A source network interaction-oriented photovoltaic model-free parameter disturbance control method and system

Abstract

The application relates to the field of new energy grid-connected control technology and discloses a photovoltaic model-free parameter anti-disturbance control method and system facing source-grid interaction, which realizes rapid inhibition and advance compensation of transient disturbance by constructing composite feedforward compensation through disturbance observation and trend prediction on real-time electrical data of each distributed photovoltaic inverter; the adaptive capacity of network control to grid state change is improved by dynamically adjusting virtual inertia, damping and output impedance in combination with inverter response, grid strength and cluster disturbance characteristics; and the utilization efficiency of support resources and the overall cooperative anti-disturbance capacity of the cluster are improved by further performing task dynamic distribution and cooperative control according to the real-time support capacity of each inverter.

Description

Technical Field

[0001] This invention relates to the field of new energy grid-connected control technology, specifically to a photovoltaic model-free parameter disturbance rejection control method and system oriented towards source-grid interaction. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] As the scale of distributed photovoltaic (PV) grid connection continues to expand, the dynamic coupling between PV inverters and the power grid is becoming increasingly stronger. Especially under conditions of grid strength changes, fault disturbances, and operation mode switching, voltage, frequency, and power disturbances may propagate between multiple grid-connected nodes. The local adjustment of a single inverter is difficult to simultaneously address the rapid suppression of disturbances, the adjustment of grid construction capabilities, and the coordinated support of the cluster.

[0004] Existing technologies mostly employ model-based parameter-based control tuning, steady-state voltage identification, source-load prediction, scenario matching, or threshold switching for photovoltaic grid-connected control. However, actual grid parameters exhibit uncertainty and time-varying characteristics, while transient disturbances possess rapid evolution and network propagation characteristics, making it difficult for preset models or fixed control strategies to reflect disturbance states and their subsequent changes in a timely manner. Furthermore, different inverters have varying available active and reactive power, as well as thermal margins, making it difficult to fully utilize the real-time support capabilities of the grid cluster using fixed grid parameters or preset task allocation methods. Therefore, existing technologies still suffer from insufficient adaptation to transient disturbance evolution, insufficient flexibility in adjusting grid parameters, and insufficient coordination in the support behavior of multiple inverters. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a model-free parameter-based disturbance rejection control method and system for photovoltaic (PV) systems oriented towards source-grid interaction. This method employs a composite feedforward disturbance rejection control system based on disturbance observation and trend prediction, and dynamically adjusts grid parameters according to grid strength, disturbance characteristics, and inverter response. Furthermore, it integrates the real-time support capabilities of each inverter for dynamic task allocation and collaborative control, achieving adaptive collaborative support for distributed PV clusters against transient disturbances.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: One or more embodiments provide a model-free parameter disturbance rejection control method for photovoltaic systems oriented towards source-grid interaction, comprising the following steps: The system acquires real-time electrical data of each distributed photovoltaic inverter, performs disturbance observation, generates current disturbance estimates and disturbance trend predictions, constructs a composite feedforward compensation strategy based on the current disturbance estimates and disturbance trend predictions, and outputs disturbance suppression control commands. After each inverter executes the disturbance suppression control command, the distributed photovoltaic inverter response data, grid strength depth index and cluster disturbance feature set are obtained, and dynamic mapping is performed to generate a flexible grid weight coefficient set. Based on the set of flexible grid weight coefficients, the virtual inertia parameters, virtual damping parameters, and equivalent output impedance parameters of each distributed photovoltaic inverter are dynamically adjusted to obtain the adaptive grid control state. Based on the adaptive network control state, combined with the flexible network weight coefficient set and the real-time capability status of each inverter, the dynamic allocation and coordinated control of support tasks are carried out. The participation timing and support output of each inverter are coordinated and adjusted to generate cluster coordinated control commands, and the cluster coordinated support actions are executed.

[0007] A further technical solution involves acquiring real-time operational electrical data from each distributed photovoltaic inverter, observing disturbances, and generating current disturbance estimates and disturbance trend predictions, including the following steps: Based on real-time operating electrical data, transient decomposition of electrical behavior is performed to extract multi-frequency disturbance components, and the dynamic correlation structure between the multi-frequency disturbance components is fused to generate the current disturbance estimate. Based on the current disturbance estimate, sequence evolution inference and coupling strength evolution analysis are performed to construct the disturbance propagation trend topology and generate disturbance trend prediction.

[0008] A further technical solution involves performing sequence evolution deduction and coupling strength evolution analysis based on the current perturbation estimate, constructing a perturbation propagation trend topology, and generating a perturbation trend prediction, including: Based on the current disturbance estimate, sequence evolution is extrapolated to predict the future disturbance change state of a single inverter; Based on the disturbance coupling relationship between multiple inverters, the coupling strength evolution analysis is performed to construct the disturbance propagation trend topology; Based on the disturbance evolution trend of a single inverter and the topology of disturbance propagation trend, the variation law of the disturbance itself and the propagation effect between clusters are analyzed to generate the disturbance trend prediction.

[0009] A further technical solution involves constructing a composite feedforward compensation strategy based on the current disturbance estimate and the disturbance trend prediction, and outputting disturbance suppression control commands, including: Based on the current disturbance estimate and the disturbance trend prediction, an instantaneous disturbance compensation channel and a trend disturbance compensation channel are constructed respectively. The outputs of the two compensation channels are weighted and fused to form a multi-level disturbance suppression compensation structure, and the original compensation signal is calculated. The compensation intensity adjustment factor is determined based on the real-time dynamic response state of the inverter. The original compensation signal is then adaptively amplitude corrected, and the corrected compensation signal is converted into a disturbance suppression control command that can be executed in the inner loop of the inverter.

[0010] A further technical solution involves acquiring distributed photovoltaic inverter response data, grid strength depth indicators, and cluster disturbance characteristic sets after each inverter executes disturbance suppression control commands. This data is then dynamically mapped to generate a flexible grid construction weight coefficient set, including the following steps: The inverter response data after the disturbance suppression control command is executed is obtained, and dynamic response performance indicators characterizing the control effect are extracted within a preset response time window. The response dynamic profile of each inverter is constructed based on the response performance indicators. The grid intensity depth index and cluster disturbance feature set of each grid connection point are obtained. A spatiotemporal graph structure is constructed with each distributed photovoltaic inverter as a node and the electrical coupling relationship between inverters as an edge. The node features and edge features are aggregated based on the spatiotemporal graph neural network to generate grid-disturbance interaction feature vector. Based on the response dynamic profile of each inverter and the grid-disturbance interaction feature vector, a grid construction demand adaptation mapping is performed, the grid construction tendency corresponding to each inverter is calculated, and a set of flexible grid construction weight coefficients is generated.

[0011] A further technical solution involves dynamically adjusting the virtual inertia parameters, virtual damping parameters, and equivalent output impedance parameters of each distributed photovoltaic inverter based on the flexible grid weight coefficient set, to obtain the adaptive grid control state, including the following steps: Based on the set of flexible grid weight coefficients, the virtual inertia parameters and virtual damping parameters of each distributed photovoltaic inverter are dynamically adjusted, and spatial mapping is performed in combination with the positional relationship or electrical connection relationship of each inverter to generate a dynamic grid capacity distribution map. Based on the real-time grid interaction status and dynamic grid capacity distribution map, the active power-frequency droop parameters and reactive power-voltage droop parameters of each distributed photovoltaic inverter are dynamically adjusted to generate voltage and frequency support control parameters that are adapted to the current grid status. Based on the adjusted voltage and frequency support control parameters, the equivalent output impedance of each distributed photovoltaic inverter is dynamically adjusted, and the virtual inertia parameter, virtual damping parameter, droop control parameter and equivalent output impedance parameter are integrated in a coordinated manner to obtain the adaptive grid control state.

[0012] A further technical solution, based on adaptive network control state, combines the flexible network weight coefficient set with the real-time capability status of each inverter to dynamically allocate and coordinate the support tasks, generating cluster coordinated control commands, including the following steps: Based on the adaptive network control state, the flexible network weight coefficient set, and the real-time capability state of each inverter, the voltage support capability, frequency support capability, and oscillation suppression capability of each inverter are evaluated in multiple dimensions. The electrical topology relationship between the inverters is then used to generate a real-time support capability map of the cluster. The current active power support demand and reactive power support demand are determined based on the real-time grid interaction status. The task allocation ratio of each inverter is determined by combining the real-time support capability map of the cluster. Under the condition of meeting the real-time support capability constraints of each inverter, the support tasks are dynamically allocated to obtain the active power support quantity and reactive power support quantity corresponding to each inverter, thus forming the support task allocation result. Based on the task allocation results of each inverter, combined with its electrical topology location, real-time capability status and flexible network weight coefficient, the participation order, response rate and power reference value correction of each inverter are optimized in a spatiotemporal manner to generate cluster collaborative control commands.

[0013] One or more embodiments provide a model-free parameter disturbance rejection control system for photovoltaic source-grid interaction, comprising: The observation and compensation control module is configured to acquire real-time operating electrical data of each distributed photovoltaic inverter, perform disturbance observation, generate current disturbance estimate and disturbance trend prediction, construct a composite feedforward compensation strategy based on the current disturbance estimate and disturbance trend prediction, and output disturbance suppression control commands. The decision mapping module is configured to obtain distributed photovoltaic inverter response data, grid strength depth index and cluster disturbance feature set after each inverter executes disturbance suppression control command, perform dynamic mapping, and generate a set of flexible grid construction weight coefficients. The parameter adjustment module is configured as a set of flexible grid weight coefficients to dynamically adjust the virtual inertia parameters, virtual damping parameters and equivalent output impedance parameters of each distributed photovoltaic inverter to obtain an adaptive grid control state. The task allocation module is configured to dynamically allocate and coordinate the support tasks based on the adaptive network control state, combined with the flexible network weight coefficient set and the real-time capability status of each inverter. It coordinates the participation timing and support output of each inverter to generate cluster coordinated control commands and executes cluster coordinated support actions.

