A Model Predictive Control-Based Optimization Method and Device for Dynamic Reactive Power Compensation of High-Voltage SVG

By using a model predictive control method, forward-looking compensation of high-voltage static var generators was achieved, which solved the problems of response lag, fixed parameters and insufficient harmonic compensation in the existing technology, and improved power quality and device life.

CN123121701APending Publication Date: 2026-11-17CONSTANCE GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611179898.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-05
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing control methods for high-voltage static var generators suffer from problems such as delayed compensation due to post-event response tracking, fixed control parameters that are difficult to adapt to changes in grid operating conditions, neglect of device switching losses, coarse constraint processing leading to waveform distortion, and limited accuracy of harmonic reactive power compensation.

Method used

A model predictive control-based approach is adopted, which achieves dual adaptation of model and control parameters through multi-step prediction and rolling optimization of discrete state-space model, combined with state estimation and online identification of grid impedance. Multi-objective optimization takes into account both compensation effect and device switching stress, explicit constraint treatment ensures smooth transition near the boundary, and the power quality is improved through the coordinated compensation of fundamental and harmonic reactive power.

Benefits of technology

It improves the compensation response speed, operating condition adaptability, device loss and harmonic suppression performance, and enhances power quality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121701A_ABST
    Figure CN123121701A_ABST
Patent Text Reader

Abstract

This embodiment provides a high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control. A discrete state-space model of the high-voltage static var generator (SGG) and the grid interconnection system is established in a synchronous rotating coordinate system. Extended Kalman filtering is used to estimate the system state, and the equivalent impedance of the grid is identified online to obtain time-varying model parameters. A reactive power demand prediction sequence is generated based on the historical reactive power sequence of the load and the operating condition identification results. A multi-objective optimization problem containing reactive power compensation deviation terms, voltage deviation terms, and control increment penalty terms is constructed in conjunction with the discrete state-space model. Under the condition of satisfying operating constraints, the optimal control sequence is obtained through rolling solution, and a harmonic compensation component is added to generate a comprehensive control command. A pulse width modulation drive signal is generated through coordinate inverse transformation and space vector modulation and sent to the power device. The optimization objective weights and prediction time domain are adjusted according to the operation monitoring results and fed back to the discrete state-space model to form an adaptive closed loop.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of reactive power compensation and power quality control technology in power systems, specifically to a high-voltage SVG dynamic reactive power compensation optimization method and device based on model predictive control. Background Technology

[0002] Existing control methods for high-voltage static var generators (SVRs) mostly employ real-time tracking strategies based on instantaneous reactive power theory, calculating compensation commands based on the reactive power deficit detected at the current moment. This reactive response mode inherently suffers from detection, calculation, and execution delays. Faced with rapid reactive power fluctuations from impact loads such as electric arc furnaces and rolling mills, the compensation commands consistently lag behind actual demands, resulting in limited voltage fluctuation suppression effectiveness. Existing control strategies typically use fixed proportional-integral parameters or preset control gains, failing to consider dynamic changes in grid operating conditions. When grid short-circuit capacity changes, parallel capacitor banks are switched on or off, or system topology alters, controllers with fixed parameters struggle to maintain ideal performance, leading to anything from decreased response speed to severe oscillations or even instability.

[0003] The power devices of high-voltage static var generators (SVRs) are subjected to significant switching and thermal stresses. Existing control methods focus on reactive power compensation while neglecting device losses. In pursuit of rapid response, frequent switching states increase device losses, reducing equipment lifespan and operational economy. When the SVR capacity approaches its rated value or the DC-side voltage approaches its limit, existing control methods lack effective constraint mechanisms. Simple amplitude limiting leads to output waveform distortion, or allowing over-limit operation compromises equipment safety, resulting in significant deterioration of control performance near constraint boundaries.

[0004] Existing methods are insufficient for compensating complex reactive power components containing high-order harmonics. In renewable energy grid-connected scenarios, harmonic reactive power generated by inverters couples with fundamental reactive power. Traditional frequency division compensation strategies suffer from inter-band interference, and insufficient accuracy in harmonic reactive power compensation limits the improvement in power quality. Related feedforward compensation methods reduce response delay by detecting the load current change rate, but they do not address multi-step prediction based on system models, rolling optimization, online identification of time-varying parameters, multi-objective collaboration, and explicit handling of hard constraints. Therefore, a dynamic reactive power compensation method for high-voltage static var generators is needed that combines system modeling, state estimation, demand forecasting, multi-objective rolling optimization, and explicit constraint handling. Summary of the Invention

[0005] To address the problems of existing high-voltage static var generator (SVG) control methods, such as post-event response tracking leading to compensation lag, fixed control parameters making it difficult to adapt to changes in grid operating conditions, neglecting device switching losses, coarse constraint processing resulting in waveform distortion, and limited accuracy of harmonic reactive power compensation, this application provides a high-voltage SVG dynamic reactive power compensation optimization method and device based on model predictive control. This method eliminates the delay in detection and calculation execution through multi-step prediction and rolling optimization of the discrete state-space model, achieves dual adaptation of the model and control parameters through state estimation and online identification of grid impedance, balances compensation effect and device switching stress through multi-objective optimization, ensures smooth transition near the boundary through explicit constraint processing, and improves power quality through coordinated compensation of fundamental and harmonic reactive power, thereby improving compensation response speed, operating condition adaptability, device losses, and harmonic suppression performance.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control, comprising: establishing the dynamic equations of the high-voltage static var generator and the grid interconnection system in a synchronous rotating coordinate system and discretizing them to obtain a discrete state-space model; sampling the three-phase voltage at the point of common coupling and the three-phase current output by the static var generator at high speed and obtaining direct-axis and quadrature-axis components through coordinate transformation; using extended Kalman filtering to estimate the system state based on the direct-axis and quadrature-axis components to obtain a smoothed state trajectory; using the smoothed state trajectory and the injected current, employing a recursive least squares method with a forgetting factor to identify the equivalent resistance and equivalent inductance of the grid online to obtain time-varying model parameters, and backfilling them into the discrete state-space model; wherein, when the time-varying model parameters exceed a preset reasonable range, the previous valid value is retained and an anomaly is identified; based on the load... The reactive power historical sequence and real-time operating condition identification results generate a reactive power demand prediction sequence, which is then combined with the discrete state-space model to construct a multi-objective optimization problem. An optimization algorithm is selected based on the current operating state, and the result of the previous cycle is used as an initial value to solve the multi-objective optimization problem under the condition of satisfying operating constraints, obtaining the optimal control sequence. Harmonic compensation components are added to the optimal control sequence based on the harmonic components of the load current to generate a comprehensive control command. Inverse coordinate transformation and space vector modulation are performed to convert the comprehensive control command into a pulse width modulation drive signal, which is then sent to the power devices of the static var generator after verification by protection logic. Based on the operating monitoring results, the optimization target weights and prediction time domain are adjusted within a preset safety range to obtain adaptive parameters, which are then fed back to the discrete state-space model to form an adaptive closed loop with the multi-objective optimization problem.

