A photovoltaic power station virtual synchronous control method and a control system thereof

Abstract

The present application relates to the field of new energy power generation grid-connected control technology, specifically to a kind of photovoltaic power station virtual synchronous control method and control system thereof, comprising: obtaining the multi-source data set of photovoltaic power station DC side, AC side electrical quantity, point of common coupling voltage and environmental data, separating it into low-frequency trend, power frequency cycle and instantaneous disturbance component by time-frequency domain hybrid decomposition. The power frequency component is injected into the equivalent circuit model composed of inductance simulating inertia and resistance simulating damping, and the virtual internal potential and power angle are obtained by circuit solving. Combined with low-frequency and residual component, a multi-dimensional reference trajectory is generated through behavior prediction network, and a dynamic error surface is constructed accordingly. The duty cycle correction sequence is calculated along the gradient direction of the error surface, and finally issued to each inverter to adjust the switching time sequence. The method improves the control accuracy and working condition adaptability, and enhances the physical interpretability of control.

Description

Technical Field

[0001] This invention relates to the field of grid-connected control technology for new energy power generation, and in particular to a virtual synchronization control method and control system for photovoltaic power plants. Background Technology

[0002] Photovoltaic power plants are connected to the grid via power electronic inverters. Lacking the rotational inertia and damping characteristics of traditional synchronous generators, they lack the active support capability to withstand grid frequency fluctuations, impacting grid stability. To endow photovoltaic power plants with frequency regulation capabilities similar to synchronous machines, virtual synchronous machine technology has been proposed. Existing technologies are generally based on mathematical models of the rotor motion equations of synchronous generators. They calculate instantaneous power by collecting voltage and current at the point of common coupling, and then substitute these data into a second-order swing equation to solve for virtual speed and power angle online, thereby generating reference current or voltage commands. These methods treat the photovoltaic power plant as a whole or control individual inverters.

[0003] Existing technical solutions have shortcomings. Actual operating data from photovoltaic power plants is mixed with various components, including slow power drift caused by changes in sunlight and temperature, power frequency cycle fluctuations, switching harmonics, and transient faults. Traditional methods often employ simple filtering, which struggles to accurately separate these physically distinct components. This results in impure input signals for the virtual synchronization control loop, degraded dynamic response quality under complex operating conditions, and even induces subsynchronous oscillations. Furthermore, calculations relying on purely mathematical models lack an intuitive correspondence at the physical circuit level. Key parameters such as virtual inertia and damping are merely coefficients in the algorithm, and their tuning and optimization depend on experience and extensive simulations. This makes it difficult to establish a clear mapping with the dynamic characteristics of the actual power grid, affecting the reliability and interpretability of the control.

[0004] The key problem that this invention needs to solve is how to accurately separate the different dynamic characteristic components in the multi-source operation data of photovoltaic power plants, and provide a virtual synchronous control with a clear physical meaning to improve the adaptability, accuracy and reliability of the control. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose a virtual synchronous control method and control system for photovoltaic power plants.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a virtual synchronization control method for a photovoltaic power station, comprising: Acquire multi-source datasets of photovoltaic power plant operation, which include instantaneous electrical quantities of the DC and AC sides of the photovoltaic inverter, three-phase voltage waveforms at the point of common coupling, and light intensity and temperature recorded by environmental monitoring devices; The multi-source dataset is subjected to time-frequency domain hybrid decomposition to separate the low-frequency component representing a slow changing trend, the power frequency component representing periodic fluctuations, and the residual component representing instantaneous disturbances. The power frequency component is injected into the pre-constructed virtual synchronous machine equivalent circuit model, which consists of an inductor branch simulating mechanical inertia and a resistor branch simulating damping characteristics. The virtual internal potential and virtual power angle are obtained by solving the circuit. The virtual internal potential, the virtual power angle, the low-frequency component, and the residual component are fed into the behavior prediction network, and the behavior prediction network outputs a multi-dimensional reference trajectory for the next control cycle. Based on the multi-dimensional reference trajectory and the real-time power frequency components, a dynamic error surface is constructed; Solve for the gradient direction of the dynamic error surface, and generate a set of duty cycle correction sequences for pulse width modulation waves along the gradient direction; The duty cycle correction sequence is sent to the pulse width modulation generator of each photovoltaic inverter in the photovoltaic power station to adjust the switching timing of each photovoltaic inverter.

[0007] Preferably, the step of performing time-frequency domain hybrid decomposition on the multi-source dataset specifically includes: The three-phase voltage waveform is subjected to phase-locked processing to obtain the synchronous rotation angle; Using the synchronous rotation angle, the instantaneous electrical quantity is transformed from the stationary coordinate system to the synchronous rotating coordinate system to obtain the direct axis component and the quadrature axis component; Sliding time windows are applied to the direct-axis components and the quadrature-axis components respectively, and empirical mode decomposition is performed within each time window to obtain a series of intrinsic mode functions arranged from high frequency to low frequency; The intrinsic mode functions with frequencies higher than a preset threshold are classified as the residual components; The eigenmode functions with frequencies between a preset frequency range are reconstructed to obtain the power frequency components; The low-frequency component is obtained by superimposing the intrinsic mode functions with frequencies below a preset threshold and the trend term.

[0008] Preferably, the step of injecting the power frequency component into a pre-constructed virtual synchronous machine equivalent circuit model, wherein the virtual synchronous machine equivalent circuit model consists of an inductor branch simulating mechanical inertia and a resistor branch simulating damping characteristics, and obtaining the virtual internal potential and virtual power angle through circuit solving, specifically includes: The voltage and current in the power frequency component are used as the port voltage and port current of the equivalent circuit model of the virtual synchronous machine. Based on Kirchhoff's voltage law and Kirchhoff's current law, list the system of differential-algebraic equations for the equivalent circuit model of the virtual synchronous machine. The differential-algebraic equations are solved using numerical integration to obtain the current and flux linkage of the inductor branch. The flux linkage of the inductor branch is defined as the amplitude of the virtual internal potential; The difference between the phase angle of the port voltage and the phase angle of the virtual internal potential is defined as the virtual power angle.

