Microgrid double-layer distributed data acquisition and control system for wide frequency oscillation

Abstract

The application relates to the technical field of micro-grid control systems, and particularly discloses a micro-grid double-layer distributed data acquisition control system for coping with wide-frequency oscillation, which comprises a bottom-layer data sensing and physical control layer, a top-layer distributed cooperative scheduling layer and a double-layer power instruction feedforward fusion unit connecting the two layers. The technical scheme of the application realizes decoupled operation of fundamental wave steady-state control and wide-frequency oscillation suppression in the actual operation process of the micro-grid through the double-layer structure of the bottom-layer data sensing and physical control layer and the top-layer distributed cooperative scheduling layer. In the bottom-layer sensing stage, a hardware-level self-adaptive band-pass decoupling unit separates real-time current for on-site damping response and actual current for steady-state cooperative calculation from the composite current, the oscillation data interference on the fundamental wave power control link is isolated, and the on-site wide-frequency damping control unit can output basic damping instructions based on the real-time current.

Description

Technical Field

[0001] This invention relates to the field of microgrid control system technology, and in particular to a two-layer distributed data acquisition and control system for microgrids that can cope with broadband oscillations. Background Technology

[0002] Existing microgrid control systems primarily employ centralized or single distributed control architectures to achieve power output and regulation. Traditional control schemes typically collect the terminal voltage and output current of inverter nodes, extract power characteristics through a unified data processing link, and rely on droop control strategies to regulate the fundamental voltage and frequency, as well as allocate active and reactive power. In this architecture, data acquisition and control command processing usually occur within the same control loop.

[0003] In practical applications where a high proportion of power electronic devices are connected to microgrids, the system may experience broadband oscillations under certain operating conditions. Existing technologies fail to effectively decouple the fundamental power flow signal from the broadband oscillation signal during the data sensing phase. This results in composite current signals containing oscillating components directly entering the steady-state power control calculation stage, causing distortion of the coordinated control commands.

[0004] Meanwhile, when broadband oscillations occur, systems relying on a single control layer lack a global scheduling mechanism for oscillation characteristic signals, making it difficult to provide targeted damping power support while maintaining the fundamental power distribution of the microgrid, which in turn leads to fluctuations in the operating status of the microgrid system. Summary of the Invention

[0005] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a two-layer distributed data acquisition and control system for microgrids to cope with broadband oscillations, thereby improving the overall stability control capability of microgrid systems under broadband oscillation conditions.

[0006] To achieve the above objectives, the first aspect of the present invention proposes a two-layer distributed data acquisition and control system for microgrids to cope with broadband oscillations, including a bottom data sensing and physical control layer, a top distributed collaborative scheduling layer, and a two-layer power command feedforward fusion unit connecting the two layers.

[0007] The underlying data sensing and physical control layer collects the composite current signal of the inverter in the microgrid, and extracts the real-time current containing a broadband oscillation component and the actual current containing only the power frequency fundamental component from the composite current signal through a hardware-level adaptive bandpass decoupling unit; the underlying data sensing and physical control layer also includes a local broadband damping control unit, which generates active power damping command and reactive power damping command based on the real-time current and the instantaneous voltage of the port.

[0008] The top-level distributed collaborative scheduling layer obtains the actual current, fundamental voltage phasor, and oscillating slow characteristic quantity of each node in the microgrid based on the distributed communication network; the fundamental power collaborative calculation unit in the top-level distributed collaborative scheduling layer outputs steady-state active power collaborative command and steady-state reactive power collaborative command, and the distributed damping gain scheduling unit outputs the damping scheduling coefficient of each node based on the oscillating slow characteristic quantity.

[0009] The dual-layer power command feedforward fusion unit receives the steady-state active power coordination command, the steady-state reactive power coordination command, the active power damping command, and the reactive power damping command, and generates inverter output voltage reference value and angular frequency reference value through orthogonal superposition, and injects them into the inverter's droop control main circuit.

[0010] To achieve the above objectives, a second aspect of the present invention proposes a two-layer distributed data acquisition and control method for microgrids to cope with broadband oscillations, applied to a two-layer distributed data acquisition and control system for microgrids to cope with broadband oscillations. The method includes the following steps:

[0011] Step S1: Collect the composite current signal of the inverter in the microgrid through the underlying data sensing and physical control layer, and extract the real-time current containing broadband oscillation components and the actual current containing only power frequency fundamental components from the composite current signal through the hardware-level adaptive bandpass decoupling unit in the underlying data sensing and physical control layer.

[0012] Step S2: The local broadband damping control unit in the underlying data sensing and physical control layer generates active damping commands and reactive damping commands based on the real-time current and instantaneous port voltage.

[0013] Step S3: The actual current, fundamental voltage phasor, and slow oscillation characteristic of each node in the microgrid are obtained through the top-level distributed collaborative scheduling layer based on the distributed communication network.

[0014] Step S4: The steady-state active power coordination command and steady-state reactive power coordination command are output by the fundamental power coordination calculation unit in the top-level distributed coordination scheduling layer, and the damping scheduling coefficient of each node is output by the distributed damping gain scheduling unit in the top-level distributed coordination scheduling layer based on the oscillation slow characteristic quantity.

[0015] Step S5: The steady-state active power coordination command, the steady-state reactive power coordination command, the active power damping command, and the reactive power damping command are received by the dual-layer power command feedforward fusion unit. The output voltage reference value and the angular frequency reference value of the inverter are generated by orthogonal superposition, and the output voltage reference value and the angular frequency reference value are injected into the droop control main circuit of the inverter.

[0016] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described microgrid dual-layer distributed data acquisition and control method for coping with broadband oscillations.

[0017] The technical solution of this invention achieves decoupled operation of fundamental steady-state control and broadband oscillation suppression in the actual operation of microgrids through a two-layer architecture of bottom-layer data sensing and physical control layer and top-layer distributed collaborative scheduling layer.

[0018] At the underlying sensing stage, a hardware-level adaptive bandpass decoupling unit separates the real-time current used for local damping response and the actual current used for steady-state collaborative calculation from the composite current. This isolates the interference of oscillation data on the fundamental power control loop, enabling the local broadband damping control unit to output basic damping commands based on the real-time current.

[0019] In the top-level coordination stage, the distributed coordination scheduling layer utilizes the actual current and slow oscillation characteristics extracted from each node to simultaneously realize the allocation of fundamental steady-state power and the calculation of damping scheduling coefficients for each node. Finally, through the dual-layer power command feedforward fusion unit, the steady-state coordination command and damping command are orthogonally superimposed, enabling the inverter to output damped power that combines global coordination scheduling characteristics while ensuring the fundamental voltage, frequency regulation and power distribution of the entire network. This improves the overall stability control capability of the microgrid system in dealing with wide-frequency oscillation conditions. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the implementation of the microgrid dual-layer distributed data acquisition and control system for coping with broadband oscillations provided by the present invention;

[0021] Figure 2 This is a diagram of composite current signal and adaptive decoupling characteristic waveform data in the microgrid dual-layer distributed data acquisition and control system for dealing with broadband oscillations provided by the present invention;

[0022] Figure 3 This is a time-frequency evolution bitmap of slow feature extraction of broadband oscillation in a microgrid dual-layer distributed data acquisition and control system for dealing with broadband oscillations provided by the present invention;

[0023] Figure 4 This is a comparison chart of the pre-damping compensation tracking of the LSTM prediction subunit in the microgrid dual-layer distributed data acquisition and control system for dealing with broadband oscillations provided by this invention.

[0024] Figure 5 This is a dynamic threshold and state deviation interaction diagram of the time-delay elastic event triggering mechanism in the microgrid dual-layer distributed data acquisition and control system for dealing with broadband oscillations provided by the present invention.

[0025] Figure 6 This is a three-dimensional surface diagram of the nonlinear mapping between time delay compensation weights and communication link characteristics in the microgrid dual-layer distributed data acquisition and control system for dealing with broadband oscillations provided by the present invention;

[0026] Figure 7 This is a dynamic saturation limiting waveform diagram of damping power considering AC / DC physical constraints in the microgrid dual-layer distributed data acquisition and control system for dealing with broadband oscillations provided by the present invention.

[0027] Figure 8 This is an adaptive reconstructed data graph of the global damping scheduling coefficient based on the margin penalty factor in the microgrid dual-layer distributed data acquisition and control system for coping with broadband oscillations provided by the present invention.

[0028] Figure 9 This is a flowchart illustrating the microgrid dual-layer distributed data acquisition and control method for dealing with broadband oscillations provided by the present invention.

[0029] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0030] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0031] The following description, with reference to the accompanying drawings, describes a dual-layer distributed data acquisition control system, method, and electronic device for microgrids to cope with broadband oscillations according to embodiments of the present invention.

[0032] Example 1:

[0033] In the current stage of power system development, with the high proportion of renewable energy and the massive integration of power electronic conversion equipment into microgrids, the source-grid-load-storage structure of microgrids exhibits highly nonlinear and time-varying characteristics. This change in physical characteristics makes the system highly susceptible to broadband oscillations ranging from several hertz to several kilohertz between the inverter control loop and the grid impedance under certain disturbances.

[0034] To improve the stability control capability of microgrids under complex operating conditions, such as Figure 1 As shown, the control system disclosed in this embodiment includes a bottom-level data perception and physical control layer, a top-level distributed collaborative scheduling layer, and a two-layer power command feedforward fusion unit that connects the two layers.

[0035] The underlying data sensing and physical control layer collects the composite current signal of the inverter in the microgrid, and extracts the real-time current containing a broadband oscillation component and the actual current containing only the power frequency fundamental component from the composite current signal through a hardware-level adaptive bandpass decoupling unit; the underlying data sensing and physical control layer also includes a local broadband damping control unit, which generates active power damping command and reactive power damping command based on the real-time current and the instantaneous voltage of the port.

[0036] The top-level distributed collaborative scheduling layer obtains the actual current, fundamental voltage phasor, and oscillating slow characteristic quantity of each node in the microgrid based on the distributed communication network; the fundamental power collaborative calculation unit in the top-level distributed collaborative scheduling layer outputs steady-state active power collaborative command and steady-state reactive power collaborative command, and the distributed damping gain scheduling unit outputs the damping scheduling coefficient of each node based on the oscillating slow characteristic quantity.

[0037] The dual-layer power command feedforward fusion unit receives the steady-state active power coordination command, the steady-state reactive power coordination command, the active power damping command, and the reactive power damping command, and generates inverter output voltage reference value and angular frequency reference value through orthogonal superposition, and injects them into the inverter's droop control main circuit.

[0038] Optionally, in the underlying physical equipment of a microgrid, the control core of the inverter is typically a heterogeneous computing platform composed of a digital signal processor and a field-programmable gate array.

[0039] In this embodiment, the hardware-level adaptive bandpass decoupling unit employs an adjustable bandwidth second-order generalized integrator array, implemented in parallel hardware via a field-programmable gate array. Because the current signal in a microgrid contains extremely complex frequency components, traditional software filtering algorithms are limited by the instruction cycle of the digital signal processor, making it difficult to achieve high-precision separation of the entire frequency band at the microsecond level.