[0007] Furthermore, it also includes: establishing AC-side dynamic equations, DC-side capacitor dynamic equations, and the Thevenin equivalent model of the power grid in a synchronous rotating coordinate system, wherein the AC-side dynamic equations describe the changes in direct-axis current and quadrature-axis current, and the DC-side capacitor dynamic equations describe the changes in DC bus voltage; discretizing the AC-side dynamic equations, the DC-side capacitor dynamic equations, and the Thevenin equivalent model of the power grid at a configurable control period to obtain a discrete state-space model with a state vector containing direct-axis current, quadrature-axis current, and DC capacitor voltage, a control vector containing direct-axis modulation voltage and quadrature-axis modulation voltage, and an output vector containing reactive power output and point of common coupling voltage.

[0008] Furthermore, it also includes: extracting the changing trend of the reactive power sequence of the load in the most recent preset quantity control cycle and identifying the current load condition type; generating a reactive demand prediction sequence containing mean prediction and uncertainty interval in the prediction time domain; constructing a multi-objective optimization problem with reactive compensation deviation term, voltage deviation term and control increment penalty term in the objective function based on the reactive demand prediction sequence and the discrete state space model; and expressing the current amplitude, reactive capacity and DC voltage constraints as a set of inequality constraints about the state variables and control variables.

[0009] Furthermore, it also includes: reducing the multi-objective optimization problem to a quadratic programming problem; selecting an optimization algorithm between the effective set method and the interior point method based on the current operating state; using the result of the previous control cycle as the initial value for hot start-up and rolling solution to obtain the optimal control sequence under the condition of satisfying the operating constraints; establishing a harmonic prediction model for the harmonic components of the load current and adding a harmonic compensation objective term to the multi-objective optimization problem; using a multi-resonant controller to achieve frequency band decoupling; ensuring fundamental reactive power compensation first when capacity is limited; and adding harmonic compensation components to the optimal control sequence to generate comprehensive control commands.

[0010] Furthermore, it also includes: performing a Park inverse transform on the integrated control command to obtain a modulation wave in the three-phase stationary coordinate system; selecting a space vector modulation strategy according to the operating state and comparing the modulation wave with a triangular carrier wave to generate a pulse width modulation drive signal; sending the pulse width modulation drive signal to the power device of the static var generator after passing through the bridge arm shoot-through protection and overcurrent protection logic verification; blocking the pulse width modulation output and registering the protection action flag when the verification fails, so that the static var generator enters a safe state.

[0011] Furthermore, it also includes: continuously monitoring the root mean square value of reactive power compensation deviation, voltage fluctuation amplitude, root mean square value of control increment, and constraint activation frequency to generate operation monitoring results; determining the change in the equivalent impedance of the power grid as a change in operating condition and triggering parameter adjustment when the change exceeds a preset threshold condition; adjusting the weights of each item of the objective function and the prediction time domain length within a preset safety range based on the operation monitoring results to obtain adaptive parameters; and feeding the adaptive parameters back to the discrete state space model with parameter identifiers to form an adaptive closed loop with the multi-objective optimization problem.

[0012] Secondly, this application provides a high-voltage SVG dynamic reactive power compensation optimization device based on model predictive control, comprising: a system modeling module, used to establish the AC side dynamic equations, DC side capacitor dynamic equations, and the Thevenin equivalent model of the grid interconnection system of the high-voltage static var generator in a synchronous rotating coordinate system, and discretize them at a configurable control period to obtain a discrete state-space model; a state estimation module, used to sample the three-phase voltage at the point of common coupling and the three-phase output current of the static var generator at high speed and obtain the direct-axis and quadrature-axis components through coordinate transformation, use extended Kalman filtering to estimate the system state based on the direct-axis and quadrature-axis components to obtain a smooth state trajectory, and use a recursive least squares method with a forgetting factor to identify the equivalent resistance and equivalent inductance of the grid online based on the smooth state trajectory and the injected current to obtain time-varying model parameters, and backfill them into the discrete state-space model, wherein the time-varying model parameters exceed a preset reasonable range. The system maintains the previous valid value and registers and identifies abnormal indicators. The prediction optimization module generates a reactive power demand prediction sequence based on the historical reactive power sequence and operating condition identification results. It then constructs a multi-objective optimization problem containing reactive power compensation deviation terms, voltage deviation terms, and control increment penalty terms using the discrete state-space model, and solves it in a rolling manner under operating constraints to obtain the optimal control sequence. The harmonic compensation module establishes a harmonic prediction model for the harmonic components of the load current and adds a harmonic compensation objective term to the multi-objective optimization problem. After frequency band decoupling using a multi-resonant controller, it adds harmonic compensation components to the optimal control sequence to generate a comprehensive control command. The modulation adaptive module performs Park inverse transform and space vector modulation on the comprehensive control command to generate a pulse width modulation drive signal. Based on the operating monitoring results, it adjusts the optimization objective weights and prediction time domain within a preset safety range to obtain adaptive parameters that flow back to the discrete state-space model.

[0013] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method.

[0014] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control.

[0015] Fifthly, this application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the aforementioned model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method.

