Photovoltaic cluster collaborative regulation system and method based on digital twinning and topological association

Abstract

This invention relates to the field of distributed energy regulation in power distribution networks, specifically disclosing a photovoltaic cluster collaborative regulation system and method based on digital twins and topology association. The system includes a cloud-based decision-making layer, an edge control layer, and a terminal execution layer. The cloud deploys a global topology database and a digital twin engine to perform day-ahead optimization scheduling and generate regional power targets. The edge layer uses an improved NSGA-III algorithm incorporating topology constraint factors for multi-objective optimization, generating localized regulation commands. The terminal layer possesses four-quadrant enhanced control and communication interruption autonomy functions. The system constructs a three-level topology model of "distribution transformer-branch line-equipment," quantifies electrical distance and topology association, and establishes a dynamic adjacency matrix. This invention achieves topology-driven precise power allocation and dynamic clustering, effectively improving photovoltaic absorption capacity, regulation response speed, and power distribution network operation safety through digital twin pre-simulation and cloud-edge-terminal closed-loop collaboration.

Description

Technical Field

[0001] This invention belongs to the field of distributed energy regulation in power distribution networks, specifically a photovoltaic cluster collaborative regulation system and method based on digital twins and topological association. Background Technology

[0002] In the field of distributed energy regulation in power distribution networks, the group regulation and control of photovoltaic clusters is crucial for ensuring stable system operation and improving the capacity for renewable energy absorption. Currently, regulation strategies based on optimization algorithms (such as particle swarm optimization and genetic algorithms) have been widely researched and applied, with their core objectives typically focusing on achieving regional power balance or maintaining voltage stability. However, these methods often treat the distribution network as a logically uniform whole when formulating strategies. Their core flaw lies in completely ignoring the decisive impact of the inherent and complex physical topology of the distribution network (such as branch impedance, line length, and switch connections) on power flow, voltage distribution, and line losses. This leads to a disconnect between the generated regulation commands and the actual electrical characteristics of the power grid, much like directing traffic without consulting a road map. While macroscopic goals may be achieved, precise, efficient, and safe power allocation cannot be realized, resulting in problems such as blind power allocation and persistently high line losses on long branches.

[0003] The rise of digital twins and edge computing technologies has provided new tools for the refined regulation of power grids. Digital twins can construct a high-fidelity model of the physical power grid in virtual space, achieving state awareness and policy prediction; edge computing can push computing power down to the network edge, enabling rapid processing and decision-making of local data. However, in current technological practices, these two technologies are often used independently or simply superimposed, failing to be deeply coupled with photovoltaic cluster regulation. A significant drawback is the lack of a closed-loop empowerment system encompassing "topology awareness - policy generation - edge execution - security protection." This results in cloud-based global optimization strategies being difficult to adapt to localized real-time topology states, edge decision-making lacking forward-looking prediction support, slow response, and a disconnect between the regulation process and security protection mechanisms, failing to meet the high-order requirements for dynamic adaptability and security in scenarios with high proportions of photovoltaic grid integration.

[0004] In the field of power grid security, islanding is a significant risk. Traditional islanding detection methods mainly rely on monitoring single electrical quantities such as voltage and frequency at the point of common coupling (PCC) to see if they exceed fixed thresholds. While this method is simple in principle, its fundamental flaw lies in its "indiscriminate" detection logic, failing to consider the specific location of photovoltaic (PV) units in the distribution network topology (e.g., whether they are at the end or near the end) to assess their islanding risk. PV units located at the electrical end, with weak connections to the main grid, are more prone to islanding, while those located near the end are less likely to form islands, and their voltage and frequency fluctuations are more likely to originate from other disturbances. Traditional methods do not differentiate between these factors, easily leading to false tripping of near-end PV units due to harmonic interference or instantaneous fluctuations, while true islanding at the end may fail to be detected in time due to conservative threshold settings. This not only threatens the safety of equipment and maintenance personnel but also reduces power supply reliability. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a photovoltaic cluster collaborative control system based on digital twins and topological association, which solves the problems of low control accuracy, slow dynamic response, and high operational risk caused by fragmented topological association, lack of cloud-edge-device collaborative closed loop, and insufficient security protection mechanisms in the prior art.

[0006] A photovoltaic cluster collaborative control system based on topology-driven and digital twin technology includes a cloud-based decision-making layer, an edge control layer, and a terminal execution layer.