[0009] Preferably, the behavior prediction network is a temporal convolutional network pre-trained with historical operating data. Its input layer receives the time series of the virtual internal potential, virtual power angle, low-frequency component and residual component, and its output layer outputs the multi-dimensional reference trajectory. The multi-dimensional reference trajectory includes the reference active power trajectory, the reference reactive power trajectory, and the reference voltage amplitude trajectory for future control cycles.

[0010] Preferably, the construction of the dynamic error surface specifically includes: Extract the actual active power, actual reactive power, and actual voltage amplitude corresponding to the power frequency component at the current moment; The difference between the actual active power and the starting point of the reference active power trajectory is calculated as the instantaneous error of active power. The difference between the actual reactive power and the starting point of the reference reactive power trajectory is calculated as the instantaneous reactive power error. The difference between the actual voltage amplitude and the starting point of the reference voltage amplitude trajectory is calculated as the instantaneous error of the voltage amplitude. Using the instantaneous errors of active power, reactive power, and voltage amplitude as independent variables, a three-dimensional error space is constructed. In the error space, a surface is defined as a function of the sum of the squares of the three instantaneous errors; this surface is the dynamic error surface.

[0011] Preferably, the step of solving the gradient direction of the dynamic error surface and generating a set of duty cycle correction sequences for pulse width modulation waves along the gradient direction specifically includes: Calculate the partial derivatives of the dynamic error surface corresponding to the instantaneous errors of active power, reactive power, and voltage amplitude; The partial derivatives form a gradient vector, and the direction of the gradient vector is the direction that makes the dynamic error surface decrease the fastest. The gradient vector is mapped onto the direct-axis voltage and quadrature-axis voltage control components of the photovoltaic inverter to obtain the voltage correction amount; Calculate the desired voltage space vector based on the voltage correction amount and the DC bus voltage; Based on the desired voltage space vector, the action time of each of the three basic voltage vectors is calculated using the nearest three-vector synthesis method. Divide the duration of each basic voltage vector by the pulse width modulation period to obtain the duty cycle correction sequence for the corresponding bridge arm.

[0012] Preferably, after obtaining the multi-source dataset of the photovoltaic power station operation, the method further includes a data credibility verification step: Cross-compare similar data from different monitoring points. If the data deviation exceeds the tolerance, an alternative strategy based on interpolation of adjacent data is initiated. Calculate the noise energy of each data channel in the multi-source dataset, and perform noise reduction filtering processing based on wavelet threshold for the channel data with noise energy exceeding the threshold.

[0013] Preferably, the step of solving the system of differential-algebraic equations using numerical integration to obtain the current and flux linkage of the inductor branch specifically includes: Discretize the differential equations in the system of differential algebraic equations into difference equations; Set the solution step size, and iteratively calculate the current and flux linkage values ​​of the inductor branch at the current moment based on the values ​​of the current and flux linkage of the inductor branch at the previous moment. In each iteration, the solution of the algebraic equation system is substituted into the difference equation to update the values ​​of the current and flux linkage of the inductor branch. The iterative process is repeated until the solution of the differential-algebraic equation system converges, and the values ​​of the current and flux linkage of the inductor branch that have finally converged are output.

[0014] Preferably, the calculation of the partial derivatives of the dynamic error surface corresponding to the instantaneous errors of active power, reactive power, and voltage amplitude specifically includes: The functional expression of the dynamic error surface is differentiated with respect to the instantaneous error variables of active power, reactive power, and voltage amplitude, respectively. In the process of differentiating the instantaneous error variable of active power, the instantaneous error variable of reactive power and the instantaneous error variable of voltage amplitude are regarded as constants to obtain the partial derivative with respect to the instantaneous error of active power. In the process of differentiating the instantaneous error variable of reactive power, the instantaneous error variable of active power and the instantaneous error variable of voltage amplitude are regarded as constants to obtain the partial derivative with respect to the instantaneous error of reactive power; In the process of differentiating the instantaneous error variable of voltage amplitude, the instantaneous error variables of active power and reactive power are treated as constants to obtain the partial derivative with respect to the instantaneous error of voltage amplitude.

[0015] Preferably, the present invention also includes a virtual synchronization control system for a photovoltaic power station, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the virtual synchronization control method for a photovoltaic power station as described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A time-frequency domain hybrid decomposition is performed on a multi-source dataset containing electrical quantities and environmental parameters to separate three types of characteristic components: low-frequency trend, power frequency period, and transient disturbance residuals. This technique decouples environmental gradual changes, fundamental power, and transient disturbances from the mixed signal, enabling subsequent control loops to perform differentiated processing on components with different physical characteristics. The purity of the input signal is improved, effectively avoiding mutual interference between different types of dynamic processes and enhancing the robustness and response accuracy of the control system under complex and fluctuating conditions. The core algorithm of virtual synchronous control only processes the power frequency component representing periodic fluctuations, thus isolating it from the negative impacts of slow-changing trends and sudden disturbances.

[0017] An equivalent circuit model is constructed, consisting of an inductive branch simulating inertia and a resistive branch simulating damping. The separated power frequency electrical quantities are injected into this model, and the virtual internal potential and power angle are directly obtained through circuit solving. This technique physicalizes the abstract virtual inertia and damping coefficients into specific circuit element parameters, and the control process is equivalent to the electromagnetic transient response of this equivalent circuit. This approach gives the dynamic characteristics of the virtual synchronous machine an intuitive physical circuit correspondence, and parameter design can draw on the influence of inductance and resistance on the dynamic response in actual power grids. The predictability and interpretability of control behavior are enhanced, allowing engineers to more directly map and tune circuit parameters based on the actual requirements of the power grid for inertia and damping, simplifying the complexity of engineering implementation and improving the inherent stability of the system. Attached Figure Description