[0016] As can be seen from the above technical solution, this application provides a high-voltage SVG dynamic reactive power compensation optimization method and device based on model predictive control. It establishes a discrete state-space model in a synchronous rotating coordinate system and performs forward-looking compensation by rolling optimization. It achieves dual adaptation by extended Kalman filter state estimation and impedance identification by recursive least squares method with forgetting factor. It balances the compensation effect and device stress by introducing a control increment penalty term into the objective function. It expresses the current, capacity and DC voltage constraints as a set of inequality constraints for explicit solution. It improves power quality by coordinating the compensation of fundamental and harmonic reactive power and multi-resonance control. It adjusts the optimization target weight and prediction time domain according to the operation monitoring results and feeds the adaptive parameters back to the discrete state-space model to form a closed loop. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0018] Figure 1 This is a flowchart illustrating the high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control in the embodiments of this application. Figure 2 This is a schematic diagram of the high-voltage SVG dynamic reactive power compensation optimization device based on model predictive control in the embodiments of this application. Figure 3 A schematic diagram illustrating the modeling, state estimation, and online impedance identification process of the static var generator and grid interconnection system in this application embodiment; Figure 4 This is a schematic diagram of the load reactive power demand prediction and multi-objective optimization problem construction and rolling solution process in the embodiments of this application; Figure 5 This is a schematic diagram of the coordinate inverse transformation, space vector modulation, and protection verification execution process of the integrated control command in the embodiments of this application. Figure 6 This is a schematic diagram of the closed-loop iterative process of operation monitoring and optimization target weights and prediction time-domain adaptive backflow in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0020] Considering that existing high-voltage static var generator (SVG) control methods often employ post-event response real-time tracking, fixed control parameters, neglect of device switching losses, coarse constraint handling, and limited harmonic reactive power compensation capabilities, this application provides a model predictive control-based optimization method for dynamic reactive power compensation of high-voltage SVG. This method employs interconnected system modeling, state estimation and online impedance identification, demand forecasting, multi-objective rolling optimization, explicit constraint handling, and harmonic collaborative compensation to perform predictive control of dynamic reactive power compensation for high-voltage SVG.

[0021] This application provides an embodiment of a high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control, see [link to relevant documentation]. Figure 1 The method specifically includes the following: Step S101: Establish the dynamic equations of the high-voltage static var generator and the grid interconnection system in the synchronous rotating coordinate system and discretize them to obtain a discrete state-space model. Use extended Kalman filtering to estimate the system state based on real-time measurements and identify the equivalent impedance of the grid online to obtain time-varying model parameters, and then fill them back into the discrete state-space model.

[0022] In this embodiment, the method is applied to the high-voltage static var generator control system. The control system adopts a closed-loop control architecture of state estimation, model prediction, rolling optimization, constraint processing and instruction execution. The hardware adopts an architecture that combines a digital signal processor and a field-programmable gate array.

[0023] In one feasible implementation, such as Figure 3 As shown, the main circuit of the static var generator adopts a three-phase bridge voltage source inverter topology and is connected to the grid bus via a connecting reactor. An AC-side dynamic equation is established in a synchronous rotating coordinate system. This equation describes the dynamic changes of the direct-axis current and quadrature-axis current. The direct-axis current is related to active power, and the quadrature-axis current is related to reactive power.

[0024] The dynamic equations of the DC-side capacitor describe the changes in the DC bus voltage, and the rate of change of the capacitor voltage is proportional to the active power flowing into the DC side. The Thevenin equivalent model is used on the grid side, where the point of common coupling voltage is determined by the equivalent electromotive force, the equivalent impedance voltage drop, and the injected current. Based on this, the continuous-time model is discretized at the configurable control period to obtain a discrete state-space model.

[0025] The discrete state-space model's state vector includes direct-axis current, quadrature-axis current, and DC capacitor voltage; the control vector includes direct-axis modulation voltage and quadrature-axis modulation voltage; and the output vector includes reactive power output and point of common coupling voltage. The grid equivalent impedance in the model parameters is a time-varying parameter and needs to be identified and updated online.

[0026] In parallel processing paths, such as Figure 3 As shown, the three-phase voltage at the point of common coupling, the three-phase current output from the static var generator, and the DC bus voltage are sampled at high speed. The voltage and current components in a synchronous rotating coordinate system are obtained through coordinate transformation. An extended Kalman filter is used to estimate the system state. The filter noise covariance is determined based on offline calibration data, and a smooth state trajectory is output for parameter identification and reading.

[0027] In one specific implementation, the equivalent impedance of the power grid is identified online using a recursive least squares method with a forgetting factor, which is calibrated based on the parameter change rate. The resistance of the connecting reactor is corrected for temperature drift using temperature compensation, while the DC capacitor uses an offline calibration value. If the identified parameters exceed a preset reasonable range, it is judged as an identification anomaly; the previous valid parameter value is retained, and an identification anomaly flag is registered.

[0028] The time-varying model parameters are backfilled after being aligned with the discrete state-space model using the parameter identifier as the primary key, and are then read in chronological order at the prediction entry point in step S102. The anomaly identification identifier is read during parameter reestimation to determine whether to re-enter the identification process, and based on this, the discrete state-space model and time-varying model parameters are delivered to the subsequent prediction optimization stage.

[0029] Step S102: Generate a reactive power demand prediction sequence based on the historical sequence of reactive power and the real-time operating condition identification results, and construct a multi-objective optimization problem in combination with the discrete state-space model. Select an optimization algorithm based on the current operating state and use the result of the previous cycle as the initial value to solve the multi-objective optimization problem in a rolling manner under the condition of satisfying the operating constraints to obtain the optimal control sequence. Add harmonic compensation components to the optimal control sequence based on the harmonic components of the load current to generate a comprehensive control command.

[0030] The discrete state-space model and time-varying model parameters output in step S101 are read into the prediction and optimization stage, such as... Figure 4 As shown, it is used as the initial condition and model parameters for multi-step prediction. When time-varying model parameters are missing, a model not ready flag is registered and the previous valid model is used.

[0031] In one specific implementation, the trend of change of the reactive power sequence of the load in the most recent preset quantity control cycle is extracted. For periodic impact loads such as electric arc furnaces, an adaptive notch filter is used to extract the impact period and phase. For random fluctuation loads, trend extrapolation is used. The current working condition type is identified by real-time feature matching and the corresponding prediction model is called.

[0032] Based on this, a reactive power demand forecast sequence is generated in the prediction time domain. The reactive power demand forecast sequence is ordered by the prediction time and includes the mean prediction and an uncertainty interval. The uncertainty interval is used for subsequent robust optimization. In each control cycle, the measured values ​​are compared with the predicted values ​​to calculate the prediction error and feed it back to correct the prediction model parameters.

[0033] In one feasible implementation, a multi-objective optimization problem is constructed by combining the discrete state-space model. The objective function includes a reactive power compensation deviation term, a voltage deviation term, and a control increment penalty term, all of which are in quadratic form. They respectively measure the deviation between reactive power output and demand, the deviation between the voltage at the point of common coupling and the reference voltage, and the direct-axis and quadrature-axis modulation voltage increments at adjacent times.