[0014] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein when executed by the processor, the computer instructions perform the steps in the above-described photovoltaic model-free parameter disturbance rejection control method for source-grid interaction.

[0015] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the above-described photovoltaic model-free parameter disturbance rejection control method for source-grid interaction.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves rapid suppression and early compensation of transient disturbances by observing and predicting the real-time electrical data of each distributed photovoltaic inverter and constructing a composite feedforward compensation. It also dynamically adjusts the virtual inertia, damping, and output impedance based on inverter response, grid strength, and cluster disturbance characteristics to improve the adaptability of grid control to changes in grid conditions. Furthermore, it dynamically allocates tasks and coordinates control based on the real-time support capabilities of each inverter to improve the utilization efficiency of cluster support resources and the overall collaborative anti-disturbance capability.

[0017] The advantages of the present invention, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0019] Figure 1 This is a flowchart of a photovoltaic model-free parameter disturbance rejection control method oriented towards source-grid interaction according to Embodiment 1 of the present invention; Figure 2 This is an example of a disturbance propagation trend topology diagram from Embodiment 1 of the present invention.

[0020] Figure 3 This is a schematic diagram of the flexible network weight coefficient mapping in Example 1 of the present invention.

[0021] Figure 4 This is an example of the virtual inertia and damping dynamic reshaping diagram of Embodiment 1 of the present invention.

[0022] Figure 5 This is a schematic diagram of a photovoltaic model-free parameter disturbance rejection control system oriented towards source-grid interaction according to Embodiment 2 of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] It should be noted that the terminology used herein is for describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0026] Example 1 In one or more of the technical solutions disclosed in the implementation methods, such as Figures 1 to 4 As shown, a model-free parameter disturbance rejection control method for photovoltaic systems with source-grid interaction includes the following steps: Step 1: Obtain real-time operating electrical data of each distributed photovoltaic inverter, perform disturbance observation, generate current disturbance estimate and disturbance trend prediction, and construct a composite feedforward compensation strategy based on the current disturbance estimate and disturbance trend prediction, and output disturbance suppression control command. Step 2: After each inverter executes the disturbance suppression control command, obtain the distributed photovoltaic inverter response data, grid strength depth index and cluster disturbance feature set, perform dynamic mapping, and generate a flexible grid construction weight coefficient set. Step 3: Based on the set of flexible grid weight coefficients, dynamically adjust the virtual inertia parameters, virtual damping parameters, and equivalent output impedance parameters of each distributed photovoltaic inverter to obtain the adaptive grid control state. Step 4: Based on the adaptive network control state, combined with the flexible network weight coefficient set and the real-time capability status of each inverter, the support tasks are dynamically allocated and coordinated. The participation timing and support output of each inverter are coordinated and adjusted to generate cluster coordinated control commands, and the cluster coordinated support actions are executed.

[0027] The method in this embodiment observes disturbances in the real-time electrical data of each distributed photovoltaic inverter to obtain the current disturbance estimate and the disturbance trend prediction. Based on this, a composite feedforward compensation strategy is constructed, enabling the control system to not only suppress disturbances that have already occurred, but also to implement compensation in advance based on the subsequent evolution trend of the disturbance, thereby improving the timeliness of response and suppression capability to rapid transient disturbances. On this basis, flexible grid construction weights are dynamically generated based on the inverter response data after disturbance suppression, grid strength, and cluster disturbance characteristics. The virtual inertia, virtual damping, and equivalent output impedance of each inverter are adjusted accordingly, so that the grid construction control parameters can adaptively change with the grid operating state and disturbance characteristics, reducing the problem of insufficient adaptability of fixed parameter control under varying grid strength conditions. Furthermore, based on the real-time capability status of each inverter, cluster support tasks are dynamically allocated and coordinated, allowing different inverters to participate in frequency, voltage, and oscillation support according to their available support capabilities. This avoids local capacity shortages or insufficient utilization of support resources caused by fixed allocation, thereby improving the dynamic adaptability, coordinated support capability, and operational stability of the distributed photovoltaic cluster under conditions of parameter uncertainty and transient disturbances.

[0028] The photovoltaic model-free parameter disturbance rejection control method described in this embodiment adopts a technical path that combines data-driven disturbance observation, flexible grid configuration dynamic mapping decision-making, and cluster collaborative control. It can realize adaptive disturbance rejection control and collaborative stability support of photovoltaic clusters in complex power grid environments. Each step is described in detail below. In step 1, real-time operating electrical data of each distributed photovoltaic inverter is obtained, which may include electrical quantity data such as voltage, current, power and frequency of the inverter grid connection point collected at a set sampling frequency; The real-time operating electrical data collected by each distributed photovoltaic inverter at continuous sampling times is represented as: ; in, Indicates the first One distributed photovoltaic inverter as well as They represent the first The voltage, current, active power, reactive power, and frequency data of an inverter at time t.

[0029] Optionally, real-time operating electrical data such as voltage and current at each inverter grid connection point can be collected. The sampling frequency can typically be set between 10kHz and 50kHz to ensure that microsecond-level transient processes can be captured.

[0030] Since power grid transient disturbances typically exhibit complex changes across multiple time scales and frequencies, it is difficult to accurately distinguish different sources of disturbance using raw electrical data directly. Therefore, disturbance observation processing is required for real-time operating electrical data. In step 1, real-time operating electrical data of each distributed photovoltaic inverter is acquired, disturbance observation is performed, and current disturbance estimates and disturbance trend predictions are generated, including the following steps: Step 11: Perform transient decomposition of electrical behavior based on real-time operating electrical data, extract multi-frequency disturbance components, and fuse the dynamic correlation structure between the multi-frequency disturbance components to generate the current disturbance estimate; Specifically, empirical mode decomposition (EMD) or variational mode decomposition (VMD) methods are used to decompose the original electrical signals, such as voltage and current at the grid connection point, into multiple intrinsic mode components at different frequency scales. ; in, This represents the original electrical signal of the i-th inverter. This represents the k-th intrinsic mode component. Let K represent the residual term, and K represent the number of modal components obtained from the decomposition.

[0031] In the above implementation, the acquired non-stationary signal, such as the grid connection point voltage, undergoes electrical behavior transient decomposition. The original signal is decomposed into a series of intrinsic mode components and a residual term. Each intrinsic mode represents an oscillation mode in different frequency bands from high to low, and the set of these components is the extracted multi-band disturbance component. After decomposition, the original electrical signal is converted into multiple frequency band disturbance components: including high-frequency disturbance components, mid-frequency disturbance components, and low-frequency disturbance components. Among them, the high-frequency disturbance component is used to characterize rapid changing processes such as switching action and transient impacts, such as power device switching noise and lightning strikes; the medium-frequency disturbance component is used to characterize oscillating transient processes; and the low-frequency disturbance component is used to characterize slow changing processes such as voltage fluctuations and changes in the power grid background.

[0032] Subsequently, dynamic correlation analysis was performed on each multi-band disturbance component. Based on the energy proportion of each disturbance component within the current time window and its cross-correlation with the reference signal under normal operating conditions, the corresponding dynamic weighting coefficients were determined. Based on dynamic weighting coefficients, the disturbance components of each frequency band are fused to obtain the current disturbance estimate. Then, the current disturbance estimate of inverter j at time t after fusion is calculated. The formula is expressed as: ; in, This is the i-th intrinsic mode component; It is the dynamic weighting coefficient corresponding to this component; Weighting coefficient The determination is based on the energy proportion of each component within the current time window or the degree of cross-correlation with the baseline signal of normal system operation, thereby ensuring that the frequency band component with the greatest impact on the current disturbance dominates the estimation. Through the above processing, the main disturbance components in the complex electrical waveform are extracted and quantified, achieving an accurate characterization of the current transient disturbance state.

[0033] For example, in a scenario comprising three photovoltaic inverters PV1, PV2, and PV3, real-time operational electrical data of the grid-connected points is acquired at a frequency of 20 kHz to generate disturbance observation results. When a remote fault causes voltage oscillation at the PV2 grid-connected point, transient electrical behavior decomposition is performed on the acquired voltage signal. Variational mode decomposition (VMD) is used to decompose it into five intrinsic mode components c1 to c5 and a residual. Analysis reveals that c1 and c2 are high-frequency switching noise, c3 is a 400 Hz subsynchronous oscillation component characterized by the fault, c4 is a 50 Hz power frequency component, and c5 is a low-frequency background fluctuation. To quantify the disturbance, the energy proportion of each component within a 20-millisecond time window is calculated, generating dynamic weighting coefficients. The energies of each component within this time window are E1=0.2J, E2=0.3J, E3=5.5J, E4=1.0J, and E5=0.5J, with a total energy of 7.5J. Therefore, the weighting coefficients for each component are w1=0.027, w2=0.040, w3=0.733, w4=0.133, and w5=0.067. At time t0, the instantaneous amplitudes of each component are 5V, -8V, -25V, 311V (power frequency fundamental peak value, undisturbed, removed), and -2V, respectively. The current disturbance estimate of PV2 is then calculated. for: ; Calculations show that the weighted contribution of the main disturbance component c3 is -18.325V, which accurately quantifies the main impact of the current subsynchronous oscillation on the voltage. Through transient decomposition of electrical behavior and dynamic weighted fusion, the most significant disturbance component can be accurately extracted and quantified from the complex power grid signal, providing a clear disturbance input for subsequent compensation control.