[0018] Figure 1 This is a flowchart of the virtual synchronization control method for photovoltaic power plants described in this invention; Figure 2 This is a flowchart of time-frequency domain hybrid decomposition; Figure 3 A flowchart for solving the equivalent circuit model of a virtual synchronous machine; Figure 4 A bar chart showing the basic vector action time distribution of voltage space vector sectors in the virtual synchronous control of a photovoltaic power station; Figure 5 A multi-dimensional trajectory comparison radar chart for virtual synchronous control of photovoltaic power plants. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 A multi-source dataset of photovoltaic power plant operation is acquired, including instantaneous electrical quantities on the DC and AC sides of the photovoltaic inverter, three-phase voltage waveforms at the point of common coupling, and irradiance and temperature recorded by environmental monitoring devices. The acquired multi-source dataset undergoes time-frequency domain hybrid decomposition to separate low-frequency components representing slow-changing trends, power frequency components representing periodic fluctuations, and residual components representing instantaneous disturbances. The decomposed power frequency components are injected into a pre-constructed virtual synchronous machine equivalent circuit model, which consists of an inductor branch simulating mechanical inertia and a resistor branch simulating damping characteristics. The virtual internal potential and virtual power angle are obtained through circuit solving. The obtained virtual internal potential and virtual power angle, along with the previously decomposed low-frequency and residual components, are fed into a behavior prediction network, which outputs a multi-dimensional reference trajectory for the next control cycle. Based on this multi-dimensional reference trajectory and the real-time acquired and decomposed power frequency components, a dynamic error surface is constructed. The gradient direction of the dynamic error surface is determined, and a set of pulse width modulation (PWM) duty cycle correction sequences for adjusting the inverter switching states is generated along this gradient direction. This duty cycle correction sequence is then sent to the PWM generators of each photovoltaic inverter within the photovoltaic power station to adjust the switching timing of each inverter, thereby achieving virtual synchronous control of the entire photovoltaic power station.

[0022] In one embodiment of the present invention, see [reference] Figure 2The specific process of time-frequency domain hybrid decomposition of multi-source datasets includes: performing phase-locked loop processing on the three-phase voltage waveforms at the point of common coupling to obtain the synchronous rotation angle of the power grid; using this synchronous rotation angle, transforming the instantaneous voltage and current, and other electrical quantities on the AC side of the photovoltaic inverter, from the stationary coordinate system to the synchronous rotating coordinate system to obtain the corresponding direct-axis and quadrature-axis components; applying a sliding time window to each of the obtained direct-axis and quadrature-axis components, and performing empirical mode decomposition on the data within each time window to obtain a series of intrinsic mode functions (IMFs) arranged from high frequency to low frequency; classifying IMFs with frequencies higher than a preset threshold as residual components; superimposing and reconstructing IMFs with frequencies between a preset power frequency range to obtain the power frequency components; and superimposing IMFs with frequencies lower than a preset threshold and the trend terms generated by empirical mode decomposition to obtain low-frequency components. After acquiring the multi-source datasets, a data reliability verification step is also included. This step cross-compares similar data from different monitoring points. If the deviation between the data from a certain monitoring point and the data from other points exceeds a preset tolerance, an interpolation strategy based on data from adjacent time points or adjacent monitoring points is initiated to generate alternative data. Simultaneously, the noise energy of each data channel in the multi-source dataset is calculated. For channel data with noise energy exceeding a set threshold, wavelet threshold-based noise reduction filtering is performed.

[0023] In practical implementation, the multi-source datasets are acquired from monitoring units deployed within the power plant. These include sensors that collect DC-side voltage and current, and AC-side three-phase voltage and current from the photovoltaic inverter; waveform recording devices that record the three-phase voltage waveform at the point of common coupling (PCC); and illuminance and temperature sensors located in the photovoltaic array area. This data is aggregated to the central controller via the station's communication network. In practice, the phase-locked loop (PLL) processing of the three-phase voltage waveform typically employs a software PLL algorithm. This algorithm precisely extracts the phase of the grid fundamental positive-sequence component from the PCC three-phase voltage waveform, thereby obtaining the synchronous rotation angle. In practice, the software PLL algorithm utilizes the coordinate transformation performed by the synchronous rotation angle. This transformation converts the collected instantaneous three-phase voltage and current on the AC side of the photovoltaic inverter to a direct-axis and quadrature-axis coordinate system that rotates synchronously with the grid voltage through Clark and Park transformations. This yields the time-varying direct-axis voltage component, quadrature-axis voltage component, direct-axis current component, and quadrature-axis current component. The sliding time window applied to the direct-axis and quadrature-axis components has a fixed window length, which covers multiple power frequency cycles. Within each time window, the direct-axis voltage component sequence within the window is independently subjected to empirical mode decomposition. The decomposition process iteratively selects components that satisfy the intrinsic mode function conditions. Similarly, the same empirical mode decomposition operation is independently performed on the quadrature-axis voltage component sequence, the direct-axis current component sequence, and the quadrature-axis current component sequence within the window.

[0024] In some embodiments, a series of intrinsic mode functions (EMFs) obtained after empirical mode decomposition are arranged in descending order of their instantaneous frequencies. A preset threshold is set to a frequency value much higher than the power frequency, such as 150 Hz. All EMFs with instantaneous frequencies higher than this preset threshold are classified as residual components characterizing instantaneous disturbances. A preset frequency range is set to a narrow band interval around the power frequency, such as 45 Hz to 55 Hz. EMFs with instantaneous frequencies falling within this preset frequency range are superimposed and reconstructed to obtain the power frequency component characterizing periodic fluctuations. The remaining EMFs with instantaneous frequencies lower than the preset threshold, along with the monotonic trend term generated during the empirical mode decomposition process, are superimposed to obtain the low-frequency component characterizing a slowly changing trend. This decomposition process is performed independently for each data sequence within each sliding time window. As the time window slides, the data in the new window undergoes a new round of decomposition.

