Distributed new energy cluster adaptive power coordinated control method and related devices

Abstract

This invention discloses a distributed new energy cluster adaptive power coordinated control method and related devices, relating to the field of new energy technology. The method includes: dividing the distribution network into several clusters, each managed by an edge controller; the edge controller quantifies the cluster's aggregate adjustable capability and encapsulates it into a standardized information packet; monitoring the state variables of the cluster boundary nodes to determine if an over-limit event has occurred; if so, generating a start command and synchronizing; responding to the start command, initializing improved alternating direction multiplier method parameters through a cloud coordinator; the edge controller constructs a local aggregate adjustable capability model based on the standardized information packet, enters an iterative process, and outputs optimization results; each edge controller generates control commands based on the local optimization solution in the optimization results and executes them; after verifying the elimination of the over-limit event, resetting is performed. This invention solves the problems of high communication resource consumption, weak privacy protection, poor algorithm adaptability, and insufficient control response in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, and in particular to a method and device for adaptive power coordination control of distributed new energy clusters. Background Technology

[0002] The penetration rate of new energy sources, represented by distributed photovoltaic and wind power, in power distribution networks continues to rise. However, the randomness, intermittency, and volatility of new energy output pose significant challenges to the operation of traditional power distribution networks. Problems such as local voltage exceeding limits and line power flow overload are becoming increasingly prominent, seriously threatening the safety, stability, and power quality of the power grid. Against this backdrop, automatic power control technology has become a key means to achieve refined regulation of distributed energy resources and ensure reliable operation of the power grid.

[0003] Currently, automatic power control for distributed renewable energy mainly relies on two technical paths: First, centralized control schemes: This scheme relies on a central controller (such as a cloud platform or dispatch master station), requiring the collection of real-time data on all distributed energy sources, loads, and topology across the network. It then generates unified control commands through global optimization calculations and distributes them to each execution unit. While theoretically capable of achieving globally optimal control, this approach faces challenges in engineering applications, including high communication bandwidth requirements, heavy computational burden, and the risk of single points of failure. Furthermore, the centralized architecture requires all operators to share core operational data, which can easily lead to privacy breaches in a multi-entity market environment. Second, distributed control schemes: To overcome the limitations of centralized control, distributed control schemes decompose the global optimization problem into multiple sub-problems, which are solved in parallel by each local controller based on local information and limited communication. A typical implementation is a coordination mechanism based on the alternating direction multiplier method, where each node exchanges state information at fixed intervals and iteratively solves the problem. This scheme reduces the pressure and computational burden at single points to some extent, but it still has the following shortcomings: the fixed-period communication mechanism causes a lot of unnecessary communication overhead when the system is running smoothly, especially in rural or remote areas with limited communication resources; at the same time, since the power flow equation of the distribution network has typical non-convex and nonlinear characteristics, standard distributed optimization algorithms converge slowly or even fail to converge when dealing with this type of problem, making it difficult to meet the real-time and reliability requirements of practical engineering.

[0004] In summary, existing technologies suffer from high communication resource consumption, weak privacy protection, poor algorithm adaptability, and insufficient control response in scenarios with a high proportion of distributed new energy access. Therefore, there is an urgent need for a new collaborative control method that can balance communication efficiency, data privacy, algorithm convergence, and control response capability. Summary of the Invention

[0005] This invention provides a distributed new energy cluster adaptive power collaborative control method and related device to solve the problems of high communication resource consumption, weak privacy protection, poor algorithm adaptability and insufficient control response in the case of high proportion of distributed new energy access.

[0006] In view of this, the first aspect of the present invention provides a distributed new energy cluster adaptive power cooperative control method, the method comprising: The power distribution network is divided into several dynamic clusters, each of which is managed by an edge controller. The edge controller quantifies the aggregation and adjustability of the cluster and encapsulates it into a standardized information package, which serves as the only data unit for interaction with the cloud coordinator. Monitor the state variables of the cluster boundary nodes to determine if an out-of-limit event has occurred. When an out-of-limit event is detected, trigger the collaborative optimization process to generate a startup command and synchronize it to the cloud coordinator and all edge controllers. In response to the start command, the improved alternating direction multiplier method parameters are initialized through the cloud coordinator. Each edge controller constructs a local aggregate adjustable capability model based on the standardized information packet and enters the iterative process: In each iteration, each edge controller solves the local optimization sub-problem in parallel based on the local aggregate adjustable capability model to obtain the boundary state variables, so that the cloud coordinator calculates the consensus error based on each boundary state variable, and dynamically adjusts the penalty factor according to the consensus error to update the coordination variables before entering the next iteration, until the convergence condition is met and a globally consistent optimization result is output; Each edge controller generates device-level control instructions based on the local optimization solution in the optimization results. After performing multi-level security checks on the device-level control instructions, they are executed in batches according to the timing. After verifying that the over-limit event that triggered the start instruction has been eliminated, the trigger flag is reset and the silent monitoring state is restored.

[0007] Optionally, dividing the distribution network into several dynamic clusters includes: The electrical distance between any two nodes is calculated based on the node impedance matrix, and the communication cost coefficient is constructed based on communication delay, bandwidth and packet loss rate. The electrical distance and the communication cost coefficient are normalized and then weighted and fused to construct a comprehensive correlation weight matrix; The optimal partitioning scheme is generated based on the comprehensive association weight matrix using the spectral clustering algorithm, and the number of clusters and the physical boundaries of each cluster are determined according to the optimal partitioning scheme.

[0008] Optionally, dividing the distribution network into several dynamic clusters further includes: Establish repartitioning trigger conditions to automatically initiate repartitioning when the network topology changes or communication performance deteriorates.

[0009] Optionally, the quantization cluster's aggregation tunability is encapsulated into a standardized information packet, including: Based on the operating data of the distributed new energy sources within the cluster, the state estimation results are obtained by weighted least squares method. The state estimation results include the voltage amplitude and phase angle of each node within the cluster. Based on the state estimation results, the adjustment potential of each unit in the cluster is calculated. The adjustment potential includes the active power reduction space of the photovoltaic unit, the active power increase space and active power reduction space of the energy storage system, and the maximum reactive power absorption capacity and maximum reactive power generation capacity of all inverters with reactive power adjustment capabilities. The adjustment potential of each unit is classified and aggregated according to power type to form a cluster-level total active power upward adjustment space, total active power downward adjustment space, total reactive power absorption space and total reactive power output space to obtain four types of aggregated adjustable capabilities. The adjustment cost function corresponding to each type of aggregated adjustable capability is established with the adjustment power as the independent variable. The feasibility of the aggregated adjustable capability is verified by power flow calculation, and the verified aggregated adjustable capability is encapsulated into a standardized adjustable capability information package.

[0010] Optionally, monitoring the state variables of cluster boundary nodes to determine whether an out-of-limit event has occurred includes: The voltage amplitude and tie-line power of the cluster boundary nodes are monitored through dual data sources. The first data source is the edge controller that collects real-time data from the local measurement device at a preset first sampling frequency. The second data source is the cloud coordinator that receives synchronous phasor data from the wide-area measurement system. Based on the preset voltage over-limit judgment conditions and the preset power over-limit judgment conditions, determine whether an over-limit event has occurred; The voltage over-limit determination conditions include: the voltage amplitude at the boundary node exceeds the preset voltage operating range, or the voltage change rate exceeds the preset voltage change rate threshold. The power limit determination conditions include: the power of the tie line exceeds the preset transmission power limit, or the power change rate exceeds the preset power change rate threshold.

[0011] Optionally, the step of triggering a collaborative optimization process to generate a startup command when an out-of-limit event is detected, and synchronizing it to the cloud coordinator and all edge controllers, includes: When an out-of-limit event is detected, the trigger unit generates a start command. The start command is encapsulated into a fixed-length data frame, which includes the event type, severity, timestamp, and trigger unit identifier. The triggering unit sends the start command to the receiver through a preset priority communication channel. When the receiver receives the start command, it verifies the integrity of the frame and sends a confirmation response to the triggering unit. The receiver includes all edge controllers and cloud coordinators participating in the collaborative optimization. After receiving the confirmation response, the triggering unit replies with a ready signal to the receiver to complete the three-way handshake; if the three-way handshake is not completed within a preset time, the triggering unit initiates command retransmission or switches to the backup communication route. After the three-way handshake is completed, all receivers and triggering units enter the optimization state in unison according to the timestamp in the start command.

[0012] Optionally, the initialization of the improved alternating direction multiplier method parameters includes: Initialize the Lagrange multiplier vectors to zero vectors; The reference values ​​for the initial boundary variables are the measured values ​​at the trigger time; The penalty parameters are initialized to the preset initial value of the penalty factor; The maximum number of iterations is set to a preset maximum number of iterations threshold, the absolute convergence tolerance is set to a preset absolute convergence tolerance value, and the relative convergence tolerance is set to a preset relative convergence tolerance value.

[0013] Optionally, each edge controller obtains boundary state variables by solving local optimization sub-problems in parallel based on the local aggregate adjustable capability model, including: Each edge controller establishes a local optimization sub-problem for this iteration based on the current penalty factor and the local aggregation adjustable capability model; The local optimization subproblem aims to minimize the cluster operating cost, which is determined based on the cost function in the local aggregate adjustable capability model. The constraints of the local optimization subproblem include the upper and lower limits of power, ramp rate limit, voltage safety boundary, and power flow balance equation of each distributed unit in the cluster, wherein the power flow balance equation of the power flow is a non-convex constraint. Each edge controller uses the LinDistFlow linearization model to convexize the non-convex constraints, transforming the local optimization subproblem into a convex optimization problem. Each edge controller solves the convex optimization problem to obtain the active power regulation and reactive power regulation of each controllable distributed unit in the cluster, and calculates the boundary state variables of the cluster based on the active power regulation and reactive power regulation.

