A Two-Level Consistency Distributed Control Method Based on Heterogeneous Resources in Virtual Power Plants
By adopting a two-layer consistent distributed control method, the problems of frequency regulation response metering and deviation allocation of heterogeneous resources in a virtual power plant are solved, realizing a fast and accurate response to the primary frequency regulation of the power grid and overcoming the challenges of global communication dependence and resource heterogeneity in traditional control.
Patent Information
- Application Number
- CN202611176057.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-04
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-08-04
AI Technical Summary
The lack of physical common connection points for heterogeneous resources within virtual power plants makes it difficult to achieve real-time distributed metering of frequency regulation response and to balance and autonomously allocate active power deviation, thus affecting the rapid and accurate response of the primary frequency regulation of the power grid.
A two-layer consensus distributed control method is adopted, which realizes data interaction through bidirectional communication in a decentralized topology. Combined with response detection consensus calculation and error compensation consensus algorithm, the frequency regulation response and active power deviation are accurately measured and evenly distributed within a fixed time period, respectively, to build a signal-material synchronous evolution architecture and realize multi-stage rolling regulation.
Under the condition of no physical common connection point, the system realizes accurate distributed measurement of frequency regulation response of virtual power plant and balanced autonomous allocation of active power deviation, ensuring the physical boundary security of heterogeneous resources and realizing rapid and accurate response of primary frequency regulation of power grid.
Smart Images

Figure CN122678233B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of virtual power plant and grid control technology, and particularly to a two-layer consistent distributed control method based on heterogeneous resources in a virtual power plant. Specifically, this method involves accurate metering of the total output of heterogeneous resources, comprehensive assessment of the generalized availability of multi-source equipment, balanced autonomous allocation of frequency regulation active power deviation, and two-layer collaborative control to achieve error-free primary frequency regulation in the virtual power plant. Background Technology
[0002] Virtual power plants play a crucial role in primary frequency regulation of the power grid by aggregating massive distributed energy resources (such as energy storage, wind and solar power, and controllable loads). To ensure rapid and accurate response during frequency regulation, a distributed control architecture is an effective way to break away from the dependence of traditional centralized control on global communication. However, because heterogeneous resources within a virtual power plant typically lack physical common connection points, it is difficult not only to achieve real-time distributed metering of the cumulative frequency regulation response of each dispersed resource, but also to achieve a reasonable and balanced distribution of active power deficit among resources.
[0003] The core challenges are as follows: First, the topologies of each aggregated resource are isolated, relying solely on sparse network interaction data. Coupled with the inherent random instability on both the source and load sides, global frequency regulation data naturally experiences asynchronous delays during transmission, making it impossible to obtain accurate cumulative responses through real-time local aggregation. Second, primary frequency regulation of the power grid requires rapid response (second-level response), while the resources within the virtual power plant exhibit significant heterogeneity, with different power conversion interfaces, AC / DC configurations, and power flow directions, resulting in extremely different levels of regulation response speed. This multi-dimensional spatiotemporal mismatch between information and things, caused by the varying physical ramp rates and temporal dynamic constants of "wind, solar, storage, and load," makes it difficult to autonomously and evenly allocate frequency regulation active power deviations as needed.
[0004] In view of this, the present invention proposes a new solution to address the above-mentioned problems. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a two-layer consensus distributed control method based on heterogeneous resources of virtual power plants, which aims to achieve a fast and accurate response to the primary frequency regulation command of the power grid.
[0006] To solve the technical problem, the solution of the present invention is:
[0007] A two-layer consensus distributed control method based on heterogeneous resources of a virtual power plant is provided, wherein the virtual power plant includes... In a distributed energy resource system without a physical common connection point, each network node interacts with the virtual power plant master station via bidirectional communication in a decentralized topology. When the grid frequency drops out of the dead zone, the distributed energy resources execute droop control through their respective grid-connected converters to enter an initial steady state with static errors. Subsequently, a two-layer consistent distributed control is initiated, specifically including the following steps:
[0008] (1) Consistency calculation of the first layer's response detection:
[0009] Within each node, maintain the global average value estimate of the active frequency regulation response increment and the global average value estimate of the residual active power error. The distributed metering virtual power plant accumulates the frequency regulation response through the fixed-time consistency update equation and obtains the converged global average active power error.
[0010] (2) Error compensation consensus algorithm for the next level:
[0011] A generalized availability rate is defined to quantitatively evaluate the relative adjustment capability of heterogeneous resources. Virtual adjustment intensity and local dynamic deviation estimates are maintained within each network node. Active power deviation correction instructions are output through the consistency update equation to drive the heterogeneous distributed energy resources to synchronously perform power ramp-up response during the algorithm iteration process.
[0012] (3) Multi-stage rolling adjustment is achieved by alternating operation of the above two-layer computer mechanism until the global remaining frequency regulation error is lower than the preset threshold, so as to realize the virtual power plant's response to the primary frequency regulation command in a gradual manner.
[0013] As a preferred embodiment of the present invention, in the communication between the virtual power plant and the distributed energy resources... The network includes one master coordinating node (i.e., the Leader node numbered 1), and... Heterogeneous follower nodes (i.e., numbered as follows) (Follower nodes); only the master coordinating node can receive the total primary frequency regulation command sent by the virtual power plant master station, so its initial deviation includes the overall target; while the initial deviation of each follower node is only negatively correlated with its own initial response.