[0007] The cloud-based decision layer is equipped with a global topology database and a digital twin engine to perform day-ahead optimization scheduling. The objective function is to minimize the total system operating cost. It also combines power balance, topology security, equipment operation and voltage quality constraints to generate regional power targets.

[0008] The edge control layer is equipped with a regional control unit, which receives the regional power target and performs multi-objective Pareto optimization using the improved NSGA-III algorithm with topological constraint factors to generate localized control commands.

[0009] The terminal execution layer includes a photovoltaic inverter, energy storage equipment, and charging pile, which are used to execute the localized control commands and have four-quadrant enhanced control and communication interruption autonomy functions.

[0010] The system is based on topology-driven architecture and constructs a three-level topology model of "distribution transformer-branch line-equipment". It characterizes the connection relationship between units by quantifying electrical distance and topological correlation, and constructs a dynamic adjacency matrix to reflect topology changes in real time.

[0011] Preferably, the three-level topology model of "distribution transformer-branch line-equipment" uses the following coding rules to achieve a unique mapping between unit identity and location:

[0012] The unit coding format for the transformer layer is "T-voltage level-transformer area ID";

[0013] The branch layer unit encoding format is "L-belonging T unit-serial number";

[0014] The device layer unit encoding format is "D-belonging L unit-type-serial number".

[0015] Preferably, the formula for calculating the electrical distance is:

[0016]

[0017] in, and These are the weighting coefficients, and , For equipment to distribution transformer physical distance, For equipment to distribution transformer The branch impedance, and The maximum length and impedance within the transformer area;

[0018] The formula for calculating the topological correlation degree is:

[0019]

[0020] in, This is the maximum electrical distance within the transformer substation area.

[0021] Preferably, the dynamic adjacency matrix The rules for assigning values ​​to its elements are as follows:

[0022] When units i and j are directly connected and the switch is closed. ;

[0023] When unit i and j are indirectly associated ;

[0024] When unit i and j are not connected or the switch is open. ;

[0025] The dynamic adjacency matrix is ​​updated every 5 minutes, and the update is based on the closed or open status of the smart switch.

[0026] Preferably, the improved NSGA-III algorithm used in the edge control layer introduces topology constraint factors including: electrical distance span within the cluster is not greater than 0.5, and / or the topology correlation degree of units within the cluster is not less than 0.5.

[0027] Preferably, the four-quadrant enhanced control of the terminal execution layer includes the following operating conditions:

[0028] Quadrant I: Energy storage absorbs active power and generates capacitive reactive power;

[0029] Quadrant II: Stored energy releases active power and generates inductive reactive power;

[0030] Quadrant III: Energy storage releases active power and absorbs reactive power;

[0031] Quadrant IV: Energy storage absorbs active power and reactive power.

[0032] Preferably, the digital twin engine constructs a 1:1 topology-level digital twin model for pre-simulating the following three core scenarios: topology mutual assistance scenario, extreme failure scenario, and security protection scenario; and adopts a 5-minute long short-term memory network LSTM rolling prediction model to improve the accuracy of the pre-simulation.

[0033] Preferably, the system further includes a security protection mechanism, which includes:

[0034] The island detection model with multiple features has the following island criterion:

[0035]

[0036] When the criterion value is ≥0.2, it is determined to be in an islanded state. This is the inverse correlation value between the equipment and the distribution transformer;

[0037] The topology sequence isolation execution logic performs circuit breaker isolation in the order from the end device to the branch line head switch.

[0038] A photovoltaic cluster coordinated control method based on topology-driven and digital twin technologies, applied to the system described above, includes the following steps:

[0039] A three-level topology model of "distribution transformer-branch line-equipment" is constructed, the electrical distance and topological correlation are calculated, and a dynamic adjacency matrix is ​​generated;

[0040] Based on the power grid's operating status and topology characteristics, the clustering of photovoltaic clusters is dynamically triggered and executed;

[0041] Perform pre-emptive optimization scheduling in the cloud, generate regional power targets, and distribute them to the edge layer;

[0042] Multi-objective collaborative optimization under topological constraints is performed at the edge layer to generate localized control commands and send them to terminal devices;

[0043] Power regulation is performed at the terminal device layer, and real-time data is fed back to the edge layer and the cloud to form a closed-loop control.

[0044] Key scenarios are simulated using digital twin models, and control strategies are dynamically optimized based on topological deviation rates.

[0045] Preferably, the step of dynamically triggering and executing the clustering partitioning of the photovoltaic cluster specifically includes:

[0046] Clustering is initiated when any of the following triggering conditions are met: the ratio of photovoltaic power output mutation value to rated power ≥ 20%, the distribution transformer load rate K ≥ 110%, and the average state of charge of energy storage. The original cluster association fails due to a load similarity of ≥85% between adjacent branches or a sudden change in topological correlation.

