Networking SVG adaptive adjustment system and method based on artificial intelligence

Abstract

The invention discloses a networking SVG adaptive adjustment system and method based on artificial intelligence. Relates to the technical field of low-voltage reactive compensation and electric energy quality treatment. Wherein the data acquisition module is used for synchronously acquiring multi-dimensional power data of a power grid; the state evaluation module is used for calculating a system state index and identifying a power grid impedance characteristic and a resonant frequency; the artificial intelligence decision module is used for outputting a virtual impedance adjustment decision and a multi-target coordination control instruction; the multi-mode control module is used for selecting a dominant control strategy; the control execution module is used for adjusting a decision and generating an SVG control instruction; the performance evaluation and updating module is used for calculating comprehensive performance indexes and evaluating system impedance change and resonant frequency deviation; and the safety protection module is used for monitoring the state of the SVG equipment and recording system operation data. Therefore, intelligent decision and virtual impedance adaptive adjustment based on artificial intelligence and dynamic coordination adaptation of multiple control targets are realized.

Description

Technical Field

[0001] This invention relates to the field of low-voltage reactive power compensation and power quality management technology, and in particular to an artificial intelligence-based network SVG adaptive adjustment system and method. Background Technology

[0002] As the global energy structure transitions towards clean and low-carbon energy, the scale of new energy power generation, such as photovoltaic and wind power, continues to expand. Due to their characteristics such as output fluctuations and unique interface features, these distributed power sources significantly alter the grid impedance and dynamic response characteristics after being connected to the grid in large numbers, posing severe challenges to power system voltage stability and power quality. Grid-connected SVG (Static Var Compensator), as a core dynamic reactive power compensation device, can rapidly generate or absorb reactive power, playing a crucial role in maintaining grid voltage stability and improving power quality. It is an important device for ensuring the safe operation of new energy grid-connected systems. However, current mainstream grid-connected SVG control technology is still based on traditional classical control theory, which has significant limitations in complex and ever-changing new energy grid-connected scenarios, making it difficult to meet the grid's high requirements for control accuracy and stability.

[0003] The shortcomings of existing SVG control technology are mainly reflected in four aspects: First, it adopts a fixed parameter control strategy based on PI regulation, and the parameters are mostly fixed according to the rated operating point of the power grid. When the system impedance changes due to factors such as distributed power source switching and load fluctuations, problems such as voltage regulation response lag, decreased reactive power compensation accuracy, and reduced stability margin are likely to occur. Second, it lacks the ability to identify the power grid impedance online, which can easily cause voltage oscillations in high-impedance power grids. Control performance deteriorates under weak power grid conditions, and it is difficult to suppress high-frequency resonance. Third, it lacks multi-objective coordination capabilities, adopts a single control mode or simple switching logic, and cannot dynamically optimize the balance of objectives such as voltage stability and reactive power compensation according to the real-time status of the power grid. Fourth, it does not deeply integrate artificial intelligence technology, lacks an intelligent decision-making mechanism based on operating data, and control parameters need to be manually adjusted, making it impossible to achieve online self-tuning and autonomous optimization of control strategies.

[0004] There are currently no effective solutions to the technical problems existing in the technology, such as insufficient adaptive capability, difficulty in multi-objective coordination optimization, lack of intelligent decision-making capability, and limited system resonance suppression effect. Summary of the Invention

[0005] The embodiments of this disclosure provide an artificial intelligence-based network SVG adaptive adjustment system to at least solve the technical problems of insufficient adaptive capability, difficulty in multi-objective coordination optimization, lack of intelligent decision-making capability, and limited system resonance suppression effect in the prior art.

[0006] According to one aspect of the present disclosure, an artificial intelligence-based grid-connected SVG adaptive regulation system is provided, comprising: a data acquisition module for synchronously acquiring multi-dimensional power data of the power grid, the multi-dimensional power data including voltage, current, power, impedance, and harmonic distortion rate; a state assessment module for calculating system state indicators and identifying grid impedance characteristics and resonant frequency based on the multi-dimensional power data, and outputting state assessment results, wherein the system state indicators include power factor, impedance phase margin, and voltage sensitivity coefficient; an artificial intelligence decision-making module for outputting virtual impedance adjustment decisions and multi-objective coordinated control commands based on the state assessment results through reinforcement learning and neural network models; and a multi-mode... The system includes a control module for selecting a dominant control strategy based on the state assessment results, which includes a direct voltage control mode and a constant voltage regulation mode; a control execution module for generating SVG control commands based on the dominant control strategy and the virtual impedance adjustment decision, and for implementing adaptive adjustment of the SVG output reactive power and voltage; a performance evaluation and update module for calculating comprehensive performance indicators based on the output results of the control execution module, evaluating system impedance changes and resonant frequency shifts, and feeding them back to the artificial intelligence decision module to update the decision commands; and a safety protection module for monitoring the SVG device status, performing fault ride-through control when the power grid is abnormal, and recording system operation data.

[0007] According to another aspect of the embodiments of this disclosure, an artificial intelligence-based adaptive regulation method for grid-connected SVG is also provided, comprising the following steps: collecting multi-dimensional power data of the power grid, including voltage, current, power, impedance, and harmonic distortion rate; calculating system state indicators and identifying power grid impedance characteristics and resonant frequency based on the multi-dimensional power data to obtain state assessment results; outputting virtual impedance adjustment decisions and multi-objective coordinated control commands through an artificial intelligence decision model based on the state assessment results; selecting a dominant control strategy based on the state assessment results; generating SVG control commands based on the dominant control strategy and the virtual impedance adjustment decisions to achieve adaptive regulation; calculating comprehensive performance indicators based on the control execution results, evaluating system impedance changes and resonant frequency shifts, and feeding them back to the artificial intelligence decision model to update commands; and monitoring the SVG device status and performing fault ride-through control when the power grid is abnormal.