[0025] In practice, the data reliability verification step is executed immediately after the central controller receives the multi-source dataset. For similar data from different monitoring points, such as light intensity data from three different weather stations, the central controller performs real-time cross-comparison. The central controller calculates the pairwise differences between the three light intensity readings. If any difference exceeds a preset tolerance value, such as 50 watts per square meter, an abnormal data point is identified. Optionally, the central controller then initiates a replacement strategy based on adjacent data interpolation. Specifically, it generates replacement data by linear interpolation based on the effective light intensity data of the previous and next time moments in the time series, or based on the effective light intensity data of spatially adjacent weather stations, to replace the data identified as abnormal.

[0026] In some embodiments, the process of calculating the noise energy of each data channel involves applying a high-pass digital filter to the original sampled data sequence of each channel to filter out the fundamental frequency and low-frequency trends, retaining the high-frequency components, and then calculating the sum of squares of the high-frequency sequence over a fixed calculation period as the noise energy estimate. It is understood that when the noise energy estimate of a channel exceeds a set threshold, it indicates that the data in that channel is significantly contaminated by high-frequency noise. Optionally, the central controller then performs wavelet threshold-based denoising filtering on the original data of that channel. The processing selects an appropriate wavelet basis function and decomposition level, performs wavelet decomposition on the signal, applies a hard or soft threshold function to the obtained wavelet coefficients for quantization, sets wavelet coefficients below the threshold to zero, and then uses the processed wavelet coefficients to reconstruct the signal. The resulting denoised data can then be used for subsequent time-frequency domain hybrid decomposition. It is understood that the above cross-comparison and denoising filtering processes are applied in parallel to all relevant data channels of the multi-source dataset to ensure that the data input to the time-frequency domain hybrid decomposition stage has sufficient reliability. In practical implementation, the condition for determining whether a component is a strict eigenmode function during empirical mode decomposition is that, throughout the entire data sequence, the number of extreme points must be equal to or at most differ by one zero-crossing point, and the mean of the upper envelope defined by local maxima and the lower envelope defined by local minima is zero at any point. This condition can be expressed mathematically as follows: for a data sequence... The first one obtained after screening Individual eigenmode functions It is necessary to ensure the average value of its upper and lower envelopes. Satisfying the relation:

[0027] in: represent The local maxima are obtained by using the upper envelope formed by cubic spline interpolation. represent The local minimum is obtained by using the lower envelope formed by cubic spline interpolation. Represents a time variable, symbol This indicates that it is approximately zero over the entire time span.

[0028] In one embodiment of the present invention, see [reference] Figure 3The specific process of injecting power frequency components into a pre-constructed virtual synchronous machine equivalent circuit model and obtaining the virtual internal potential and virtual power angle through circuit solution includes: using the voltage and current in the power frequency components as the port voltage and port current inputs of the virtual synchronous machine equivalent circuit model; formulating a system of differential-algebraic equations for the virtual synchronous machine equivalent circuit model based on Kirchhoff's voltage and current laws; solving the system of differential-algebraic equations using numerical integration to obtain the current and flux linkage values ​​of the inductor branch of the model; specifically, discretizing the differential equations in the system of differential-algebraic equations into difference equations; setting a solution step size; and iteratively calculating the current and flux linkage values ​​of the inductor branch at the current moment based on the current and flux linkage values ​​of the inductor branch at the previous moment; and in each iteration step, substituting the solution of the algebraic equation system into the difference equations, and then using the difference equations to update the current and flux linkage values ​​of the inductor branch. Repeat this iterative process until the solution to the differential-algebraic equation system converges, and output the final converged values ​​of the inductor branch current and flux linkage. Define the flux linkage of the inductor branch obtained from the solution as the magnitude of the virtual internal potential. Define the difference between the phase angle of the model port voltage and the phase angle of the virtual internal potential as the virtual power angle.

[0029] In practical implementation, the power frequency components include the direct-axis voltage power frequency component, quadrature-axis voltage power frequency component, direct-axis current power frequency component, and quadrature-axis current power frequency component after coordinate transformation and reconstruction. In practical implementation, the pre-constructed virtual synchronous machine equivalent circuit model is defined in the simulation environment of the central controller. The virtual synchronous machine equivalent circuit model includes an inductor branch simulating mechanical inertia and a resistor branch simulating damping characteristics, with the inductor and resistor branches connected in a specific topology. Using the voltage and current in the power frequency components as the port voltage and port current of the virtual synchronous machine equivalent circuit model means using the direct-axis voltage power frequency component and the quadrature-axis voltage power frequency component as the port voltage input of the virtual synchronous machine equivalent circuit model in the synchronous rotating coordinate system, and using the direct-axis current power frequency component and the quadrature-axis current power frequency component as the port current input of the virtual synchronous machine equivalent circuit model.

[0030] In practical implementation, a system of differential-algebraic equations for the equivalent circuit model of the virtual synchronous machine is written based on Kirchhoff's voltage and current laws. The differential equations describe the dynamic relationship between flux linkage and voltage in the inductor branch, while the algebraic equations describe the Ohm's law constraints and port connection relationships in the resistor branch. Numerical integration is used to solve the system of differential-algebraic equations to obtain the current and flux linkage in the inductor branch. The solution process is executed within a fixed control cycle. In practical implementation, discretizing the differential equations in the system of differential-algebraic equations into difference equations is the first step in numerical integration. For example, using the forward Euler method, the differential equation describing the rate of change of flux linkage is transformed into a linear equation concerning the flux linkage values ​​at the current and next time steps. A solution step size is set, typically less than or equal to the execution cycle of the entire control method. Based on the current and flux linkage values ​​of the inductor branch at the previous time step, the current and flux linkage values ​​of the inductor branch at the current time step are iteratively calculated.

[0031] In some embodiments, during each iteration, the solution of the algebraic equation system needs to be substituted into the difference equation. The algebraic equation system of the virtual synchronous machine equivalent circuit model is used to solve for the voltage and current relationships of all branches based on the currently known port voltages and the branch variables from the previous iteration. The voltage or current values ​​related to the inductor branch obtained from the algebraic equation system are substituted into the discretized difference equation to update the current and flux linkage values ​​of the inductor branch. The iteration process is repeated until the solution of the differential algebraic equation system converges. The convergence condition is that the changes in the current and flux linkage values ​​of the inductor branch calculated in two consecutive iterations are both less than a preset minimum threshold. The finally converged current and flux linkage values ​​of the inductor branch are output as the internal state variables of the virtual synchronous machine equivalent circuit model in the current control cycle. It can be understood that the above iterative solution process is performed independently in each control cycle, and the final converged state of the previous cycle is used as the initial value for the current iteration.

