Distributed new energy networking type unit current and voltage coordinated regulation and control method based on multi-layer coordination

Abstract

The invention discloses a distributed new energy networking type unit current and voltage coordinated regulation and control method based on multilayer coordination. The method comprises the steps that a power reference value is transmitted to a global scheduling layer in real time; the multi-objective optimization function is issued to the regional coordination layer in real time for execution; receiving the global power to ensure that the global power converges preferentially in regional coordination; each node enables the state to gradually correspond to a complete bureau target based on the own state; the local control layer adopts self-adaptive droop control; according to the method, the problems of inertia deficiency and insufficient virtual inertia response are solved, and the voltage support strength and the operation reliability of a power system are greatly improved; the problem of limited voltage supporting capability is solved; the problem that distributed control lacks global coordination is solved; the problem of high-frequency current fluctuation is solved, the resistance of the system to the high-frequency current fluctuation is greatly improved, and equipment safety and stable operation of a power grid are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of power system automation control, and more specifically, to a method for coordinated current and voltage regulation of distributed renewable energy grid-connected generating units based on multi-layer coordination. Background Technology

[0002] With the acceleration of the global energy transition, new energy technologies, represented by grid-connected wind turbines and distributed photovoltaics, are developing rapidly. Their installed capacity in the power system has exceeded 30%, and this proportion continues to rise under the impetus of my country's "dual carbon" target.

[0003] In terms of system support capabilities, the stability of traditional power systems relies on the physical inertia and excitation regulation capabilities of synchronous generators. These units can provide a natural buffer when there is an active power imbalance through rotor kinetic energy storage. When there is an active power deficit in the grid, the generator rotor releases kinetic energy to delay the current drop; when there is an active power surplus, the rotor absorbs energy to suppress the current surge. However, new energy units are connected to the grid through power electronic converters and have no physical inertia. Their control logic is mainly based on tracking active power commands and lacks the damping response characteristics of synchronous generators. Moreover, although grid-connected wind turbines have the potential for virtual inertia regulation, the virtual parameters in existing technologies are mostly configured with fixed values, which cannot adapt to complex operating conditions.

[0004] In terms of voltage support, traditional synchronous generators can directly output reactive power to regulate node voltage through the excitation system, and can provide strong support of 1.5-2.0 times the rated current during short-circuit faults; however, the voltage control of new energy units relies on the "grid voltage tracking" mode, and their reactive power output is limited by the converter capacity and control strategy. When a serious fault such as a three-phase short circuit occurs in the grid, if the voltage drops below 0.2 per unit value for more than 100ms, most new energy units will trigger low voltage protection and disconnect from the grid.

[0005] At the control and coordination level, traditional centralized dispatching adopts a "data acquisition-central decision-command issuance" model, with communication delays generally exceeding 150ms, making it difficult to cope with the second-level power fluctuations of grid-connected wind turbine clusters. Meanwhile, purely distributed control lacks a global optimization mechanism, easily leading to power offsetting between regions. Furthermore, the high-frequency switching characteristics of power electronic equipment combined with the low inertia characteristics of the system can cause drastic current fluctuations, threatening the insulation life of the equipment. Traditional control methods have many shortcomings when dealing with grid-connected renewable energy:

[0006] (1) Lack of inertia and insufficient virtual inertia response: The stability of traditional power systems relies on the physical inertia of synchronous generators to provide buffering, while new energy units use power electronic converters for grid connection, lacking physical inertia and inertial response. Although grid-connected wind turbines can simulate the response of synchronous machines through virtual inertia adjustment, most of the virtual inertia parameters in the existing technology are fixed values ​​and cannot be dynamically adjusted according to the grid status. Such fixed parameter settings cannot effectively cope with the rapid fluctuations of the system under extreme conditions, resulting in poor suppression of current fluctuations by wind turbines, and even causing secondary fluctuations, affecting the stability of the grid.

[0007] (2) Limited voltage support capability: Traditional synchronous generators output reactive power through the excitation system, which can quickly respond to faults such as grid short circuits and provide strong voltage support. In contrast, the voltage control of new energy units mainly relies on the "grid voltage tracking" mode, which is limited by the capacity of the converter and the control strategy. When a large-scale grid fault occurs, the voltage recovery capability of new energy units is weak. A continuous voltage drop may trigger undervoltage protection and lead to grid disconnection, prolonging the grid recovery time and reducing grid reliability.

[0008] (3) Distributed control lacks global coordination: The centralized dispatching system in traditional power systems relies on data acquisition and central decision-making to issue instructions. However, when a high proportion of new energy sources are connected to the grid, distributed control methods often lack a global optimization mechanism and cannot effectively coordinate the working status of various new energy equipment. For example, wind turbine clusters may experience power offsetting between regions due to conflicts in local voltage regulation targets, resulting in reverse power flow fluctuations in the power grid; such power fluctuations not only affect the stability of the power grid, but may also lead to circuit breaker malfunctions and equipment damage.

[0009] (4) High-frequency current fluctuation problem: The combination of the high-frequency switching characteristics of power electronic equipment and the low inertia characteristics of the power system may cause high-frequency current fluctuations in a wide frequency band. These high-frequency current fluctuations pose a serious threat to the insulation system of power equipment and the operational stability of the power grid. Actual measurement data from some new energy collection stations show that the amplitude of such high-frequency current fluctuations can reach 30% of the rated current, which greatly shortens the service life of the equipment and is not easy to suppress through traditional control methods.

[0010] Therefore, it is essential to design a method for coordinated current and voltage regulation of distributed renewable energy grid-connected units based on multi-layer coordination. Summary of the Invention

[0011] To address the problems existing in the prior art, the present invention aims to provide a multi-layer coordinated distributed new energy grid-connected unit current and voltage collaborative control method. This method addresses the issues raised in the background art, such as insufficient inertia and virtual inertia response leading to poor suppression of current fluctuations by wind turbines, which may even trigger secondary fluctuations and affect grid stability; limited voltage support capacity resulting in weak voltage recovery capability of new energy units, with continuous voltage drops potentially triggering undervoltage protection and causing grid disconnection, prolonging grid recovery time and reducing grid reliability; lack of global coordination in distributed control leading to power offsetting between regions, causing reverse power flow fluctuations and equipment damage; and high-frequency current fluctuations significantly shortening equipment lifespan and being difficult to suppress using traditional control methods.

[0012] To solve the above problems, the present invention adopts the following technical solution.

[0013] A multi-layered, coordinated current and voltage control method for distributed renewable energy grid-connected generating units includes:

[0014] The global scheduling layer uses evolutionary gradient regularized reinforcement learning to calculate the power reference value of each region based on the real-time operation data of distributed new energy units, environment and power grid.

[0015] A multi-objective optimization function is constructed, with current deviation, node voltage deviation, wind turbine output deviation and energy storage SOC deviation as optimization indicators. The voltage weight is optimized, and finally, power allocation instructions for each region are generated. The cycle of wind turbine regions with grid structure is shortened and sent to the regional coordination layer for execution in real time.

[0016] After receiving the global power reference value, the regional coordination layer uses a weighted dynamic consensus algorithm to achieve coordinated control of current and voltage. It defines node state variables (current and voltage amplitude), adjacency matrix, and weight matrix. The weight of grid-type units is higher than that of conventional units to ensure that they converge first in regional coordination.

[0017] Each node iteratively updates its state based on its own state, the states of its neighbors, and global reference instructions, introducing convergence acceleration terms and constraint coefficients to gradually align the state with the global objective.

[0018] The local control layer implements a differentiated control strategy. Conventional units adopt adaptive droop control. Small deviations (≤2%) weaken the droop coefficient to suppress overshoot, while large deviations restore the initial coefficient and enhance regulation capability.

[0019] Furthermore, firstly, hierarchical control initialization is performed on distributed renewable energy units, collecting key parameters such as real-time current, voltage, unit type, and control mode to establish a characteristic parameter matrix. Unit type is used to distinguish control priority: Level 1 is grid-connected wind turbines, Level 2 is conventional wind power, and Level 3 is photovoltaic power generation. Control mode is used to distinguish droop or VSM operating states: 0 = standby, 1 = droop, 2 = VSM. To achieve uniformity of voltage dimensions for different units, 10kV is used as the reference voltage, and the unit voltage is per-unit standardized using the following function expression:

[0020]

[0021]

[0022] in This is the real-time voltage of the generator unit. To standardize voltage and provide standardized input for subsequent voltage stability judgment and control algorithms, hierarchical control initialization clarifies the unit control priority and operating mode, establishes a characteristic parameter matrix, and clarifies the control priority of grid-connected wind turbines, etc. Voltage standardization unifies the dimensions, providing standardized input for subsequent voltage stability judgment and control algorithms, and improving the accuracy and consistency of control.

[0023] Furthermore, the global scheduling layer, as the top-level control unit, constructs a multi-objective optimization function based on standardized data:

[0024]

[0025] in This represents the maximum current deviation at the node. This represents the maximum voltage deviation at the node. The total output deviation of the grid-type wind turbine; This refers to the state of charge deviation of energy storage.