[0034] The initial values ​​of each weight are determined based on dimensional normalization and adjusted within a preset range according to the operating conditions. For impact load scenarios, the weight of the reactive power compensation deviation term is increased; for device temperature rise scenarios, the weight of the control increment penalty term is increased. The constraints are expressed as a set of inequalities for current amplitude, reactive power capacity, DC voltage, and modulation voltage. The lower limit of the DC voltage ensures the modulation margin, while the upper limit ensures the device's withstand voltage safety.

[0035] Optionally, the multi-objective optimization problem is reduced to a quadratic programming problem and then solved in a rolling manner. The optimization algorithm is selected between the effective set method and the interior point method according to the current running state. The result of the previous control cycle is used as the initial value for hot start, and the maximum solution time is set as a preset proportion of the control cycle. If the solution fails to converge due to timeout, the current optimal feasible solution is used. If there is no feasible solution, a safe rollback is adopted and the solution timeout flag is registered.

[0036] In another embodiment, the harmonic components of the load current are extracted through spectral analysis and transformed into a corresponding rotating coordinate system to be represented as DC. A harmonic prediction model is established, and a harmonic compensation objective term is added to the multi-objective optimization problem. When capacity is limited, fundamental reactive power compensation is prioritized. The optimal control sequence obtained by adding harmonic compensation components generates a comprehensive control command for reading in step S103.

[0037] Step S103: Perform coordinate inverse transformation and space vector modulation to convert the integrated control command into a pulse width modulation drive signal, and after protection logic verification, send it to the power device of the static var generator. According to the operation monitoring results, adjust the optimization target weight and prediction time domain within the preset safety range to obtain adaptive parameters, and feed the adaptive parameters back to the discrete state space model to form an adaptive closed loop with the multi-objective optimization problem.

[0038] The integrated control instruction output in step S102 is read into the instruction execution stage, such as... Figure 5As shown, the integrated control command includes a direct-axis modulated voltage component and a quadrature-axis modulated voltage component, and is read in time sequence at the execution entry point with the control cycle number as the primary key.

[0039] In one specific implementation, the integrated control command is subjected to an inverse Parker transform to obtain the modulated wave in a three-phase stationary coordinate system. A modulation strategy is selected based on the operating state: space vector modulation is used during normal operation to achieve higher DC voltage utilization; an overmodulation algorithm is used to extend the linear modulation range in the overmodulation region; and a specific modulation strategy is employed during fault ride-through to ensure safety.

[0040] The modulated wave is compared with a triangular carrier wave to generate a pulse width modulation (PWM) signal. The carrier frequency is coordinated with the control period to avoid aliasing, and the switching frequency is calibrated based on a tradeoff between device switching losses and harmonic content. The PWM signal is then sent to the gate of an insulated-gate bipolar transistor after dead-time insertion and drive amplification.

[0041] Before the pulse width modulation signal is output, it undergoes logic verification for bridge arm shoot-through protection and overcurrent protection. If the verification fails, the pulse width modulation output is blocked and a protection action flag is registered. The static var generator enters a safe state. The protection action flag is read at the alarm entry point to trigger the recovery process, thereby completing the transmission of the drive signal for the current control cycle.

[0042] Optionally, such as Figure 6 As shown, continuous monitoring of the root mean square value of reactive power compensation deviation, voltage fluctuation amplitude, root mean square value of control increment, and constraint activation frequency generates operational monitoring results. These results are written to the monitoring cache in chronological order of sampling time. The root mean square value of control increment reflects the switching stress of the device, and the constraint activation frequency reflects the degree to which the operation approaches the boundary.

[0043] When the identified change in the equivalent impedance of the power grid exceeds a preset threshold or the performance index exceeds the allowable range, it is judged as a change in operating condition and parameter adjustment is triggered. The preset threshold is determined based on on-site operation statistics. Accordingly, within a preset safety range, the weights of each term in the objective function and the prediction time domain length are adjusted to obtain adaptive parameters. For weak power grids with larger grid impedance, the prediction time domain is shortened to improve robustness, while for strong power grids with smaller grid impedance, the prediction time domain is extended.

[0044] The adaptive parameters, using the parameter identifier as the primary key, flow back to the discrete state-space model and the multi-objective optimization problem, allowing steps S101 and S102 to read them at the parameter update entry point, forming an adaptive closed loop of detection, optimization, monitoring, and updating. The adaptively adjusted parameters are constrained by a preset safety range to prevent extreme parameters from causing system instability. Adjustments exceeding the limit are truncated at boundary values ​​and a parameter exceeding the limit identifier is registered.

[0045] As can be seen from the above description, the high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control provided in this application can perform forward-looking compensation through multi-step prediction and rolling optimization of discrete state-space model, achieve dual adaptation of model and control parameters through state estimation and online identification of grid impedance, balance the compensation effect and device switching stress through control incremental penalty term, explicitly handle operating constraints through inequality constraint set, and improve power quality through synergistic compensation of fundamental and harmonic reactive power.

[0046] In one embodiment of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method of this application, see [link to relevant documentation]. Figure 3 It can also specifically include the following: Step S201: Establish the AC side dynamic equation, the DC side capacitor dynamic equation, and the Thevenin equivalent model of the power grid in the synchronous rotating coordinate system. The AC side dynamic equation describes the changes in the direct-axis current and the quadrature-axis current, and the DC side capacitor dynamic equation describes the changes in the DC bus voltage. Step S202: Discretize the AC side dynamic equation, the DC side capacitor dynamic equation, and the Thevenin equivalent model of the power grid at the configurable control period to obtain a discrete state-space model with the state vector containing the direct-axis current, quadrature-axis current, and DC capacitor voltage, the control vector containing the direct-axis modulation voltage and quadrature-axis modulation voltage, and the output vector containing the reactive power output and the point of common coupling voltage.

