Model-Predictive Frequency Control Method Based on Coordination of Inertia and Power in New Energy Power Stations

Abstract

This application belongs to the field of power electronics technology and relates to a model predictive frequency control method for coordinating inertia and power in renewable energy power plants. It addresses the system stability problem caused by the high proportion of renewable energy penetration in heterogeneous renewable energy power plants by coordinating the prediction and optimization of virtual inertia and active power reference values. The method includes: constructing a nonlinear model predictive control optimization model based on time-varying weights; obtaining smoothed frequency deviation estimates, power disturbance estimates, and reconstructed frequency change rates at the current moment using a Kalman filter-based state observation model; and inputting the frequency deviation estimates, power disturbance estimates, and reconstructed frequency change rates into the nonlinear model predictive control optimization model based on time-varying weights to obtain the virtual inertia reference values ​​and active power reference values ​​at the current moment. This application can effectively reduce the maximum frequency deviation of the system, eliminate control signal oscillations, and ensure stable operation of each state variable within the safety boundary under extreme disturbance conditions.

Description

Technical Field

[0001] This application belongs to the field of power electronics technology and relates to power system frequency stability control and new energy grid connection technology. Specifically, it relates to a model prediction frequency control method that coordinates the inertia and power of new energy power plants. Background Technology

[0002] Driven by the global energy transition, the penetration rate of new energy sources such as wind power and photovoltaics in the power system continues to rise. The modern power system is undergoing a major transformation from being dominated by traditional synchronous generators to being dominated by power electronic devices. The system is exhibiting significant heterogeneity, with traditional synchronous generators gradually being replaced by grid-connected and grid-connected inverters.

[0003] However, existing grid-connected equipment typically operates in maximum power point tracking (MPPT) mode, where its output power is decoupled from the grid frequency, effectively lacking the ability to proactively respond to grid frequency changes. As the proportion of such grid-connected equipment continues to increase, the system's equivalent inertia and primary frequency regulation capability decrease significantly. When encountering sudden load changes or drastic fluctuations in power generation, the power system is highly susceptible to a sharp increase in the rate of frequency change, significantly weakening its disturbance immunity and potentially leading to frequency instability or cascading power outages under extreme conditions.

[0004] To improve the frequency stability of power grids with a high proportion of renewable energy, Virtual Synchronous Generator (VSG) technology has emerged. By simulating the physical characteristics of the rotor motion equations of a synchronous generator, VSG can provide a certain amount of inertia and damping support for grid-connected inverters. However, traditional VSG control typically uses fixed inertia and damping coefficients. This static parameter setting is difficult to effectively adapt to the complex and variable actual operating conditions of renewable energy power plants. Specifically, in the initial stage of a disturbance, the system needs a large inertia to suppress the rate of frequency change and limit the drop in the maximum frequency deviation; however, in the frequency recovery stage, excessive inertia means that too much kinetic energy accumulates in the system, which not only slows down the speed of frequency recovery to the rated operating point and increases the settling time, but may also cause secondary frequency oscillations in severe cases. On the other hand, while a smaller inertia helps with rapid steady-state recovery, it cannot provide sufficient transient frequency support. Therefore, fixed-parameter VSG control struggles to achieve an ideal balance between transient frequency suppression and dynamic recovery performance.

[0005] To overcome the shortcomings of fixed-parameter control schemes, existing technologies have proposed some adaptive parameter adjustment strategies or introduced model predictive control architectures. However, most existing methods are based on decoupled control development, treating adaptive inertia adjustment as an independent external module and active power point tracking as another independent component. This decoupled structure severely ignores the inherent nonlinear coupling characteristics within the rotor motion equations. Since time-varying inertia continuously alters the dynamic properties of the power system, traditional power controllers designed based on nominal system models cannot adapt to these real-time dynamic changes, leading to conflicts between providing inertia support and active power injection control. Ultimately, this results in suboptimal system dynamic recovery performance and makes it difficult to achieve global synergistic optimization between frequency security support, control command smoothness, and the safety of power station energy storage operation. Summary of the Invention

[0006] This application addresses the problems existing in the prior art by providing a model predictive frequency control method that coordinates inertia and power in new energy power stations. This method solves the technical problems that fixed parameters cannot simultaneously take into account transient frequency suppression and dynamic recovery speed, and that traditional decoupled control architectures are prone to conflicts between inertia support and active power injection, leading to energy storage exceeding limits and control oscillations.

[0007] In its first aspect, this application provides a model predictive frequency control method for coordinating inertia and power in renewable energy power plants, comprising: Model construction steps: Using a nonlinear discrete prediction model as the base model, a multi-objective cost function with multiple objectives of improving frequency dynamic performance and optimizing operational economy is constructed as the objective function, and the active power constraints of the power station's photovoltaic storage module and the SOC constraints of the energy storage module are used as constraints to construct a nonlinear model prediction control optimization model based on time-varying weights. Estimation steps: Use a state observation model based on Kalman filtering to obtain the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment; Prediction steps: Input the frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate into the nonlinear model predictive control optimization model to obtain the virtual inertia reference value and active power reference value at the current moment.

