Generator excitation regulation method and system based on digital predictive control

Abstract

The application discloses a generator excitation regulation method and system based on digital predictive control, which establishes a seven-order state space mathematical model of a generator, and adopts an extended Kalman filter to perform real-time state estimation; a model predictive control algorithm is executed based on a rolling horizon optimization strategy, and an optimal control sequence is solved through quadratic programming; a multi-disturbance observer is integrated to estimate external disturbances in real time and perform feedforward compensation; a recursive least square method is adopted to identify generator parameters on line; and a double-channel control structure is designed to realize coordinated control of fast response and accurate regulation. The application shortens the system response time to below 50 ms, controls the steady-state voltage deviation within ±0.5%, and significantly improves the dynamic response performance and control accuracy of the generator excitation regulation system, and enhances the stability of the power system.

Description

Technical Field

[0001] This invention relates to the field of power system control technology, and in particular to a generator excitation regulation method and system based on digital predictive control, applicable to voltage regulation and stability control of large generator sets. Background Technology

[0002] The generator excitation regulation system is a key device in a power system for maintaining voltage stability and improving system dynamic performance, and its control performance directly affects the safe and stable operation of the power grid. With the continuous expansion of the power system and the increase in the proportion of new energy integration, higher requirements are placed on the response speed and control accuracy of the generator excitation regulation system.

[0003] Traditional generator excitation regulation systems mainly employ analog or simple digital controllers, typically based on PID control strategies to achieve voltage regulation. These systems have a relatively simple structure, but their control algorithms are rather simplistic and difficult to adapt to complex grid operating conditions. Another type of more advanced excitation regulator uses digital control technology, implementing more complex control algorithms through microprocessors, but it still relies on traditional linear control theory.

[0004] The most typical excitation regulation system in the existing technology adopts the proportional-integral-derivative (PID) control principle. By detecting the deviation between the generator terminal voltage and the setpoint, and performing calculations through proportional, integral, and derivative stages, it outputs an excitation regulation signal to control the magnitude of the excitation current. The control principle of this method is based on linear control with error feedback, and the dynamic response characteristics of the system are improved by adjusting the PID parameters.

[0005] However, existing technologies have the following key problems: First, the response speed is slow. The response time of traditional PID control is usually between 100-200ms, which is difficult to meet the rapid adjustment requirements during power grid faults. Second, the control accuracy is low. The steady-state voltage deviation reaches ±1-2%, which can easily cause voltage instability during load changes or power grid disturbances, affecting the safe and stable operation of the power system. Summary of the Invention

[0006] The purpose of this invention is to provide a generator excitation regulation method and system based on digital predictive control, which aims to solve the technical problems of slow response speed and low control accuracy of existing excitation regulation systems, and realize fast response and high-precision control of generator excitation systems.

[0007] To achieve the above objectives, the technical solution provided by the present invention is as follows: A generator excitation regulation method based on digital predictive control includes the following steps: Real-time operating data of generator terminal voltage, stator current and speed are acquired, and a seventh-order state-space mathematical model is established, including stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity and power angle. Real-time state estimation is performed through extended Kalman filter to obtain high-precision system state estimation information. Based on the system state estimation information, a rolling time-domain optimization strategy is adopted to execute the model predictive control algorithm. The prediction time domain is set to 20 control cycles and the control time domain is set to 5 control cycles. An objective function containing the output weight matrix and the control weight matrix is ​​established, and the optimal control sequence is calculated by a quadratic programming solution method. Real-time monitoring of grid frequency changes, load power disturbances and system parameter drift is performed. External disturbance information is estimated through a state-extended dimension disturbance observer. The disturbance information is fed forward into the prediction model to correct the prediction equation and obtain the prediction result after disturbance compensation. Based on the parameterized model, the recursive least squares method is used to identify the synchronous reactance, transient reactance, subtransient reactance and time constant of the generator online, and the model parameters are updated periodically to obtain adaptively updated model parameters; Based on the adaptively updated model parameters, a dual-channel control structure is used for coordinated control. The fast channel achieves fast transient response through pole placement, while the precise channel eliminates steady-state error through integral control. The weighting coefficients of the two channels are dynamically adjusted according to the voltage deviation to obtain the final excitation control signal.

[0008] The establishment of a seventh-order state-space mathematical model, including stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity, and power angle, and the real-time state estimation using an extended Kalman filter, yields high-precision system state estimation information, including: Based on the electromagnetic and mechanical parameters of the generator, a state vector x=[id,iq,if,ikd,ikq,ω,δ]T is constructed. The system matrix A and input matrix B are determined according to the synchronous reactance, transient reactance, subtransient reactance and rotational inertia, and a standardized state-space model is obtained. The standardized state-space model is subjected to nonlinear state estimation processing. An extended Kalman filter is used, and the high-precision system state estimation information is obtained by recursive calculation through the state prediction equation and the state update equation.

[0009] The model predictive control algorithm employs a rolling time-domain optimization strategy, setting the prediction time domain to 20 control cycles and the control time domain to 5 control cycles. An objective function is established, including an output weight matrix and a control weight matrix. The optimal control sequence is calculated using a quadratic programming solution method, including: Based on the system state estimation information, an objective function including an output tracking term and a control increment penalty term is established, and the output weight matrix Q and the control weight matrix R are set to obtain the complete optimization objective function. Constraints are set for the objective function, including excitation current constraints, excitation voltage constraints, and rotor angular velocity constraints. The constrained optimization problem is transformed into a quadratic programming problem, which is solved using the effective set method to obtain the optimal control sequence.

[0010] The process of estimating external disturbance information using a state-extended perturbation observer, feeding this disturbance information forward into the prediction model to correct the prediction equation, and obtaining the disturbance-compensated prediction result includes: Real-time monitoring and classification of power grid frequency changes, load power disturbances and system parameter drifts are performed. A state-extended disturbance observer is designed to construct extended state equations by treating unknown disturbances as extended states, thereby obtaining an extended state model containing disturbance information. For the extended state model, the observer equation is used to calculate the estimated values ​​of various disturbances. The frequency disturbance, power disturbance and parameter disturbance information are fed forward to the corresponding terms of the prediction equation to obtain the prediction result after disturbance compensation.

