maintenance platform.

[0144] When the error > 5%, set the tertiary (advanced or emergency) response accordingly. At this time, the following can be done: perform hard disconnection protection (to prevent equipment damage); initiate a manual inspection work order (to locate the sensor or power grid fault point); update the model parameter library (iteratively train based on fault data).

[0145] Thus, by setting corresponding response measures for different error calculation situations, even in the case of large errors during the calculation process, it is possible to respond in a timely manner and avoid accidents.

[0146] Step S12: Construct an output matrix of the common coupling point voltage based on the line resistance, and solve the common coupling point voltage based on the output matrix;

[0147] Step S13: Based on the multi - band harmonic frequencies, the total grid current, and the common coupling point voltage, and considering the harmonic distortion level, construct a state space;

[0148] The constructed state space is as follows:

[0149]

[0150] Among them, reflects the current harmonic distortion level; the multi - band harmonic frequencies characterize the harmonic frequency bands, the common coupling point voltage and the total grid current characterize the grid operation state. characterizes the state vector (including real - time grid operation indicators).

[0151] Step S14: Generate an action space with the virtual impedance parameter as the update parameter based on the state space;

[0152] Specifically, first generate a candidate parameter adjustment amount for the virtual impedance parameter to be optimized according to the state space; then construct an action space for parameter adjustment based on the candidate parameter adjustment amount, and set the parameter update strategy for the action space; among them, the parameter update strategy is to use the sum of the virtual impedance parameter before update and the candidate parameter adjustment amount as the virtual impedance parameter after update.

[0153] First is the generation of the action space. The policy network outputs: according to generate the candidate parameter adjustment amount and , that is:

[0154]

[0155] Among them, represents the action vector (the adjustment amount of virtual impedance parameters); represents the increment of the virtual impedance gain of the h-th harmonic (generated by the TD3 policy network); represents the increment of the damping ratio of the h-th harmonic (generated by the policy network).

[0156] Secondly, it is parameter update:

[0157]

[0158] Among them, the superscript "new" represents the updated parameter, and the superscript "old" represents the parameter before update. Action space constraint is limited to ±10%. Action space constraint is limited to 0.1 - 0.9.

[0159] Step S15: Construct a reward function according to the virtual impedance, multi-band harmonic frequencies, and harmonic distortion levels;

[0160] The constructed reward function is as follows:

[0161]

[0162] Among them, R represents the immediate reward function, which is used to comprehensively evaluate harmonic suppression and energy consumption; , are weight coefficients, which are used to dynamically balance THD suppression and energy consumption; the optimization goal is to maximize , that is, to reduce both and the energy consumption of virtual impedance adjustment; represents the frequency-domain characteristic of the virtual impedance (i.e., at value.

[0163] Step S16: Construct an action-value function according to the reward function, and construct a policy network with a double Q-network structure according to the action-value function;

[0164] The action-value function is used to evaluate the long-term return of the state-action pair, and the calculation formula is as follows:

[0165]

[0166] The policy network is updated as follows:

[0167]

[0168] Among them, Denotes the discount factor, which is used to weigh the importance of current and future rewards; Denotes the policy network parameters (Actor network weight matrix); Denotes the Q-network parameters (Critic network weight matrix).

[0169] There are also other model parameters in the policy network. For example, D represents the experience replay buffer, which is used to store historical state-action-reward sequences, with a capacity of 10 6 entries; N represents the number of batch training samples, that is, the amount of data for a single gradient update, with a default of 256 entries per batch; t represents the time step, which is used to discretize the control period, and the control period is 1 ms (synchronized with PWM).

[0170] Step S17: Iteratively update the policy network within the action space range, and minimize the loss function during the iterative update process;

[0171] The iterative update logic of the policy network is as follows:

[0172] Calculate the target value through the double Q-network structure of the TD3 algorithm , and minimize the loss function .

[0173] Compared with the traditional trial-and-error method, through the iterative update of the policy network, the response speed can be increased by 50%, and the parameters can adapt to the changes in the power grid topology.