[0026] Weights satisfy Optimization is achieved using evolutionary gradient regularization meta-reinforcement learning;

[0027] Inner loop: Perturb the policy parameters to generate p samples, calculate the reward value and estimate the direction of the evolution gradient to quickly adapt to the current working conditions;

[0028] Outer loop: Aggregates gradient updates of meta-parameters across multiple scenarios, enabling the strategy to generalize across scenarios;

[0029] Finally, active / reactive power reference values ​​covering the entire region are generated every 1 to 3 minutes and distributed to the regional coordination layer to optimize multiple objectives. The global scheduling layer coordinates key deviations such as current and voltage through multi-objective optimization functions, combined with evolutionary gradient regularization reinforcement learning. The inner loop quickly adapts to the operating conditions, while the outer loop enhances the cross-scenario generalization ability. Power reference values ​​for the entire region are generated periodically to provide accurate basis for regional coordination and improve the adaptability of regulation and global synergy.

[0030] Furthermore, the regional coordination layer models the N units in this region as a weighted graph network, with the communication matrix satisfying symmetry and normalization properties, and its functional expression is:

[0031]

[0032] in, Let i be the degree of node i. The communication delay between nodes i and j. This is an attenuation factor that ensures the weight of links with poor communication quality is automatically reduced.

[0033] Regional coordinated regulation based on a weighted adaptive dynamic consensus algorithm:

[0034]

[0035] in, For the first After the nth iteration, the th Updated status values ​​of the desktop unit; For the first During the nth iteration, the 1st The current real-time status value of the unit; This represents the number of iterations. Assign unit number; This is the correction factor for the neighbor collaboration term; For the first Neighbor node set of the Taiwan machine group Perform summation; For the first Taiwanese crew and neighboring crew Communication weights between them; For the first Taiwanese crew and neighboring crew The regulatory weights between them; For the first In the next iteration, the neighboring units The current real-time status value; This is the correction factor for the global constraint term; For the first During the nth iteration, the 1st Global constraints on the Taiwan unit; This is the correction coefficient for the convergence acceleration term; For the first During the nth iteration, the 1st Global reference status values ​​of the unit;

[0036] The weight of the network-type nodes is set to 1.8, and that of the conventional nodes is set to 1.0. γ is the adaptive step size, ensuring that the iteration convergence is ≤25 steps and the regional voltage deviation is ≤±3%. A three-level response is triggered when the voltage deviation exceeds the limit.

[0037] 3%~5%: Increase virtual inertia to mitigate fluctuations;

[0038] 5%~8%: Simultaneously utilize the regional energy storage system to compensate for power;

[0039] 8%: Rapidly restore voltage by removing ≤5% of non-critical loads;

[0040] The regional coordination layer uses a weighted graph network model, with communication weights dynamically adjusted according to quality, and combines adaptive algorithms to achieve coordinated control of generating units; network-type generating units converge first, iterate quickly and have small voltage deviations, and the three-level response mechanism effectively responds to over-limit situations and ensures regional stability.

[0041] Furthermore, the local control layer implements the regional coordination results at the individual unit execution level. Conventional wind power and photovoltaic units adopt adaptive droop control, with the function expression as follows:

[0042]

[0043]

[0044] in, For the current regulation deviation of the unit; This is the current reference value; : is the active current droop factor; This represents the real-time active power output value of the generator unit. This is a reference value for the active power of the generator unit; This is the attenuation correction term for current regulation; This is the target voltage value after unit adjustment; This is the voltage reference value; The reactive power-voltage droop factor; This refers to the real-time reactive power output value of the generator set. This is a reference value for the reactive power of the generator unit; This is the absolute value of the voltage deviation; This is the attenuation correction term for voltage regulation;

[0045] The expression for the droop coefficient function is as follows:

[0046]

[0047] in, The active-current droop factor is the same as the previous text. Consistent; The reactive power-voltage droop factor; This refers to the rated active power of the generator unit. This refers to the rated reactive power of the generator unit.

[0048] The exponential term is used to suppress over-adjustment under small deviations (e.g., the adjustment intensity is automatically reduced by 30% when the current deviation is <0.05 per unit). The grid-type wind turbine introduces virtual inertial control as shown in the function expression below:

[0049]

[0050] in, This represents the unit's real-time virtual inertia value. This serves as the initial virtual inertia reference value for the unit. This is the virtual inertia adjustment coefficient; This is the absolute value of the rate of change of current;

[0051] A hybrid control system combining droop and virtual synchronous machines is employed, where the electromagnetic power function expression for transient mode is:

[0052]

[0053] in, This refers to the output electromagnetic power of a grid-type wind turbine in VSM mode. For the virtual electromotive force in VSM mode; The real-time voltage of the power grid at the unit's connection point; Virtual synchronization reactance in VSM mode; virtual electromotive force With grid voltage The phase difference between them; For the real-time angular velocity of the grid-type wind turbine; The rated angular velocity of the unit; This is the power angle ratio adjustment coefficient; The rated power angle of the generator unit;

[0054] Steady-state mode: The droop coefficient is 1 / 3 of that of conventional units, while also considering regulation accuracy; during mode switching, an exponential decay transition strategy, as shown in the function expression below, is used to ensure power fluctuations are ≤3%, and the function expression is:

[0055]

[0056] in: For grid-type wind turbines during mode switching, at a certain moment Real-time control output; At the same time Below, the control output of the grid-type wind turbine in VSM mode; At the same time Below, the control output of the grid-type fan in droop mode; The current moment; This is the start time of the mode switch; Transition time for mode switching; The attenuation coefficient; For the dynamic weighting coefficients of the VSM mode control output; The dynamic weighting coefficients for the output of the droop mode control are used. The local control layer implements control commands through differentiated control: the adaptive droop control of conventional units suppresses excessive adjustment with small deviations, and the grid-type wind turbines adopt droop-VSM hybrid control and introduce virtual inertia. The mode switching is smooth and the power fluctuation is ≤3%, which effectively enhances the system stability and provides accurate and stable underlying support for global and regional control.

[0057] Furthermore, evolutionary gradient-regularized meta-reinforcement learning is employed, based on historical operational data covering three full years of seasonal operating conditions, extreme weather, and load fluctuation scenarios, to construct a meta-reinforcement learning task set. Each task It includes input features such as wind speed, illumination, and load curves, and output targets, and initializes meta-parameters. and inner loop learning rate , outer loop learning rate Set the number of disturbance samples p=30 and the disturbance intensity... This ensures that the initial strategy has the potential to generalize across tasks;

[0058] For the task Based on the current meta-parameters Generate initial policy parameters Perform k=5 steps of stochastic gradient descent, with the function expression as follows:

[0059]

[0060] in: For the first meta-learning task set The task in the first The strategy parameters are updated within the loop. For the first The task in the first The strategy parameters are updated within the loop. The task number in the meta-learning task set; Update the number of steps for the inner loop iteration; The learning rate for the inner loop; For policy parameters Gradient operator for partial derivatives; For the first Each task has parameters as follows The loss function value at that time; For the first Each task has parameters as follows At that time, loss function For strategy parameters The gradient vector;

[0061] The expression for the loss function is:

[0062]

[0063] The strategy parameter is At that time, in the meta-learning task set, the first Task The loss value below; These are the policy parameters in the MERGER algorithm; For the first in the meta-learning task set One task; This is the weighting coefficient for the current deviation term; This is the absolute value of the current deviation; This is the weighting coefficient for the voltage deviation term; This is the absolute value of the voltage deviation; This represents the weighting coefficient for the energy storage SOC deviation term; This represents the absolute value of the deviation between the energy storage state of charge and the target value.

[0064] Measure the current strategy in the task The adjustment deviation on the updated parameters. The function expression for generating p=30 perturbation samples is:

[0065]

[0066] in, : The first in the meta-learning task set The first task, the first After the inner loop update, the generated first step The strategy parameters corresponding to each perturbation sample; : The first in the meta-learning task set The first task, the first The base strategy parameters are updated after the loop within the step; : Refers to the number of the perturbation sample; : refers to the random perturbation vector; : refers to a mean of 0 and a covariance matrix of The multivariate normal distribution;

[0067] Calculate the loss for each sample And the gradient is estimated through an evolutionary strategy, the function expression of which is:

[0068]

[0069] For loss function For strategy parameters The gradient vector; The normalization coefficients for the gradient estimation; To The calculation results of each perturbation sample are summed; For the first The random perturbation vector corresponding to each perturbation sample; For the first The policy parameters corresponding to each perturbation sample Below, loss function The actual value of; For all The average of the loss values ​​of each perturbation sample;

[0070] Summarize the evolutionary gradient of all tasks The meta-parameters are updated using gradient descent, and the function expression is:

[0071]

[0072] in, These are the meta-parameters in the MERGER algorithm; The learning rate for the outer loop; The average coefficients of the gradient across multiple tasks; To concentrate on tasks Sum the gradient results of each task. Task number 1 to For the first Task Below, loss function For meta-parameters The gradient vector; As an operator for taking partial derivatives with respect to the meta-parameters, evolutionary gradient regularization meta-reinforcement learning constructs a task set based on historical data from multiple scenarios. The inner loop quickly adapts to the current operating conditions, while the outer loop improves the generalization ability across scenarios. By optimizing the deviations of current, voltage, and energy storage SOC, it enhances the adaptability of the strategy and the accuracy of regulation, resulting in more efficient convergence.