[0032] Optionally, the discretization process can employ different numerical integration schemes, and the update relation of the forward Euler method can be expressed by the following formula:

[0033] in: Representing the The flux linkage value of the inductor branch at the next iteration Representing the The flux linkage value of the inductor branch at the next iteration This represents the set solution step size time interval. Representative at the The voltage value applied across the inductor branch is obtained by solving a system of algebraic equations during the next iteration.

[0034] In some embodiments, the flux linkage of the inductor branch obtained by the solution is defined as the amplitude of the virtual internal potential. The phase angle of the virtual internal potential is obtained through the virtual rotor motion equation in the equivalent circuit model of the integrator virtual synchronous machine, or directly taken from the phase-locked angle of the power frequency component of the port voltage plus an additional angle generated by the power regulator. The difference between the phase angle of the port voltage and the phase angle of the virtual internal potential is defined as the virtual power angle, which is directly obtained from the result of phase-locked processing of the three-phase voltage waveform at the point of common coupling. It can be understood that the amplitude of the virtual internal potential and the virtual power angle together constitute the key variables describing the internal electrical state of the virtual synchronous machine. Optionally, in a more complex topology of the equivalent circuit model of the virtual synchronous machine, multiple inductor or capacitor elements may be included. In this case, the dimension of the differential algebraic equation system increases, but the core process of discretization and iterative solution using numerical integration methods remains consistent. The final output virtual internal potential and virtual power angle will be used as part of the input time series of the behavior prediction network.

[0035] In one embodiment of the present invention, the behavior prediction network is a temporal convolutional network pre-trained with historical operating data. Its input layer receives time-series data of virtual internal potential, virtual power angle, low-frequency components, and residual components, and its output layer outputs a multi-dimensional reference trajectory for the next control cycle. This multi-dimensional reference trajectory includes a reference active power trajectory, a reference reactive power trajectory, and a reference voltage amplitude trajectory for the next control cycle. The specific process of constructing the dynamic error surface includes: extracting the actual active power, actual reactive power, and actual voltage amplitude corresponding to the power frequency component at the current moment; calculating the difference between the actual active power and the reference active power trajectory at the trajectory's starting point as the instantaneous active power error; calculating the difference between the actual reactive power and the reference reactive power trajectory's starting point as the instantaneous reactive power error; and calculating the difference between the actual voltage amplitude and the reference voltage amplitude trajectory's starting point as the instantaneous voltage amplitude error. Using the instantaneous active power error, the instantaneous reactive power error, and the instantaneous voltage amplitude error as independent variables, a three-dimensional error space is constructed. In this error space, a surface is defined as a function of the sum of the squares of the three instantaneous errors; this surface is called the dynamic error surface.

[0036] In practical implementation, the behavioral prediction network is deployed in the central controller of the photovoltaic power plant, based on the time series of virtual internal potential and virtual power angle output from the equivalent circuit model of the virtual synchronous machine, as well as the time series of low-frequency components and residual components obtained after time-frequency domain hybrid decomposition. The behavioral prediction network is a temporal convolutional network pre-trained with historical operating data, which includes long-term collected data on various electrical and environmental quantities of the photovoltaic power plant. The temporal convolutional network is trained through supervised learning to learn the mapping relationship from the input state to the future output trajectory. In practical implementation, the input layer of the behavioral prediction network receives the time series of virtual internal potential, virtual power angle, low-frequency components, and residual components. These time series are aligned with fixed timestamps and organized into a multi-dimensional tensor format before being input into the temporal convolutional network. The output layer of the behavioral prediction network outputs a multi-dimensional reference trajectory for the next control cycle. The multi-dimensional reference trajectory is given in the form of a discrete time point sequence, covering the time window from the current moment to the end of a complete future control cycle.

[0037] In some embodiments, the multi-dimensional reference trajectory specifically includes a reference active power trajectory, a reference reactive power trajectory, and a reference voltage amplitude trajectory for future control cycles. The reference active power trajectory describes the active power change curve that the photovoltaic power station should follow over a future period, as predicted by the temporal convolutional network; the reference reactive power trajectory describes the reactive power change curve that should be followed; and the reference voltage amplitude trajectory describes the voltage amplitude change curve at the point of common coupling that should be followed. It can be understood that the starting point of the multi-dimensional reference trajectory corresponds to the reference value at the current control moment, while subsequent points on the trajectory correspond to the reference values ​​at various future sampling moments. Optionally, the structure of the temporal convolutional network includes causal dilated convolutional layers, residual connections, and fully connected output layers. The dilated convolutional design enables the network to capture long-term dependencies in the input time series.

[0038] In practical implementation, the process of constructing the dynamic error surface begins with extracting the actual electrical quantities corresponding to the power frequency components at the current moment. Power calculations are performed on the real-time acquired and decomposed power frequency components to obtain the actual active power, actual reactive power, and actual voltage amplitude at the current moment. The difference between the actual active power and the starting point of the reference active power trajectory is calculated as the instantaneous error of active power. The starting point of the reference active power trajectory is the reference active power value output by the time convolutional network corresponding to the current moment. The difference between the actual reactive power and the starting point of the reference reactive power trajectory is calculated as the instantaneous error of reactive power. The difference between the actual voltage amplitude and the starting point of the reference voltage amplitude trajectory is calculated as the instantaneous error of voltage amplitude. It can be understood that the instantaneous errors of active power, reactive power, and voltage amplitude collectively characterize the degree of deviation between the current operating state of the photovoltaic power station and the expected trajectory of the behavior prediction network from the starting point.