[0034] After obtaining the current disturbance estimates of each distributed photovoltaic inverter, we further analyze the evolution of the disturbance within a single inverter and its propagation relationship among multiple inverters, and predict the development trend of the disturbance in the future period. Step 12: Based on the current perturbation estimate, perform sequence evolution deduction and coupling strength evolution analysis to construct the perturbation propagation trend topology and generate perturbation trend predictions, including: Step 121: Based on the current disturbance estimate, perform sequence evolution deduction to predict the future disturbance change state of a single inverter.

[0035] Specifically, after obtaining the current disturbance estimate for each distributed photovoltaic inverter, a sequence analysis is performed on the disturbance estimate over a continuous time period to obtain the development trend of the disturbance itself. The current disturbance estimate from multiple consecutive sampling times is constructed into a disturbance time series: ; A sliding time window can be set, which can be set from 50 to 200 milliseconds, to capture segments of perturbation evolution across time scales.

[0036] Based on the above disturbance time series, the changes in disturbance amplitude, growth rate and decay trend are analyzed to extract the evolution law of the disturbance itself and predict the disturbance change state of a single inverter in the future.

[0037] Step 122: Perform coupling strength evolution analysis based on the disturbance coupling relationship between multiple inverters, and construct the disturbance propagation trend topology; Specifically, based on the information correlation between the disturbance estimation sequences of each inverter, the transfer entropy method is used to calculate the disturbance coupling strength between any two inverters: ; in, It represents the information coupling strength of the disturbance propagated from inverter i to inverter j at time t, and quantifies the possibility and strength of the disturbance from one inverter to another at time t.

[0038] Based on the coupling strength between multiple inverters, a time-varying coupling strength evolution map is constructed. : ; Where N represents the number of photovoltaic inverters.

[0039] In the above implementation, for any pair of inverters i and j, a nonlinear causal analysis tool, such as transfer entropy, is used for calculation. By continuously calculating the transfer entropy between all inverter pairs within a sliding time window, the propagation direction and intensity attenuation of the disturbance among multiple inverters are dynamically captured, forming an N×N dimensional coupling strength matrix that evolves over time, where N is the number of inverters, and the matrix elements represent the disturbance influence intensity of one inverter on another at time t. This time-series matrix constitutes the coupling strength evolution map.

[0040] Furthermore, a threshold screening is performed on the coupling strength evolution graph. A coupling strength threshold is set to remove low-strength correlations, and the remaining valid correlations are converted into a disturbance propagation trend topology: topology nodes represent distributed photovoltaic inverters; directed edges represent disturbance propagation directions; and edge weights represent disturbance propagation strengths. Optionally, the coupling strength threshold can be set to 0.6-0.8; In this embodiment, a coupling strength threshold is set for the coupling strength evolution graph A(t), and elements below this threshold are set to zero to filter out weak correlations and noise interference. The processed matrix is ​​treated as an adjacency matrix of a weighted directed graph, with the inverter as a node and non-zero elements... This represents a directed edge from node j to node i, with weight . Based on this topology, key nodes, main propagation paths, and potential oscillating loops of disturbance propagation are identified, revealing the disturbance propagation patterns within the cluster. Using methods such as degree centrality and betweenness centrality, the graph is analyzed to identify key nodes along propagation paths with high out-degree (i.e., main disturbance sources) or high betweenness centrality (i.e., key propagation paths). Simultaneously, loop detection algorithms are used to discover potential oscillating loop structures. This analyzed and structured dynamically weighted directed graph represents the disturbance propagation trend topology. Figure 2 As shown, each node represents a photovoltaic inverter, and the directed arrows between nodes indicate the direction of disturbance propagation. Figure 2 The black node PV2 is a key node, representing a high centrality or betweenness in the disturbance propagation network, and is the main source or inevitable path for disturbance propagation. Finally, a disturbance trend prediction is generated based on this topology. By combining the current disturbance estimate, coupling strength, and evolution rate (i.e., the rate of change of coupling strength) of each node, the propagation range and impact level of the disturbance in the future are predicted.

[0041] Step 123: Based on the single inverter disturbance evolution trend obtained in Step 121 and the disturbance propagation trend topology obtained in Step 122, analyze the disturbance's own variation law and the inter-cluster propagation effect, and generate the disturbance trend prediction quantity.

[0042] Specifically, based on the disturbance intensity, coupling intensity, and rate of change of coupling intensity of each node in the disturbance propagation trend topology, the propagation range, affected nodes, degree of impact, and propagation level of the disturbance in the future time period are predicted, and finally the disturbance trend prediction quantity is generated. Through the above steps, the analysis of single-point disturbance states is extended to the prediction of cluster disturbance propagation trends, providing predictive disturbance inputs for subsequent composite feedforward compensation control.

[0043] For inverter j in the cluster, the predicted disturbance trend at time t+Δt is... , can be represented as: ; in, This represents the start time of the current analysis window; This is the length of the window; if the window contains 50 sampling points, then the value is 50. Let J represent the set of all neighboring inverters that have a direct impact on inverter j in the disturbance propagation trend topology; It is the attenuation constant; The electrical distance between inverters i and j, such as impedance magnitude; Let be the standardized, filtered causal coupling strength from inverter i to inverter j at time t. This formula integrates the self-evolution trend of the disturbance and the propagation effect from other units within the cluster. The final set of predicted values ​​for each node constitutes the complete disturbance trend prediction.

[0044] For example, after obtaining the time series of the current disturbance estimates for each inverter, the disturbance trend prediction is generated. Taking PV2 as an example, its disturbance sequence over the past 2.5 milliseconds (50 sampling points) is analyzed to predict the trend 10 milliseconds later. This prediction combines two aspects: first, analyzing the evolution of the disturbance itself through sequence evolution deduction; and second, analyzing the propagation impact of other inverters within the cluster through coupling strength evolution analysis. Assume that at time t and t-1, the current disturbance estimates for PV2 are -25V and -22V, respectively. Simultaneously, the disturbance propagation trend topology constructed by calculating the transfer entropy shows a strong coupling relationship of 0.75 from PV1 to PV2. Based on the propagation effect formula including an electrical distance of 0.1Ω and an attenuation constant of 5 (1 / Ω), the -30V disturbance from PV1 will have a superimposed impact on PV2. Finally, the disturbance trend prediction for PV2 is obtained by adding its own evolution trend to the propagation effect. : ; For simplicity, the above estimations are based solely on data from times t and t-1. By integrating time series forecasting with physical causal propagation, it is possible to provide early warning of a complex disturbance far more severe than the current one, buying valuable time to shift from passive response to proactive defense.

[0045] Step 1, the process of constructing a composite feedforward compensation strategy based on the current disturbance estimate and the disturbance trend prediction, and outputting disturbance suppression control commands, includes: Step 101: Construct an instantaneous disturbance compensation channel and a trend disturbance compensation channel based on the current disturbance estimate and the disturbance trend prediction, respectively, and perform weighted fusion of the outputs of the two compensation channels to form a multi-level disturbance suppression compensation structure, and calculate the original compensation signal; Specifically, the current disturbance estimate characterizes the current disturbance state and generates an instantaneous reverse compensation component based on the current disturbance amplitude to quickly offset the impact of the current disturbance; the disturbance trend prediction characterizes the disturbance change trend over a future time period and generates an advance compensation component based on the predicted disturbance change to adjust the inverter output response in advance. This is for the original compensation signal of inverter j at time t. , can be represented as: ; in, and These are dimensionless current and future disturbance compensation coefficients, whose values ​​can be optimized through offline simulation or tuned through online self-learning algorithms such as reinforcement learning, typically ranging from 0.8 to 1.2. This constructs a hierarchical, multi-level disturbance suppression and compensation structure with both real-time response and look-ahead adjustment, used to calculate the initial compensation signal for each inverter.

[0046] The above implementation method couples and compensates the acquired current disturbance estimate with the disturbance trend prediction. This fusion is not a simple linear superposition, but rather constitutes a dual-channel compensation structure. The current disturbance estimate is mainly used to generate an immediate reverse compensation component to quickly respond to the disturbance that has already occurred; the disturbance trend prediction is used to generate a proactive compensation component to anticipate upcoming disturbance changes. Through this processing, the current disturbance suppression requirements and future disturbance development trends are fused to form a composite compensation signal that simultaneously possesses rapid response capabilities and predictive adjustment capabilities.

[0047] Step 102: Determine the compensation intensity adjustment factor based on the real-time dynamic response state of the inverter, perform adaptive amplitude correction on the original compensation signal obtained in Step 101, and convert the corrected compensation signal into a disturbance suppression control command that can be executed in the inner loop of the inverter. Specifically, the current operating status parameters of the inverter are obtained, a dynamic compensation intensity adjustment factor is calculated based on the obtained operating status parameters, and the compensation intensity is adjusted through the dynamic compensation intensity adjustment factor. Optionally, the current operating status parameters of the inverter can be obtained, including: DC side bus voltage margin, AC side output current margin, and power device temperature status, etc. For calculating the compensation intensity adjustment factor at time t The calculation formula is: ; in, This is the real-time measured value of the DC-side bus voltage; The minimum permissible DC voltage is 85% of the rated value; This is the rated DC voltage, i.e., the rated DC bus voltage; Let t be the effective value of the inverter output current. The maximum allowable output current of the inverter is 120% of the rated current. This refers to the junction temperature of the power device. Rated operating temperature; This is the temperature protection threshold.

[0048] When the inverter approaches its maximum output power limit or its internal temperature approaches the alarm threshold, the compensation intensity adjustment factor will adaptively decrease, and vice versa. This adjustment mechanism ensures the safety and feasibility of the compensation action.