[0008] This invention forms a collaborative closed loop through parameter transmission, data input, and result output: the data acquisition module synchronously collects multi-dimensional power data and inputs it into the state assessment module; the state assessment module outputs state assessment results based on this data, which are then input into the artificial intelligence decision-making module and the multi-mode control module respectively; the artificial intelligence decision-making module outputs virtual impedance adjustment decisions and multi-objective coordinated control commands, and the multi-mode control module outputs the dominant control strategy, both of which are input into the control execution module; the control execution module generates SVG control commands and performs adaptive adjustment, then inputs the output results into the performance evaluation and update module; the performance evaluation and update module outputs evaluation results and feeds them back to the artificial intelligence decision-making module to optimize decisions, while the safety protection module monitors the SVG equipment and grid operating conditions throughout the process, performs fault ride-through control, and records operating data, ultimately achieving adaptive and stable regulation of the SVG to the grid. This invention realizes real-time acquisition of multi-dimensional grid operating data, accurate assessment of system status, intelligent decision-making and virtual impedance adaptive adjustment based on artificial intelligence, and dynamic coordination and adaptation of multiple control objectives, solving the technical problems of insufficient adaptive capability, difficulty in multi-objective coordinated optimization, lack of intelligent decision-making capability, and limited system resonance suppression effect in existing technologies. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this disclosure and form part of this invention, illustrate exemplary embodiments of the present disclosure and are used to explain the disclosure, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of the AI-based adaptive adjustment system for web SVG according to an embodiment of this disclosure. Detailed Implementation

[0010] To enable those skilled in the art to better understand the technical solutions of this disclosure, 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 some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0011] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0012] According to this embodiment, an artificial intelligence-based adaptive adjustment system for SVG network construction is provided. It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0013] To enable those skilled in the art to better understand the present disclosure, the technical solutions of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present disclosure, and not all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present disclosure.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0016] Figure 1 This is a schematic diagram of a network SVG adaptive adjustment system based on artificial intelligence, as described in this embodiment. (Refer to...) Figure 1 As shown, the system includes: The data acquisition module is used to synchronously acquire multi-dimensional power data of the power grid, including voltage, current, power, impedance, and harmonic distortion rate. The status assessment module is used to calculate system status indicators and identify grid impedance characteristics and resonant frequency based on the multi-dimensional power data, and output the status assessment results, wherein the system status indicators include power factor, impedance phase margin and voltage sensitivity coefficient. An artificial intelligence decision-making module is used to output virtual impedance adjustment decisions and multi-objective coordinated control commands based on the state assessment results through reinforcement learning and neural network models. A multi-mode control module is used to select a dominant control strategy based on the state evaluation results. The dominant control strategy includes a direct voltage control mode and a constant voltage regulation mode. The control execution module is used to generate SVG control commands based on the dominant control strategy and the virtual impedance adjustment decision, and is also used to realize adaptive adjustment of the reactive power and voltage output of the SVG. The performance evaluation and update module is used to calculate comprehensive performance indicators based on the output of the control execution module, evaluate system impedance changes and resonant frequency shifts, and feed them back to the artificial intelligence decision-making module to update decision instructions; and The safety protection module is used to monitor the status of SVG equipment, perform fault ride-through control when the power grid is abnormal, and record system operation data.

[0017] Specifically, such as Figure 1 As shown, the core functions, implementation logic, and inter-module collaboration relationships of each module are as follows: As the basic input unit of the system, the data acquisition module's core function is to synchronously collect multi-dimensional key power data during the power grid's operation. The voltage and current data are the basis for calculating power and impedance. Power data reflects the power grid's energy transmission status, impedance data reflects the impedance characteristics of power grid lines and equipment, and harmonic distortion rate data characterizes the power quality status. The synchronicity of the acquisition process ensures the temporal correlation and effectiveness of various data. Its complete power data set provides comprehensive and reliable raw data support for the subsequent state assessment module, which is a prerequisite for the system to achieve precise regulation.

[0018] The state assessment module takes multi-dimensional power data output from the data acquisition module as input and completes two core processing steps through a specific algorithm: First, it calculates system state indicators such as power factor, impedance phase margin, and voltage sensitivity coefficient. The power factor reflects energy utilization efficiency, the impedance phase margin measures system operational stability, and the voltage sensitivity coefficient characterizes the voltage response to changes in operating conditions. Second, it identifies the grid impedance characteristics and resonant frequency. These grid impedance characteristics determine the system's dynamic interaction patterns, and the resonant frequency is a key risk point for inducing grid oscillations. This module integrates the above calculation results and identification information to output a comprehensive state assessment result reflecting the grid's operating status (normal performance + potential risks), providing clear judgment criteria for the artificial intelligence decision-making module and the multi-mode control module, thus realizing the transformation from data acquisition to decision-making basis.

[0019] As the intelligent core of the system, the artificial intelligence decision-making module takes the evaluation results output by the state assessment module as input and, relying on the collaborative computation of reinforcement learning and neural network models, outputs two types of key decision information: First, virtual impedance adjustment decisions, which clarify the specific adjustment parameters and directions of virtual resistance and virtual inductance, and are the core technical means to change the SVG output characteristics and adapt to changes in grid impedance; second, multi-objective coordinated control commands, which clarify the macro-control priorities of the system (such as prioritizing voltage stabilization, prioritizing resonance suppression, prioritizing reactive power compensation, etc.), and resolve coordination conflicts between multiple control objectives. These two types of outputs together constitute a decision-making system for macro-direction and specific parameters, providing precise guidance for subsequent control execution.

[0020] The multi-mode control module takes the state assessment results output by the state assessment module as input and selects an appropriate dominant control strategy based on the differences in grid operating conditions: when grid voltage fluctuations are severe and stability is insufficient, the direct voltage control mode is selected to achieve rapid voltage response and stable control; when the grid operates smoothly and voltage fluctuations are within a reasonable range, the constant voltage regulation mode is selected to accurately match reactive power demand and optimize power quality. This module ensures precise adaptation of the control mode to grid operating conditions through targeted strategy switching, providing a clear execution framework for the control execution module.

[0021] The control execution module, as the physical output unit of the system, takes the dominant control strategy of the multi-mode control module and the virtual impedance adjustment decision of the artificial intelligence decision module as dual inputs. It generates SVG control commands (such as PWM drive signals and parameter adjustment commands) to translate the decision information into actual physical actions. On the one hand, it achieves precise adjustment of the reactive power output of the SVG to match the reactive power demand of the power grid; on the other hand, it achieves adaptive control of the SVG output voltage to maintain grid voltage stability. This module is the key link connecting intelligent decision-making with actual adjustment effects, directly achieving the core output goals of stabilizing the power grid and optimizing power quality.

[0022] The performance evaluation and update module takes the output of the control execution module as input and performs two tasks: first, it calculates a comprehensive performance index, which quantifies the system's adjustment effect in multiple dimensions such as voltage stability, harmonic suppression, and reactive power compensation accuracy; second, it evaluates changes in system impedance characteristics and resonant frequency shifts, and monitors the dynamic evolution of the power grid state in real time after control execution. Its output is directly fed back to the artificial intelligence decision-making module, which then optimizes subsequent decision commands by updating model parameters, forming an iterative closed loop of "execution-evaluation-optimization." This ensures that the system's decision-making capability continuously adapts to changes in power grid operating conditions and enhances its adaptive adjustment performance.