[0014] Optionally, the cloud coordinator calculates the consensus error based on each of the boundary state variables, and dynamically adjusts the penalty factor according to the consensus error, including: The cloud coordinator calculates the original residual vector and dual residual vector for the current iteration based on the boundary state variables uploaded by each edge controller. The original residual vector and the dual residual vector together constitute the quantization of the current consensus error. The cloud coordinator compares the 2-norm of the original residual vector with the 2-norm of the dual residual vector, and dynamically adjusts the penalty factor based on the comparison result.

[0015] Optionally, the step of dynamically adjusting the penalty factor based on the comparison result includes: When the 2-norm of the original residual vector is greater than the 2-norm of the dual residual vector, the penalty factor is amplified by an increasing factor, but not exceeding the preset upper limit. When the 2-norm of the dual residual vector is greater than the 2-norm of the original residual vector, the penalty factor is reduced by a decreasing factor, but not lower than the preset lower limit. The penalty factor remains unchanged when the 2-norm of the original residual vector and the 2-norm of the dual residual vector are in relative equilibrium.

[0016] Optionally, the step of generating device-level control instructions based on the local optimized solution in the optimization results, performing multi-level security checks on the device-level control instructions, executing them in batches according to timing, and resetting the trigger flag after verifying that the over-limit event that triggered the start instruction has been eliminated, and restoring the silent monitoring state, includes: The power regulation contained in the local optimization solution in the optimization results is converted into device-level control commands. The device-level control commands include the active power load reduction ratio and reactive power setpoint of the photovoltaic inverter, and the charging and discharging power and duration of the energy storage system. The device-level control commands are subjected to security verification, which includes verification of the device's safe operating range, verification of fast power flow prediction, and verification of control coordination. Based on the response speed of each unit, a batch execution strategy is formulated, and the device-level control commands are sequentially sent to the corresponding units through an encrypted confirmation mechanism in accordance with the batch execution strategy, so that the corresponding units execute the device-level control commands. The execution status of each unit is monitored in real time. When the actual output deviates from the command, the equipment-level control command of other units in the same batch or subsequent batches is adjusted to compensate for the deviation. After all device-level control commands have been executed, verify whether the boundary state variables have returned to a safe range. If so, record the data, reset the trigger flag, and restore the silent monitoring state.

[0017] A second aspect of the present invention provides a distributed new energy cluster adaptive power coordination control system, the system comprising: The first control submodule is used to divide the distribution network into several dynamic clusters, wherein each cluster is managed by an edge controller. The edge controller quantifies the aggregation and adjustability of the cluster and encapsulates it into a standardized information package, which serves as the only data unit for interaction with the cloud coordinator. The second control submodule is used to monitor the state variables of the cluster boundary nodes to determine whether an over-limit event has occurred. When an over-limit event is detected, a collaborative optimization process is triggered to generate a startup command and synchronize it to the cloud coordinator and all edge controllers. The third control submodule is used to respond to the start command, initialize the improved alternating direction multiplier method parameters through the cloud coordinator, and each edge controller constructs a local aggregate adjustable capability model based on the standardized information packet and enters the iterative process: In each iteration, each edge controller solves the local optimization sub-problem in parallel based on the local aggregate adjustable capability model to obtain the boundary state variables, so that the cloud coordinator calculates the consensus error based on each boundary state variable, and dynamically adjusts the penalty factor according to the consensus error to update the coordination variables before entering the next iteration, until the convergence condition is met and a globally consistent optimization result is output; The fourth control submodule is used to generate device-level control instructions through each edge controller based on the local optimization solution in the optimization results, perform multi-level security verification on the device-level control instructions and execute them in batches in sequence, and reset the trigger flag after verifying that the over-limit event that triggered the start instruction has been eliminated, and restore the silent monitoring state.

[0018] A third aspect of the present invention provides a distributed new energy cluster adaptive power coordination control device, the device comprising a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the steps of the distributed new energy cluster adaptive power cooperative control method as described in the first aspect above, according to the instructions in the program code.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium for storing program code for executing the distributed new energy cluster adaptive power cooperative control method described in the first aspect above.

[0020] As can be seen from the above technical solutions, the present invention has the following advantages: First, regarding communication efficiency, this invention fundamentally changes the fixed-cycle communication mode in traditional distributed control through an event-driven mechanism. The system remains silent during normal operation, only initiating communication and optimization processes when genuine adjustments are needed, such as detecting voltage or power exceeding limits. This on-demand response mechanism reduces the consumption of most unnecessary communication resources, making it particularly suitable for rural and remote power distribution networks with weak communication infrastructure, while simultaneously enabling rapid detection and response to grid disturbances.

[0021] Secondly, regarding privacy protection, the information interaction paradigm established by this invention based on "adjustable capability" aggregation fundamentally solves the problem of commercial privacy leakage under the participation of multiple investment entities. Each edge cluster only needs to report its cluster-level aggregation adjustment capability and cost coefficient to the cloud coordinator, without disclosing core data such as its internal topology, load, and unit output plans. This design principle of "capability visible, details hidden" ensures the overall coordination and optimization effect while strictly protecting the commercial secrets of each operating entity, meeting the development needs of diversified participation in the power market environment.

[0022] Third, regarding algorithm performance, this invention makes key improvements to the standard ADMM algorithm to address the non-convex characteristics of power flow in distribution networks. By employing linearization approximation or convex relaxation techniques in the local optimization problem, the non-convex problem is transformed into a convex problem that can be solved efficiently. Simultaneously, an adaptive penalty factor strategy is introduced to dynamically adjust algorithm parameters based on boundary consensus errors. These two improvements significantly enhance the convergence and computational efficiency of the algorithm in real distribution network environments, showing a significant improvement in convergence speed compared to the standard ADMM algorithm. Furthermore, it exhibits good convergence stability under different operating scenarios, effectively enhancing the engineering practical value of the method.

[0023] Fourth, regarding system practicality, this invention constructs a complete intelligent closed-loop control system. From state monitoring and event triggering to optimization calculation and instruction execution, a strict closed-loop control process is formed. The system possesses multi-level security verification, batch-based sequential execution, and adaptive learning capabilities, ensuring the safe and reliable execution of control instructions and continuously optimizing its own parameters during operation. This highly structured design enables the system to adapt to complex and ever-changing actual operating environments, providing reliable technical support for the safe and stable operation of distribution networks with a high proportion of distributed renewable energy access.

[0024] In summary, the adaptive power coordination control method for distributed new energy clusters provided by this invention addresses the shortcomings of existing technologies in scenarios with high proportions of distributed new energy access, which suffer from high communication resource consumption, weak privacy protection capabilities, poor algorithm adaptability, and insufficient control response. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a distributed new energy cluster adaptive power cooperative control method provided in an embodiment of the present invention; Figure 2 A general technical framework diagram provided for embodiments of the present invention; Figure 3 The dynamic cluster partitioning algorithm flow provided in this embodiment of the invention; Figure 4 This is a schematic diagram of adjustable capability quantization modeling provided in an embodiment of the present invention; Figure 5 A flowchart illustrating the event triggering mechanism provided in this embodiment of the invention; Figure 6 The improved ADMM algorithm iterative loop provided in the embodiments of the present invention; Figure 7 This is a flowchart of control instruction generation and execution provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of a distributed new energy cluster adaptive power collaborative control system provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0028] Please see Figure 1 and Figure 2 The present invention provides a distributed new energy cluster adaptive power cooperative control method, comprising: Step 101: Divide the power distribution network into several dynamic clusters, where each cluster is managed by an edge controller. The edge controller quantifies the aggregation and adjustability of the cluster and encapsulates it into a standardized information package, which serves as the only data unit for interaction with the cloud coordinator. It should be noted that, firstly, the distribution network is divided into several dynamic clusters based on electrical coupling. This is understandable, as efficient distributed collaboration requires dividing the vast distribution network into several easily manageable collaborative units. This sub-step defines the physical boundaries of the subsequent "edge clusters," ensuring tight electrical connections within the clusters. This makes the local optimization problems within each cluster relatively independent, reducing mutual interference between clusters. Each cluster is managed by an edge controller. The aforementioned distribution network partitioning directly defines the physical jurisdiction of each edge controller, ensuring that each dynamically partitioned cluster is managed by an independent edge controller, thus laying the organizational foundation for all subsequent localized operations. Next, the aggregation adjustability of the clusters is standardized and quantified through the edge controller. This invention defines "cluster aggregation adjustability" as the core unit of system information interaction and transforms it into executable control parameters through a standardized modeling and quantification process. In summary, step 101 lays the physical foundation through "dynamic cluster partitioning," achieves information abstraction through "adjustable capability quantification," and finally constructs an infrastructure that ensures both global coordination and achieves the dual goals of efficient communication and data privacy through the interaction paradigm of "capability-detail separation," as detailed in the corresponding implementation examples below.

[0029] Step 102: Monitor the state variables of the cluster boundary nodes to determine whether an over-limit event has occurred. When an over-limit event is detected, trigger the collaborative optimization process to generate a start command and synchronize it to the cloud coordinator and all edge controllers. It should be noted that this step designs an event-driven collaborative triggering strategy. Based on the system architecture construction and information interaction mechanism definition completed in step 101, a corresponding dynamic response mechanism needs to be established. This step, based on the cluster boundary division and status monitoring point configuration established in the system initialization phase, constructs an intelligent switching mechanism from normal monitoring to optimization calculation. The event triggering mechanism established in this step directly relies on the physical boundary and monitoring system defined in step 101. The triggering signal it generates will be the sole source of instructions and control timing reference for initiating the subsequent distributed collaborative optimization solution phase. It can be understood that this step, through the event triggering mechanism in the distributed new energy cluster collaborative control, focuses on achieving rapid response to abnormal situations by monitoring the status variables of cluster boundary nodes. The system continuously monitors key status variables such as voltage amplitude and tie-line power of boundary nodes. When voltage exceeds limits (e.g., exceeding the operating range or rate of change exceeds limits) or power exceeds limits (e.g., exceeding transmission limits or rate of change exceeds limits), the collaborative optimization process is immediately triggered and a start command is generated. This ensures that the problem is detected and control is initiated in a timely manner, and the start command is synchronized to the cloud coordinator and all edge controllers, achieving efficient startup of distributed collaborative control. See the corresponding implementation examples below for details.