[0047] The threshold of the triggering condition is dynamically adjusted by using a near-end strategy to optimize the PPO algorithm.

[0048] Based on the collaborative potential score, the cells are divided into highly collaborative clusters using an improved K-means algorithm. The collaborative potential score integrates topological correlation, load similarity, and photovoltaic fluctuation consistency.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] By constructing a three-level topology model of "distribution transformer-branch line-equipment" and quantifying the relationship between electrical distance and topological correlation, the control strategy of photovoltaic clusters can be precisely matched with the physical connection characteristics of the distribution network. This fundamentally solves the problems of blind power allocation, excessive long branch line losses, and insufficient reactive power compensation accuracy caused by ignoring topological characteristics in traditional control, thereby significantly improving the local absorption capacity of photovoltaic output, effectively reducing the overall system line loss, and enhancing the stability of grid operation.

[0051] Leveraging a three-tiered cloud-edge-device collaborative architecture and digital twin pre-simulation technology, the system achieves a rapid closed loop from global optimization to local execution. The cloud layer is responsible for forward-looking strategy generation, the edge layer is responsible for multi-objective real-time optimization considering local topology constraints, and the terminal devices possess rapid response and autonomous capabilities to handle communication interruptions. This collaborative mechanism reduces the control response time from minutes to seconds, significantly improving the system's dynamic adaptability and response speed to photovoltaic fluctuations and load changes. Simultaneously, strategy pre-simulation verification avoids ineffective or risky operations, ensuring the economic efficiency and security of control measures.

[0052] By integrating topological features and multiple electrical quantities to construct a multi-dimensional security protection system, the system's proactive defense capabilities have achieved a qualitative leap. In particular, incorporating topological correlation into islanding detection criteria and combining it with topological sequence isolation logic significantly reduces the false positive rate caused by harmonic interference, enabling rapid and accurate fault location and isolation, effectively curbing the spread of faults. This not only greatly enhances the safety resilience of the distribution network in the face of extreme faults but also provides higher safety assurance for operation and maintenance personnel, laying a solid foundation for the safe and stable operation of new power systems. Attached Figure Description

[0053] Figure 1 This is a diagram of the architecture of the present invention;

[0054] Figure 2 Here is a flowchart of the dynamic clustering mechanism;

[0055] Figure 3 A three-level decision-making flowchart for cloud, edge, and endpoint;

[0056] Figure 4 A timeline diagram for the hierarchical structure and fault isolation of the rigid-flexible collaborative model;

[0057] Figure 5 A flowchart for topology-level digital twin pre-simulation and multi-dimensional security protection. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] like Figure 1 As shown:

[0060] Example 1: Topology-Driven Dynamic Clustering and Mesh Optimization Mechanism

[0061] This embodiment describes how the system achieves precise clustering and optimized control of photovoltaic clusters through dynamic clustering, and its process corresponds to... Figure 2 (Flowchart of dynamic clustering mechanism).

[0062] Topology construction and data acquisition (corresponding) Figure 2 (Steps for "Topology Construction" and "Real-time Data Acquisition")

[0063] The system is implemented in a 10kV distribution network area. First, a three-level topology model of "transformer-branch line-equipment" is constructed. For example, the transformer layer unit is coded as T-10kV-A01, a branch line under it is coded as LT-10kV-A01-03, and a photovoltaic device on this branch is coded as DLT-10kV-A01-03-PV-05. The system is configured so that edge nodes collect the power, energy storage SOC, and load power of the photovoltaic device every minute, and the branch line harmonic detector uploads harmonic distortion rate (THD) data every 100ms.

[0064] Clustering triggering and threshold optimization (corresponding) Figure 2 (Steps for "whether the power grid condition meets the clustering requirements" and "dynamically optimizing the threshold using the PPO algorithm")

[0065] like Figure 2 As shown in the flowchart, the system continuously monitors the grid's operating status. When the output of the photovoltaic device DLT-10kV-A01-03-PV-05 suddenly reaches 30% of its rated power within 5 minutes (exceeding the preset threshold of 20%), it is determined that the clustering requirement is met, and clustering is initiated. Simultaneously, the system calls the Proximity Policy Optimization (PPO) algorithm, which dynamically optimizes the photovoltaic fluctuation threshold from 20% to 22% based on 12-dimensional features such as season, time period, and average load factor.