[0014] As a preferred embodiment of the present invention, in step (1), at the initial moment of performing response detection consistency calculation, the global active response average value estimate is... and global active power error average estimate Assign values as follows:
[0015]
[0016] In the formula, For nodes The active response value, i.e., the active measurement value. With initial active power latch value difference; The primary frequency regulation command sent to the virtual power plant master station; the variables in each variable This refers to the initial moment of the calculation, i.e., the communication period. .
[0017] As a preferred embodiment of the present invention, in step (1), the grid-connected converter of each network node executes the following fixed-time consistent update equation with fine-tuning and acceleration terms:
[0018]
[0019] In the formula, The index of the period; This is for simulating step size; For the set of neighboring nodes; It is the topological adjacency coefficient, if a node Its neighboring nodes If communication exists, then ,otherwise ; To fine-tune the exponentiation of the term, ; To accelerate the exponentiation of terms ; , To fine-tune the approach gain, , To accelerate drive gain; The operator is a sign-preserving power function (i.e. This is used to safely perform nonlinear power calculations without changing the flow of information.
[0020] As a preferred embodiment of the present invention, the update equations share a common communication step size and a nonlinear power parameter; through dynamic adjustment , , , By having fine-tuning and acceleration terms play a dominant role in different stages of algorithm execution, the algorithm can ensure that it runs at a constant speed within a fixed time. Internally and globally uniformly convergent; convergence deadline The following constraint must be satisfied:
[0021]
[0022] In the formula, The algebraic connectivity of the topological Laplace matrix is determined by the topology of the nodes within the virtual power plant.
[0023] As a preferred embodiment of the present invention, for adjusting the global active power response estimate The first set of gain parameters and A symmetric time-domain allocation strategy is adopted: the time of the fine-tuning approach term and the acceleration driving term are both set to the same control margin time. To achieve smooth convergence, For adjusting the global active power error estimate The second set of gain parameters , An asymmetric time-domain allocation strategy is adopted to accelerate the time of the driving term. and fine-tuning approach term time satisfy A relatively generous start-up time is reserved to enhance instruction propagation capability and suppress physical overshoot, while the final approach time is tightened to quickly eliminate residual errors. .
[0024] As a preferred embodiment of the present invention, in step (2), the generalized availability rate is defined. The calculation formula is as follows:
[0025] ,
[0026] In the formula, Adjust the gain to the intrinsic reference. , , These are respectively real-time state variables, physical golden operating points, and normalized intervals;
[0027] For different types of heterogeneous resources connected to the virtual power plant, the relevant parameters are defined as follows: Energy storage resources: Real-time state of charge ; The optimal healthy state of charge; The difference between the maximum and minimum states of charge; wind and solar resources: This is the real-time load reduction factor; This is the median value of the safety load reduction factor; For the entire adjustment range; controllable loads such as charging piles: The extent to which the charged amount is sufficient relative to the expected target charge level; Fully meet user expectations; This is the normalized charging range.
[0028] As a preferred embodiment of the present invention, in step (2), the node In the Active power deviation correction command carried out per communication cycle for:
[0029] ;
[0030] In the formula, This refers to the initial availability of latches in a single round of optimization; It refers to virtual adjustment intensity, which ensures the balanced allocation of heterogeneous resources by gradually converging in a distributed communication network.
[0031] At the initial moment of executing the secondary error compensation consensus algorithm, for and local dynamic deviation estimates Assign values as follows:
[0032]
[0033] In the formula, For nodes The global average active power error after the first-layer response detection algorithm converges; among the variables This refers to the initial moment of the calculation, i.e., the communication period. .
[0034] As a preferred embodiment of the present invention, in step (2), each node in the 1st... The following dynamic update equation is executed in each cycle:
[0035]
[0036] In the formula, This is for simulating step size; For the set of neighboring nodes; It is the topological adjacency coefficient; For error feedback gain constant; last term As a differential dynamic input compensation term at the digital instruction level, the difference between the current step and the previous step instruction is introduced in a rolling manner to ensure that the algorithm converges absolutely at the instruction level and achieves self-balancing.
[0037] As a preferred embodiment of the present invention, in step (3), the two-layer computing mechanism is run alternately. The global response error (i.e., the absolute value of the difference between the total primary frequency modulation command and the total active power response) and the actual convergence steps of the second-layer algorithm both show a decreasing trend until the global response error is less than the set value. :
[0038]
[0039] In the formula, The primary frequency regulation command sent by the virtual power plant master station is received by the main coordinating node; For nodes The effective response.
[0040] Compared with the prior art, the advantages of the present invention are:
[0041] (1) This invention proposes a response detection consistency algorithm, which accurately measures the actual total response of a virtual power plant within a fixed time in a distributed topology without common connection points, breaking the dependence of traditional centralized control on global communication. At the same time, the global average error is synchronously distributed to each node, which greatly saves the communication time for the second-layer algorithm to recalculate the global state.