[0030] Step 103: In response to the start command, the improved alternating direction multiplier method parameters are initialized through the cloud coordinator. Each edge controller constructs a local aggregate adjustable capability model based on the standardized information packet and enters the iterative process: In each iteration, each edge controller solves the local optimization sub-problem in parallel based on the local aggregate adjustable capability model to obtain the boundary state variables, so that the cloud coordinator calculates the consensus error based on each boundary state variable, and dynamically adjusts the penalty factor according to the consensus error to update the coordination variables before entering the next iteration, until the convergence condition is met and a globally consistent optimization result is output; It should be noted that this step is an improved distributed collaborative optimization solution, which is a logical necessity and paradigm summary of the previous two steps 101 and 102. It establishes the basic principles of information interaction during system operation, ensuring privacy protection and efficient communication. Its execution flow is as follows: First, after the limit event is triggered, the system initiates an improved alternating direction multiplier method (ADMM) iterative process, achieving coordinated optimization of cluster decisions through four closely connected computational stages. These four stages include: parameter initialization, parallel local optimization, boundary coordination, and convergence judgment, forming a closed-loop iterative structure. Understandably, after responding to the start command, the cloud coordinator first initializes the improved alternating direction multiplier method parameters (such as Lagrange multipliers, boundary variable reference values, etc.), and each edge controller builds a local aggregate adjustable capability model based on the standardized information packet and starts the iterative process. In each iteration, the edge controller solves the local optimization subproblem with the goal of minimizing operating costs (considering constraints such as power upper and lower limits and power flow balance) in parallel, calculates the boundary state variables and uploads them to the cloud. The cloud calculates the consensus error composed of the original residual and the dual residual based on each boundary state variable, dynamically adjusts the penalty factor to update the coordination variables, and iterates until the convergence condition is met (such as the residual being less than the tolerance or reaching the maximum number of iterations), and finally outputs a globally consistent optimization result, as shown in the corresponding implementation examples below.

[0031] Step 104: Each edge controller generates device-level control instructions based on the local optimization solution in the optimization results. After performing multi-level security checks on the device-level control instructions, they are executed in batches according to the timing. After verifying that the over-limit event that triggered the start instruction has been eliminated, the trigger flag is reset and the silent monitoring state is restored.

[0032] It should be noted that this step is the value loop point of the entire method, as it transforms the "paper" optimization scheme generated in step 103 into physical actions in the actual power grid. Understandably, each edge controller generates device-level control commands (such as the active power load reduction ratio of the photovoltaic inverter, the charging and discharging power of the energy storage system, etc.) based on the local optimization solution in the optimization results. After multi-level safety verification (equipment safe operating range, fast power flow prediction, and control coordination verification), these commands are executed in batches according to the device response speed. During execution, deviations are monitored in real time and dynamically compensated. After all commands are executed, it is verified whether the boundary state variables have returned to the safe range. After confirming that the over-limit event has been eliminated, the trigger flag is reset, and the silent monitoring state is restored, forming a complete control closed loop. See the corresponding implementation examples below for details.

[0033] This invention provides a distributed new energy cluster adaptive power collaborative control method, which achieves a complete closed loop from system initialization, intelligent triggering, collaborative optimization to control execution through four logically rigorous and interconnected steps. Its internal logic is as follows: the system initialization and information aggregation stage establishes the physical foundation and information interaction specifications for distributed collaboration, defining the system architecture and data interfaces for all subsequent operations; the event-driven collaborative triggering stage continuously monitors boundary state variables and initiates regulation when limits are exceeded, realizing on-demand switching from normal monitoring to optimized calculation; the improved distributed collaborative optimization solution stage, based on the system architecture and trigger signals provided by the preceding stages, generates globally optimized control decisions through an improved algorithm; and the control command generation and execution stage transforms the optimized decisions into specific equipment commands, completing the closed loop from calculation to physical control. These four stages are sequentially connected and cyclically interact, forming a complete automatic power control process.

[0034] In one embodiment, see Figure 3 In step 101, the distribution network is divided into several dynamic clusters, including: Step 1011: Calculate the electrical distance between any two nodes based on the node impedance matrix, and construct the communication cost coefficient based on communication delay, bandwidth and packet loss rate; Specifically, firstly, the impedance matrix of any two nodes is calculated. electrical distance between : ; The node admittance matrix is ​​formed based on the distribution network topology and line parameters, and the node impedance matrix is ​​obtained by inversion operation, thereby obtaining the self impedance and mutual impedance of the required node. , It is self-impedance. This represents mutual impedance. The smaller the electrical distance, the stronger the voltage interaction between nodes, and the higher the degree of coupling.

[0035] Then, a communication quality assessment is conducted, including comprehensively considering factors such as communication latency, bandwidth, and reliability (e.g., packet loss rate), and a communication cost coefficient is defined. The larger the value of this coefficient, the worse the communication conditions between nodes.

[0036] Communication performance directly affects the real-time performance and reliability of distributed control. For each communication link... Define the communication cost coefficient : ; In the formula: For link Communication delay (ms); The maximum allowable communication delay (ms) of the system; For link Available bandwidth (Mbps); To control the minimum bandwidth required (Mbps); For link The packet loss rate; This represents the maximum packet loss rate allowed by the system. The weighting coefficients for latency, bandwidth, and packet loss rate are respectively, satisfying... ; The larger the value, the worse the communication quality of the link. When partitioning, it should be avoided as the main communication link within the cluster.

[0037] Step 1012: Normalize the electrical distance and the communication cost coefficient, then weight and fuse them to construct a comprehensive correlation weight matrix; Specifically, a comprehensive correlation weight matrix is ​​established between nodes. : ; in and These are the normalization functions for electrical distance and communication cost, respectively. and These are the weighting coefficients. The larger the value, the more suitable the nodes are to be assigned to the same cluster.

[0038] Step 1013: Use the spectral clustering algorithm to generate the optimal partitioning scheme based on the comprehensive association weight matrix, and determine the number of clusters and the physical boundaries of each cluster according to the optimal partitioning scheme.

[0039] Specifically, firstly, based on the constructed comprehensive weight matrix... Expanding the matrix, each element quantifies the combined correlation strength between the corresponding nodes in terms of both electrical coupling and communication quality. Subsequently, the Laplace matrix corresponding to this weight matrix is ​​calculated. This is the difference between the degree matrix and the weight matrix. Next, eigenvalue decomposition is performed on this Laplacian matrix, and the first... The eigenvectors corresponding to the smallest non-zero eigenvalues, where the number of clusters is... The eigenvectors can be automatically determined using metrics such as eigenvalue gaps. Then, these eigenvectors are combined column-wise to form a feature space matrix, and their row vectors are partitioned using the K-means clustering algorithm. Finally, based on this clustering result, the original network nodes are assigned to different groups, thus forming multiple tightly interconnected and reliably communicating collaborative clusters. This process achieves an optimal balance between electrical performance and communication efficiency, and lays a direct physical architectural foundation for subsequent edge controller deployment and distributed collaborative optimization.

[0040] Step 1014: Set repartition trigger conditions to automatically initiate repartitioning when the network topology changes or communication performance deteriorates.

[0041] Specifically, a repartitioning trigger condition is established. When the network topology changes significantly or communication performance deteriorates (e.g., topology change monitoring: detecting line connection / disconnection, distributed power supply start / stop, etc. through smart meters, PMU (phasor measurement unit), or circuit breaker status signals; communication performance deterioration monitoring: real-time monitoring of communication latency, packet loss rate, and bandwidth; if a threshold is exceeded, repartitioning is triggered), the system automatically initiates repartitioning to maintain the optimal cluster structure. The repartitioning method is as follows: First, the partitioning result directly defines the physical jurisdiction of each edge controller, ensuring that each dynamically partitioned cluster is managed by an independent edge controller, thus laying the organizational foundation for all subsequent localized operations. Second, on top of this management architecture, each edge controller independently performs standardized modeling and quantification of "adjustable capabilities" within its own cluster based on locally collected data. This process is completed entirely within the cluster boundary, achieving local information aggregation and privacy protection. Finally, in the optimization phase that requires global coordination, namely the improved distributed collaborative optimization solution in step 103, each cluster solves the optimization sub-problems constrained by local "adjustability" in parallel according to the previously defined boundaries. The partitioning results thus constitute the fundamental basis of distributed parallel computing and the structural framework for coordinated optimization.

[0042] In one embodiment, in step 101, please refer to Figure 4 The aggregation and tunability capabilities of the quantified cluster are encapsulated into standardized information packages, including: Step 1015: Based on the operating data of the distributed new energy sources within the cluster, state estimation is performed using the weighted least squares method to obtain the state estimation result, which includes the voltage amplitude and phase angle of each node within the cluster. Specifically, the edge controller first collects operational data from all distributed renewable energy units within its managed cluster, including real-time output of photovoltaic inverters, state of charge and charging / discharging power of energy storage systems, and wind turbine operating status. Simultaneously, it collects power consumption data from load nodes within the cluster and network topology connections. Based on this real-time measurement data, the edge controller performs state estimation calculations to obtain the voltage amplitude and phase angle of each node within the cluster, establishing an accurate perception of the current operating status.

[0043] The edge controller uses weighted least squares to perform state estimation calculations to obtain the voltage amplitude and phase angle of each node within the cluster, thus establishing an accurate perception of the current operating state. The specific implementation process is as follows: First, real-time measurements within the cluster are collected, including node injected power, voltage amplitude, and branch current amplitude. Second, measurement equations are established. ,in For measurement vectors, This represents a nonlinear functional relationship between the measured quantity and the state variable. Let this be the measurement error vector. Then, construct the objective function. ,in The measurement error covariance matrix is ​​typically a diagonal matrix, where the diagonal elements represent the accuracy weights of each measurement device. Finally, the solution is obtained to make... Minimize state variables ,in A vector containing the voltage magnitudes of all nodes to be estimated. With voltage phase vector This process effectively eliminates noise and bad data by optimizing and fitting redundant measurements, thereby outputting high-precision system state estimation results.