[0039] In some embodiments, a three-dimensional error space is constructed using the instantaneous errors of active power, reactive power, and voltage amplitude as independent variables. The three orthogonal coordinate axes of the error space represent the instantaneous errors of active power, reactive power, and voltage amplitude, respectively. In the three-dimensional error space, a surface is defined as a function of the sum of the squares of the three instantaneous errors; this surface is the dynamic error surface.

[0040] In one embodiment of the present invention, the specific process of solving the gradient direction of the dynamic error surface and generating a duty cycle correction sequence along that direction includes calculating the partial derivatives of the dynamic error surface corresponding to the instantaneous errors of active power, reactive power, and voltage amplitude. These three partial derivatives constitute a gradient vector, the direction of which represents the direction that causes the dynamic error surface value to decrease the fastest. This gradient vector is mapped onto the direct-axis and quadrature-axis voltage control components of the photovoltaic inverter in a synchronous rotating coordinate system to obtain the required voltage correction amount. Based on this voltage correction amount and the current DC bus voltage, the desired voltage space vector is calculated. Based on the desired voltage space vector, the nearest three-vector synthesis method is used to calculate the duration of each of the three basic voltage vectors within one pulse width modulation period. The duration of each basic voltage vector is divided by the pulse width modulation period to obtain the duty cycle correction sequence for the corresponding inverter arm.

[0041] In practical implementation, the function value of the dynamic error surface is determined by the instantaneous errors of active power, reactive power, and voltage amplitude. The partial derivatives of the dynamic error surface corresponding to these errors are calculated by differentiating the function expression of the dynamic error surface. A gradient vector is formed by the partial derivatives of the instantaneous errors of active power, reactive power, and voltage amplitude. The direction of the gradient vector represents the direction in which the function value of the dynamic error surface decreases the fastest. In practical implementation, the gradient vector is mapped onto the direct-axis and quadrature-axis voltage control components of the photovoltaic inverter. The mapping process relies on the static relationship model between the output power and voltage of the photovoltaic inverter. This model converts the direction of power error change into the correction direction of the voltage control quantity, thereby obtaining the direct-axis voltage correction and quadrature-axis voltage correction.

[0042] Based on the direct-axis voltage correction, quadrature-axis voltage correction, and the current DC bus voltage, the desired voltage space vector is calculated. The calculation process includes converting the direct-axis and quadrature-axis voltage correction components from the synchronous rotating coordinate system back to the stationary two-phase coordinate system. In practice, based on the desired voltage space vector, the nearest three-vector synthesis method is used. This method first determines the sector where the desired voltage space vector is located in the space vector diagram, then selects two adjacent basic voltage vectors and the zero vector as the three basic voltage vectors for synthesis. The duration of each of the three basic voltage vectors within one pulse width modulation (PWM) cycle is calculated. The duration calculation is based on the volt-second balance principle, ensuring that the weighted sum of the times of the three basic voltage vectors and the zero vector equals the desired voltage space vector within a single PWM cycle. Dividing the duration of each basic voltage vector by the PWM cycle yields the duty cycle correction sequence for the corresponding inverter arm. This duty cycle correction sequence directly determines the on and off times of each switching device in the next PWM cycle.

[0043] In some embodiments, the duration of action of the fundamental voltage vector in the most recent three-vector synthesis method can be obtained by solving a set of equations based on the components of the desired voltage space vector in a Cartesian coordinate system. The relationship between the components of the desired voltage space vector in the α-β coordinate system and the duration of action of the fundamental voltage vector can be expressed as:

[0044]

[0045] in: This represents the pulse width modulation period. and These represent the α-axis and β-axis components of the desired voltage space vector, respectively. and The α-axis and β-axis components represent the first fundamental voltage vector. and The α-axis and β-axis components represent the second fundamental voltage vector. and These represent the durations of the first and second fundamental voltage vectors, respectively. The third fundamental vector is the zero vector, and its duration is... It can be understood that the solution to the above system of equations provides the precise action time of the non-zero fundamental voltage vector used to synthesize the desired voltage space vector.

[0046] Optionally, the correspondence between the basic voltage vector and the operating time under different switching states can be represented by a simplified reference table, which clarifies the basic voltage vector sequence required after sector identification and its calculated duty cycle. For example, for a common three-phase two-level inverter, refer to Table 1 for the distribution of its basic voltage vector and operating time.

[0047] Table 1: Time Allocation Table for Basic Voltage Vector Action

[0048] In some embodiments, the process of converting the action time into a duty cycle correction sequence involves dividing the action time of each basic voltage vector by the pulse width modulation period to obtain a set of values ​​between 0 and 1. This set of values, after appropriate sorting and format conversion, forms a control word that can be directly sent to the pulse width modulation generator of a specific photovoltaic inverter. It can be understood that the generation of the duty cycle correction sequence is a key step in closed-loop control, enabling the output voltage of the photovoltaic inverter to track the desired voltage space vector determined by the gradient descent direction, thereby achieving coordinated and rapid adjustment of active power, reactive power, and voltage amplitude.

[0049] See Figure 4 This is a bar chart showing the basic vector action time distribution of voltage space vector sectors in the virtual synchronous control of a photovoltaic power plant. This chart belongs to the voltage space vector modulation stage of the virtual synchronous control of the photovoltaic power plant, and is used to visually present the time distribution rules of the basic vectors in different sectors. It is the core data support for achieving the "desired voltage space vector synthesis". By observing the differences in the vector action time of each sector, the selection strategy of the basic vector can be adjusted to improve the synthesis accuracy of the voltage space vector; based on the time distribution of each sector, the insertion method of the zero vector can be optimized to reduce inverter switching losses; if the vector action time of a certain sector deviates from the normal range, it can help locate anomalies in the voltage modulation stage. In one embodiment of the present invention, the specific steps for calculating the partial derivatives of the dynamic error surface corresponding to the instantaneous errors of active power, reactive power, and voltage amplitude are as follows: The functional expression of the dynamic error surface is differentiated with respect to the instantaneous error variables of active power, reactive power, and voltage amplitude, respectively. In the differentiation of the instantaneous error variable of active power, the instantaneous error variables of reactive power and voltage amplitude are treated as constants, thereby obtaining the partial derivative with respect to the instantaneous error of active power. Similarly, in the differentiation of the instantaneous error variable of reactive power, the instantaneous error variables of active power and voltage amplitude are treated as constants, thereby obtaining the partial derivative with respect to the instantaneous error of reactive power. Finally, in the differentiation of the instantaneous error variable of voltage amplitude, the instantaneous error variables of active power and reactive power are treated as constants, thereby obtaining the partial derivative with respect to the instantaneous error of voltage amplitude.