[0008] In conjunction with the first aspect, in some embodiments, the nonlinear discrete prediction model is represented as follows in the model building step:

[0009] in:

[0010] In the formula, for Frequency deviation at time, for Photovoltaic active power at any given time for The active power of energy storage at any given time. for The state of charge of the stored energy at any given moment; for Frequency deviation at time, for Photovoltaic active power at any given time for The active power of energy storage at any given time. for The state of charge of the stored energy at any given moment; This is the disturbance estimate. The sampling period is For the rated frequency, Let be the equivalent time constant of photovoltaic power. The equivalent time constant of the energy storage system, The rated installed capacity of the energy storage system. The damping coefficient; for The virtual inertia reference value at time t. for Reference value of photovoltaic active power at any given time. for Reference value of active power of energy storage at any given time; for The equivalent inertial constant at time t. The VSG inertial time constant, The proportion of VSG capacity to the total capacity of the energy storage system. The proportion of grid-connected photovoltaic capacity to the total energy storage system capacity. This refers to the proportion of grid-type energy storage capacity to the total capacity of the energy storage system.

[0011] In conjunction with the first aspect, in some embodiments, in the model construction step, the multi-objective cost function includes a frequency dynamic performance function, a control cost function, and an operational economy function; the frequency dynamic performance function is used to penalize frequency deviation and frequency change rate; the control cost function is used to penalize control increments to ensure smooth operation of the actuator and prevent frequent fluctuations in virtual inertia reference values ​​and active power reference values; the operational economy function is used to minimize power output deviation to improve economic dispatching effect and delay battery aging; the multi-objective cost function is expressed as:

[0012] In the formula, For a multi-objective cost function, To predict the time domain, For frequency dynamic performance function, To control the cost function, To run the economy function, For time, For time steps.

[0013] In conjunction with the first aspect, in some embodiments, the frequency dynamic performance function is expressed as:

[0014] in:

[0015] In the formula, It is an adaptive weighting factor used to dynamically adjust the penalty intensity for frequency deviation and frequency change rate; This is the frequency deviation weighting coefficient. for Frequency deviation at any given moment; The frequency change rate weighting coefficient, for Rate of change of frequency at time, This is the sensitivity coefficient. The maximum permissible deviation of the nominal frequency. This is the maximum permissible value for the nominal frequency change rate.

[0016] In conjunction with the first aspect, in some embodiments, the control cost function is expressed as:

[0017] In the formula, For virtual inertia control incremental weighting coefficients, for The virtual inertia control increment at any given moment; The active power command control incremental weighting coefficient is used. for Active power command control increment at any given time; This is the virtual inertia deviation weighting coefficient. for Virtual inertia at any given moment This is the virtual inertia rating.

[0018] In conjunction with the first aspect, in some embodiments, the operational economy function is expressed as:

[0019] In the formula, This is the active power output weighting coefficient. for The active power of the energy storage module at all times. for The active power of the photovoltaic module at any given time.

[0020] In conjunction with the first aspect, in some embodiments, the active power constraint of the power station's optical storage module in the model building step is expressed as:

[0021] In the formula, for The active power adjustment of the corresponding module at all times. This represents the maximum active power reserve capacity of the module. The SOC constraint of the energy storage module is expressed as follows:

[0022] In the formula, In order to be in Always The predicted value of the state of charge of the energy storage module at any time. This represents the lower limit of the state of charge of the energy storage module. This represents the upper limit of the state of charge of the energy storage module.

[0023] In conjunction with the first aspect, in some embodiments, the method for obtaining the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment using a Kalman filter-based state observation model in the estimation step includes: The external power disturbance is modeled as a slowly changing augmented state, and the state vector is defined as follows: The control input is The measurement output is The state-space model based on Kalman filtering is constructed as follows:

[0024] in, ; In the formula, For frequency deviation, For power disturbance, This is the transpose of the matrix. This is the active power command value. This is a frequency measurement value. , For the system matrix, For the output matrix, To conform to the Gaussian white noise distribution of process noise, For measurement noise that conforms to a Gaussian white noise distribution, This is the equivalent damping coefficient. It is the equivalent inertial constant. For time; The Kalman filter is used for time and measurement updates, and recursive updates are performed to calculate the optimal estimate, thus obtaining the optimal state estimate. , This is the estimated value of the frequency deviation. This is an estimate of the power disturbance. The time update is represented as:

[0025] In the formula, for The prior state estimate at time t. for The posterior state estimate at time t. for The prior error covariance matrix at time t. for The posterior error covariance matrix at time t; Discretized system matrix The input matrix is ​​discretized. For process noise The covariance matrix; The measurement update is represented as follows:

[0026] In the formula, for Kalman gain at time step For measuring noise The covariance matrix, for The posterior state estimate at time t. for The posterior error covariance matrix at time t. It is the identity matrix; Using the optimal state estimate and the physical constraints of the rocking equation, a smooth and physically consistent rate of frequency change is reconstructed, expressed as:

[0027] In the formula, for The rate of change of the reconstructed frequency at any given moment for The equivalent inertial constant at time t. for The input is controlled at all times, namely the active power command value.

[0028] In conjunction with the first aspect, in some embodiments, the methods for obtaining the current virtual inertia reference value and active power reference value include: State trajectory prediction based on a nonlinear discrete prediction model; Under the conditions of satisfying the active power constraints of the photovoltaic storage module and the SOC constraints of the energy storage module, a nonlinear programming solver is used to minimize the objective function in the prediction time domain to generate the control increment sequence. The first element of the control increment sequence is used as the virtual inertia reference value and active power reference value at the current moment.

[0029] In conjunction with the first aspect, in some embodiments, the method further includes: at the next sampling time, returning to the estimation step and continuing to predict the virtual inertia reference value and active power reference value for the next time step.