[0044] The accurate voltage amplitude and phase angle information obtained from the state estimation are directly used in several key subsequent steps of this invention: First, the result is the basis for calculating the adjustment potential of each distributed unit in the cluster. Only with accurate node voltages can the active and reactive power adjustable space of photovoltaic, energy storage and other equipment be reliably assessed. Second, after generating the cluster aggregation adjustable capability, its feasibility needs to be verified through power flow calculation. Accurate initial state estimation is the prerequisite for the reliable convergence of this verification calculation. Finally, the boundary node voltage and phase angle output by the state estimation are used as real-time and reliable boundary state quantities to monitor whether voltage or power exceedance occurs, thereby determining whether to initiate global collaborative optimization.

[0045] Step 1016: Calculate the adjustment potential of each unit in the cluster based on the state estimation results. The adjustment potential includes the active power reduction space of the photovoltaic unit, the active power increase space and active power reduction space of the energy storage system, and the maximum reactive power absorption capacity and maximum reactive power generation capacity of all inverters with reactive power adjustment capabilities. It should be noted that, based on accurate perception of the operating status, the edge controller evaluates the adjustment potential of each unit within the cluster through parallel computing. Specifically, for photovoltaic units, their active power reduction potential equals the difference between the current actual output and the minimum technical output, where the minimum output needs to comprehensively consider real-time illumination conditions and the inverter's minimum operating capacity limit. For energy storage systems, their active power increase potential equals the current remaining dischargeable power, which is determined by the difference between the maximum discharge power of the energy storage converter and the current discharge power, taking into account the state-of-charge upper limit constraint. Conversely, their active power reduction potential equals the current remaining rechargeable power, which is determined by the difference between the maximum charging power of the converter and the current charging power, taking into account the state-of-charge upper limit constraint. For all inverters with reactive power regulation capabilities (including photovoltaic and energy storage converters), their maximum reactive power absorption capacity is calculated based on their current operating point and apparent capacity curve. With maximum reactive power generation capacity This process requires real-time verification of the safe operating boundaries of various devices, such as power, voltage, and current, to ensure that the calculated adjustment potential is strictly within the equipment's safety constraints.

[0046] Step 1017: Classify and aggregate the adjustment potential of each unit according to power type to form a cluster-level total active power upward adjustment space, total active power downward adjustment space, total reactive power absorption space and total reactive power output space to obtain four types of aggregated adjustable capabilities, and establish adjustment cost functions corresponding to each type of aggregated adjustable capability with adjustment power as the independent variable. Specifically, firstly, the regulation potential of each unit is categorized and aggregated according to power type to form four core regulation capabilities at the cluster level: total active power adjustment space. Total active power reduction space Total reactive power absorption space Space for total reactive power output .in, =∑ Energy storage capacity + other adjustable resource capacity; =∑ Photovoltaic downgrade capacity + Energy storage downgrade capacity; =∑Inverter reactive power absorption capacity; =∑Reactive power output capacity of the inverter.

[0047] Then, for each type of regulation capability, a corresponding regulation cost function is established. Active power regulation costs typically consider factors such as fuel cost and opportunity cost, while reactive power regulation costs mainly consider equipment losses and opportunity cost. These cost functions can be modeled as linear or quadratic functions of the regulated power (specifically, in the local optimization problem of step 103, the cost function...). As part of the objective function, used to minimize the adjustment cost): ;in The cost coefficient is obtained by weighted aggregation based on the actual situation of each unit in the cluster (for example, by weighted aggregation of the cost coefficients of each unit in the cluster, and the weights can be set according to factors such as equipment type, adjustment efficiency, and operating costs).

[0048] Step 1018: Verify the feasibility of the aggregated adjustable capability through power flow calculation, and encapsulate the verified aggregated adjustable capability into a standardized adjustable capability information package.

[0049] Specifically, after completing capacity aggregation and cost modeling, the edge controller verifies the feasibility of the provided regulation capabilities through power flow calculations (e.g., performing power flow calculations under a given regulation capability to check for constraints such as voltage overruns or line overloads. If no violations occur, it is considered feasible), ensuring that providing the nominal regulation capability does not trigger grid overruns. Finally, the four types of regulation capabilities and their corresponding cost coefficients are encapsulated into a standardized "adjustable capability information package" (e.g., packaging Pup, Pdown, Qabs, Qinj, and their cost coefficients into a structured data packet in JSON format). This information package serves as the sole interactive data between the edge cluster and the cloud coordinator, providing unified boundary conditions for subsequent collaborative optimization. Understandably, this standardized modeling and quantification process ensures that the reported regulation capabilities accurately reflect the cluster's true control potential while fully considering economic and safety requirements, laying a reliable data foundation for network-wide collaborative optimization.

[0050] Furthermore, in one embodiment, the standardized information packet of step 1018 is used as the sole data unit for interaction with the cloud coordinator.

[0051] Specifically, after completing the quantitative modeling of the cluster's adjustable capabilities, the process enters the information exchange phase. The edge controller uses the standardized and aggregated "adjustable capability" information package as the sole data exchange unit, reporting it to the cloud coordinator periodically or via event triggering.

[0052] The information packet's content and format are as follows: It employs a standardized data structure containing four core dimensions: active power upscaling space, active power downscaling space, reactive power absorption space, reactive power generation space, and corresponding adjustment cost coefficients. Each dimension is normalized to ensure data comparability across different clusters. The information packet uses a lightweight encapsulation format, significantly reducing communication load.

[0053] Information processing in the cloud: After receiving adjustable capability information packets from each edge controller, the cloud coordinator constructs a global optimization model based on this aggregated information. The cloud does not need to, and does not receive, any specific information about the internal workings of the cluster, including detailed data such as network topology, real-time output of distributed units, and load curves. This "black box" interaction mode allows the cloud to focus only on the control boundary conditions of each cluster, greatly simplifying the complexity of global coordination.

[0054] Data protection at the edge: At the edge, all core data involving business privacy and operational secrets are strictly kept locally. This data includes sensitive information such as the real-time power generation of photovoltaic power plants, user electricity consumption characteristics, precise state of charge of energy storage systems, and internal network topology connections. By transforming raw data into standardized capability indicators, a privacy protection barrier is established at the source of information transmission.

[0055] Two-way information flow design: This interaction paradigm adopts a two-way asymmetric design, meaning that the edge transmits highly aggregated adjustable capability information to the cloud, while the cloud sends globally coordinated boundary target values ​​to the edge. This design ensures both the effectiveness of global optimization and the security of data privacy for all participants, providing a feasible technical path for distributed collaborative control of new energy sources involving multiple stakeholders.

[0056] In summary, step 101 lays the physical foundation through "dynamic cluster partitioning," achieves information abstraction through "adjustable capability quantification," and ultimately constructs an infrastructure that guarantees both global coordination and achieves the dual goals of efficient communication and data privacy through the interaction paradigm of "capability-detail separation." This is fundamentally different from the naked data exchange in existing technologies and is the primary step in achieving innovation in this method.

[0057] In one embodiment, in step 102, please refer to Figure 5 Monitor the state variables of cluster boundary nodes to determine if an out-of-bounds event has occurred, including: Step 1021: Monitor the voltage amplitude and tie-line power of the cluster boundary node through dual data sources, wherein: the first data source is the edge controller collecting real-time data from the local measurement device at a preset first sampling frequency, and the second data source is the cloud coordinator receiving synchronous phasor data from the wide-area measurement system; Specifically, by constructing a complete condition monitoring network, the monitored objects directly correspond to the cluster boundary connection points and interconnection lines defined in the initialization phase. The condition monitoring network employs a dual data source protection mechanism: firstly, real-time data from local measurement devices is collected via edge controllers at a sampling frequency of no less than 1kHz; secondly, the cloud coordinator synchronously receives phasor data from the wide-area measurement system at a sampling frequency of 50Hz. The real-time data collected by the edge controllers mainly includes the state variables of the power grid and equipment operating parameters, specifically covering voltage amplitude and phase angle, active and reactive power. In practice, these data are reflected in aspects such as the actual output of photovoltaic units. The state of charge (SOC) of the energy storage system and the active power of the load nodes. Specific measurements are taken. The monitoring data then undergoes a three-stage processing flow: first, digital filtering and outlier removal; second, state estimation to improve data accuracy; and third, a sliding window algorithm to smooth instantaneous fluctuations. All data is accompanied by timestamps and quality indicators, providing a reliable basis for triggering judgments.

[0058] Furthermore, to improve the quality of monitoring data relied upon for event triggering judgment, the following specific technical means are used to enhance the accuracy of state estimation: First, a redundant design is adopted in the measurement configuration to ensure that key state quantities are monitored by multiple independent measuring devices or sensors based on different principles. This significantly increases the number of measurement equations compared to the number of state variables to be estimated, providing a data foundation for eliminating random errors. Second, in the estimation model, weight coefficients corresponding to the accuracy of the measuring equipment are assigned to each measurement value, forming the objective function of the weighted least squares method. Data from synchronous phasor measurement units are given a larger weight, while low-precision or easily disturbed data are given a smaller weight, thereby mathematically suppressing the influence of noise. Third, during the solution iteration process, by calculating standardized residuals and setting thresholds, bad data that significantly deviates from physical laws is automatically identified and eliminated. Finally, the estimated value obtained after algorithm convergence has a statistically minimized mean square error compared to the actual system state, thus outputting highly reliable state quantities such as voltage and power, providing a reliable basis for subsequent triggering judgment.

[0059] Step 1022: Determine whether an over-limit event has occurred based on the preset voltage over-limit judgment conditions and the preset power over-limit judgment conditions; The voltage over-limit determination conditions include: the voltage amplitude at the boundary node exceeds the preset voltage operating range, or the voltage change rate exceeds the preset voltage change rate threshold. The power limit determination conditions include: the power of the tie line exceeds the preset transmission power limit, or the power change rate exceeds the preset power change rate threshold.