[0050] In practical implementation, the functional expression of the dynamic error surface is differentiated with respect to the instantaneous error variables of active power, reactive power, and voltage amplitude, respectively. The differentiation follows the rules for calculating partial derivatives of multivariable functions. Specifically, in the differentiation of the instantaneous error variable of active power, the instantaneous error variables of reactive power and voltage amplitude are treated as constants. This means that when differentiating the instantaneous error variable of active power, the terms in the functional expression containing these variables are considered constant terms, and their derivatives are zero, thus obtaining the partial derivatives with respect to the instantaneous error of active power.

[0051] In some embodiments, during the differentiation of the reactive power instantaneous error variable, the active power instantaneous error variable and the voltage amplitude instantaneous error variable are treated as constants. The terms in the function expression containing these variables contribute zero during differentiation, thus obtaining the partial derivative with respect to the reactive power instantaneous error. In a specific implementation, during the differentiation of the voltage amplitude instantaneous error variable, the active power instantaneous error variable and the reactive power instantaneous error variable are treated as constants. The terms in the function expression containing these variables are not considered during differentiation, thus obtaining the partial derivative with respect to the voltage amplitude instantaneous error. It can be understood that the above three independent differentiation processes are executed sequentially or in parallel, ultimately producing three scalar results as components of the gradient vector on the corresponding coordinate axes.

[0052] Optionally, the functional expression of the dynamic error surface is the sum of the squares of the instantaneous active power error variable, the squares of the instantaneous reactive power error variable, and the squares of the instantaneous voltage amplitude error variable, and its partial derivatives have a concise form. The partial derivative with respect to the instantaneous active power error variable is calculated as twice the instantaneous active power error variable, the partial derivative with respect to the instantaneous reactive power error variable is calculated as twice the instantaneous reactive power error variable, and the partial derivative with respect to the instantaneous voltage amplitude error variable is calculated as twice the instantaneous voltage amplitude error variable. In some embodiments, the calculation of partial derivatives is completed in real time in the arithmetic logic unit of the central controller, and the values ​​of the instantaneous active power error variable, the instantaneous reactive power error variable, and the instantaneous voltage amplitude error variable come from the real-time construction of the dynamic error surface. The differentiation operation itself can be implemented by directly applying the derivative formula in code, without the need for numerical difference approximation, because the functional form of the dynamic error surface is a known and simple quadratic polynomial. Optionally, when the function definition of the dynamic error surface takes other forms, such as including cross terms or weighting coefficients of different error variables, the calculation of partial derivatives needs to be re-derived based on the new function expression and the corresponding differentiation operation needs to be performed. However, the core differentiation rules and the principle of treating other variables as constants remain unchanged. The three partial derivatives finally calculated will be directly used to assemble the gradient vector.

[0053] See Figure 5 This is a multi-dimensional trajectory comparison radar chart for virtual synchronous control of a photovoltaic power station. This chart is part of the performance evaluation of virtual synchronous control in photovoltaic power stations, used to visually present the degree of fit between the control effect and the target trajectory. It is a core tool for verifying the effectiveness of the virtual synchronous control algorithm. By measuring the overlap between the actual trajectory and the reference trajectory, the tracking accuracy of the virtual synchronous control is quantified; dimensions with significant differences between the actual and reference trajectories are identified as key areas for adjusting control algorithm parameters; if the actual trajectory continues to deviate from the reference trajectory, anomalies in the control process can be detected. The spatial difference between the "actual trajectory" and the "reference trajectory" visually quantifies the degree of deviation between the control effect and the ideal target, avoiding the abstractness of purely numerical comparisons.

[0054] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A virtual synchronous control method for a photovoltaic power station, characterized in that, include: Acquire multi-source datasets of photovoltaic power plant operation, which include instantaneous electrical quantities of the DC and AC sides of the photovoltaic inverter, three-phase voltage waveforms at the point of common coupling, and light intensity and temperature recorded by environmental monitoring devices; The multi-source dataset is subjected to time-frequency domain hybrid decomposition to separate the low-frequency component representing a slow changing trend, the power frequency component representing periodic fluctuations, and the residual component representing instantaneous disturbances. The power frequency component is injected into the pre-constructed virtual synchronous machine equivalent circuit model, which consists of an inductor branch simulating mechanical inertia and a resistor branch simulating damping characteristics. The virtual internal potential and virtual power angle are obtained by solving the circuit. The virtual internal potential, the virtual power angle, the low-frequency component, and the residual component are fed into the behavior prediction network, and the behavior prediction network outputs a multi-dimensional reference trajectory for the next control cycle. Based on the multi-dimensional reference trajectory and the real-time power frequency components, a dynamic error surface is constructed; Solve for the gradient direction of the dynamic error surface, and generate a set of duty cycle correction sequences for pulse width modulation waves along the gradient direction; The duty cycle correction sequence is sent to the pulse width modulation generator of each photovoltaic inverter in the photovoltaic power station to adjust the switching timing of each photovoltaic inverter.

2. The virtual synchronization control method for a photovoltaic power station according to claim 1, characterized in that, The time-frequency domain hybrid decomposition of the multi-source dataset specifically includes: The three-phase voltage waveform is subjected to phase-locked processing to obtain the synchronous rotation angle; Using the synchronous rotation angle, the instantaneous electrical quantity is transformed from the stationary coordinate system to the synchronous rotating coordinate system to obtain the direct axis component and the quadrature axis component; Sliding time windows are applied to the direct-axis components and the quadrature-axis components respectively, and empirical mode decomposition is performed within each time window to obtain a series of intrinsic mode functions arranged from high frequency to low frequency; The intrinsic mode functions with frequencies higher than a preset threshold are classified as the residual components; The eigenmode functions with frequencies between a preset frequency range are reconstructed to obtain the power frequency components; The low-frequency component is obtained by superimposing the intrinsic mode functions with frequencies below a preset threshold and the trend term.