[0040] This embodiment leverages the hardware parallel processing advantages of field-programmable gate arrays (FPGAs) to construct multiple parallel second-order generalized integrator modules, forming an arrayed processing structure. Each second-order generalized integrator module contains two mutually orthogonal internal feedback loops, enabling zero steady-state error tracking of AC signals at a specific resonant frequency.

[0041] During signal decoupling, the system dynamically adjusts the sampling rate and passband range for the real-time current according to the microgrid's system operating condition level, ensuring that the output data for the actual current always maintains a preset fixed narrowband output. This hardware-software co-design allows the system to physically separate two independent data streams:

[0042] One is the actual current channel for conventional fundamental power control, whose cutoff frequency is strictly limited to near the power frequency to filter out all high-frequency switching ripple and mid-frequency oscillation components, ensuring the purity of the steady-state control signal.

[0043] The other is a real-time current channel for broadband oscillation suppression, which has a wider frequency response window to capture sudden broadband oscillations in the power grid without attenuation, providing real physical feedback for subsequent damping control.

[0044] like Figure 2 As shown, based on the simulation test results of the system encountering a wideband disturbance at 0.2 seconds, the given composite current signal and adaptive decoupling characteristic waveform data graph contains three sets of sub-graphs arranged vertically. The horizontal axis of all sub-graphs is time, and the vertical axis is current amplitude.

[0045] The blue composite current curve at the top of the figure maintains a reference sinusoidal shape with an amplitude of approximately 100 amperes until 0.2 seconds later. After being disturbed at 0.2 seconds, the curve is superimposed with multiple frequency components, resulting in significant distortion and high-frequency oscillation of the envelope.

[0046] The red curve in the middle of the figure shows the actual current of the fundamental frequency wave at the bottom layer of the hardware-level adaptive bandpass decoupling unit. When the disturbance is cut in at 0.2 seconds, the red curve shows a transient overshoot of about 5 amps in a short time interval of 0.2 seconds to 0.25 seconds. Then it quickly converges and recovers to the smooth reference sine trajectory. This shows that the decoupling mechanism can effectively separate and ensure the stable operation of the fundamental frequency control signal. It also objectively reflects the short-term transient delay defect of the digital filter under step change conditions.

[0047] The green real-time current curve of the broadband oscillation at the bottom of the figure represents the transient component separated by decoupling. This green curve rapidly surges from a flat state to an oscillation peak of nearly 20 amperes at 0.2 seconds, reconstructing the dynamic decay process of the broadband disturbance and providing direct data support for local broadband damping control. Furthermore, the green curve still retains a slight fluctuation of about 1 ampere before the 0.2-second stable section, truly presenting the unavoidable signal residue phenomenon caused by the limitation of hardware calculation accuracy in industrial sites, reserving technical improvement dimensions for subsequent filter order upgrades and noise reduction algorithm optimization.

[0048] For example, the system divides the microgrid's system operating conditions into three levels: Level 1 steady-state operating condition, Level 2 fluctuating operating condition, and Level 3 oscillating operating condition. The hardware acquisition parameters of the underlying data sensing and physical control layer, the damping gain of the local broadband damping control unit, and the iteration cycle and communication scheduling strategy of the top-level distributed collaborative scheduling layer are all synchronously and adaptively adapted to the switching of the system operating conditions.

[0049] In actual microgrid operation monitoring, the classification of operating conditions can be set as the total harmonic distortion rate of the inverter port output voltage and the rate of change of the fundamental frequency.

[0050] When the microgrid is in the first level steady-state condition, that is, when the voltage harmonic distortion rate and frequency fluctuation are both within the normal operating threshold, the sampling of the real-time current adopts the first sampling rate and the first passband range, and the nonlinear consensus power flow algorithm adopts the first iteration step size.

[0051] When the system is under the second level of fluctuation, that is, when the system is affected by sudden changes in external load or intermittent fluctuations in the output of new energy sources, and the voltage or frequency begins to show a tendency to exceed the limit, but has not yet formed a continuous divergent oscillation, the sampling of the real-time current adopts the second sampling rate and the second passband range, and the long short-term memory network oscillation prediction subunit is activated to perform pre-damped feedforward, and the nonlinear consensus power flow algorithm adopts the second iteration step size.

[0052] When the system is in the third-level oscillation condition, i.e., when the system has experienced significant wideband resonance and the current waveform exhibits severe envelope distortion, a third sampling rate and a third passband range are used for sampling the real-time current, and the nonlinear consensus power flow algorithm uses a third iteration step size; wherein, the first sampling rate is less than the second sampling rate, and the second sampling rate is less than the third sampling rate; the first passband range is less than the second passband range, and the second passband range is less than the third passband range; and the first iteration step size is greater than the second iteration step size.

[0053] Specifically, the design basis for the above operating parameters is as follows: Under the first steady-state operating condition, the system does not need to suppress high-frequency oscillations. Using a smaller first sampling rate and first passband range can significantly reduce the power consumption of the analog-to-digital converter and the computing bus load of the microprocessor. At the same time, using a larger first iteration step size can reduce the data interaction frequency of the top-level distributed communication network and save communication bandwidth.

[0054] When entering the second level of fluctuation condition, the system gradually widens the signal observation window by increasing the sampling rate and the passband range to the second level, and then reduces the step size to the second iteration size to accelerate the update frequency of control commands and prevent the fluctuation from worsening further.

[0055] When entering the third-level oscillation condition, the wideband oscillation frequency may reach several kilohertz. At this time, the system must perform full-load sampling at the highest third sampling rate and extend the passband range to the third passband range to cover the possible subsynchronous, supersynchronous and high-frequency harmonic oscillation bands.

[0056] Meanwhile, since the system is in a critical state with extremely low stability margin at this time, the nonlinear consistent power flow algorithm must use the smallest third iteration step size to ensure that the steady-state coordination command and damping command can be output in the finest granular smoothness on the time axis, and prevent secondary disturbances to the microgrid due to excessive control step size.

[0057] It should also be noted that the suppression of wideband oscillations not only depends on real-time electrical quantity control, but also requires quantitative analysis of the macroscopic evolution trend of the oscillation signal.

[0058] In this embodiment, the underlying data sensing and physical control layer further includes an oscillation slow feature extraction unit; the oscillation slow feature extraction unit obtains the oscillation residual current by subtracting the real-time current from the actual current, and extracts the dominant oscillation frequency, oscillation energy level and oscillation power flow angle from the oscillation residual current based on a set sliding time window, and packages them as the oscillation slow feature quantity along with the fundamental steady-state data and uploads them to the distributed communication network. The fundamental steady-state data at least includes the actual current and the fundamental voltage phasor.

[0059] In this step, the oscillating residual current essentially represents the abnormal broadband disturbance component remaining after completely stripping away the conventional integer harmonics such as the fifth and seventh harmonics and the 50 Hz or 60 Hz fundamental frequency. By creating a sliding time window through a fixed-length first-in-first-out queue in the microprocessor memory, the system can perform fast Fourier transform or matrix beam analysis on the oscillating residual current over a historical period.

[0060] Among them, the dominant oscillation frequency refers to the frequency component with the highest amplitude in the spectrum analysis; the oscillation energy level is obtained by integrating the square of the residual current amplitude within the sliding time window, which is used to characterize the potential destructive force of the oscillation on the physical equipment; the oscillation power flow angle is obtained by calculating the phase difference between voltage and current at the same characteristic frequency, and this parameter determines whether the current inverter node is the source or the victim of broadband oscillation interference.

[0061] By using these three types of features as slow oscillation features, the top-level scheduling network can grasp the oscillation status of all nodes in the network without receiving massive amounts of high-frequency sampling waveforms from the bottom layer, and can do so through a low-bandwidth network.

[0062] like Figure 3 As shown, combining broadband oscillation residual data collected and separated from the microgrid's underlying layer, the time-frequency evolution bitmap extracted from this broadband oscillation slow characteristic intuitively displays the transient physical evolution process of the microgrid system within a 0.6-second observation period. The horizontal axis of the bitmap represents time, scaled in seconds, and the vertical axis represents frequency, scaled in Hertz. The color depth in the bitmap transitions from a cool blue to a warm red, corresponding to the energy levels in the legend, objectively reflecting the transient energy intensity at different time-frequency nodes.

[0063] As can be seen from the thermal distribution in the figure, at 0.1 seconds, the system first excited a low-frequency broadband oscillation. This dominant oscillation band showed a stable horizontal evolution trend over time and initially exhibited a deep red clustering state representing high energy density. Then, at 0.3 seconds, a high-frequency oscillation of 500 Hz was excited a second time. This high-frequency oscillation band showed a downward frequency drift trend over time, and during this process, the energy level remained in a high deep red clustering state.

[0064] The superimposed white dashed box in the figure accurately reproduces the boundary trajectory of the sliding time window with a length of 0.04 seconds set in the system's underlying control. By continuously advancing this sliding time window along the time axis, the oscillation slow feature extraction unit can effectively lock the dominant oscillation frequency with the highest energy level within the current time slice, thereby reducing the dimensionality of the dense original waveform data into slowly varying feature values ​​that can be transmitted over low-bandwidth networks, effectively reducing the communication channel load of the top-level distributed collaborative scheduling layer.

[0065] Meanwhile, from the edge distribution of high-energy color blocks in the bitmap, it can be observed that, affected by the digital truncation effect of the fixed-length sliding time window, obvious frequency spectrum leakage and energy tailing phenomena appear at the upper and lower frequency band edges of the energy accumulation area. In addition, light blue environmental measurement noise is widely distributed in the background area. This reflects that the current fixed-window-length feature extraction algorithm still has certain measurement ambiguity under the mutual constraints of time resolution and frequency resolution. This proves that when the system is to deal with more complex high-dynamic broadband disturbance conditions in the future, there is still room for technical upgrades and optimizations in introducing adaptive variable window length control for the time window function and the underlying digital noise reduction algorithm.

[0066] Optionally, at the local physical control level, in order to achieve a fast response to oscillations, the local broadband damping control unit includes a local adaptive damping subunit and a long short-term memory network oscillation prediction subunit; the local adaptive damping subunit dynamically adjusts the local active power damping gain and the local reactive power damping gain according to the real-time oscillation amplitude; the long short-term memory network oscillation prediction subunit outputs an oscillation growth rate factor based on the time-series data composed of the oscillation residual current, thereby generating a pre-damping compensation amount; the local broadband damping control unit combines the damping scheduling coefficient, the local active power damping gain, the local reactive power damping gain, and the pre-damping compensation amount to generate the final total active power damping gain and the total reactive power damping gain, for use in outputting the active power damping command and the reactive power damping command.