[0049] Furthermore, the magnitude of the original compensation signal is adjusted using the calculated compensation intensity adjustment factor to obtain the disturbance suppression control command; for voltage compensation, the original compensation signal is multiplied by the compensation intensity adjustment factor to obtain the final compensation voltage command. : ; The compensation voltage command is decomposed into d-axis and q-axis components in the dq rotating coordinate system. Through the equivalent admittance stage, it is converted into two dynamic correction quantities for the inverter's inner-loop current reference value. These are used to calculate the active current correction quantity along the d-axis and the reactive current correction quantity along the q-axis of inverter j at time t. and , respectively represented as: ; in, and These are the equivalent resistance and equivalent reactance at the inverter's grid connection point, respectively. This represents the phase angle of the disturbance voltage. These two correction values... and These are the final output disturbance suppression control commands, which are directly superimposed on the current reference value generated by the original power control loop of the inverter. By controlling the inverter to output the corresponding compensation current, the coordinated compensation and calibration of grid-side disturbances are achieved.

[0050] For example, after obtaining the current disturbance estimate of PV2 (-25V) and the predicted disturbance trend (-42.05V), a disturbance suppression control command is generated for PV2. First, coupled compensation fusion is performed to form the original compensation signal. The current disturbance compensation coefficient is set to 1.0, and the future disturbance compensation coefficient is set to 0.9. The original compensation signal... (V). This negative value indicates that a compensation voltage opposite to the disturbance needs to be generated. Next, the adaptive adjustment mechanism for disturbance rejection compensation strength is activated. The status of PV2 is monitored in real time, obtaining the following: DC side voltage 780V, rated DC voltage 800V, minimum allowable voltage 680V; output current RMS value 80A, maximum allowable current 120A; power device junction temperature 95℃, rated operating temperature 75℃, temperature protection threshold 125℃. The compensation strength adjustment factor is calculated according to the formula. Finally, a correction command for the current loop is generated. If the disturbance voltage phase angle is 90 degrees at this moment, it manifests as a reactive disturbance, and the grid-connected equivalent reactance is 0.5Ω. Then the active current correction amount... Reactive current correction (A). This -70A reactive current correction is the final output disturbance suppression control command, which will be superimposed on the original reactive current reference value. Through a multi-level compensation structure, the command not only responds quickly to the current disturbance of -25V, but also prepares in advance to deal with the future disturbance of -42.05V, achieving a dynamic balance between disturbance suppression and equipment safety.

[0051] Step 2 involves obtaining the response data, grid strength depth index, and cluster disturbance characteristic set formed by real-time operating electrical data of each inverter after each distributed photovoltaic inverter executes disturbance suppression control commands. This data describes the inverter's current disturbance response capability, grid support requirements, and cluster disturbance status, and generates a corresponding flexible grid construction weight coefficient set through dynamic mapping decision-making.

[0052] In step 2, based on the execution effect data of the disturbance suppression control command, combined with the obtained grid strength depth index of each grid connection point in the cluster, and fused with the cluster disturbance feature set formed by the real-time operating electrical data, the execution mode dynamic mapping decision is made to generate a flexible grid construction weight coefficient set. Step 2: After each inverter executes the disturbance suppression control command, obtain the distributed photovoltaic inverter response data, grid strength depth index, and cluster disturbance characteristic set, perform dynamic mapping, and generate a flexible grid construction weight coefficient set, including the following steps: Step 21: Obtain inverter response data after the disturbance suppression control command is executed, extract dynamic response performance indicators that characterize the control effect within a preset response time window, and construct the response dynamic profile of each inverter based on the response performance indicators. The response data of each distributed photovoltaic inverter comes from the short-time response process after the execution of the disturbance suppression control command. Each set of response data corresponds to the actual control effect of an inverter under the current disturbance state. The inverter response data can include dynamic change data such as grid connection point voltage, current, frequency, active power and reactive power. Specifically, within a short time window after the disturbance suppression control command is issued, such as 100-500 milliseconds, the execution effect data such as voltage and current at the inverter grid connection point are obtained.

[0053] Furthermore, within a preset response time window, the response data is analyzed to extract response performance indicators reflecting the inverter's dynamic response performance, including but not limited to voltage sag recovery time after disturbance suppression, voltage overshoot, and frequency recovery rate. These response performance indicators are then used to construct a response dynamic feature vector, i.e., a response dynamic profile. : ; in, This represents the dynamic profile of the response of the j-th inverter. This represents the m-th response performance index.

[0054] These metrics are integrated into a single vector, the dynamic response profile, which objectively characterizes the inverter's dynamic response quality under specific disturbances using the current control strategy. This step converts the actual response effect after the disturbance suppression control command is executed into quantifiable dynamic performance characteristics, providing input for subsequent network configuration adaptation.

[0055] Step 22: Obtain the grid strength depth index and cluster disturbance feature set of each grid connection point. Construct a spatiotemporal graph structure with each distributed photovoltaic inverter as a node and the electrical coupling relationship between inverters as an edge. Aggregate the node features and edge features based on the spatiotemporal graph neural network to generate grid-disturbance interaction feature vector. Specifically, the grid strength depth index corresponding to each grid connection point is first obtained. The grid strength depth index is used to characterize the grid strength and power quality status at the connection location of each distributed photovoltaic inverter. It is obtained by detecting and calculating the operating status of the grid connection point. The grid strength depth index is calculated by collecting voltage and current change data at the grid connection point and combining them with grid parameters, which may include short-circuit ratio (SCR), equivalent grid impedance, voltage stability margin, background harmonic distortion rate, and three-phase imbalance. In some embodiments, the cluster disturbance feature set is used to describe the overall dynamic state of the current photovoltaic cluster after being disturbed, and may include disturbance amplitude, disturbance frequency components, disturbance change trend, disturbance propagation direction, disturbance impact range and / or the degree of disturbance to each inverter. The disturbance features corresponding to each inverter are summarized to form the cluster disturbance feature set C: ; in, Let N represent the disturbance characteristics corresponding to the j-th inverter, and N represent the number of photovoltaic inverters.

[0056] Furthermore, based on the grid strength depth index and cluster disturbance feature set, a spatiotemporal graph structure is constructed with inverters as nodes and electrical coupling relationships as edges. In this structure, the graph nodes represent each distributed photovoltaic inverter, and the node features include the grid strength depth index; the graph edges represent the electrical coupling relationships between inverters, and the edge features include disturbance propagation intensity, disturbance direction, and coupling change trend. Based on the constructed spatiotemporal graph structure, node and edge features are extracted and input into a pre-trained spatiotemporal graph neural network (GNN) for fusion processing. The GNN aggregates neighbor information through graph convolutional layers and captures dynamic relationships using a temporal attention mechanism, outputting a power grid-disturbance interaction feature vector. ; in, The term d represents the fused grid-disturbance interaction feature, used to characterize the current grid strength, disturbance state, and cluster propagation relationship.

[0057] Optionally, during the training of the spatiotemporal graphical neural network (GNN), a data pair consisting of historical power grid disturbance data and corresponding optimal control strategies is used to conduct supervised learning of the overall intensity, main frequency components, and spatial distribution characteristics of the disturbances, with the goal of minimizing reconstruction error and strategy matching loss.

[0058] Step 23: Fuse the dynamic profiles of each inverter response obtained in Step 21 with the grid-disturbance interaction feature vectors obtained in Step 22, and perform grid construction demand adaptation mapping based on the fused features to calculate the grid construction tendency corresponding to each inverter and generate a set of flexible grid construction weight coefficients. After obtaining the response dynamic profile of each distributed photovoltaic inverter and the grid-disturbance interaction feature vector, the above features are input into the grid construction mode adaptation model to dynamically calculate the grid construction requirements corresponding to the current operating state of each inverter and generate the flexible grid construction weight coefficient corresponding to each inverter. It is possible to respond to dynamic profiles. Interactive feature vectors of power grid and disturbance The input is a pre-trained grid configuration pattern adaptation model. This model can be a deep neural network, a fuzzy logic reasoning model, or a reinforcement learning decision model. It is trained using historical power grid disturbance scenario data and corresponding optimal grid configuration control strategies to establish a mapping relationship between input features and grid configuration control requirements.

[0059] For the j-th inverter, its flexible grid weighting coefficient The calculation is as follows: in: This represents the flexible network weighting coefficient of the j-th inverter; This represents the basic offset of the model, used to characterize the inverter's default network configuration tendency in the absence of input features; This represents the dynamic characteristic of the m-th response of the j-th inverter; The weight parameters represent the dynamic characteristics of the response and are used to characterize the degree of influence of different response performance indicators on network construction requirements. This represents the d-th feature in the power grid-disturbance interaction feature vector; Indicates the i-th power grid-disturbance characteristic pair Influence weight; M represents the number of dynamic features of the response; D represents the dimension of the power grid-disturbance interaction characteristics.

[0060] When the flexible grid weighting coefficient approaches 1, it indicates that the inverter should be more inclined towards the grid-connected GFM operating mode, actively providing voltage and frequency support; when it approaches 0, it is more inclined towards the grid-following GFL operating mode, mainly for power injection. Figure 3 As shown in the figure, the X-axis represents grid strength, measured by the short-circuit ratio (SCR), and the Y-axis represents disturbance level. When the grid strength is low (i.e., a weak grid), the SCR value is small, and the disturbance level is high, the color tends to be darker, indicating that the weight coefficient is close to 1, and the inverter is more inclined to adopt a grid configuration mode to actively support the grid. For each inverter in the cluster, a corresponding weight coefficient is generated independently. Integrating these weight coefficients yields the flexible grid configuration weight coefficient set.