[0060] Specifically, the triggering judgment is based on a dynamic threshold system, comprising two dimensions: absolute threshold and rate of change threshold. The absolute threshold is set according to power grid safety standards, including hard constraints such as upper and lower voltage operating limits and line transmission capacity; the rate of change threshold is set to address the fluctuation characteristics of new energy sources, used to capture the system's rapid dynamic processes. The triggering judgment adopts a fully distributed architecture, with each edge controller and cloud coordinator having equal judgment authority. The judgment logic follows these principles: Voltage triggering: Boundary node voltage exceeds the range of 0.95-1.05 pu, or the rate of change exceeds 0.1 pu / s; Power trigger: The tie line power exceeds the transmission limit, or the rate of change exceeds 10% / s of the rated value.

[0061] In one embodiment, in step 102, please refer to Figure 5 When an out-of-limit event is detected, a collaborative optimization process is triggered to generate a startup command, which is then synchronized to the cloud coordinator and all edge controllers, including: Step 1023: When an out-of-limit event is detected, the triggering unit generates a start command. The start command is encapsulated into a fixed-length data frame, which includes the event type, severity, timestamp, and triggering unit identifier. Step 1024: The triggering unit sends the start command to the receiver through a preset priority communication channel. When the receiver receives the start command, it verifies the integrity of the frame and replies with an acknowledgment response to the triggering unit. The receiver includes all edge controllers and cloud coordinators participating in the collaborative optimization. Step 1025: After receiving the confirmation response, the triggering unit replies with a ready signal to the receiver to complete the three-way handshake; if the three-way handshake is not completed within a preset time, the triggering unit initiates command retransmission or switches to the backup communication route. Step 1026: After the three-way handshake is completed, all receivers and triggering units enter the optimization state in accordance with the timestamp in the start command.

[0062] It should be noted that, Figure 5 The three startup modes are differentiated response strategies based on event priority sorting, with the core difference being response latency and scheduling order: Immediate startup: corresponds to the highest priority event (such as...) Figure 5 In the event of a "voltage limit violation" (e.g., the system unconditionally interrupts the current task with the shortest possible delay and directly enters the optimization process), the system will immediately proceed to the optimization process. Fast startup: This applies to the next highest priority event (e.g., [missing information]). Figure 5The "rate of change exceeding the limit" strategy initiates optimization as soon as resources allow or a higher-priority task is completed, with a response latency higher than "immediate start" but still relatively short. Sequential start: Applicable to non-urgent batch events (such as "power exceeding the limit" in the figure), the system sorts multiple events according to preset rules (such as severity of exceeding the limit, occurrence sequence) and processes them sequentially, avoiding instantaneous resource overload and achieving orderly scheduling. These three strategies together constitute a hierarchical coordination mechanism for event response, ensuring priority handling of critical issues while also balancing the overall communication and computational load of the system.

[0063] Specifically, once the triggering conditions are met, i.e., when an over-limit event is confirmed, the collaborative optimization process is initiated, implemented as follows: When an over-limit event is detected, the triggering unit, i.e., the edge controller or cloud coordinator, immediately generates a structured start command. The core data fields of this command include: an event type code, such as 01 indicating voltage over-limit and 02 indicating power overload; a quantified value of severity, such as the specific percentage of voltage over-limit or the megawatts of power over-limit; a timestamp of the event occurrence accurate to milliseconds; and a triggering unit identifier. This command is encapsulated as a fixed-length data frame and transmitted through a virtual private network with quality of service guarantees. This priority communication channel ensures that the transmission latency and packet loss rate of the control command are below a preset threshold by configuring differential service code point marking or establishing a dedicated virtual link. The transmission process employs a three-way handshake confirmation mechanism: the sender first sends a start command, the receiver verifies the frame integrity and replies with an acknowledgment, and the sender replies with a final ready signal upon receiving the acknowledgment, thereby establishing a synchronization state between the two communicating parties. If the handshake is not completed within a set time, the system will resend the start command or switch to a backup route. After confirming synchronization, all units participating in the collaborative optimization will uniformly enter the optimization calculation cycle according to the timestamp in the instruction, thereby achieving precise synchronization of coordinated actions across the entire network.

[0064] Furthermore, in one embodiment, the present invention allocates dedicated computing resources for the optimization process and establishes an independent timeout protection mechanism. While optimization is starting, the normal operation of the monitoring system is maintained, achieving parallel processing of control and monitoring. The specific process is as follows: When the collaborative optimization process is triggered by a limit-crossing event, the system's underlying scheduler immediately allocates dedicated computing resources for this optimization task. This includes allocating a dedicated CPU core, reserving a specific amount of memory space, and allocating a dedicated communication buffer. This ensures that the optimization iteration is not interfered with by system data acquisition or human-computer interaction, guaranteeing computational efficiency and real-time performance. Simultaneously, the system establishes an independent timeout protection mechanism, setting a maximum allowable time threshold for the entire optimization process. If the optimization calculation fails to converge to a feasible solution within this time limit, the protection mechanism will forcibly terminate the current iteration loop and trigger a degradation processing strategy, such as switching to a pre-set conservative control strategy or issuing an alarm, to prevent the system from becoming unresponsive due to optimization process stagnation.

[0065] The entire optimization computation is performed within a dedicated resource container. Its core process involves executing the improved distributed collaborative optimization solution process detailed in step 103 of this invention, including parallel local optimization by each edge controller, boundary information exchange, and global coordination by the cloud coordinator. Crucially, while the optimization computation is intensively performed, the system's state monitoring function continues to operate on its original resource track, collecting and processing grid data in real time. This design achieves parallel processing of control and monitoring: the monitoring system continuously tracks the grid state, providing the latest boundary data input to the optimization algorithm and continuously judging whether new, more urgent triggering events occur during the optimization process; while the optimization process focuses on solving for the optimal control command. Both exchange data securely and with low latency through shared memory or message queues, thereby efficiently completing the collaborative optimization computation task while ensuring the observability of the system's global state.

[0066] In one embodiment, in step 103, please refer to Figure 6 Initialize the parameters of the improved alternating direction multiplier method, including: Step 1031: Initialize the Lagrange multiplier vector to a zero vector; initialize the boundary variable reference value to the measured value at the trigger time; initialize the penalty parameter to the preset initial value of the penalty factor; set the maximum number of iterations to the preset maximum number of iterations threshold, the absolute convergence tolerance to the preset absolute convergence tolerance value, and the relative convergence tolerance to the preset relative convergence tolerance value.

[0067] Specifically, this step is the preparatory work for the optimization process. It receives the start instruction from step 102 and sets initial values ​​for subsequent iterations. Initial boundary variable reference values. Typically, the parameters are determined based on the current measurement value at the trigger moment (from the monitoring results in step 102). Parameter initialization and configuration are fundamental steps in initiating the optimization process. After confirming the event trigger, the cloud coordinator first sends an optimization start command to the relevant edge clusters. Subsequently, the system initializes the core parameters of the improved alternating direction multiplier method (ADMM), including: Lagrange multiplier vectors: ; In the formula, It is a Lagrange multiplier vector with a dimension equal to the number of boundary constraints, and an initial value of zero vector.

[0068] Boundary variable reference values: Determined based on current measurements

[0069] Penalty parameters: ; In the formula, It is a reference vector of boundary state quantities (such as boundary voltage and switching power), with the initial value being the measured value at the trigger moment.

[0070] Maximum number of iterations: ; Convergence tolerance: ; Among them, Lagrange multipliers The dimension m is equal to the number of boundary constraints, and the boundary variables. Dimensions This is equal to the number of boundary state variables. These parameters are distributed to each edge controller via a secure communication link, ensuring that all nodes begin optimization calculations based on uniform initial conditions.

[0071] In one embodiment, in step 103, please refer to Figure 6 Each edge controller obtains boundary state variables by solving local optimization sub-problems in parallel based on the local aggregation adjustable capability model, including: Step 1032: Each edge controller establishes a local optimization sub-problem for this iteration based on the current penalty factor and the local aggregation adjustable capability model; The local optimization subproblem aims to minimize the cluster operating cost, which is determined based on the cost function in the local aggregate adjustable capability model. The constraints of the local optimization subproblem include the upper and lower limits of power, ramp rate limit, voltage safety boundary, and power flow balance equation of each distributed unit in the cluster, wherein the power flow balance equation of the power flow is a non-convex constraint. Step 1033: Each edge controller uses the LinDistFlow linearization model to convexize the non-convex constraints, transforming the local optimization subproblem into a convex optimization problem; Step 1034: Each edge controller solves the convex optimization problem to obtain the active power adjustment and reactive power adjustment of each controllable distributed unit in the cluster, and calculates the boundary state variables of the cluster based on the active power adjustment and reactive power adjustment.

[0072] Specifically, each edge controller utilizes the current penalty factor and the local aggregation tunable capability model defined in step 101, and receives coordination variables from the cloud, to solve for the local optimal solution in parallel. The use of convexity techniques such as LinDistFlow ensures that the local problem is necessarily solvable, which is the foundation for the convergence of the entire ADMM algorithm and solves the non-convex convergence problem mentioned in the background. Parallel local optimization is the core of the distributed architecture. After receiving the coordination variables, each edge controller independently solves its local automatic power control sub-problem. The mathematical model of the local optimization sub-problem is: ; The following conditions must be met simultaneously: (Inequality constraints); (Equality constraints); (Feasible region constraint).