[0067] In engineering applications, Long Short-Term Memory (LSTM) networks, comprising input, forget, and output gates, effectively overcome the gradient vanishing problem in traditional recurrent neural networks when processing long-sequence current signals. The oscillation prediction subunit of the LSM network is pre-trained offline using a large number of microgrid broadband oscillation fault waveforms and deployed online in the underlying digital signal processor. Its input is the oscillation residual current sequence from multiple past grid cycles, and its output is the growth rate factor of the oscillation amplitude within the next few milliseconds. If an exponential increase in oscillation amplitude is predicted, this subunit generates a forward-looking pre-damping compensation amount in advance, thereby overcoming the inherent physical delays in hardware measurement and instruction computation.

[0068] like Figure 4 As shown, based on the dynamic response test data of the microgrid's underlying local broadband damping control unit when encountering sudden disturbances, the tracking comparison chart of the pre-damped compensation of the long short-term memory network prediction subunit objectively demonstrates the transient execution characteristics of the system in the feedforward prediction stage. The horizontal axis of the attached chart represents time, scaled in seconds, and the vertical axis represents oscillation amplitude, scaled in amperes.

[0069] In the figure, the blue solid line represents the actual oscillation amplitude envelope curve obtained by the underlying hardware acquisition and fitting, and the red dashed line represents the prediction predamping compensation curve output online by the prediction subunit of the long short-term memory network.

[0070] As can be seen from the waveform evolution process, at 0.1 seconds, the microgrid is subjected to external disturbances that trigger broadband oscillations. The blue actual oscillation amplitude curve rises rapidly and reaches a peak of about 15 amperes in a very short time.

[0071] The red prediction curve begins to show a clear upward trend around 0.095 seconds, achieving a lead prediction of approximately five milliseconds on the time axis. This lead characteristic in time phase confirms the effectiveness of the Long Short-Term Memory Network prediction subunit in outputting the oscillation growth rate factor based on historical residual time-series data. It provides physical operational margin for the underlying data sensing and physical control layer to generate forward-looking pre-damping compensation, thereby compensating to some extent for the inherent latency of inverter hardware measurement and control command calculation.

[0072] Meanwhile, observing the high frequency of broadband oscillations in the red dashed line from 0.1 to 0.15 seconds reveals that the prediction curve not only exhibits an overshoot peak of approximately 2 amperes, but also shows continuous high-frequency numerical fluctuations and glitches along the falling edge of the curve. This objectively reflects that the generalization accuracy and output smoothness of the current offline-trained neural network model are still limited when facing highly dynamic and nonlinear abrupt changes. This indicates that there is still substantial room for optimization and upgrading of the online weight dynamic fine-tuning algorithm of the neural network and the post-smoothing filtering stage of the prediction results in future practical engineering applications of this control system.

[0073] To define the internal gain fusion logic of the local broadband damping control unit, a specific mathematical model for calculating the total gain is given here, as shown in the following formula:

[0074] ;

[0075] ;

[0076] In the formula, This represents the total active power damping gain; This represents the local active power damping gain; This represents the damping scheduling coefficient issued by the top level; This represents the pre-damping compensation amount generated by the oscillation prediction subunit of the Long Short-Term Memory Network for the active channel; This represents the total reactive power damping gain. This represents the local reactive power damping gain; This represents the pre-damping compensation amount generated by the oscillation prediction subunit of the long short-term memory network for the reactive channel.

[0077] The above formula clearly demonstrates the linear superposition and multiplicative modulation relationship between local adaptive control, top-level scheduling constraints, and neural network feedforward prediction in the final damping gain generation process. This is achieved through the damping scheduling coefficient. The multiplicative modulation effect allows top-level scheduling instructions to directly scale the damping response intensity of the bottom layer.

[0078] For example, for the global coordinated control of the top layer of the microgrid, the fundamental power coordinated calculation unit adopts a nonlinear consistent power flow algorithm based on graph theory. The nonlinear consistent power flow algorithm takes the minimum global voltage deviation, the minimum frequency deviation, and the minimum active and reactive power sharing error as optimization objectives, and the iteration step size and iteration period of the nonlinear consistent power flow algorithm are dynamically adjusted according to the system operating condition level of the microgrid.

[0079] In traditional droop control, the difference in line impedance between each node and the common connection point often leads to a drop in output voltage and frequency when the microgrid is islanded, and active and reactive power cannot be accurately and evenly distributed among multiple parallel inverters according to capacity.

[0080] This embodiment introduces a nonlinear consensus power flow algorithm based on multi-agent graph theory, abstracting each inverter in the microgrid as a node in a graph-based communication network. Each node only needs to exchange the fundamental steady-state data with its adjacent nodes in the communication topology. Through distributed iterative computation, the states of all nodes (such as virtual voltage per unit value and active power output ratio) are made to converge to a globally unified reference value.

[0081] This algorithm eliminates the need for a central controller to collect information from the entire network, effectively avoiding the risk of system crashes caused by single points of failure and improving the robustness of the control layer.

[0082] For example, to clarify the nonlinear consistency state update logic based on graph theory, the formula is defined as follows:

[0083] ;

[0084] In the formula, Representing the The node at the th The local cooperative state quantity after the next iteration is specifically mapped to the base value of the steady-state active power cooperative command or steady-state reactive power cooperative command that needs to be generated subsequently. Representing the The node at the th Local cooperative state variables at the next iteration; This represents the iteration step size under the current operating condition, namely one of the aforementioned first iteration step size, second iteration step size, or third iteration step size; accumulation symbol Covering the microgrid communication topology and the first All adjacent nodes of the nth node One node; Represents the weight elements in the adjacency matrix of the communication topology. If the node With nodes If a direct communication link exists, the value is 1 (a normal number); otherwise, the value is 0. Representing adjacent nodes Transmitted to nodes via a distributed communication network The local cooperative state variables.

[0085] Through this iterative rule, the power state deviation of any node in the system will spread to the entire network within a finite number of iterations, achieving a gradual approach to zero for the fundamental power allocation error.

[0086] Specifically, regarding the global scheduling of broadband oscillations, the distributed damping gain scheduling unit constructs a dual-objective optimization model and uses a distributed consensus iterative algorithm to solve for the damping scheduling coefficients of each node. The optimization objective of the dual-objective optimization model is to minimize the weighted sum of the global oscillation energy and the total damping loss. The global oscillation energy is obtained by summing the oscillation energy levels of each node in the microgrid, and the total damping loss is obtained by summing the squares of the damping scheduling coefficients of each node in the microgrid. The two are weighted by a preset damping loss weighting coefficient.

[0087] If the system relies solely on the local broadband damping control unit at the bottom layer for self-response, local node damping output saturation is highly likely, leading to inverter overcurrent protection triggering. Furthermore, different nodes have different geographical locations and line impedances, resulting in significant differences in their efficiency in suppressing network-wide oscillations. By constructing a dual-objective optimization model, the system can find the optimal Pareto balance between suppressing the intensity of network-wide broadband oscillations (i.e., global oscillation energy) and reducing the additional hardware heat and energy losses incurred by the inverter to provide damping (i.e., total damping loss).

[0088] For example, the objective function of this bi-objective optimization model is shown in the following formula:

[0089] ;

[0090] In the formula, The comprehensive evaluation function value representing the bi-objective optimization; accumulation symbol. Covers all nodes participating in coordinated scheduling within the microgrid ; Represents the first uploaded through the underlying layer The oscillation energy level of each node; This represents the preset damping loss weighting coefficient. The magnitude of this coefficient determines whether the system is more inclined to quickly smooth out oscillations or protect the lifespan of the inverter hardware under the current operating conditions. Representing the The damping scheduling coefficients need to be calculated for each node.

[0091] The distributed damping gain scheduling unit solves the objective function using distributed optimization algorithms such as subgradient descent, obtains the optimal damping scheduling coefficient matrix that satisfies the network constraints, and distributes it to each underlying physical node. This ensures that under oscillation conditions, the node with the strongest suppression capability and the largest hardware margin undertakes the main damping task.

[0092] It should also be noted that both collaborative computation for the fundamental frequency and gain scheduling for the oscillation rely on a highly reliable information exchange and transmission medium.

[0093] In this embodiment, the distributed communication network is configured with a communication link self-healing unit and a dynamic resource scheduling unit. The communication link self-healing unit monitors the link quality factor in real time. When the link quality factor is lower than a first preset threshold, it selects the optimal backup link through a water wave optimization algorithm and performs seamless switching. The dynamic resource scheduling unit performs hierarchical scheduling of transmitted data according to the link quality factor, giving priority to ensuring the transmission of the actual current, the fundamental voltage phasor, and the oscillation slow characteristic quantity.

[0094] Microgrid communication networks are typically deployed in industrial areas with strong electromagnetic interference or in remote areas with harsh natural environments. Fiber optic communication may face physical damage, while backup links based on 5G mobile communication technology or wireless local area networks may experience signal attenuation.

[0095] The communication link self-healing unit continuously calculates the link quality factor of each communication path. The water wave optimization algorithm simulates the propagation, refraction, and wave breaking mechanism of water waves, and has strong global search capabilities and fast convergence speed. It can quickly plan alternative routes that bypass congested or faulty nodes from the huge network topology at the moment when the performance of the main link degrades.

[0096] Meanwhile, when the link quality factor decreases but a switchover has not yet been triggered, the dynamic resource scheduling unit imposes strict priority sorting restrictions on network packets, discarding non-time-sensitive data packets such as equipment temperature monitoring and environmental parameter collection, freeing up limited network bandwidth resources, and prioritizing the transmission of core closed-loop data participating in the underlying loop control and top-level iterative calculation.

[0097] To clarify the evaluation logic of the link quality factor in the communication link self-healing unit, the following formula is provided for explanation:

[0098] ;

[0099] In the formula, The link quality factor represents the communication link; Represents the communication delay weighting coefficient; This represents the maximum tolerable communication delay time allowed by the system. This represents the real-time communication latency measured using a time synchronization protocol. Represents the data packet integrity weighting coefficient; This represents the data packet delivery success rate within a specific statistical window.

[0100] When the real-time latency increases or the packet loss rate increases, the link quality factor calculated by this formula will decrease significantly. When its value is lower than the aforementioned first preset threshold, the water wave optimization algorithm is triggered to reconstruct the network topology.

[0101] Optionally, after completing the decoupling of the underlying data and the collaborative calculation of the top layer, the control system must smoothly integrate these two scheduling commands for different frequency bands and apply them to the physical modulation hardware of the inverter.

[0102] In this embodiment, the specific data fusion logic of the dual-layer power command feedforward fusion unit is as follows: calculate the difference between the steady-state reactive power coordination command and the fundamental reactive power calculated based on the actual current to obtain a first difference; calculate the difference between the steady-state active power coordination command and the fundamental active power calculated based on the actual current to obtain a second difference; multiply the vector formed by the first difference and the second difference with the operating condition adaptive steady-state droop coefficient matrix to obtain a first adjustment amount; multiply the vector formed by the reactive power damping command and the active power damping command with the operating condition adaptive damping feedforward reference matrix to obtain a second adjustment amount; add the first adjustment amount and the second adjustment amount to obtain a command matrix composed of the inverter output voltage reference value and the angular frequency reference value.