[0047] Step S101, the system modeling step, establishes the dynamic equations of the interconnected system in a synchronously rotating coordinate system, such as... Figure 3 As shown. The main circuit of the static var generator adopts a three-phase bridge voltage source inverter topology and is connected to the grid bus via a connecting reactor. The dynamic equation of the AC side describes the dynamic changes of the direct-axis current and quadrature-axis current, which are jointly determined by the resistance of the connecting reactor, the inductance of the connecting reactor, the grid angular frequency, and the direct-axis and quadrature-axis components of the modulation voltage. The dynamic equation of the DC side capacitor describes the change of the DC bus voltage, and the rate of change of the capacitor voltage is proportional to the active power flowing into the DC side. The Thevenin equivalent model of the grid side characterizes the grid short-circuit capacity with equivalent electromotive force and equivalent impedance. The voltage at the point of common coupling is jointly determined by the equivalent electromotive force, the equivalent impedance voltage drop, and the injected current.

[0048] The AC-side dynamic equations, the DC-side capacitor dynamic equations, and the Thevenin equivalent model of the power grid are discretized at the configurable control period to generate a discrete state-space model, such as... Figure 3As shown. The state vector of the discrete state-space model contains direct-axis current, quadrature-axis current, and DC capacitor voltage; the control vector contains direct-axis modulation voltage and quadrature-axis modulation voltage; and the output vector contains reactive power output and point of common coupling voltage. Among the model parameters, the grid equivalent resistance and grid equivalent inductance are time-varying parameters and need to be identified and updated online. Based on this, the discrete state-space model is written to the model cache using the model version number as the primary key. When the control period value deviates from the modulation accuracy requirement, a period out-of-bounds flag is registered and corrected according to the boundary value, for step S102 to read sequentially at the prediction entry point.

[0049] In one embodiment of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method of this application, see [link to relevant documentation]. Figure 3 It can also specifically include the following: Step S301: The three-phase voltage at the point of common coupling and the three-phase current output from the static var generator are sampled at high speed and the direct-axis and quadrature-axis components are obtained through coordinate transformation. The extended Kalman filter is used to estimate the system state based on the direct-axis and quadrature-axis components to obtain a smooth state trajectory. Step S302: Based on the smooth state trajectory and the injected current, the recursive least squares method with forgetting factor is used to identify the equivalent resistance and equivalent inductance of the power grid online to obtain the time-varying model parameters. When the time-varying model parameters exceed the preset reasonable range, the previous valid value is maintained and an abnormality is identified.

[0050] The state estimation step in step S101 involves high-speed sampling of the measured quantities and performing coordinate transformation, such as... Figure 3 As shown, the three-phase voltage at the point of common coupling, the three-phase current output from the static var generator, and the DC bus voltage are sampled at high speed. After coordinate transformation, the direct-axis and quadrature-axis components in the synchronous rotating coordinate system are obtained. An extended Kalman filter is used to estimate the system state based on the direct-axis and quadrature-axis components. The process noise covariance and measurement noise covariance of the filter are determined based on offline calibrated steady-state and transient data. A smooth state trajectory is output to filter out measurement noise and provide a basis for parameter identification.

[0051] The smoothed state trajectory and injected current are used to identify the equivalent impedance of the power grid online via a recursive least squares method with a forgetting factor, such as... Figure 3 As shown. The recursive least squares method identifies the equivalent resistance and equivalent inductance of the power grid based on the voltage at the point of common coupling, the injected current, and the estimated equivalent electromotive force of the power grid, thus obtaining time-varying model parameters. The forgetting factor is calibrated according to the parameter change rate to track slowly changing parameters and maintain robustness to rapid disturbances. When the time-varying model parameters exceed a preset reasonable range, they are identified as an anomaly, and the previous valid value is retained and an anomaly identifier is registered. The preset reasonable range is determined based on the equipment nameplate parameters and system characteristics as a preset percentage of the nominal value. The time-varying model parameters use the parameter identifier as the primary key for step S201 to read at the model update entry point.

[0052] In one embodiment of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method of this application, see [link to relevant documentation]. Figure 4 It can also specifically include the following: Step S401: Extract the changing trend of the load reactive power sequence of the most recent preset quantity control cycle and identify the current load condition type, and generate a reactive demand forecast sequence containing mean forecast and uncertainty interval in the forecast time domain. Step S402: Based on the reactive power demand prediction sequence and the discrete state-space model, construct a multi-objective optimization problem with an objective function containing reactive power compensation deviation term, voltage deviation term and control increment penalty term, and express the current amplitude, reactive power capacity and DC voltage constraints as a set of inequality constraints about state variables and control variables.

[0053] Step S201, the demand forecasting process, extracts trends from the load reactive power sequence and identifies operating conditions, such as... Figure 4 As shown, the trend of change in reactive power sequence of load in the most recent preset quantity control cycle is extracted. For periodic impact loads such as electric arc furnaces, an adaptive notch filter is used to extract the impact period and phase. For random fluctuation loads, trend extrapolation is used. The current operating condition type of electric arc furnace during melting, refining and tapping periods is identified by real-time feature matching and the corresponding prediction model is called. A reactive power demand prediction sequence containing mean prediction and uncertainty interval is generated in the prediction time domain. A shorter prediction step size and a larger uncertainty margin are used during the melting period.

[0054] The reactive power demand prediction sequence is combined with the discrete state-space model to construct a multi-objective optimization problem, such as... Figure 4 As shown. The objective function includes a reactive power compensation deviation term, a voltage deviation term, and a control increment penalty term, all in quadratic form. The reactive power compensation deviation term measures the deviation between reactive power output and demand in the predicted time domain; the voltage deviation term measures the deviation between the voltage at the point of common coupling and the reference voltage; and the control increment penalty term measures the direct-axis modulation voltage increment and the quadrature-axis modulation voltage increment at adjacent time points. Based on this, the current amplitude, reactive power capacity, and DC voltage constraints are expressed as a set of inequalities regarding state variables and control variables. The multi-objective optimization problem uses the control cycle number as the primary key for step S501 to read at the solution entry point.

[0055] In one embodiment of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method of this application, see [link to relevant documentation]. Figure 4 It can also specifically include the following: Step S501: Reduce the multi-objective optimization problem to a quadratic programming problem, select the optimization algorithm between the effective set method and the interior point method according to the current running state, use the result of the previous control cycle as the initial value for hot start, and obtain the optimal control sequence by rolling solution under the condition of satisfying the running constraints. Step S502: Establish a harmonic prediction model for the harmonic components of the load current and add a harmonic compensation objective term to the multi-objective optimization problem. Use a multi-resonant controller to achieve frequency band decoupling. When the capacity is limited, ensure the fundamental reactive power compensation first. Add harmonic compensation components to the optimal control sequence to generate comprehensive control commands.