[0073] Furthermore, the weighted dynamic consistency algorithm:

[0074] The N grid-connected renewable energy units in the region, including grid-connected wind turbines, conventional wind power, and photovoltaics, are modeled as an undirected graph network: G = (V, E), where the node set V represents the units, and the edge set E defines the communication connections; the state vector is initialized. Indicates the initial current adjustment amount, per unit value; Core parameter settings: Communication matrix Satisfying symmetry and normalization The expression for the control weight function is:

[0075]

[0076] in, In the weighted dynamic consensus algorithm for the regional coordination layer, nodes With nodes The regulatory weighting coefficients between them; For nodes With nodes The node degree of the communication network in which it is located;

[0077] Based on communication latency every 10ms and node deviation The function expression for updating the elements of the communication matrix is:

[0078]

[0079] in For the first During the nth iteration, the 1st Taiwanese crew and neighboring crew Communication weighting coefficients between them; This is a communication delay attenuation term; This is the delay attenuation coefficient; For the first During the nth iteration, the 1st Taiwanese crew and neighboring crew Real-time communication delay between them; For the first The set of all neighbor nodes of the Taiwan machine group The summation of the communication delay attenuation terms is the normalized denominator of the communication weights; The target unit number currently participating in status coordination; To be compatible with the target unit Adjacent neighboring unit numbers; The algorithm iterates through an undirected graph network to model unit connections, and the symmetric normalization of the communication matrix ensures balanced control. The weights are updated in real time every 10ms, dynamically adapted based on communication delay, and automatically reduce the impact of poor links. The control weights are optimized by combining node degree, which improves the accuracy and response speed of regional unit coordination and enhances the robustness of the system.

[0080] Furthermore, based on the voltage reference command issued by the regional coordination layer Per-unit value, corresponding to a 10kV rated voltage and active power distribution command, is the reference output of a single conventional wind turbine. Reference output of a single photovoltaic unit ;

[0081] Configure the core parameters for adaptive droop control:

[0082] Initial value of active current droop factor: Ensure that for every 100A increase in current deviation, the active power output is adjusted by ±5kW;

[0083] Initial value of reactive power-voltage droop factor: =0.06V / kVar, ensuring that the reactive power output can be adjusted by ±333kVar for every 0.02 per-unit increase in voltage deviation;

[0084] Attenuation coefficient: It is used to reduce the adjustment intensity in scenarios with small deviations, so as to avoid frequent fluctuations.

[0085] Real-time acquisition of reactive power and voltage deviation from the generator set output, and dynamic correction of the droop factor according to the following logic: When Per-unit value: Triggers the decay mechanism; the corrected coefficient function expression is:

[0086]

[0087] n is the actual effective coefficient in the power grid control scenario; e is a natural constant, with a fixed value of approximately 2.718 in mathematics;

[0088] when When setting per-unit values, restore the initial coefficients. Simultaneously, the reactive power output limit is set to prevent exceeding the converter capacity; the reactive power-voltage droop control function expression is:

[0089]

[0090] in, This is the reactive power command value for the generator set; This is the reactive power reference value; This is the voltage-reactive power regulation coefficient; For the voltage deviation value of the node where the unit is located, it accurately responds to the regional coordination command, and the core parameter configuration has clear adjustment accuracy; it dynamically corrects the droop coefficient, attenuates small deviations to prevent frequent fluctuations, and restores the initial strength for large deviations; it links the reactive power output limit to avoid overcapacity and ensures stable and efficient operation of the unit.

[0091] Furthermore, the configuration and execution of the grid-type wind turbine droop-VSM hybrid control; using the rate of change of current (ROCOT) as the core switching condition, is acquired in real time by the PMU with a sampling interval of 0.1s, and the hybrid control parameters are configured as follows:

[0092] Transient mode trigger threshold: ROCOT > 50 A / s;

[0093] Steady-state mode trigger threshold: 50A / s;

[0094] Sag factor: Taken as 1 / 2 of that of a conventional unit in steady state, i.e. ;

[0095] VSM mode core parameters: electromotive force E, synchronous reactance X, initial value of power angle. .

[0096] In transient mode: VSM control is immediately activated, and the electromagnetic power function expression is called as follows:

[0097]

[0098] in, 1.05 is the electromagnetic power of the synchronous motor; 1.05 is the terminal voltage adjustment coefficient or actual terminal voltage per unit value of the synchronous motor; 1.0 is the excitation electromotive force per unit value of the synchronous motor; 0.2 is the direct-axis synchronous reactance per unit value of the synchronous motor; 5.25 is the simplified coefficient of the formula. In This refers to the power angle of the synchronous motor;

[0099] In steady-state mode: trigger mode switching is smooth, using an exponential transition function.

[0100]

[0101]

[0102] in, The dynamic response coefficient of the virtual synchronous motor; It is a natural constant, with a fixed mathematical value of approximately 2.718; This is the dynamic adjustment coefficient for droop control;

[0103] The expression for the droop function is:

[0104]

[0105]

[0106] in: This is the corrected rate of change of the power grid frequency; The initial rate of change of frequency before correction; This is the reference moment of inertia of the power grid; The virtual moment of inertia provided for grid-type units; This refers to the change in active power of the generating unit or the power grid area; The current-to-power conversion factor; This refers to the change in current within the generating unit or power grid area.

[0107] The control value function expression for calculating the output current of a grid-type wind turbine is:

[0108]

[0109] in, This is the current command value for the generator set; This is a current reference value; The current change is used as the core switching condition, and the PMU high-frequency acquisition of 0.1s ensures timely response; transient activation of VSM control enhances the anti-disturbance capability, and steady-state use of adaptive droop coefficient improves the regulation accuracy; exponential transition avoids mode switching fluctuations, corrects the frequency change rate and accurately controls the current, strengthens the grid support capability and improves system stability.

[0110] Furthermore, the rotor kinetic energy release and virtual inertia activation:

[0111] Real-time monitoring of the unit's ROCOT will trigger kinetic energy release when the following conditions are met:

[0112] Grid-type wind turbine: |ROCOT|>100A / s At this time, the power increases by 20%, and ROCOT reaches 120A / s;

[0113] For conventional units: |ROCOT|>150A / s, the photovoltaic output drops sharply by 30%, and ROCOT reaches 160A / s;

[0114] Rotational inertia parameters: grid-type fan Conventional wind power Photovoltaics, because they have no rotor structure, rely on energy storage for assistance;

[0115] The expression for the function to calculate the kinetic energy release is:

[0116]

[0117] in: The rotational kinetic energy of rotating components in the power grid; This is a fixed coefficient in the rotational kinetic energy formula; Let be the moment of inertia of the rotating component; The current angular velocity of the rotating component; The rated angular velocity of the rotating component; The difference in squares of angular velocities;

[0118] Virtual inertia activation: Simultaneously with the release of kinetic energy, the current change rate function is expressed as:

[0119]

[0120] in, This is the corrected rate of change of the power grid frequency; The initial rate of change of frequency before correction; This is the reference moment of inertia of the power grid; This is the proportional term for inertia correction; The virtual moment of inertia provided for grid-type units;

[0121] Recovery logic: When ROCOT falls back to the threshold, the virtual inertia is gradually reduced to the initial value at a rate of 0.5 seconds per cycle to avoid secondary fluctuations. ROCOT is monitored in real time, and kinetic energy release is triggered differently according to the unit type. Grid-connected wind turbines and conventional units respond accurately to fluctuations. The virtual inertia activation corrects the frequency change rate, and photovoltaics compensate through energy storage. When falling back, the inertia is gradually restored to avoid secondary fluctuations, quickly smooth out grid disturbances, and enhance the system's anti-disturbance capability and stability.

[0122] The distributed new energy grid-connected unit current and voltage coordinated control system based on multi-layer coordination includes a data acquisition and preprocessing module, a global scheduling module, a regional coordination module, and a local control module. Each module forms a closed loop through data interaction to achieve coordinated control of unit current and voltage.

[0123] The data acquisition and preprocessing module is responsible for initializing unit control, acquiring real-time operating data, and standardizing voltage to provide standardized data for regulation.

[0124] The global scheduling module calculates power reference values ​​based on preprocessed data using evolutionary gradient regularization meta-reinforcement learning, generates power allocation instructions through multi-objective optimization, and sends them to the regional coordination module.

[0125] The regional coordination module models the unit as a weighted graph network and achieves coordinated control through a weighted dynamic consensus algorithm. When the voltage deviation exceeds the limit, a graded response is triggered, and then the control command is sent to the local control module.

[0126] The local control module implements differentiated strategies: conventional units adopt adaptive droop control, while grid-type wind turbines adopt droop-VSM hybrid control, ensuring the control effect through mode switching and energy release.

