Photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game

Abstract

This invention relates to the field of power and its automation technology, and discloses a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. The method includes: identifying grid impedance from real-time operating data of distributed energy agents to obtain the current output distribution state; determining the preliminary parameter adjustment range for voltage fluctuation suppression based on the current output distribution state and a preset physical constraint embedding algorithm; adaptively and dynamically adjusting virtual inertia and damping coefficient to generate parameter configurations; determining a preliminary allocation strategy based on the parameter configurations; constructing an output adjustment matrix based on the preliminary allocation strategy; when a sudden disturbance is detected, each distributed energy agent autonomously identifies the disturbance type and determines the disturbance response priority based on the output adjustment matrix; and autonomously updating inverter control parameters locally based on the disturbance response priority until the photovoltaic cluster stability index reaches the target threshold. This method solves problems such as reliance on centralized communication, lack of physical constraints, and insufficient coordination capabilities.

Description

Technical Field

[0001] This invention relates to the field of power and its automation technology, and in particular to a method for intelligent control of photovoltaic clusters based on physical constraint embedding and dynamic game theory. Background Technology

[0002] With the high proportion of distributed photovoltaic (PV) power grids being integrated into distribution networks, problems such as voltage exceeding limits and harmonic exceedances are becoming increasingly prominent. Existing collaborative control methods mainly fall into two categories. Centralized control relies on a central controller, which suffers from communication delays and single-point-of-failure risks, making it difficult to adapt to the plug-and-play requirements of distributed PV. Distributed control often employs droop control with fixed parameters or virtual synchronous machine control; however, these methods do not fully consider the physical constraints of the power grid (such as impedance characteristics) and the dynamic game relationship between nodes, resulting in limited voltage regulation and harmonic suppression effects, and a lack of a rapid autonomous response mechanism based on local information. Therefore, existing technologies require a PV cluster control method that can embed power grid physical constraints and achieve autonomous collaboration of local controllers to address the technical problems of voltage exceeding limits, harmonic exceedances, difficulties in renewable energy integration, strong communication dependence, and insufficient transient response capabilities. Summary of the Invention

[0003] The main objective of this invention is to provide a smart control method for photovoltaic clusters based on physical constraint embedding and dynamic game theory, aiming to solve technical problems such as voltage exceeding limits, harmonic exceedance, difficulty in absorbing new energy sources, strong communication dependence, and insufficient transient response capability.

[0004] To achieve the aforementioned objectives, the first aspect of this invention proposes a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory, wherein the photovoltaic cluster comprises several distributed energy intelligent agents, and the method includes: The real-time operating data of each distributed energy intelligent agent is acquired, the grid impedance is identified in the real-time operating data, the output distribution characteristics of each distributed energy intelligent agent are analyzed, and the current output distribution status is obtained. Based on the current power distribution state and the preset physical constraint embedding algorithm, the preliminary parameter adjustment range required for voltage fluctuation suppression is intelligently analyzed; When the initial parameter adjustment range exceeds the preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate the parameter configuration of each of the distributed energy intelligent agents. Based on the parameter configuration, the initial allocation strategy for distributed reactive power support is determined through real-time interaction and dynamic game among the distributed energy intelligent agents. Based on the preliminary allocation strategy of the distributed reactive power support, the harmonic suppression parameters among the distributed energy agents are optimized, and the power output adjustment matrix is ​​constructed. When a sudden interference is detected, each of the distributed energy intelligent agents is controlled to autonomously identify the type of interference and determine the priority of interference response based on local measurement information according to the output adjustment matrix. Based on the interference response priority, the inverter control parameters are updated locally until the photovoltaic cluster stability index reaches the target threshold.

[0005] Furthermore, the step of acquiring real-time operating data of each distributed energy intelligent agent, performing grid impedance identification on the real-time operating data, analyzing the output distribution characteristics of each distributed energy intelligent agent, and obtaining the current output distribution state includes: Acquire photovoltaic power generation data and grid connection status information collected by each of the distributed energy intelligent agents; Based on the photovoltaic power generation data, the grid access status information, and the preset recursive least squares algorithm, the grid impedance is identified, and the output distribution characteristics of each of the distributed energy intelligent agents in the photovoltaic cluster are analyzed to obtain the current output distribution status.

[0006] Furthermore, the step of intelligently analyzing the preliminary parameter adjustment range required for voltage fluctuation suppression based on the current power distribution state and a preset physical constraint embedding algorithm includes: According to the preset physical constraint embedding algorithm, the physical characteristics of the power grid impedance are incorporated as constraints into the training process of the neural network. The current power output distribution state is input into the neural network to predict the voltage change trend based on potential instability factors in historical voltage fluctuation records, and to determine the preliminary parameter adjustment range required for voltage fluctuation suppression.

[0007] Furthermore, when the initial parameter adjustment range exceeds a preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate parameter configurations for each of the distributed energy intelligent agents, including: When the initial parameter adjustment range exceeds the preset threshold range, the degree to which the initial parameter adjustment range exceeds the preset threshold range is calculated; Obtain the electrical location information of each of the distributed energy intelligent agents; Based on the degree of excess and the electrical location information of each of the distributed energy intelligent agents, the virtual inertia adjustment amount and damping coefficient adjustment amount of each of the distributed energy intelligent agents are determined by real-time dynamic calculation. The virtual inertia setting value of each of the distributed energy intelligent agents is updated according to the virtual inertia adjustment amount, and the damping coefficient setting value of each of the distributed energy intelligent agents is updated according to the damping coefficient adjustment amount, thereby generating the parameter configuration of each of the distributed energy intelligent agents.

[0008] Furthermore, the step of determining the initial allocation strategy for distributed reactive power support based on the parameter configuration and through collaborative interaction and dynamic game among the distributed energy agents includes: Based on the parameter configuration, the local voltage gradient information of adjacent nodes is exchanged in real time through a preset distributed consensus protocol; Each of the distributed energy intelligent agents is regarded as an intelligent agent, and a preliminary allocation strategy for distributed reactive power support is determined based on the dynamic game among the photovoltaic cluster intelligent agents.

[0009] Furthermore, the step of exchanging local voltage gradient information of neighboring nodes in real time through a preset distributed consensus protocol based on the parameter configuration includes: Based on the parameter configuration, calculate the local voltage gradient information of each of the distributed energy intelligent agents; Through a preset distributed consensus protocol, each of the distributed energy agents sends local voltage gradient information to neighboring distributed energy agents and receives local voltage gradient information sent by neighboring distributed energy agents.

[0010] Furthermore, the step of treating each of the distributed energy intelligent agents as intelligent agents and determining the initial allocation strategy for distributed reactive power support based on the dynamic game among the photovoltaic cluster intelligent agents includes: Construct game payoff functions for each of the aforementioned distributed energy intelligent agents; Based on the game payoff function, the reactive power output setting value of each of the distributed energy intelligent agents is iteratively updated. When the convergence condition of the game payoff function is met, the reactive power output setting value of each of the distributed energy intelligent agents is used as the initial allocation strategy for the distributed reactive power support.

[0011] Furthermore, the step of optimizing the harmonic suppression parameters among the distributed energy agents and constructing an output adjustment matrix based on the initial allocation strategy of the distributed reactive power support includes: Based on the preliminary allocation strategy of the distributed reactive power support and the preset Nash equilibrium solution algorithm, the harmonic suppression parameters among the distributed energy agents are optimized. Based on the optimized harmonic suppression parameters, the harmonic impedance of each distributed energy intelligent agent is reshaped to construct the output adjustment matrix after harmonic impedance reshaping.

[0012] Furthermore, upon detecting a sudden interference, the step of controlling each of the distributed energy intelligent agents to autonomously identify the interference type and determine the interference response priority based on local measurement information according to the output adjustment matrix includes: When a sudden interference is detected, based on the plug-and-play support mechanism under the preset global communication-free architecture, each of the distributed energy intelligent agents autonomously determines the type of interference according to the locally measured voltage change rate, and determines the corresponding interference response priority according to the type of interference.

[0013] Furthermore, the step of automatically updating the inverter control parameters locally based on the interference response priority until the photovoltaic cluster stability index reaches the target threshold includes: According to the interference response priority, each of the distributed energy intelligent agents locally updates the inverter control parameters to perform output adjustment. Monitor the execution feedback data of each of the distributed energy intelligent agents to determine whether the stability indicators of the photovoltaic cluster have reached the target threshold. When the stability index of the photovoltaic cluster reaches the target threshold, the target adjustment result is obtained, and the response control for the current sudden disturbance ends.