[0103] The essence of this data fusion logic lies in the fact that, in the conventional inverter control of a microgrid, active power primarily affects the system's angular frequency, while reactive power primarily affects the system's node voltage. The fusion logic described above can be fully expressed using the following matrix operations:

[0104] ;

[0105] In the formula, the left side of the matrix This represents the final generated inverter output voltage reference value. This represents the final generated angular frequency reference value; in the first term on the right side of the equals sign, and The adaptive steady-state droop coefficient matrix for the aforementioned operating condition is formed, representing the conventional reactive-voltage droop coefficient and active-frequency droop coefficient, respectively. This represents the steady-state reactive power coordination command. This represents the fundamental reactive power calculated based on the actual current, and the difference between the two corresponds to the aforementioned first difference. This represents the steady-state active power coordination command. This represents the fundamental active power calculated based on the actual current, and the difference between the two corresponds to the second difference mentioned above; in the second term on the right side of the equation, and The adaptive damping feedforward reference matrix for the aforementioned operating condition is formed, representing the reactive damping feedforward coefficient and the active damping feedforward coefficient, respectively. The reactive power damping command represents the output of the underlying local broadband damping control unit. This represents the active damping command.

[0106] As can be seen from the matrix equation, the steady-state cooperative deviation is formed by the traditional droop coefficient array to form the first adjustment quantity for maintaining the steady-state operation of the system fundamental wave, while the damping command containing full-band oscillation suppression information is formed by the feedforward matrix to form the second adjustment quantity for dynamically smoothing broadband oscillations.

[0107] The two are losslessly orthogonally superimposed at the final output, and the resulting command matrix is ​​directly fed into the inverter's space vector pulse width modulator, thereby controlling the on / off duty cycle sequence of the inverter's insulated gate bipolar transistor, transforming the digital-level collaborative algorithm into the physical-level electromagnetic power output.

[0108] Based on the current technical status of microgrids when facing the access of high-penetration power electronic devices, it can be seen that traditional control schemes, due to the lack of hardware decoupling mechanisms between the fundamental wave and broadband oscillation signals, as well as the lack of a global damping coordination system across nodes, are prone to large-scale grid disconnection or even microgrid disconnection once grid-connected resonance or islanded broadband oscillation occurs.

[0109] The solution in this embodiment enables the microgrid to accurately and smoothly schedule the capacity of idle inverters across the entire network without affecting the fundamental voltage and the on-demand allocation of active and reactive loads when facing any complex load changes or broadband oscillations caused by network topology changes. This effectively suppresses dangerous oscillation energy in multiple frequency bands, and effectively improves the safety and stability margin of modern power system microgrid operation and its self-healing survivability in harsh communication environments.

[0110] Example 2:

[0111] In practical industrial microgrid applications, such as wind-solar-storage complementary microgrid clusters in remote areas or offshore wind power aggregation systems, the top-level distributed communication network (such as the data acquisition and monitoring control system communication network) is often in a physical environment with strong electromagnetic interference.

[0112] In addition, due to the wide spatial distribution of communication nodes, the differences in the processing capabilities of communication routing nodes, and the dynamic reconfiguration of microgrid topology, random packet loss and asymmetric delay in transmission time are very likely to occur in the communication link.

[0113] When dealing with the high-frequency dynamic physical process of broadband oscillation in microgrids, if the top-level control system relies entirely on the traditional fixed-period time-triggered communication mechanism, on the one hand, a large amount of redundant and repetitive state data will be injected into the communication channel when the system is in a steady state, resulting in the ineffective occupation of the limited communication bandwidth and causing channel congestion; on the other hand, when the microgrid experiences severe disturbances, the fixed-period communication mechanism cannot adaptively increase the data interaction frequency according to the urgency of the drastic changes in the system state.

[0114] More seriously, when the communication network experiences asymmetric delays or high packet loss rates due to congestion or physical interference, if the top-level nonlinear consensus power flow algorithm continues to indiscriminately trust and use "outdated" data transmitted from all adjacent nodes, it will cause the algorithm's iteration direction to deviate from the actual physical system state, and may even trigger "information-physical coupling resonance" where information system delays and physical power grid oscillations are superimposed, leading to complete instability of the microgrid.

[0115] To solve this technical problem, the technical solution of this embodiment includes the following:

[0116] Optionally, the top-level distributed collaborative scheduling layer further includes a time-delay resilient event-triggered communication subsystem configured in the distributed communication network. This time-delay resilient event-triggered communication subsystem is software-architecturally independent of the underlying inverter high-frequency pulse width modulation control task, and is typically deployed within the application layer program of the microgrid central controller or the edge computing gateway of each distributed node.

[0117] The core design concept of this subsystem is to break away from the traditional periodic data broadcasting mechanism and instead adopt an on-demand communication mode driven by changes in the system's physical state. Under this architecture, communication resources will only be actively used for data reporting and interaction when the node's own power grid operating characteristics undergo substantial changes and the magnitude of these changes exceeds the set safety threshold.

[0118] Through this mechanism, the system can fundamentally reduce the communication traffic of the entire network and smooth out peaks and valleys, reserving sufficient communication bandwidth to cope with emergency situations such as broadband oscillations.

[0119] Furthermore, the latency-elastic event-triggered communication subsystem includes a local state observation module and an adaptive triggering determination module. These two modules run in parallel as high-priority real-time task threads within the processor's microprocessor core, jointly constituting the data flow control mechanism of the sending node. They are responsible for rigorously evaluating and approving each set of scheduling instructions and running data about to enter the protocol stack's transmission queue.

[0120] For example, the local state observation module is used to calculate in real time the absolute value of the state error between the scheduling state parameters of this node at the current moment and the scheduling state parameters successfully sent to the distributed communication network last time, to obtain the node-level state deviation. In this operation step, the scheduling state parameters are explicitly defined to include the actual current, the fundamental voltage phasor, and the oscillation slow characteristic quantity.

[0121] In engineering implementation, these three types of data with different physical dimensions cannot be directly compared numerically. Therefore, the local state observation module first allocates a set of historical state registers with timestamps in memory to cache the scheduling state parameter matrix that was successfully sent through the network interface controller and received an acknowledgment reply from the underlying link layer last time.

[0122] When a new digital sampling cycle begins, the local state observation module extracts the latest calculated scheduling state parameter matrix. To eliminate the magnitude differences between different physical dimensions (such as the ampere value of current, the volt and angle values ​​of voltage phasors, and the Hertz value of oscillation frequency), the local state observation module performs per-unit processing on the above two sets of matrices, mapping all state parameters to the same numerical range.

[0123] Subsequently, a vector subtraction operation is performed between the current standardized matrix and the historical cached standardized matrix, and the weighted Euclidean norm or Chebyshev distance of the difference vector is calculated. This distance quantification is used as the node-level state deviation. This node-level state deviation objectively and accurately reflects the degree of deviation of the physical trajectory of this node during this period.

[0124] Specifically, in order to avoid the system's adaptability from deteriorating under different operating conditions due to a single fixed triggering condition, the adaptive triggering determination module sets a dynamic triggering threshold that includes an exponential decay term and a static constant term.

[0125] In microgrid communication control, if a low static threshold is used as the judgment standard immediately after a data transmission, the system state will often experience non-essential fluctuations due to measurement noise or transient decay in a short period of time due to the electromechanical transient inertia of the physical system. This can easily lead to frequent and continuous false triggering of communication in a short period of time, causing network communication congestion.

[0126] To suppress this phenomenon, the dynamic trigger threshold, which includes an exponential decay term and a static constant term, is defined and calculated in real time using the following formula:

[0127]

[0128] In the formula, the left side of the equation is Representing the first in a microgrid Each node at the current time The dynamic trigger threshold calculated at each moment; the right side of the equation The representative is the first The static constant term pre-tuned by each node defines the maximum state drift dead zone that the system can tolerate under long-term steady-state operation. Only state changes exceeding this dead zone limit are considered to have network interaction value. The initial amplitude coefficient representing the exponentially decaying term, whose value is usually positively correlated with the severity of the current system operating condition; natural constant. The mathematical basis for exponential decay; Representative regarding the first The exponential decay rate coefficient set for each node determines how fast the dynamic threshold decreases over time. The larger the decay rate coefficient, the faster the threshold falls back to the static dead zone. Representative from the first From the last successful data packet transmission by each node to the current time The absolute time interval that has elapsed.

[0129] The dynamic physical model constructed using the above formula can achieve the following control logic: When a node has just completed a data transmission, the time interval is extremely small, the exponential decay term is close to its initial maximum value, the dynamic trigger threshold is at a high level, and the system enters the trigger dead zone suppression stage, effectively shielding the pseudo-state changes caused by sensor measurement white noise or high-frequency ripple of switching elements; as time goes on, the time interval increases, the exponential decay term decays rapidly with a smooth curve to approach zero, the dynamic trigger threshold gradually decreases and stabilizes at the level of the static constant term, so that after a reasonable suppression period, the system can restore high sensitivity detection of small but continuous physical state deviations.

[0130] like Figure 5 As shown, combining the communication adaptive control strategy of the top-level distributed collaborative scheduling layer in complex physical environments, the dynamic threshold and state deviation interaction diagram of this delay elastic event triggering mechanism intuitively demonstrates the numerical determination evolution process of the sender's on-demand wake-up mechanism. The horizontal axis of the attached diagram represents time, scaled in seconds, and the vertical axis represents state deviation and threshold, scaled in per-unit values.

[0131] In the figure, the blue solid line represents the node-level state deviation curve calculated in real time by the local state observation module, the red solid line represents the dynamic trigger threshold curve generated by the adaptive trigger judgment module, and the green solid dots indicate the actual communication wake-up trigger pulse generated when the judgment condition is met.

[0132] As can be observed from the waveform evolution trajectory, whenever a green communication wake-up trigger pulse is generated, the red dynamic trigger threshold curve will quickly jump to a high level near the 0.5 per-unit value, and then decline smoothly in an exponential manner and gradually approach the lower limit of the static constant of the 0.1 per-unit value. This dynamic adjustment process shields the invalid fluctuations caused by the transient adjustment of the physical system after data transmission.

[0133] Within a time interval of 1.5 to 2.5 seconds, the blue curve rises sharply due to significant external disturbances encountered by the microgrid and breaks through the red dynamic attenuation defense line. The system generates a green trigger pulse in a timely manner at the intersection of the two curves to broadcast data. This on-demand communication mechanism reduces the communication bandwidth occupancy rate during periods without disturbances.

[0134] However, within the 3.5 to 4-second time range, the local state deviation of the blue signal continuously experiences high-frequency, irregular, and minute fluctuations near the lower limit of the static constant of 0.1 per unit, causing the system to trigger multiple dense green communication pulses consecutively within this extremely short time window. This objectively reflects that the existing dynamic trigger threshold model has a relatively fixed tuning of the static constant term, and will still exhibit local redundant communication phenomena when dealing with the underlying electrical measurement environment with strong random white noise. This indicates that the trigger determination logic still has room for technical upgrades and optimizations in the future by introducing an adaptive reference base technology based on real-time dynamic estimation of environmental background noise.