[0042] (2) This invention introduces a generalized availability rate, which comprehensively considers the physical constraints of heterogeneous resources such as energy storage SOC, wind and solar load shedding, and controllable load adequacy, and transforms them into the same dimensionless per-unit value for evaluation. This breaks down the physical barriers between resources, making it easier to uniformly call the subsequent consensus algorithm and balance the distribution of active power.
[0043] (3) This invention proposes an error compensation consistency algorithm and constructs a synchronous evolution architecture for information and physical equipment, so that the discrete algorithm iteration and physical equipment ramping are seamlessly aligned in the time domain and converge simultaneously. The introduction of instruction differential mechanism and feedforward mechanism realizes the balanced and autonomous allocation of frequency modulation active power deviation among multi-source heterogeneous resources based on generalized availability.
[0044] (4) This invention constructs a two-layer distributed control architecture for response detection and error compensation. This architecture uses a rolling time-domain mechanism of "one layer of metering and two layers of allocation" to gradually approach zero-error primary frequency modulation by adjusting the physical limit gap of the equipment through multiple rounds of alternating adaptive adjustment, while ensuring the safety of the physical boundary of heterogeneous resources. Attached Figure Description
[0045] Figure 1 This is a diagram of a two-layer consensus distributed control algorithm architecture for heterogeneous resources in a virtual power plant.
[0046] Figure 2 Flowchart of the consensus algorithm (layer 1) for response detection.
[0047] Figure 3 This is a flowchart of the error compensation consensus algorithm (second layer).
[0048] Figure 4 This is a heterogeneous resource line and communication topology for a virtual power plant based on IEEE 69 nodes. Detailed Implementation
[0049] The present invention will now be described in further detail with reference to the accompanying drawings.
[0050] Part One: Implementation Method of the Invention
[0051] To address the difficulties in total output measurement and heterogeneous resource coordination in existing technologies, this invention proposes a two-layer consensus distributed control method based on heterogeneous resources of a virtual power plant. The overall control algorithm architecture is as follows: Figure 1as shown. This method achieves differential-free primary frequency regulation through a two-layer coordination mechanism, including: a response detection consensus algorithm (the first layer), configured to realize distributed accurate metering of the total frequency regulation response of distributed resources without a physical common connection point; and an error compensation consensus algorithm (the second layer), configured to realize secondary correction based on the virtual power plant primary frequency regulation command deviation. This method can realize accurate distributed measurement of the frequency regulation response increment of the virtual power plant in the absence of a common physical connection; by introducing the generalized availability to uniformly evaluate various resource constraints and establishing a closed-loop feedback mechanism, the two-layer algorithm operates in synchronous iteration with physical equipment, so that the active power deviation of frequency regulation can be evenly and autonomously allocated, while ensuring the physical boundary safety of various resources, and finally gradually achieving differential-free primary frequency regulation.
[0052] The virtual power plant described includes distributed energy resources (DERs) without common connection points, which realize data interaction with the virtual power plant master station through centerless topology two-way communication. network nodes, including 1 master coordination node (Leader node, numbered 1) and -1 heterogeneous follower nodes (Follower nodes, numbered ). When the grid frequency continuously drops out of the dead band and exceeds the trigger time , the distributed energy resources perform droop control through their respective grid-connected converters, and latch the real-time active power output and formally start the two-layer consensus method after the droop control reaches a steady state with static error.
[0053] 1. The first-layer response detection consensus calculation: each node maintains a global active power frequency regulation response increment average estimator and a global residual active power error average estimator inside, and the accumulated frequency regulation response of the virtual power plant is metered in a distributed manner through a fixed-time consensus update equation, and the converged global average active power error is obtained.
[0054] The control iteration process of the first-layer response detection consensus algorithm according to the present invention is as Figure 2 shown, and its contents are summarized in sequence as: [Dual-core state construction] response estimation / error estimation ; [Decentralized initialization] only the Leader node receives the total command , and the Follower nodes are initialized by local measurement ; [Dual-power fixed-time driving] the fine adjustment term (0 < p < 1) eliminates steady-state static error, the acceleration term (q > 1) quickly suppresses deviation, and the deterministic time upper bound converges; [Accurate metering and cross-layer coordination] total response solution: Global average error distribution: initial value for the two-layer algorithm; [Algorithm effectiveness and core advantages] Decentralized accurate measurement / low-latency cross-layer synchronization.
[0055] First, each node runs a response detection consistency algorithm to achieve real-time distributed metering of the total output of the virtual power plant's distributed resources in the absence of physical common connection points. Each node maintains two state variables: a global active response average estimate. and global active power error average estimate This allows the algorithm to obtain accurate global response measurements when it converges, and also enables the pre-allocation of active power gaps. It eliminates the secondary communication overhead caused by the secondary algorithm in the traditional architecture for solving or waiting for the global state, and provides a critical time margin for the power ramp-up of physical devices.
[0056] At the initial moment of the response detection consensus algorithm startup, the two state variables are assigned values as follows:
[0057]
[0058] In the formula, For nodes The active response value, i.e., the active measurement value. With initial latch value difference; This is the primary frequency regulation command sent to the virtual power plant master station. The variables include... This refers to the initial moment of the calculation, i.e., the communication period. .