[0014] A second aspect of this invention proposes a photovoltaic cluster intelligent control device based on physical constraint embedding and dynamic game theory, comprising: The first processing unit is used to acquire real-time operating data of each distributed energy intelligent agent, perform grid impedance identification on the real-time operating data, analyze the output distribution characteristics of each distributed energy intelligent agent, and obtain the current output distribution status. The second processing unit is used to intelligently analyze the preliminary parameter adjustment range required for voltage fluctuation suppression based on the current power distribution state and the preset physical constraint embedding algorithm. The third processing unit is used to adaptively and dynamically adjust the virtual inertia and damping coefficient when the preliminary parameter adjustment range exceeds the preset threshold range, and generate the parameter configuration of each of the distributed energy intelligent agents. The fourth processing unit is used to determine the initial allocation strategy for distributed reactive power support based on the parameter configuration and through real-time interaction and dynamic game between the distributed energy intelligent agents. The fifth processing unit is used to optimize the harmonic suppression parameters among the distributed energy agents and construct the output adjustment matrix based on the initial allocation strategy of the distributed reactive power support. The sixth processing unit is used to control each of the distributed energy intelligent agents to autonomously identify the type of interference and determine the priority of interference response based on local measurement information when a sudden interference is detected, according to the output adjustment matrix. The seventh processing unit is used to automatically update the inverter control parameters locally according to the interference response priority until the photovoltaic cluster stability index reaches the target threshold.

[0015] A third aspect of the present invention provides a local controller, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game as described in the first aspect.

[0016] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game as described in the first aspect.

[0017] In this embodiment, the reliance on a central controller is completely eliminated. All operations are based on distributed computing using local controllers and limited communication between adjacent nodes. This eliminates the need for global synchronization or centralized optimization, significantly reducing communication bandwidth requirements and the risk of single points of failure. This allows the photovoltaic cluster to operate stably even in environments with weak communication or momentary network outages.

[0018] The physical characteristics of the power grid are deeply integrated into control decisions. By embedding physical constraints and customizing virtual inertia and damping coefficients according to electrical location, it is ensured that all control commands do not violate physical boundaries such as voltage limits and equipment capacity. This avoids infeasible outputs that may result from purely data-driven or experience-based control, and improves the safety and reliability of the control strategy.

[0019] Joint optimization of voltage control and power quality was achieved. By coordinating the reactive power output of each node through distributed game theory, voltage fluctuations were effectively suppressed. By optimizing harmonic suppression parameters and constructing an output adjustment matrix, the voltage was improved without causing harmonic amplification or resonance, thus solving the problem of conflict between voltage regulation and harmonic suppression in traditional methods.

[0020] It possesses plug-and-play rapid emergency response capabilities. Each node can autonomously identify the type of interference and determine its priority based solely on its local voltage change rate, without waiting for remote commands; it updates inverter parameters locally according to priority and judges cluster stability indicators by exchanging simple location information with neighbors, achieving millisecond-level fully distributed recovery control and overcoming the response latency of centralized architectures.

[0021] In summary, the embodiments of the present invention achieve fully distributed autonomous collaboration from steady-state optimization to transient response without the need for a central controller and global communication, thus solving problems such as reliance on centralized communication, lack of physical constraints, and insufficient collaboration capabilities. Attached Figure Description

[0022] Figure 1 This invention provides a schematic flowchart of a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 2This invention provides a schematic flowchart of steps S11-S12 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 3 This invention provides a schematic flowchart of steps S21-S22 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 4 This invention provides a schematic flowchart of steps S31 to S34 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 5 This invention provides a schematic flowchart of steps S41-S42 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 6 This invention provides a schematic flowchart of steps S411~S412 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 7 This invention provides a schematic flowchart of steps S421 to S423 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 8 This invention provides a schematic flowchart of steps S51-S52 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 9 This invention provides a schematic flowchart of steps S71 to S73 in a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. Figure 10 This invention provides a schematic diagram of a photovoltaic cluster intelligent control device based on physical constraint embedding and dynamic game theory. Figure 11 A schematic diagram of a local controller provided in an embodiment of the present invention.

[0023] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any modules and all combinations of one or more associated listed items.

[0026] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0027] Reference Figure 1 This invention provides a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory. The method provided by this invention uses local controllers equipped in each distributed energy intelligent agent as the execution entity. Each local controller independently collects data from its own node, performs calculations and decisions, communicates with neighboring nodes, and executes the final control command. The method includes the following steps S1-S7: S1: Obtain real-time operating data of each distributed energy intelligent agent, identify grid impedance on the real-time operating data, analyze the output distribution characteristics of each distributed energy intelligent agent, and obtain the current output distribution status.

[0028] In this embodiment, the local controllers configured in each distributed energy intelligent entity (such as photovoltaic inverters, energy storage converters, etc.) serve as the execution entities. Each local controller collects real-time operating data of its own node through its own sensors, and obtains operating data of neighboring nodes through limited communication between adjacent nodes. The acquired real-time operating data may include voltage, current, active power, reactive power, etc.

[0029] The real-time operating data is subjected to grid impedance identification, and the output distribution characteristics of each distributed energy intelligent entity are analyzed to obtain the current output distribution status. The results of impedance identification directly affect the accuracy of subsequent voltage sensitivity analysis, output distribution characteristic judgment, and virtual inertia adjustment.

[0030] The local controller can flexibly select the impedance identification implementation method based on computing resources, communication conditions, and on-site power quality. For example, a recursive least squares algorithm can be used for online recursive updates, utilizing a forgetting factor to track the time-varying characteristics of impedance; a Kalman filter algorithm can also be used to obtain a more robust estimate in environments with strong noise or harmonics; for nodes with limited computing power, the fundamental component can be extracted by fast Fourier transform and then periodically solved using the least squares method; for critical nodes with extremely high accuracy requirements, a small-amplitude characteristic frequency disturbance signal can be injected for short-term active identification; in scenarios with sufficient historical data, lightweight neural networks or random forest models can also be deployed for impedance prediction.

[0031] After obtaining the grid impedance of each node, the local controller also needs to analyze the power output distribution characteristics of each node in the photovoltaic cluster. Specifically, each local controller collects the impedance identification results and current output power of neighboring nodes through neighbor communication. Combining this with its own data, it determines the relative position of each node according to the preset electrical topology, such as the beginning, middle, or end, and calculates the proportion of each node's active power to the total active power of the cluster, as well as the sensitivity of each node's power change to the voltage of neighboring nodes. This information is then summarized to form the current power output distribution status and stored in the local controller's memory in a structured data format. Each node's record includes its own and other related node identifiers, electrical location tags, current active / reactive power, local equivalent impedance, power ratio, and voltage sensitivity factor.

[0032] In one embodiment, S1 may include S11~S12, such as Figure 2 As shown, S11~S12 are as follows: S11: Obtain photovoltaic power generation data and grid connection status information collected by each of the distributed energy intelligent agents.

[0033] In this embodiment, the photovoltaic power generation data may include the DC voltage, DC current, active power, reactive power output of the photovoltaic array, and the percentage of the current power generation relative to the rated power. Grid connection status information may include the instantaneous values ​​of the three-phase voltage and current at the grid connection point, frequency, phase angle, and voltage harmonic distortion rate.

[0034] S12: Based on the photovoltaic power generation data, the grid access status information, and the preset recursive least squares algorithm, the grid impedance is identified, and the output distribution characteristics of each of the distributed energy intelligent agents in the photovoltaic cluster are analyzed to obtain the current output distribution status.

[0035] After obtaining the photovoltaic power generation data and grid connection status information of the local node and adjacent nodes, the local controller uses a recursive least squares algorithm to identify the grid impedance online.

[0036] The local controller uses the voltage and current at the grid connection point as inputs to establish a linear regression model—voltage drop equals current multiplied by impedance. The voltage sample value at each moment is used as the observation value, and the current sample value as the regression variable. The estimated values ​​of the impedance parameters (including equivalent resistance and equivalent reactance) are continuously updated using the RLS recursive formula.

[0037] After obtaining the equivalent impedance estimate for each node, the local controller combines the active and reactive power data of each node to analyze the output distribution characteristics of the entire photovoltaic cluster. The output distribution characteristics describe the relative position of each distributed energy agent in the cluster, its power contribution, and its impact on voltage fluctuations.

[0038] Specifically, the electrical distance between a node and the main network access point can be determined based on the impedance identification results. The greater the impedance, the farther the distance. Combined with the node's power output value, the proportion of the node in the total output of the cluster can be calculated, as well as the sensitivity of the node's power change to the voltage of adjacent nodes, i.e., the derivative of voltage with respect to power.