[0127] Compared with the prior art, the advantages of this invention are:

[0128] Addressing the issues of inertia deficiency and insufficient virtual inertia response: By employing a hybrid control mechanism of droop-VSM for grid-connected wind turbines and a dynamic virtual inertia adjustment mechanism, the limitations of traditional fixed virtual parameters are completely overcome. Based on the current rate of change ROCOT mode switching logic, VSM control is activated during transients to release rotor kinetic energy. Grid-connected wind turbines experience a 20% power surge when |ROCOT|>100A / s, while conventional units respond precisely when |ROCOT|>150A / s, quickly buffering active power imbalance. The virtual inertia value is dynamically adjusted according to the grid status. Combined with the cross-scenario generalization capability of evolutionary gradient regularization reinforcement learning, the inner loop quickly adapts to the operating conditions, while the outer loop optimizes the meta-parameters, ensuring effective suppression of current fluctuations even under extreme conditions. Simultaneously, rotor kinetic energy recovery adopts a gradual strategy of 0.5s / cycle to avoid secondary disturbances, compensating for the lack of physical inertia in new energy units and significantly improving the grid's buffering capacity and stability in response to power fluctuations.

[0129] To address the limited voltage support capability, a multi-layered voltage support system was constructed from global to local levels. The global scheduling layer sets the voltage deviation weight to no less than 30%, prioritizing voltage stability through multi-objective optimization. The regional coordination layer establishes a three-level voltage response mechanism: increasing virtual inertia when the voltage deviation is 3%~5%, invoking energy storage compensation when it is 5%~8%, and cutting off non-critical loads when it is 8%, quickly curbing voltage drops. The local control layer dynamically corrects the reactive power-voltage regulation coefficient through adaptive droop control, attenuating the regulation intensity to avoid frequent fluctuations when the deviation is small, and restoring the initial coefficient and linking it with the converter capacity limit when the deviation is large, ensuring accurate and controllable reactive power output. This system significantly improves the voltage recovery capability of new energy units during grid faults, effectively avoiding the risk of grid disconnection when the voltage drops below 0.2 per unit, shortening grid recovery time, and greatly improving the voltage support strength and operational reliability of the power system.

[0130] Addressing the lack of global coordination in distributed control: By constructing a multi-layered coordination architecture of "global scheduling - regional coordination - local control," this architecture overcomes the dual dilemmas of high latency in traditional centralized scheduling and the lack of global optimization in distributed control. The global scheduling layer employs evolutionary gradient regularized meta-reinforcement learning, trained based on historical data from over three years of scenarios, generating global power reference values ​​every 1-3 minutes to achieve precise optimization across scenarios. The regional coordination layer uses a weighted dynamic consensus algorithm, setting the weight of network-type units to 1.8 (1.0 for conventional units), and dynamically updating the communication matrix every 10ms based on latency to ensure priority convergence for critical units and avoid regional power offsetting. The local control layer executes instructions differentiated according to unit type, forming a closed loop of "global optimization - regional collaboration - individual unit implementation." This architecture not only solves the latency problem of centralized scheduling dealing with second-level fluctuations but also compensates for the lack of a global perspective in distributed control, achieving efficient collaboration between regions and units, and reducing the risk of reverse power flow fluctuations and equipment malfunctions.

[0131] Addressing the issue of high-frequency current fluctuations: This invention employs a multi-layered collaborative and differentiated control strategy to specifically suppress current fluctuations caused by the superposition of high-frequency switching of power electronic equipment and low system inertia. Conventional units utilize adaptive droop control, which weakens the regulation coefficient by 30% when the current deviation is ≤2% through an exponential decay term, preventing over-regulation from exacerbating fluctuations. Grid-connected wind turbines introduce dynamic virtual inertia, adjusting the inertia value in real time according to ROCOT to quickly smooth out high-frequency disturbances across a wide frequency band. Simultaneously, mode switching uses an exponential decay transition strategy to ensure power fluctuations are ≤3%, reducing the impact of drastic current changes on equipment insulation. In actual testing scenarios, this method can control the amplitude of high-frequency current fluctuations within 30% of the rated current, significantly lower than the effect of traditional regulation, effectively extending the insulation life of power equipment. Compared to the limitations of traditional methods in suppressing high-frequency fluctuations, this invention significantly improves the system's resilience to high-frequency current fluctuations through a multi-layered prevention and control approach of "global buffering - regional collaboration - precise suppression of individual units," ensuring equipment safety and stable grid operation. Attached Figure Description

[0132] Figure 1 This is a diagram illustrating the three-layer coordination control architecture and data interaction of the present invention.

[0133] Figure 2 This is a comparison chart of the optimization effects of the MERGER algorithm of this invention and traditional algorithms;

[0134] Figure 3 This is a diagram showing the convergence curve and dynamic change of communication weights of the weighted dynamic consensus algorithm of this invention. Detailed Implementation

[0135] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0136] Please see Figure 1-3 The present invention provides an embodiment 1, a method for coordinated current and voltage regulation of distributed renewable energy grid-connected generating units based on multi-layer coordination, comprising:

[0137] Based on real-time operation data of distributed new energy units, environment and power grid, evolutionary gradient regularization reinforcement learning is used to calculate the power reference value of each region, and the data is transmitted to the global scheduling layer in real time through industrial Ethernet and 5G dual-mode communication.

[0138] First, the distributed renewable energy units are initialized with hierarchical control. Key parameters such as real-time current, voltage, unit type, and control mode are collected to establish a characteristic parameter matrix. Unit type is used to distinguish control priority: Level 1 is for grid-connected wind turbines, Level 2 for conventional wind power, and Level 3 for photovoltaic power generation. Control mode is used to distinguish between droop and VSM operating states: 0 = standby, 1 = droop, 2 = VSM. To achieve uniform voltage dimensions for different units, 10kV is used as the reference voltage, and the unit voltage is normalized using the following function expression:

[0139]

[0140]

[0141] in This is the real-time voltage of the generator unit. To standardize voltage and provide standardized input for subsequent voltage stability judgment and control algorithms, hierarchical control initialization clarifies the unit control priority and operating mode, establishes a characteristic parameter matrix, and clarifies the control priority of grid-connected wind turbines, etc.; voltage standardization unifies the dimensions, providing standardized input for subsequent voltage stability judgment and control algorithms, and improving the accuracy and consistency of control.

[0142] Evolutionary gradient-regularized meta-reinforcement learning constructs a meta-reinforcement learning task set based on historical operational data covering three full years of seasonal operating conditions, extreme weather, and load fluctuation scenarios. Each task It includes input features such as wind speed, illumination, and load curves, and output targets, and initializes meta-parameters. and inner loop learning rate , outer loop learning rate Set the number of disturbance samples p=30 and the disturbance intensity... This ensures that the initial strategy has the potential to generalize across tasks;

[0143] For the task Based on the current meta-parameters Generate initial policy parameters Perform k=5 steps of stochastic gradient descent, with the function expression as follows:

[0144]

[0145] in: For the first meta-learning task set The task in the first The strategy parameters are updated within the loop. For the first The task in the first The strategy parameters are updated within the loop. The task number in the meta-learning task set; Update the number of steps for the inner loop iteration; The learning rate for the inner loop; For policy parameters Gradient operator for partial derivatives; For the first Each task has parameters as follows The loss function value at that time; For the first Each task has parameters as follows At that time, loss function For strategy parameters The gradient vector;

[0146] The expression for the loss function is:

[0147]

[0148] The strategy parameter is At that time, in the meta-learning task set, the first Task The loss value below; These are the policy parameters in the MERGER algorithm; For the first in the meta-learning task set One task; This is the weighting coefficient for the current deviation term; This is the absolute value of the current deviation; This is the weighting coefficient for the voltage deviation term; This is the absolute value of the voltage deviation; This represents the weighting coefficient for the energy storage SOC deviation term; This represents the absolute value of the deviation between the energy storage state of charge and the target value.

[0149] Measure the current strategy in the task The adjustment deviation on the updated parameters. The function expression for generating p=30 perturbation samples is:

[0150]

[0151] in, : The first in the meta-learning task set The first task, the first After the inner loop update, the generated first step The strategy parameters corresponding to each perturbation sample; : The first in the meta-learning task set The first task, the first The base strategy parameters are updated after the loop within the step; : Refers to the number of the perturbation sample; : refers to the random perturbation vector; : refers to a mean of 0 and a covariance matrix of The multivariate normal distribution;

[0152] Calculate the loss for each sample And the gradient is estimated through an evolutionary strategy, the function expression of which is:

[0153]

[0154] For loss function For strategy parameters The gradient vector; The normalization coefficients for the gradient estimation; To The calculation results of each perturbation sample are summed; For the first The random perturbation vector corresponding to each perturbation sample; For the first The policy parameters corresponding to each perturbation sample Below, loss function The actual value of; For all The average of the loss values ​​of each perturbation sample;

[0155] Summarize the evolutionary gradient of all tasks The meta-parameters are updated using gradient descent, and the function expression is:

[0156]

[0157] in, These are the meta-parameters in the MERGER algorithm; The learning rate for the outer loop; The average coefficients of the gradient across multiple tasks; To concentrate on tasks Sum the gradient results of each task. Task number 1 to For the first Task Below, loss function For meta-parameters The gradient vector; For operators that take partial derivatives with respect to the primary parameters, evolutionary gradient regularization-based primary reinforcement learning constructs a task set based on historical data from multiple scenarios. The inner loop quickly adapts to the current operating conditions, while the outer loop improves cross-scenario generalization ability. By optimizing the deviations of current, voltage, and energy storage SOC, it enhances policy adaptability and control accuracy, resulting in more efficient convergence.