[0030] Compared with the prior art, the advantages and positive effects of this application are as follows: This application provides a model predictive frequency control method for coordinating inertia and power in new energy power plants. It uses a nonlinear discrete predictive model as the base model, a multi-objective cost function (constructed with multiple objectives of improving frequency dynamic performance and optimizing operational economy) as the objective function, and active power constraints of the photovoltaic-storage modules and SOC constraints of the energy storage modules as constraints. A time-varying weighted nonlinear model predictive control optimization model is constructed. A Kalman filter-based state observation model is used to obtain the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment. These estimates are then input into the nonlinear model predictive control optimization model to obtain the virtual inertia reference value and active power reference value at the current moment. This application incorporates the virtual inertia reference value and active power reference value into a unified nonlinear model predictive control, directly addressing the inherent nonlinear coupling characteristics in the rotor motion equations and achieving coordinated optimization of the virtual inertia reference value and active power reference value. The introduction of a time-varying weighting mechanism for the phase plane trajectory solves the problem of slow recovery in traditional methods. By leveraging the state observation model with Kalman filtering, disturbances are accurately estimated, effectively eliminating high-frequency oscillations in the control signal and extending the service life of the underlying hardware. The safety boundaries of active power and the SOC (State of Charge) of the energy storage module are directly embedded as hard constraints into the predictive control optimization. When the system approaches its operating limits, the power regulation burden can be automatically and smoothly transferred, effectively preventing over-discharge of the energy storage battery and ensuring the safety of the site equipment under extreme conditions. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the model prediction frequency control method for coordinating inertia and power in new energy power stations as described in the embodiments of this application. Figure 2 This is a schematic flowchart illustrating the method for obtaining the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment in an embodiment of this application. Figure 3This is a schematic flowchart illustrating the method for obtaining the current virtual inertia reference value and active power reference value according to an embodiment of this application. Figure 4 This is a schematic diagram of the heterogeneous new energy power station topology according to an embodiment of this application; Figure 5 This is a schematic diagram of the improved IEEE 9-node system topology according to an embodiment of this application; Figure 6 This is a schematic diagram comparing the system frequency response under four different control strategies in an embodiment of this application under a step load disturbance. Figure 7 This is a schematic diagram comparing the equivalent virtual inertia constant response under four different control strategies in an embodiment of this application under a step load disturbance. Figure 8 This is a schematic diagram comparing the active power command values ​​of the energy storage system under four different control strategies under a step load disturbance according to an embodiment of this application. Figure 9 This is a schematic diagram comparing the active power command values ​​of a photovoltaic system under four different control strategies under a step load disturbance, as described in an embodiment of this application. Figure 10 This is a schematic diagram of the active power disturbance of photovoltaic under cloud cover conditions according to an embodiment of this application; Figure 11 This is a schematic diagram comparing the system frequency response under four different control strategies under random photovoltaic fluctuations, as described in the embodiments of this application. Figure 12 This is a schematic diagram comparing the equivalent virtual inertia constant responses under four different control strategies for random photovoltaic fluctuations in embodiments of this application; Figure 13 This is a schematic diagram of the dynamic response of the system frequency and equivalent virtual inertia constant of the system under the control method proposed in this application when the SOC is close to the lower limit of the SOC. Figure 14 This diagram illustrates the active power command of the energy storage system, the active power command of the photovoltaic system, and the dynamic response of the energy storage SOC under the control method proposed in this application when the system approaches the lower limit of SOC.

[0032] In the diagram, 1. Power grid, 2. Common coupling point, 3. n 4. Grid-connected photovoltaic units. n Individual grid-type photovoltaic units, 5. n Individual grid-type energy storage units, 6, Node 1, 7, Node 2, 8, Node 3, 9, Node 4, 10, Node 5, 11, Node 6, 12, Node 7, 13, Node 8, 14, Node 9, 15, Load 1, 16, Load 2, 17, Load 3, 18, Synchronous phasor measurement device, 19, Thermal power unit G1, 20, Thermal power unit G2, 21, New energy power station. Detailed Implementation

[0033] The present application will now be described in detail with reference to the accompanying drawings through exemplary embodiments. However, it should be understood that, without further description, elements, structures, and features in one embodiment may be advantageously incorporated into other embodiments.

[0034] This application provides a model predictive frequency control method for coordinating inertia and power in renewable energy power plants. Addressing the system stability issues caused by the high proportion of renewable energy penetration in heterogeneous renewable energy power plants, this method performs coordinated prediction and optimization of virtual inertia reference values ​​and active power reference values. The following detailed description of this model predictive frequency control method for coordinating inertia and power in renewable energy power plants, in conjunction with accompanying drawings and embodiments, provides a comprehensive overview.

[0035] See Figure 1 The first aspect of this application provides a model predictive frequency control method for coordinating inertia and power in new energy power stations, including: S1. Model construction steps: Using the nonlinear discrete prediction model as the base model, a multi-objective cost function with multiple objectives of improving frequency dynamic performance and optimizing operational economy is constructed as the objective function. The active power constraints of the power station's photovoltaic storage module and the SOC constraints of the energy storage module are used as constraints to construct a nonlinear model prediction control optimization model based on time-varying weights.