[0039] In this embodiment, efficient acquisition of real-time operating data of the photovoltaic cluster and online impedance identification are achieved. Each local controller can obtain the necessary photovoltaic power and grid status information through lightweight communication between nodes. A recursive least squares algorithm is used for grid impedance identification, which can recursively update equivalent resistance and reactance online, quickly track photovoltaic power output fluctuations and grid structure changes, and has low computational complexity and fast convergence speed, perfectly suited to the limited computing power of the local controller. Based on the identification results, the output distribution characteristics can be analyzed to accurately determine the relative position, power contribution, and voltage sensitivity of each node on the feeder, providing accurate basic status information for subsequent physical constraint embedding and personalized parameter configuration.

[0040] S2: Based on the current output distribution state and the preset physical constraint embedding algorithm, intelligently analyze the preliminary parameter adjustment range required for voltage fluctuation suppression.

[0041] A pre-set physical constraint embedding algorithm is used. The core of this algorithm is to take the current power distribution state as a constraint and incorporate it into the decision-making process of the control parameter adjustment range. Based on the pre-set physical constraint embedding algorithm, the algorithm can intelligently analyze the preliminary parameter adjustment range required for voltage fluctuation suppression.

[0042] The pre-defined physical constraint embedding algorithm can be a trained lightweight neural network or a linearized predictive control model. It linearizes the power grid equations near the current operating point and then directly calculates the parameter adjustment range required to meet voltage safety limits by solving a small-scale quadratic programming problem with physical constraints. Alternatively, it can generate a parameter adjustment interval table in advance through offline simulation for different power distribution states (such as the power ratio at the feeder end, voltage deviation, etc.), allowing for direct lookup at the table during runtime.

[0043] In one implementation, S2 may include S21~S22, such as Figure 3 As shown, S21~S22 are as follows: S21: Based on the preset physical constraint embedding algorithm, the physical characteristics of the power grid impedance are incorporated as constraints into the training process of the neural network.

[0044] In this embodiment, the preset physical constraint embedding algorithm is a trained lightweight neural network. During training, the inputs include historical power output distribution and load disturbances, while the outputs are voltage fluctuation amplitude and adjustment parameters. This network not only learns the data mapping relationship between inputs and outputs, but more importantly, it embeds the physical characteristics of the power grid into the network's loss function or network structure in the form of constraints.

[0045] Specifically, physical constraints can include: Ohm's law (the relationship between voltage drop and current / impedance), Kirchhoff's voltage law (the algebraic sum of loop voltages is zero), upper and lower limits for allowable voltage at each node (e.g., 0.93~1.07 pu), and saturation limits for the inverter's reactive power output. For example, a penalty term can be added to the loss function so that the loss value increases significantly when the voltage adjustment scheme predicted by the network causes the calculated voltage of a node to exceed the physical limit; or, when designing the network structure, a physical constraint projection layer can be added after the output layer to map the network's original output to the feasible region.

[0046] S22: Input the current power output distribution state into the neural network to predict the voltage change trend based on potential instability factors in historical voltage fluctuation records, and determine the preliminary parameter adjustment range required for voltage fluctuation suppression.

[0047] The current power distribution state obtained in step S1 is used as a feature vector and input into a pre-trained neural network embedded with physical constraints. This network uses potential instability factors learned from historical voltage fluctuation records during the training phase to predict the current voltage change trend.

[0048] The network output consists of two parts: one is the possible change curve of the voltage of each node in the future period; the other is the range of parameters that the local controller should initially adjust in order to suppress voltage fluctuations, such as the range in which the virtual inertia needs to be increased and the range in which the damping coefficient needs to be decreased.

[0049] In this embodiment, physical characteristics such as grid impedance are embedded as constraints into the training process of the neural network. This allows the network to naturally adhere to physical boundaries such as Ohm's law, node voltage limits, and inverter capacity when predicting voltage change trends. This avoids unreasonable control commands that might arise from purely data-driven models, significantly improving the scientific validity and safety of the output results. Simultaneously, by using a physically constrained optimized neural network to infer the current power distribution, potential instability factors can be accurately identified from historical voltage fluctuation records, thereby quickly predicting voltage change trends and providing preliminary parameter adjustment ranges.

[0050] S3: When the initial parameter adjustment range exceeds the preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate the parameter configuration of each of the distributed energy intelligent agents.

[0051] The local controller obtains an initial parameter adjustment range through a physical constraint embedding algorithm. This initial range may exceed the safety limits specified by the equipment manufacturer or the thresholds allowed by the grid operation procedures. To ensure equipment safety and take into account global stability, the local controller needs to correct the portion exceeding the threshold and distribute the global adjustment requirements differentially to each distributed energy agent, generating personalized parameter configurations.

[0052] When the initial parameter adjustment range exceeds the preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate the parameter configuration for each of the distributed energy intelligent agents. There are multiple methods for this dynamic adjustment, which can be flexibly selected based on communication conditions, computing resources, and control precision requirements.

[0053] For example, a distributed consensus algorithm can be used to allow nodes to negotiate the allocation of excess capacity, ensuring that the final adjustment amount for each node is proportional to its remaining adjustable capacity, thus preventing some nodes from over-adjusting while others remain idle. Alternatively, an auction or bidding mechanism can be used, where each node declares its required adjustment amount based on its sensitivity to voltage fluctuations and its remaining capacity, and then a distributed optimization solution is obtained. For nodes with abundant computing resources, a quadratic programming problem with the objective of minimizing voltage deviation can be solved in real time to directly determine the optimal parameter configuration for each node.

[0054] In one embodiment, S3 may include S31 to S34, such as Figure 4 As shown, S31~S34 are as follows: S31: When the preliminary parameter adjustment range exceeds the preset threshold range, calculate the degree to which the preliminary parameter adjustment range exceeds the preset threshold range.

[0055] The local controller pre-stores the parameter safety threshold ranges for the inverter at this node. For example, the allowable adjustment range for virtual inertia is 0.5~10 seconds, and the allowable adjustment range for damping coefficient is 0~5. When the initial parameter adjustment range given in step S2 exceeds the device's upper limit, the local controller determines that the preset threshold has been exceeded. Subsequently, the system calculates the degree of exceedance. The degree of exceedance can be expressed as an absolute value or a relative percentage. For example, if the required target value of 10.5 seconds exceeds the upper limit of 10 seconds by 0.5 seconds, the degree of exceedance is 5%. For the damping coefficient, if the initial adjustment requirement is reduced to -0.1, the degree of exceedance is the absolute value of the exceedance relative to the lower limit of 0, which is 0.1. This degree of exceedance will be used for subsequent adjustment allocation. If the initial adjustment range does not exceed any threshold, the local controller can directly use this range as the final parameter configuration without performing subsequent exceedance processing steps.

[0056] S32: Obtain the electrical location information of each of the distributed energy intelligent agents.

[0057] To rationally distribute the global excess capacity across different nodes, the local controller needs to acquire the electrical location information of each distributed energy agent. This electrical location information primarily includes the node's voltage level and its topological location within the photovoltaic cluster, such as the beginning, middle, or end of a feeder. The topological location can be indirectly determined through the impedance identification results in step S1: nodes with higher impedance values ​​are typically located at the end of the feeder, while nodes with lower impedance values ​​are closer to the main grid connection point.

[0058] Each local controller can construct an electrical topology map of the entire cluster by exchanging impedance identification results and power data with neighboring nodes, thereby determining the relative positional relationship of each node.

[0059] S33: Based on the degree of excess and the electrical position information of each of the distributed energy intelligent agents, the virtual inertia adjustment amount and damping coefficient adjustment amount of each of the distributed energy intelligent agents are determined by real-time dynamic calculation.

[0060] Based on the degree of excess and the electrical location information of each distributed energy agent, a weighted allocation algorithm is used to dynamically calculate the adjustment amount that each node should bear.

[0061] Nodes located at the end of a feeder or at a lower voltage level are more sensitive to voltage fluctuations, therefore their virtual inertia should increase more significantly, and their damping coefficient should decrease more significantly as well. Nodes located at the beginning of a feeder or at a higher voltage level can accommodate smaller adjustments. For example, an electrical sensitivity factor can be set, with a factor of 1.0 for end nodes, 0.6 for intermediate nodes, and 0.3 for beginning nodes. The voltage level will also adjust this factor.

[0062] Then the virtual inertia adjustment of a certain node = global overshoot level × the sensitivity factor of that node.