[0158] A multi-objective optimization function is constructed, with current deviation, node voltage deviation, wind turbine output deviation and energy storage SOC deviation as optimization indicators. The voltage weight is no less than 30%. Finally, power allocation instructions for each region are generated every 2 minutes. The cycle for wind turbine regions with grid structure is shortened to 1 minute and sent to the regional coordination layer for execution in real time.

[0159] The global scheduling layer, acting as the top-level control unit, constructs a multi-objective optimization function based on standardized data:

[0160]

[0161] in This represents the maximum current deviation at the node. This represents the maximum voltage deviation at the node. The total output deviation of the grid-type wind turbine; This refers to the state of charge deviation of energy storage.

[0162] Weights satisfy Optimization is achieved using evolutionary gradient regularization meta-reinforcement learning;

[0163] Inner loop: Perturb the policy parameters to generate p samples, calculate the reward value and estimate the direction of the evolution gradient to quickly adapt to the current working conditions;

[0164] Outer loop: Aggregates gradient updates of meta-parameters across multiple scenarios, enabling the strategy to generalize across scenarios;

[0165] Finally, every 1 to 3 minutes, active / reactive power reference values ​​covering the entire region are generated and sent to the regional coordination layer to optimize multiple objectives. The global scheduling layer coordinates key deviations such as current and voltage through multi-objective optimization functions, and combines evolutionary gradient regularization reinforcement learning. The inner loop quickly adapts to the operating conditions, and the outer loop enhances the cross-scenario generalization ability. The power reference values ​​of the entire region are generated regularly to provide accurate basis for regional coordination and improve the adaptability of regulation and global coordination.

[0166] After receiving the global power reference value, the regional coordination layer uses a weighted dynamic consensus algorithm to achieve coordinated control of current and voltage. It defines node state variables (current and voltage amplitude), adjacency matrix, and weight matrix. The weight of grid-type units is higher than that of conventional units to ensure that they converge first in regional coordination.

[0167] The regional coordination layer models the N units in this region as a weighted graph network. The communication matrix satisfies symmetry and normalization properties, and its functional expression is:

[0168]

[0169] in, Let i be the degree of node i. The communication delay between nodes i and j. This is an attenuation factor that ensures the weight of links with poor communication quality is automatically reduced.

[0170] Regional coordinated regulation based on a weighted adaptive dynamic consensus algorithm:

[0171]

[0172] in, For the first After the nth iteration, the th Updated status values ​​of the desktop unit; For the first During the nth iteration, the 1st The current real-time status value of the unit; This represents the number of iterations. Assign unit number; This is the correction factor for the neighbor collaboration term; For the first Neighbor node set of the Taiwan machine group Perform summation; For the first Taiwanese crew and neighboring crew Communication weights between them; For the first Taiwanese crew and neighboring crew The regulatory weights between them; For the first In the next iteration, the neighboring units The current real-time status value; This is the correction factor for the global constraint term; For the first During the nth iteration, the 1st Global constraints on the Taiwan unit; This is the correction coefficient for the convergence acceleration term; For the first During the nth iteration, the 1st Global reference status values ​​of the unit;

[0173] The weight of the network-type nodes is set to 1.8, and that of the conventional nodes is set to 1.0. γ is the adaptive step size, ensuring that the iteration convergence is ≤25 steps and the regional voltage deviation is ≤±3%. A three-level response is triggered when the voltage deviation exceeds the limit.

[0174] 3%~5%: Increase virtual inertia to mitigate fluctuations;

[0175] 5%~8%: Simultaneously utilize the regional energy storage system to compensate for power;

[0176] 8%: Rapidly restore voltage by removing ≤5% of non-critical loads;

[0177] The regional coordination layer uses a weighted graph network model, with communication weights dynamically adjusted according to quality, and combines adaptive algorithms to achieve coordinated control of generating units; network-type generating units converge first, iterate quickly and have small voltage deviations, and a three-level response mechanism effectively deals with over-limit situations, ensuring regional stability.

[0178] Weighted dynamic consistency algorithm:

[0179] The N grid-connected renewable energy units in the region, including grid-connected wind turbines, conventional wind power, and photovoltaics, are modeled as an undirected graph network: G = (V, E), where the node set V represents the units, and the edge set E defines the communication connections; the state vector is initialized. Indicates the initial current adjustment amount, per unit value; Core parameter settings: Communication matrix Satisfying symmetry and normalization The expression for the control weight function is:

[0180]

[0181] in, In the weighted dynamic consensus algorithm for the regional coordination layer, nodes With nodes The regulatory weighting coefficients between them; For nodes With nodes The node degree of the communication network in which it is located;

[0182] Based on communication latency every 10ms and node deviation The function expression for updating the elements of the communication matrix is:

[0183]

[0184] in For the first During the nth iteration, the 1st Taiwanese crew and neighboring crew Communication weighting coefficients between them; This is a communication delay attenuation term; This is the delay attenuation coefficient; For the first During the nth iteration, the 1st Taiwanese crew and neighboring crew Real-time communication delay between them; For the first The set of all neighbor nodes of the Taiwan machine group The summation of the communication delay attenuation terms is the normalized denominator of the communication weights; The target unit number currently participating in status coordination; To be compatible with the target unit Adjacent neighboring unit numbers; The algorithm iterates through an undirected graph network to model unit relationships, and the symmetric normalization of the communication matrix ensures balanced control. The weights are updated in real time at 10ms, dynamically adapted based on communication delay, and automatically reduce the impact of poor links. The control weights are optimized by combining node degree, which improves the accuracy and response speed of regional unit coordination and enhances the robustness of the system.

[0185] Each node iteratively updates its state based on its own state, the states of its neighbors, and global reference instructions, introducing convergence acceleration terms and constraint coefficients to gradually align the state with the global objective.

[0186] The local control layer implements a differentiated control strategy. Conventional units adopt adaptive droop control. Small deviations (≤2%) weaken the droop coefficient to suppress overshoot, while large deviations restore the initial coefficient and enhance regulation capability.

[0187] The local control layer implements the regional coordination results to individual machine execution. Conventional wind and solar power units use adaptive droop control, with the function expression as follows:

[0188]

[0189]

[0190] in, For the current regulation deviation of the unit; This is the current reference value; : is the active current droop factor; This represents the real-time active power output value of the generator unit. This is a reference value for the active power of the generator unit; This is the attenuation correction term for current regulation; This is the target voltage value after unit adjustment; This is the voltage reference value; The reactive power-voltage droop factor; This refers to the real-time reactive power output value of the generator set. This is a reference value for the reactive power of the generator unit; This is the absolute value of the voltage deviation; This is the attenuation correction term for voltage regulation;

[0191] The expression for the droop coefficient function is as follows:

[0192]

[0193] in, The active-current droop factor is the same as the previous text. Consistent; The reactive power-voltage droop factor; This refers to the rated active power of the generator unit. This refers to the rated reactive power of the generator unit.

[0194] The exponential term is used to suppress over-adjustment under small deviations (e.g., the adjustment intensity is automatically reduced by 30% when the current deviation is <0.05 per unit). The grid-type wind turbine introduces virtual inertial control as shown in the function expression below:

[0195]

[0196] in, This represents the unit's real-time virtual inertia value. This serves as the initial virtual inertia reference value for the unit. This is the virtual inertia adjustment coefficient; This is the absolute value of the rate of change of current;

[0197] A hybrid control system combining droop and virtual synchronous machines is employed, where the electromagnetic power function expression for transient mode is:

[0198]

[0199] in, This refers to the output electromagnetic power of a grid-type wind turbine in VSM mode. For the virtual electromotive force in VSM mode; The real-time voltage of the power grid at the unit's connection point; Virtual synchronization reactance in VSM mode; virtual electromotive force With grid voltage The phase difference between them; For the real-time angular velocity of the grid-type wind turbine; The rated angular velocity of the unit; This is the power angle ratio adjustment coefficient; The rated power angle of the generator unit;

[0200] Steady-state mode: The droop coefficient is 1 / 3 of that of conventional units, taking into account regulation accuracy; during mode switching, an exponential decay transition strategy, as shown in the function expression below, is used to ensure power fluctuations are ≤3%, providing precise and stable underlying support for regional coordination and global scheduling. The function expression is as follows:

[0201]

[0202] in: For grid-type wind turbines during mode switching, at a certain moment Real-time control output; At the same time Below, the control output of the grid-type wind turbine in VSM mode; At the same time Below, the control output of the grid-type fan in droop mode; The current moment; This is the start time of the mode switch; Transition time for mode switching; The attenuation coefficient; For the dynamic weighting coefficients of the VSM mode control output; As the dynamic weighting coefficient of the droop mode control output, the local control layer implements the control command through differentiated control: the adaptive droop control of conventional units suppresses excessive adjustment with small deviations, and the grid-type wind turbine adopts droop-VSM hybrid control and introduces virtual inertia. The mode switching is smooth and the power fluctuation is ≤3%, which effectively enhances the system stability and provides accurate and stable underlying support for global and regional control.