[0036] Specifically, in one embodiment of this application, the nonlinear discrete prediction model is expressed as:

[0037] in:

[0038] In the formula, for Frequency deviation at time, for Photovoltaic active power at any given time for The active power of energy storage at any given time. for The state of charge of the stored energy at any given moment; for Frequency deviation at time, for Photovoltaic active power at any given time for The active power of energy storage at any given time. for The state of charge of the stored energy at any given moment; This is the disturbance estimate. The sampling period is For the rated frequency, Let be the equivalent time constant of photovoltaic power. The equivalent time constant of the energy storage system, The rated installed capacity of the energy storage system. The damping coefficient; for The virtual inertia reference value at time t. for Reference value of photovoltaic active power at any given time. for Reference value of active power of energy storage at any given time; for The equivalent inertial constant at time t. The VSG inertial time constant, The proportion of VSG capacity to the total capacity of the energy storage system. The proportion of grid-connected photovoltaic capacity to the total energy storage system capacity. This refers to the proportion of grid-type energy storage capacity to the total capacity of the energy storage system.

[0039] In this embodiment, the virtual inertia reference value and the active power reference value can be directly processed by the nonlinear discrete prediction model to handle the inherent nonlinear coupling characteristics during rotor motion, thereby achieving the synergistic optimization of the virtual inertia reference value and the active power reference value.

[0040] Specifically, in one embodiment of this application, the multi-objective cost function includes a frequency dynamic performance function, a control cost function, and an operational economy function; the frequency dynamic performance function is used to penalize frequency deviation and frequency change rate; the control cost function is used to penalize control increments to ensure smooth operation of the actuator and prevent frequent fluctuations in virtual inertia reference values ​​and active power reference values; the operational economy function is used to minimize power output deviation to improve economic dispatching effect and delay battery aging.

[0041] Specifically, the multi-objective cost function is expressed as:

[0042] In the formula, For a multi-objective cost function, To predict the time domain, For frequency dynamic performance function, To control the cost function, To run the economy function, For time, For time steps.

[0043] In this embodiment, a multi-objective cost function is constructed, consisting of a frequency dynamic performance function, a control cost function, and an operational economy function. This function achieves synergistic optimization from three dimensions: dynamic stability of power grid frequency, control execution quality, and economic operation of the system. It effectively overcomes the shortcomings of traditional single-objective frequency regulation strategies, such as slow frequency recovery, large control fluctuations, high equipment losses, and poor economic efficiency.

[0044] Specifically, the frequency dynamic performance function is expressed as:

[0045] in:

[0046] In the formula, It is an adaptive weighting factor used to dynamically adjust the penalty intensity for frequency deviation and frequency change rate; This is the frequency deviation weighting coefficient. for Frequency deviation at any given moment; The frequency change rate weighting coefficient, for Rate of change of frequency at time, This is the sensitivity coefficient. The maximum permissible deviation of the nominal frequency. This is the maximum permissible value for the nominal frequency change rate.

[0047] In this embodiment, an adaptive weighting factor is introduced. This adaptive weighting factor is designed based on the trajectory characteristics of the system in the Δf-RoCoF phase plane. By dynamically adjusting the penalty granularity of frequency deviation and frequency change rate through the adaptive weighting factor, the weight is automatically reduced for small-amplitude disturbances to relax the adjustment constraints and improve the adjustment response speed. For high-power impact disturbances, the weight is adaptively increased to strengthen the constraints on frequency deviation and frequency fluctuation rate, suppress continuous frequency shift, accelerate frequency recovery, and solve the problem of slow frequency recovery in traditional methods.

[0048] Specifically, the control cost function is expressed as:

[0049] In the formula, For virtual inertia control incremental weighting coefficients, for The virtual inertia control increment at any given moment; The active power command control incremental weighting coefficient is used. for Active power command control increment at any given time; This is the virtual inertia deviation weighting coefficient. for Virtual inertia at any given moment This is the virtual inertia rating.

[0050] In this embodiment, a penalty is imposed on the control increment by a control cost function to constrain the frequent jumps and large fluctuations between the virtual inertia reference value and the active power reference value. This smooths the output of frequency modulation control commands, avoids reciprocating adjustments and frequent actions of the actuator, reduces equipment electromechanical losses, and extends equipment lifespan; it also eliminates secondary frequency fluctuations induced by control oscillations, ensuring a smooth and continuous adjustment process and guaranteeing stable operation of the actuator.

[0051] Specifically, the operational economy function is expressed as:

[0052] In the formula, This is the active power output weighting coefficient. for The active power of the energy storage module at all times. for The active power of the photovoltaic module at any given time.

[0053] In this embodiment, the operating economy function aims to minimize power output deviation, accurately matching the grid's economic dispatch output demand, reducing the deviation between actual and planned power output, lowering redundant energy consumption and dispatch costs, and improving dispatch operation economy. Simultaneously, it can smooth out fluctuations in high-power, frequent charging and discharging of energy storage, alleviate battery capacity decay, delay equipment aging, reduce maintenance and replacement costs, and achieve a synergistic improvement in frequency regulation performance, equipment lifespan, and operational efficiency.

[0054] Specifically, the active power constraint of the power station's optical storage module is expressed as follows:

[0055] In the formula, for The active power adjustment of the corresponding module at all times. This represents the maximum active power reserve capacity of the module.

[0056] It should be noted that the active power regulation of the photovoltaic and energy storage modules (including photovoltaic modules and energy storage modules) at the power station is limited by the maximum reserve capacity of the photovoltaic and energy storage modules at the power station. Therefore, constraining the active power of the photovoltaic and energy storage modules at the power station within the range of their maximum active power reserve capacity can accurately match the actual adjustable capacity of the photovoltaic and energy storage modules at the power station and effectively avoid problems such as power over-limit during the optimization process.