[0061] For example, after PV2 executes a -70A reactive current compensation command, its effect is evaluated and a decision is made on whether to adjust the grid configuration. First, execution effect data is collected, and a dynamic response profile is extracted. Analysis shows that within 100 milliseconds after the compensation command is issued, the voltage dip recovery time at the PV2 grid connection point is 150ms, still with a 5% voltage overshoot. This dynamic response profile is quantified as a performance vector R2=[150,0.05]. Simultaneously, a grid-disturbance interaction image is constructed. The grid strength depth index at the grid connection point is obtained; the short-circuit ratio is measured at 2.5, indicating a weak grid; the total background harmonic distortion rate is 4.5%, indicating some power quality issues. These indicators constitute the grid strength depth index S2=[2.5,0.045]. Meanwhile, a cluster disturbance image compiled from data from all inverters shows that the current disturbance is mainly a grid-wide subsynchronous oscillation with a medium intensity level. The cluster disturbance feature set is represented as C=[medium,sub-sync], where medium indicates a moderate disturbance intensity level and sub-sync indicates a subsynchronous oscillation disturbance type. A graph neural network (GNN) is used to fuse S2 and C to generate a grid-disturbance interaction map G. Finally, the response dynamic profile R2 and the grid-disturbance interaction features G are input into a pre-trained grid configuration adaptation model, outputting the flexible grid configuration weight coefficients of PV2. The value is 0.7. Similarly, calculations are performed on PV1 and PV3, yielding... PV1 is closer to the disturbance source and requires stronger support. The final set of flexible network weight coefficients is {0.8, 0.7, 0.4}.

[0062] In step 3 of this embodiment, after obtaining the set of flexible grid construction weight coefficients, the internal control parameters of the inverters are dynamically adjusted according to the grid construction requirements of each inverter, so that each inverter has the grid construction capability to match the current grid strength and disturbance state. In this embodiment, the adaptive grid construction control state refers to the control parameter combination state formed by each distributed photovoltaic inverter after dynamically adjusting the virtual inertia parameter, virtual damping parameter, voltage-frequency droop parameter, and equivalent output impedance parameter according to the flexible grid construction weight coefficients, the real-time interaction state of the grid, and the disturbance level. It is used to characterize the current grid construction support capability of each inverter and its degree of matching with the grid state. Step 3: Based on the flexible grid weight coefficient set, dynamically adjust the virtual inertia parameters, virtual damping parameters, and equivalent output impedance parameters of each distributed photovoltaic inverter to obtain the adaptive grid control state, including the following steps: Step 31: Based on the set of flexible grid weight coefficients, dynamically adjust the virtual inertia parameters and virtual damping parameters of each distributed photovoltaic inverter, and perform spatial mapping in combination with the positional relationship or electrical connection relationship of each inverter to generate a dynamic grid capacity distribution map. Step 311: Based on the flexible grid weighting coefficient, dynamically adjust the virtual inertia parameters and virtual damping parameters of each distributed photovoltaic inverter to obtain the adjusted virtual inertia parameters and virtual damping parameters. Specifically, this step involves constructing a dynamic network capability distribution map. This is based on flexible network weighting coefficients. The virtual inertia and damping characteristics of each inverter are adjusted in a coordinated manner. Virtual inertia and damping are key parameters for grid-connected inverters to mimic synchronous generators and provide frequency support. This reshaping process is achieved through a parameter mapping function based on the flexible grid weighting coefficient. Virtual inertia of inverter j With virtual damping The adjustment process is represented as follows: ; in, This represents the virtual inertia parameter after adjustment for the j-th inverter; This represents the virtual damping parameter of the j-th inverter after adjustment; and These are the adjustment ranges of the virtual damping coefficient, representing the maximum and minimum values, used to adjust the damping ratio of the frequency response; Step 312: Map the adjusted virtual inertia parameters and virtual damping parameters of all inverters in the cluster according to the inverter position relationship or electrical connection relationship to form a dynamic network capability distribution map; like Figure 4As shown, the changes in virtual inertia and virtual damping of three photovoltaic inverters (PV1, PV2, and PV3) after obtaining different flexible grid construction weight coefficients of 0.8, 0.7, and 0.4, respectively, were compared. Visualizing the adjusted virtual inertia and virtual damping values ​​of all inverters in the cluster on a geographical or electrical topology creates a dynamic grid construction capability distribution map, which intuitively displays the current distribution pattern of inertia and damping support capabilities within the cluster.

[0063] Step 32: Based on the real-time grid interaction status and dynamic grid capacity distribution map, dynamically adjust the active power-frequency droop parameters and reactive power-voltage droop parameters of each distributed photovoltaic inverter to generate voltage and frequency support control parameters that are adapted to the current grid status. First, the real-time grid interaction status is acquired, including deviations and rates of change in grid frequency and voltage. Then, combined with a dynamic grid capacity distribution map, the voltage-frequency support behavior of each inverter is dynamically adjusted. Specifically, by adjusting the voltage-frequency droop control parameters of the inverters, autonomous power allocation among different inverters is achieved. Droop control is the core mechanism for autonomous power allocation by grid-connected inverters. The calculation involves determining the output angular frequency of inverter j in the adjusted voltage-frequency droop characteristics. and voltage amplitude The formula is adjusted as follows: ; in, and These are the rated angular frequency and AC voltage amplitude; and It refers to the active and reactive power output of the inverter; The active power droop factor is expressed in rad / (s·pu). The smaller the value, the stronger the frequency support. This is the reactive power droop factor; the smaller the value, the stronger the voltage support.

[0064] Adjust the above control parameters according to the grid disturbance status, and the output angular frequency of inverter j is adjusted. and voltage amplitude Including the output angular frequency of inverter j Voltage amplitude Virtual inertia With virtual damping This enables different inverters to have varying degrees of voltage and frequency support capabilities, forming an adaptive grid construction strategy.

[0065] Step 33: Based on the adjusted voltage and frequency support control parameters, dynamically adjust the equivalent output impedance of each distributed photovoltaic inverter, and integrate the virtual inertia parameter, virtual damping parameter, droop control parameter and equivalent output impedance parameter to obtain the adaptive grid control state. Specifically, by adjusting the parameters of the inverter's voltage control loop and current control loop, the inverter's equivalent output impedance can be dynamically varied between different resistive, inductive, or capacitive characteristics: when it is necessary to enhance grid connection capability, the inductive characteristic of the output impedance is increased, so that the inverter exhibits power exchange characteristics similar to a synchronous generator; when it is necessary to reduce support strength, the output impedance characteristics are adjusted so that the inverter maintains stable power output.

[0066] For example, the output impedance of the inverter can be shaped by adjusting the proportional-integral (PI) parameters of the voltage and current loops. This dynamic coordination of the output impedance ensures stable power exchange between the inverters within the cluster and good interaction with the external power grid. After this series of dynamic adjustments and coordination of parameters, the entire photovoltaic cluster reaches a stable operating state that matches the current grid-disturbance environment.

[0067] Finally, the resulting adaptive network control state can be expressed as: ; in, These represent the virtual inertia, virtual damping, active droop, reactive droop, and equivalent output impedance of the j-th inverter, respectively.

[0068] For example, based on the generated flexible network weight coefficient set {0.8, 0.7, 0.4}, the control parameters of the three inverters PV1, PV2, and PV3 are dynamically adjusted. First, the virtual inertia and damping characteristics are collaboratively reshaped to construct a dynamic network capability distribution map. If the maximum virtual inertia parameter of the inverter is 4s, the virtual damping coefficient adjustment range is... =20(pu∙s), =100(pu∙s). Taking PV2 as an example, its weight is 0.7. Adjusted virtual inertia. ( Adjusted virtual damping (pu∙s). Similarly, PV1 can be calculated. =3.2s, =84; PV3 =1.6s, =52. These parameter values ​​are marked on the cluster topology map to form a dynamic network capability distribution map, showing that PV1 and PV2 regions form strong inertial support centers. Next, an adaptive network construction strategy is generated. A grid frequency drop to 49.90Hz due to disturbance is detected. To provide frequency support, the droop coefficient needs to be adjusted. For PV2, with its higher network density, it should bear more support responsibility; therefore, its active power droop coefficient is reduced. If the adjusted... The value is 0.005. PV2 outputs 1.0 pu of active power at this time, then its output angular frequency is... (rad / s), thereby actively raising the frequency. Finally, dynamic coordination of the output impedance characteristics is performed, making the output impedance of PV1 and PV2 more inductive to simulate the characteristics of a synchronous generator, while PV3 maintains a relatively high impedance characteristic to ensure the stability of power exchange within the cluster. Through flexible grid weighting coefficients, the abstract "mode switching" decision is concretized into the adjustment of a series of key control parameters such as virtual inertia and damping, enabling the photovoltaic cluster to transform from a pure power source into a virtual synchronous generator cluster with dynamic stability characteristics that can actively participate in grid regulation.

[0069] Step 4: Based on the adaptive network control state, and combining the flexible network weight coefficient set with the real-time capability status of each inverter, dynamically allocate and coordinate the support tasks, and generate cluster coordinated control commands, including the following steps: Step 41: Based on the adaptive network control state, the flexible network weight coefficient set, and the real-time capability state of each inverter, conduct a multi-dimensional evaluation of the voltage support capability, frequency support capability, and oscillation suppression capability of each inverter, and combine the electrical topology relationship between the inverters to generate a real-time support capability map of the cluster. The real-time capability status of each distributed photovoltaic inverter is used to characterize the actual support margin that the inverter can currently provide. The real-time capability status can include the available power on the DC side, the current thermal margin of the inverter, and the DC side capacitor voltage level. Combined with the flexible grid construction weight coefficient, the current voltage support, frequency support, and power oscillation suppression capability of the photovoltaic cluster are evaluated.