[0135] Optionally, the adaptive triggering determination module generates a communication wake-up command if and only if the node-level state deviation is greater than the dynamic triggering threshold, allowing the node to broadcast the updated scheduling state parameters to the distributed communication network.

[0136] In the specific microprocessor's underlying execution logic, once the condition is met, the adaptive triggering module generates a hardware interrupt or software semaphore through the operating system kernel. This communication wake-up command not only triggers the network protocol stack to encapsulate, verify, and encrypt the latest scheduling status parameters (including high-precision actual current values, amplitude and phase angle of the fundamental voltage, and oscillation slow characteristics containing the dominant oscillation frequency and energy level), but also activates the Ethernet controller or wireless RF front-end chip in low-power listening mode.

[0137] The awakened physical communication layer then injects the encapsulated data frame into the physical medium of the distributed communication network through a predetermined multicast or broadcast address. For sampling periods that do not meet the trigger conditions, the microprocessor will directly discard the currently stored backup state parameters, continue to execute local closed-loop calculations, and not send any messages to the outside world, thereby maintaining the idle state of the physical channel to the greatest extent possible.

[0138] For example, the above-mentioned event-triggered sending mechanism effectively solves the network bandwidth congestion problem, but the complete closed loop of microgrid distributed collaborative scheduling must also deal with the data distortion problem caused by poor communication links at the receiving end.

[0139] In a distributed control network, due to differences in the physical distance between different nodes, different queuing and scheduling strategies of relay routers, and uneven spatiotemporal distribution of electromagnetic interference intensity in the environment, a node will inevitably face non-uniform communication delays and varying degrees of data packet corruption and loss when receiving data broadcast by all its neighboring nodes.

[0140] Specifically, the top-level distributed cooperative scheduling layer is configured with delay compensation weights when receiving data from neighboring nodes. In the iterative calculation process of nonlinear consensus power flow algorithms and distributed consensus damping optimization, the traditional adjacency matrix weights are usually statically determined only by the physical connectivity of the communication topology or the electrical admittance of the lines.

[0141] However, when data from a neighboring node experiences a high latency of hundreds of milliseconds during transmission, the power grid state reflected in that data (such as the oscillation energy or voltage phasor at a certain moment) has already developed a significant phase lag compared to the current actual physical power grid state. If the receiving node continues to use static weights to treat this high-latency data equally with other low-latency health data, it is equivalent to introducing an unknown lag element into the closed-loop control system. This will directly weaken the system's phase margin and, when dealing with high-frequency sensitive conditions such as wideband oscillations, it is highly likely to induce the entire distributed control algorithm to diverge.

[0142] It is also important to note that, in order to quantify and dynamically adjust the negative impact of communication degradation, the delay compensation weight is negatively correlated with the real-time packet loss rate and asymmetric communication delay time of the communication link. This negative correlation means that the longer the delay of the communication link and the more severe the packet loss, the lower the weight of the data transmitted from that link should be in the collaborative computing at the receiving end.

[0143] Therefore, this embodiment constructs a precise adjustment model for the delay compensation weights, as shown in the following formula:

[0144] ;

[0145] In the formula, the left side of the equation is Represents this node Receiving adjacent nodes When sending data, the delay compensation weights are allocated for that specific communication link; the constant 1 in the denominator on the right side of the equation is used to prevent mathematical singularities where the divisor is 0, ensuring the boundedness and convergence of the weights; Representative node For nodes The delay sensitivity penalty factor set in the communication link is used to adjust the system's tolerance for time delay; The node represents the time measured via the network's precise time protocol. To the node The asymmetric communication delay time between them is usually obtained by subtracting the sending timestamp from the receiving timestamp and deducting the system's inherent clock skew. This represents the set packet loss sensitivity penalty factor, used to adjust the system's tolerance for data integrity loss; This represents the real-time packet loss rate, calculated within a set sliding time observation window, which is the ratio of the number of lost data packets to the total expected number of received data packets.

[0146] As can be seen from the above formula, both the increase in asymmetric communication delay time and the rise in real-time packet loss rate will lead to a linear increase in the value of the denominator, thus making the delay compensation weight of the communication link exhibit a smooth and strictly monotonically decreasing nonlinear decay characteristic.

[0147] like Figure 6 As shown, combined with the adaptive compensation mechanism of the top-level distributed collaborative scheduling layer in complex communication environments, the three-dimensional surface diagram of the nonlinear mapping between the delay compensation weight and the communication link characteristics intuitively displays the result of the weight reduction penalty calculation of the receiver for the bad communication data packets.

[0148] In the three-dimensional coordinate system of the attached figure, the first horizontal axis represents the asymmetric communication delay time, the second horizontal axis represents the real-time packet loss rate, and the vertical axis represents the delay compensation weight. The three-dimensional grid surface and the color mapping from bright yellow to dark blue in the figure represent the spatial distribution characteristics of the weight under different communication quality combinations.

[0149] From the spatial trend of the surface, it can be observed that when the asymmetric communication delay time approaches 0 and the real-time packet loss rate approaches 0, the surface located in the origin region of the coordinate system turns bright yellow. At this time, the delay compensation weight is maintained at the highest level close to 1 per unit value, indicating that the system gives high weight to the adjacent node status data transmitted by the healthy communication link.

[0150] As the asymmetric communication delay time extends to 0.5 seconds or the real-time packet loss rate climbs to 20%, the surface exhibits a smooth and non-linear monotonically decreasing downward trend along both horizontal coordinate axes, and the color gradually transitions to dark blue, representing low weight. This objectively reflects that the system effectively weakens the negative impact of poor communication links on the convergence stability of global collaborative instructions by reducing the weight of lagging state data.

[0151] Meanwhile, in the extremely harsh region where the asymmetric communication delay time is greater than 0.4 seconds and the real-time packet loss rate is greater than 15%, the dark blue low-weight surface shows slight burr undulation deformation. This truly records the local computational distortion caused by the digital quantization truncation error and measurement noise when the underlying digital communication module is evaluating extremely low-quality signals. This objectively shows that when dealing with extremely harsh boundary communication conditions accompanied by high-intensity random interference, the nonlinear mapping model of this weight compensation algorithm still has room for further optimization and upgrading in terms of the smoothness of numerical output and the ability to resist noise.

[0152] Specifically, the top-level distributed cooperative scheduling layer uses the delay compensation weight to reduce the weight of the state data of adjacent nodes that experience communication delays, in order to dynamically weaken the impact of high-latency communication links on the convergence stability of global cooperative instructions during the calculation of the nonlinear consensus power flow algorithm.

[0153] When calculating the state update matrix in the nonlinear consensus power flow algorithm, the system directly multiplies the delay compensation weights calculated by the above formula and applies them to the elements of the topological adjacency weight matrix of the corresponding link. When a severe communication failure occurs in an adjacent node, causing significant data lag, its corresponding delay compensation weight will approach zero.

[0154] At this point, when performing the next collaborative iteration calculation, this node will almost ignore the lagging state parameters transmitted by the faulty node, and instead rely more on its own local historical state feedback and data from other neighboring nodes with good communication. This weight-based soft isolation mechanism, compared to the hard cutoff strategy of directly disconnecting the communication connection, not only avoids the algebraic graph theory singularity problem caused by abrupt changes in the connectivity of the system topology matrix, but also enables the smooth reintegration and fusion of data from neighboring nodes as the communication link quality recovers through the automatic recovery of the delay compensation weights.

[0155] Given the current state of microgrid control systems in harsh communication environments, most existing solutions assume an ideal, latency-free, and infinitely bandwidth-free communication channel between the top-level energy management system and the bottom-level controller. However, when applied to real-world environments, traditional control strategies often suffer from network congestion due to large amounts of periodic redundant data, and generate control command oscillations when encountering occasional communication delays. Ultimately, these strategies not only fail to effectively mitigate broadband oscillations but may also become a contributing factor to grid instability.

[0156] This embodiment deeply integrates a latency-elastic event-triggered communication subsystem into the top-level distributed collaborative scheduling layer, constructing a transmitter-based on-demand wake-up mechanism based on dynamic thresholds and a receiver-based adaptive compensation mechanism based on communication quality degradation penalties. This overall scheme endows the microgrid control system with extremely strong cyber-physical fusion robustness.

[0157] In practical operation, the system significantly conserves valuable communication channel resources in industrial settings and ensures that even under adverse conditions such as high latency or packet loss at some network nodes, the collaborative control based on the nonlinear consensus algorithm maintains extremely high mathematical convergence stability. This solution plugs the risk of broadband oscillation runaway at the critical intermediate link of information transmission, providing a practical engineering implementation path for the safe, stable, and economical operation of microgrids in complex electromagnetic environments and unreliable communication networks.

[0158] Example 3:

[0159] In existing microgrid control systems, when top-level collaborative algorithms issue commands such as voltage regulation, frequency support, or oscillation suppression, they typically assume that the underlying grid-connected inverter is an ideal power source that ignores hardware boundaries.

[0160] However, in actual industrial microgrid operation scenarios, the output capability of inverters is limited by extremely strict physical laws. When the microgrid experiences broadband oscillations, the system needs the inverter to output high-frequency damping power to smooth out the oscillations.

[0161] Among them, the output active damping power requires the inverter to directly exchange energy with the DC side (such as photovoltaic arrays, energy storage battery clusters or DC bus capacitors); while the output reactive damping power, although it does not consume average energy, will cause a sharp rise in the effective value of the AC side current, resulting in huge conduction and switching losses of power switching devices such as insulated gate bipolar transistors inside the inverter.

[0162] Ignoring these physical constraints and blindly forcing the inverter to execute the excessively large damping scheduling coefficient issued by the top level will directly lead to a deep drop or an excessive surge in the DC bus capacitor voltage, triggering the DC overvoltage or undervoltage protection to disconnect from the grid; or it will cause the junction temperature of the power switching devices to exceed the physical limit in a short period of time, causing thermal runaway or even hardware failure.

[0163] To address the technical problem of the disconnect between physical hardware constraints and abstract algorithm scheduling, the technical solution of this embodiment includes the following:

[0164] Optionally, the underlying data sensing and physical control layer also includes AC / DC physical constraints and power margin assessment units configured at each node.

[0165] This unit, acting as a secure bridge connecting the physical hardware and control software, is independently deployed within the inverter's underlying digital signal processor, maintaining the same interrupt priority as the high-frequency sampling control task. Its core function is to continuously and comprehensively monitor and evaluate the inverter's hardware operating status, and to translate the hardware's maximum load-bearing capacity into a digitally usable power boundary, thereby providing an insurmountable safety baseline for any subsequent control commands.

[0166] For example, the AC / DC physical constraint and power margin assessment unit collects the DC bus voltage state parameters of the inverter and the instantaneous thermal load state parameters of the power switching devices inside the inverter in real time, and calculates and outputs the local active power damping power margin and the local reactive power damping power margin in real time in combination with the rated capacity of the inverter.