[0023] As the system's safety assurance unit, the safety protection module monitors the operating status of the SVG device in real time (such as device temperature, operating current, insulation status, etc.) and senses the power grid conditions in real time. When abnormal conditions such as short circuits or voltage surges / dips occur in the power grid, fault ride-through control is immediately executed to ensure that the SVG device does not disconnect from the grid or suffer damage during the fault, while reducing secondary impacts on the power grid. In addition, this module synchronously records system operating data (including control strategy content, device operating parameters, system response when a fault occurs, etc.), providing historical data support for fault tracing and strategy optimization, and comprehensively ensuring the safe and reliable operation of the system.

[0024] In summary, through the logical connection of "data acquisition - status assessment - intelligent decision-making - mode selection - execution adjustment - performance feedback - security protection", a complete adaptive adjustment closed loop is formed. The inputs and outputs of each module are interconnected and work together to ultimately achieve accurate adaptation and stable and efficient adjustment of SVG to changes in power grid operating conditions.

[0025] As described in the background section, the shortcomings of existing SVG control technology are mainly reflected in four aspects: First, it adopts a fixed parameter control strategy based on PI regulation, and the parameters are mostly fixed according to the rated operating point of the power grid. When the system impedance changes due to factors such as distributed power source switching and load fluctuations, problems such as voltage regulation response lag, decreased reactive power compensation accuracy, and reduced stability margin are likely to occur. Second, it lacks the ability to identify the power grid impedance online, which can easily cause voltage oscillations in high-impedance power grids. Control performance deteriorates under weak power grid conditions, and it is difficult to suppress high-frequency resonance. Third, it lacks multi-objective coordination capabilities, adopts a single control mode or simple switching logic, and cannot dynamically optimize the balance of objectives such as voltage stability and reactive power compensation according to the real-time status of the power grid. Fourth, it does not deeply integrate artificial intelligence technology, lacks an intelligent decision-making mechanism based on operating data, and control parameters need to be manually adjusted, making it impossible to achieve online self-tuning and autonomous optimization of control strategies.

[0026] In view of this, the present invention forms a collaborative closed loop through parameter transmission, data input, and result output in sequence: the data acquisition module synchronously collects multi-dimensional power data and inputs it into the state assessment module; the state assessment module outputs state assessment results based on the data, which are then input into the artificial intelligence decision-making module and the multi-mode control module respectively; the artificial intelligence decision-making module outputs virtual impedance adjustment decisions and multi-objective coordinated control commands, and the multi-mode control module outputs the dominant control strategy, both of which are input into the control execution module; the control execution module generates SVG control commands and performs adaptive adjustment, then inputs the output results into the performance evaluation and update module; the performance evaluation and update module outputs evaluation results and feeds them back to the artificial intelligence decision-making module to optimize the decision, while the safety protection module monitors the SVG equipment and grid operating conditions throughout the process, performs fault ride-through control, and records operating data, ultimately achieving adaptive and stable regulation of the SVG to the grid. Thus, the present invention achieves real-time acquisition of multi-dimensional grid operating data, accurate assessment of system status, intelligent decision-making based on artificial intelligence and adaptive virtual impedance adjustment, and dynamic coordination and adaptation of multiple control objectives; thereby solving the technical problems of insufficient adaptive capability, difficulty in multi-objective coordinated optimization, lack of intelligent decision-making capability, and limited system resonance suppression effect in the prior art.

[0027] Optionally, the state assessment module includes a fast-change assessment unit, a medium-speed assessment unit, and a slow-change assessment unit, wherein: the fast-change assessment unit is used to assess the state of the power grid in the case of instantaneous fluctuations; the medium-speed assessment unit is used to assess the state of the power grid in the case of dynamic changes; the slow-change assessment unit is used to assess the state of the topology drift scenario; and calculates system state indicators and identifies power grid impedance characteristics and resonant frequencies based on the multi-dimensional power data, and outputs the state assessment results.

[0028] Specifically, the feature extraction and state assessment module achieves accurate adaptive assessment for different power grid change rates through differentiated settings of fast-changing assessment units, medium-speed assessment units, and slow-changing assessment units. Simultaneously, by linking and fusing high-frequency resonance identification results with system state indicators, it ensures the comprehensiveness and effectiveness of the state assessment results. The fast-change assessment unit (10ms level) is designed to assess instantaneous fluctuations in the power grid, including voltage surges / dips, short-term inrush currents, and transient faults, which are characterized by rapid changes in the power grid state and short durations. This unit uses fast-response data processing logic to capture the power data characteristics corresponding to such instantaneous changes in real time, ensuring timely perception and assessment of sudden power grid conditions. The medium-speed assessment unit (100ms level) focuses on dynamic changes in the power grid, including gradual changes in load power, stable adjustment of distributed power output, and slow fluctuations in power grid frequency. This unit continuously monitors and analyzes the evolution trend of power data during such dynamic changes through tracking data processing, thereby achieving a complete assessment of the power grid transition state. The slow-change assessment unit (1s level) is specifically designed to assess topology drift scenarios, which include slow changes in grid line parameters due to aging and environmental factors, long-term stable shifts in the topology structure after the switching of distributed power sources, and long-term drift of equipment parameters. These are state changes with slow rates of change and lasting impact. Through long-term monitoring and cumulative analysis, this unit captures the potential impact of such slow-change characteristics on the grid operating status, ensuring the effective identification of changes in the basic operating conditions of the grid. Simultaneously, this module links the identification results of high-frequency resonant frequency and amplitude with system status indicators such as power factor, impedance phase margin, and voltage sensitivity coefficient. The identification results of high-frequency resonant frequency and amplitude are used to reflect specific safety risks existing in the power grid, while system status indicators are used to reflect the normal operating performance of the power grid (such as power utilization efficiency, stability, and dynamic response characteristics). Together, they constitute the core content of the status assessment results, which not only covers the normal operating status information of the power grid but also does not omit key safety risk factors. This provides a comprehensive and accurate decision-making basis for subsequent artificial intelligence decision-making modules and multi-mode coordinated control modules, ensuring the pertinence and effectiveness of control strategies.