[0056] After the multi-objective optimization problem described in step S401 is solved, it is reduced to a quadratic programming problem and solved in a rolling manner, such as... Figure 4 As shown, the optimal algorithm is selected between the effective set method and the interior point method based on the current operating state. The effective set method is used for normal operating conditions, and the interior point method is used for boundary operating conditions. The result of the previous control cycle is used as the initial value for hot start to reduce the number of iterations. The maximum solution time is set as a preset proportion of the control cycle. If the solution fails to converge due to timeout, the current optimal feasible solution is used. If there is no feasible solution, a safe rollback based on the result of the previous cycle is used and the solution timeout flag is registered. Based on this, the optimal control sequence is obtained under the condition of satisfying the operating constraints.

[0057] A harmonic prediction model is established for the harmonic components of the load current, and a harmonic compensation objective term is added to the multi-objective optimization problem, such as... Figure 4 As shown, harmonic components are extracted from the load current through spectral analysis and transformed into a DC quantity in a corresponding rotating coordinate system. The harmonic prediction model uses a relatively long prediction step size. The harmonic compensation target term measures the deviation between the output of each harmonic and the harmonic demand, and the priority of the fundamental harmonic over lower harmonics and vice versa is reflected by weights. A multi-resonant controller is used to achieve frequency band decoupling. When capacity is limited, fundamental reactive power compensation is prioritized. Based on this, a comprehensive control command is generated by adding harmonic compensation components to the optimal control sequence, and the control cycle number is used as the primary key for step S601 to read at the execution entry point.

[0058] In one embodiment of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method of this application, see [link to relevant documentation]. Figure 5 It can also specifically include the following: Step S601: Perform Park inverse transformation on the integrated control command to obtain the modulation wave in the three-phase stationary coordinate system, select a space vector modulation strategy according to the operating state, and compare the modulation wave with the triangular carrier to generate a pulse width modulation drive signal; Step S602: After the pulse width modulation drive signal is verified by the bridge arm shoot-through protection and overcurrent protection logic, it is sent to the power device of the static var generator. If the verification fails, the pulse width modulation output is blocked and the protection action flag is registered so that the static var generator enters a safe state.

[0059] The integrated control command described in step S501 is used to generate a pulse width modulation drive signal through inverse Parker transform and space vector modulation, such as... Figure 5As shown, the integrated control command is subjected to Park inverse transform to obtain the modulation wave in the three-phase stationary coordinate system. The modulation strategy is selected according to the operating state. During normal operation, space vector modulation is used to obtain higher DC voltage utilization and lower harmonic content. In the overmodulation region, an overmodulation algorithm is used to extend the linear modulation range. During fault ride-through, a specific modulation strategy is used. The modulation wave is compared with the triangular carrier to generate a pulse width modulation drive signal. The carrier frequency is coordinated with the control period to avoid aliasing.

[0060] The pulse width modulation drive signal is sent to the power device after being verified by protection logic, such as Figure 5 As shown, the pulse width modulation (PWM) drive signal, after dead-time insertion and drive amplification, is sent to the gate of the insulated-gate bipolar transistor (IGBT). It first undergoes bridge arm shoot-through protection and overcurrent protection logic checks to ensure compliance with device safety constraints. If the checks fail, the PWM output is blocked and a protection action flag is registered, causing the static var generator (SVR) to enter a safe state. Based on this, the drive signal for the current control cycle is issued. The protection action flag is read at the alarm entry point to trigger the recovery process and is then read at the monitoring entry point in step S701.

[0061] In one embodiment of the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method of this application, see [link to relevant documentation]. Figure 6 It can also specifically include the following: Step S701: Continuously monitor the root mean square value of reactive power compensation deviation, voltage fluctuation amplitude, root mean square value of control increment and constraint activation frequency to generate operation monitoring results. When the change of the grid equivalent impedance exceeds the preset threshold condition, it is judged as a change in operating condition and parameter adjustment is triggered. Step S702: Based on the operation monitoring results, adjust the weights of each item of the objective function and the prediction time domain length within a preset safety range to obtain adaptive parameters. Then, return the adaptive parameters to the discrete state space model with parameter identifiers to form an adaptive closed loop with the multi-objective optimization problem.

[0062] After step S601 is completed, the control effect is continuously monitored and operational monitoring results are generated, such as... Figure 6 As shown. Continuous monitoring of the root mean square value of reactive power compensation deviation, voltage fluctuation amplitude, root mean square value of control increment, and constraint activation frequency generates operational monitoring results. The root mean square value of control increment reflects the switching stress of the devices, and the constraint activation frequency reflects the degree to which the operation approaches the boundary. Based on this, when the change in the equivalent impedance of the power grid exceeds a preset threshold condition or the performance index exceeds the allowable range, it is determined as a change in operating condition and parameter adjustment is triggered. The preset threshold condition is determined based on on-site operational statistics. The operational monitoring results are written into the monitoring cache in chronological order of sampling time.

[0063] The operational monitoring results are used to adjust the optimization target weights and prediction time domain within a preset safety range, such as... Figure 6As shown, when impact load characteristics are detected, the weight of the reactive power compensation deviation term is increased; when voltage over-limit risk is detected, the weight of the voltage deviation term is increased; when device temperature rise is detected, the weight of the control increment penalty term is increased. For weak grids with high grid impedance, the prediction time domain is shortened; for strong grids with low grid impedance, the prediction time domain is extended to obtain adaptive parameters. Based on this, the adaptive parameters are fed back to the discrete state-space model in step S101 and the multi-objective optimization problem in step S402 using parameter identifiers as the primary key. Out-of-bounds adjustment amounts are truncated according to boundary values ​​and parameter over-bounds identifiers are registered, forming an adaptive closed loop of detection, optimization, monitoring, and updating.