[0057] Specifically, the SOC constraint of the energy storage module is expressed as follows:

[0058] In the formula, In order to be in Always The predicted value of the state of charge of the energy storage module at any time. This represents the lower limit of the state of charge of the energy storage module. This represents the upper limit of the state of charge of the energy storage module.

[0059] It should be noted that deep charging and discharging of energy storage modules will accelerate the aging of positive and negative electrode materials and the decomposition of electrolyte, resulting in a shortened lifespan of the energy storage module. In order to prevent the energy storage module from being degraded due to overcharging or over-discharging, the predicted SOC value of the energy storage module is constrained within a safe range, deep charging and discharging are avoided, the cycle degradation of the energy storage module is significantly slowed down, the entire life cycle of the energy storage module is extended, and the equipment replacement cost is reduced.

[0060] S2. Estimation steps: Use the state observation model based on Kalman filtering to obtain the smoothed frequency deviation estimate, power disturbance estimate and reconstructed frequency change rate at the current moment.

[0061] Specifically, in one embodiment of this application, see [link to embodiment]. Figure 2 Methods for obtaining the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment using a state observation model based on Kalman filtering include: S21. Model the external power disturbance as a slowly changing augmented state, and define the state vector as follows: The control input is The measurement output is The state-space model based on Kalman filtering is constructed as follows:

[0062] in, ; In the formula, For frequency deviation, For power disturbance, This is the transpose of the matrix. This is the active power command value. This is a frequency measurement value. , For the system matrix, For the output matrix, To conform to the Gaussian white noise distribution of process noise, For measurement noise that conforms to a Gaussian white noise distribution, This is the equivalent damping coefficient. It is the equivalent inertial constant. For time.

[0063] It should be noted that traditional state modeling only considers controllable inputs, and external power disturbances are attributed to model errors, resulting in a mismatch between the model and the actual power grid dynamics. In this embodiment, external power disturbances are expanded to represent system states, transforming unknown disturbances into filterable and observable state variables, rather than treating them as random model noise. This reduces model mismatch errors, comprehensively depicts all power grid dynamics (controllable regulation + unknown external disturbances), and improves the fitting accuracy of the state model to the actual power grid dynamics. Simultaneously, slow power disturbances that were previously impossible to measure directly can be estimated, enabling real-time online observation of disturbances.

[0064] S22. Perform time and measurement updates using Kalman filtering on the state-space model, and recursively update to calculate the optimal estimate, obtaining the optimal state estimate. , This is the estimated value of the frequency deviation. This is an estimate of the power disturbance.

[0065] The time update is represented as:

[0066] In the formula, for The prior state estimate at time t. for The posterior state estimate at time t. for The prior error covariance matrix at time t. for The posterior error covariance matrix at time t; for The system matrix (i.e., the target research object) after time discretization. for The input matrix after time discretization. For process noise The covariance matrix.

[0067] It should be noted that the prior error covariance matrix and posterior error covariance matrix The initial value of the posterior error covariance matrix is ​​calculated in real time and updated on a rolling basis using the Kalman filter recursive formula; while the initial value of the posterior error covariance matrix is... In this embodiment of the application, the pre-defined positive definite matrix is ​​specifically initialized as follows:

[0068] Process noise covariance matrix In this embodiment of the application, the pre-designed symmetric positive definite matrix is ​​specifically set as a diagonal matrix:

[0069] System Matrix and input matrix It is based on The system parameters, which are updated in real time, are calculated and discretized. Specifically, the discretized system matrix is ​​obtained by using the first-order forward Euler method. and input matrix They are:

[0070]

[0071] In the formula, for The equivalent inertial constant of the power grid at time t. This is the equivalent damping coefficient. The system controls the sampling period (which is set to 0.1s in this embodiment).

[0072] The measurement update is represented as follows:

[0073] In the formula, for Kalman gain at time step For measuring noise The covariance matrix, for The posterior state estimate at time t. for The posterior error covariance matrix at time t. It is an identity matrix.

[0074] Specifically, the measurement noise covariance matrix It is a pre-defined matrix. Since the system's measurement output (frequency deviation) in this embodiment is a one-dimensional scalar, the measurement noise covariance matrix... It is expressed as a one-dimensional scalar constant. In the embodiments of this application, its specific value is set as follows: .

[0075] In this embodiment, time updates rely on system electromechanical dynamics for state prediction, utilizing historical dynamic patterns to smooth instantaneous noise. This eliminates on-site sensor noise, transmission glitches, and pulse interference. Power grid frequency and power sensors commonly exhibit Gaussian noise and instantaneous jumps; recursive filtering can suppress random noise, preventing it from propagating to subsequent frequency change calculations. Measurement updates introduce real-time measured data to correct prediction biases. Confidence weights for prediction and measurement are dynamically allocated using the covariance matrix, recursively minimizing the mean square error of state estimation and outputting the statistically optimal state value, simultaneously correcting augmented disturbance states. Through time and measurement updates, a joint optimal estimation of disturbance and electromechanical state is achieved.

[0076] S23. Using the optimal state estimate and the physical constraints of the rocking equation, a smooth and physically consistent rate of frequency change is reconstructed, expressed as:

[0077] In the formula, for The rate of change of the reconstructed frequency at any given moment for The equivalent inertial constant at time t. for The input is controlled at all times, namely the active power command value.