[0167] During actual data acquisition, the system utilizes high-precision high-voltage Hall voltage sensors or isolated voltage distribution resistor networks to continuously sample the actual voltage across the DC bus capacitors at a microsecond-level sampling rate for the DC bus voltage state parameters. Since the high-frequency throughput of active damping power directly causes the charging and discharging of the bus capacitors, this voltage state parameter objectively reflects the current adequacy of the DC side energy buffer.

[0168] Regarding the instantaneous thermal load parameters of the power switching devices inside the inverter, since directly measuring the junction temperature of the semiconductor chip is difficult to achieve in engineering, this system obtains the reference temperature by using a negative temperature coefficient thermistor arranged on the heat dissipation substrate of the power module. Combined with the real-time calculation of the instantaneous value of the three-phase current flowing through the device and the pulse width modulation switching frequency, the virtual junction temperature data and heat accumulation rate of the silicon wafer inside the power switching device are calculated online in real time through a preset equivalent thermal resistance-capacitance network model.

[0169] The virtual junction temperature data, together with the heat accumulation rate, constitutes the instantaneous thermal load state parameters, which truly reflect how much additional reactive current the hardware can withstand under the current operating conditions without thermal damage.

[0170] Specifically, in order to accurately convert the aforementioned physical state parameters into digital margin indices that can be used by the control algorithm, the AC / DC physical constraint and power margin evaluation unit performs calculations using a specific mathematical derivation model. The following formula is provided for illustration:

[0171] ;

[0172] ;

[0173] Of the two formulas above, the first sub-formula is used to calculate the local active power damping margin. This represents the local active power damping margin calculated at the end. It represents the energy conversion efficiency coefficient from the DC side to the AC side of the inverter, and is used to compensate for the loss of actual output active power due to switching losses. The equivalent physical capacitance of the inverter DC-side bus support capacitor; This represents the actual measured voltage value of the DC-side bus obtained in real time through sensors, which belongs to the aforementioned DC-side bus voltage state parameters. This represents the minimum threshold voltage allowed by the inverter on the DC side bus to maintain normal grid-connected control. The estimated time constant representing the typical period of broadband oscillation in a microgrid; Representing the first in a microgrid The fundamental active power that each node is outputting based on actual current calculations.

[0174] The physical meaning of this first sub-formula is that it calculates the maximum remaining energy that the DC-side capacitor can continuously release at rated efficiency without triggering undervoltage protection. After deducting the basic active power load it has already undertaken, the remaining power value is the limit margin that the inverter can use for output active power damping without causing DC-side collapse.

[0175] The second sub-formula is used to calculate the local reactive power damping power margin. This represents the local reactive power damping margin calculated at the end. The rated capacity of the aforementioned inverter reflects the inverter's maximum apparent power output capability under standard operating conditions. This represents the thermal derating factor calculated based on instantaneous thermal load state parameters. When the calculated virtual junction temperature of the device approaches the maximum safe temperature, this factor will gradually decay nonlinearly from 1 to 0, forcibly reducing the upper limit of the available apparent power of the inverter to prevent burnout. It also represents the fundamental active power; Representing the The fundamental reactive power of each node, calculated based on the actual current and currently being output.

[0176] The physical meaning of this second sub-formula is that, using the power triangle rule, under the apparent power derating constraint caused by the current thermal state, the upper limit of the total reactive power available to the inverter is calculated, and the fundamental reactive power that has already been output is further deducted, finally obtaining a dynamic safety margin that can be used to provide high-frequency reactive power damping.

[0177] Optionally, after completing the accurate physical margin calculation, these boundary conditions need to be used to enforce constraints on the control commands.

[0178] The local broadband damping control unit embeds a dynamic saturation limiting submodule. This submodule uses the local active power damping margin and the local reactive power damping margin as safety boundary conditions to perform saturation truncation control on the generated active power damping command and the reactive power damping command. In the specific low-level code execution flow, the dynamic saturation limiting submodule is connected in series between the original damping command generation logic and the final pulse width modulation generator.

[0179] Once the local broadband damping control unit has preliminarily calculated the theoretically required active and reactive damping command values ​​based on parameters such as real-time current, the dynamic saturation limiting submodule will immediately compare these theoretical values ​​with the values ​​calculated by the above formula. and Perform numerical comparison.

[0180] If the theoretically calculated active power damping command value exceeds the local active power damping margin, the module will forcibly truncate the final active power damping command issued to the hardware for execution to the upper limit of the local active power damping margin; conversely, if the theoretical value is within the safety boundary, it will be transmitted as is. Through this stringent hard limiting mechanism, the inverter possesses underlying hardware self-protection capabilities under any extreme operating conditions or even in the event of a failure of the top-level dispatch command, effectively preventing major operational accidents such as DC over / under voltage or AC overheating and grid disconnection caused by forcibly responding to damping demands.

[0181] like Figure 7 As shown, combining the AC / DC physical constraints of the inverter's underlying layer with the dynamic limiting logic of the power margin assessment unit, the damping power dynamic saturation limiting waveform diagram considering AC / DC physical constraints consists of two related sub-diagrams, which intuitively demonstrate the forced constraint execution process of the physical hardware boundary on the transient damping control command.

[0182] Both subplots have time on the x-axis, with the scale unit being seconds. The upper subplot has DC bus voltage on the y-axis, with the scale unit being volts. The red dashed line in the figure represents the minimum threshold set at 700 volts, and the blue solid line represents the actual measured voltage of the DC bus.

[0183] The vertical axis of the lower subplot represents active power, with the unit of measurement being kilowatts. The light blue dashed line in the figure represents the theoretical active damping command calculated from the underlying theory. The thin green solid line represents the dynamic envelope of the local active damping power margin calculated in real time based on DC voltage. The dark blue thick solid line represents the actual output active damping power after boundary truncation.

[0184] As can be seen from the waveform time-domain evolution, the system encountered a disturbance at 0.2 seconds. The theoretical active damping command quickly generated a high-frequency sinusoidal oscillation demand with an amplitude of up to 50 kilowatts. At the same time, the actual measured voltage of the upper sub-graph was affected by the physical influence of the damping power extraction, and dropped rapidly from 800 volts to approach the minimum threshold of 700 volts. Due to this objective physical condition, the green local active damping power margin envelope of the lower sub-graph rapidly contracted from the initial 60 kilowatts to around 30 kilowatts.

[0185] When the light blue theoretical active damping command repeatedly breaks through the green dynamic margin envelope within the 0.22-second to 0.3-second interval, the dark blue actual output active damping power waveform is forcibly truncated within the safety margin boundary, exhibiting obvious clipping and limiting characteristics. This objectively verifies that the system can avoid over-limit shutdown accidents caused by deep discharge of DC-side capacitors in extreme oscillation conditions.

[0186] Careful observation of the local dynamic characteristics at the time of truncation reveals that the actual output active damping power is accompanied by a tiny high-frequency spike overshoot of about 0.5 kW at the moment of contact with the green margin boundary. Furthermore, the green dynamic margin envelope itself exhibits continuous high-frequency subtle jitter due to the influence of the voltage sensor's wideband measurement noise. This objectively reflects that the current hardware hard-limiting logic is limited in dealing with the discrete sampling delay of analog-to-digital conversion and the smoothness of nonlinear voltage envelope filtering. This indicates that the underlying boundary truncation execution stage still has substantial room for technical upgrades in the future, particularly in the introduction of dead-zone feedforward compensation algorithms and multi-stage Kalman filters.

[0187] For example, relying solely on bottom-level self-protection cutoff is far from sufficient. If a node unilaterally cuts off the output of damping power, it means that the total damping energy used to smooth broadband oscillations in the microgrid is insufficient.

[0188] If the top-level coordination system is unaware of this and continues to issue the original scheduling instructions to the node, the broadband oscillations of the entire microgrid will not be effectively suppressed. Therefore, the AC / DC physical constraint and power margin assessment unit will also upload margin penalty factors characterizing the depletion degree of the local active power damping power margin and the local reactive power damping power margin to the top-level distributed coordination scheduling layer via the distributed communication network.

[0189] This design establishes a margin alarm channel in the microgrid control system that runs directly from the underlying physical hardware to the top-level collaborative scheduling layer, enabling the global collaborative scheduling algorithm to obtain the available power margin status of each underlying physical node in real time.

[0190] Specifically, to quantify this level of hardware exhaustion and enable it to directly participate in complex top-level mathematical optimization operations, the system precisely models the margin penalty factor. The following formula illustrates the algorithm for generating this margin penalty factor:

[0191] ;

[0192] In the formula, This represents the margin penalty factor finally calculated for the i-th node; The penalty weighting coefficient, which is preset for the active power channel, is used to adjust the proportion of the impact of active power margin depletion on the overall penalty intensity. This represents the measured value of the active damping power that the inverter at node i is currently outputting. This represents the local active power damping margin of the i-th node, calculated using the aforementioned formula. This represents a very small positive bias constant, used to place in the denominator to prevent crashes caused by illegal mathematical operations involving division by zero when the margin is completely exhausted and the value drops to zero. This represents the penalty weighting coefficient preset for the reactive power channel; This represents the measured value of the reactive damping power that the inverter at node i is currently outputting. This represents the local reactive power damping margin of the i-th node, calculated using the aforementioned formula; natural constant. An exponential growth model was constructed.

[0193] The above equation clearly shows that when the actual damped output of the i-th node is much smaller than the remaining margin, the exponent of the exponential term approaches zero, and the margin penalty factor remains near a small basic constant. However, once the actual output approaches or even equals the current remaining limit margin, the fractional term increases rapidly, and the characteristics of the exponential function cause the margin penalty factor to spike geometrically. This nonlinear penalty mechanism can amplify the imminent exhaustion of node hardware into a highly significant numerical signal, which is then reported through the communication network.

[0194] It is also worth noting that after receiving this critical hardware status signal at the top level, the system performed a corresponding global policy reconfiguration.

[0195] When solving for the damping scheduling coefficients of each node, the distributed damping gain scheduling unit in the top-level distributed collaborative scheduling layer introduces the margin penalty factor as a dynamic penalty constraint into the bi-objective optimization model.

[0196] To achieve adaptive energy redistribution of global damping power, the system expands and reconstructs the original objective function in Example 1, directly coupling the uploaded margin penalty factor as an adaptive penalty weight into the damping loss term of the objective function. The reconstructed bi-objective optimization update model is shown in the following formula:

[0197] ;

[0198] In the formula, Represents the global comprehensive evaluation objective function reconstructed after introducing hardware physical constraints; accumulation symbol Covers all nodes participating in coordinated scheduling within the microgrid. ; Representing the Each node extracts and uploads the oscillation energy level from the underlying layer; This represents the system's preset basic damping loss weighting coefficient; For the first The margin penalty factor is reported in real time by each node through the communication network; Representing the The damping scheduling coefficients of each node to be solved.

[0199] Since the optimization objective of the distributed damping gain scheduling unit is to find The function's minimum value, when a certain node Due to the depletion of its own hardware margin, its corresponding When this value increases dramatically, the penalty gradient weight of this term in the objective function will rise sharply. To ensure the monotonically converging descent of the global objective function, the optimization solver will automatically avoid this high-penalty node on the mathematical iteration gradient, thereby significantly reducing the damping scheduling coefficient corresponding to this node. The numerical values ​​are used to smoothly transfer the damping task to a healthy node with a smaller margin penalty factor, thus forming an adaptive energy redistribution closed loop with rigorous mathematical logic support.