[0059] Since only the Leader node can receive the primary frequency regulation command sent by the virtual power plant master station, its initial deviation includes the overall target, while the initial deviations of the other Follower nodes are only negatively correlated with their own initial responses.
[0060] In the For each cycle, the grid-connected converter of each network node executes the following fixed-time consistent update equation with fine-tuning and acceleration terms:
[0061]
[0062] In the formula, This is the sequence number of the communication cycle; This is for simulating step size; For the set of neighboring nodes; It is the topological adjacency coefficient, if the node with neighbors If communication exists, then ,otherwise ; For the power of the fine-tuning term ( ), To accelerate the exponentiation of the term ( ); , To fine-tune the approach gain, , To accelerate drive gain; The operator is a sign-preserving power function (i.e. This is used to safely perform nonlinear power calculations without changing the flow of information.
[0063] The dynamic collaborative driving mechanism of fine-tuning and acceleration terms is explained in detail below: When the state deviation between neighboring nodes is large (the absolute value of the difference is greater than 1), due to the higher-order powers... The amplification effect, the acceleration term Achieving superlinear dynamic large gain drives rapid diffusion of the entire network state in the early stages of iteration; as node states converge, local deviations enter the fine-tuning range (absolute difference less than 1), and the acceleration term rapidly decays to near zero due to higher-order powers, while the fine-tuning term... Based on fractional powers It has a large differential gain when the system is close to the equilibrium point, thus forcing the residual error to be quickly cleared to zero within a fixed time.
[0064] The above update equations share the communication step size and nonlinear power parameters. This invention dynamically adjusts these parameters. , , , This allows the fine-tuning and acceleration terms to play a dominant driving role at different stages of the algorithm's execution, ensuring that the algorithm achieves its goals within a fixed time frame. Internally and globally uniformly convergent, convergence deadline The following constraint must be satisfied:
[0065]
[0066] In the formula, The algebraic connectivity of the topological Laplace matrix is determined by the topology of the nodes within the virtual power plant.
[0067] , , , The specific method for determining the value is as follows:
[0068] For adjusting the global active response estimate The first set of gain parameters , Since the initial measurement state evolves relatively smoothly in the network, a symmetric time-domain allocation strategy is adopted, setting the time of both the fine-tuning approach term and the acceleration driving term to the same control margin time. ( To achieve smooth convergence, a specific example of the symmetric time-domain allocation strategy is as follows:
[0069]
[0070] For adjusting the global active power error estimate The second set of gain parameters , Because it contains the total primary frequency modulation command received only by the Leader node. Therefore, an asymmetric time-domain allocation strategy is adopted to accelerate the time of the driving term. and fine-tuning approach term time satisfy A relatively generous start-up time is reserved to enhance instruction propagation capability and suppress physical overshoot, while the final approach time is tightened to quickly eliminate residual errors. A specific example of an asymmetric time-domain allocation strategy is as follows:
[0071]
[0072] The convergence condition for the response detection consistency algorithm is defined as follows:
[0073]
[0074] That is, the deviation of the global active power response average estimate and the deviation of the global active power error average estimate between two adjacent iterations of all nodes in the network are both less than the preset threshold. , At this point, we have:
[0075]
[0076] That is, the response estimates of all nodes. All converge to the global active response average, and the error estimate is... All converge to the global average active power error. The Leader node can be determined by the convergence value. The actual total response power of the entire virtual power plant can be directly calculated. and the convergence value The initial deviation value is injected into the second-layer algorithm to complete the handover between the two layers:
[0077]
[0078] At this point, the error compensation consensus algorithm is activated, and each node adjusts its active power command based on the deviation.
[0079] 2. Second-layer error compensation consensus algorithm: Define generalized availability to quantitatively evaluate the relative adjustment capability of heterogeneous resources, maintain virtual adjustment intensity and local dynamic deviation estimates within each network node, and output active power deviation correction instructions through consensus update equation to drive heterogeneous distributed energy resources to synchronously perform power ramp-up response during algorithm iteration.
[0080] The control iteration process of the second-level error compensation consensus algorithm described in this invention is as follows: Figure 3 As shown, its content can be summarized as follows: [Dual-core feedback construction] Adjusting intensity / Dynamic bias; [Multi-source heterogeneous per-unit standardization] Normalization and quantification of heterogeneous resource availability; [Seamless error closed-loop injection] Global average error with first-layer convergence. Direct injection, with an initial adjustment strength of 0 for smooth entry; [Dual-core update and self-balancing mechanism] Gradual convergence ensures "equitable distribution based on energy". Dynamic zeroing lock for "zero-difference response across the entire network," introducing differential compensation to achieve dynamic command feedback; [Balanced autonomous command allocation] Decentralized autonomous output based on available energy:
Algorithm Performance and Core Advantages
[0081] To overcome the strong heterogeneity and constraint uncertainties of various distributed energy resources (DERs), a generalized availability rate is introduced. This is used to quantitatively assess its relative regulatory capability in participating in primary frequency modulation and secondary correction:
[0082]
[0083] In the formula, The inherent baseline adjustment gain is determined by the resource physical hardware response bandwidth, ramp rate, and economic cost. , , These are the real-time state quantity, the physical golden operating point, and the normalized interval, respectively.