[0063] The damping coefficient adjustment can be achieved using a similar formula, but since the damping coefficient needs to be reduced, the higher the sensitivity factor, the greater the reduction. This calculation is performed dynamically in real time, meaning it is recalculated each time an out-of-range situation is detected to adapt to changes in the grid topology or output distribution.

[0064] S34: Update the set value of the virtual inertia of each of the distributed energy intelligent agents according to the virtual inertia adjustment amount, update the set value of the damping coefficient of each of the distributed energy intelligent agents according to the damping coefficient adjustment amount, and generate the parameter configuration of each of the distributed energy intelligent agents.

[0065] Based on the virtual inertia adjustment and damping coefficient adjustment calculated in step S33, update the virtual inertia setting and damping coefficient setting of this node respectively.

[0066] The update method can be: New virtual inertia setting = Current baseline value + Virtual inertia adjustment amount New damping coefficient setting value = Current reference value - Damping coefficient adjustment amount If the initial adjustment range does not exceed the threshold, the new setpoint can be directly taken as the median of the initial adjustment range output in step S2 or determined according to other local control strategies. After the update is completed, each local controller generates a personalized parameter configuration for its node, including a new virtual inertia value, a new damping coefficient value, and other potentially relevant control parameters.

[0067] In this embodiment, when the initial parameter adjustment range exceeds the safety threshold, the degree of exceedance can be accurately calculated. Combined with the electrical location information of each distributed energy intelligent agent, the adjustment amount required to suppress voltage fluctuations is differentially allocated to each node. This dynamic and personalized parameter configuration method ensures that the virtual inertia and damping coefficient of each inverter do not exceed the physical limits of the equipment, while also providing stronger dynamic support to voltage-sensitive areas such as feeder ends. The resulting personalized parameter configuration scheme for the entire photovoltaic cluster significantly improves the voltage stability and operational safety of the photovoltaic cluster under weak grid conditions compared to uniform and fixed parameter settings. It also avoids the problem of some nodes being over-adjusted while other nodes are underutilized, achieving a balance between global collaboration and local optimization.

[0068] S4: Based on the parameter configuration, a preliminary allocation strategy for distributed reactive power support is determined through real-time interaction and dynamic game among the distributed energy intelligent agents.

[0069] After generating personalized parameter configurations for each distributed energy agent, the photovoltaic cluster needs to further coordinate the reactive power output of each node in order to achieve rapid voltage recovery and overall economic balance.

[0070] Based on the parameter configuration, a preliminary allocation strategy for distributed reactive power support is determined through real-time interaction and dynamic game theory among distributed energy agents. Without a central controller, each node exchanges only local information with its neighbors. A consensus protocol enables all nodes to reach an agreement on voltage-reactive power sensitivity. Then, within a game theory framework, each node iteratively adjusts its reactive power output, ultimately converging the entire cluster into a coordinated reactive power allocation scheme.

[0071] This embodiment does not restrict the strategies for interaction and game. The specific implementation method adopted depends on the communication reliability, node computing power, real-time requirements, and complexity of the running scenario.

[0072] For example, a distributed consensus protocol can be used to first exchange local voltage gradient information: each node calculates the voltage based on its local parameter configuration and voltage measurements. Reactive power sensitivity (e.g., obtaining the gradient through inertial filtering and damping adjustment of the difference between local and neighboring voltages) is assessed. Then, an average consensus or maximum consensus algorithm is used to repeatedly exchange gradient data with neighbors, ensuring that the gradient estimates of all nodes converge. For scenarios with poor communication conditions or a large number of nodes, event-triggered or asynchronous consensus protocols can be used, initiating communication only when the gradient deviation exceeds a threshold, thus reducing network load. The game theory payoff function can also be designed differently based on node type.

[0073] In one embodiment, S4 may include S41~S42, such as Figure 5As shown, S41~S42 are as follows: S41: Based on the parameter configuration, the local voltage gradient information of adjacent nodes is exchanged in real time through a preset distributed consensus protocol.

[0074] Based on the parameter configuration, local voltage gradient information of neighboring nodes is exchanged in real time through a preset distributed consensus protocol. Without obtaining global information, a set of gradient information reflecting voltage-reactive power sensitivity is obtained by utilizing local measurements and neighbor exchanges.

[0075] Each node, configured with parameters such as its virtual inertia and damping coefficient, calculates its local voltage gradient characteristics based on local and neighboring voltage data. It then repeatedly exchanges this gradient information with neighboring nodes through a pre-defined distributed consensus protocol, iterating multiple times until the gradient estimates of all nodes converge. Depending on different communication environments and convergence requirements, the consensus protocol can employ synchronous, asynchronous, or event-triggered modes to balance communication load and convergence speed.

[0076] In one embodiment, S41 may include S411~S412, such as Figure 6 As shown, S411~S412 are as follows: S411: Calculate the local voltage gradient information of each of the distributed energy intelligent agents according to the parameter configuration.

[0077] Each distributed energy agent first collects the real-time voltage amplitude of its local node using a local voltage sensor. Then, based on a pre-defined communication topology, it obtains the real-time voltage amplitudes of all directly connected neighboring nodes through a distributed consensus protocol. For each neighbor, the voltage of the local node is subtracted from the voltage of the neighbor to obtain the original voltage difference for the corresponding branch.

[0078] The original voltage difference is subjected to inertial filtering based on the virtual inertia setting in the parameter configuration. For example, a first-order inertial filter is used to smoothly attenuate the rapid changes in the voltage difference. The filtering strength is related to the size of the virtual inertia.

[0079] The damping adjustment of the filtered voltage difference is usually performed by multiplying the filtered difference by the damping coefficient to obtain the local gradient contribution of the branch.

[0080] The gradient contributions of all branches are summarized to form the local voltage gradient information of this node.

[0081] S412: Through a preset distributed consensus protocol, each of the distributed energy agents sends local voltage gradient information to neighboring distributed energy agents and receives local voltage gradient information sent by neighboring distributed energy agents.

[0082] The calculated local voltage gradient information is packaged into standard data frames and sent to all physically adjacent nodes via the communication network. Simultaneously, each node continuously listens to its port to receive gradient data from its neighbors.

[0083] After receiving neighbor data, the node updates its gradient estimate according to a preset consensus iteration formula. For example, the average consensus algorithm is used to take the weighted average of its own estimate and the neighbor's estimate as the new estimate.

[0084] Repeat the sending and updating process until the difference between two adjacent estimates is less than a pre-set convergence threshold, or the maximum number of iterations is reached. After convergence, the gradient estimates of all nodes become consistent.

[0085] In this embodiment, by utilizing local voltage sensors and neighboring voltage data, and through inertial filtering and damping adjustment, the interference of voltage measurement noise and instantaneous fluctuations on gradient estimation is effectively suppressed. This results in generated local voltage gradient information that is both sensitive and stable, providing high-quality foundational data for subsequent consensus negotiation. A distributed consensus protocol allows each node to exchange gradient information with its neighboring nodes, converging the gradient estimation of the entire network to a consistent state without relying on global communication. This significantly reduces communication bandwidth consumption and the risk of single-point failures, while accelerating information fusion. The synergistic effect of these two mechanisms enables the photovoltaic cluster to quickly and reliably achieve a unified understanding of voltage-reactive power sensitivity under dynamically changing operating conditions. This lays a solid collaborative foundation for subsequent dynamic game theory in reactive power support, thereby improving the robustness and real-time performance of the entire control system.

[0086] S42: Treat each of the distributed energy intelligent agents as an intelligent agent, and determine the initial allocation strategy for distributed reactive power support based on the dynamic game among the photovoltaic cluster intelligent agents.

[0087] Based on the aforementioned gradient consistency, each distributed energy agent is regarded as an agent with self-interested behavior, and the initial allocation strategy for reactive power support is solved through dynamic game theory.

[0088] Each node constructs a game payoff function that includes voltage deviation penalties, reactive power output costs, and penalties for differences in output between neighboring nodes. A distributed iterative algorithm is then used to seek the Nash equilibrium solution. For example, the optimal response iteration method can be used, where each node, given its current neighbor's output, calculates the reactive power output value that maximizes its own payoff in each round and updates and broadcasts this value to its neighbors, iterating repeatedly until convergence. Alternatively, a distributed gradient ascent method can be used, where each node gradually adjusts its output value along the first-order gradient direction of its own payoff function with respect to output, with the step size decreasing with the number of iterations to ensure convergence.

[0089] In one embodiment, S42 may include S421-S423, such as... Figure 7 As shown, S421~S423 are as follows: S421: Construct a game payoff function for each of the distributed energy agents.