[0070] A photovoltaic cluster refers to an operating cluster consisting of multiple distributed photovoltaic inverters, which are the operating set of multiple distributed photovoltaic inverters and their electrical connections that participate in the current coordinated control. This step, based on the stable operation of the photovoltaic cluster, performs a multi-dimensional capability coupling assessment by linking the flexible grid weighting coefficient set with the real-time capability status of each inverter. This multi-dimensional capability coupling assessment evaluates not only the capability of individual inverters but also the overall cluster capability after coupling through the grid configuration defined by the flexible grid weighting coefficient set. For example, the total inertial support capability provided by the cluster is not simply the sum of the virtual inertia of individual inverters but also depends on the electrical connections and impedances between them.

[0071] One specific implementation method involves calculating voltage support capability indicators. , represented as: ; in, This represents the current real-time voltage of the DC-side capacitor of inverter i; This is the rated voltage of the DC-side support capacitor of the inverter; This represents the available reactive power capacity of inverter i in its current real-time operating state. This represents the rated reactive power capacity of inverter i; This represents the number of inverters.

[0072] Similarly, the active frequency support capability, reactive voltage support capability, and oscillation suppression capability can be calculated separately.

[0073] By associating the support capability indicators of each inverter with the position of the inverter in the electrical topology, a real-time support capability map of the cluster is formed.

[0074] Furthermore, the cluster voltage support capability index and the local support capability corresponding to each inverter are correlated with the positional relationship of the inverter in the geographical location or electrical topology to form a real-time support capability map of the cluster.

[0075] Step 42: Determine the current active power support demand and reactive power support demand based on the real-time grid interaction status, and determine the task allocation ratio of each inverter in combination with the real-time support capability map of the cluster. Under the condition of meeting the real-time support capability constraints of each inverter, the support tasks are dynamically allocated to obtain the active power support amount and reactive power support amount corresponding to each inverter, and the support task allocation result is formed. Step 421: Based on the real-time power grid interaction status, identify the current active and reactive power grid support tasks; Specifically, it acquires real-time grid interaction status parameters such as grid connection point voltage, system frequency, and power oscillation status, and determines the support tasks to be executed based on preset judgment conditions.

[0076] When the grid frequency is detected to be lower than a set threshold, such as 49.8Hz, a frequency support task is determined; when the grid connection point voltage is detected to be lower than a set range, a voltage recovery task is determined; when continuous power oscillation is detected, an oscillation suppression task is determined.

[0077] Step 422: Based on the real-time support capability map of the cluster, dynamically allocate the support capabilities of the support tasks based on the inverter.

[0078] Based on the total demand corresponding to the current grid support tasks and the real-time support capability map of the power grid cluster, the task allocation ratio for each inverter is calculated, and the initial support task volume for each inverter is determined according to the task allocation ratio. Taking reactive power support tasks as an example, when the grid requires a total reactive power support volume... At that time, the corresponding task allocation is determined based on the real-time support capability of each inverter. : ; in, This represents the reactive power support task allocated to the i-th inverter, where N represents the number of inverters. Assign a support ratio to the i-th inverter, and calculate the initial reactive power support workload corresponding to the i-th inverter based on this ratio. : Furthermore, the amount of tasks allocated... With the inverter's currently available reactive power capacity Compare, when satisfied At that time, The actual reactive power support workload of the inverter; otherwise, the support workload of the inverter is limited to: The excess capacity is then redistributed according to the local support capacity indicators of other inverters that still have remaining reactive power capacity, until the total support demand is met or the maximum support capacity that the current cluster can provide is reached. Through the above processing, the support task allocation results for each inverter are obtained: ; in, This represents the i-th inverter. This indicates the amount of active power support allocated to this inverter. This indicates the amount of reactive power support allocated to this inverter.

[0079] Step 43: Based on the supporting task allocation results, perform time-series coordination and spatial collaborative optimization to generate cluster collaborative control instructions; Based on the active power support task and reactive power support task corresponding to each inverter, and combined with the inverter's positional relationship in the electrical topology, real-time capability status, and flexible network weight coefficient, the response sequence, response rate, and power reference value correction of each inverter are collaboratively optimized.

[0080] Specifically, when a voltage sag occurs, the inverter closest to the fault location and with sufficient reactive power capacity is prioritized for rapid reactive power compensation. Simultaneously, other inverters are controlled to respond in a coordinated manner according to a preset timing sequence, forming a spatially coordinated and temporally ordered support control process. For example, in response to a voltage sag, the inverter closest to the fault location and with sufficient reactive power capacity may be instructed to perform rapid reactive power compensation first, while other inverters follow at a slightly slower rate, forming a spatially coordinated and temporally ordered support matrix. This optimization process ensures that the cluster's response is both fast and stable, reducing power oscillations that might be caused by simultaneous, synchronized operation of all inverters.

[0081] In some embodiments, the generated cluster collaborative control command includes the active power reference value correction amount, reactive power reference value correction amount, response rate and action timing information corresponding to each inverter, and is sent to each distributed photovoltaic inverter, so that multiple inverters can complete grid support tasks according to the collaborative strategy.

[0082] For example, after a photovoltaic cluster enters a new stable operating state, assuming a new disturbance occurs in the power grid, resulting in a regional voltage dip, cluster coordinated control commands need to be generated. First, a multi-dimensional capability coupling assessment is performed to generate a real-time support capability map of the cluster. Taking the assessment of reactive power voltage support capability as an example, the rated DC voltage of the inverters is set to 800V, and the rated reactive power capacity is 5Mvar for each inverter. At time t, the states of each inverter are as follows: the flexible grid construction weight coefficients for PV1, PV2, and PV3 are 0.8, 0.7, and 0.4, respectively; the DC voltages are 790V, 780V, and 810V, respectively; and the available reactive power capacities are 4.5Mvar, 4.0Mvar, and 4.8Mvar, respectively. The voltage support capability index is calculated according to the formula: ; The value comprehensively reflects the current reactive power support potential of the cluster. Next, a dynamic matching network for support tasks is constructed. Real-time grid interaction status indicates that the grid-connected bus needs to inject 10Mvar of reactive power to restore the voltage to normal levels. This 10Mvar task is allocated based on the composition of each inverter's support capacity indicators, grid weight, and availability margin, forming a task-capacity collaborative mapping relationship. For example, PV1 is allocated 4.5Mvar, PV2 3.5Mvar, and PV3 2Mvar. Finally, timing and spatial coordination optimization is performed based on this mapping relationship. The instruction is that PV1, which has the closest electrical distance, will output its full 4.5Mvar within 20ms; PV2 will begin injecting reactive power at a slower rate at 30ms; and PV3 will act as a backup, deciding whether to intervene after 50ms based on the voltage recovery status. This series of specific power share and response timing instructions constitutes the final cluster collaborative control instructions. The cluster's real-time support capability map characterizes the support capability and available support margin of each inverter. Based on the task matching relationship, the grid support requirements are mapped to the real-time support capabilities of each inverter, thereby optimizing the allocation of support tasks among the inverters and improving the utilization efficiency and collaborative support capabilities of the cluster's support resources.

[0083] A further technical solution, in step 43, involves coordinated control of the power output of each distributed photovoltaic inverter, including timing coordination and spatial optimization of the support behavior of each inverter, specifically including the following steps: Step 431: Based on the time evolution path of the cluster's real-time support capability map and the real-time power grid interaction status, determine the key time windows and spatial influence domains corresponding to the support behaviors, and construct a spatiotemporal coordinated control framework. The time evolution path of the real-time grid interaction state is used to characterize the changing trends of grid connection point voltage, grid frequency, and power oscillation state over time, such as within 0-2 seconds after a fault or disturbance occurs. Based on the time evolution path, key time windows corresponding to different support tasks are determined.

[0084] This embodiment can identify the key time window for supporting behavior by predicting the real-time power grid interaction status. For example, 20-100 milliseconds after a voltage dip is the key window for reactive power support, while 200 milliseconds to 2 seconds after a frequency drop is the key window for primary frequency regulation response. When continuous power oscillation is detected, the time interval in which the oscillation amplitude exceeds the set threshold is determined as the key time window corresponding to oscillation suppression. Furthermore, based on the electrical topology of the photovoltaic cluster, the fault location, and the local support capability of each inverter, the spatial influence domain corresponding to different support tasks is determined. The spatial influence domain is used to characterize the electrical area where the inverter has an effective support effect on the current grid disturbance, that is, which areas of inverters can produce the most effective results when participating in support. Furthermore, by associating the key time window with the spatial influence domain, a spatiotemporal coordinated control framework is obtained, which is used to limit the response time range and participation range of each inverter in performing different support tasks.

[0085] Step 432: Within the spatiotemporal coordinated control framework, based on the allocation results of the support tasks corresponding to each inverter, the participation timing, output adjustment rate and action intensity of each inverter are differentiated and arranged to generate cluster coordinated control commands. In some embodiments, based on the active power support task and reactive power support task of each inverter obtained in the preceding steps, and combined with the spatial influence domain of each inverter and the key time window of the corresponding support task, the start-up time, power rise time and support output strength of each inverter are determined respectively. For voltage recovery tasks, inverters located within the fault-affected area and with available reactive power capacity exceeding a set threshold are prioritized to rapidly increase reactive power output within the corresponding critical time window. The remaining inverters then output reactive power sequentially based on their own support capabilities and electrical distance from the fault location to participate in the support. For frequency support tasks, the active power support start-up time and rise rate of each inverter are set differently according to the available active power margin and network weight of each inverter, so that the active power support output of different inverters is gradually put into operation according to the preset timing sequence.

[0086] For the task of suppressing power oscillations, the power regulation phase of different inverters is set differently according to the oscillation state and the phase response characteristics of each inverter, so that some inverters perform advanced power compensation and some inverters perform lagging power compensation, so as to form a synergistic damping effect on power oscillations.