[0029] Through the above methods, the feature extraction and state assessment module achieves comprehensive and accurate assessment of three different rate-of-change scenarios in the power grid: instantaneous fluctuations, dynamic changes, and topology drift. It can not only capture sudden power grid states in a timely manner, continuously track transitional states, and effectively identify long-term slow-changing states, but also integrate the identification results of high-frequency resonant frequency and amplitude with system state indicators such as power factor, impedance phase margin, and voltage sensitivity coefficient. This allows the state assessment results to simultaneously cover the normal operating performance of the power grid and key safety risk information, thereby outputting comprehensive, accurate, and realistic assessment data. This provides a solid and reliable basis for the subsequent AI decision-making module to generate decision instructions and for the multi-mode coordinated control module to select the dominant control strategy, ensuring the pertinence and effectiveness of the entire system's regulation logic.

[0030] Optionally, the artificial intelligence decision-making module includes a reinforcement learning unit and a neural network unit, wherein: the reinforcement learning unit is configured to use harmonic distortion rate and impedance phase margin as the state space and virtual impedance adjustment amount as the action space to output decision instructions; and the neural network unit is configured to use the multi-dimensional power data as input and output compensation amount for the virtual impedance adjustment decision through a neural network model.

[0031] Specifically, the AI ​​decision-making module achieves intelligent decision output that combines global decision-making with local accuracy through the collaborative operation of reinforcement learning units and neural network units. Its core logic and the functions of each unit are as follows: As the core of global decision-making, the reinforcement learning unit constructs a decision framework adapted to the regulation needs of SVG by clearly defining the state space and action space. The state space is defined as a set of parameters including harmonic distortion rate and impedance phase margin. Harmonic distortion rate is a core indicator characterizing power quality and directly reflects the degree of voltage waveform distortion; impedance phase margin is a key parameter for measuring system stability and determines the system's ability to withstand fluctuations in operating conditions. Together, they constitute the dimension by which the unit perceives the core operating state of the power grid, ensuring that the decision-making basis accurately covers the two core requirements of power quality and stability. The action space is defined as a set of parameters including virtual impedance adjustment. Virtual impedance adjustment is a core control method for changing the output characteristics of SVG and adapting to changes in power grid state. This definition enables the reinforcement learning unit to directly output targeted decision commands regarding the direction and magnitude of impedance adjustment, providing macroscopic directional guidance for subsequent control execution.

[0032] It should be noted that the state space s(t) for designing and training the improved reinforcement learning agent is designed as follows:

[0033] Where PM is the phase margin, |Z|ratio is the impedance ratio, Pflow is the power fluctuation rate, and PF is the power factor. The action space a(t) is defined as:

[0034] The adjustment amounts of the virtual resistor and inductor are discretized into 121 action combinations (11×11 grid). The reward function R(t) is innovatively defined as follows:

[0035] Where β is the boundary penalty term (activated when the parameter approaches the physical limit). The neural network unit, acting as a precision correction unit, takes multi-dimensional power data (voltage, current, power, etc.) output from the data acquisition module as input. Through the nonlinear mapping and fitting capabilities of the neural network model, it outputs a compensation amount to correct the virtual impedance adjustment decision. This compensation amount can effectively correct potential deviations in the initial decision output by the reinforcement learning unit. For example, by combining the real-time operating status of the SVG device (such as load and temperature) or subtle changes in grid conditions (such as local impedance fluctuations), the virtual impedance adjustment amount can be finely optimized to ensure accurate adaptation between the decision command and the actual operating conditions.

[0036] When the two work together, the reinforcement learning unit outputs global decision commands and virtual impedance adjustment directions based on the macroscopic state of the power grid (power quality, stability), while the neural network unit outputs localized correction compensation amounts based on the microscopic features of multi-dimensional power data. This results in a collaborative decision-making outcome of global decision and precise correction, which not only ensures the globality and adaptability of the decision but also improves its accuracy and reliability. This provides high-quality decision support for the control execution module and significantly enhances the system's adaptive adjustment capability to changes in power grid operating conditions.

[0037] Through the above methods, global optimal decision-making based on the core requirements of power grid power quality and stability is achieved by defining the state space and action space. Furthermore, the precise fitting of the network SVG operating state parameters by the sensor structure outputs correction compensation, effectively correcting the initial decision deviation. Finally, virtual impedance adjustment decision and power prediction results at the assessment point are formed, which are global, adaptive, and accurate. This provides high-quality decision support for the multi-mode coordinated control module and significantly improves the system's intelligent decision-making capability and the condition adaptability of decision commands.

[0038] Optionally, the multi-mode control module selects the dominant control strategy according to the voltage stability requirements, including: when the voltage fluctuation exceeds the preset range, a direct voltage control strategy is adopted; when the voltage fluctuates within the stable range, a constant voltage differential regulation strategy is adopted.

[0039] Specifically, the multi-mode control module uses voltage stability requirements as the core decision-making basis. By pre-setting voltage stability judgment criteria, it achieves targeted selection and switching between two dominant control strategies. Its core logic, applicable scenarios, and adjustment objectives are as follows: The preset range is a voltage stability threshold pre-set by the system based on the grid's rated operating standards and the SVG equipment's regulation capabilities. It is used to clearly define the severity of voltage fluctuations and provide a quantitative basis for control strategy switching. When the voltage fluctuation exceeds this preset range due to sudden changes in distributed power output, severe load fluctuations, or external faults, it indicates that the grid is in an emergency condition with a high risk of voltage instability. At this time, the module adopts a direct voltage control strategy. This strategy aims to quickly suppress fluctuations and ensure grid stability in an emergency. By directly adjusting the amplitude and phase of the SVG output voltage, it achieves instantaneous reactive power response, which can offset the impact of abnormal voltage fluctuations in a short time and quickly pull the grid voltage back to the stable range, highlighting the regulation characteristics of rapid response and high priority for stability maintenance.

[0040] When the power grid operates stably and the voltage fluctuation is within the preset stable range, it indicates that the power grid is in normal operating condition. At this time, the module adopts a constant voltage differential regulation strategy. This strategy aims to accurately maintain the voltage and optimize the reactive power compensation accuracy. By setting a fixed voltage-reactive power regulation slope, the reactive power output of the SVG is adaptively adjusted with differential characteristics as the power grid voltage changes. It can stably maintain the power grid voltage near the rated value and accurately match the real-time reactive power demand of the power grid, avoiding power quality fluctuations or increased equipment losses due to over-regulation, highlighting the characteristics of precise regulation and strong adaptability to operating conditions.

[0041] The differentiated selection and switching of the two dominant control strategies enables comprehensive adaptation to emergency and normal operating conditions. This ensures both rapid stabilization under grid instability risks and precise and economical regulation under normal operation. It provides the control execution module with an optimal control framework that adapts to the operating conditions, ensuring that the SVG can achieve efficient adaptive regulation under various voltage fluctuation scenarios.