[0090] A game payoff function is constructed for each of the distributed energy agents. This payoff function may include a voltage deviation penalty term, a reactive power output cost term, and a coordination consistency term with neighboring distributed energy agents.

[0091] The revenue function is typically designed as a differentiable concave function. The voltage deviation penalty term can be the negative of the square of the difference between the local voltage and the rated voltage, meaning the larger the deviation, the lower the revenue. The reactive power cost term can be the negative of the square of the reactive power, meaning the greater the output, the higher the cost. The coordination consistency term can be the negative of the square of the difference in reactive power output between adjacent nodes to encourage smoothing. The weighting coefficients of each term can be dynamically adjusted by each node based on its electrical location, equipment aging, or real-time electricity price.

[0092] S422: Based on the game payoff function, iteratively update the reactive power output setting value of each of the distributed energy intelligent agents.

[0093] In this embodiment, either a distributed gradient descent algorithm or a distributed Nash equilibrium algorithm can be used. When using the distributed gradient descent algorithm, each node calculates the gradient of its own reward function with respect to reactive power output, and then adjusts its output value along the gradient's ascending direction with a certain step size. This step size typically decreases with the number of iterations to ensure convergence. When using the distributed Nash equilibrium algorithm, each node analytically solves for the optimal response function given its neighbor's policy, and then directly updates its output to that optimal response value. After each update, the node broadcasts the new output value to its neighbors and receives new values ​​from them, iterating in a loop.

[0094] When using the distributed Nash equilibrium solution algorithm, each node initializes its reactive power output setting based on its current operating state. Then, it enters a loop iteration. In each iteration, each node receives the current reactive power output values ​​of all its neighbors and substitutes them into its own game payoff function. At this point, all variables in the payoff function except for the node's own reactive power output are known constants. The node finds the reactive power output value that maximizes its own payoff by solving a univariate optimization problem; this value is the optimal response for that round. The node broadcasts its new reactive power output value to all its neighbors and simultaneously receives new values ​​from its neighbors. All nodes update their output synchronously or asynchronously and then enter the next iteration. This process is repeated, and as the number of iterations increases, the reactive power output values ​​of each node gradually stabilize. When the change in reactive power output of all nodes between two adjacent rounds is less than a preset threshold, the algorithm converges. At this point, the output values ​​of all nodes constitute a Nash equilibrium, i.e., a stable reactive power support allocation strategy.

[0095] S423: When the convergence condition of the game payoff function is met, the reactive power output setting value of each of the distributed energy intelligent agents is used as the initial allocation strategy for the distributed reactive power support.

[0096] This step defines the iteration termination conditions and outputs the final strategy. Common convergence criteria include: the maximum change in reactive power output of all nodes is less than a preset threshold, or the rate of change of the revenue function value of all nodes is less than a set percentage, or the sum of squared norms of the gradients of all nodes between two adjacent iterations is less than a threshold. When the conditions are met, each node freezes its current reactive power output setting as the initial allocation strategy.

[0097] In this embodiment, by constructing a benefit function that includes voltage deviation, reactive power cost, and neighbor collaboration, and employing a distributed Nash equilibrium iteration that only requires neighbor information, each node spontaneously forms a network-wide coordinated reactive power allocation scheme while pursuing its own profit maximization. This mechanism requires no central controller, has low computational complexity, converges quickly, and the final Nash equilibrium solution has unilateral stability, significantly improving the robustness, self-organization capability, and voltage control economy of the photovoltaic cluster in weak communication environments.

[0098] S5: Based on the preliminary allocation strategy of the distributed reactive power support, optimize the harmonic suppression parameters among the distributed energy agents and construct the output adjustment matrix.

[0099] The initial allocation strategy for distributed reactive power support further optimizes the harmonic suppression parameters among various distributed energy agents and constructs an output adjustment matrix accordingly. This improves power quality while controlling voltage and provides a pre-calculated control benchmark for rapid response to sudden disturbances.

[0100] In optimizing harmonic suppression parameters, one approach is to treat each node as a game player. Each node constructs its own harmonic suppression cost function, which typically includes a penalty term for the local harmonic current distortion rate and the opportunity cost of the impact of reactive power output changes on harmonics. Each node uses a distributed Nash equilibrium solution algorithm to exchange its current harmonic suppression parameters with neighboring nodes, iteratively updating its parameter values ​​until it converges to a Nash equilibrium point, thereby obtaining the optimal harmonic suppression parameters for each node.

[0101] Another approach is to employ a joint optimization strategy, incorporating both harmonic suppression parameters and reactive power output into a unified game payoff function, and solving it synchronously using a multi-objective distributed optimization algorithm to achieve better global coordination.

[0102] In one implementation, S5 may include S51~S52, such as Figure 8 As shown, S51~S52 are as follows: S51: Based on the preliminary allocation strategy of the distributed reactive power support and the preset Nash equilibrium solution algorithm, optimize the harmonic suppression parameters among the distributed energy agents.

[0103] Each distributed energy agent's local controller first constructs a harmonic suppression cost function based on the local reactive power output setpoint determined in step S4 and the current voltage and current measurement data. This cost function typically contains two core parts: the first part is a penalty term for the local harmonic current distortion rate, whose value increases sharply as the harmonic current injected into the grid by the node increases, encouraging the node to actively suppress the harmonics it generates; the second part is the cost considering the impact of changes in reactive power output on harmonic suppression capability, i.e., if the node's current reactive power output is already close to the inverter's capacity limit, then its remaining capacity available for harmonic suppression is small, and the corresponding opportunity cost is higher. The cost function can also introduce a coordination consistency term between adjacent nodes to ensure a smooth transition of harmonic suppression parameters between adjacent nodes, avoiding local overcompensation or undercompensation.

[0104] After constructing the cost function, each node interacts with its neighbors using a pre-defined distributed Nash equilibrium solution algorithm to find the Nash equilibrium point of the cost function. In each iteration, each node receives the harmonic suppression parameters currently used by its neighbors.

[0105] Assuming neighbor parameters remain constant, each node seeks the harmonic suppression parameter value that minimizes its own cost function. This univariate optimization problem can be solved analytically by setting the derivative to zero, or numerically when the cost function is non-convex. The new parameters obtained are the node's optimal response to the current neighbor's policy. The node broadcasts this optimal response value to all neighboring nodes and simultaneously receives new values ​​broadcast by its neighbors. All nodes update their parameters synchronously or asynchronously before entering the next iteration. This process is repeated until the parameter changes of all nodes in two consecutive iterations are less than a preset threshold, at which point convergence is considered achieved. At this point, the harmonic suppression parameters of all nodes constitute a pure policy Nash equilibrium point, i.e., the optimal harmonic suppression parameters.

[0106] Each local controller updates its local harmonic suppression coefficient based on this optimal parameter for subsequent inverter control.

[0107] S52: Based on the optimized harmonic suppression parameters, the harmonic impedance of each distributed energy intelligent agent is reshaped to construct the output adjustment matrix after harmonic impedance reshaping.

[0108] After obtaining the optimal harmonic suppression parameters for each node, this sub-step further adjusts the electrical characteristics of the nodes and constructs an output adjustment matrix for rapid response to sudden disturbances.

[0109] First, the local controller of each distributed energy intelligent entity calculates the target harmonic impedance value that the node should present at the harmonic frequency based on the optimal harmonic suppression parameters obtained in sub-step S51. The target impedance typically includes two parts: equivalent resistance and equivalent reactance. Its value should be set so that the node exhibits the expected resistive, inductive, or capacitive characteristics to harmonic currents, thereby suppressing harmonic amplification or avoiding resonance with the line impedance. For example, if it is desired that the node absorbs harmonic currents, a smaller resistance value can be set for the target impedance; if it is desired to isolate upstream harmonics, a larger inductive impedance can be set.

[0110] After calculating the target harmonic impedance value, each node reshapes the harmonic impedance by adjusting the inverter's control loop parameters. Specific adjustment methods include: modifying the proportional-integral controller coefficients in the inverter's voltage or current loop to change the frequency characteristics of the output impedance; connecting a virtual impedance loop in series or parallel in the control loop to simulate a voltage drop corresponding to the target impedance value using an algorithm; or directly utilizing the compensation current generation function of the active power filter to make the inverter present the required equivalent impedance at harmonic frequencies. The local controller monitors the harmonic components of the grid connection point voltage and current in real time, and uses closed-loop regulation to make the actual equivalent harmonic impedance approach the target value.