[0078] In this embodiment, the noiseless optimal state obtained by Kalman filtering is substituted into the swing equation to reconstruct the frequency change rate. The frequency change rate is limited by the hard constraint of power grid electromechanical dynamics, and the frequency change rate observation is accurate and smooth. While eliminating measurement noise, it ensures consistency with the power balance equation.

[0079] S3. Prediction Step: Input the frequency deviation estimate, power disturbance estimate and reconstructed frequency change rate into the nonlinear model predictive control optimization model to obtain the virtual inertia reference value and active power reference value at the current moment.

[0080] Specifically, in one embodiment of this application, see [link to embodiment]. Figure 3 Methods for obtaining the current virtual inertia reference value and active power reference value include: S31. Predicting state trajectory based on a nonlinear discrete prediction model; S32. Under the conditions of satisfying the active power constraints of the photovoltaic storage module and the SOC constraints of the energy storage module, a nonlinear programming solver is used to minimize the objective function in the prediction time domain to generate the control increment sequence. S33. Use the first element of the control increment sequence as the virtual inertia reference value and active power reference value at the current moment.

[0081] It should be noted that a nonlinear programming (NLP) solver is a set of programs / software libraries that rely on numerical algorithms to solve nonlinear programming mathematical optimization problems.

[0082] In this embodiment, the safety boundaries of active power and SOC (State of Charge) of the energy storage module are directly embedded into the predictive control optimization as hard constraints. When the system approaches its operating limit, the power regulation burden can be automatically and smoothly transferred, effectively preventing the over-discharge of the energy storage battery and ensuring the safety of the site equipment under extreme conditions.

[0083] In one embodiment of this application, the method further includes: at the next sampling time, returning to step S2, and continuing to predict the virtual inertia reference value and active power reference value at the next time.

[0084] To verify the effectiveness of the model predictive frequency control method for the coordinated inertia and power of new energy power plants described in this application, this paper focuses on... Figure 4 The diagram shows a heterogeneous renewable energy power plant. Taking the improved IEEE 9-node system as an example, experimental testing was conducted, and a hardware-in-the-loop real-time test platform based on the improved IEEE 9-node system was constructed. The experimental platform mainly includes a real-time simulator and a digital controller. The real-time simulator is used to simulate the millisecond-level dynamic response of the main power grid and renewable energy power plants. The digital controller interacts with the real-time simulator through analog / digital I / O interfaces, acquiring grid-connected voltage, current, and frequency signals in real time, and performing predictive optimization of the nonlinear predictive control optimization model proposed in this application.

[0085] See Figure 5 The improved IEEE 9-node system consists of three parts: conventional synchronous generator sets, grid-connected photovoltaic units employing maximum power point tracking (MPPT) control, and grid-connected photovoltaic-storage units using the method described in this application. Key parameters of the improved IEEE 9-node system are shown in Table 1.

[0086] Table 1 Key parameters of the improved IEEE 9-node system

[0087] In Table 1, As the system's baseline capacity, The proportion of VSG capacity to the total capacity of the energy storage system. To determine the proportion of grid-connected photovoltaic capacity to the total energy storage system capacity, The proportion of grid-connected photovoltaic capacity to the total energy storage system capacity. The proportion of grid-type energy storage capacity to the total capacity of the energy storage system. This refers to the rated installed capacity of the energy storage system. The VSG inertial time constant, This refers to the proportion of the steam turbine's output power to its total steam turbine output power. This represents the product of the turbine's mechanical power gain coefficient and droop coefficient. As the reference inertial time constant, The virtual inertial time constant of VSG. The damping coefficient of the GFM converter. This refers to the frequency regulation coefficient of the GFM converter; This is the inherent delay of the GFM converter. Let be the equivalent time constant of photovoltaic power. The equivalent time constant of the energy storage system, The sampling period is This is the virtual inertia deviation weighting coefficient. This is the active power output weighting coefficient; This is the frequency deviation weighting coefficient. The frequency change rate weighting coefficient; For virtual inertia control incremental weighting coefficients, The active power command control incremental weighting coefficient; This is the sensitivity coefficient. N p For prediction in the time domain.

[0088] To comprehensively evaluate the control performance of the method proposed in this application, four sets of comparative experiments were designed. The specific definitions of each control strategy are as follows: Strategy 1: The model predictive frequency control method for coordinating inertia and power in new energy power plants proposed in this application combines virtual inertia reference values ​​and active power reference values ​​(including active power reference values ​​for energy storage). and photovoltaic active power reference value It is incorporated into a unified optimization control for collaborative solution.

[0089] Strategy 2: Traditional MPC strategy, which fixes the virtual inertia to the nominal value and only optimizes the active power output of the photovoltaic storage unit.

[0090] Strategy 3: Adaptive inertia combined with traditional MPC strategy. The virtual inertia is independently adjusted based on the segmentation rules of the frequency change rate RoCoF and the frequency deviation signal, and does not participate in the optimization solution.

[0091] Strategy 4: Traditional VSG control strategy, which provides frequency support by increasing the primary frequency modulation coefficient.

[0092] To verify the dynamic performance of the control method described in this application, a step load was applied at t=1s to increase the disturbance. Figure 6-9 Table 2 shows a comparison of the system dynamic response under four control strategies.

[0093] Table 2 Comparison of system dynamic response under step load disturbance

[0094] In Table 2, f min For the minimum frequency, Δ f max For the maximum frequency deviation, t s To adjust the system time, H max This represents the maximum equivalent virtual inertia.