[0042] Optionally, the control execution module includes a virtual impedance adjustment unit and a PWM drive unit, wherein: the virtual impedance adjustment unit is used to adjust the virtual impedance parameters of the SVG in real time according to the virtual impedance adjustment decision; the PWM drive unit is used to generate a PWM signal according to the dominant control strategy to control the output of the SVG power module.

[0043] Specifically, the control execution module, as the core execution unit connecting intelligent decision-making and physical regulation, transforms decision commands and control strategies into the actual operation of the SVG through the functional division and coordination of the virtual impedance adjustment unit and the PWM drive unit. Its core logic, implementation method, and function are as follows: The virtual impedance adjustment unit takes the virtual impedance adjustment decision output by the artificial intelligence decision module as input. Its core function is to adjust the virtual impedance parameters (including virtual resistance and virtual inductance) of the SVG in real time. The virtual impedance parameters directly determine the output impedance characteristics of the SVG. By receiving the adjustment parameters and direction in the decision command, this unit quickly responds and dynamically modifies the relevant parameters, so that the SVG output impedance can accurately adapt to the changes in the power grid impedance characteristics, thereby achieving the goal of suppressing power grid resonance and optimizing system stability.

[0044] The PWM drive unit takes the dominant control strategy output by the multi-mode control module as input, and its core function is to generate the corresponding PWM (Pulse Width Modulation) signal. Based on the requirements of the dominant control strategy (direct voltage control mode or constant voltage differential regulation mode), this unit generates a PWM drive signal that meets the requirements of control accuracy and response speed through techniques such as carrier phase shifting and pulse width modulation. This signal directly acts on the power module of the SVG (such as IGBTs and other power devices), controlling the turn-on and turn-off timing of the power module. Ultimately, it transforms the control strategy into the actual voltage and current output of the SVG, achieving precise compensation for reactive power in the power grid and stable voltage control.

[0045] When the two units work together, the virtual impedance regulation unit is responsible for optimizing the output characteristics of the SVG to adapt to the power grid, while the PWM drive unit is responsible for converting the control strategy into specific power output actions. The two form an execution system of "characteristic adaptation + precise drive", which ensures that the virtual impedance adjustment decision of the artificial intelligence decision module and the dominant control strategy of the multi-mode control module can be implemented efficiently. Ultimately, the SVG can adaptively regulate the reactive power and voltage of the power grid, achieving the core goal of stabilizing the power grid operation and optimizing power quality.

[0046] Optionally, the performance evaluation and update module is configured to trigger the online learning of the artificial intelligence decision-making module to update the decision instructions when the system operating point changes to a threshold.

[0047] Specifically, the core function of this performance evaluation and update module is to monitor changes in system operating status and trigger targeted instruction optimization processes. The specific implementation of its triggering logic and parameter update method is as follows: The system operating point refers to the current combination of operating conditions of the power grid and the SVG (Static Var Generator), including the comprehensive state of key parameters such as grid voltage, power demand, impedance characteristics, and SVG output status. The preset threshold is a pre-set standard for judging changes in operating conditions, used to define whether the shift in the operating point has affected control performance. This module monitors changes in the system operating point in real time. When the detected shift in the operating point reaches the preset threshold, it indicates that the existing parameters of the BP neural network in the current artificial intelligence decision-making module are no longer able to accurately adapt to the changed operating conditions. At this time, incremental learning of the BP neural network is triggered. This learning method does not require a full retraining of the neural network, but only makes local adjustments to the network parameters based on the latest collected operating condition data. This ensures the timeliness of parameter updates, improves update efficiency, and avoids system regulation interruptions caused by full training.

[0048] During the update execution phase, for neural network units, the gradient backpropagation algorithm is used to update their weight parameters. This algorithm calculates the error between the network output and the actual expected result, propagates it backward along the error gradient, and adjusts the weights layer by layer, so that the correction compensation output of the BP neural network is more in line with the current working conditions, thereby improving the accuracy of decision correction. For reinforcement learning policy networks, an experience replay mechanism is used to update their parameters. This mechanism stores the state, action, and reward data of the system's historical operation as experience samples. During the update, samples are randomly selected to train the policy network, avoiding decision bias caused by the network's over-reliance on current working condition data. At the same time, it makes full use of historical experience to improve the global optimization capability of the policy network.

[0049] Through the above methods, the parameters of the artificial intelligence decision-making module are dynamically optimized, ensuring that its decision-making ability always adapts to changes in system operating conditions and continuously maintains the system's adjustment accuracy and stability.

[0050] Optionally, the safety protection module includes a virtual impedance boundary setting unit and a fault recording unit, wherein: the virtual impedance boundary setting unit is used to limit the virtual impedance adjustment range to prevent overmodulation; and the fault recording unit stores system response data under abnormal power grid conditions in association with fault types.

[0051] Specifically, the safety protection module, through the functional division of the virtual impedance boundary setting unit and the fault recording unit, respectively realizes the active constraint on the operation of the network SVG equipment and the traceability management of fault data, providing dual protection for the safe and stable operation of the system. Its core logic and implementation method are as follows: The core function of the virtual impedance boundary setting unit is to construct a safety constraint barrier for equipment operation by defining the adjustment limits of virtual impedance. Virtual impedance adjustment is a key means for grid-connected SVG to adapt to changes in grid operating conditions, but the adjustment amount is not unlimited. If the virtual impedance adjustment exceeds a reasonable range, it will cause the SVG's power conversion circuit to enter an overmodulation state, leading to output voltage waveform distortion, power quality degradation, and in severe cases, damage to power devices such as IGBTs due to excessive voltage or current stress. Based on this, this unit pre-sets the upper and lower limits of virtual impedance adjustment according to the hardware parameters of the grid-connected SVG (such as the rated capacity of power devices and circuit topology limits) and grid operation requirements, forming a clear adjustment boundary. When the system executes virtual impedance adjustment decisions, this unit verifies in real time whether the adjustment amount is within the set boundary. If it exceeds the boundary, a limiting mechanism is triggered to ensure that the virtual impedance is always adjusted within a safe range, avoiding overmodulation from the source and ensuring the safety of the equipment and the power quality of the grid.