[0073] in: , is the running cost function for cluster k; Active power adjustment cost function; Reactive power adjustment cost function; In the formula, The boundary coupling matrix has a structure consisting of clusters. The electrical connectivity with neighboring clusters determines the local decision variables. Mapped to boundary state variables. For cluster The decision variable vector typically includes the active power regulation of each controllable distributed unit within the cluster. and reactive power regulation . For cluster The operating cost function, specifically expressed as active power regulation cost. With reactive power adjustment costs The sum of these coefficients is derived from the cost coefficients determined during the cluster aggregation adjustable capability modeling process. The Lagrange multiplier vector associated with the boundary consistency constraints has a dimension equal to the number of system boundary constraints and is used in the iteration to coordinate the optimization solutions of each cluster to satisfy the global coupling relationship. This is a globally shared vector of boundary variable reference values, representing the target values ​​of temporary boundary state quantities (such as boundary node voltage and tie line power) during the iteration process. The penalty parameter is used to adjust the weight of the boundary consistency constraint in the objective function, and its value is dynamically adjusted according to the adaptive strategy proposed in this invention. This represents all the inequality constraints that cluster k must satisfy, mainly including the upper and lower power limits, ramp rate limits, and voltage safety boundaries of each distributed unit within it. The equality constraints that cluster k must satisfy are typically power balance equations based on a linearized power flow model. Decision variables The feasible region is defined by the aforementioned inequality constraints, equality constraints, and physical limitations of the equipment.

[0074] To handle non-convexity problems, the LinDistFlow linearization model is used for convexity transformation: ; ; in, This represents the voltage magnitude (per-unit value or actual value) at the end node i of the path from the root node (number 0) to node i. It represents the voltage amplitude at the root node of the distribution network (usually the connection point or slack node of the upstream power grid) and is used as a reference voltage in calculations. This represents the set of all branches along the path from the root node 0 to node i. This indicates the resistance (per-unit value or actual value in Ω) of branch ij. This represents the reactance (per-unit value or actual value in Ω) of branch ij. This represents the active power flowing through branch ij (per unit value or actual value kW / MW). This represents the reactive power flowing through branch ij (per unit value or actual value kvar / Mvar). This represents the active power loss (per unit value or actual value kW / MW) on branch ij.

[0075] This linearization model transforms non-convex power flow constraints into a linear form, ensuring that the local problem is a convex optimization problem, which can be solved efficiently using algorithms such as the interior-point method. In the parallel local optimization solution step, each edge controller solves the convex local optimization problem, and the most direct result obtained is its decision variable vector. The optimal solution is found by optimizing the active and reactive power planned adjustments of each controllable distributed unit within cluster k. Specifically, the core parameters optimized in this process are these adjustments, which directly constitute the decision variables. The elements mainly include: active power regulation commands for all photovoltaic inverters, energy storage converters, and other units within the cluster. And reactive power regulation commands for all grid-connected inverters with reactive power regulation capabilities. These optimization results play a crucial role in subsequent processes: First, they are used to calculate the boundary state variables of this cluster, namely the boundary node voltage and switching power, and are reported as key information to the cloud coordinator in the "Boundary Coordination and Information Exchange" sub-step of step 103 to calculate the global consensus error and drive iteration; finally, when the global algorithm converges, these final optimized values, confirmed by global coordination, are used to... and These commands will be directly passed to the "Control Command Generation and Execution" step 104. There, they will be parsed, security verified, and assigned as specific device-level control commands, which will then be sent to the physical execution unit, thus completing the full closed loop from optimization calculation to actual power regulation. Therefore, the power regulation commands obtained from the local optimization solution are the core link realizing the entire invention from collaborative calculation to physical control.

[0076] In one embodiment, in step 103, the cloud coordinator calculates the consensus error based on each of the boundary state variables and dynamically adjusts the penalty factor according to the consensus error, including: Step 1035: The cloud coordinator calculates the original residual vector and dual residual vector of the current iteration based on the boundary state variables uploaded by each edge controller. The original residual vector and the dual residual vector together constitute the quantization of the current consensus error. Step 1036: The cloud coordinator compares the 2-norm of the original residual vector with the 2-norm of the dual residual vector, and dynamically adjusts the penalty factor based on the comparison result, including: When the 2-norm of the original residual vector is greater than the 2-norm of the dual residual vector, the penalty factor is amplified by an increasing factor, but not exceeding the preset upper limit. When the 2-norm of the dual residual vector is greater than the 2-norm of the original residual vector, the penalty factor is reduced by a decreasing factor, but not lower than the preset lower limit. The penalty factor remains unchanged when the 2-norm of the original residual vector and the 2-norm of the dual residual vector are in relative equilibrium.

[0077] Specifically, this sub-step is crucial for achieving global coordination. Each cluster reports its local optimization results (boundary information), and the cloud coordinator calculates the consensus error based on this. The "boundary state quantities" exchanged here are the same set of physical quantities monitored in step 102, but their purposes differ: step 102 uses them to determine whether a trigger has occurred, while here they are used for coordination and optimization. After completing their local optimization, each edge controller uploads the solved boundary state quantities to the cloud coordinator.

[0078] The exchanged information includes: boundary node voltage vectors: This represents a vector consisting of the voltage magnitudes of the boundary nodes across all relevant clusters, where dimension m equals the total number of monitored boundary nodes in the system. Boundary exchange power vector: This represents a vector consisting of the active and reactive power exchanged along all boundary tie lines, with each pair... The power corresponding to a boundary tie line.

[0079] After collecting boundary information from all clusters, the cloud coordinator calculates the boundary consensus error: Original residuals: ; Dual residuals: ; In the formula, This represents the original residual vector at the t-th iteration. It quantifies the local optimization results for all K clusters. With the current global consensus goal The difference between them is a key indicator for measuring the original feasibility (boundary consistency) of an optimization problem. Let represent the dual residual vector at the t-th iteration. It consists of a penalty factor. The change in adjacent iterations of the boundary variable reference value z The product obtained by multiplication reflects the update magnitude of the dual variable λ, which is an indicator of dual feasibility (gradient optimality).

[0080] The improved ADMM algorithm uses an adaptive strategy to dynamically adjust the penalty factor ρ, and its update rule is as follows: Define the penalty factor update function: ; Where the function Determined by the following piecewise function: Case 1: When the 2-norm of the original residual vector is significantly greater than the 2-norm of the dual residual vector: If satisfied The penalty factor is calculated according to the increasing factor. Enlarge, but not exceeding the preset upper limit. : ; Case 2: When the 2-norm of the dual residual vector is significantly greater than the 2-norm of the original residual vector: If satisfied The penalty factor is then calculated using the decreasing factor. Shrink, but not below the preset lower limit : ; Case 3: When the 2-norm of the original residual vector and the 2-norm of the dual residual vector are in relative equilibrium: If neither of the above two conditions is met, the penalty factor will remain unchanged. ; in, Let represent the penalty factor in the t-th iteration. It is a positive scalar used to weight the boundary consistency constraint term in the objective function, and its value is dynamically adjusted according to the adaptive rule. This represents the relative threshold parameter, a scalar greater than 1 (recommended value is 5). It is used to determine whether the norm of the original residual and the dual residual is "significantly" greater than that of the other, thus determining the direction of the penalty factor adjustment. This represents the increment factor, a scalar greater than 1 (recommended value is 1.5). It is used to scale the penalty factor when the original residuals are dominant. This represents the decreasing factor, a scalar greater than 1 (recommended value is 1.5). It is used to proportionally reduce the penalty factor when the dual residual is dominant. This represents the upper limit of the penalty factor, a preset positive scalar (recommended value is 100), used to prevent the penalty factor from being too large and causing numerical problems. Indicates: Lower bound of the penalty factor, a preset positive scalar (recommended value is 0.1), used to prevent the penalty factor from being too small, which would lead to slow convergence.

[0081] The adaptive penalty factor strategy is one of the key improvements of this invention. It dynamically adjusts the penalty parameters based on the consensus error. By increasing the penalty when the convergence speed is slow and finely adjusting it when it is close to convergence, the convergence process is effectively accelerated and the real-time performance of the control is improved.

[0082] In summary, the original residuals and dual residuals This is a key intermediate result calculated after each iteration to evaluate the algorithm's state, not the final optimized output. The adaptive penalty factor strategy compares the norms of these two residuals and, based on preset parameters (…),… , , , , Dynamic adjustment The aim is to balance the convergence speed of primal feasibility and dual feasibility, thereby accelerating the convergence of the entire distributed optimization process.

[0083] Furthermore, in one embodiment, the convergence judgment and iterative control process in step 103 is as follows: Specifically, this sub-step controls the termination of the optimization loop. If convergence fails, the coordination variables are updated and the system returns to steps 1035-1036 to continue iteration; if convergence has occurred, it means a globally feasible optimal solution has been found, and the system immediately proceeds to step 104 to execute the proposed solution. The system determines convergence based on the relative error criterion: Original residual convergence condition: ; Dual residual convergence condition: ; in, The number of clusters participating in the optimization; For absolute tolerance; relative tolerance If all the above conditions are met, the algorithm is considered converged, and the optimization result is output; otherwise, the coordination variable is updated. Lagrange multipliers update: ; Boundary variable update: ; Return to the parallel local optimization step and continue iterating. Also, if the number of iterations reaches... If the calculation fails, the computation will be forcibly terminated to ensure real-time requirements are met.

[0084] In one embodiment, see Figure 7 Step 104 includes: Step 1041: Convert the power regulation amount contained in the local optimization solution in the optimization result into device-level control instructions. The device-level control instructions include the active power load reduction ratio and reactive power setpoint of the photovoltaic inverter, and the charging and discharging power and duration of the energy storage system. Specifically, after the distributed collaborative optimization solution in step 103 converges, each edge controller generates specific device-level control commands based on the final determined local optimization solution. The control command generation process includes two core steps: command parsing and power allocation. Command parsing converts the power adjustment output by the optimization algorithm into control parameters that the device can recognize, while power allocation rationally distributes the total adjustment to specific execution units based on the adjustability and operating status of each unit within the cluster. For photovoltaic inverters, the control commands include active power descent ratio and reactive power setpoints; for energy storage systems, the commands specify the charging and discharging power and duration; for wind turbines, the commands involve pitch angle adjustment and converter control parameters.