[0095] Experimental results show that the control method described in this application has significant advantages in frequency transient support and dynamic recovery. Under the same load impact, the control method described in this application issues the instantaneous active power injection command earlier to curb the increase of frequency deviation. Maximum frequency deviation Δ f max Compared to existing conventional strategies, the reductions were 7.1%, 3.6%, and 8.1% respectively, effectively suppressing system frequency drops. Thanks to the predictive time-domain advantages and time-varying inertia adjustment mechanism of the nonlinear model predictive control optimization model, the system settling time was significantly reduced. t s The recovery time is reduced to 5.49 seconds. Compared to heuristic adaptive adjustment, the control method described in this application can accurately predict trends and rapidly reduce virtual inertia during the frequency recovery period, overcoming the drawback of slow recovery and enabling the system to return to steady state more quickly and smoothly.

[0096] To evaluate the robustness of the proposed method under real meteorological conditions, a system was constructed as follows: Figure 10 The cloud disturbance model is shown. See also... Figure 10 The simulated photovoltaic array encountered cloud cover between t=1s and t=9s, resulting in a power reduction of about 20MW, and was superimposed with random high-frequency noise. Figures 11-12 The system dynamic response of four control strategies under random photovoltaic fluctuations is compared. Table 3 details the frequency performance statistics of each strategy.

[0097] Table 3 Comparison of System Dynamic Responses under Random Photovoltaic Fluctuations

[0098] In Table 3, STD represents the standard deviation of frequency.

[0099] Experimental results show that the control method described in this application will reduce the maximum frequency deviation Δ f maxThe frequency deviation is limited to 0.1837 Hz. Compared with traditional fixed parameter control and heuristic adaptive control, the maximum frequency deviation is reduced by approximately 11.9% and 4.1%, respectively, verifying that the look-ahead prediction capability of the control method described in this application can more accurately coordinate the planning of photovoltaic power generation and effectively improve the system safety margin. In terms of random noise suppression, the frequency standard deviation (STD) of the control method described in this application is only 0.0767, and the waveform is the smoothest. Traditional heuristic strategies, due to their high dependence on the frequency differential signal, are susceptible to high-frequency noise interference, leading to violent oscillations of virtual inertia and increasing equipment losses. In contrast, the control method described in this application effectively isolates local random noise through the control increment penalty mechanism in the objective function and optimal state observation. While avoiding ineffective high-frequency adjustment, it achieves smooth adaptive adjustment of virtual inertia, balancing transient frequency support effect and the operational stability of the underlying actuator.

[0100] To verify the constraint handling capability of the control method described in this application under the extreme state of energy storage capacity, an extreme test condition was set: the initial SOC of the energy storage was set to 0.202, approaching the preset lower limit SOC. min =0.2. Under this state, a power notch perturbation is applied, and the experimental results are as follows: Figure 13-14 As shown.

[0101] In the initial stage of disturbance, since the energy storage SOC has not yet reached the safe lower limit, the control strategy prioritizes smoothing out frequency deviations. The system rapidly increases virtual inertia and energy storage output to curb the rate of frequency change, and the SOC decreases rapidly. As the SOC approaches and smoothly stabilizes at the safe lower limit, the inequality boundary constraints within the optimized control of the control method described in this application are automatically activated. At this time, the control method described in this application uses the predictive time domain to anticipate the risk of exceeding limits, dynamically adjusts the optimization priority, actively reduces the energy storage reference power, and simultaneously increases the photovoltaic reference power to maintain system power balance. The above dynamic process shows that the control method described in this application has a strict hard constraint handling capability. Under extreme operating conditions, the control method described in this application can automatically shift the control priority to the equipment safety boundary, achieve smooth coordination of photovoltaic and energy storage output, effectively prevent over-discharge of energy storage batteries, and effectively ensure the physical safety and long-term operating life of the underlying equipment.

[0102] In summary, this application proposes a model predictive frequency control method for new energy power plants that coordinates inertia and power. It constructs a unified predictive control optimization model, directly addresses the nonlinear coupling characteristics of the rotor motion equations based on real-time state estimation and rolling optimization, and achieves coordinated optimization of virtual inertia and active power commands. This significantly reduces the maximum frequency deviation and frequency change rate under step disturbances; simultaneously, it effectively eliminates high-frequency oscillations in the control signal under random noise environments, greatly improving the transient stability and robustness of the system. By directly internalizing the physical boundary constraints of the equipment into the optimization solution process, it ensures that the system can automatically and smoothly execute power reduction under extreme operating conditions, guaranteeing frequency support performance while strictly maintaining the operational safety of the underlying hardware.

[0103] The above embodiments are used to explain this application, not to limit it. Any modifications and changes made to this application within the spirit and scope of the claims shall fall within the protection scope of this application.