[0200] For example, this mathematical optimization process based on a penalty mechanism is transformed into an adaptive energy transfer and collaborative control process in the actual operation of a physical microgrid.

[0201] Assume the microgrid contains node A and node B. Due to cloud cover, the DC bus voltage of node A is low, resulting in a very small active power damping margin calculated by the AC / DC physical constraints and power margin assessment unit, and consequently a very large margin penalty factor.

[0202] Meanwhile, the energy storage battery connected to node B has sufficient power, and the power device temperature is low, so its margin penalty factor remains at a low safe value. When dealing with the broadband oscillation of the entire network, the top-level distributed damping gain scheduling unit will detect this difference and quickly reconstruct the originally evenly distributed damping scheduling coefficient, actively reducing the damping scheduling coefficient of node A to an extremely low level, allowing node A to maintain only basic fundamental wave power generation.

[0203] Simultaneously, the damping energy gap required to suppress broadband oscillations across the entire network is transferred to node B, significantly increasing the damping dispatch coefficient of node B. Node B then releases a large amount of damping power within its ample physical boundaries to smooth out oscillations. This adaptive energy redistribution mechanism breaks the rigid parameter tuning mode in traditional control systems, endowing the microgrid system with extremely strong group cooperation flexibility under extreme operating conditions.

[0204] like Figure 8 As shown, combined with the dual-objective optimization and reconfiguration mechanism of the top-level distributed collaborative scheduling layer, the adaptive reconfiguration data graph of the global damping scheduling coefficient based on the margin penalty factor consists of two related subgraphs, which objectively demonstrate the adaptive redistribution process of global damping power in the face of local node hardware exhaustion in a microgrid.

[0205] The horizontal axis of both subgraphs represents time, with the scale unit being seconds. The vertical axis of the upper subgraph represents the margin penalty factor. The dark red solid line in the graph represents the hardware fatigue state of the first node, calculated and uploaded in real time through the underlying physical boundary.

[0206] The vertical axis of the lower subplot represents the damping control coefficient. The solid blue line in the figure represents the damping control coefficient of the first node, and the dashed green line represents the damping control coefficient of the second node.

[0207] From the waveform evolution process, it can be observed that in the initial oscillation range of 0.5 seconds to 2 seconds, since the hardware margins of both nodes are in a healthy state, the dark red margin penalty factor is maintained near the low safety constant of 1.5. At this time, the blue solid line overlaps with the green dashed line, and the system distributes the damping scheduling coefficient evenly at the numerical level of 0.45, with both inverters undertaking the task of oscillation smoothing.

[0208] At 2 seconds, the first node's available active damping power margin is severely depleted due to its own physical limits. The dark red curve in the upper subplot rises rapidly under the drive of the exponential model and approaches a high penalty value of 26.5 near 3 seconds.

[0209] Driven by this dynamic penalty boundary condition, the optimization algorithm in the lower subgraph responds rapidly. The blue damping scheduling coefficient curve of the first node decreases exponentially and converges to an extremely low level of 0.1, forcibly removing the damping output burden from this fatigued node. Simultaneously, the green damping scheduling coefficient curve of the second node rises synchronously and stabilizes at a high level of 0.8, taking over the main damping task required to smooth broadband oscillations across the entire network. This process verifies that the distributed cooperative scheduling layer can achieve safe transfer of damping energy based on the physical health of nodes.

[0210] However, during the coefficient reconstruction transition interval of 2 to 4 seconds, both the blue solid line and the green dashed line exhibited mutually coupled oscillations and local numerical jitters with a frequency of approximately 4 Hz and an amplitude of approximately 3%. This truly reflects that when the current distributed consensus iterative algorithm deals with abrupt changes in large gradient penalty constraints, there is a certain dynamic convergence lag and iterative optimization oscillation in its step-size optimization process for finding the minimum value. This indicates that the bi-objective optimization model still has clear room for technical upgrades in the future, particularly in the introduction of adaptive variable step-size gradient descent algorithms and the improvement of the iterative convergence damping ratio.

[0211] Based on existing technologies, this embodiment summarizes the overall effectiveness of the solution. Traditional microgrid broadband oscillation control schemes often focus on AC side signal processing and control law design, severely neglecting the physical impact of damping energy interaction on the thermal load of inverter DC energy storage components and AC switching components. This design flaw leads to the damping commands, originally intended for system stabilization, often becoming the direct cause of physical hardware overload damage when facing severe or prolonged broadband oscillations, resulting in cascaded grid disconnection and shutdown accidents of the inverter cluster.

[0212] In this embodiment, the AC / DC physical constraints and power margin assessment unit are deployed deep into the hardware layer, establishing a hard limiting defense line based on physical limits. At the same time, by constructing a margin penalty factor mechanism, the feedback path from the fatigue state of the hardware layer to the control brain is opened, enabling the top-level distributed collaborative scheduling to perform targeted adaptive energy redistribution.

[0213] The entire technical solution not only effectively avoids the risk of damage beyond the limit when a single inverter participates in damping control, ensuring the service life of microgrid hardware assets; more importantly, by relaying the adaptive damping power of fatigue nodes through global healthy nodes, it significantly enhances the continuous survivability and global stability recovery capability of the microgrid system under long-term complex and severe oscillation conditions, realizing a major leap from pure software algorithm control to deep integration of software, hardware, physical and information control.

[0214] Example 4:

[0215] Traditional microgrid control architectures often employ a single control level. When processing complex grid signals including power electronic conversion devices, they typically cannot effectively decouple the fundamental power flow signal from the broadband oscillation signal physically or algorithmically during the data sensing phase. This single data processing link causes composite signals containing high-frequency or low-frequency oscillation components to directly enter the steady-state power allocation stage, leading to distortion of coordinated commands. Simultaneously, the system lacks a global scheduling mechanism for oscillatory characteristics, making it difficult to achieve a balance between steady-state power distribution and dynamic damping.

[0216] In order to overcome the specific problems existing in the above-mentioned background art and to achieve the beneficial effects of the present invention, such as Figure 9 As shown, this embodiment provides a two-layer distributed data acquisition and control method for microgrids to cope with broadband oscillations. The method is applied to a two-layer distributed data acquisition and control system for microgrids to cope with broadband oscillations, and includes the following steps:

[0217] Step S1: Collect the composite current signal of the inverter in the microgrid through the underlying data sensing and physical control layer, and extract the real-time current containing broadband oscillation components and the actual current containing only the power frequency fundamental wave component from the composite current signal through the hardware-level adaptive bandpass decoupling unit in the underlying data sensing and physical control layer.

[0218] In specific industrial microgrid physical application environments, the execution of step S1 heavily relies on high-precision power sensing equipment and underlying digital signal processing hardware. A wideband Hall current transformer is configured at the inverter's AC output port to continuously monitor the actual current of the microgrid under complex operating conditions. The analog signal acquired by this sensor is converted into a discrete digital sequence, i.e., the composite current signal, via a high-speed analog-to-digital converter.

[0219] In actual engineering research and development and control equipment deployment processes, in order to avoid occupying a large number of clock cycles of the main control chip, the complex nonlinear control algorithms and feature extraction logic involved in this step are usually modeled in advance in a software environment that supports advanced matrix operations and dynamic system simulation, including physical object modeling, control law pre-simulation, and visualization analysis of running data graphs.

[0220] After parameter tuning in the virtual environment, automatic code generation technology is used to convert the control logic into hardware description language or low-level C language source code, which is then directly burned into a field-programmable gate array (FPGA) or a dedicated digital signal processor (DSP) for execution. The hardware-level adaptive bandpass decoupling unit, relying on the parallel computing architecture of the FPGA, performs real-time filtering of the composite current signal through a hardware digital filter array, strictly separating it in the frequency domain and outputting a high-frequency broadband real-time current signal and a low-pass filtered actual current signal. This achieves decoupling of steady-state control data from transient damping data at the very front end of the physical link.

[0221] Step S2: The local broadband damping control unit in the underlying data sensing and physical control layer generates active damping commands and reactive damping commands based on the real-time current and instantaneous port voltage.

[0222] In the actual inverter's underlying control cabinet, the local broadband damping control unit is mainly implemented by a high-speed software control loop on the main control processing board. After the microprocessor obtains the real-time current, which contains real and violent oscillation characteristics without smoothing, it performs coordinate transformation and instantaneous electromagnetic power calculation with the instantaneous port voltage collected by the voltage transformer. The active and reactive components calculated at this time directly reflect the current transient energy fluctuation state of the microgrid.

[0223] The underlying control program calculates the active and reactive power damping commands required to suppress wideband oscillations in this frequency band based on the set adaptive damping control law. From a physical perspective, the execution of the active power damping command directly utilizes the transient physical energy of the inverter's DC-side capacitors or energy storage battery clusters for throughput, while the execution of the reactive power damping command alters the turn-on and turn-off timing of the inverter's AC-side bridge arm insulated-gate bipolar transistors to output reactive current.

[0224] The independent operation of this step enables the underlying power electronic equipment to have a local first line of physical defense against broadband oscillations without relying on external communication commands.

[0225] Step S3: The actual current, fundamental voltage phasor, and slow oscillation characteristic of each node in the microgrid are obtained through the top-level distributed collaborative scheduling layer based on the distributed communication network.

[0226] This control step extends the system's data domain from the local physical device level to the central communication network of the entire microgrid system. In practical operation scenarios, this distributed communication network is typically constructed using industrial Ethernet switches, fiber optic ring networks, or low-latency industrial wireless communication gateways.

[0227] The edge computing gateways or synchronous phasor measurement devices of each distributed node in the microgrid are responsible for standardizing the pure actual current extracted in step S1, the fundamental voltage phasor calculated by the phase-locked loop, and the slow oscillation characteristic quantity characterizing the macroscopic evolution trend of the oscillation after dimensionality reduction processing into standardized protocol packets. Subsequently, these important data packets are transmitted and exchanged point-to-point or multicast in the distributed communication network.

[0228] This step ensures that each computing node participating in the global scheduling of the microgrid can update the steady-state electrical quantity topology and oscillation characteristic spectrum of the entire network in real time in its local register, providing an accurate, reliable, and frequency-reduced data foundation for subsequent multi-agent collaborative computing.

[0229] Step S4: The steady-state active power coordination command and steady-state reactive power coordination command are output by the fundamental power coordination calculation unit in the top-level distributed coordination scheduling layer, and the damping scheduling coefficient of each node is output by the distributed damping gain scheduling unit in the top-level distributed coordination scheduling layer based on the oscillation slow characteristic quantity.

[0230] In terms of specific system hardware deployment, the top-level distributed collaborative scheduling layer typically runs centrally on the industrial control server in the microgrid's central control room, or is distributed across high-performance station control terminals in each sub-microgrid system. The fundamental power collaborative calculation unit executes a consensus power flow iterative algorithm, and based on the clean fundamental data obtained in step S3, calculates a steady-state collaborative command aimed at eliminating steady-state voltage and frequency deviations in the microgrid and achieving equal distribution of active and reactive power equipment capacity.