[0052] The fault recording unit focuses on data tracing and management under abnormal power grid conditions, achieving accurate retention of fault information. Abnormal power grid conditions include sudden events that may affect system operation, such as power grid short circuits, voltage spikes / drops, and significant frequency deviations. System response data covers key information such as the control strategy of the network SVG (e.g., the execution logic of fault ride-through control), operating parameters (e.g., current, voltage, temperature), decision command outputs, and equipment status changes at the time of the anomaly. When detecting an abnormal power grid condition, this unit automatically associates and binds the fault type (e.g., short circuit fault, voltage dip fault) with the corresponding system response data, storing it in a designated storage medium according to a preset format. This associated storage method ensures the integrity of fault information—avoiding a disconnect between fault type and response data—and provides a clear basis for subsequent fault analysis, facilitating technicians to trace the cause of the fault, evaluate the system's fault ride-through effect, and provide reliable historical data support for optimizing safety protection strategies and improving the system's fault resistance capabilities.

[0053] Through the above methods, fault analysis and strategy optimization are supported from the data level, which together constitute the core protection logic of the security protection module, and comprehensively improve the operational security and maintainability of the network SVG system.

[0054] Optionally, the system also includes a digital twin module for building a power grid simulation model, predicting resonance points, and optimizing the parameters of the artificial intelligence decision-making module based on the evaluation results of the performance evaluation and update module.

[0055] Specifically, the digital twin module, as the system's virtual simulation and forward-looking optimization unit, enhances the adaptability and foresight of the artificial intelligence decision-making module by constructing a high-fidelity virtual image, conducting risk prediction, and optimizing closed-loop parameters. Its core logic, implementation method, and synergistic effects are as follows: The core foundation of the digital twin module is the construction of a power grid simulation model. This model takes the actual topology of the physical power grid (such as line connections and distributed power source deployment locations), equipment parameters (such as SVG rated capacity, line impedance characteristics, and load type parameters), and historical operating data (such as multi-dimensional power data and fault records) as input. It employs professional simulation techniques such as impedance analysis to build a high-fidelity digital twin that is synchronized with and consistent with the physical power grid in real time. This simulation model can accurately reproduce the operating status, dynamic response characteristics, and operating condition changes of the physical power grid, providing a reliable virtual experimental platform for subsequent resonant point prediction and parameter optimization.

[0056] In terms of resonance point prediction, the digital twin module relies on the dynamic simulation capabilities of the aforementioned simulation model, combined with real-time collected multi-dimensional power data and performance evaluation results, to predict potential resonance risks in the power grid in advance. By simulating the impact of different operating conditions (such as load changes, distributed power source switching, and virtual impedance adjustment) on the impedance characteristics of the power grid, it accurately calculates the frequency points (i.e., resonance points) that may induce resonance and the corresponding risk levels. This breaks through the limitations of traditional passive resonance response and provides forward-looking risk warning information for the artificial intelligence decision-making module, enabling it to avoid resonance-induced areas in advance when making decisions, thereby improving the preventiveness and safety of regulation strategies.

[0057] In the parameter optimization phase, the digital twin module uses the evaluation results (such as comprehensive performance indicators, system impedance changes, and resonant frequency shifts) output by the performance evaluation and update module as a basis to conduct iterative parameter optimization in a virtual scenario. The evaluation results are fed back to the digital twin model to simulate and adjust the core parameters of the artificial intelligence decision-making module (such as the state space weights of the reinforcement learning unit and the network structure parameters of the neural network unit). Simulations are then used to verify the adjustment effects of different parameter combinations (such as voltage stability, harmonic suppression rate, and resonance avoidance capability), selecting the optimal parameter combination that best suits the current operating conditions and possesses forward-looking adaptability. Subsequently, this module feeds back the optimal parameters to the artificial intelligence decision-making module, driving it to update the decision model parameters. This ensures that the decision instructions not only adapt to the current physical power grid state but also to anticipated trends in operating condition changes.

[0058] This module achieves forward-looking prediction of resonance risk through "physical-virtual" two-way interaction, providing a preventive basis for decision-making. On the other hand, it improves the efficiency and safety of parameter optimization of the artificial intelligence decision-making module by replacing physical trial and error with virtual simulation, further ensuring the stability and power quality of the physical power grid operation.

[0059] Optionally, an AI-based adaptive regulation method for SVG (Static Var Generator) in a power grid is characterized by the following steps: collecting multi-dimensional power data from the power grid, including voltage, current, power, impedance, and harmonic distortion rate; calculating system state indicators and identifying grid impedance characteristics and resonant frequencies based on the multi-dimensional power data to obtain state assessment results; outputting virtual impedance adjustment decisions and multi-objective coordinated control commands through an AI decision model based on the state assessment results; selecting a dominant control strategy based on the state assessment results; generating SVG control commands based on the dominant control strategy and the virtual impedance adjustment decisions to achieve adaptive regulation; calculating comprehensive performance indicators based on the control execution results, evaluating system impedance changes and resonant frequency shifts, and feeding them back to the AI ​​decision model to update commands; and monitoring the SVG device status and performing fault ride-through control when the power grid is abnormal.

[0060] Specifically, this AI-based grid-based SVG adaptive regulation method achieves precise adaptation and stable regulation of power grid operating conditions through a seven-step closed-loop process: "data acquisition - state assessment - intelligent decision-making - strategy selection - execution regulation - performance feedback - security protection". The core logic, implementation methods, and connection relationships of each step are as follows: Multi-dimensional power data acquisition: As the fundamental input for the methodology, high-precision sensor arrays deployed at grid connection points and assessment points simultaneously collect five key power data categories: voltage, current, power, impedance, and harmonic distortion rate. Voltage and current data are the basis for calculating power and analyzing power quality; power data reflects the power grid's energy transmission status; impedance data reflects the impedance characteristics of grid lines and equipment (directly affecting system matching); and harmonic distortion rate data characterizes the degree of voltage / current waveform distortion (a core power quality indicator). The acquisition process emphasizes synchronicity and comprehensiveness, ensuring that the data fully covers the core characteristics of grid operation, providing reliable raw data support for subsequent condition assessments.

[0061] The state assessment results are generated by using multi-dimensional power data as input and employing specialized algorithms to perform two core processing steps: First, calculating system state indicators (such as power factor, impedance phase margin, and voltage sensitivity coefficient). The power factor reflects energy utilization efficiency, the impedance phase margin measures system operational stability, and the voltage sensitivity coefficient characterizes the voltage's response to changes in operating conditions. Second, identifying the grid impedance characteristics and resonant frequency. Impedance characteristics determine the system's dynamic interaction patterns, and the resonant frequency is a key risk point for inducing grid oscillations. By integrating the above calculation results and identified information, a comprehensive state assessment result reflecting the grid's "normal operating performance + potential risks" is generated, providing a clear basis for subsequent decision-making and strategy selection.