[0066] Dynamic partitioning based on multi-feature fusion (corresponding) Figure 2 (Steps in "Dynamically Partitioning Clusters Based on Multi-Feature Fusion of K-means Clustering with Potential Scores")

[0067] The system calculates the cooperative potential score between branch line LT-10kV-A01-03 and its neighboring branch line LT-10kV-A01-01. This score incorporates the topological correlation (calculated as follows). = 0.78), load similarity (calculated Sim = 88%), and photovoltaic fluctuation consistency. The calculated potential score is 85. Based on this high score, the improved K-means algorithm forces these two branches into the same "photovoltaic fluctuation smoothing cluster".

[0068] Cluster control and closed-loop feedback (corresponding) Figure 2 (The steps of "cluster-differentiated regulation" and "closed-loop feedback")

[0069] The system implements a unified power stabilization strategy for the cluster, instructing units within the cluster to coordinate adjustments. After the strategy is implemented, the cluster's curtailment rate decreases from 8% to 3%. This adjustment effect is used as feedback data input to the reward function of the PPO algorithm, driving continuous iterative optimization of the clustering threshold, resulting in... Figure 2 The closed loop shown.

[0070] Example 2: Topology-enabled dynamic autonomous system of cloud-edge-device

[0071] This embodiment demonstrates the collaborative decision-making and execution process of the system under a three-tier cloud-edge-device architecture, and its overall process corresponds to... Figure 3 (Cloud-Edge-Device Three-Level Decision-Making Flowchart).

[0072] The cloud recently optimized the scheduling (corresponding to) Figure 3 (The steps in "Cloud → Current Optimized Scheduling")

[0073] like Figure 3 At the process's starting point, the cloud-based digital twin engine performs pre-optimized scheduling. The goal is to minimize the total system operating cost, while satisfying power balance and branch current requirements. Equipment power upper and lower limits and voltage quality The solution is obtained under the given constraints. After the solution is obtained, a "regional power target" is issued to the edge layer. For example, the upper limit of the total photovoltaic output of the edge region Alpha during the midday peak photovoltaic power generation period is required to be 5.2 MW.

[0074] Multi-objective collaborative optimization at the edge layer (corresponding) Figure 3 (Steps in "Edge Layer → Multi-Objective Optimization")

[0075] After receiving the target, the edge region control unit performs local optimization using an improved NSGA-III algorithm. The optimization objectives include minimizing voltage deviation. Three-phase unbalance (≤2%) and number of equipment operations. In the Pareto solution generation process, the algorithm introduces a topology constraint factor of "electrical distance span within the cluster ≤ 0.5", and prioritizes retaining solutions that meet this constraint, ultimately generating localized control commands.

[0076] Device-side execution and real-time feedback (corresponding) Figure 3 (The steps in "Device Side → Execution and Autonomy" and "Real-time Feedback")

[0077] At the device level (such as the energy storage converter), four-quadrant enhanced control is executed according to edge layer instructions. During the peak solar power generation period at noon (corresponding to quadrant I), the energy storage device absorbs active power (+P) and generates capacitive reactive power (+Q) to suppress frequency and voltage rise. If communication is interrupted, the device automatically activates an autonomous strategy: if the local voltage exceeds 1.05 pu, it automatically reduces the load until the voltage recovers. Real-time data from the device (load power, solar power fluctuation, SOC) is fed back to the edge layer.

[0078] Global iteration (corresponding) Figure 3 (Feedback and Iteration Steps)

[0079] The edge layer aggregates real-time data and control effects, synchronizing them to the cloud. The cloud uses this data to update the global topology database and digital twin model, supporting the next round of optimized scheduling and forming a complete system. Figure 3 The closed loop shown is "cloud target - edge optimization - terminal execution - feedback iteration".

[0080] Example 3: A topologically layered coupled rigid-flexible collaborative model of "optical-storage-charging-off" systems

[0081] This embodiment illustrates the differentiated control strategies of the system under normal and fault conditions, and its hierarchical structure and fault isolation timing correspond to... Figure 4 (Timing diagram of hierarchical and fault isolation in rigid-flexible collaborative model).

[0082] Flexible layer power linkage (under normal operating conditions, corresponding to) Figure 4 (The "flexible layer" part)

[0083] During normal midday operation, the system detected 150 kW of excess photovoltaic power in a certain distribution area that needed to be absorbed. This was based on topological correlation. electrical distance The system follows the formula Distribute power regulation. One located at the end ( = 0.9), closely related ( The energy storage unit with a voltage ratio of 0.85 was assigned a 45 kW absorption capacity. Meanwhile, for voltage-sensitive nodes (… ≥ 0.8), the system is inversely proportional to impedance. Allocate reactive power compensation to ensure stable voltage at the end.