[0231] Meanwhile, the distributed damping gain scheduling unit substitutes the slow oscillation characteristics uploaded by each node into the global objective optimization function for numerical solution. This calculation process directly evaluates the damping capability and current hardware status of all physical nodes in the network, outputting globally targeted damping scheduling coefficients. The fundamental power collaborative calculation unit and the distributed damping gain scheduling unit are independent at the computational physical layer, performing command calculations for the power frequency steady-state domain and wideband transient domain of the power grid, respectively, ensuring that the global equalization logic and the dynamic damping allocation logic do not interfere with each other at the algorithm execution level.

[0232] Step S5: The steady-state active power coordination command, the steady-state reactive power coordination command, the active power damping command, and the reactive power damping command are received by the dual-layer power command feedforward fusion unit. The output voltage reference value and the angular frequency reference value of the inverter are generated by orthogonal superposition, and the output voltage reference value and the angular frequency reference value are injected into the droop control main circuit of the inverter.

[0233] In the final stage of the closed-loop control method, the multi-dimensional command data streams converge again within the digital signal controller of the underlying physical inverter. The dual-layer power command feedforward fusion unit receives macroscopic steady-state cooperative commands parsed from the network protocol stack, as well as high-frequency transient damping commands calculated from the underlying local loop.

[0234] Since the physical orthogonal decoupling of the fundamental frequency and the oscillation frequency band has been achieved in the previous data sensing stage (step S1), in the command terminal fusion stage here, the controller directly uses vector addition to mathematically superimpose the steady-state command, which represents the intention of global steady-state equalization of the microgrid, and the damping command, which represents the intention of broadband transient suppression. The output voltage reference value parameter and angular frequency reference value parameter generated after superposition are directly fed into the space vector pulse width modulation module of the inverter, and converted into a drive level signal with a specific pulse width duty cycle, which is ultimately applied to the physical power switching devices of the inverter.

[0235] In summary, during the actual operation of the microgrid, the method in this embodiment strictly adheres to the operating principle of a two-layer architecture, achieving decoupled operation of fundamental steady-state control and broadband oscillation suppression. At the bottom-level sensing and processing stage, hardware-level filtering decoupling isolates high-frequency oscillation data from interfering with the fundamental power control process. At the top-level coordination stage, the isolated extracted data synchronously realizes the allocation of fundamental steady-state power and the calculation of damping scheduling coefficients for each node.

[0236] Ultimately, by orthogonally superimposing commands at the inverter terminal, the inverter can output damped power incorporating global coordinated dispatch characteristics while ensuring the fundamental voltage, frequency regulation, and power distribution of the entire grid. This method solves the command distortion problem caused by the lack of decoupling of composite signals in existing technologies, effectively improving the overall stability control capability and hardware reliability of microgrid systems under wideband oscillation conditions.

[0237] Example 5:

[0238] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0239] like Figure 10 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0240] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0241] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0242] The memory 103 stores a computer program corresponding to the microgrid two-layer distributed data acquisition and control method for dealing with broadband oscillations according to the above embodiments of the present invention. This computer program is executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0243] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 10The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0244] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A two-layer distributed data acquisition and control system for microgrids to cope with broadband oscillations, characterized in that, It includes a bottom-level data perception and physical control layer, a top-level distributed collaborative scheduling layer, and a two-layer power command feedforward fusion unit that connects the two layers; The underlying data sensing and physical control layer collects the composite current signal of the inverter in the microgrid, and extracts the real-time current containing a broadband oscillation component and the actual current containing only the power frequency fundamental component from the composite current signal through a hardware-level adaptive bandpass decoupling unit; the underlying data sensing and physical control layer also includes a local broadband damping control unit, which generates active power damping command and reactive power damping command based on the real-time current and the instantaneous voltage of the port. The top-level distributed collaborative scheduling layer obtains the actual current, fundamental voltage phasor, and slow oscillation characteristic of each node in the microgrid based on the distributed communication network. The fundamental power collaborative calculation unit in the top-level distributed collaborative scheduling layer outputs steady-state active power collaborative instructions and steady-state reactive power collaborative instructions, and the distributed damping gain scheduling unit outputs the damping scheduling coefficient of each node based on the oscillation slow characteristic quantity. The dual-layer power command feedforward fusion unit receives the steady-state active power coordination command, the steady-state reactive power coordination command, the active power damping command, and the reactive power damping command, and generates inverter output voltage reference value and angular frequency reference value through orthogonal superposition, and injects them into the inverter's droop control main circuit.

2. The system according to claim 1, characterized in that, The hardware-level adaptive bandpass decoupling unit adopts an adjustable bandwidth second-order generalized integrator array, which is implemented in parallel hardware through a field-programmable gate array. The system dynamically adjusts the sampling rate and passband range for the real-time current according to the system operating level of the microgrid, so that the output data for the actual current always maintains a preset fixed narrowband output. The system classifies the operating conditions of the microgrid into three levels: Level 1 steady-state operating condition, Level 2 fluctuating operating condition, and Level 3 oscillating operating condition. The hardware acquisition parameters of the underlying data sensing and physical control layer, the damping gain of the local broadband damping control unit, and the iteration cycle and communication scheduling strategy of the top-level distributed collaborative scheduling layer are all synchronously and adaptively adapted to the switching of the system operating condition level.

3. The system according to claim 1, characterized in that, The underlying data perception and physical control layer also includes an oscillation slow feature extraction unit; The oscillation slow feature extraction unit obtains the oscillation residual current by subtracting the real-time current from the actual current, and extracts the dominant oscillation frequency, oscillation energy level, and oscillation power flow angle from the oscillation residual current based on a set sliding time window. These are packaged as the oscillation slow feature quantity and uploaded to the distributed communication network along with the fundamental steady-state data. The fundamental steady-state data includes at least the actual current and the fundamental voltage phasor.

4. The system according to claim 1, characterized in that, The local broadband damping control unit includes a local adaptive damping subunit and a long short-term memory network oscillation prediction subunit. The local adaptive damping subunit dynamically adjusts the local active damping gain and local reactive damping gain according to the real-time oscillation amplitude; the long short-term memory network oscillation prediction subunit outputs the oscillation growth rate factor based on the time-series data composed of the oscillation residual current, thereby generating the pre-damping compensation amount. The local broadband damping control unit combines the damping scheduling coefficient, the local active power damping gain, the local reactive power damping gain, and the pre-damping compensation amount to generate the final total active power damping gain and total reactive power damping gain, which are used to output the active power damping command and the reactive power damping command.

5. The system according to claim 2, characterized in that, The fundamental power collaborative calculation unit adopts a graph-based nonlinear consensus power flow algorithm; The nonlinear consistent power flow algorithm aims to minimize the global voltage deviation, frequency deviation, and active and reactive power sharing error. Furthermore, the iteration step size and iteration period of the nonlinear consistent power flow algorithm are dynamically adjusted according to the system operating condition level of the microgrid.

6. The system according to claim 3, characterized in that, The distributed damping gain scheduling unit constructs a dual-objective optimization model and uses a distributed consensus iterative algorithm to solve for the damping scheduling coefficients of each node. The optimization objective of the bi-objective optimization model is to minimize the weighted sum of global oscillation energy and total damping loss. The global oscillation energy is obtained by summing the oscillation energy levels of each node in the microgrid, and the total damping loss is obtained by summing the squares of the damping scheduling coefficients of each node in the microgrid. The two are weighted by a preset damping loss weighting coefficient.

7. The system according to claim 1, characterized in that, The specific data fusion logic of the dual-layer power command feedforward fusion unit is as follows: The difference between the steady-state reactive power coordination command and the fundamental reactive power calculated based on the actual current is calculated to obtain a first difference value; the difference between the steady-state active power coordination command and the fundamental active power calculated based on the actual current is calculated to obtain a second difference value; the vector formed by the first difference value and the second difference value is multiplied by the operating condition adaptive steady-state droop coefficient matrix to obtain a first adjustment amount; The second adjustment amount is obtained by multiplying the vector formed by the reactive damping command and the active damping command with the operating condition adaptive damping feedforward reference matrix. The first adjustment amount and the second adjustment amount are added together to obtain a command matrix composed of the inverter output voltage reference value and the angular frequency reference value.

8. The system according to claim 1, characterized in that, The distributed communication network is equipped with a communication link self-healing unit and a dynamic resource scheduling unit. The communication link self-healing unit monitors the link quality factor in real time. When the link quality factor is lower than the first preset threshold, it selects the optimal backup link through the water wave optimization algorithm and performs seamless switching. The dynamic resource scheduling unit performs hierarchical scheduling of transmitted data based on the link quality factor, prioritizing the transmission of the actual current, the fundamental voltage phasor, and the slow oscillation characteristic.

9. The system according to claim 5, characterized in that, Under the first level steady-state condition, the sampling of the real-time current adopts a first sampling rate and a first passband range, and the nonlinear consistent power flow algorithm adopts a first iteration step size; Under the second-level fluctuation condition, the sampling of the real-time current adopts the second sampling rate and the second passband range, the long short-term memory network oscillation prediction subunit is activated to perform pre-damped feedforward, and the nonlinear consensus power flow algorithm adopts the second iteration step size; Under the third-level oscillation condition, the sampling of the real-time current adopts a third sampling rate and a third passband range, and the nonlinear consensus power flow algorithm adopts a third iteration step size; Wherein, the first sampling rate is less than the second sampling rate, and the second sampling rate is less than the third sampling rate; the first passband range is less than the second passband range, and the second passband range is less than the third passband range; the first iteration step size is greater than the second iteration step size.

10. The system according to claim 1, characterized in that, The underlying data sensing and physical control layer also includes AC / DC physical constraints and power margin evaluation units configured at each node; The AC / DC physical constraint and power margin assessment unit collects the DC bus voltage state parameters of the inverter and the instantaneous thermal load state parameters of the power switching devices inside the inverter in real time. Combined with the rated capacity of the inverter, it calculates and outputs the local active power damping power margin and the local reactive power damping power margin in real time. The local broadband damping control unit is embedded with a dynamic saturation limiting submodule. The dynamic saturation limiting submodule uses the local active power damping power margin and the local reactive power damping power margin as safety boundary conditions to perform saturation cutoff control on the generated active power damping command and the reactive power damping command. The AC / DC physical constraint and power margin assessment unit will also upload the margin penalty factor, which characterizes the depletion degree of the local active power damping power margin and the local reactive power damping power margin, to the top-level distributed cooperative scheduling layer through the distributed communication network. When solving for the damping scheduling coefficients of each node, the distributed damping gain scheduling unit in the top-level distributed collaborative scheduling layer introduces the margin penalty factor as a dynamic penalty constraint into the bi-objective optimization model. By adaptively reducing the damping scheduling coefficients of nodes with depleted power margins, it achieves adaptive energy redistribution of global damping power.

Images