[0111] Based on the electrical characteristics after harmonic impedance reshaping, each node collaboratively constructs an output adjustment matrix. This matrix is ​​a square matrix with dimensions equal to the number of nodes multiplied by the number of nodes. Its elements represent the coupling sensitivity of the voltage or current at the grid connection point of other nodes when the active or reactive power output of a node changes. A typical method for constructing this matrix is ​​as follows: each local controller uses the equivalent model of the inverter after harmonic impedance reshaping, combined with the grid line impedance parameters identified in step S1, to calculate the transmission coefficient between the output power change of its node and the voltage change of adjacent nodes. Specifically, this calculation can use the corresponding elements in the inverse of the node admittance matrix, or solve for the voltage-power sensitivity through linearized power flow equations. Each node sends its calculated transmission coefficients related to itself to the corresponding node via a distributed consensus protocol, or to a temporary coordinator. Ultimately, all nodes obtain the complete output adjustment matrix, or each node obtains the necessary information for its own row / column.

[0112] In this embodiment, the coordinated optimization of harmonic suppression parameters and the adaptive construction of the output adjustment matrix are realized in a fully distributed framework, laying a precise and efficient control foundation for rapid response under subsequent sudden disturbances.

[0113] S6: When a sudden interference is detected, according to the output adjustment matrix, each of the distributed energy intelligent agents is controlled to autonomously identify the type of interference based on local measurement information and determine the priority of the interference response.

[0114] In this embodiment, the detection of sudden interference is based entirely on local measurement information from each distributed energy agent, without the need for global communication or central coordination. Each node can collect the voltage signal at the grid connection point in real time, calculate the voltage change rate, and determine whether interference has occurred and its type based on this. Wavelet transform or fast Fourier transform can also be used to extract the time-frequency characteristics of the voltage waveform to determine whether interference has occurred and its type.

[0115] Upon detecting a sudden disturbance, the distributed energy agent is controlled, based on the output adjustment matrix, to autonomously identify the disturbance type and determine the disturbance response priority according to local measurement information. Each distributed energy agent first autonomously identifies the disturbance type using local measurement information, and then, in conjunction with the output adjustment matrix, determines its own response priority under the current disturbance.

[0116] Specifically, the output adjustment matrix records the power-voltage coupling sensitivity between each node. It is not used directly to determine the type of interference, but rather to dynamically adjust the priority after determining the type, so that nodes with greater influence and more critical positions can obtain higher response priority, thereby achieving more reasonable distributed coordination.

[0117] In one implementation, when a sudden interference is detected, based on a pre-defined plug-and-play support mechanism under a global communication-free architecture, each of the distributed energy intelligent agents autonomously determines the type of interference according to the locally measured voltage change rate, and determines the corresponding interference response priority according to the type of interference.

[0118] Each local controller collects the instantaneous voltage value of the grid-connected point in real time at a fixed sampling frequency, typically 128 points per cycle or one point per millisecond. The controller calculates the effective voltage value or fundamental amplitude, and then calculates the voltage rate of change, i.e., the change in voltage amplitude per unit time. Common calculation methods include dividing the voltage difference between two consecutive sampling points by the sampling interval, or using the least-squares linear fitting slope within a sliding window to suppress random noise. To improve anti-interference capability, the voltage signal can be preprocessed with first-order low-pass filtering or moving average. When the absolute value of the voltage rate of change exceeds a preset trigger threshold, such as exceeding zero times the rated voltage per millisecond, the controller determines that a sudden interference has been detected and then enters the type determination process.

[0119] When autonomously identifying interference types, the controller compares the measured voltage change rate with a preset interference type threshold range. If the voltage change rate is negative and its absolute value exceeds the first threshold (e.g., below -0.2 times the rated voltage per millisecond), it is judged as a voltage sag. If the voltage change rate is positive and exceeds the second threshold (e.g., above +0.2 times the rated voltage per millisecond), it is judged as a voltage surge. If the absolute value of the voltage change rate is small (e.g., below 0.05 times the rated voltage per millisecond), but at the same time the total harmonic distortion rate suddenly increases above the preset harmonic threshold (e.g., increases by 5%), it is judged as harmonic interference. To make the judgment more robust, a hysteresis comparator can be used: the interference type is only confirmed when the voltage change rate exceeds the threshold for multiple consecutive sampling points (e.g., three consecutive sampling points), avoiding false triggering due to instantaneous spikes. In addition, the absolute offset of the voltage amplitude can be used as an auxiliary criterion. For example, when the voltage is below 0.9 times the rated value or above 1.1 times the rated value, even if the voltage change rate is slightly lower, it is judged as the corresponding interference.

[0120] After determining the interference type, each local controller queries a pre-stored interference response priority mapping table. This mapping table defines the baseline response priority for each interference type—voltage sag, voltage surge, and harmonic interference—for this node. For example, for voltage sag, nodes at the end of the feeder typically have the highest priority; for harmonic interference, nodes closer to the harmonic source or with higher harmonic impedance have higher priority. The mapping table can be configured offline based on electrical location before node commissioning, or it can be automatically generated automatically using simple rules, such as based on impedance identification results, when the node is plug-and-play. Based on the currently identified interference type, the node reads the corresponding priority level from the mapping table, such as high, medium, and low, or a specific numerical level, as the basis for subsequent output adjustments.

[0121] In this embodiment, each distributed energy intelligent agent does not require a central controller or global communication. It can quickly and autonomously complete the identification of the type of sudden interference and the determination of the response priority by relying solely on local voltage measurement. This provides an accurate and timely triggering basis for updating local control parameters and adjusting output in subsequent steps.

[0122] S7: Based on the interference response priority, update the inverter control parameters locally until the photovoltaic cluster stability index reaches the target threshold.

[0123] Each distributed energy intelligent agent's local controller updates the inverter's control parameters locally based on its own priority. At the same time, it determines whether the stability indicators of the entire photovoltaic cluster have recovered to within the target threshold through local measurements and limited neighbor information exchange. Once the target is met, the response control ends.

[0124] In one embodiment, S7 may include S71~S73, such as Figure 9 As shown, S71~S73 are as follows: S71: According to the interference response priority, each of the distributed energy intelligent agents locally updates the inverter control parameters to perform output adjustment.

[0125] Each distributed energy intelligent agent's local controller autonomously updates the inverter's control parameters locally based on the disturbance response priority and the output adjustment matrix, thereby executing corresponding output adjustments. High-priority nodes act first and with larger adjustment ranges, while low-priority nodes respond more slowly or with smaller adjustment ranges.

[0126] Each node, based on its priority level, queries a pre-defined adjustment step size or target value from a pre-defined adjustment mapping table. For example, in the event of a voltage sag, a high-priority node needs to rapidly increase reactive current injection to support the voltage; its reactive power output target value can be set to 80% of the current available capacity. For medium-priority nodes, this is set to 50%, and for low-priority nodes, it is set to 20%. The nodes directly update the reactive current reference value or voltage-reactive power droop coefficient in the inverter control loop based on the target value. Another approach is to use integral regulation: nodes are set with different integral rates according to their priority; the higher the priority, the larger the integral coefficient, and the faster the output adjustment speed, thus allowing high-priority nodes to complete the adjustment before low-priority nodes.

[0127] S72: Monitor the execution feedback data of each of the distributed energy intelligent agents to determine whether the stability index of the photovoltaic cluster has reached the target threshold.

[0128] Each local controller collects feedback data such as voltage, current, active power output, and reactive power output from its local node in real time. The sampling frequency is typically consistent with the control cycle, for example, once every ten milliseconds. Based on this data, the node calculates its contribution to local stability metrics. Common stability metrics include voltage deviation (the absolute value of the difference between the actual voltage and the rated voltage), frequency deviation (the difference between the actual frequency and the rated frequency), and total harmonic distortion (THD). The node can directly obtain voltage deviation and harmonic distortion through local measurements, while frequency deviation requires a phase-locked loop (PLL) or zero-crossing detection.

[0129] To determine the overall stability of the cluster, each node needs to obtain global or at least local stability metrics information. Since this step does not rely on global communication, nodes typically achieve distributed consensus by exchanging stability metric compliance information with neighboring nodes. For example, after completing local adjustments, each node checks if its local voltage deviation is less than a threshold, such as no more than two percent of the rated voltage. If it meets the threshold, it broadcasts a "compliant" flag to neighboring nodes; if it does not, it broadcasts a "non-compliant" flag. After receiving flags from all neighbors, if each node meets the threshold and all its neighbors also meet the threshold, it determines that the cluster stability metrics have reached the target threshold.