[0084] Rigid layer fault isolation (fault condition, corresponding) Figure 4 (Rigid layer section and fault isolation timing)

[0085] When the system detects that a certain branch line simultaneously has voltage ,frequency Furthermore, rigid isolation is immediately triggered when the multi-feature island criterion is ≥0.2. For example... Figure 4 As shown in the timing diagram, isolation is performed in topological order:

[0086] Terminal equipment (D≥0.8): It will trip first within 20-30ms to cut off the source of the fault.

[0087] Mid-section equipment (0.3≤D<0.8): Delay 40-50ms for circuit breaker tripping and secondary isolation.

[0088] Branch line start-up switch (D<0.3): Completes full isolation within 100ms to ensure that the fault does not spread across the branch line.

[0089] During the circuit breaker tripping process, the edge layer simultaneously activates the emergency plan to ensure power supply to critical loads.

[0090] Example 4: Topology-level digital twin pre-simulation and multi-dimensional security protection

[0091] This embodiment illustrates how the system utilizes digital twins for proactive simulation and security protection, and its core process corresponds to... Figure 5 (Topology-level digital twin pre-simulation and multi-dimensional security protection).

[0092] Twin model construction and synchronization (corresponding) Figure 5 (Steps in "Topology-level Digital Twin Construction")

[0093] A 1:1 topology-level digital twin model is built in the cloud, importing a three-level topology structure of "distribution transformer-branch line-equipment" and labeling parameters such as impedance, length, and current carrying capacity. The model receives real-time data (dynamic adjacency matrix, voltage, frequency, SOC, photovoltaic output) uploaded from the edge layer every 100ms and updates its own status accordingly.

[0094] Core Scene Preview and LSTM Prediction (corresponding) Figure 5 (5-minute LSTM rolling prediction step)

[0095] Before implementing the new strategy, the digital twin model simulates three scenarios: "topology mutual assistance," "extreme failure," and "safety protection." The simulation uses 5-minute LSTM rolling prediction, taking as input the topology correlation time-series data and meteorological data from the past 24 hours, and outputting trends such as photovoltaic power output and load power for the next 5 minutes. For example, the simulation simulates the mutual assistance strategy of "transmitting power from a photovoltaic area to a heavily loaded area," verifying that it can reduce line loss rate and restore voltage to a safe range.

[0096] Multi-dimensional security protection and deviation correction (corresponding) Figure 5 (Steps for "Multi-feature Island Detection", "Bias Judgment" and "Initiation Correction")

[0097] Security protection: The system operates a multi-feature island detection model, and its criteria are as follows:

[0098] .

[0099] When the criterion is ≥0.2, trigger as follows Figure 5 The "topological sequence isolation" process is shown. To suppress false positives, only those weakly related to the distribution transformer are isolated. Device startup detection is performed with a delay of 20ms for confirmation. (< 0.5)

[0100] Deviation correction: System-defined topology deviation rate Perform deviation judgment. When If the accuracy exceeds 5% and persists for three consecutive cycles, initiate a correction process: for example, adjust the threshold for dynamic clustering or recalibrate the branch impedance parameters of the twin model to ensure the accuracy of the simulation and the effectiveness of the strategy. Figure 5 The diagram shows a rapid closed loop of "topology-driven → pre-decision → protection execution → feedback correction".

[0101] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0102] The accompanying drawings of the embodiments disclosed in this invention only involve structures relevant to the embodiments disclosed in this invention. Other structures can be referred to with common designs. Unless otherwise specified, the same embodiment and different embodiments of this invention can be combined with each other.

[0103] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A photovoltaic cluster coordinated control system based on topology-driven and digital twin technologies, characterized in that, It includes a cloud-based decision-making layer, an edge control layer, and a terminal execution layer; The cloud-based decision layer is equipped with a global topology database and a digital twin engine to perform day-ahead optimization scheduling. The objective function is to minimize the total system operating cost. It also combines power balance, topology security, equipment operation and voltage quality constraints to generate regional power targets. The edge control layer is equipped with a regional control unit, which receives the regional power target and performs multi-objective Pareto optimization using the improved NSGA-III algorithm with topological constraint factors to generate localized control commands. The terminal execution layer includes a photovoltaic inverter, energy storage equipment, and charging pile, which are used to execute the localized control commands and have four-quadrant enhanced control and communication interruption autonomy functions. The system is based on topology-driven architecture and constructs a three-level topology model of "distribution transformer-branch line-equipment". It characterizes the connection relationship between units by quantifying electrical distance and topological correlation, and constructs a dynamic adjacency matrix to reflect topology changes in real time.