[0011] The recursive least squares method is used to identify the synchronous reactance, transient reactance, subtransient reactance, and time constant of the generator online, and the model parameters are updated periodically to obtain adaptively updated model parameters, including: Based on the parameterized model y(k)=φᵀ(k)θ(k)+e(k), a parameter vector θ(k) containing synchronous reactance Xd, transient reactance X'd, subtransient reactance X"d and time constants T'd0 and T"d0 is constructed to obtain the complete parameter identification model. The parameter identification model is processed by a recursive least squares algorithm. Through the recursive update of the gain matrix K(k) and covariance matrix P(k), the identified parameters are periodically updated to the coefficient matrix A and input matrix B of the state space model to obtain the adaptively updated model parameters.

[0012] The system employs a dual-channel control structure for coordinated control. The fast channel achieves rapid transient response through pole placement, while the precision channel eliminates steady-state error through integral control. The weighting coefficients of the two channels are dynamically adjusted based on the voltage deviation to obtain the final excitation control signal, including: Based on the adaptively updated model parameters, a fast-path controller is designed. The pole placement method is used to place the dominant pole of the closed-loop system at the desired position. The fast transient response is achieved through the state feedback control law uf(k)=-Kfx(k), and the fast-path control output uf(k) is obtained. To address the deviation between the voltage setpoint and the actual voltage, a precise channel controller is designed. An integral control law ui(k) = ui(k-1) + Ki[yr(k) - y(k)] is used to process the deviation, eliminating steady-state error and obtaining the precise channel control output ui(k). Weights are assigned to the fast channel control output uf(k) and the precise channel control output ui(k). The weight coefficient α is dynamically adjusted according to the magnitude of the voltage deviation |yr(k)-y(k)|. When the deviation is large, α is set to favor fast response, and when the deviation is small, α is set to favor precise adjustment, thus obtaining the dynamically adjusted weight coefficient α. Based on the dynamically adjusted weight coefficient α, a dual-channel coordinated control process is performed using the weighted synthesis algorithm u(k)=αuf(k)+(1-α)ui(k) to achieve an organic combination of fast response and high-precision adjustment, thereby obtaining the final excitation control signal.

[0013] The present invention also provides a generator excitation regulation system based on digital predictive control, comprising: The data acquisition module is used to acquire real-time operating data of generator terminal voltage, stator current, and speed; The state estimation module is used to establish a seventh-order state-space mathematical model that includes stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity, and power angle, and performs real-time state estimation through an extended Kalman filter. The predictive control module is used to execute the model predictive control algorithm based on system state estimation information and employ a rolling time-domain optimization strategy to calculate the optimal control sequence using a quadratic programming solution method. The disturbance observation module is used to monitor grid frequency changes, load power disturbances and system parameter drift in real time, and to estimate external disturbance information through the state-extended disturbance observer; The parameter identification module is used to identify generator parameters online based on a parameterized model and using the recursive least squares method. The dual-channel control module is used to coordinate control based on adaptively updated model parameters using a dual-channel control structure to obtain the final excitation control signal.

[0014] The beneficial effects of this invention are: By establishing a seventh-order state-space mathematical model of the generator and using an extended Kalman filter for real-time state estimation, the accuracy of obtaining system state information is improved, providing accurate basic data for subsequent control. By employing a model predictive control algorithm with rolling time-domain optimization, the system's future behavior can be predicted and optimized, significantly improving the system's dynamic response performance. By integrating multiple disturbance observers and feedforward compensation techniques, the system's ability to suppress external disturbances is enhanced, and the robustness of the control system is improved. Online parameter identification was performed using the recursive least squares method, which enabled adaptive updating of model parameters and allowed the control system to adapt to changes in generator parameters. An innovative dual-channel control structure was designed, which organically combines fast response and precise adjustment, achieving excellent performance with a response time of less than 50ms and a steady-state voltage deviation controlled within ±0.5%. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the generator excitation regulation method based on digital predictive control according to the present invention. Figure 2 This is a schematic diagram of the seventh-order state-space model of the generator of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0018] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0019] In this document, the term "and / or" merely describes a relationship, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0020] 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, not all, of the embodiments of the present invention. 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.

[0021] like Figure 1 As shown, the generator excitation regulation method based on digital predictive control provided by the present invention includes the following steps: Step S1: Obtain real-time operating data of generator terminal voltage, stator current and speed, establish a seventh-order state-space mathematical model including stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity and power angle, and perform real-time state estimation through extended Kalman filter to obtain high-precision system state estimation information.

[0022] Based on the electromagnetic and mechanical parameters of the generator, a state vector x=[id,iq,if,ikd,ikq,ω,δ]T is constructed. The system matrix A and input matrix B are determined according to the synchronous reactance, transient reactance, subtransient reactance and rotational inertia, thus obtaining a standardized state-space model.

[0023] like Figure 2 As shown, the seventh-order state-space model of the generator includes four parts: stator circuit, excitation circuit, damping winding circuit, and mechanical motion equations. The typical values ​​of the system matrix A are: A(1,1)=-Rs / Ld=-0.015 / 0.18=-0.083, A(2,2)=-Rs / Lq=-0.083, A(3,3)=-Rf / Lf=-0.001 / 0.165=-0.006, where Rs=0.015pu is the stator resistance and Rf=0.001pu is the excitation winding resistance.

[0024] The standardized state-space model is subjected to nonlinear state estimation processing using an extended Kalman filter. High-precision system state estimation information is obtained through recursive calculations using the state prediction equation and the state update equation. The state prediction equation of the extended Kalman filter is x(k+1|k)=f[x(k|k),u(k)]+w(k), where f(·) is the nonlinear state transition function and w(k) is the process noise.