[0203] Voltage reference commands issued by the regional coordination layer Per-unit value, corresponding to a 10kV rated voltage and active power distribution command, is the reference output of a single conventional wind turbine. Reference output of a single photovoltaic unit ;

[0204] Configure the core parameters for adaptive droop control:

[0205] Initial value of active current droop factor: Ensure that for every 100A increase in current deviation, the active power output is adjusted by ±5kW;

[0206] Initial value of reactive power-voltage droop factor: =0.06V / kVar, ensuring that the reactive power output can be adjusted by ±333kVar for every 0.02 per-unit increase in voltage deviation;

[0207] Attenuation coefficient: It is used to reduce the adjustment intensity in scenarios with small deviations, so as to avoid frequent fluctuations.

[0208] Real-time acquisition of reactive power and voltage deviation from the generator set output, and dynamic correction of the droop factor according to the following logic: When Per-unit value: Triggers the decay mechanism; the corrected coefficient function expression is:

[0209]

[0210] n is the actual effective coefficient in the power grid control scenario; e is a natural constant, with a fixed value of approximately 2.718 in mathematics;

[0211] when When setting per-unit values, restore the initial coefficients. Simultaneously, the reactive power output limit is set to prevent exceeding the converter capacity; the reactive power-voltage droop control function expression is:

[0212]

[0213] in, This is the reactive power command value for the generator set; This is the reactive power reference value; This is the voltage-reactive power regulation coefficient; For the voltage deviation value of the node where the unit is located, it accurately responds to the regional coordination command, and the core parameter configuration has clear adjustment accuracy; it dynamically corrects the droop coefficient, attenuates small deviations to prevent frequent fluctuations, and restores the initial strength for large deviations; it links the reactive power output limit to avoid overcapacity and ensures stable and efficient operation of the unit;

[0214] Configuration and execution of hybrid control for grid-type wind turbine droop-VSM; using the rate of change of current (ROCOT) as the core switching condition, real-time data acquisition via PMU, sampling interval 0.1s, and configuration of hybrid control parameters:

[0215] Transient mode trigger threshold: ROCOT > 50 A / s;

[0216] Steady-state mode trigger threshold: 50A / s;

[0217] Sag factor: Taken as 1 / 2 of that of a conventional unit in steady state, i.e. ;

[0218] VSM mode core parameters: electromotive force E, synchronous reactance X, initial value of power angle. .

[0219] In transient mode: VSM control is immediately activated, and the electromagnetic power function expression is called as follows:

[0220]

[0221] in, 1.05 is the electromagnetic power of the synchronous motor; 1.05 is the terminal voltage adjustment coefficient or actual terminal voltage per unit value of the synchronous motor; 1.0 is the excitation electromotive force per unit value of the synchronous motor; 0.2 is the direct-axis synchronous reactance per unit value of the synchronous motor; 5.25 is the simplified coefficient of the formula. In This refers to the power angle of the synchronous motor;

[0222] In steady-state mode: trigger mode switching is smooth, using an exponential transition function.

[0223]

[0224]

[0225] in, The dynamic response coefficient of the virtual synchronous motor; It is a natural constant, with a fixed mathematical value of approximately 2.718; This is the dynamic adjustment coefficient for droop control;

[0226] The expression for the droop function is:

[0227]

[0228]

[0229] in: This is the corrected rate of change of the power grid frequency; The initial rate of change of frequency before correction; This is the reference moment of inertia of the power grid; The virtual moment of inertia provided for grid-type units; This refers to the change in active power of the generating unit or the power grid area; The current-to-power conversion factor; This refers to the change in current within the generating unit or power grid area.

[0230] The control value function expression for calculating the output current of a grid-type wind turbine is:

[0231]

[0232] in, This is the current command value for the generator set; This is a current reference value; The current change is used as the core switching condition, and the PMU high-frequency acquisition of 0.1s ensures timely response; transient activation of VSM control enhances the anti-disturbance capability, and steady-state use of adaptive droop coefficient improves the regulation accuracy; exponential transition avoids mode switching fluctuations, corrects the frequency change rate and accurately controls the current, strengthens the grid support capability and improves system stability.

[0233] Rotor kinetic energy release and virtual inertia activation:

[0234] Real-time monitoring of the unit's ROCOT will trigger kinetic energy release when the following conditions are met:

[0235] Grid-type wind turbine: |ROCOT|>100A / s At this time, the power increases by 20%, and ROCOT reaches 120A / s;

[0236] For conventional units: |ROCOT|>150A / s, the photovoltaic output drops sharply by 30%, and ROCOT reaches 160A / s;

[0237] Rotational inertia parameters: grid-type fan Conventional wind power Photovoltaics, because they have no rotor structure, rely on energy storage for assistance;

[0238] The expression for the function to calculate the kinetic energy release is:

[0239]

[0240] in: The rotational kinetic energy of rotating components in the power grid; This is a fixed coefficient in the rotational kinetic energy formula; Let be the moment of inertia of the rotating component; The current angular velocity of the rotating component; The rated angular velocity of the rotating component; The difference in squares of angular velocities;

[0241] Virtual inertia activation: Simultaneously with the release of kinetic energy, the current change rate function is expressed as:

[0242]

[0243] in, This is the corrected rate of change of the power grid frequency; The initial rate of change of frequency before correction; This is the reference moment of inertia of the power grid; This is the proportional term for inertia correction; The virtual moment of inertia provided for grid-type units;

[0244] Recovery Logic: When ROCOT falls back to within the threshold, the virtual inertia is gradually reduced to the initial value at a rate of 0.5 seconds per cycle to avoid secondary fluctuations. ROCOT is monitored in real time, and kinetic energy release is triggered differently according to the unit type. Grid-connected wind turbines and conventional units respond accurately to fluctuations. The virtual inertia activation corrects the frequency change rate, and photovoltaics compensate through energy storage. During the fall, the inertia is gradually restored to avoid secondary fluctuations, quickly smooth out grid disturbances, and enhance the system's anti-disturbance capability and stability.

[0245] The distributed new energy grid-connected unit current and voltage coordinated control system based on multi-layer coordination includes a data acquisition and preprocessing module, a global scheduling module, a regional coordination module, and a local control module. Each module forms a closed loop through data interaction to achieve coordinated control of unit current and voltage.

[0246] The data acquisition and preprocessing module is responsible for initializing unit control, acquiring real-time operating data, and standardizing voltage to provide standardized data for regulation.

[0247] The global scheduling module calculates power reference values ​​based on preprocessed data using evolutionary gradient regularization meta-reinforcement learning, generates power allocation instructions through multi-objective optimization, and sends them to the regional coordination module.

[0248] The regional coordination module models the unit as a weighted graph network and achieves coordinated control through a weighted dynamic consensus algorithm. When the voltage deviation exceeds the limit, a graded response is triggered, and then the control command is sent to the local control module.

[0249] The local control module implements differentiated strategies: conventional units adopt adaptive droop control, while grid-type wind turbines adopt droop-VSM hybrid control, ensuring the control effect through mode switching and energy release.

[0250] In summary, this invention addresses the issues of inertia deficiency and insufficient virtual inertia response: By employing a hybrid control mechanism of droop-VSM for grid-connected wind turbines and a dynamic virtual inertia adjustment mechanism, it fundamentally overcomes the limitations of traditional fixed virtual parameters. Based on the current rate of change ROCOT mode switching logic, VSM control is activated during transients to release rotor kinetic energy. Grid-connected wind turbines experience a 20% power surge when |ROCOT|>100A / s, while conventional units respond precisely when |ROCOT|>150A / s, rapidly buffering active power imbalance. The virtual inertia value is dynamically adjusted according to the grid status. Combined with the cross-scenario generalization capability of evolutionary gradient regularization reinforcement learning, the inner loop quickly adapts to operating conditions, while the outer loop optimizes meta-parameters, ensuring effective suppression of current fluctuations even under extreme conditions. Simultaneously, rotor kinetic energy recovery employs a gradual strategy of 0.5s / cycle to avoid secondary disturbances, compensating for the lack of physical inertia in new energy units and significantly improving the grid's buffering capacity and stability in response to power fluctuations.

[0251] To address the limited voltage support capability, a multi-layered voltage support system was constructed from global to local levels. The global scheduling layer sets the voltage deviation weight to no less than 30%, prioritizing voltage stability through multi-objective optimization. The regional coordination layer establishes a three-level voltage response mechanism: increasing virtual inertia when the voltage deviation is 3%~5%, invoking energy storage compensation when it is 5%~8%, and cutting off non-critical loads when it is 8%, quickly curbing voltage drops. The local control layer dynamically corrects the reactive power-voltage regulation coefficient through adaptive droop control, attenuating the regulation intensity to avoid frequent fluctuations when the deviation is small, and restoring the initial coefficient and linking the converter capacity limit when the deviation is large, ensuring accurate and controllable reactive power output. This system significantly improves the voltage recovery capability of new energy units during grid faults, effectively avoiding the risk of grid disconnection when the voltage drops below 0.2 per unit, shortening grid recovery time, and greatly improving the voltage support strength and operational reliability of the power system.