[0084] This study addresses three typical heterogeneous resources within a virtual power plant.
[0085] The specific definitions of the above parameters are as follows:
[0086] (1) Energy storage resources
[0087] Represents the real-time state of charge; Represents the preset optimal healthy state of charge (with maximum margin in both charging and discharging); This represents the safe operating range.
[0088] (2) Scenic resources
[0089] This is the real-time load reduction factor; This represents the median value of the safety reduction factor; This is the full adjustment range.
[0090] (3) Controllable loads of charging piles
[0091] The extent to which the charged amount is sufficient relative to the expected target charge level; Fully meet user expectations; This is the normalized charging range.
[0092] Based on generalized availability, a virtual adjustment intensity is introduced. and local dynamic deviation estimates Two state variables are used to construct an "allocation-feedback" collaborative update architecture. To ensure the stability of the allocation baseline, the generalized availability rate is latched as an initial value in each round of optimization. .node In the Active power deviation correction command carried out per communication cycle for:
[0093]
[0094] In the formula, This refers to virtual regulation intensity, which guides active power deviation correction commands through gradual convergence in a distributed communication network. The output ensures that heterogeneous resources are allocated in a balanced manner according to energy; through... Dynamic zeroing locks in the gradual elimination of active power deviation across the entire network.
[0095] At the initial moment of algorithm startup (communication cycle) The initial values for the state variables of each node are assigned as follows:
[0096]
[0097] In the formula, For nodes The global average active power error after the first-layer response detection algorithm converges; among the variables This refers to the initial moment of the calculation, i.e., the communication period. .
[0098] In the In each cycle, each node executes the following dynamic update equation:
[0099]
[0100] In the formula, This is for simulating step size; For the set of neighboring nodes; It is the topological adjacency coefficient; This is the error feedback gain constant. (Last term) As a differential dynamic input compensation term at the digital instruction level, the difference between the current step and the previous step instruction is introduced in a rolling manner to ensure that the algorithm converges absolutely at the instruction level and achieves self-balancing.
[0101] The convergence condition for the error-compensated consensus algorithm is defined as follows:
[0102]
[0103] That is, the deviation of virtual adjustment intensity and the estimated deviation of local dynamic deviation between two adjacent iterations of each node in the entire network are both less than the preset threshold. , At this point, we have:
[0104]
[0105] That is, the virtual adjustment intensity of all nodes in the network reaches global consistency, and the active power deviation is autonomously distributed according to energy balance.
[0106] During the operation of the error compensation consensus algorithm, the virtual regulation intensity and local dynamic deviation state variables of each node evolve synchronously with the active power output of the physical equipment. As the two-layer algorithm gradually eliminates the remaining frequency regulation gap, the actual output of the nodes synchronously climbs to a new steady-state value.
[0107] 3. By alternately operating the above two-level computer mechanism, multi-stage rolling adjustment is achieved until the global remaining frequency regulation error is lower than the preset threshold, so as to realize the virtual power plant's response to the primary frequency regulation command in a gradual manner.
[0108] After the error compensation consensus algorithm completes its execution, the system automatically restarts the response detection consensus algorithm for the next round of the two-layer consensus algorithm. This process is repeated until error-free primary frequency regulation is gradually achieved, meaning the global response error (the absolute value of the difference between the total primary frequency regulation command and the total active power response) is less than the set value. :
[0109]
[0110] In the formula, The primary frequency regulation command received by the Leader node from the virtual power plant master station; For nodes The effective response.
[0111] As the rolling iteration process progresses, the global response error decreases round by round, and the actual convergence steps of the second-layer algorithm show a decreasing trend. The computation time of subsequent rounds will be significantly shorter than that of the first round. Multiple rounds of dual-layer collaborative optimization can be completed within the golden response time of one frequency modulation, achieving rapid and gradual elimination of the frequency modulation gap.
[0112] Part Two: A Specific Example
[0113] like Figure 4 As shown, this embodiment constructs a distributed collaborative control architecture for a virtual power plant based on an improved IEEE 69-node distribution network system. The system aggregates... A distributed energy resource (DER) without a physical common connection point consists of 1 leader node and 19 follower nodes, including 6 photovoltaic nodes, 4 wind power nodes, 4 energy storage nodes, and 6 controllable load nodes for charging piles. The arrangement of logical communication numbers (01 to 20) does not depend on the geographical order of the physical nodes, but is arranged according to the master-slave attribute of the resource and the ease of closing the ring communication link.
[0114] like Figure 4 As shown in the "Leader-Follower Communication Topology" in the lower right corner, 20 DER nodes exchange data through bidirectional communication links. The main topology forms a closed bidirectional ring network of 01-02-03-...-20-01. To improve information convergence speed, the system is configured with sparse network links across nodes, including bidirectional direct communication between nodes 01 and 11, 03 and 13, 06 and 16, and 08 and 18. In the eigenvalue spectrum of the entire network Laplacian matrix constructed accordingly, the second smallest eigenvalue (algebraic connectivity) is solved to obtain an exact analytical value:
[0115]
[0116] The topological adjacency coefficient matrix based on this example Defined as:
[0117] a. Basic ring link: for any ,have (Among them, 20's neighbor is 1).