[0174] Step S18: When the harmonic distortion level is less than the preset fluctuation threshold and the reward function reaches a steady state, stop the iteration, and output the virtual impedance parameters obtained after the last iteration as the optimal virtual impedance parameters.

[0175] When the fluctuation of 10 consecutive iterations is less than 0.1% and the reward function R reaches a steady state, output the current optimal parameters , .

[0176] Step 105, change the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression.

[0177] When the optimal virtual impedance parameters are obtained through parameter optimization, the equivalent output impedance of the inverter can be changed through the virtual impedance under the optimal virtual impedance parameters , (that is, by injecting the virtual impedance , into the virtual impedance ) to achieve harmonic suppression.

[0178] Furthermore, harmonic energy recovery can also be achieved through a three-level bidirectional DC / AC converter. Specifically, taking the optimal virtual impedance parameter as a constraint condition, the optimized harmonic power is calculated; the energy recovery efficiency is calculated based on the optimized harmonic power, and the reward function of the optimized deep reinforcement learning is adjusted based on the energy recovery efficiency to achieve the recycling of bidirectional harmonic energy.

[0179] Among them, the optimized harmonic power is calculated by the following formula:

[0180]

[0181] Constraint effect: When optimized to the high-frequency band, increases, making compressed within a safe range.

[0182] The energy conversion efficiency is calculated by the following formula :

[0183]

[0184] Among them, is the DC power. is the switching loss; is the core loss; is the effective value of the AC current; is the switching frequency.

[0185] Output value, and then through the feedback parameter, is fed back to the reward function R to correct the weight coefficient. Among them, the adaptive adjustment rule for weight coefficient correction is , .

[0186] Thus, through the recycling of bidirectional harmonic energy, 15% - 20% of harmonic energy can be recovered, reducing the power consumption of the governance system itself.

[0187] Exemplarily, the traditional APF algorithm is compared with the technical solution provided in the embodiment of the present invention. The comparison results of the key indicators are shown in Table 1 below:

[0188] Table 1: Comparison results of key indicators between the traditional scheme and the scheme of the present invention

[0189]

[0190] Combined with Table 1, it can be seen that by adopting the technical solution of the present invention, compared with the traditional processing method, the response time is faster, the suppression range of the harmonic distortion level is larger, and the adaptive ability is better. And due to the adoption of the recycling of bidirectional harmonic energy, the power consumption is smaller.

[0191] In an embodiment of the present invention, a harmonic suppression method based on multi-modal virtual impedance cooperative control and deep reinforcement learning is provided. In the first step, the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side are obtained as basic data for subsequent harmonic suppression calculations. In the second step, real-time phase correction is performed on the original electrical signal to obtain multi-band harmonic frequencies, so as to perform digital noise suppression based on the real-time phase correction, correct the error in real time, and realize the dynamic harmonic decomposition and adaptive tracking of the original electrical signal. In the third step, a virtual impedance is constructed based on the multi-band harmonic frequencies, and an impedance characteristic matrix is generated based on the impedance parameters and filtering parameters. Thus, by constructing a frequency-band adaptive virtual impedance network and combining the adaptive sub-band virtual impedance for cooperative control, the risk of resonance of the traditional LCL filter can be avoided through sub-band independent adjustment, and sub-band harmonic resonance suppression can be achieved. In the fourth step, the impedance characteristic matrix is used to optimize the parameters of the virtual impedance based on deep reinforcement learning to obtain the optimal virtual impedance parameters. Thus, by optimizing the parameters of the virtual impedance through an intelligent algorithm, a parameter combination with a larger suppression range of the harmonic distortion level and a better harmonic suppression effect can be obtained. In the fifth step, the equivalent output impedance of the inverter is changed through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression. Further, the optimal virtual impedance parameters can also be used as constraint conditions to calculate and optimize the harmonic power, calculate the energy recovery efficiency based on the optimized harmonic power, and adjust and optimize the reward function of the deep reinforcement learning based on the energy recovery efficiency to realize the two-way recycling of harmonic energy and reduce the power consumption required for harmonic suppression. The technical solution of the present invention effectively solves the three major problems of strong harmonic time-variation, high governance energy consumption, and parameter solidification in the high-penetration photovoltaic scenario through the four-stage cooperation of signal decomposition-impedance control-intelligent optimization-energy cycle.