[0252] Addressing the lack of global coordination in distributed control: By constructing a multi-layered coordination architecture of "global scheduling - regional coordination - local control," this architecture overcomes the dual dilemmas of high latency in traditional centralized scheduling and the lack of global optimization in distributed control. The global scheduling layer employs evolutionary gradient regularized meta-reinforcement learning, trained on historical data from over three years of scenarios, generating global power reference values ​​every 1-3 minutes to achieve precise optimization across scenarios. The regional coordination layer uses a weighted dynamic consensus algorithm, setting the weight of network-type units to 1.8 (1.0 for conventional units), and dynamically updating the communication matrix every 10ms based on latency to ensure priority convergence for critical units and avoid regional power offsetting. The local control layer executes instructions differentiated according to unit type, forming a closed loop of "global optimization - regional collaboration - individual unit implementation." This architecture not only solves the latency problem of centralized scheduling dealing with second-level fluctuations but also compensates for the lack of a global perspective in distributed control, achieving efficient collaboration between regions and units, and reducing the risk of reverse power flow fluctuations and equipment malfunctions.

[0253] Addressing the issue of high-frequency current fluctuations: This invention employs a multi-layered collaborative and differentiated control strategy to specifically suppress current fluctuations caused by the superposition of high-frequency switching of power electronic equipment and low system inertia. Conventional units utilize adaptive droop control, which weakens the regulation coefficient by 30% when the current deviation is ≤2% through an exponential decay term, preventing over-regulation from exacerbating fluctuations. Grid-connected wind turbines introduce dynamic virtual inertia, adjusting the inertia value in real time according to ROCOT to quickly smooth out high-frequency disturbances across a wide frequency band. Simultaneously, mode switching uses an exponential decay transition strategy to ensure power fluctuations are ≤3%, reducing the impact of drastic current changes on equipment insulation. In actual testing scenarios, this method can control the amplitude of high-frequency current fluctuations within 30% of the rated current, significantly lower than the effect of traditional regulation, effectively extending the insulation life of power equipment. Compared to the limitations of traditional methods in suppressing high-frequency fluctuations, this invention significantly improves the system's resilience to high-frequency current fluctuations through a multi-layered prevention and control approach of "global buffering - regional collaboration - precise suppression of individual units," ensuring equipment safety and stable grid operation.

[0254] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A method for coordinated current and voltage regulation of distributed renewable energy grid-connected generating units based on multi-layer coordination, characterized in that: include: The global scheduling layer uses evolutionary gradient regularized reinforcement learning to calculate the power reference value of each region based on the real-time operation data of distributed new energy units, environment and power grid. A multi-objective optimization function is constructed, with current deviation, node voltage deviation, wind turbine output deviation and energy storage SOC deviation as optimization indicators. The voltage weight is optimized, and finally, power allocation instructions for each region are generated. The cycle of wind turbine regions with grid structure is shortened and sent to the regional coordination layer for execution in real time. After receiving the global power reference value, the regional coordination layer uses a weighted dynamic consensus algorithm to achieve coordinated control of current and voltage. It defines node state variables (current and voltage amplitude), adjacency matrix, and weight matrix. The weight of grid-type units is higher than that of conventional units to ensure that they converge first in regional coordination. Each node iteratively updates its state based on its own state, the states of its neighbors, and global reference instructions, introducing convergence acceleration terms and constraint coefficients to gradually align the state with the global objective. The local control layer implements a differentiated control strategy. Conventional units adopt adaptive droop control. Small deviations (≤2%) weaken the droop coefficient to suppress overshoot, while large deviations restore the initial coefficient and enhance regulation capability.

2. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination as described in claim 1, characterized in that: First, the distributed renewable energy units are initialized with hierarchical control. Key parameters such as real-time current, voltage, unit type, and control mode are collected to establish a characteristic parameter matrix. Unit type is used to distinguish control priority: Level 1 is for grid-connected wind turbines, Level 2 for conventional wind power, and Level 3 for photovoltaic power generation. Control mode is used to distinguish between droop and VSM operating states: 0 = standby, 1 = droop, 2 = VSM. To achieve uniform voltage dimensions for different units, 10kV is used as the reference voltage, and the unit voltage is normalized using the following function expression: in This is the real-time voltage of the generator unit. This standardizes the voltage, providing a standardized input for subsequent voltage stability assessment and control algorithms.

3. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination as described in claim 2, characterized in that: The global scheduling layer, acting as the top-level control unit, constructs a multi-objective optimization function based on standardized data: in This represents the maximum current deviation at the node. This represents the maximum voltage deviation at the node. The total output deviation of the grid-type wind turbine; This refers to the state of charge deviation of energy storage. Weights satisfy Optimization is achieved using evolutionary gradient regularization meta-reinforcement learning; Inner loop: Perturb the policy parameters to generate p samples, calculate the reward value and estimate the direction of the evolution gradient to quickly adapt to the current working conditions; Outer loop: Aggregates gradient updates of meta-parameters across multiple scenarios, enabling the strategy to generalize across scenarios; Finally, active / reactive power reference values ​​covering the entire area are generated every 1 to 3 minutes and sent to the regional coordination layer.

4. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination as described in claim 1, characterized in that: The regional coordination layer models the N units in this region as a weighted graph network. The communication matrix satisfies symmetry and normalization properties, and its functional expression is: in, Let i be the degree of node i. The communication delay between nodes i and j. This is an attenuation factor that ensures the weight of links with poor communication quality is automatically reduced. Regional coordinated regulation based on a weighted adaptive dynamic consensus algorithm: in, For the first After the nth iteration, the th Updated status values ​​of the desktop unit; For the first During the nth iteration, the 1st The current real-time status value of the unit; This represents the number of iterations. Assign unit number; This is the correction factor for the neighbor collaboration term; For the first Neighbor node set of the Taiwan machine group Perform summation; For the first Taiwanese crew and neighboring crew Communication weights between them; For the first Taiwanese crew and neighboring crew The regulatory weights between them; For the first In the next iteration, the neighboring units The current real-time status value; This is the correction factor for the global constraint term; For the first During the nth iteration, the 1st Global constraints on the Taiwan unit; This is the correction coefficient for the convergence acceleration term; For the first During the nth iteration, the 1st Global reference status values ​​of the unit; The weight of the network-type node is set to 1.8, and that of the conventional node is set to 1.

0. γ is the adaptive step size, ensuring that the iteration convergence is ≤25 steps and the regional voltage deviation is ≤±3%. When the voltage deviation exceeds the limit, a third-level response is triggered. 3%~5%: Increase virtual inertia to mitigate fluctuations; 5%~8%: Simultaneously utilize the regional energy storage system to compensate for power; 8%: Remove ≤5% of non-critical loads to quickly restore voltage.

5. The method for coordinated current and voltage regulation of distributed renewable energy grid-connected generating units based on multi-layer coordination as described in claim 1, characterized in that: The local control layer implements the regional coordination results to individual machine execution. Conventional wind and solar power units use adaptive droop control, with the function expression as follows: in, For the current regulation deviation of the unit; This is the current reference value; : is the active current droop factor; This represents the real-time active power output value of the generator unit. This is a reference value for the active power of the generator unit; This is the attenuation correction term for current regulation; This is the target voltage value after unit adjustment; This is the voltage reference value; This is the reactive power-voltage droop factor; This refers to the real-time reactive power output value of the generator set. This is a reference value for the reactive power of the generator unit. This is the absolute value of the voltage deviation; This is the attenuation correction term for voltage regulation; The expression for the droop coefficient function is as follows: in, The active-current droop factor is the same as the previous text. Consistent; This is the reactive power-voltage droop factor; This refers to the rated active power of the generator unit. This refers to the rated reactive power of the generator unit. The exponential term is used to suppress over-adjustment under small deviations (e.g., the adjustment intensity is automatically reduced by 30% when the current deviation is <0.05 per unit). The grid-type wind turbine introduces virtual inertial control as shown in the function expression below: in, This represents the unit's real-time virtual inertia value. This serves as the initial virtual inertia reference value for the unit. This is the virtual inertia adjustment coefficient; This is the absolute value of the rate of change of current; A hybrid control system combining droop and virtual synchronous machines is employed, where the electromagnetic power function expression in transient mode is: in, This refers to the output electromagnetic power of a grid-type wind turbine in VSM mode. For the virtual electromotive force in VSM mode; The real-time voltage of the power grid at the unit's connection point; Virtual synchronization reactance in VSM mode; virtual electromotive force With grid voltage The phase difference between them; For the real-time angular velocity of the grid-type wind turbine; The rated angular velocity of the unit; This is the power angle ratio adjustment coefficient; The rated power angle of the generator unit; Steady-state mode: The droop coefficient is 1 / 3 of that of conventional units, while also considering regulation accuracy; during mode switching, an exponential decay transition strategy, as shown in the function expression below, is used to ensure power fluctuations are ≤3%, and the function expression is: in: For grid-type wind turbines during mode switching, at a certain moment Real-time control output; At the same time Below, the control output of the grid-type wind turbine in VSM mode; At the same time Below, the control output of the grid-type fan in droop mode; The current moment; This is the start time of the mode switch; Transition time for mode switching; The attenuation coefficient; For the dynamic weighting coefficients of the VSM mode control output; The dynamic weighting coefficients for controlling the output in droop mode.