[0025] The state update equation is: x(k+1|k+1)=x(k+1|k)+K(k+1)[y(k+1)-h(x(k+1|k))] Where h(·) is the nonlinear observation function, and the Kalman gain is: K(k+1)=P(k+1|k)HT(k+1)[H(k+1)P(k+1|k)HT(k+1)+R(k+1)]-1 The covariance matrix P(k+1|k) is recursively updated using the Jacobian matrix F(k) and the observation matrix H(k+1). The process noise covariance matrix Q is set to diag[0.01,0.01,0.05,0.02,0.02,0.001,0.001], and the observation noise covariance matrix R is set to diag[0.001,0.001,0.0001]. System state estimation information with an accuracy of over 99.5% is obtained through recursive calculation.

[0026] Specifically, the high-precision system state estimation information obtained by employing an extended Kalman filter and recursively calculating through state prediction equations and state update equations includes: For the standardized state-space model, the process noise covariance matrix Q and the observation noise covariance matrix R are set. The matrix parameters are determined according to the system dynamic characteristics and measurement accuracy to obtain the filter noise characteristic configuration. The extended Kalman filter is linearized by calculating the linearization matrix of the nonlinear state transition function using the Jacobian matrix F(k) and the linearization matrix of the nonlinear observation function using the observation matrix H(k+1), thus obtaining the linearization parameters at the current time step. Based on the filter noise characteristics configuration and linearization parameters, calculate the Kalman gain: K(k+1)=P(k+1|k)HT(k+1)[H(k+1)P(k+1|k)HT(k+1)+R(k+1)]-1 The optimal estimation is achieved by recursively updating the error covariance matrix P(k+1|k), thus obtaining the error covariance information of the state estimate. The error covariance information of the state estimation is recursively updated. State prediction and state update steps are performed in each control cycle. High-precision state estimation is achieved through filtering algorithm to obtain real-time state estimation values ​​including stator current id, iq, excitation current if, angular velocity ω and power angle δ, thus obtaining the high-precision system state estimation information.

[0027] This invention further optimizes the state estimation process. For the standardized state-space model, the system sets the process noise covariance matrix Q=diag[0.01,0.01,0.05,0.02,0.02,0.001,0.001] and the observation noise covariance matrix R=diag[0.001,0.001,0.0001] based on the generator's dynamic characteristics and measurement accuracy. This configuration takes into account the relatively high noise levels of the excitation current if and the damping winding currents ikd and ikq, while the noise levels of the angular velocity ω and the power angle δ are relatively low, enabling the filter to have differentiated filtering capabilities for different state variables.

[0028] Linearizing the extended Kalman filter is a crucial step in ensuring its effective operation in nonlinear systems. In each control cycle, the system calculates the Jacobian matrix F(k) = ∂f / ∂x|x = x(k|k), u = u(k) of the nonlinear state transition function f(x,u) at the current operating point, and the linearization matrix H(k+1) = ∂h / ∂x|x = x(k+1|k) of the nonlinear observation function h(x). For example, when the generator switches from rated operating condition of 280MW to partial load of 220MW, the element F(3,7) of the Jacobian matrix F(k) changes from 0.056 to 0.043, reflecting the change in the dynamic effect of the power angle change on the excitation current.

[0029] Based on the filter noise characteristics and linearization parameters, the system calculates the Kalman gain K(k+1) = P(k+1|k)HT(k+1)[H(k+1)P(k+1|k)HT(k+1)+R(k+1)]⁻¹, where the error covariance matrix P(k+1|k) is recursively updated using the formula P(k+1|k) = F(k)P(k|k)FT(k)+Q(k). In actual operation, the initial covariance matrix P(0|0) is set to 0.1×I⁷×⁷. After approximately 50 control cycles, the covariance matrix converges to a stable value, and the state estimation error is significantly reduced. In particular, the estimation error of the excitation current if is reduced from the initial 5% to below 0.5%, and the estimation error of the angular velocity ω is controlled within 0.01%.

[0030] Within each control cycle, the system performs two steps: state prediction and state update. The state prediction step uses the current control input and state estimates to predict the state at the next moment through a nonlinear state transition function; the state update step then corrects the prediction results using actual measurements. For example, when a sudden change in terminal voltage is detected, the filtering algorithm can accurately estimate the changes in the excitation current *if* and the power angle *δ* within 3ms, providing timely and accurate state information for subsequent control. Ultimately, the system obtains high-precision real-time estimates of seven state variables, including stator current *id*, *iq*, excitation current *if*, angular velocity *ω*, and power angle *δ*, with an estimation accuracy exceeding 99.5%, meeting the requirements of high-performance control.

[0031] Step S2: Based on the system state estimation information, a rolling time-domain optimization strategy is adopted to execute the model predictive control algorithm. The prediction time domain is set to 20 control cycles and the control time domain is set to 5 control cycles. An objective function containing the output weight matrix and the control weight matrix is ​​established, and the optimal control sequence is calculated by a quadratic programming solution method.

[0032] Based on the system state estimation information, an objective function including an output tracking term and a control increment penalty term is established. The output weight matrix Q and the control weight matrix R are set, resulting in the complete optimization objective function. Objective function: J=∑i=1~Np||y(k+i|k)-yr(k+i)||²Q+∑j=0~Nc-1||Δu(k+j)||²R Where Np=20 is the prediction time domain, Nc=5 is the control time domain, Q is the output weight matrix, and R is the control weight matrix.

[0033] Constraints are set for the objective function, including excitation current constraint -Ifmax≤If≤Ifmax, excitation voltage constraint -Ufmax≤Uf≤Ufmax, and rotor angular velocity constraint ωmin≤ω≤ωmax. This transforms the constrained optimization problem into a quadratic programming problem, which is solved using the effective set method to obtain the optimal control sequence. Within each control cycle, based on the current state and the prediction model, the system output response for the next 20 cycles is predicted, and the optimal control sequence u*(k), u*(k+1), ..., u*(k+Nc-1) is calculated. The first control variable u*(k) is taken as the control output for the current cycle.