[0085] Step 1042: Perform security verification on the device-level control commands. The security verification includes verification of the device's safe operating range, verification of fast power flow prediction, and verification of control coordination. Specifically, before executing control commands, the edge controller performs multi-level safety checks on the generated device-level control commands. The first level of verification checks whether the command exceeds the device's safe operating range, including power limits, voltage tolerance, and thermal stability constraints (i.e., checking if the command exceeds the device's nameplate parameters). The second level of verification predicts the system state after command execution through fast power flow calculations to ensure no new safety risks are introduced (i.e., performing fast power flow calculations to predict the system state after command execution and checking for new voltage exceedances or overloads). The third level of verification assesses the coordination of control commands to avoid control conflicts between different devices (i.e., checking for conflicts between commands from different devices, such as simultaneous charging and discharging requirements). If any verification fails, the system will initiate a command correction mechanism or re-trigger optimization calculations.

[0086] Step 1043: Based on the response speed of each unit, formulate a batch execution strategy, and in accordance with the batch execution strategy, issue the device-level control instructions to the corresponding units in sequence through an encrypted confirmation mechanism, so that the corresponding units execute the device-level control instructions; Specifically, considering the differences in response characteristics among different devices, the execution timing of device-level control commands is precisely coordinated. Based on the classification of device response speeds, a batch execution strategy is formulated: first, units with fast response capabilities are adjusted (such as energy storage converters and photovoltaic inverters); second, devices with medium response speeds (such as wind turbines) are adjusted; and finally, mechanical regulating devices are adjusted as needed. The execution interval for each batch is set according to the dynamic characteristics of the power grid to ensure a smooth transition in the control process.

[0087] Step 1044: Monitor the execution status of each unit in real time. When the actual output deviates from the command, adjust the equipment-level control commands of other units in the same batch or subsequent batches to compensate for the deviation. Specifically, verified control commands are sent to each distributed renewable energy unit via a reliable communication channel. Command transmission employs encryption and confirmation mechanisms to ensure transmission integrity and reliability. During execution, the edge controller monitors the equipment's execution status in real time, including deviations between actual output and commands, and changes in equipment operating parameters. If the execution deviation exceeds a threshold, the system automatically activates a compensation mechanism to adjust the output of other units to make up for the deviation.

[0088] The compensation mechanism is as follows: During the monitoring of control command execution, if the edge controller detects that the deviation between the actual output of a device and the command value exceeds a preset threshold, the system will automatically initiate a local closed-loop compensation mechanism. This mechanism aims to quickly restore the overall control objective of the cluster without re-triggering global collaborative optimization. Its specific implementation process is as follows: (1) Deviation Judgment and Compensation Calculation: The system compares the feedback data of each device with the issued control commands in real time. When the absolute value of the execution deviation ΔP (or ΔQ) of a certain unit (denoted as device A) exceeds its dynamic threshold (e.g., 5% of the command value or a fixed power limit), the compensation process is triggered. The total power to be compensated is the excess deviation, i.e., ΔP_comp = ΔP_A (or ΔQ_comp = ΔQ_A), and the goal is to ensure that the total output of the cluster is consistent with the total amount required by the optimization command.

[0089] (2) Quick screening of available backup resources: The edge controller immediately queries its latest local "adjustable capacity" model and real-time status to screen other controllable units in the cluster that have not yet deviated and still have corresponding adjustment margins, forming a backup resource pool. When screening, priority is given to devices with fast response speed (such as energy storage converters) and high adjustment accuracy.

[0090] (3) Optimized allocation and instruction generation of compensation: The system does not simply distribute the compensation equally, but performs a fast local optimization calculation. The goal is to minimize the additional costs or disturbances to the original operating plan caused by compensation while satisfying the total compensation constraint. For example, allocation can be made according to the proportion of remaining adjustment capacity, or economic allocation can be made based on the marginal cost of the secondary cost function of each standby unit. After calculation, compensation instructions are generated for one or more standby units (such as equipment B and C).

[0091] (4) Issuance of compensation instructions and verification of effects: The generated compensation instructions are issued to the selected backup units through a secure channel. Subsequently, the controller continuously monitors the execution status of the compensation units and the recovery status of the total output at the cluster boundary. If the total output of the cluster successfully returns to the target range after compensation, the compensation is successful and the event log is recorded. If the requirements are still not met due to insufficient backup resources or after compensation, the event will be upgraded to a new local or global limit violation event, which may trigger a new round of event-driven collaborative optimization as described in step 102.

[0092] Step 1045: After all device-level control commands have been executed, verify whether the boundary state variables have returned to the safe range. If so, record the data, reset the trigger flag, and restore the silent monitoring state.

[0093] Specifically, after all device-level control commands have been executed, the system enters the effect verification phase. By collecting boundary state variables and key node operating data, it verifies whether voltage overruns or line overloads have been eliminated. Simultaneously, it assesses whether the system's operating state has returned to a safe range. Upon successful verification, the system records all data from the entire control process, including the triggering cause, optimization results, and control effects, and resets the trigger flag (i.e., reset condition) to restore normal monitoring status, preparing for the next event response.

[0094] Regarding the reset conditions, it should be noted that after the optimized control is completed, the system automatically enters the state recovery phase, using a multi-dimensional verification mechanism to ensure the safety and stability of the power grid operation before resuming silent monitoring. Specific reset verification conditions include: the boundary node voltage must remain consistently stable within the normal safe range of 0.95-1.05 per-unit values, and this stable state must be maintained for at least three consecutive sampling periods to eliminate misjudgments caused by instantaneous fluctuations; the tie-line transmission power must decrease to below 90% of the line thermal stability limit to reserve necessary safety margins for system operation; simultaneously, the system requires that the rate of change of each state variable tends to stabilize, and its fluctuation amplitude must not exceed the preset stabilization threshold to ensure the system's dynamic characteristics return to stability. These three conditions must be met simultaneously for the system to execute a reset operation, clear the trigger flag, release optimized computing resources, switch from the "optimized control" state back to the "silent monitoring" state, and complete the full closed loop of this event trigger.

[0095] The system establishes a complete event archive, recording key data such as triggering causes, control effects, and response times. Based on historical data analysis, it automatically optimizes triggering thresholds and monitoring strategies, forming a self-improving intelligent learning mechanism. This step, through precise state perception, intelligent trigger judgment, reliable process coordination, and an adaptive learning mechanism, constructs an efficient event-driven control system, providing accurate start signals and necessary preparation conditions for subsequent optimization calculations. In summary, this event triggering mechanism fully utilizes the system architecture and parameter foundation established in step 101, achieving seamless integration from resource preparation to optimized control, ensuring that the system maximizes the conservation of communication and computing resources while guaranteeing control effectiveness.

[0096] In summary, through the detailed explanation and logical correlation analysis of the above four steps, it can be seen that the distributed new energy cluster adaptive power cooperative control method of the present invention provides a highly structured and adaptive closed-loop control scheme. Initialization in step 101 is fundamental; event triggering in step 102 is key to energy saving and agility; improvement and optimization in step 103 ensures performance and reliability; and control execution in step 104 is the ultimate manifestation of value. These four steps are interdependent and interconnected, together forming a complete automatic power control solution that is highly efficient in communication, protects privacy, ensures reliable convergence, and is responsive, systematically solving all the shortcomings of existing technologies mentioned in the background section.

[0097] The above describes a distributed new energy cluster adaptive power cooperative control method provided in the embodiments of the present invention. The following describes a distributed new energy cluster adaptive power cooperative control system provided in the embodiments of the present invention.

[0098] Please see Figure 8The present invention provides a distributed new energy cluster adaptive power coordination control system, comprising: The first control submodule 201 is used to divide the power distribution network into several dynamic clusters, wherein each cluster is managed by an edge controller. The edge controller quantifies the aggregation and adjustability of the cluster and encapsulates it into a standardized information package, which serves as the only data unit for interaction with the cloud coordinator. The second control submodule 202 is used to monitor the state variables of the cluster boundary nodes to determine whether an over-limit event has occurred. When an over-limit event is detected, a collaborative optimization process is triggered to generate a start command and synchronize it to the cloud coordinator and all edge controllers. The third control submodule 203 is used to respond to the start command, initialize the improved alternating direction multiplier method parameters through the cloud coordinator, and each edge controller constructs a local aggregate adjustable capability model based on the standardized information packet and enters the iterative process: In each iteration, each edge controller solves the local optimization sub-problem in parallel based on the local aggregate adjustable capability model to obtain the boundary state variables, so that the cloud coordinator calculates the consensus error based on each boundary state variable, and dynamically adjusts the penalty factor according to the consensus error to update the coordination variables before entering the next iteration, until the convergence condition is met and a globally consistent optimization result is output; The fourth control submodule 204 is used to generate device-level control instructions through each edge controller based on the local optimization solution in the optimization results, perform multi-level security verification on the device-level control instructions and execute them in batches in sequence, and reset the trigger flag after verifying that the over-limit event that triggered the start instruction has been eliminated, and restore the silent monitoring state.

[0099] This invention provides a distributed new energy cluster adaptive power cooperative control system, offering a highly structured and adaptive closed-loop control scheme. The initialization of the first control submodule 201 is fundamental; the event triggering of the second control submodule 202 is crucial for energy saving and agility; the improvement and optimization of the third control submodule 203 ensures performance and reliability; and the control execution of the fourth control submodule 204 is the ultimate realization of its value. These four steps are interdependent and interconnected, collectively forming a complete automatic power control solution that is highly efficient in communication, protects privacy, ensures reliable convergence, and provides agile response, systematically addressing all the shortcomings of existing technologies mentioned in the background section.

[0100] Furthermore, this embodiment of the invention also provides a distributed new energy cluster adaptive power coordination control device, the device including a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the steps of the distributed new energy cluster adaptive power cooperative control method as described in the above method embodiments, according to the instructions in the program code.

[0101] Furthermore, this embodiment of the invention also provides a computer-readable storage medium for storing program code, which is used to execute the distributed new energy cluster adaptive power cooperative control method described in the above method embodiments.