[0087] Taking the active power support process of the j-th inverter as an example, its actual active power adjustment at time t is... Represented as: ; ; in, It is its allocated share of total active power support; It is a time function; The start time for inverter j to begin responding to the support task; This refers to the rise time required for inverter j to go from initial response to reaching full support output. This orchestration and adjustment process ensures that the cluster's support capacity is released precisely and systematically at the most needed time and place.

[0088] Within the spatiotemporal coordinated control framework of this embodiment, the output adjustment rate and phase response of each inverter are interleaved. This means that not all inverters start responding at the same speed and time. For example, for the power oscillation suppression task, some inverters may be instructed to perform advance compensation, while others may perform lag compensation, actively injecting damping through phase interleaving. Based on the task and capability coordination mapping relationship generated in the previous step, the role of each inverter in different support tasks is dynamically adjusted, specifically reflected in the differentiation of participation timing and intensity. For example, in a complex fault, an inverter with high reactive power capacity but low virtual inertia setting may be instructed to output reactive power at maximum intensity during the critical time window of the early stage of voltage recovery, while contributing less active power in the subsequent frequency regulation stage.

[0089] For example, after generating the cluster coordinated control command, it needs to be further refined into executable power output coordination actions. First, a spatiotemporal coordinated control framework is constructed. It is predicted that a voltage sag will last for the next 2 seconds, with the first 100 milliseconds being the critical time window for voltage recovery. Spatially, this sag mainly affects the feeders where PV1 and PV2 are located. Therefore, the spatiotemporal coordinated control framework is determined to focus on mobilizing PV1 and PV2 for strong support within the [0, 100ms] time window. Second, staggered orchestration and dynamic adjustment are performed within this framework. Based on the task and capacity coordination mapping relationship, PV1 is allocated a total active power support share of 4.5Mvar. A fast response timing sequence is planned for it, with a startup time of 20ms and a rise time of 40ms. For PV2, its task is 3.5Mvar, with a startup time of 35ms and a rise time of 60ms. At t=50ms, i.e., 30ms after PV1 startup and 15ms after PV2 startup, the time function is calculated... ; Therefore, their actual active power adjustment at this moment is (MW); (MW). This arrangement allows PV1 to quickly take over as the main inverter, while PV2 follows smoothly, forming a tiered response. Through a spatiotemporal coordinated control framework and staggered arrangement, the macroscopic coordinated strategy is transformed into a microscopic power response curve for each inverter accurate to the millisecond level.

[0090] Furthermore, in step 4, based on the obtained cluster collaborative control instructions, the power output of each distributed photovoltaic inverter is coordinated and controlled to execute cluster collaborative support actions, including the following steps: Step 441: Based on the cluster collaborative control command, perform dynamic trajectory planning for the power output of each inverter to generate a power output collaborative scheduling sequence; Specifically, based on the active power support task, reactive power support task, start-up time, and power rise time corresponding to each inverter in the cluster collaborative control command, the power adjustment trajectory of each inverter from the current power operating point to the target support power operating point is determined. Among them, the power adjustment trajectory is used to characterize the active power reference value and reactive power reference value corresponding to each moment during the execution of the support task, so that the inverter gradually reaches the target support power according to the aforementioned determined participation sequence and output adjustment rate, rather than jumping directly from the current power value to the target power value.

[0091] Furthermore, the power adjustment trajectory of each inverter is discretized according to the preset control cycle to obtain the power reference value of each inverter at the continuous control moment. The power reference values ​​of each inverter participating in the current support task are then combined in time order to form a power output coordinated scheduling sequence.

[0092] This enables each inverter to perform power adjustments according to the predetermined start-up time, power change rate, and target support amount, reducing the secondary disturbances to the grid connection point voltage and frequency caused by the cluster's large-scale power adjustments.

[0093] Through the aforementioned dynamic trajectory planning, the power output of each inverter gradually transitions from its current operating state to the target support state according to the preset participation sequence and power change rate, achieving continuous and smooth adjustment of the support power and avoiding secondary disturbances such as voltage, frequency, and power oscillations caused by sudden power command changes. At the same time, by forming a coordinated scheduling sequence of the power adjustment trajectories of each inverter according to a unified time base, the support output of different inverters remains coordinated in the time dimension, improving the stability and controllability of the cluster's coordinated support process.

[0094] Step 442: Based on the real-time power grid interaction status, perform elastic correction and transient response enhancement on the power output coordinated scheduling sequence to achieve adaptive power adjustment; Specifically, during the execution of the power output coordinated scheduling sequence, the actual voltage, actual frequency, and power oscillation status at the grid connection point are continuously acquired, and the real-time grid interaction status is compared with the expected grid status corresponding to the current scheduling time to determine the deviation between the actual response status and the expected response status. When the deviation exceeds the set range, the current power scheduling command is corrected according to the direction and degree of the deviation.

[0095] Taking reactive power regulation during voltage recovery as an example, when the actual voltage at the grid connection point is detected to be lower than the expected voltage recovery target value at the current moment, the reactive power support of the corresponding inverter is increased. For the j-th inverter, the reactive power command it finally executes at time t is... , can be represented as: ; ; in, This indicates the reactive power command that the j-th inverter finally executes at time t; This is the original reactive power dispatch command value; The gain coefficient is a constant greater than zero. Let be the target value for voltage recovery at time t; The actual value of the grid connection point voltage obtained in real time at time t; This is a saturation function used to limit the magnitude of the correction, preventing overshoot and overcorrection; x represents the deviation between the expected voltage recovery target value and the actual voltage at the grid connection point. The threshold is adjusted to reflect the maximum allowable voltage deviation.

[0096] When the actual voltage recovery rate is slower than expected, the corresponding voltage deviation is positive, and voltage support is enhanced by increasing the reactive power command. As the actual voltage gradually approaches the expected voltage recovery target value, the correction amount decreases accordingly. The correction amplitude is limited by a saturation function to avoid over-adjustment of the reactive power command due to excessive instantaneous voltage deviation.

[0097] For the frequency support process, the same feedback correction method can be used to dynamically correct the active power scheduling commands of each inverter based on the deviation between the actual frequency and the expected frequency recovery state.

[0098] Step 443: Based on the real-time capability status of each inverter, the modified power command is constrained, and the support tasks that exceed the execution capability of a single inverter are dynamically redistributed to obtain the modified power control command. After obtaining the corrected active power command and reactive power command for each inverter, the system combines the current available active power margin, available reactive power capacity, thermal margin, and current limit of each inverter to determine whether the corrected power command is within the current executable range of the corresponding inverter.

[0099] When the j-th inverter can fully execute the corresponding corrected power command, the power command remains unchanged; when the corrected power command exceeds the inverter's current executable capacity, its actual power command is limited to the current capacity boundary, and the remaining unexecuted support quantity is determined as the support quantity to be reallocated.

[0100] Based on the real-time support capacity map of the cluster, inverters that still have remaining support margins are identified from other inverters, and the support amount to be reallocated is redistributed according to the remaining support capacity of each inverter until the required support amount is met or the current cluster reaches its maximum available support capacity.

[0101] Through the above processing, when a single inverter is constrained by capacity, current or thermal state, the support tasks within the cluster are dynamically transferred, so that the actual total support volume of the cluster is kept as close as possible to the current grid support requirements.

[0102] Step 444: Based on the modified power control command, perform synchronous and coordinated control on each inverter to complete the cluster collaborative support action.

[0103] The active power command and reactive power command, after being adjusted for flexibility and capacity constraints, are sent to the corresponding distributed photovoltaic inverters. Based on a unified time base, each inverter is controlled to perform corresponding power regulation according to the previously determined start-up time, power change rate, and intensity. Specifically, a Precision Time Protocol (PTP) can be used to synchronize the time base of each inverter participating in the collaborative support, enabling each inverter to perform power adjustment according to its corresponding collaborative control sequence under a unified time base.

[0104] The steps described above in this embodiment, through adaptive power coordination and dynamic balancing, enable the cluster to make online corrections to the predetermined support plan based on the real-time grid status. When the operating status, such as voltage and frequency, deviates from the expected response trajectory, the support strength is adjusted in a timely manner, improving the cluster's tracking ability and response speed to transient disturbances. Simultaneously, when a single inverter is limited by power, current, or thermal margin, its unfinished support tasks are transferred to other inverters that still have remaining support capacity, reducing the impact of local capacity constraints on the overall support effect of the cluster, improving the utilization rate of support resources and the continuity of cluster support. Furthermore, by coordinating the support actions of each inverter through a unified time base, each inverter outputs in a coordinated manner according to the predetermined timing and adjustment requirements, reducing power fluctuations and control conflicts caused by asynchronous responses, thereby improving the stability, coordination, and robustness of the photovoltaic cluster's collaborative support process.

[0105] For example, in the final execution phase, dynamic trajectory planning is performed on PV1 and PV2 based on the instructions. If the reactive power support task of PV1 increases from 0 Mvar to 4.5 Mvar, a power output trajectory is planned that smoothly rises with an S-shaped curve within 40 ms, forming a power output coordinated scheduling sequence with a time resolution of 10 ms. At t=40 ms, the original scheduling sequence instruction value is 1.8 Mvar. Next, this sequence is elastically corrected. Continuous monitoring of the grid connection point voltage reveals that due to unmodeled dynamics in the grid impedance, the actual voltage at t=40 ms is 0.88 pu, while the expected voltage recovery trajectory under coordinated support should be 0.90 pu. Under this deviation, the output is that the final reactive power instruction executed by PV1 is corrected to 2.3 Mvar. Finally, dynamic power balancing and behavior synchronization are performed. During the elastic correction, it is found that the output current of PV1 is close to its 120A limit, and it cannot fully output the current corresponding to 2.3 Mvar. The 0.2Mvar task that failed to complete was dynamically reassigned to PV2, which still had spare capacity at the time. Simultaneously, PTP high-precision clock synchronization ensured that PV1's 2.1Mvar output and PV2's additional 0.2Mvar output occurred precisely at the same moment. Through this series of operations, the cluster ultimately completed the collaborative support action accurately. The elastic correction mechanism can correct execution deviations caused by model inaccuracies or environmental changes in real time.