6. The method for coordinated current and voltage regulation of distributed renewable energy grid-connected generating units based on multi-layer coordination as described in claim 1, characterized in that: Evolutionary gradient-regularized meta-reinforcement learning constructs a meta-reinforcement learning task set based on historical operational data covering three full years of seasonal operating conditions, extreme weather, and load fluctuation scenarios. Each task It includes input features such as wind speed, illumination, and load curves, and output targets, and initializes meta-parameters. and inner loop learning rate , outer loop learning rate Set the number of disturbance samples p=30 and the disturbance intensity... This ensures that the initial strategy has the potential to generalize across tasks; For the task Based on the current meta-parameters Generate initial policy parameters Perform k=5 steps of stochastic gradient descent, with the function expression as follows: in: For the first meta-learning task set The task in the first The strategy parameters are updated within the loop. For the first The task in the first The strategy parameters are updated within the loop. The task number in the meta-learning task set; Update the number of steps for the inner loop iteration; The learning rate for the inner loop; For policy parameters Gradient operator for partial derivatives; For the first Each task has parameters as follows The loss function value at that time; For the first Each task has parameters as follows At that time, loss function For strategy parameters The gradient vector; The expression for the loss function is: The strategy parameter is At that time, in the meta-learning task set, the first Task The loss value below; These are the policy parameters in the MERGER algorithm; For the first in the meta-learning task set One task; This is the weighting coefficient for the current deviation term; This is the absolute value of the current deviation; This is the weighting coefficient for the voltage deviation term; This is the absolute value of the voltage deviation; This represents the weighting coefficient for the energy storage SOC deviation term; This represents the absolute value of the deviation between the energy storage state of charge and the target value. Measure the current strategy in the task The control deviation on the updated parameters The function expression for generating p=30 perturbation samples is: in, : The first in the meta-learning task set The first task, the first After the inner loop update, the generated first... The policy parameters corresponding to each perturbation sample; : The first in the meta-learning task set The first task, the first The base strategy parameters are updated after the loop within the step; : Refers to the number of the perturbation sample; : refers to the random perturbation vector; : refers to a mean of 0 and a covariance matrix of The multivariate normal distribution; Calculate the loss for each sample And the gradient is estimated through an evolutionary strategy, the function expression of which is: For loss function For strategy parameters The gradient vector; The normalization coefficients for the gradient estimation; To The calculation results of each perturbation sample are summed; For the first The random perturbation vector corresponding to each perturbation sample; For the first The policy parameters corresponding to each perturbation sample Below, loss function The actual value of; For all The average of the loss values ​​of each perturbation sample; Summarize the evolutionary gradient of all tasks The meta-parameters are updated using gradient descent, and the function expression is: in, These are the meta-parameters in the MERGER algorithm; The learning rate for the outer loop; The average coefficients of the gradient across multiple tasks; To concentrate on tasks Sum the gradient results of each task. Task number 1 to For the first Task Below, loss function For meta-parameters The gradient vector; An operator for taking partial derivatives with respect to the primary parameters.

7. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination according to claim 1, characterized in that: Weighted dynamic consistency algorithm: The N grid-connected renewable energy units in the region, including grid-connected wind turbines, conventional wind power, and photovoltaics, are modeled as an undirected graph network: G = (V, E), where the node set V represents the units, and the edge set E defines the communication connections; the state vector is initialized. Indicates the initial current adjustment amount, per unit value; Core parameter settings: Communication matrix Satisfying symmetry and normalization The expression for the control weight function is: in, In the weighted dynamic consensus algorithm for the regional coordination layer, nodes With nodes The regulatory weighting coefficients between them; For nodes With nodes The node degree of the communication network in which it is located; Based on communication latency every 10ms and node deviation The function expression for updating the elements of the communication matrix is: in For the first During the nth iteration, the 1st Taiwanese crew and neighboring crew Communication weighting coefficients between them; This is a communication delay attenuation term; This is the delay attenuation coefficient; For the first During the nth iteration, the 1st Taiwanese crew and neighboring crew Real-time communication delay between them; For the first The set of all neighbor nodes of the Taiwan machine group The summation of the communication delay attenuation terms is the normalized denominator of the communication weights; The target unit number currently participating in status coordination; To be compatible with the target unit Adjacent neighboring unit numbers; This represents the number of iterations of the algorithm.

8. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination as described in claim 1, characterized in that: Voltage reference commands issued by the regional coordination layer Per-unit value, corresponding to a 10kV rated voltage and active power distribution command, is the reference output of a single conventional wind turbine. Reference output of a single photovoltaic unit ; Configure the core parameters for adaptive droop control: Initial value of active current droop factor: Ensure that for every 100A increase in current deviation, the active power output is adjusted by ±5kW; Initial value of reactive power-voltage droop factor: =0.06V / kVar, ensuring that the reactive power output can be adjusted by ±333kVar for every 0.02 per-unit increase in voltage deviation; Attenuation coefficient: This is used to weaken the adjustment force in scenarios with small deviations to avoid frequent fluctuations; Real-time acquisition of reactive power and voltage deviation from the generator set output, and dynamic correction of the droop factor according to the following logic: When Per-unit value: Triggers the decay mechanism; the corrected coefficient function expression is: n is the actual effective coefficient in the power grid control scenario; e is a natural constant, with a fixed value of approximately 2.718 in mathematics; when When setting per-unit values, restore the initial coefficients. Simultaneously, the reactive power output limit is set to prevent exceeding the converter capacity; the reactive power-voltage droop control function expression is: in, This is the reactive power command value for the generator set; This is the reactive power reference value; This is the voltage-reactive power regulation coefficient; This represents the voltage deviation value at the node where the generator unit is located.

9. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination as described in claim 8, characterized in that: The configuration and execution of the hybrid control for the grid-type wind turbine droop-VSM are described below; using the rate of change of current (ROCOT) as the core switching condition, data is collected in real time via PMU with a sampling interval of 0.1s, and hybrid control parameters are configured as follows: Transient mode trigger threshold: ROCOT > 50 A / s; Steady-state mode trigger threshold: 50A / s; Sag factor: Taken as half of that of a conventional unit in steady state, i.e. ; VSM mode core parameters: electromotive force E, synchronous reactance X, initial value of power angle. ; In transient mode: VSM control is immediately activated, and the electromagnetic power function expression is called as follows: in, 1.05 is the electromagnetic power of the synchronous motor; 1.05 is the terminal voltage adjustment coefficient or actual terminal voltage per unit value of the synchronous motor; 1.0 is the excitation electromotive force per unit value of the synchronous motor; 0.2 is the direct-axis synchronous reactance per unit value of the synchronous motor; 5.25 is the simplified coefficient of the formula. In This refers to the power angle of the synchronous motor; In steady-state mode: trigger mode switching is smooth, using an exponential transition function. in, The dynamic response coefficient of the virtual synchronous motor; It is a natural constant, with a fixed mathematical value of approximately 2.718; This is the dynamic adjustment coefficient for droop control; The expression for the droop function is: in: This is the corrected rate of change of the power grid frequency; The initial rate of change of frequency before correction; This is the reference moment of inertia of the power grid; The virtual moment of inertia provided for grid-type units; This refers to the change in active power of the generating unit or the power grid area; The current-to-power conversion factor; This refers to the change in current within the generating unit or power grid area. The control value function expression for calculating the output current of a grid-type wind turbine is: in, This is the current command value for the generator set; This is a current reference value; This represents the change in current.

10. The method for coordinated current and voltage regulation of distributed new energy grid-connected generating units based on multi-layer coordination according to claim 9, characterized in that: The rotor kinetic energy release and virtual inertia activation: Real-time monitoring of the unit's ROCOT will trigger kinetic energy release when the following conditions are met: Grid-type wind turbine: |ROCOT|>100A / s At this time, the power increases by 20%, and ROCOT reaches 120A / s; For conventional units: |ROCOT|>150A / s, the photovoltaic output drops sharply by 30%, and ROCOT reaches 160A / s; Rotational inertia parameters: grid-type fan Conventional wind power Photovoltaics, because they have no rotor structure, rely on energy storage for assistance; The expression for the function to calculate the kinetic energy release is: in: The rotational kinetic energy of rotating components in the power grid; This is a fixed coefficient in the rotational kinetic energy formula; The moment of inertia of the rotating component; The current angular velocity of the rotating component; The rated angular velocity of the rotating component; The difference in squares of angular velocities; Virtual inertia activation: Simultaneously with the release of kinetic energy, the current change rate function is expressed as: in, This is the corrected rate of change of the power grid frequency; The initial rate of change of frequency before correction; This is the reference moment of inertia of the power grid; This is the proportional term for inertia correction; The virtual moment of inertia provided for grid-type units; Recovery logic: When ROCOT falls back to within the threshold, the virtual inertia is gradually reduced to the initial value at a rate of 0.5 seconds per cycle to avoid secondary fluctuations.

Images