[0062] Artificial Intelligence Decision Output: The state assessment results are input into the artificial intelligence decision-making model (including reinforcement learning and neural network units). Through collaborative computation, the model outputs two core decision pieces of information: first, virtual impedance adjustment decisions, clarifying the specific adjustment parameters and directions for virtual resistance and virtual inductance (directly determining the SVG output characteristics); second, multi-objective coordinated control commands, defining the macro-control priorities of the system (such as prioritizing voltage stabilization, prioritizing resonance suppression, prioritizing reactive power compensation optimization, etc.), resolving coordination conflicts between multiple control objectives. This step enhances the accuracy and adaptability of decision-making through artificial intelligence technology, providing dual guidance of "specific parameters + macro-direction" for regulation execution.

[0063] Dominant control strategy selection: Based on the state assessment results, an appropriate dominant control strategy is selected according to the grid voltage stability requirements. When grid voltage fluctuations exceed the preset stable range (emergency conditions, such as sudden changes in distributed power output or external faults), a direct voltage control strategy with fast response is selected; when the voltage fluctuates within the stable range (normal conditions), a constant voltage differential regulation strategy with high regulation accuracy is selected. The core of strategy selection is operating condition adaptation, ensuring that the control mode accurately matches the grid operating state, providing the optimal control framework for subsequent execution stages.

[0064] Adaptive regulation execution: As the physical output of the method, it takes the dominant control strategy and virtual impedance adjustment decisions as dual inputs to generate SVG control commands (such as PWM drive signals, virtual impedance parameter adjustment commands, etc.) and translates them into actual regulation actions. On the one hand, it adjusts the virtual impedance parameters to adapt to the grid impedance characteristics, suppressing resonance risks; on the other hand, it controls the output of the SVG power module to achieve precise compensation for grid reactive power and stable voltage control. This step transforms abstract decisions and strategies into physical regulation behaviors, directly achieving the core goal of "stabilizing the grid and optimizing power quality".

[0065] Performance Feedback and Command Update: Taking the control execution results as input, two core tasks are carried out: first, calculating comprehensive performance indicators (quantifying the multi-dimensional adjustment effects such as voltage stability, harmonic suppression rate, and reactive power compensation accuracy); second, evaluating changes in system impedance characteristics and resonant frequency shift (understanding the dynamic evolution of the power grid state after control execution). The evaluation results are fed back to the artificial intelligence decision-making model, driving the model to update decision parameters through online learning (such as optimizing the reinforcement learning strategy network and adjusting neural network weights), outputting updated commands adapted to new operating conditions, forming an iterative closed loop of "execution-evaluation-optimization," ensuring that the system's decision-making capability continuously adapts to changes in power grid operating conditions.

[0066] Equipment monitoring and fault protection: Throughout the entire regulation process, the operating status of the SVG equipment itself is monitored in real time (such as equipment temperature, operating current, insulation status, IGBT module health status, etc.), while the grid conditions are also sensed: When the grid experiences abnormal conditions such as short circuits or voltage surges / dips, the fault ride-through control mechanism is immediately activated to ensure that the SVG equipment does not disconnect from the grid or suffer damage during the fault, while reducing the secondary impact on the grid and providing full-process safety assurance for the stable operation of the system.

[0067] By sequentially connecting and iteratively advancing the above steps, the targeted (based on real-time data and status) and intelligent (AI decision-making) nature of the adjustment process is ensured, as well as the safety and continuity of operation (performance feedback and safety protection), ultimately achieving adaptive adjustment of the grid SVG to changes in grid operating conditions.

[0068] Optionally, the artificial intelligence decision-making model employs reinforcement learning and neural network in synergy. The reinforcement learning uses harmonic distortion rate and impedance phase margin as state inputs and outputs a virtual impedance adjustment amount. The neural network uses the multi-dimensional power data as inputs and outputs a compensation amount to correct the decision.

[0069] Specifically, the AI ​​decision-making model constructs a two-layer intelligent decision-making system of "global decision-making + precise correction" by dividing the functions and synergistically integrating reinforcement learning and neural networks. This system ensures both the globality and adaptability of decisions, while also improving their accuracy and adaptability to different operating conditions. Its core logic, implementation methods, and collaborative relationships are as follows: 1. Enhance the overall decision-making role of reinforcement learning Reinforcement learning serves as the "global decision-making core" of the model, taking the core state parameters of power grid operation as input and outputting macroscopic virtual impedance adjustment commands. The specific logic is as follows: State input definition: Harmonic distortion rate and impedance phase margin are explicitly defined as the core state inputs. Harmonic distortion rate is a key indicator characterizing the power quality of the power grid, directly reflecting the degree of distortion of voltage / current waveforms and determining the power quality optimization requirements. Impedance phase margin is a core parameter for measuring the system's operational stability, directly reflecting the system's ability to resist operating condition fluctuations and avoid oscillations. Together, they constitute the core dimension of reinforcement learning's perception of the power grid's operational status, ensuring that the decision-making basis can accurately cover the two core control objectives of "power quality" and "operational stability".

[0070] Decision output includes a virtual impedance adjustment, which is a core control parameter (such as the specific adjustment direction and magnitude of virtual resistance and virtual inductance) that changes the output impedance characteristics of the grid SVG to adapt to changes in grid operating conditions. Reinforcement learning continuously interacts with the grid's operating status and optimizes the decision logic based on a preset reward function (such as voltage stability and harmonic suppression rate as targets), ensuring that the output virtual impedance adjustment achieves global-level operating condition adaptation and providing a clear macro-level decision-making framework for subsequent adjustments.

[0071] 2. The precise correction function of neural networks The neural network, acting as the model's "precision correction unit," takes multi-dimensional power data as input and outputs targeted compensation amounts, thereby refining and optimizing the initial decisions made during reinforcement learning. Input data range: The input consists of multi-dimensional power data (including basic power parameters such as voltage, current, power, and impedance). This type of data can comprehensively reflect the micro-operational conditions of the power grid (such as local voltage fluctuations and subtle load changes) and the operating status of the SVG device itself (such as device load level and operating temperature), providing rich micro-feature support for the correction logic.