[0034] The addition of excitation current constraints, excitation voltage constraints, and rotor angular velocity constraints transforms the constraint optimization problem into a quadratic programming problem, which is solved using the effective set method to obtain the optimal control sequence, including: For the aforementioned optimization objective function, upper and lower limit constraints for excitation current -Ifmax≤If≤Ifmax, upper and lower limit constraints for excitation voltage -Ufmax≤Uf≤Ufmax, and operating range constraints for rotor angular velocity ωmin≤ω≤ωmax are set to obtain a complete set of constraints. The set of constraints is mathematically transformed to construct the inequality constraint matrix Au≤b and the equality constraint matrix Aequ=beq, thus transforming the original constrained optimization problem into the standard quadratic programming problem min(1 / 2)uTHu+fTu, resulting in the standardized quadratic programming problem. For the standardized quadratic programming problem, the effective set method is used for iterative solution. By dynamically identifying and updating the active constraint set and verifying the optimality condition of KKT, the optimal control sequence in the future control time domain is calculated in each control cycle, resulting in a multi-step optimized control sequence containing u*(k) to u*(k+Nc-1). The multi-step optimized control sequence is subjected to rolling optimization processing. The first control quantity u*(k) is extracted as the actual control output of the current cycle. The optimization problem is solved again in the next control cycle to ensure the real-time performance and rolling optimization characteristics of the control, and the optimal control sequence is obtained.

[0035] This invention employs a highly efficient quadratic programming algorithm. For the objective function, the system sets strict constraints: excitation current constraint -3000A ≤ If ≤ 3000A, considering the thermal capacity limitation of the excitation winding; excitation voltage constraint -400V ≤ Uf ≤ 400V, considering the output capability of the excitation power supply; rotor angular velocity constraint 0.95ωn ≤ ω ≤ 1.05ωn, ensuring the generator operates within a safe range. These constraints together constitute a complete set of constraints, guaranteeing that the control output meets physical and safety limitations.

[0036] Mathematically transforming the set of constraints is a key step in converting the original problem into a standard quadratic programming problem. The system constructs an inequality constraint matrix Au≤b, where A contains the coefficients of all constraints and b contains the corresponding upper and lower limits of the constraints; simultaneously, it constructs an equality constraint matrix Aequ=beq, representing the constraints of the system's dynamic equations. Through this transformation, the original constrained optimization problem is transformed into the standard form min(1 / 2)uTHu+fTu, where H=2(BTQB+R) is the Hessian matrix, and f=2BT(Ax-yr) is the linear term, both calculated from the system matrix, weight matrix, and reference trajectory.

[0037] For standardized quadratic programming problems, the system employs the effective set method for iterative solution. This method first identifies the active constraint set (i.e., boundary constraints), and then solves the unconstrained optimization problem within the subspace defined by the active constraints. During iteration, the sign of the Lagrange multipliers determines whether the active constraint set needs updating, and the optimal solution is determined through the KKT optimality test. For example, when the excitation current approaches the upper limit of 3000A, the corresponding constraint is activated; the algorithm adds this constraint to the active set and resolves under the new constraints. This method typically converges to the optimal solution within 5-10 iterations, with a computation time of less than 0.5ms, meeting the real-time requirement of a 1ms control cycle.

[0038] Rolling optimization of multi-step optimized control sequences is a core characteristic of model predictive control. In each control cycle, the system calculates the optimal control sequence u*(k) to u*(k+Nc-1) for the next Nc = 5 control cycles, but only executes the first control variable u*(k). For example, at t = 10ms, the system calculates the control sequence [150V, 120V, 90V, 60V, 30V], but only executes the first control variable 150V; in the next control cycle t = 11ms, the system re-solves the optimization problem based on new measurements and state estimates, obtaining a new control sequence. This rolling optimization strategy ensures the real-time performance and adaptability to environmental changes while maintaining the ability to predict future behavior.

[0039] Step S3: Monitor the grid frequency change, load power disturbance and system parameter drift in real time, estimate the external disturbance information through the state extended dimension disturbance observer, feed the disturbance information forward into the prediction model to correct the prediction equation, and obtain the prediction result after disturbance compensation.

[0040] Real-time monitoring and classification of power grid frequency changes, load power disturbances, and system parameter drift are performed. A state-extended disturbance observer is designed to construct extended state equations by treating unknown disturbances as extended states, resulting in an extended state model containing disturbance information. The extended state equation is [x(k+1);d(k+1)]=[A,Bd;0,I][x(k);d(k)]+[B;0]u(k), where d(k) represents the disturbance vector.

[0041] For the extended state model, the observer equation is used to calculate the estimated values ​​of various disturbances. Frequency disturbance, power disturbance, and parameter disturbance information are fed forward and corrected into the corresponding terms of the prediction equation to obtain the disturbance-compensated prediction result. The observer equation is [x̂(k+1);d̂(k+1)]=[A,Bd;0,I][x̂(k);d̂(k)]+[B;0]u(k)+L[y(k)-Cx̂(k)]. The observer gain matrix L is designed to converge the observation error. The estimated disturbance information d̂(k) is fed forward into the prediction model, and the prediction equation is corrected to x(k+i|k)=A 1 x(k|k)+∑[j=0 to i-1]Aʲ[Bu(k+i-1-j)+Bdd̂(k)], significantly improves prediction accuracy and control performance through disturbance compensation.

[0042] The method involves calculating estimates of various disturbances using observer equations, and then feeding forward the frequency disturbance, power disturbance, and parameter disturbance information to the corresponding terms of the prediction equation to obtain the disturbance-compensated prediction result, including: Feature extraction and classification are performed on the power grid frequency change Δf, load power disturbance ΔP and system parameter drift. Frequency disturbance model, power disturbance model and parameter disturbance model are established, and corresponding disturbance feature vectors df, dP and dθ are constructed respectively to obtain a classified multi-disturbance model. For the classified multi-perturbation model, a multi-input perturbation observer is designed. By designing the pole configuration of the observer gain matrix L, the perturbation observation error is brought together. The frequency perturbation amplitude, power perturbation change and parameter drift are estimated in real time to obtain an accurate perturbation estimation vector d̂=[d̂f,d̂P,d̂θ]T. Based on the disturbance estimation vector, a classification feedforward compensation strategy is established, which compensates the frequency disturbance d̂f into the angular frequency prediction term ω(k+i|k), the power disturbance d̂P into the electromagnetic torque prediction term Te(k+i|k), and the parameter disturbance d̂θ into the corresponding elements of the system matrix A, thereby obtaining targeted compensation strategy parameters. Based on the compensation strategy parameters, the state prediction equation of the model predictive control, x(k+i|k)=A, is corrected. 1 x(k|k)+∑[j=0 to i-1]Aʲ[Bu(k+i-1-j)+Bdd̂(k)], through feedforward compensation of multiple types of perturbations, significantly improves the accuracy and robustness of the prediction model, and obtains the prediction result after perturbation compensation.