[0064] To implement all or part of the aforementioned method, this application provides an embodiment of an apparatus for implementing the model predictive control-based high-voltage SVG dynamic reactive power compensation optimization method. See [link to embodiment]. Figure 2 The device specifically includes the following components: The system modeling module is used to establish the AC side dynamic equations, DC side capacitor dynamic equations, and the Thevenin equivalent model of the grid interconnection system of the high voltage static var generator in a synchronous rotating coordinate system, and to discretize them at the configurable control period to obtain a discrete state space model. The state estimation module is used to estimate the system state based on real-time measurements using extended Kalman filtering, and to identify the equivalent impedance of the power grid online using recursive least squares with a forgetting factor to obtain time-varying model parameters and backfill them into the discrete state space model. The prediction and optimization module is used to generate a reactive power demand prediction sequence based on the historical sequence of reactive power and the operating condition identification results. It combines the discrete state space model to construct a multi-objective optimization problem containing reactive power compensation deviation term, voltage deviation term and control increment penalty term, and obtains the optimal control sequence by rolling solution under the condition of satisfying the operating constraints. The harmonic compensation module is used to establish a harmonic prediction model for the harmonic components of the load current and add a harmonic compensation objective term to the multi-objective optimization problem. After using a multi-resonant controller to achieve frequency band decoupling, it adds harmonic compensation components to the optimal control sequence to generate comprehensive control commands. The modulation adaptive module is used to perform Park inverse transform and space vector modulation on the integrated control command to generate pulse width modulation drive signal, and adjust the optimization target weight and prediction time domain within a preset safety range according to the operation monitoring results to obtain adaptive parameters that are fed back to the discrete state space model.

[0065] The system modeling module establishes the dynamic equations of the AC side of the static var generator, the dynamic equations of the DC side capacitor, and the Thevenin equivalent model of the power grid in a synchronous rotating coordinate system, and discretizes them at the configurable control cycle to obtain a discrete state-space model. The state vector of the discrete state-space model includes the direct-axis current, the quadrature-axis current, and the DC capacitor voltage. The state estimation module uses an extended Kalman filter to estimate the system state based on the voltage and current components after high-speed sampling and coordinate transformation to obtain a smooth state trajectory. Then, it uses a recursive least squares method with a forgetting factor to identify the equivalent resistance and equivalent inductance of the power grid online based on the smooth state trajectory and the injected current to obtain time-varying model parameters and backfill them into the discrete state-space model.

[0066] The prediction and optimization module extracts the changing trend of the historical reactive power sequence of the load and identifies the current operating condition type to generate a reactive demand prediction sequence containing mean prediction and uncertainty interval. Combined with the discrete state-space model, it constructs a multi-objective optimization problem containing reactive power compensation deviation terms, voltage deviation terms, and control increment penalty terms. The current amplitude, reactive power capacity, and DC voltage constraints are expressed as a set of inequalities and reduced to a quadratic programming problem. Under the condition of satisfying the operating constraints, the optimal control sequence is obtained through rolling solution. The harmonic compensation module establishes a harmonic prediction model for the harmonic components of the load current and adds a harmonic compensation objective term to the multi-objective optimization problem. After using a multi-resonant controller to achieve frequency band decoupling, it adds harmonic compensation components to the optimal control sequence to generate comprehensive control commands.

[0067] The modulation adaptive module performs a Park inverse transform on the comprehensive control command to obtain the modulation wave in the three-phase stationary coordinate system. It then selects a space vector modulation strategy and compares it with a triangular carrier wave to generate a pulse width modulation (PWM) drive signal. After verification by bridge arm shoot-through protection and overcurrent protection logic, the signal is sent to the power device. If the verification fails, the PWM output is blocked, putting the static var generator into a safe state. The modulation adaptive module continuously monitors reactive power compensation deviation, voltage fluctuation, control increment, and constraint activation frequency. When the change in the grid equivalent impedance exceeds a preset threshold, it adjusts the optimization target weights and prediction time domain within a preset safety range to obtain adaptive parameters. These adaptive parameters are then fed back to the discrete state space model to form an adaptive closed loop with the multi-objective optimization problem.

[0068] As can be seen from the above description, the high-voltage SVG dynamic reactive power compensation optimization device based on model predictive control provided in this application embodiment can perform predictive control and adaptive adjustment of the dynamic reactive power compensation of the high-voltage static var generator through the coordinated operation of five modules: system modeling, state estimation, predictive optimization, harmonic compensation and modulation adaptation.

[0069] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method.

[0070] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0071] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method.

[0072] In this embodiment, a discrete state-space model is established in a synchronous rotating coordinate system, and forward reactive power compensation is performed by rolling optimization in the prediction time domain. Dual adaptation of the model and control parameters is achieved by extended Kalman filter state estimation and online impedance identification using recursive least squares method with forgetting factor. The compensation effect and device switching stress are balanced by introducing a control increment penalty term into the objective function. Current, capacity, and DC voltage constraints are expressed as a set of inequalities and explicitly incorporated into the quadratic programming solution. Power quality is improved by synergistic compensation of load harmonic components and frequency band decoupling of the multi-resonant controller. The optimization target weights and prediction time domain are adjusted according to the operation monitoring results, and the adaptive parameters are fed back to the discrete state-space model to form a closed loop.

[0073] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. 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.

Claims

1. A method for optimizing dynamic reactive power compensation of high-voltage SVG based on model predictive control, characterized in that, The method includes: The dynamic equations of the high-voltage static var generator (SVM) and grid interconnection system in a synchronous rotating coordinate system are established and discretized to obtain a discrete state-space model. The three-phase voltage at the point of common coupling and the three-phase current output by the SVM are sampled at high speed and the direct-axis and quadrature-axis components are obtained through coordinate transformation. An extended Kalman filter is used to estimate the system state based on the direct-axis and quadrature-axis components to obtain a smooth state trajectory. Based on the smooth state trajectory and the injected current, a recursive least squares method with a forgetting factor is used to identify the equivalent resistance and equivalent inductance of the grid online to obtain time-varying model parameters, which are then backfilled into the discrete state-space model. When the time-varying model parameters exceed a preset reasonable range, the previous valid value is retained and an anomaly is identified. Based on the historical sequence of reactive power and the real-time operating condition identification results, a reactive demand prediction sequence is generated, and a multi-objective optimization problem is constructed by combining the discrete state space model. Based on the current operating state, an optimization algorithm is selected and the result of the previous cycle is used as the initial value. Under the condition of satisfying the operating constraints, the multi-objective optimization problem is solved in a rolling manner to obtain the optimal control sequence. Based on the harmonic components of the load current, a harmonic compensation component is added to the optimal control sequence to generate a comprehensive control command. The integrated control command is converted into a pulse width modulation drive signal by performing coordinate inverse transformation and space vector modulation. After being verified by protection logic, the signal is sent to the power device of the static var generator. Based on the operation monitoring results, the optimization target weight and prediction time domain are adjusted within a preset safety range to obtain adaptive parameters. The adaptive parameters are then fed back to the discrete state space model to form an adaptive closed loop with the multi-objective optimization problem.