[0192] For better illustration, refer to Figure 2 , which shows a schematic diagram of the overall process of a harmonic suppression method based on virtual impedance cooperative control provided by an embodiment of the present invention. It should be noted that this embodiment only briefly describes the general process of harmonic suppression based on virtual impedance cooperative control. The specific implementation process of each step can be understood by referring to the relevant content in the foregoing embodiments and will not be elaborated here. It can be understood that the present invention places no restrictions on this.

[0193] Step 201: Obtain the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side;

[0194] Step 202: Perform real-time phase correction on the original electrical signal by combining HHT signal decomposition and Kalman filtering to obtain multi-band harmonic frequencies;

[0195] Step 203: Take the gain coefficient and damping ratio as the virtual impedance parameters to be optimized. Based on the multi-band harmonic frequencies, construct a frequency-band adaptive virtual impedance, and based on Kirchhoff's law, generate an impedance characteristic matrix according to the impedance parameters and filtering parameters.

[0196] Step 204: Use the impedance characteristic matrix to optimize the parameters of the virtual impedance based on deep reinforcement learning, obtain the optimal virtual impedance parameters, and change the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression.

[0197] Step 205: Take the optimal virtual impedance parameters as the constraint conditions, calculate the optimized harmonic power, calculate the energy recovery efficiency according to the optimized harmonic power, and adjust the reward function of the optimized deep reinforcement learning based on the energy recovery efficiency.

[0198] Refer to Figure 3 , which shows the structural block diagram of a harmonic suppression device based on virtual impedance collaborative control provided by an embodiment of the present invention. Specifically, it may include:

[0199] A data acquisition unit 301, configured to acquire the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side.

[0200] A phase correction unit 302, configured to perform real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies.

[0201] A virtual impedance and characteristic matrix construction unit 303, configured to construct a virtual impedance based on the multi-band harmonic frequencies, and generate an impedance characteristic matrix based on the impedance parameters and the filtering parameters.

[0202] A parameter optimization unit 304, configured to use the impedance characteristic matrix to optimize the parameters of the virtual impedance based on deep reinforcement learning to obtain the optimal virtual impedance parameters.

[0203] A harmonic suppression unit 305, configured to change the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression.

[0204] In an optional embodiment, the phase correction unit 302 includes:

[0205] A signal decomposition unit, configured to decompose the original electrical signal to obtain the non-steady harmonic components.

[0206] A real-time phase error correction unit, configured to extract the instantaneous frequency according to the non-steady harmonic components, and perform real-time phase error correction on the instantaneous phase of the non-steady harmonic components based on Kalman filtering to obtain the phase correction error.

[0207] An instantaneous frequency optimization unit, configured to optimize the instantaneous frequency according to the phase correction error to obtain an optimized instantaneous frequency;

[0208] A multi-band harmonic frequency calculation unit, configured to calculate a multi-band harmonic frequency through integration according to the optimized instantaneous frequency.

[0209] In an alternative embodiment, the virtual impedance and characteristic matrix construction unit 303 includes:

[0210] A virtual impedance construction unit, configured to use the gain coefficient and damping ratio as virtual impedance parameters to be optimized, and construct a frequency band adaptive virtual impedance according to the multi-band harmonic frequency to achieve harmonic resonance suppression in sub-bands.

[0211] In an alternative embodiment, the impedance parameters include line resistance and line inductance; the filtering parameters include filtering inductance, filtering capacitance and damping resistance; the impedance characteristic matrix includes a system matrix and an input matrix; the virtual impedance and characteristic matrix construction unit 303 includes an impedance characteristic matrix generation unit, and the impedance characteristic matrix generation unit is specifically configured to:

[0212] Based on Kirchhoff's law, construct a differential equation set according to the line resistance, the line inductance, the filtering inductance, the filtering capacitance and the damping resistance;

[0213] Convert the differential equation set into a standard state equation form to obtain a system matrix and an input matrix.