[0043] This invention achieves refined disturbance classification processing. The system extracts features and classifies and identifies grid frequency variation Δf, load power disturbance ΔP, and system parameter drift. Frequency disturbance is calculated by the deviation between the measured grid frequency and the nominal value of 50Hz; power disturbance is identified by the rate of change of active power P and reactive power Q; parameter disturbance is identified by comparing the deviation patterns between model predictions and actual measurements. The system establishes frequency disturbance models, power disturbance models, and parameter disturbance models respectively, constructing corresponding disturbance feature vectors df, dP, and dθ to form a classified multi-disturbance model.

[0044] For the classified multi-perturbation model, a multi-input perturbation observer was designed. The observer gain matrix L was designed using the pole placement method, placing the observer poles at s=-100±j100, enabling rapid convergence of perturbation observation errors. In actual operation, the estimation accuracy of frequency perturbation reached ±0.005Hz, power perturbation reached ±2MW, and parameter perturbation reached ±3%. For example, when the grid frequency drops from 50.00Hz to 49.95Hz, the perturbation observer accurately estimates the frequency perturbation value of -0.05Hz within 15ms and predicts the corresponding power perturbation of approximately 15MW, forming an accurate perturbation estimation vector d̂=[d̂f,d̂P,d̂θ]T.

[0045] Based on the disturbance estimation vector, the system establishes a classification feedforward compensation strategy. Frequency disturbance d̂f is directly compensated into the angular frequency prediction term ω(k+i|k), with the correction formula ω(k+i|k)=ω_base(k+i|k)+d̂f; power disturbance d̂P is compensated into the electromagnetic torque prediction term Te(k+i|k), with the correction formula Te(k+i|k)=Te_base(k+i|k)+d̂P / ω; parameter disturbance d̂θ is compensated into the corresponding elements of the system matrix A, for example, synchronous reactance disturbance is compensated into the elements related to Xd in matrix A. This targeted compensation strategy ensures that disturbance information is accurately applied to the corresponding parts of the prediction model.

[0046] Based on the compensation strategy parameters, the state prediction equation for the modified model predictive control of the system is x(k+i|k)=A 1 x(k|k)+∑[j=0 to i-1]Aʲ[Bu(k+i-1-j)+Bdd̂(k)]. Through feedforward compensation for multiple types of disturbances, the accuracy of the prediction model is significantly improved. For example, under grid frequency fluctuations, the voltage prediction error of traditional predictive control reaches ±0.5kV, while after adding disturbance compensation, the prediction error is reduced to ±0.1kV. Under load abrupt changes, the voltage prediction accuracy increases from 95% to 99%. This high-precision prediction result provides a more reliable decision-making basis for the control system, significantly improving the system's robustness and disturbance resistance.

[0047] Step S4: Based on the parameterized model, the synchronous reactance, transient reactance, subtransient reactance and time constant of the generator are identified online using the recursive least squares method. The model parameters are updated periodically to obtain adaptively updated model parameters.

[0048] Based on the parameterized model y(k)=φᵀ(k)θ(k)+e(k), a parameter vector θ(k) containing synchronous reactance Xd, transient reactance X'd, subtransient reactance X"d and time constants T'd0 and T"d0 is constructed to obtain the complete parameter identification model.

[0049] The parameter identification model is processed using a recursive least squares algorithm. Through recursive updates of the gain matrix K(k) and covariance matrix P(k), the identified parameters are periodically updated into the coefficient matrix A and input matrix B of the state-space model to obtain the adaptively updated model parameters. The recursive least squares algorithm is θ̂(k) = θ̂(k-1) + K(k)[y(k) - φᵀ(k)θ̂(k-1)], the gain matrix K(k) = P(k-1)φ(k)[1 + φᵀ(k)P(k-1)φ(k)]⁻¹, the covariance matrix P(k) = [IK(k)φᵀ(k)]P(k-1) / λ, and the forgetting factor λ = 0.995. The model parameters are updated every 10 control cycles, and the identified parameters are updated into the coefficient matrix A and input matrix B of the state-space model in real time to ensure that the prediction model always reflects the true characteristics of the generator.

[0050] Step S5: Based on the adaptively updated model parameters, a dual-channel control structure is used for coordinated control. The fast channel achieves fast transient response through pole placement, while the precise channel eliminates steady-state error through integral control. The weighting coefficients of the two channels are dynamically adjusted according to the voltage deviation to obtain the final excitation control signal.

[0051] Based on the adaptively updated model parameters, a fast-path controller is designed. The pole placement method is used to place the dominant pole of the closed-loop system at the desired location. A fast transient response is achieved through the state feedback control law uf(k) = -Kfx(k), resulting in the fast-path control output uf(k). The fast-path controller uses the pole placement method to place the dominant pole of the closed-loop system at the desired location s = -50 ± j50. The feedback gain matrix Kf is obtained by solving the characteristic equation det(sI-A+BKf) = (s+50-j50)(s+50+j50)... = 0.

[0052] To address the deviation between the voltage setpoint and the actual voltage, a precision channel controller is designed. An integral control law, ui(k) = ui(k-1) + Ki[yr(k) - y(k)], is used to process the error and eliminate steady-state error, yielding the precision channel control output ui(k). The precision channel uses integral control to eliminate steady-state error; the integral control law is ui(k) = ui(k-1) + Ki[yr(k) - y(k)], and the integral gain Ki = 0.1.