2. The high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control according to claim 1, characterized in that, The process of establishing the dynamic equations of the high-voltage static var generator and the grid interconnection system in a synchronous rotating coordinate system and discretizing them to obtain a discrete state-space model includes: In a synchronous rotating coordinate system, dynamic equations for the AC side, dynamic equations for the DC side capacitor, and the Thevenin equivalent model of the power grid are established respectively. The dynamic equations for the AC side describe the changes in the direct-axis current and the quadrature-axis current. The dynamic equations for the DC side capacitor describe the changes in the DC bus voltage. The AC-side dynamic equation, the DC-side capacitor dynamic equation, and the Thevenin equivalent model of the power grid are discretized at the configurable control period to obtain a discrete state-space model with a state vector containing direct-axis current, quadrature-axis current, and DC capacitor voltage, a control vector containing direct-axis modulation voltage and quadrature-axis modulation voltage, and an output vector containing reactive power output and point of common coupling voltage.

3. The high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control according to claim 1, characterized in that, The process of generating a reactive power demand prediction sequence based on the historical reactive power sequence and real-time operating condition identification results, and then constructing a multi-objective optimization problem using the discrete state-space model, includes: Extract the trend of change of the reactive power sequence of the load in the most recent preset quantity control cycle and identify the current load condition type. Generate a reactive demand forecast sequence containing mean forecast and uncertainty interval in the forecast time domain. Based on the reactive power demand prediction sequence and the discrete state-space model, a multi-objective optimization problem is constructed with the objective function containing reactive power compensation deviation term, voltage deviation term and control increment penalty term, and the current amplitude, reactive power capacity and DC voltage constraints are expressed as a set of inequality constraints with respect to state variables and control variables.

4. The high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control according to claim 1, characterized in that, The optimal control sequence is obtained by selecting an optimization algorithm based on the current operating state and using the result of the previous cycle as the initial value to solve the multi-objective optimization problem under the condition of satisfying the operating constraints. A comprehensive control command is generated by adding harmonic compensation components to the optimal control sequence based on the harmonic components of the load current, including: The multi-objective optimization problem is reduced to a quadratic programming problem. Based on the current operating state, the optimization algorithm is selected between the effective set method and the interior point method. The result of the previous control cycle is used as the initial value for hot start-up, and the optimal control sequence is obtained by rolling solution under the condition of satisfying the operating constraints. A harmonic prediction model is established for the harmonic components of the load current, and a harmonic compensation objective term is added to the multi-objective optimization problem. A multi-resonant controller is used to achieve frequency band decoupling. When the capacity is limited, the fundamental reactive power compensation is guaranteed first. The harmonic compensation component is added to the optimal control sequence to generate a comprehensive control command.

5. The high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control according to claim 1, characterized in that, The execution of coordinate inverse transformation and space vector modulation converts the comprehensive control command into a pulse width modulation drive signal, which is then sent to the power device of the static var generator after being verified by protection logic. This includes: The Park inverse transform is performed on the integrated control command to obtain the modulation wave in the three-phase stationary coordinate system. A space vector modulation strategy is selected according to the operating state, and the modulation wave is compared with the triangular carrier wave to generate a pulse width modulation drive signal. The pulse width modulation drive signal is sent to the power device of the static var generator after being verified by the bridge arm shoot-through protection and overcurrent protection logic. If the verification fails, the pulse width modulation output is blocked and the protection action flag is registered so that the static var generator enters a safe state.

6. The high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control according to claim 5, characterized in that, The step of adjusting the optimization target weights and prediction time domain within a preset safety range based on operational monitoring results to obtain adaptive parameters, and then feeding these adaptive parameters back to the discrete state space model to form an adaptive closed loop with the multi-objective optimization problem, includes: The system continuously monitors the root mean square value of reactive power compensation deviation, voltage fluctuation amplitude, root mean square value of control increment, and constraint activation frequency to generate operation monitoring results. When the change in the equivalent impedance of the power grid exceeds the preset threshold, it is judged as a change in operating condition and triggers parameter adjustment. Based on the operational monitoring results, the weights of each term of the objective function and the prediction time domain length are adjusted within a preset safety range to obtain adaptive parameters. The adaptive parameters are then fed back to the discrete state space model with parameter identifiers to form an adaptive closed loop with the multi-objective optimization problem.

7. A high-voltage SVG dynamic reactive power compensation optimization device based on model predictive control, characterized in that, The device includes: The system modeling module is used to establish the AC side dynamic equations, DC side capacitor dynamic equations, and the Thevenin equivalent model of the grid interconnection system of the high voltage static var generator in a synchronous rotating coordinate system, and to discretize them at the configurable control period to obtain a discrete state space model. The state estimation module is used to sample the three-phase voltage at the point of common coupling and the three-phase current output from the static var generator at high speed, and obtain the direct-axis and quadrature-axis components through coordinate transformation. An extended Kalman filter is used to estimate the system state based on the direct-axis and quadrature-axis components to obtain a smooth state trajectory. Based on the smooth state trajectory and the injected current, a recursive least squares method with a forgetting factor is used to identify the equivalent resistance and equivalent inductance of the power grid online to obtain time-varying model parameters, which are then backfilled into the discrete state space model. When the time-varying model parameters exceed a preset reasonable range, the previous valid value is retained and an anomaly is identified. The prediction and optimization module is used to generate a reactive power demand prediction sequence based on the historical sequence of reactive power and the operating condition identification results. It combines the discrete state space model to construct a multi-objective optimization problem containing reactive power compensation deviation term, voltage deviation term and control increment penalty term, and obtains the optimal control sequence by rolling solution under the condition of satisfying the operating constraints. The harmonic compensation module is used to establish a harmonic prediction model for the harmonic components of the load current and add a harmonic compensation objective term to the multi-objective optimization problem. After using a multi-resonant controller to achieve frequency band decoupling, it adds harmonic compensation components to the optimal control sequence to generate comprehensive control commands. The modulation adaptive module is used to perform Park inverse transform and space vector modulation on the integrated control command to generate pulse width modulation drive signal, and adjust the optimization target weight and prediction time domain within a preset safety range according to the operation monitoring results to obtain adaptive parameters that are fed back to the discrete state space model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the high-voltage SVG dynamic reactive power compensation optimization method based on model predictive control as described in any one of claims 1 to 6.