Claims

1. A new energy station inertia and power coordinated model predictive frequency control method, characterized in that, include: Model construction steps: Using a nonlinear discrete prediction model as the base model, a multi-objective cost function with multiple objectives of improving frequency dynamic performance and optimizing operational economy is constructed as the objective function. The active power constraints of the solar-storage modules and the SOC constraints of the energy storage modules are used as constraints to construct a nonlinear model predictive control optimization model based on time-varying weights. The nonlinear discrete prediction model is expressed as follows: in: In the formula, for Frequency deviation at time, for Photovoltaic active power at any given time for The active power of energy storage at any given time. for The state of charge of the stored energy at any given moment; for Frequency deviation at time, for Photovoltaic active power at any given time for The active power of energy storage at any given time. for The state of charge of the stored energy at any given moment; This is the disturbance estimate. The sampling period is For the rated frequency, Let be the equivalent time constant of photovoltaic power. The equivalent time constant of the energy storage system, The rated installed capacity of the energy storage system. The damping coefficient; for The virtual inertia reference value at time t. for Reference value of photovoltaic active power at any given time. for Reference value of active power of energy storage at any given time; for The equivalent inertial constant at time t. The VSG inertial time constant, The proportion of VSG capacity to the total capacity of the energy storage system. The proportion of grid-connected photovoltaic capacity to the total energy storage system capacity. The proportion of grid-type energy storage capacity to the total capacity of the energy storage system Estimation steps: Use a state observation model based on Kalman filtering to obtain the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment; Prediction steps: Input the frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate into the nonlinear model predictive control optimization model to obtain the virtual inertia reference value and active power reference value at the current moment.

2. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 1, characterized in that, In the model construction step, the multi-objective cost function includes a frequency dynamic performance function, a control cost function, and an operational economy function. The frequency dynamic performance function is used to penalize frequency deviation and frequency change rate. The control cost function is used to penalize control increments to ensure smooth operation of the actuator and prevent frequent fluctuations in virtual inertia reference values ​​and active power reference values. The operational economy function is used to minimize power output deviation to improve economic dispatching effectiveness and delay battery aging. The multi-objective cost function is expressed as follows: In the formula, For a multi-objective cost function, To predict the time domain, For frequency dynamic performance function, To control the cost function, To run the economy function, For time, For time steps.

3. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 2, characterized in that, The frequency dynamic performance function is expressed as follows: in: In the formula, It is an adaptive weighting factor used to dynamically adjust the penalty intensity for frequency deviation and frequency change rate; This is the frequency deviation weighting coefficient. for Frequency deviation at any given moment; The frequency change rate weighting coefficient, for Rate of change of frequency at time, This is the sensitivity coefficient. The maximum permissible deviation of the nominal frequency. This is the maximum permissible value for the nominal frequency change rate.

4. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 2, characterized in that, The control cost function is expressed as: In the formula, For virtual inertia control incremental weighting coefficients, for The virtual inertia control increment at any given moment; The active power command control incremental weighting coefficient is used. for Active power command control increment at any given time; This is the virtual inertia deviation weighting coefficient. for Virtual inertia at any given moment This is the virtual inertia rating.

5. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 2, characterized in that, The operational economy function is expressed as follows: In the formula, This is the active power output weighting coefficient. for The active power of the energy storage module at all times. for The active power of the photovoltaic module at any given time.

6. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 1, characterized in that, In the model construction steps, the active power constraint of the power station's photovoltaic storage module is expressed as follows: In the formula, for The active power adjustment of the corresponding module at all times. This represents the maximum active power reserve capacity of the module. The SOC constraint of the energy storage module is expressed as follows: In the formula, In order to be in Always The predicted value of the state of charge of the energy storage module at any time. This represents the lower limit of the state of charge of the energy storage module. This represents the upper limit of the state of charge of the energy storage module.

7. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 1, characterized in that, In the estimation step, the methods for obtaining the smoothed frequency deviation estimate, power disturbance estimate, and reconstructed frequency change rate at the current moment using a state observation model based on Kalman filtering include: The external power disturbance is modeled as a slowly changing augmented state, and the state vector is defined as follows: The control input is The measurement output is The state observation model based on the Kalman filter is constructed as follows: in, ; In the formula, For frequency deviation, For power disturbance, This is the transpose of the matrix. This is the active power command value. This is a frequency measurement value. , For the system matrix, For the output matrix, To conform to the Gaussian white noise distribution of process noise, For measurement noise that conforms to a Gaussian white noise distribution, This is the equivalent damping coefficient. It is the equivalent inertial constant. For time; Perform time and measurement updates using Kalman filtering on the state-space model, recursively update and calculate the optimal estimate to obtain the optimal state estimate. , This is the estimated value of the frequency deviation. This is an estimate of the power disturbance; The time update is represented as: In the formula, for The prior state estimate at time t. for The posterior state estimate at time t. for The prior error covariance matrix at time t. for The posterior error covariance matrix at time t; The system matrix is ​​the discretized matrix. The input matrix is ​​discretized. For process noise The covariance matrix; The measurement update is represented as follows: In the formula, for Kalman gain at time step For measuring noise The covariance matrix, for The posterior state estimate at time t. for The posterior error covariance matrix at time t. It is the identity matrix; Using the optimal state estimate and the physical constraints of the rocking equation, a smooth and physically consistent rate of frequency change is reconstructed, expressed as: In the formula, for The rate of change of the frequency of reconstruction at any given moment for The equivalent inertial constant at time t. for The input is controlled at all times, namely the active power command value.

8. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 1, characterized in that, Methods for obtaining the current virtual inertia reference value and active power reference value include: State trajectory prediction based on a nonlinear discrete prediction model; Under the conditions of satisfying the active power constraints of the photovoltaic storage module and the SOC constraints of the energy storage module, a nonlinear programming solver is used to minimize the objective function in the prediction time domain to generate the control increment sequence. The first element of the control increment sequence is used as the virtual inertia reference value and active power reference value at the current moment.

9. The model predictive frequency control method for coordinating inertia and power in new energy power stations as described in claim 8, characterized in that, The method further includes: at the next sampling time, returning to the estimation step and continuing to predict the virtual inertia reference value and active power reference value at the next time step.

Images