[0053] Weights are assigned to the fast-channel control output uf(k) and the precise-channel control output ui(k). The weighting coefficient α is dynamically adjusted based on the voltage deviation |yr(k) - y(k)|. When the deviation is large, α is set to favor fast response; when the deviation is small, α is set to favor precise adjustment, resulting in the dynamically adjusted weighting coefficient α. The final control output is u(k) = αuf(k) + (1-α)ui(k), where the weighting coefficient α is dynamically adjusted according to the system operating state: when the voltage deviation |yr(k) - y(k)| > 0.02, α = 0.8 favors fast response; when the voltage deviation |yr(k) - y(k)| < 0.005, α = 0.2 favors precise adjustment.

[0054] Based on the dynamically adjusted weighting coefficient α, a dual-channel coordinated control process is performed using the weighted synthesis algorithm u(k) = αuf(k) + (1-α)ui(k) to achieve an organic combination of fast response and high-precision adjustment, resulting in the final excitation control signal. Through dual-channel coordinated control, the system response time is shortened to within 50ms, and the steady-state voltage deviation is controlled within ±0.5%.

[0055] The following section uses the excitation regulation system of a 300MW steam turbine generator set as an example to illustrate the specific implementation process of the technical solution of this invention. The generator has a rated voltage of 20kV, a rated current of 8660A, a synchronous reactance Xd = 1.8pu, a transient reactance X'd = 0.3pu, a subtransient reactance X"d = 0.2pu, a transient time constant T'd0 = 8s, and a subtransient time constant T"d0 = 0.05s.

[0056] In step S1, the system uses a TMS320F28377D processor to establish a seventh-order state-space model of the generator. The state variables are defined as x=[id, iq, if, ikd, ikq, ω, δ]T, representing the stator d-axis current, q-axis current, excitation current, d-axis damping winding current, q-axis damping winding current, angular velocity, and power angle, respectively. Typical values ​​for the system matrix A are: A(1,1)=-Rs / Ld=-0.015 / 0.18=-0.083, A(2,2)=-Rs / Lq=-0.083, A(3,3)=-Rf / Lf=-0.001 / 0.165=-0.006, where Rs=0.015pu is the stator resistance and Rf=0.001pu is the excitation winding resistance. The process noise covariance matrix Q of the Kalman filter is set to 0.01×I7×7, and the observation noise covariance matrix R is set to 0.001×I3×3. Within each 1ms control cycle, the system collects real-time data such as terminal voltage Vt=20.2kV, active power P=280MW, and reactive power Q=50MVar. State estimates are calculated using the Kalman filter, including an excitation current estimate of if_est=1850A and an angular velocity estimate of ω_est=314.2rad / s.

[0057] The model predictive control algorithm in step S2 is implemented as follows: The voltage setpoint Vref = 20kV is set, the current detected terminal voltage Vt = 19.8kV, and the voltage deviation is -0.2kV. The prediction time domain Np = 20, the control time domain Nc = 5, and the sampling period Ts = 0.001s. The weight matrix Q = diag[100, 1, 1] in the objective function emphasizes voltage tracking performance, and the control weight matrix R = 0.1. The excitation current constraint is set to -3000A ≤ If ≤ 3000A, and the excitation voltage constraint is -400V ≤ Uf ≤ 400V. The optimal control sequence is calculated using a quadratic programming solver. The first control quantity is Uf = +150V, corresponding to an excitation current increment ΔIf = +120A. In the next control cycle, when the terminal voltage rises to 19.95kV, the new optimal control quantity is adjusted to Uf = +50V, achieving precise voltage regulation.

[0058] During the disturbance observation process in step S3, the system detected a drop in grid frequency from 50.00Hz to 49.95Hz, with a frequency deviation Δf = -0.05Hz. Based on the frequency change, the disturbance observer estimated a load power increase of ΔP = +15MW, corresponding to a disturbance vector of d = [0, 0, 0, 0, 0, -0.05, 0]T. The disturbance information was fed forward into the prediction model, and the corrected state prediction equation considered the impact of the frequency drop on generator dynamics. Simultaneously, a bus voltage fluctuation of ±0.1kV was detected, which the system identified as a grid-side disturbance. Through disturbance feedforward, the excitation regulation was adjusted to respond in advance, avoiding significant voltage fluctuations. The disturbance observer achieved an observation accuracy of over 95%, significantly improving the system's disturbance immunity.

[0059] In step S4, during parameter identification, the system executes a recursive least squares algorithm update every 10ms. The initial parameter estimate is set to Xd_est = 1.8pu. After 1000 control cycles of online identification, the parameters converge to Xd_est = 1.85pu, with a relative error of 2.8%. The time constant identification result is T'd0_est = 7.8s, which deviates from the actual value of 8s by 2.5%. The forgetting factor λ = 0.995 ensures the convergence and tracking capability of parameter identification. When the generator operating condition changes from the rated load of 280MW to 220MW, the estimated value of the synchronous reactance is adjusted accordingly to 1.82pu, reflecting the adaptive characteristics of parameter identification. The updated model parameters are applied to the predictive control algorithm in real time, ensuring control accuracy.

[0060] In the coordinated execution of dual-channel control in step S5, the fast channel responds immediately when the system detects a voltage step disturbance. Pole configuration sets the dominant closed-loop pole at -50±j50, corresponding to a damping ratio ζ=0.707 and a natural frequency ωn=70.7rad / s. The fast channel's state feedback control adjusts the excitation current from 1850A to 2100A within 5ms, achieving rapid voltage recovery. Subsequently, the precision channel's integral control gradually eliminates steady-state error. The integral gain Ki=0.1 ensures the voltage stabilizes at 19.98kV after 40ms, with a steady-state deviation of only -0.1%. The weighting coefficient α is dynamically adjusted based on the voltage deviation: α=0.8 for the initial 5ms, emphasizing fast response; and α=0.2 for precise adjustment during the steady-state phase. The final total system response time is 45ms, a 70% improvement compared to the 150ms response time of traditional PID control, with steady-state accuracy improved from ±1.5% to ±0.1%.

[0061] This invention also provides a generator excitation regulation system based on digital predictive control, including a data acquisition module, a state estimation module, a predictive control module, a disturbance observation module, a parameter identification module, and a dual-channel control module. The functions of each module correspond one-to-one with the steps of the above method, collectively achieving high-performance control of the generator excitation system.