2. The system according to claim 1, characterized in that, The three-level topology model of "distribution transformer-branch line-equipment" uses the following coding rules to achieve a unique mapping between unit identity and location: The unit coding format for the transformer layer is "T-voltage level-transformer area ID"; The branch layer unit encoding format is "L-belonging T unit-serial number"; The device layer unit encoding format is "D-belonging L unit-type-serial number".

3. The system according to claim 1, characterized in that, The formula for calculating the electrical distance is: in, and These are the weighting coefficients, and , For equipment to distribution transformer physical distance, For equipment to distribution transformer The branch impedance, and The maximum length and impedance within the transformer area; The formula for calculating the topological correlation degree is: in, This is the maximum electrical distance within the transformer substation area.

4. The system according to claim 1, characterized in that, The dynamic adjacency matrix The rules for assigning values ​​to its elements are as follows: When units i and j are directly connected and the switch is closed. ; When unit i and j are indirectly associated ; When unit i and j are not connected or the switch is open. ; The dynamic adjacency matrix is ​​updated every 5 minutes, and the update is based on the closed or open status of the smart switch.

5. The system according to claim 1, characterized in that, The improved NSGA-III algorithm used in the edge control layer introduces topology constraint factors including: electrical distance span within the cluster is not greater than 0.5, and / or the topology correlation degree of units within the cluster is not less than 0.

5.

6. The system according to claim 1, characterized in that, The four-quadrant enhanced control of the terminal execution layer includes the following operating conditions: Quadrant I: Energy storage absorbs active power and generates capacitive reactive power; Quadrant II: Stored energy releases active power and generates inductive reactive power; Quadrant III: Energy storage releases active power and absorbs reactive power; Quadrant IV: Energy storage absorbs active power and reactive power.

7. The system according to claim 1, characterized in that, The digital twin engine constructs a 1:1 topology-level digital twin model for pre-simulating the following three core scenarios: topology mutual assistance scenario, extreme failure scenario, and security protection scenario; and adopts a 5-minute long short-term memory network LSTM rolling prediction model to improve the accuracy of the pre-simulation.

8. The system according to claim 1, characterized in that, The system also includes a security protection mechanism, which includes: The island detection model with multiple features has the following island criterion: When the criterion value is ≥0.2, it is determined to be in an islanded state. This is the inverse correlation value between the equipment and the distribution transformer; The topology sequence isolation execution logic performs circuit breaker isolation in the order from the end device to the branch line head switch.

9. A photovoltaic cluster coordinated control method based on topology-driven and digital twin approaches, characterized in that, The method is applied to the system as described in any one of claims 1-8 and includes the following steps: A three-level topology model of "distribution transformer-branch line-equipment" is constructed, the electrical distance and topological correlation are calculated, and a dynamic adjacency matrix is ​​generated; Based on the power grid's operating status and topological characteristics, the clustering of photovoltaic clusters is dynamically triggered and executed; Perform pre-emptive optimization scheduling in the cloud, generate regional power targets, and distribute them to the edge layer; Multi-objective collaborative optimization under topological constraints is performed at the edge layer to generate localized control commands and send them to terminal devices; Power regulation is performed at the terminal device layer, and real-time data is fed back to the edge layer and the cloud to form a closed-loop control. Key scenarios are simulated using digital twin models, and control strategies are dynamically optimized based on topological deviation rates.

10. The method according to claim 9, characterized in that, The step of dynamically triggering and executing the clustering partitioning of the photovoltaic cluster specifically includes: Clustering is initiated when any of the following triggering conditions are met: the ratio of photovoltaic power output mutation value to rated power ≥ 20%, the distribution transformer load rate K ≥ 110%, and the average state of charge of energy storage. The original cluster association fails due to a load similarity of ≥85% between adjacent branches or a sudden change in topological correlation. The threshold of the triggering condition is dynamically adjusted by using a near-end strategy to optimize the PPO algorithm. Based on the collaborative potential score, the cells are divided into highly collaborative clusters using an improved K-means algorithm. The collaborative potential score integrates topological correlation, load similarity, and photovoltaic fluctuation consistency.

Images