[0130] S73: When the stability index of the photovoltaic cluster reaches the target threshold, the target adjustment result is obtained, and the response control of the current sudden disturbance ends.

[0131] The node saves all parameter values ​​related to the current response, including its current virtual inertia, damping coefficient, reactive power setpoint, harmonic suppression coefficient, and inverter control loop parameters, to its local non-volatile storage unit. These values ​​can be used as initial references in the event of similar disturbances in the future, and can also be used for fault tracing and system analysis. Simultaneously, the node generates a sudden disturbance response completion flag. This flag can be a local state variable or a brief notification sent to the monitoring system or neighboring nodes via the communication interface.

[0132] After generating the completion flag, each local controller exits the current emergency disturbance response control process and resumes the normal operation mode of the photovoltaic cluster. Normal operation mode typically refers to regular control based on personalized parameter configurations and initial strategies, no longer maintaining the over-regulation parameters used in emergency situations.

[0133] In this embodiment, each distributed energy agent can autonomously adjust its control parameters according to its own priority under sudden interference, and determine the stable state of the cluster through local feedback and simple bit information exchange with its neighbors. After reaching the target, it smoothly exits the emergency response. The entire process does not require global communication or centralized decision-making, realizing plug-and-play, robust distributed collaborative control, effectively ensuring the voltage stability, frequency stability and harmonic suppression performance of the photovoltaic cluster under various interferences.

[0134] In this embodiment, the reliance on a central controller is completely eliminated. All operations are based on distributed computing using local controllers and limited communication between adjacent nodes. This eliminates the need for global synchronization or centralized optimization, significantly reducing communication bandwidth requirements and the risk of single points of failure. This allows the photovoltaic cluster to operate stably even in environments with weak communication or momentary network outages.

[0135] The physical characteristics of the power grid are deeply integrated into control decisions. By embedding physical constraints and customizing virtual inertia and damping coefficients according to electrical location, it is ensured that all control commands do not violate physical boundaries such as voltage limits and equipment capacity. This avoids infeasible outputs that may result from purely data-driven or experience-based control, and improves the safety and reliability of the control strategy.

[0136] Joint optimization of voltage control and power quality was achieved. By coordinating the reactive power output of each node through distributed game theory, voltage fluctuations were effectively suppressed. By optimizing harmonic suppression parameters and constructing an output adjustment matrix, the voltage was improved without causing harmonic amplification or resonance, thus solving the problem of conflict between voltage regulation and harmonic suppression in traditional methods.

[0137] It possesses plug-and-play rapid emergency response capabilities. Each node can autonomously identify the type of interference and determine its priority based solely on its local voltage change rate, without waiting for remote commands; it updates inverter parameters locally according to priority and judges cluster stability indicators by exchanging simple location information with neighbors, achieving millisecond-level fully distributed recovery control and overcoming the response latency of centralized architectures.

[0138] In summary, the embodiments of the present invention achieve fully distributed autonomous collaboration from steady-state optimization to transient response without the need for a central controller and global communication, thus solving problems such as reliance on centralized communication, lack of physical constraints, and insufficient collaboration capabilities.

[0139] Reference Figure 10 This invention also provides a photovoltaic cluster intelligent control device based on physical constraint embedding and dynamic game theory, comprising: The first processing unit 100 is used to acquire real-time operating data of each distributed energy intelligent agent, perform grid impedance identification on the real-time operating data, analyze the power output distribution characteristics of each distributed energy intelligent agent, and obtain the current power output distribution status. The second processing unit 101 is used to intelligently analyze the preliminary parameter adjustment range required for voltage fluctuation suppression based on the current power distribution state and the preset physical constraint embedding algorithm. The third processing unit 102 is used to adaptively and dynamically adjust the virtual inertia and damping coefficient when the preliminary parameter adjustment range exceeds the preset threshold range, and generate the parameter configuration of each of the distributed energy intelligent agents. The fourth processing unit 103 is used to determine the initial allocation strategy for distributed reactive power support based on the parameter configuration and through real-time interaction and dynamic game between the distributed energy intelligent agents. The fifth processing unit 104 is used to optimize the harmonic suppression parameters between the distributed energy agents and construct the output adjustment matrix based on the preliminary allocation strategy of the distributed reactive power support. The sixth processing unit 105 is used to control each of the distributed energy intelligent agents to autonomously identify the type of interference and determine the priority of interference response based on local measurement information when a sudden interference is detected, according to the output adjustment matrix. The seventh processing unit 106 is used to automatically update the inverter control parameters locally according to the interference response priority until the photovoltaic cluster stability index reaches the target threshold.

[0140] Further, the first processing unit is specifically used for: Acquire photovoltaic power generation data and grid connection status information collected by each of the distributed energy intelligent agents; Based on the photovoltaic power generation data, the grid access status information, and the preset recursive least squares algorithm, the grid impedance is identified, and the output distribution characteristics of each of the distributed energy intelligent agents in the photovoltaic cluster are analyzed to obtain the current output distribution status.

[0141] Furthermore, the second processing unit is specifically used for: According to the preset physical constraint embedding algorithm, the physical characteristics of the power grid impedance are incorporated as constraints into the training process of the neural network. The current power output distribution state is input into the neural network to predict the voltage change trend based on potential instability factors in historical voltage fluctuation records, and to determine the preliminary parameter adjustment range required for voltage fluctuation suppression.

[0142] Furthermore, the third processing unit is specifically used for: When the initial parameter adjustment range exceeds the preset threshold range, the degree to which the initial parameter adjustment range exceeds the preset threshold range is calculated; Obtain the electrical location information of each of the distributed energy intelligent agents; Based on the degree of excess and the electrical location information of each of the distributed energy intelligent agents, the virtual inertia adjustment amount and damping coefficient adjustment amount of each of the distributed energy intelligent agents are determined by real-time dynamic calculation. The virtual inertia setting value of each of the distributed energy intelligent agents is updated according to the virtual inertia adjustment amount, and the damping coefficient setting value of each of the distributed energy intelligent agents is updated according to the damping coefficient adjustment amount, thereby generating the parameter configuration of each of the distributed energy intelligent agents.

[0143] Furthermore, the fourth processing unit is specifically used for: Based on the parameter configuration, the local voltage gradient information of adjacent nodes is exchanged in real time through a preset distributed consensus protocol; Each of the distributed energy intelligent agents is regarded as an intelligent agent, and a preliminary allocation strategy for distributed reactive power support is determined based on the dynamic game among the photovoltaic cluster intelligent agents.

[0144] Furthermore, the fourth processing unit is specifically used for: Based on the parameter configuration, calculate the local voltage gradient information of each of the distributed energy intelligent agents; Through a preset distributed consensus protocol, each of the distributed energy agents sends local voltage gradient information to neighboring distributed energy agents and receives local voltage gradient information sent by neighboring distributed energy agents.

[0145] Furthermore, the fourth processing unit is specifically used for: Construct game payoff functions for each of the aforementioned distributed energy intelligent agents; Based on the game payoff function, the reactive power output setting value of each of the distributed energy intelligent agents is iteratively updated. When the convergence condition of the game payoff function is met, the reactive power output setting value of each of the distributed energy intelligent agents is used as the initial allocation strategy for the distributed reactive power support.

[0146] Furthermore, the fifth processing unit is specifically used for: Based on the preliminary allocation strategy of the distributed reactive power support and the preset Nash equilibrium solution algorithm, the harmonic suppression parameters among the distributed energy agents are optimized. Based on the optimized harmonic suppression parameters, the harmonic impedance of each distributed energy intelligent agent is reshaped to construct the output adjustment matrix after harmonic impedance reshaping.

[0147] Furthermore, the sixth processing unit is specifically used for: When a sudden interference is detected, based on the plug-and-play support mechanism under the preset global communication-free architecture, each of the distributed energy intelligent agents autonomously determines the type of interference according to the locally measured voltage change rate, and determines the corresponding interference response priority according to the type of interference.

[0148] Furthermore, the seventh processing unit is specifically used for: According to the interference response priority, each of the distributed energy intelligent agents locally updates the inverter control parameters to perform output adjustment. Monitor the execution feedback data of each of the distributed energy intelligent agents to determine whether the stability indicators of the photovoltaic cluster have reached the target threshold. When the stability index of the photovoltaic cluster reaches the target threshold, the target adjustment result is obtained, and the response control for the current sudden disturbance ends.