[0062] This invention establishes a seventh-order state-space mathematical model of the generator and uses an extended Kalman filter for real-time state estimation; it executes a model predictive control algorithm based on a rolling time-domain optimization strategy; it integrates multiple disturbance observers to estimate external disturbances in real time and performs feedforward compensation; it uses recursive least squares method to identify generator parameters online; and it designs a dual-channel control structure to achieve coordinated control of fast response and precise regulation. This reduces the system response time to within 50ms and the steady-state voltage deviation to within ±0.5%, significantly improving the dynamic response performance and control accuracy of the generator excitation regulation system and enhancing the stability of the power system.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0064] It should be noted that those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention. If such modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include such modifications and variations.

[0065] This disclosure also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of a generator excitation regulation method based on digital predictive control as described in the above-described method embodiments. The storage medium can be volatile or non-volatile computer-readable storage.

[0066] Furthermore, this disclosure also provides a computer program product storing a computer program. When the computer program is run by a processor, it executes the steps of a generator excitation regulation method based on digital predictive control provided in any of the above embodiments of this disclosure. For details, please refer to the above method embodiments, which will not be repeated here.

[0067] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium, which can be a volatile or non-volatile computer-readable storage medium. In another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0068] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and apparatuses described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this disclosure, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0069] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0070] In addition, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0071] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0072] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A generator excitation regulation method based on digital predictive control, characterized in that, Includes the following steps: Real-time operating data of generator terminal voltage, stator current and speed are acquired, and a seventh-order state-space mathematical model is established, including stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity and power angle. Real-time state estimation is performed through extended Kalman filter to obtain high-precision system state estimation information. Based on the system state estimation information, a rolling time-domain optimization strategy is adopted to execute the model predictive control algorithm. The prediction time domain is set to 20 control cycles and the control time domain is set to 5 control cycles. An objective function containing the output weight matrix and the control weight matrix is ​​established, and the optimal control sequence is calculated by a quadratic programming solution method. Real-time monitoring of grid frequency changes, load power disturbances and system parameter drift is performed. External disturbance information is estimated through a state-extended dimension disturbance observer. The disturbance information is fed forward into the prediction model to correct the prediction equation and obtain the prediction result after disturbance compensation. Based on the parameterized model, the recursive least squares method is used to identify the synchronous reactance, transient reactance, subtransient reactance and time constant of the generator online, and the model parameters are updated periodically to obtain adaptively updated model parameters; Based on the adaptively updated model parameters, a dual-channel control structure is used for coordinated control. The fast channel achieves fast transient response through pole placement, while the precise channel eliminates steady-state error through integral control. The weighting coefficients of the two channels are dynamically adjusted according to the voltage deviation to obtain the final excitation control signal.

2. The method according to claim 1, characterized in that, The establishment of a seventh-order state-space mathematical model, including stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity, and power angle, and the real-time state estimation using an extended Kalman filter, yields high-precision system state estimation information, including: Based on the electromagnetic and mechanical parameters of the generator, a state vector x=[id,iq,if,ikd,ikq,ω,δ]T is constructed. The system matrix A and input matrix B are determined according to the synchronous reactance, transient reactance, subtransient reactance and rotational inertia, and a standardized state-space model is obtained. The standardized state-space model is subjected to nonlinear state estimation processing. An extended Kalman filter is used, and the high-precision system state estimation information is obtained by recursive calculation through the state prediction equation and the state update equation.

3. The method according to claim 1, characterized in that, The model predictive control algorithm employs a rolling time-domain optimization strategy, setting the prediction time domain to 20 control cycles and the control time domain to 5 control cycles. An objective function is established, including an output weight matrix and a control weight matrix. The optimal control sequence is calculated using a quadratic programming solution method, including: Based on the system state estimation information, an objective function including an output tracking term and a control increment penalty term is established, and the output weight matrix Q and the control weight matrix R are set to obtain the complete optimization objective function. Constraints are set for the objective function, including excitation current constraints, excitation voltage constraints, and rotor angular velocity constraints. The constrained optimization problem is transformed into a quadratic programming problem, which is solved using the effective set method to obtain the optimal control sequence.

4. The method according to claim 1, characterized in that, The process of estimating external disturbance information using a state-extended perturbation observer, feeding this disturbance information forward into the prediction model to correct the prediction equation, and obtaining the disturbance-compensated prediction result includes: Real-time monitoring and classification of power grid frequency changes, load power disturbances and system parameter drifts are performed. A state-extended disturbance observer is designed to construct extended state equations by treating unknown disturbances as extended states, thereby obtaining an extended state model containing disturbance information. For the extended state model, the observer equation is used to calculate the estimated values ​​of various disturbances. The frequency disturbance, power disturbance and parameter disturbance information are fed forward to the corresponding terms of the prediction equation to obtain the prediction result after disturbance compensation.

5. The method according to claim 1, characterized in that, The recursive least squares method is used to identify the synchronous reactance, transient reactance, subtransient reactance, and time constant of the generator online, and the model parameters are updated periodically to obtain adaptively updated model parameters, including: Based on the parameterized model y(k)=φᵀ(k)θ(k)+e(k), a parameter vector θ(k) containing synchronous reactance Xd, transient reactance X'd, subtransient reactance X"d and time constants T'd0 and T"d0 is constructed to obtain the complete parameter identification model. The parameter identification model is processed by a recursive least squares algorithm. Through the recursive update of the gain matrix K(k) and covariance matrix P(k), the identified parameters are periodically updated to the coefficient matrix A and input matrix B of the state space model to obtain the adaptively updated model parameters.