[0072] Compensation Output and Function: Through the nonlinear mapping and fitting capabilities of the neural network model, a compensation amount is output to correct the virtual impedance adjustment decision. The core function of this compensation amount is to correct potential deviations in the initial virtual impedance adjustment amount output by reinforcement learning. For example, the global decision of reinforcement learning may not fully consider micro-factors such as real-time load changes of SVG devices and subtle fluctuations in local impedance of the power grid. The neural network can capture such features based on multi-dimensional power data to generate an accurate compensation amount, making up for the local adaptation deficiencies of the global decision.

[0073] 3. The synergistic integration mechanism between the two Reinforcement learning and neural networks work together to form the final decision: reinforcement learning outputs a global virtual impedance adjustment based on the macroscopic state of the power grid (power quality, stability), building a macroscopic decision framework adapted to the current operating conditions; neural networks output targeted compensation based on the microscopic features of multi-dimensional power data, refining the macroscopic adjustment; finally, through the fusion of the two (such as superimposing the virtual impedance adjustment and compensation with adaptive weights), a final virtual impedance decision command that balances globality and accuracy is obtained, providing high-quality decision support for the control execution module, ensuring that the adjustment actions of the grid SVG not only meet the global control requirements of the power grid, but also accurately adapt to changes in microscopic operating conditions, significantly improving the adaptive adjustment performance of the system.

[0074] Thus, this invention enables real-time acquisition of multi-dimensional power grid operation data, accurate assessment of system status, intelligent decision-making based on artificial intelligence and adaptive adjustment of virtual impedance, as well as dynamic coordination and adaptation of multiple control objectives; thereby solving the technical problems of insufficient adaptive capability, difficulty in multi-objective coordination optimization, lack of intelligent decision-making capability and limited system resonance suppression effect in the prior art.

[0075] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0076] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0077] In the embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The device 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. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0078] 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.

[0079] Furthermore, the functional units in the various embodiments of the present invention 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0080] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part 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 the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0081] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based network configuration SVG adaptive adjustment system, characterized in that, The system comprises: a data acquisition module for synchronously acquiring multi-dimensional power data of the power grid, the multi-dimensional power data including voltage, current, power, impedance, and harmonic distortion rate; a state evaluation module for calculating system state indicators and identifying power grid impedance characteristics and resonance frequency based on the multi-dimensional power data, and outputting state evaluation results, wherein the system state indicators include power factor, impedance phase margin, and voltage sensitivity coefficient; an artificial intelligence decision module for outputting virtual impedance adjustment decisions and multi-objective coordinated control instructions based on the state evaluation results through reinforcement learning and neural network model; a multi-mode control module for selecting a dominant control strategy according to the state evaluation results, the dominant control strategy including direct voltage control mode and constant voltage regulation mode; a control execution module for generating SVG control instructions according to the dominant control strategy and the virtual impedance adjustment decisions, and for realizing adaptive adjustment of SVG output reactive power and voltage; a performance evaluation and update module for calculating comprehensive performance indicators based on the output results of the control execution module, evaluating system impedance changes and resonance frequency shifts, and feeding back to the artificial intelligence decision module to update decision instructions; and a safety protection module for monitoring SVG device state, executing fault ride-through control when the power grid is abnormal, and recording system operation data. The state evaluation module comprises fast-changing evaluation unit, medium-speed evaluation unit, and slow-changing evaluation unit, wherein:

2. The system of claim 1, wherein, the fast-changing evaluation unit is configured to evaluate the state of the power grid in transient fluctuation scenarios; the medium-speed evaluation unit is configured to evaluate the state of the power grid in dynamic change scenarios; the slow-changing evaluation unit is configured to evaluate the state of the power grid in topology drift scenarios; and system state indicators are calculated and power grid impedance characteristics and resonance frequency are identified based on the multi-dimensional power data, and state evaluation results are outputted. The artificial intelligence decision module comprises reinforcement learning unit and neural network unit, wherein:

3. The system of claim 1, wherein, the reinforcement learning unit is configured to take harmonic distortion rate and impedance phase margin as state space, and take virtual impedance adjustment amount as action space, and output decision instructions; and the neural network unit is configured to take the multi-dimensional power data as input, and output compensation amount for the virtual impedance adjustment decisions through neural network model. The multi-mode control module selects the dominant control strategy according to voltage stability requirements, including:

4. The system of claim 1, wherein, when voltage fluctuation exceeds a preset range, direct voltage control strategy is adopted; when voltage fluctuation is within a stable range, constant voltage differential regulation strategy is adopted. The control execution module comprises virtual impedance regulation unit and PWM driving unit, wherein:

5. The system of claim 1, wherein, the virtual impedance regulation unit is configured to adjust virtual impedance parameters of the SVG in real time according to the virtual impedance adjustment decisions; the PWM driving unit is configured to generate PWM signals according to the dominant control strategy, and control the output of the SVG power module. The performance evaluation and update module is configured to trigger online learning of the artificial intelligence decision module to update decision instructions when system operating point changes reach a threshold value.

6. The system of claim 3, wherein, ​ 7. The system of claim 1, wherein, The security protection module includes a virtual impedance boundary setting unit and a fault recording unit, wherein: the virtual impedance boundary setting unit is used to define a virtual impedance adjustment range to prevent over-modulation; and the fault recording unit stores system response data under abnormal grid conditions in association with fault types.

8. The system of claim 1, wherein, Also included is a digital twin module for constructing a grid simulation model, predicting resonance points, and optimizing parameters of the artificial intelligence decision-making module based on evaluation results of the performance evaluation and update module. 9.A method for artificial intelligence-based adaptive adjustment of network configuration SVG, characterized in that, The method includes the following steps: Collecting multi-dimensional power data of the grid, including voltage, current, power, impedance, and harmonic distortion rate; Calculating system state indicators and identifying grid impedance characteristics and resonance frequencies based on the multi-dimensional power data to obtain state evaluation results; Based on the state evaluation results, outputting virtual impedance adjustment decisions and multi-objective coordinated control instructions through an artificial intelligence decision-making model; Selecting a dominant control strategy based on the state evaluation results; Generating SVG control instructions based on the dominant control strategy and the virtual impedance adjustment decisions to achieve adaptive adjustment; Calculating comprehensive performance indicators based on control execution results to evaluate system impedance changes and resonance frequency shifts, and feeding back to the artificial intelligence decision-making model to update instructions; And Monitoring the state of the SVG device and executing fault ride-through control when the grid is abnormal.

10. The method of claim 9, wherein, The artificial intelligence decision-making model uses a combination of reinforcement learning and neural networks, where reinforcement learning takes harmonic distortion rate and impedance phase margin as state input and outputs virtual impedance adjustment amount, and neural networks take the multi-dimensional power data as input and output compensation amount to correct the decision.

Images