[0214] In an alternative embodiment, the parameter optimization unit 304 includes:

[0215] A grid current processing unit, configured to reconstruct the grid current component by combining the system matrix and the input matrix, and perform harmonic separation on the grid current component to obtain the total grid current;

[0216] A common coupling point voltage solving unit, configured to construct an output matrix about the common coupling point voltage according to the line resistance, and solve the common coupling point voltage based on the output matrix;

[0217] A state space construction unit, configured to construct a state space based on the multi-band harmonic frequency, the total grid current and the common coupling point voltage, and simultaneously consider the harmonic distortion level;

[0218] An action space construction unit, configured to generate an action space with the virtual impedance parameter as the update parameter based on the state space;

[0219] A reward function construction unit, configured to construct a reward function according to the virtual impedance, the multi-band harmonic frequency and the harmonic distortion level;

[0220] A policy network construction unit, configured to construct an action value function according to the reward function, and construct a policy network with a double Q network structure according to the action value function;

[0221] An iterative update unit, configured to iteratively update the policy network within the action space range, and minimize a loss function during the iterative update process;

[0222] An optimal virtual impedance parameter output unit, configured to stop the iteration when the harmonic distortion level is less than a preset fluctuation threshold and the reward function reaches a steady state, and output the virtual impedance parameter obtained after the last iteration as the optimal virtual impedance parameter.

[0223] In an alternative embodiment, the action space construction unit includes:

[0224] A candidate parameter adjustment amount generation unit, configured to generate a candidate parameter adjustment amount of the virtual impedance parameter to be optimized according to the state space;

[0225] An action space construction subunit, configured to construct an action space for parameter adjustment based on the candidate parameter adjustment amount, and set a parameter update strategy for the action space; the parameter update strategy is to use the sum of the virtual impedance parameter before update and the candidate parameter adjustment amount as the virtual impedance parameter after update.

[0226] In an alternative embodiment, the device further includes:

[0227] An optimized harmonic power calculation unit, configured to calculate optimized harmonic power by using the optimal virtual impedance parameter as a constraint condition;

[0228] A reward function adjustment and optimization unit, configured to calculate an energy recovery efficiency according to the optimized harmonic power, and adjust and optimize the reward function of the deep reinforcement learning based on the energy recovery efficiency.

[0229] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For related parts, please refer to the partial description of the foregoing method embodiment.

[0230] An embodiment of the present invention further provides an electronic device, the device including a processor and a memory:

[0231] The memory is used to store program code and transmit the program code to the processor;

[0232] The processor is configured to execute the harmonic suppression method based on virtual impedance collaborative control according to any embodiment of the present invention according to the instructions in the program code.

[0233] An embodiment of the present invention further provides a computer-readable storage medium, which is used to store program codes for executing the harmonic suppression method based on virtual impedance collaborative control according to any embodiment of the present invention.

[0234] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0235] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0236] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0237] In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0238] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0239] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of the present invention.

Claims

1. A harmonic suppression method based on virtual impedance collaborative control, characterized in that Including: Obtain the original electrical signal, impedance parameters on the grid side, and filter parameters on the inverter side; Perform real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies; Construct a virtual impedance based on the multi-band harmonic frequencies, and generate an impedance characteristic matrix based on the impedance parameters and the filter parameters; Use the impedance characteristic matrix to perform parameter optimization of the virtual impedance based on deep reinforcement learning to obtain optimal virtual impedance parameters; Change the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression.

2. The harmonic suppression method based on virtual impedance collaborative control according to claim 1, wherein The performing real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies includes: Decompose the original electrical signal to obtain the non-steady harmonic components; Extract the instantaneous frequency according to the non-steady harmonic components, and perform real-time phase error correction on the instantaneous phase of the non-steady harmonic components based on Kalman filtering to obtain a phase correction error; Optimize the instantaneous frequency according to the phase correction error to obtain an optimized instantaneous frequency; According to the optimized instantaneous frequency, obtain the multi-band harmonic frequencies through integral calculation.