6. The method according to claim 1, characterized in that, The system employs a dual-channel control structure for coordinated control. The fast channel achieves rapid transient response through pole placement, while the precision channel eliminates steady-state error through integral control. The weighting coefficients of the two channels are dynamically adjusted based on the voltage deviation to obtain the final excitation control signal, including: Based on the adaptively updated model parameters, a fast-path controller is designed. The pole placement method is used to place the dominant pole of the closed-loop system at the desired position. The fast transient response is achieved through the state feedback control law uf(k)=-Kfx(k), and the fast-path control output uf(k) is obtained. To address the deviation between the voltage setpoint and the actual voltage, a precise channel controller is designed. An integral control law ui(k) = ui(k-1) + Ki[yr(k) - y(k)] is used to process the deviation, eliminating steady-state error and obtaining the precise channel control output ui(k). Weights are assigned to the fast channel control output uf(k) and the precise channel control output ui(k). The weight coefficient α is dynamically adjusted according to the magnitude of the voltage deviation |yr(k)-y(k)|. When the deviation is large, α is set to favor fast response, and when the deviation is small, α is set to favor precise adjustment, thus obtaining the dynamically adjusted weight coefficient α. Based on the dynamically adjusted weight coefficient α, a dual-channel coordinated control process is performed using the weighted synthesis algorithm u(k)=αuf(k)+(1-α)ui(k) to achieve an organic combination of fast response and high-precision adjustment, thereby obtaining the final excitation control signal.

7. The method according to claim 2, characterized in that, The high-precision system state estimation information is obtained by recursively calculating using an extended Kalman filter and state prediction equations, including: For the standardized state-space model, the process noise covariance matrix Q and the observation noise covariance matrix R are set. The matrix parameters are determined according to the system dynamic characteristics and measurement accuracy to obtain the filter noise characteristic configuration. The extended Kalman filter is linearized by calculating the linearization matrix of the nonlinear state transition function using the Jacobian matrix F(k) and the linearization matrix of the nonlinear observation function using the observation matrix H(k+1), thus obtaining the linearization parameters at the current time step. Based on the filter noise characteristics configuration and linearization parameters, calculate the Kalman gain: K(k+1)=P(k+1|k)HT(k+1)[H(k+1)P(k+1|k)HT(k+1)+R(k+1)]-1 The optimal estimation is achieved by recursively updating the error covariance matrix P(k+1|k), thus obtaining the error covariance information of the state estimate. The error covariance information of the state estimation is recursively updated. State prediction and state update steps are performed in each control cycle. High-precision state estimation is achieved through filtering algorithm to obtain real-time state estimation values ​​including stator current id, iq, excitation current if, angular velocity ω and power angle δ, thus obtaining the high-precision system state estimation information.

8. The method according to claim 3, characterized in that, The addition of excitation current constraints, excitation voltage constraints, and rotor angular velocity constraints transforms the constraint optimization problem into a quadratic programming problem, which is solved using the effective set method to obtain the optimal control sequence, including: For the aforementioned optimization objective function, upper and lower limit constraints for excitation current -Ifmax≤If≤Ifmax, upper and lower limit constraints for excitation voltage -Ufmax≤Uf≤Ufmax, and operating range constraints for rotor angular velocity ωmin≤ω≤ωmax are set to obtain a complete set of constraints. The set of constraints is mathematically transformed to construct the inequality constraint matrix Au≤b and the equality constraint matrix Aequ=beq, thus transforming the original constrained optimization problem into the standard quadratic programming problem min(1 / 2)uTHu+fTu, resulting in the standardized quadratic programming problem. For the standardized quadratic programming problem, the effective set method is used for iterative solution. By dynamically identifying and updating the active constraint set and verifying the optimality condition of KKT, the optimal control sequence in the future control time domain is calculated in each control cycle, resulting in a multi-step optimized control sequence containing u*(k) to u*(k+Nc-1). The multi-step optimized control sequence is subjected to rolling optimization processing. The first control quantity u*(k) is extracted as the actual control output of the current cycle. The optimization problem is solved again in the next control cycle to ensure the real-time performance and rolling optimization characteristics of the control, and the optimal control sequence is obtained.

9. The method according to claim 4, characterized in that, The method involves calculating estimates of various disturbances using observer equations, and then feeding forward the frequency disturbance, power disturbance, and parameter disturbance information to the corresponding terms of the prediction equation to obtain the disturbance-compensated prediction result, including: Feature extraction and classification are performed on the power grid frequency change Δf, load power disturbance ΔP and system parameter drift. Frequency disturbance model, power disturbance model and parameter disturbance model are established, and corresponding disturbance feature vectors df, dP and dθ are constructed respectively to obtain a classified multi-disturbance model. For the classified multi-perturbation model, a multi-input perturbation observer is designed. By designing the pole configuration of the observer gain matrix L, the perturbation observation error is brought together. The frequency perturbation amplitude, power perturbation change and parameter drift are estimated in real time to obtain an accurate perturbation estimation vector d̂=[d̂f,d̂P,d̂θ]T. Based on the disturbance estimation vector, a classification feedforward compensation strategy is established, which compensates the frequency disturbance d̂f into the angular frequency prediction term ω(k+i|k), the power disturbance d̂P into the electromagnetic torque prediction term Te(k+i|k), and the parameter disturbance d̂θ into the corresponding elements of the system matrix A, thereby obtaining targeted compensation strategy parameters. Based on the compensation strategy parameters, the state prediction equation of the model predictive control, x(k+i|k)=A, is corrected. 1 x(k|k)+∑[j=0 to i-1]Aʲ[Bu(k+i-1-j)+Bdd̂(k)], through feedforward compensation of multiple types of perturbations, significantly improves the accuracy and robustness of the prediction model, and obtains the prediction result after perturbation compensation.

10. A generator excitation regulation system based on digital predictive control, characterized in that, include: The data acquisition module is used to acquire real-time operating data of generator terminal voltage, stator current, and speed; The state estimation module is used to establish a seventh-order state-space mathematical model that includes stator d-axis current, stator q-axis current, excitation current, damping winding d-axis current, damping winding q-axis current, rotor angular velocity, and power angle, and performs real-time state estimation through an extended Kalman filter. The predictive control module is used to execute the model predictive control algorithm based on system state estimation information and employ a rolling time-domain optimization strategy to calculate the optimal control sequence using a quadratic programming solution method. The disturbance observation module is used to monitor grid frequency changes, load power disturbances and system parameter drift in real time, and to estimate external disturbance information through the state-extended disturbance observer; The parameter identification module is used to identify generator parameters online based on a parameterized model and using the recursive least squares method. The dual-channel control module is used to coordinate control based on adaptively updated model parameters using a dual-channel control structure to obtain the final excitation control signal.

Images