3. The virtual synchronization control method for a photovoltaic power station according to claim 2, characterized in that, The process involves injecting the power frequency component into a pre-constructed virtual synchronous machine equivalent circuit model. This virtual synchronous machine equivalent circuit model consists of an inductor branch simulating mechanical inertia and a resistor branch simulating damping characteristics. Virtual internal potential and virtual power angle are obtained through circuit solving. Specifically, this includes: The voltage and current in the power frequency component are used as the port voltage and port current of the equivalent circuit model of the virtual synchronous machine. Based on Kirchhoff's voltage law and Kirchhoff's current law, list the system of differential-algebraic equations for the equivalent circuit model of the virtual synchronous machine. The differential-algebraic equations are solved using numerical integration to obtain the current and flux linkage of the inductor branch. The flux linkage of the inductor branch is defined as the amplitude of the virtual internal potential; The difference between the phase angle of the port voltage and the phase angle of the virtual internal potential is defined as the virtual power angle.

4. The virtual synchronization control method for a photovoltaic power station according to claim 3, characterized in that, The behavior prediction network is a temporal convolutional network pre-trained with historical operating data. Its input layer receives the time series of the virtual internal potential, virtual power angle, low-frequency component and residual component, and its output layer outputs the multi-dimensional reference trajectory. The multi-dimensional reference trajectory includes the reference active power trajectory, the reference reactive power trajectory, and the reference voltage amplitude trajectory for future control cycles.

5. The virtual synchronization control method for a photovoltaic power station according to claim 4, characterized in that, The construction of the dynamic error surface specifically includes: Extract the actual active power, actual reactive power, and actual voltage amplitude corresponding to the power frequency component at the current moment; The difference between the actual active power and the starting point of the reference active power trajectory is calculated as the instantaneous error of active power. The difference between the actual reactive power and the starting point of the reference reactive power trajectory is calculated as the instantaneous reactive power error. The difference between the actual voltage amplitude and the starting point of the reference voltage amplitude trajectory is calculated as the instantaneous error of the voltage amplitude. Using the instantaneous errors of active power, reactive power, and voltage amplitude as independent variables, a three-dimensional error space is constructed. In the error space, a surface is defined as a function of the sum of the squares of the three instantaneous errors; this surface is the dynamic error surface.

6. The virtual synchronization control method for a photovoltaic power station according to claim 5, characterized in that, The process of solving the gradient direction of the dynamic error surface and generating a set of duty cycle correction sequences for pulse width modulation waves along the gradient direction specifically includes: Calculate the partial derivatives of the dynamic error surface corresponding to the instantaneous errors of active power, reactive power, and voltage amplitude; The partial derivatives form a gradient vector, and the direction of the gradient vector is the direction that makes the dynamic error surface decrease the fastest. The gradient vector is mapped onto the direct-axis voltage and quadrature-axis voltage control components of the photovoltaic inverter to obtain the voltage correction amount; Calculate the desired voltage space vector based on the voltage correction amount and the DC bus voltage; Based on the desired voltage space vector, the action time of each of the three basic voltage vectors is calculated using the nearest three-vector synthesis method. Divide the duration of each basic voltage vector by the pulse width modulation period to obtain the duty cycle correction sequence for the corresponding bridge arm.

7. The virtual synchronization control method for a photovoltaic power station according to claim 1, characterized in that, After obtaining the multi-source dataset of the photovoltaic power plant operation, the process also includes a data credibility verification step: Cross-compare similar data from different monitoring points. If the data deviation exceeds the tolerance, an alternative strategy based on interpolation of adjacent data is initiated. Calculate the noise energy of each data channel in the multi-source dataset, and perform noise reduction filtering processing based on wavelet threshold for the channel data with noise energy exceeding the threshold.

8. The virtual synchronization control method for a photovoltaic power station according to claim 3, characterized in that, The method of solving the system of differential-algebraic equations using numerical integration to obtain the current and flux linkage of the inductor branch specifically includes: Discretize the differential equations in the system of differential algebraic equations into difference equations; Set the solution step size, and iteratively calculate the current and flux linkage values ​​of the inductor branch at the current moment based on the values ​​of the current and flux linkage of the inductor branch at the previous moment. In each iteration, the solution of the algebraic equation system is substituted into the difference equation to update the values ​​of the current and flux linkage of the inductor branch. The iterative process is repeated until the solution of the differential-algebraic equation system converges, and the values ​​of the current and flux linkage of the inductor branch that have finally converged are output.

9. The virtual synchronization control method for a photovoltaic power station according to claim 6, characterized in that, The calculation of the partial derivatives of the dynamic error surface corresponding to the instantaneous errors of active power, reactive power, and voltage amplitude specifically includes: The functional expression of the dynamic error surface is differentiated with respect to the instantaneous error variables of active power, reactive power, and voltage amplitude, respectively. In the process of differentiating the instantaneous error variable of active power, the instantaneous error variable of reactive power and the instantaneous error variable of voltage amplitude are regarded as constants to obtain the partial derivative with respect to the instantaneous error of active power. In the process of differentiating the instantaneous error variable of reactive power, the instantaneous error variable of active power and the instantaneous error variable of voltage amplitude are regarded as constants to obtain the partial derivative with respect to the instantaneous error of reactive power; In the process of differentiating the instantaneous error variable of voltage amplitude, the instantaneous error variables of active power and reactive power are treated as constants to obtain the partial derivative with respect to the instantaneous error of voltage amplitude.

10. A virtual synchronous control system for a photovoltaic power station, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the virtual synchronous control method for a photovoltaic power station as described in any one of claims 1 to 9.

Images