3. The harmonic suppression method based on virtual impedance collaborative control according to claim 1, wherein The constructing a virtual impedance based on the multi-band harmonic frequencies includes: Take the gain coefficient and damping ratio as the virtual impedance parameters to be optimized, and construct a frequency-band adaptive virtual impedance according to the multi-band harmonic frequencies to achieve sub-band harmonic resonance suppression.

4. The harmonic suppression method based on virtual impedance collaborative control according to claim 1, characterized in that, The impedance parameters include line resistance and line inductance; the filter parameters include filter inductance, filter capacitance, and damping resistance; The impedance characteristic matrix includes a system matrix and an input matrix; the generating an impedance characteristic matrix based on the impedance parameters and the filter parameters includes: Based on Kirchhoff's law, construct a differential equation system according to the line resistance, the line inductance, the filter inductance, the filter capacitance, and the damping resistance; Convert the differential equation system into the standard state equation form to obtain the system matrix and the input matrix.

5. The harmonic suppression method based on virtual impedance collaborative control according to claim 4, wherein The using the impedance characteristic matrix to perform parameter optimization of the virtual impedance based on deep reinforcement learning to obtain optimal virtual impedance parameters includes: Reconstruct the grid current component by combining the system matrix and the input matrix, and perform harmonic separation on the grid current component to obtain the total grid current; Construct an output matrix about the voltage at the point of common coupling according to the line resistance, and solve the voltage at the point of common coupling based on the output matrix; Construct a state space based on the multi-band harmonic frequencies, the total grid current, and the voltage at the point of common coupling, while considering the harmonic distortion level; Generate an action space with the virtual impedance parameters as the update parameters based on the state space; Construct a reward function according to the virtual impedance, the multi-band harmonic frequencies, and the harmonic distortion level; Construct an action value function according to the reward function, and construct a policy network with a double Q-network structure according to the action value function; Iteratively update the policy network within the range of the action space, and minimize the loss function during the iterative update process; When the harmonic distortion level is less than the preset fluctuation threshold and the reward function reaches a steady state, stop the iteration and output the virtual impedance parameters obtained after the last iteration as the optimal virtual impedance parameters.

6. The harmonic suppression method based on virtual impedance collaborative control according to claim 5, wherein The generation of the action space with the virtual impedance parameters as the update parameters based on the state space includes: Generating a candidate parameter adjustment amount of the virtual impedance parameter to be optimized according to the state space; Constructing an action space for parameter adjustment based on the candidate parameter adjustment amount and setting a parameter update strategy for the action space; the parameter update strategy is to use the sum of the virtual impedance parameter before update and the candidate parameter adjustment amount as the virtual impedance parameter after update.

7. The harmonic suppression method based on virtual impedance collaborative control according to any one of claims 1 to 6, characterized in that It further includes: Taking the optimal virtual impedance parameter as a constraint condition to calculate the optimized harmonic power; Calculating the energy recovery efficiency according to the optimized harmonic power and adjusting the reward function of the optimized deep reinforcement learning based on the energy recovery efficiency.

8. A harmonic suppression device based on virtual impedance collaborative control, characterized in that, It includes: A data acquisition unit for acquiring the original electrical signal, impedance parameters on the grid side, and filtering parameters on the inverter side; A phase correction unit for performing real-time phase correction on the original electrical signal to obtain multi-band harmonic frequencies; A virtual impedance and characteristic matrix construction unit for constructing a virtual impedance based on the multi-band harmonic frequencies and generating an impedance characteristic matrix based on the impedance parameters and the filtering parameters; A parameter optimization unit for performing parameter optimization of the virtual impedance based on deep reinforcement learning using the impedance characteristic matrix to obtain the optimal virtual impedance parameters; A harmonic suppression unit for changing the equivalent output impedance of the inverter through the virtual impedance under the optimal virtual impedance parameters to achieve harmonic suppression.

9. An electronic device, characterized in that, The device includes a processor and a memory: The memory is used to store program codes and transmit the program codes to the processor; The processor is used to execute the harmonic suppression method based on virtual impedance collaborative control according to any one of claims 1-7 according to the instructions in the program codes.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program codes, and the program codes are used to execute the harmonic suppression method based on virtual impedance collaborative control according to any one of claims 1-7.

Images