[0118] b. Relay shortcut link: Configure the adjacency coefficient of the following 4 pairs of specific relay nodes to 1:
[0119] ; ; ; .
[0120] c. Other elements: In addition to the links and diagonal elements specified above. Apart from that, all other communication adjacency coefficients are 0.
[0121] To address the aforementioned structure, this invention provides a two-layer consensus distributed control algorithm based on heterogeneous resources of a virtual power plant. This two-layer distributed control algorithm specifically includes the following steps:
[0122] 1. First-layer response detection consistency algorithm (distributed precise metering)
[0123] The goal of the first-layer algorithm is to calculate the actual total response of the virtual power plant and distribute the global active power error in a topology without a center or common physical connection points.
[0124] (1) Initialization of state variables (communication cycle) )
[0125] Each node maintains an estimate of the global response average. and global error average estimate Its initial state is assigned the following values:
[0126]
[0127] In the formula, For nodes The active response value, i.e., the active measurement value. With initial latch value difference; This is the primary frequency regulation command sent to the virtual power plant master station. Since only the Leader node can receive the primary frequency regulation command, its initial deviation includes the overall target, while the initial deviations of the other Follower nodes are only negatively correlated with their own initial responses.
[0128] (2) Fixed-time consistent iterative update
[0129] In the Within each communication cycle, each node executes a nonlinear update equation with fine-tuning and acceleration terms:
[0130]
[0131] In the formula, The simulation step size is set to 0.005s; For the power of the fine-tuning term (take) In this example, we take 0.6 to ensure that the residual error is quickly cleared to zero when the state variables are close to being consistent. To accelerate the term to the power (take) In this example, we take 1.4 to achieve a fast startup response under large instruction mutations in the early stages of disturbance. , To fine-tune the approach gain, , To accelerate drive gain; The operator is a sign-preserving power function (i.e. ).
[0132] , , , The method for determining the value is as follows:
[0133] To ensure the fast and error-free control of the two-level consensus algorithm, the time of the first-level response detection consensus algorithm is reduced. The time should be controlled within 0.2 seconds to allow more time for the physical equipment to climb the slope.
[0134]
[0135] for , Since there are already distributed initial measurement values globally, the consistency process is relatively smooth. The fine-tuning approach term time and the acceleration driving term time are both set to 0.1s. , The calculation method is as follows:
[0136]
[0137] for , The virtual power plant master station sends the primary frequency regulation command. Received only by the Leader node, an asymmetric time allocation is used to ensure the spike signal spreads quickly and smoothly throughout the discrete network. The acceleration driver is set to a more generous 0.13s, while the fine-tuning approach term is tightened to 0.07s. , The calculation method is as follows:
[0138]
[0139] Substituting the calculated values into the consistency equation will enable the algorithm to converge uniformly within 0.2 seconds.
[0140] (3) State convergence and transition
[0141] After convergence, the estimates of all nodes reach a consensus, i.e. , At this point, the Leader node can calculate the actual total response of the entire virtual power plant:
[0142]
[0143] Each node will calculate the current global average active power error. As a control input, it is injected into the second-layer algorithm.
[0144] 2. Second-level error compensation consistency algorithm (autonomous balancing allocation of active power deviation)
[0145] (1) Generalized availability of heterogeneous resources calculate
[0146] Before running the sub-layer error allocation, each node calculates its current generalized availability based on its own highly heterogeneous resource characteristics. This is used to quantify its relative adjustment capability and physical constraints.
[0147]
[0148] For the three typical heterogeneous resources in this embodiment, the specific operational constraint parameters are defined as follows:
[0149] a. Energy storage resources (such as DER 01,07,11,17):
[0150] Real-time state of charge; The optimal healthy operating point; normalized interval .
[0151] b. Scenic resources (such as DER03,04,05,06,09,10,13,15,16,20):
[0152] This is the real-time load reduction factor; (i.e., the median value of the safety load reduction factor); full adjustment range .
[0153] c. Controllable loads such as charging piles (e.g., DER 02,08,12,14,18,19):
[0154] The charge level is sufficient; Fully meet user expectations; normalized interval .
[0155] To ensure the stability of the allocation baseline in a single round of optimization, the generalized availability rate is latched at its initial value when the next level of optimization begins. .
[0156] (2) The secondary algorithm establishes a closed-loop feedback loop at the digital instruction level based on the error issued by the first layer and the availability of each node.
[0157] (2.1) Initialization of state variables (communication cycle) )
[0158] Each node maintains virtual adjustment intensity and local dynamic deviation estimates The initial values are assigned as follows:
[0159]
[0160] The active power deviation correction command undertaken by each node is dynamically represented as follows:
[0161] .
[0162] (2.2) Iteration of dynamic update equations:
[0163] In the In each cycle, each node executes the following dynamic update equation:
[0164]
[0165] In the formula, The simulation step size is set to 0.005s; This is the error feedback gain constant.
[0166] Last item The differential dynamic input compensation term at the digital instruction level is introduced by rolling the difference between the current step and the previous step instruction to ensure that the algorithm converges absolutely at the instruction level and achieves self-balancing.