[0102] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0103] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0104] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0106] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for adaptive power coordination control of distributed new energy clusters, characterized in that, include: The power distribution network is divided into several dynamic clusters, each of which is managed by an edge controller. The edge controller quantifies the aggregation and adjustability of the cluster and encapsulates it into a standardized information package, which serves as the only data unit for interaction with the cloud coordinator. Monitor the state variables of the cluster boundary nodes to determine if an out-of-limit event has occurred. When an out-of-limit event is detected, trigger the collaborative optimization process to generate a startup command and synchronize it to the cloud coordinator and all edge controllers. In response to the start command, the improved alternating direction multiplier method parameters are initialized through the cloud coordinator. Each edge controller constructs a local aggregate adjustable capability model based on the standardized information packet and enters the iterative process: In each iteration, each edge controller solves the local optimization sub-problem in parallel based on the local aggregate adjustable capability model to obtain the boundary state variables, so that the cloud coordinator calculates the consensus error based on each boundary state variable, and dynamically adjusts the penalty factor according to the consensus error to update the coordination variables before entering the next iteration, until the convergence condition is met and a globally consistent optimization result is output; Each edge controller generates device-level control instructions based on the local optimization solution in the optimization results. After performing multi-level security checks on the device-level control instructions, they are executed in batches according to the timing. After verifying that the over-limit event that triggered the start instruction has been eliminated, the trigger flag is reset and the silent monitoring state is restored.

2. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The division of the distribution network into several dynamic clusters includes: The electrical distance between any two nodes is calculated based on the node impedance matrix, and the communication cost coefficient is constructed based on communication delay, bandwidth and packet loss rate. The electrical distance and the communication cost coefficient are normalized and then weighted and fused to construct a comprehensive correlation weight matrix; The optimal partitioning scheme is generated based on the comprehensive association weight matrix using the spectral clustering algorithm, and the number of clusters and the physical boundaries of each cluster are determined according to the optimal partitioning scheme.

3. The distributed new energy cluster adaptive power cooperative control method according to claim 2, characterized in that, The method of dividing the distribution network into several dynamic clusters also includes: Establish repartitioning trigger conditions to automatically initiate repartitioning when the network topology changes or communication performance deteriorates.

4. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The quantization cluster's aggregation and adjustment capabilities are encapsulated into standardized information packets, including: Based on the operating data of the distributed new energy sources within the cluster, the state estimation results are obtained by weighted least squares method. The state estimation results include the voltage amplitude and phase angle of each node within the cluster. Based on the state estimation results, the adjustment potential of each unit in the cluster is calculated. The adjustment potential includes the active power reduction space of the photovoltaic unit, the active power increase space and active power reduction space of the energy storage system, and the maximum reactive power absorption capacity and maximum reactive power generation capacity of all inverters with reactive power adjustment capabilities. The adjustment potential of each unit is classified and aggregated according to power type to form a cluster-level total active power upward adjustment space, total active power downward adjustment space, total reactive power absorption space and total reactive power output space to obtain four types of aggregated adjustable capabilities. The adjustment cost function corresponding to each type of aggregated adjustable capability is established with the adjustment power as the independent variable. The feasibility of the aggregated adjustable capability is verified by power flow calculation, and the verified aggregated adjustable capability is encapsulated into a standardized adjustable capability information package.

5. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The monitoring of the state variables of the cluster boundary nodes to determine whether an out-of-limit event has occurred includes: The voltage amplitude and tie-line power of the cluster boundary nodes are monitored through dual data sources. The first data source is the edge controller that collects real-time data from the local measurement device at a preset first sampling frequency. The second data source is the cloud coordinator that receives synchronous phasor data from the wide-area measurement system. Based on the preset voltage over-limit judgment conditions and the preset power over-limit judgment conditions, determine whether an over-limit event has occurred; The voltage over-limit determination conditions include: the voltage amplitude at the boundary node exceeds the preset voltage operating range, or the voltage change rate exceeds the preset voltage change rate threshold. The power limit determination conditions include: the power of the tie line exceeds the preset transmission power limit, or the power change rate exceeds the preset power change rate threshold.

6. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The process of triggering a collaborative optimization procedure upon detecting an out-of-limit event generates a startup command and synchronizes it to the cloud coordinator and all edge controllers, including: When an out-of-limit event is detected, the trigger unit generates a start command. The start command is encapsulated into a fixed-length data frame, which includes the event type, severity, timestamp, and trigger unit identifier. The triggering unit sends the start command to the receiver through a preset priority communication channel. When the receiver receives the start command, it verifies the integrity of the frame and sends a confirmation response to the triggering unit. The receiver includes all edge controllers and cloud coordinators participating in the collaborative optimization. After receiving the confirmation response, the triggering unit replies with a ready signal to the receiver to complete the three-way handshake; if the three-way handshake is not completed within a preset time, the triggering unit initiates command retransmission or switches to the backup communication route. After the three-way handshake is completed, all receivers and triggering units enter the optimization state in unison according to the timestamp in the start command.

7. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The initialization parameters for the improved alternating direction multiplier method include: Initialize the Lagrange multiplier vectors to zero vectors; The reference values ​​for the initial boundary variables are the measured values ​​at the trigger time; The penalty parameters are initialized to the preset initial value of the penalty factor; The maximum number of iterations is set to a preset maximum number of iterations threshold, the absolute convergence tolerance is set to a preset absolute convergence tolerance value, and the relative convergence tolerance is set to a preset relative convergence tolerance value.

8. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, Each edge controller obtains boundary state variables by solving local optimization sub-problems in parallel based on the local aggregation adjustable capability model, including: Each edge controller establishes a local optimization sub-problem for this iteration based on the current penalty factor and the local aggregation adjustable capability model; The local optimization subproblem aims to minimize the cluster operating cost, which is determined based on the cost function in the local aggregate adjustable capability model. The constraints of the local optimization subproblem include the upper and lower limits of power, ramp rate limit, voltage safety boundary, and power flow balance equation of each distributed unit in the cluster, wherein the power flow balance equation of the power flow is a non-convex constraint. Each edge controller uses the LinDistFlow linearization model to convexize the non-convex constraints, transforming the local optimization subproblem into a convex optimization problem. Each edge controller solves the convex optimization problem to obtain the active power regulation and reactive power regulation of each controllable distributed unit in the cluster, and calculates the boundary state variables of the cluster based on the active power regulation and reactive power regulation.

9. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The cloud coordinator calculates the consensus error based on each of the boundary state variables, and dynamically adjusts the penalty factor according to the consensus error, including: The cloud coordinator calculates the original residual vector and dual residual vector for the current iteration based on the boundary state variables uploaded by each edge controller. The original residual vector and the dual residual vector together constitute the quantization of the current consensus error. The cloud coordinator compares the 2-norm of the original residual vector with the 2-norm of the dual residual vector, and dynamically adjusts the penalty factor based on the comparison result.

10. The distributed new energy cluster adaptive power cooperative control method according to claim 9, characterized in that, The dynamic adjustment of the penalty factor based on the comparison results includes: When the 2-norm of the original residual vector is greater than the 2-norm of the dual residual vector, the penalty factor is amplified by an increasing factor, but not exceeding the preset upper limit. When the 2-norm of the dual residual vector is greater than the 2-norm of the original residual vector, the penalty factor is reduced by a decreasing factor, but not lower than the preset lower limit. The penalty factor remains unchanged when the 2-norm of the original residual vector is in relative balance with the 2-norm of the dual residual vector.

11. The distributed new energy cluster adaptive power cooperative control method according to claim 1, characterized in that, The process of generating device-level control instructions based on the local optimized solution in the optimization results, performing multi-level security checks on the device-level control instructions, executing them in batches according to timing, and resetting the trigger flag after verifying that the over-limit event that triggered the start instruction has been eliminated, and restoring the silent monitoring state, includes: The power regulation contained in the local optimization solution in the optimization results is converted into device-level control commands. The device-level control commands include the active power load reduction ratio and reactive power setpoint of the photovoltaic inverter, and the charging and discharging power and duration of the energy storage system. The device-level control commands are subjected to security verification, which includes verification of the device's safe operating range, verification of fast power flow prediction, and verification of control coordination. Based on the response speed of each unit, a batch execution strategy is formulated, and the device-level control commands are sequentially sent to the corresponding units through an encrypted confirmation mechanism in accordance with the batch execution strategy, so that the corresponding units execute the device-level control commands. The execution status of each unit is monitored in real time. When the actual output deviates from the command, the equipment-level control command of other units in the same batch or subsequent batches is adjusted to compensate for the deviation. After all device-level control commands have been executed, verify whether the boundary state variables have returned to a safe range. If so, record the data, reset the trigger flag, and restore the silent monitoring state.

12. A distributed new energy cluster adaptive power coordination control system, characterized in that, include: The first control submodule is used to divide the distribution network into several dynamic clusters, wherein each cluster is managed by an edge controller. The edge controller quantifies the aggregation and adjustability of the cluster and encapsulates it into a standardized information package, which serves as the only data unit for interaction with the cloud coordinator. The second control submodule is used to monitor the state variables of the cluster boundary nodes to determine whether an over-limit event has occurred. When an over-limit event is detected, a collaborative optimization process is triggered to generate a startup command and synchronize it to the cloud coordinator and all edge controllers. The third control submodule is used to respond to the start command, initialize the improved alternating direction multiplier method parameters through the cloud coordinator, and each edge controller constructs a local aggregate adjustable capability model based on the standardized information packet and enters the iterative process: In each iteration, each edge controller solves the local optimization sub-problem in parallel based on the local aggregate adjustable capability model to obtain the boundary state variables, so that the cloud coordinator calculates the consensus error based on each boundary state variable, and dynamically adjusts the penalty factor according to the consensus error to update the coordination variables before entering the next iteration, until the convergence condition is met and a globally consistent optimization result is output; The fourth control submodule is used to generate device-level control instructions through each edge controller based on the local optimization solution in the optimization results, perform multi-level security verification on the device-level control instructions and execute them in batches in sequence, and reset the trigger flag after verifying that the over-limit event that triggered the start instruction has been eliminated, and restore the silent monitoring state.

13. A distributed new energy cluster adaptive power coordination control device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the distributed new energy cluster adaptive power cooperative control method according to any one of claims 1-11 according to the instructions in the program code.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the distributed new energy cluster adaptive power cooperative control method according to any one of claims 1-11.

Images