[0149] Reference Figure 11 The present invention also provides a local controller, the internal structure of which can be as follows: Figure 11As shown. The local controller includes a processor, memory, network interface, and database connected via a system bus. The processor in the local controller provides computing and control capabilities. The memory of the local controller includes non-volatile storage media and internal memory. The non-volatile storage media stores operating devices, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the local controller stores signals, etc. The network interface of the local controller is used for communication with external terminals via a network connection. Furthermore, the local controller may also be equipped with input devices and a display screen, etc. When the computer program is executed by the processor, it implements a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory, including the following steps: acquiring an image to be identified; inputting the image to be identified into a text position detection model for text position detection to obtain multiple text box regions; inputting the multiple text box regions into a flip determination model for flip determination to obtain a flip flag corresponding to each text box region; detecting whether each flip flag is true; if so, flipping the corresponding text box region to obtain a flipped region. The flipped image corresponding to the flipped region is input into the text recognition model to obtain the recognized text. Those skilled in the art will understand that... Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the local controller on which the present application is applied.

[0150] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements a photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory, including the following steps: acquiring real-time operating data of each distributed energy intelligent agent, identifying grid impedance on the real-time operating data, analyzing the power output distribution characteristics of each of the distributed energy intelligent agents, and obtaining the current power output distribution state. Based on the current power distribution state and the preset physical constraint embedding algorithm, the preliminary parameter adjustment range required for voltage fluctuation suppression is intelligently analyzed; When the initial parameter adjustment range exceeds the preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate the parameter configuration of each of the distributed energy intelligent agents. Based on the parameter configuration, the initial allocation strategy for distributed reactive power support is determined through real-time interaction and dynamic game among the distributed energy intelligent agents. Based on the preliminary allocation strategy of the distributed reactive power support, the harmonic suppression parameters among the distributed energy agents are optimized, and the power output adjustment matrix is ​​constructed. When a sudden interference is detected, each of the distributed energy intelligent agents is controlled to autonomously identify the type of interference and determine the priority of interference response based on local measurement information according to the output adjustment matrix. Based on the interference response priority, the inverter control parameters are updated locally until the photovoltaic cluster stability index reaches the target threshold.

[0151] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0152] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0153] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game, the photovoltaic cluster comprising a plurality of distributed energy intelligent agents, characterized in that, The method includes: The system acquires real-time operational data collected by each distributed energy intelligent agent, identifies grid impedance on the real-time operational data, analyzes the output distribution characteristics of each distributed energy intelligent agent, and obtains the current output distribution status. Based on the current power distribution state and the preset physical constraint embedding algorithm, the preliminary parameter adjustment range required for voltage fluctuation suppression is intelligently analyzed; When the initial parameter adjustment range exceeds the preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate the parameter configuration of each of the distributed energy intelligent agents. Based on the parameter configuration, the initial allocation strategy for distributed reactive power support is determined through real-time interaction and dynamic game among the distributed energy intelligent agents. Based on the preliminary allocation strategy of the distributed reactive power support, the harmonic suppression parameters among the distributed energy agents are optimized, and the power output adjustment matrix is ​​constructed. When a sudden interference is detected, each of the distributed energy intelligent agents is controlled to autonomously identify the type of interference and determine the priority of interference response based on local measurement information according to the output adjustment matrix. Based on the interference response priority, the inverter control parameters are updated locally until the photovoltaic cluster stability index reaches the target threshold.

2. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game according to claim 1, characterized in that, The process of acquiring real-time operational data from each distributed energy intelligent agent, performing grid impedance identification on the real-time operational data, analyzing the output distribution characteristics of each distributed energy intelligent agent, and obtaining the current output distribution state includes: Acquire photovoltaic power generation data and grid connection status information collected by each of the distributed energy intelligent agents; Based on the photovoltaic power generation data, the grid access status information, and the preset recursive least squares algorithm, the grid impedance is identified, and the output distribution characteristics of each of the distributed energy intelligent agents in the photovoltaic cluster are analyzed to obtain the current output distribution status. 3.The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game according to claim 1, characterized in that, The step of intelligently analyzing the preliminary parameter adjustment range required for voltage fluctuation suppression based on the current power distribution state and a preset physical constraint embedding algorithm includes: According to the preset physical constraint embedding algorithm, the physical characteristics of the power grid impedance are incorporated as constraints into the training process of the neural network. The current power output distribution state is input into the neural network to predict voltage change trends based on potential instability factors in historical voltage fluctuation records, and to intelligently analyze the preliminary parameter adjustment range required for voltage fluctuation suppression.

4. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory according to claim 1, characterized in that, When the initial parameter adjustment range exceeds a preset threshold range, the virtual inertia and damping coefficient are adaptively and dynamically adjusted to generate parameter configurations for each of the distributed energy intelligent agents, including: When the initial parameter adjustment range exceeds the preset threshold range, the degree to which the initial parameter adjustment range exceeds the preset threshold range is calculated; Obtain the electrical location information of each of the distributed energy intelligent agents; Based on the degree of excess and the electrical location information of each of the distributed energy intelligent agents, the virtual inertia adjustment amount and damping coefficient adjustment amount of each of the distributed energy intelligent agents are determined by real-time dynamic calculation. The virtual inertia setting value of each of the distributed energy intelligent agents is updated according to the virtual inertia adjustment amount, and the damping coefficient setting value of each of the distributed energy intelligent agents is updated according to the damping coefficient adjustment amount, thereby generating the parameter configuration of each of the distributed energy intelligent agents.

5. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory according to any one of claims 1 to 4, characterized in that, The step of determining a preliminary allocation strategy for distributed reactive power support based on the parameter configuration, through collaborative interaction and dynamic game theory among the distributed energy agents, includes: Based on the parameter configuration, the local voltage gradient information of adjacent nodes is exchanged in real time through a preset distributed consensus protocol; Based on the dynamic game among distributed energy agents, a preliminary allocation strategy for distributed reactive power support is determined.

6. The intelligent control method for photovoltaic clusters based on physical constraint embedding and dynamic game theory according to claim 5, characterized in that, The step of exchanging local voltage gradient information of neighboring nodes in real time based on the parameter configuration and through a preset distributed consensus protocol includes: Based on the parameter configuration, calculate the local voltage gradient information of each of the distributed energy intelligent agents; Through a preset distributed consensus protocol, each of the distributed energy agents sends local voltage gradient information to neighboring distributed energy agents and receives local voltage gradient information sent by neighboring distributed energy agents.

7. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory according to claim 5, characterized in that, The preliminary allocation strategy for distributed reactive power support, determined based on dynamic game theory among distributed energy agents, includes: Construct game payoff functions for each of the aforementioned distributed energy intelligent agents; Based on the game payoff function, the reactive power output setting value of each of the distributed energy intelligent agents is iteratively updated. When the convergence condition of the game payoff function is met, the reactive power output setting value of each of the distributed energy intelligent agents is used as the initial allocation strategy for the distributed reactive power support.

8. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory according to any one of claims 1 to 4, characterized in that, The step of optimizing the harmonic suppression parameters among the distributed energy agents and constructing an output adjustment matrix based on the initial allocation strategy of the distributed reactive power support includes: Based on the preliminary allocation strategy of the distributed reactive power support and the preset Nash equilibrium solution algorithm, the harmonic suppression parameters among the distributed energy agents are optimized. Based on the optimized harmonic suppression parameters, the harmonic impedance of each distributed energy intelligent agent is reshaped to construct the output adjustment matrix after harmonic impedance reshaping.

9. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory according to any one of claims 1 to 4, characterized in that, When a sudden interference is detected, the process of controlling each of the distributed energy intelligent agents to autonomously identify the interference type and determine the interference response priority based on local measurement information, according to the output adjustment matrix, includes: When a sudden interference is detected, based on the plug-and-play support mechanism under the preset global communication-free architecture, each of the distributed energy intelligent agents autonomously identifies the type of interference according to the locally measured voltage change rate, and determines the corresponding interference response priority according to the type of interference.

10. The photovoltaic cluster intelligent control method based on physical constraint embedding and dynamic game theory according to any one of claims 1 to 4, characterized in that, The step of automatically updating the inverter control parameters locally based on the interference response priority until the photovoltaic cluster stability index reaches the target threshold includes: According to the interference response priority, each of the distributed energy intelligent agents locally updates the inverter control parameters to perform output adjustment. Monitor the execution feedback data of each of the distributed energy intelligent agents to determine whether the stability indicators of the photovoltaic cluster have reached the target threshold. When the stability index of the photovoltaic cluster reaches the target threshold, the target adjustment result is obtained, and the response control for the current sudden disturbance ends.

Images