[0167] Error feedback gain constant The specific method for determining the value is as follows:
[0168] Since the secondary algorithm operates synchronously with the physical layer output, and considering the ramp-up speed of heterogeneous resources, the overall time is constrained to 1.1 seconds, ensuring at least two rounds of the two-layer consensus algorithm can be performed within the first 3 seconds of the primary frequency regulation of the power grid. With a simulation step size of 0.005 seconds, the maximum number of iterations for the secondary algorithm is 220. Secondary feedback gain. The assignment to the bidirectional formula constraint of information-matter stability:
[0169]
[0170] Right now The system must satisfy the stability eigenvalue constraints of the state matrix to prevent system divergence and physical overshoot, while the estimated local dynamic deviation of the entire network must asymptotically decay to the zero domain (convergence > 99%) within a finite 220 steps. The benchmark for the average generalized availability of heterogeneous resources across the entire network in this example is... Taking 0.45, the final result of this example is... The value is 2.5.
[0171] By substituting the above calculated values into the consistency algorithm, seamless alignment of the digital iteration step size of power ramping and the physical ramping rate of heterogeneous physical devices across the network on the time axis can be achieved.
[0172] (2.3) Control command output:
[0173] When the sub-algorithm tends to converge, the virtual adjustment intensity is... To achieve global consistency, each node outputs active power deviation correction commands. Strictly proportional to its own generalized availability It achieves balanced adaptive allocation based on heterogeneous adjustment capabilities, and the physical layer ramp-up is in place.
[0174] 3. Multi-stage rolling time-domain alternating adjustment
[0175] After a round of two-level consensus distributed control algorithm is completed, the system re-enters steady state. However, due to real-time changes in the status of each resource or physical overruns, there may still be a slight adjustment gap in the system.
[0176] At this point, the system automatically triggers the first-layer response detection consistency algorithm from step one again, re-accurately measuring the total output of distributed resources. The two-layer architecture employs a multi-stage, rolling alternating mechanism of "one-layer metering, two-layer allocation," ensuring that energy storage does not exceed the safe SoC boundary, wind and solar power do not exceed maximum output, and charging piles meet user needs, ultimately progressively achieving the overall primary frequency regulation command. Fast and error-free response.
[0177] In summary, the proposed two-layer consistent distributed control method based on heterogeneous resources of a virtual power plant achieves fixed-time accurate distributed metering of the actual total output of dispersed resources in a topology without physical common connection points or a central hub; dimensionless generalized availability rate standardization comprehensive evaluation of multi-source strong heterogeneous resource regulation capabilities and physical constraints; and balanced autonomous allocation of frequency regulation active power deviation based on availability rate. Furthermore, by introducing a parallel evolution architecture of response and error dual-state variables in the first layer, it solves the coupling problem of centralized secondary communication delay and bandwidth consumption caused by the second-layer algorithm in traditional two-layer distributed control in order to solve or wait for the global state. It realizes local pre-embedding of error information, zero-delay injection across layers, and smooth entry. Under the premise of ensuring the safety of diverse physical boundaries such as the safe operating range of energy storage SoC, the depth of wind and solar load reduction, and the adequacy of charging piles, it ultimately achieves efficient, fast, and error-free distributed response to the primary frequency regulation command of the power grid.
[0178] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.
Claims
1. A two-layer consensus distributed control method based on heterogeneous resources of a virtual power plant, wherein the virtual power plant includes... A distributed energy resource without a physical common connection point, where each network node and the virtual power plant master station achieve data interaction through bidirectional communication in a decentralized topology; characterized in that... When the grid frequency drops out of the dead zone, distributed energy resources execute droop control through their respective grid-connected converters to enter an initial steady state with static errors; subsequently, two-layer consensus distributed control is initiated, specifically including the following steps: (1) Consistency calculation of the first layer's response detection: Within each network node, the global average active frequency regulation response increment and the global average residual active power error are maintained. The frequency regulation response is accumulated by the distributed metering virtual power plant through the fixed-time consistency update equation, and the converged global average active power error is obtained. The grid-connected converter of each network node executes the following fixed-time consistent update equation with fine-tuning and acceleration terms: ; In the formula, For nodes In the The average estimate of the global active response over one communication cycle. For nodes After the first After the first communication cycle update, in the... The average estimate of the global active response over one communication cycle. For nodes neighboring nodes In the The average estimate of the global active response over a communication cycle; For nodes In the The estimated average global active power error over one communication cycle. For nodes After the first After the first communication cycle update, in the... The estimated average global active power error over one communication cycle. For nodes neighboring nodes In the The estimated average global active power error over a communication cycle; This is the sequence number of the communication cycle; This is for simulating step size; For the set of neighboring nodes; It is the topological adjacency coefficient, if a node Its neighboring nodes If communication exists, then ,otherwise ; To fine-tune the exponentiation of the term, ; To accelerate the exponentiation of terms, ; , To fine-tune the approach gain, , To accelerate drive gain; The operator is a sign-preserving power function, which is used to safely perform nonlinear power calculations without changing the information flow. (2) Error compensation consensus algorithm for the next level: A generalized availability rate is defined to quantitatively evaluate the relative adjustment capability of heterogeneous resources. Virtual adjustment intensity and local dynamic deviation estimates are maintained within each network node. Active power deviation correction instructions are output through the consistency update equation to drive the heterogeneous distributed energy resources to synchronously perform power ramp-up response during the algorithm iteration process. Among them, each network node in the th Each cycle executes the following dynamic consistency update equation for error compensation: ; In the formula, For nodes In the Virtual adjustment intensity per communication cycle, For nodes After the first After the first communication cycle update, in the... Virtual adjustment intensity per communication cycle, For nodes neighboring nodes In the Virtual adjustment intensity per communication cycle; For nodes In the Local dynamic deviation estimate for each communication cycle For nodes After the first After the first communication cycle update, in the... Local dynamic deviation estimate for each communication cycle For nodes neighboring nodes In the Local dynamic deviation estimate for each communication cycle; This is for simulating step size; For the set of neighboring nodes; It is the topological adjacency coefficient; For error feedback gain constant; last term As a differential dynamic input compensation term at the digital instruction level, the difference between the current step and the previous step instruction is introduced in a rolling manner to ensure that the algorithm converges absolutely at the instruction level and achieves self-balancing. (3) Multi-stage rolling adjustment is achieved by alternating operation of the above two-layer computer mechanism until the global remaining frequency regulation error is lower than the preset threshold, so as to realize the virtual power plant's response to the primary frequency regulation command in a gradual manner.