[0106] Example 2 Based on Example 1, this example provides a photovoltaic model-free parameter disturbance rejection control system for source-grid interaction, such as... Figure 5 As shown, it includes: The observation and compensation control module is configured to acquire real-time operating electrical data of each distributed photovoltaic inverter, perform disturbance observation, generate current disturbance estimate and disturbance trend prediction, construct a composite feedforward compensation strategy based on the current disturbance estimate and disturbance trend prediction, and output disturbance suppression control commands. The decision mapping module is configured to obtain distributed photovoltaic inverter response data, grid strength depth index and cluster disturbance feature set after each inverter executes disturbance suppression control command, perform dynamic mapping, and generate a set of flexible grid construction weight coefficients. The parameter adjustment module is configured as a set of flexible grid weight coefficients to dynamically adjust the virtual inertia parameters, virtual damping parameters and equivalent output impedance parameters of each distributed photovoltaic inverter to obtain an adaptive grid control state. The task allocation module is configured to dynamically allocate and coordinate the support tasks based on the adaptive network control state, combined with the flexible network weight coefficient set and the real-time capability status of each inverter. It coordinates the participation timing and support output of each inverter to generate cluster coordinated control commands and executes cluster coordinated support actions.

[0107] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0108] Example 3 Based on Embodiment 1, this embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the photovoltaic model-free parameter disturbance rejection control method for source-grid interaction described in Embodiment 1.

[0109] Example 4 Based on Embodiment 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they complete the steps in the photovoltaic model-free parameter disturbance rejection control method for source-grid interaction described in Embodiment 1.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0111] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A model-free parameter disturbance rejection control method for photovoltaic systems with source-grid interaction, characterized in that, Includes the following steps: The system acquires real-time electrical data of each distributed photovoltaic inverter, performs disturbance observation, generates current disturbance estimates and disturbance trend predictions, constructs a composite feedforward compensation strategy based on the current disturbance estimates and disturbance trend predictions, and outputs disturbance suppression control commands. After each inverter executes the disturbance suppression control command, the distributed photovoltaic inverter response data, grid strength depth index and cluster disturbance feature set are obtained, and dynamic mapping is performed to generate a flexible grid weight coefficient set. Based on the set of flexible grid weight coefficients, the virtual inertia parameters, virtual damping parameters, and equivalent output impedance parameters of each distributed photovoltaic inverter are dynamically adjusted to obtain the adaptive grid control state. Based on the adaptive network control state, combined with the flexible network weight coefficient set and the real-time capability status of each inverter, the dynamic allocation and coordinated control of support tasks are carried out. The participation timing and support output of each inverter are coordinated and adjusted to generate cluster coordinated control commands, and the cluster coordinated support actions are executed.

2. The photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in claim 1, characterized in that, The process involves acquiring real-time electrical data from each distributed photovoltaic inverter, observing disturbances, and generating current disturbance estimates and trend predictions, including the following steps: Based on real-time operating electrical data, transient decomposition of electrical behavior is performed to extract multi-frequency disturbance components, and the dynamic correlation structure between the multi-frequency disturbance components is fused to generate the current disturbance estimate. Based on the current disturbance estimate, sequence evolution inference and coupling strength evolution analysis are performed to construct the disturbance propagation trend topology and generate disturbance trend prediction.

3. The photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in claim 2, characterized in that, Based on the current perturbation estimate, sequence evolution inference and coupling strength evolution analysis are performed to construct the perturbation propagation trend topology and generate perturbation trend predictions, including: Based on the current disturbance estimate, sequence evolution is extrapolated to predict the future disturbance change state of a single inverter; Based on the disturbance coupling relationship between multiple inverters, the coupling strength evolution analysis is performed to construct the disturbance propagation trend topology; Based on the disturbance evolution trend of a single inverter and the topology of disturbance propagation trend, the variation law of the disturbance itself and the propagation effect between clusters are analyzed to generate the disturbance trend prediction.

4. The photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in claim 1, characterized in that, The process of constructing a composite feedforward compensation strategy based on the current disturbance estimate and the disturbance trend prediction, and outputting disturbance suppression control commands, includes: Based on the current disturbance estimate and the disturbance trend prediction, an instantaneous disturbance compensation channel and a trend disturbance compensation channel are constructed respectively. The outputs of the two compensation channels are weighted and fused to form a multi-level disturbance suppression compensation structure, and the original compensation signal is calculated. The compensation intensity adjustment factor is determined based on the real-time dynamic response state of the inverter. The original compensation signal is then adaptively amplitude corrected, and the corrected compensation signal is converted into a disturbance suppression control command that can be executed in the inner loop of the inverter.

5. The photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in claim 1, characterized in that, After each inverter executes the disturbance suppression control command, the distributed photovoltaic inverter response data, grid strength depth index, and cluster disturbance characteristic set are acquired, dynamically mapped, and a flexible grid construction weight coefficient set is generated, including the following steps: The inverter response data after the disturbance suppression control command is executed is obtained, and dynamic response performance indicators characterizing the control effect are extracted within a preset response time window. The response dynamic profile of each inverter is constructed based on the response performance indicators. The grid intensity depth index and cluster disturbance feature set of each grid connection point are obtained. A spatiotemporal graph structure is constructed with each distributed photovoltaic inverter as a node and the electrical coupling relationship between inverters as an edge. The node features and edge features are aggregated based on the spatiotemporal graph neural network to generate grid-disturbance interaction feature vector. Based on the response dynamic profile of each inverter and the grid-disturbance interaction feature vector, a grid construction demand adaptation mapping is performed, the grid construction tendency corresponding to each inverter is calculated, and a set of flexible grid construction weight coefficients is generated.

6. The photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in claim 1, characterized in that, Based on the flexible grid weight coefficient set, the virtual inertia parameters, virtual damping parameters, and equivalent output impedance parameters of each distributed photovoltaic inverter are dynamically adjusted to obtain the adaptive grid control state, including the following steps: Based on the set of flexible grid weight coefficients, the virtual inertia parameters and virtual damping parameters of each distributed photovoltaic inverter are dynamically adjusted, and spatial mapping is performed in combination with the positional relationship or electrical connection relationship of each inverter to generate a dynamic grid capacity distribution map. Based on the real-time grid interaction status and dynamic grid capacity distribution map, the active power-frequency droop parameters and reactive power-voltage droop parameters of each distributed photovoltaic inverter are dynamically adjusted to generate voltage and frequency support control parameters that are adapted to the current grid status. Based on the adjusted voltage and frequency support control parameters, the equivalent output impedance of each distributed photovoltaic inverter is dynamically adjusted, and the virtual inertia parameter, virtual damping parameter, droop control parameter and equivalent output impedance parameter are integrated in a coordinated manner to obtain the adaptive grid control state.

7. The photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in claim 1, characterized in that, Based on the adaptive network control state, and combining the flexible network weight coefficient set with the real-time capability status of each inverter, dynamic allocation and coordinated control of support tasks are performed to generate cluster coordinated control commands, including the following steps: Based on the adaptive network control state, the flexible network weight coefficient set, and the real-time capability state of each inverter, the voltage support capability, frequency support capability, and oscillation suppression capability of each inverter are evaluated in multiple dimensions. The electrical topology relationship between the inverters is then used to generate a real-time support capability map of the cluster. The current active power support demand and reactive power support demand are determined based on the real-time grid interaction status. The task allocation ratio of each inverter is determined by combining the real-time support capability map of the cluster. Under the condition of meeting the real-time support capability constraints of each inverter, the support tasks are dynamically allocated to obtain the active power support quantity and reactive power support quantity corresponding to each inverter, thus forming the support task allocation result. Based on the task allocation results of each inverter, combined with its electrical topology location, real-time capability status and flexible network weight coefficient, the participation order, response rate and power reference value correction of each inverter are optimized in a spatiotemporal manner to generate cluster collaborative control commands.

8. A model-free parameter disturbance rejection control system for photovoltaic systems oriented towards source-grid interaction, characterized in that, include: The observation and compensation control module is configured to acquire real-time operating electrical data of each distributed photovoltaic inverter, perform disturbance observation, generate current disturbance estimate and disturbance trend prediction, construct a composite feedforward compensation strategy based on the current disturbance estimate and disturbance trend prediction, and output disturbance suppression control commands. The decision mapping module is configured to obtain distributed photovoltaic inverter response data, grid strength depth index and cluster disturbance feature set after each inverter executes disturbance suppression control command, perform dynamic mapping, and generate a set of flexible grid construction weight coefficients. The parameter adjustment module is configured as a set of flexible grid weight coefficients to dynamically adjust the virtual inertia parameters, virtual damping parameters and equivalent output impedance parameters of each distributed photovoltaic inverter to obtain an adaptive grid control state. The task allocation module is configured to dynamically allocate and coordinate the support tasks based on the adaptive network control state, combined with the flexible network weight coefficient set and the real-time capability status of each inverter. It coordinates the participation timing and support output of each inverter to generate cluster coordinated control commands and executes cluster coordinated support actions.

9. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, complete the steps in the photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the steps in the photovoltaic model-free parameter disturbance rejection control method for source-grid interaction as described in any one of claims 1-7.

Images