2. The method according to claim 1, characterized in that, In use for communication between virtual power plants and the distributed energy resources The network consists of one master coordinating node and... There are several heterogeneous follower nodes; only the master coordinating node can receive the overall primary frequency regulation command sent by the virtual power plant master station, so its initial deviation includes the overall target; while the initial deviation of each follower node is only negatively correlated with its own initial response.
3. The method according to claim 1, characterized in that, In step (1), at the initial moment of the response detection consistency calculation, the global active response average value estimate is calculated. and global active power error average estimate Assign values as follows: ; In the formula, For nodes The active response value, i.e., the active measurement value. With initial active power latch value difference; The primary frequency regulation command sent to the virtual power plant master station; the variables in each variable This refers to the initial moment of the calculation, i.e., the communication period. .
4. The method according to claim 1, characterized in that, The update equations share a common communication step size and a nonlinear power parameter; through dynamic adjustment... , , , By having fine-tuning and acceleration terms play a dominant role in different stages of algorithm execution, the algorithm can ensure that it runs at a constant speed within a fixed time. Internally and globally uniformly convergent; convergence deadline The following constraint must be satisfied: ; In the formula, The algebraic connectivity of the topological Laplace matrix is determined by the topology of the nodes within the virtual power plant.
5. The method according to claim 1, characterized in that, For adjusting the global active response estimate The first set of gain parameters and A symmetric time-domain allocation strategy is adopted: the time of the fine-tuning approach term and the acceleration driving term are both set to the same control margin time. To achieve smooth convergence, ; For adjusting the global active power error estimate The second set of gain parameters , An asymmetric time-domain allocation strategy is adopted: to make the acceleration driving term time and fine-tuning approach term time satisfy A relatively generous start-up time is reserved to enhance instruction propagation capability and suppress physical overshoot, while the final approach time is tightened to quickly eliminate residual errors. .
6. The method according to claim 1, characterized in that, In step (2), the time-varying generalized availability rate is defined. The calculation formula is as follows: , ; In the formula, Adjust the gain to the intrinsic reference. , , These are respectively real-time state variables, physical golden operating points, and normalized intervals; For different types of heterogeneous resources connected to the virtual power plant, the relevant parameters are defined as follows: Energy storage resources: Real-time state of charge ; The optimal healthy state of charge; The difference between the maximum and minimum states of charge; wind and solar resources: This is the real-time load reduction factor; This is the median value of the safety load reduction factor; For the entire adjustment range; controllable loads such as charging piles: The extent to which the charged amount is sufficient relative to the expected target charge level; Fully meet user expectations; This is the normalized charging range.
7. The method according to claim 1, characterized in that, In step (2), the node In the Active power deviation correction command carried out per communication cycle for: ; In the formula, This refers to the initial availability of latches in a single round of optimization; It refers to virtual adjustment intensity, which ensures the balanced allocation of heterogeneous resources by gradually converging in a distributed communication network. At the initial moment of executing the secondary error compensation consensus algorithm, for and local dynamic deviation estimates Assign values as follows: ; In the formula, For nodes The global average active power error after the first-layer response detection algorithm converges; among the variables This refers to the initial moment of the calculation, i.e., the communication period. .
8. The method according to claim 1, characterized in that, In step (3), the two-layer computing mechanism is run alternately. The global response error and the actual number of convergence steps of the second-layer algorithm both show a decreasing trend until the global response error is less than the set value. : ; In the formula, The primary frequency regulation command sent by the virtual power plant master station is received by the main coordinating node; For nodes The effective response.
Citation Information
Patent Citations
Frequency adjustment method and device using active support of virtual power plant, and storage medium
CN117674197A
Optimized scheduling method for multi-resource type virtual power plant
CN122292357A
