Method and system for state monitoring and load control of a motor-generator group for reactive power compensation of a power grid

Abstract

The application discloses a method and system for state monitoring and load control of a phase modifier group for reactive power compensation of a power grid, and belongs to the technical field of automatic control of power systems, and mainly solves the problem that the existing phase modifier group does not fully consider the health state of equipment when responding to the reactive power demand of the power grid, thereby causing the life consumption to be intensified or the control target to be single. In the application, signals of grid voltage, frequency, unit vibration, temperature and excitation current are collected in real time, reactive power deficiency is calculated, vibration frequency band energy features are extracted, and a health state index HSI is constructed in combination with adaptive wavelet base selection. With the minimum equipment life consumption and the optimal grid voltage stability as multiple targets, a particle swarm algorithm is adopted to dynamically generate a load control instruction, and weight adaptive adjustment is realized based on the HSI, and an online life consumption model is introduced as a constraint condition. Finally, the reactive power output is adjusted through the excitation system, and the collaborative control of grid stability and equipment life extension is realized.

Description

Technical Field

[0001] This invention relates to the field of power system automatic control technology. More specifically, this invention relates to a method and system for monitoring the status of synchronous condenser units and controlling loads for reactive power compensation in power grids. Background Technology

[0002] Maintaining grid voltage stability is a crucial task in power system operation. As key equipment that provides reactive power support and performs voltage regulation, the rationality of the control strategy of synchronous condensers directly affects the safe and stable operation of the power grid and the service life of the equipment itself.

[0003] Traditional synchronous condenser load control methods primarily rely on real-time grid operating parameters, such as node voltage deviations or system frequency deviations, to determine reactive power output commands. While these methods prioritize meeting the grid's instantaneous demand and can achieve rapid voltage regulation in most cases, their control logic does not adequately consider the operating state of the synchronous condenser itself. As large rotating electrical machines, synchronous condensers are subjected to a combination of mechanical, thermal, and electrical stresses on their internal components, such as bearings, rotors, and stator windings, leading to gradual aging and even failure. If a unit is already in a potentially sub-optimal or excessively worn state and is subjected to high loads based on grid demand commands, it may accelerate insulation aging and mechanical wear, significantly shorten its service life, and even induce sudden failures, posing a threat to grid security.

[0004] On the other hand, some improved control methods attempt to incorporate equipment condition factors, but they typically only use a limited number of single condition parameters (such as winding temperature) as threshold constraints for over-limit alarms. This is a relatively passive protection approach rather than an active optimization strategy. This method struggles to comprehensively and quantitatively assess the overall health status of the equipment and cannot achieve a refined and forward-looking trade-off between grid demand and equipment condition. The fundamental difficulty lies in the fact that equipment condition degradation is a gradual process involving multiple physical fields, and a single parameter cannot fully characterize its health level. Constructing a quantitative indicator that can reflect the overall health status of the equipment in real time and effectively link it with grid control objectives is a significant technical challenge.

[0005] Furthermore, even when considering both equipment status and grid stability, traditional optimization control methods often employ fixed-weight strategies to transform multi-objective optimization problems into single-objective solutions. However, both grid operating conditions and equipment health are dynamic. When equipment is in good condition, the control strategy can be more inclined to meet grid demands; conversely, when equipment is in poor condition, the strategy should prioritize equipment safety. Fixed-weight schemes cannot adapt to these dynamic changes and struggle to achieve an optimal, adaptive balance between the two. The challenge lies in designing a mechanism that can automatically adjust the optimization bias based on real-time conditions, placing high demands on the algorithm's adaptability.

[0006] Meanwhile, assessing equipment lifespan loss is itself a challenge. Lifespan loss is a cumulative, non-linear process closely related to the historical operating conditions of the load. Achieving accurate online lifespan loss assessment requires complex models and a large amount of computation. Embedding this computational process into real-time control loops that require rapid response and using it as an effective constraint or optimization objective has presented a challenge in engineering practice, highlighting the conflict between computational efficiency and model accuracy.

[0007] Therefore, the main problems with existing technologies are: the load control strategy of synchronous condensers fails to deeply coordinate with its real-time and comprehensive health status assessment; when dealing with the two competing objectives of grid voltage stability and equipment lifespan loss, there is a lack of an effective mechanism for adaptively and dynamically adjusting the optimization focus; and it is difficult to achieve online and rapid calculation and feedback of lifespan loss. This makes it difficult for the control strategy to maximize the service life of equipment while ensuring grid safety, and there is an urgent need for a new method that can integrate condition monitoring and load control in a unified and comprehensive manner. Summary of the Invention

[0008] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0009] Another objective of this invention is to provide a method and system for monitoring the status and controlling the load of synchronous condensers for reactive power compensation in power grids. This addresses the technical problem that in existing power grid reactive power compensation systems, the load control of synchronous condensers often relies solely on grid demand, lacking consideration of the real-time health status of the units themselves. This can lead to control commands potentially exacerbating equipment wear and even causing malfunctions. Furthermore, traditional methods struggle to dynamically balance the conflicting objectives of ensuring grid voltage stability and extending equipment lifespan.

[0010] To achieve these objectives and other advantages of the present invention, a method for monitoring the status of synchronous condenser units and controlling loads for reactive power compensation in a power grid is provided, comprising the following steps: S1. Voltage and frequency signals are collected in real time by voltage transformers and current transformers arranged at key nodes of the power grid, and vibration, temperature and excitation current signals are collected in real time by vibration sensors, temperature sensors and excitation current sensors installed on the main body of the synchronous condenser. S2. Calculate the current reactive power deficit of the power grid based on voltage and frequency signals; S3. Wavelet packet decomposition is performed on the vibration signal to extract frequency band energy features. A wavelet basis adaptive selection algorithm based on Shannon entropy is used to select a matching wavelet basis function for the current vibration signal. The vibration signal, temperature signal and excitation current signal are normalized. The extracted frequency band energy features, the processed normalized temperature signal and the normalized excitation current signal are combined into a feature vector. The feature vector is then input into the health status assessment model to obtain the real-time health state index (HSI) of the synchronous condenser group. S4. Based on reactive power deficit and real-time health status index (HSI), the optimal load control command is dynamically calculated and output to the excitation control system of the synchronous condenser group through a multi-objective particle swarm optimization algorithm. The multi-objective particle swarm optimization algorithm takes minimizing the equipment life loss of the synchronous condenser group and optimizing the grid voltage stability as common optimization objectives. An adaptive weight strategy based on the real-time health status index is introduced to dynamically adjust the weight coefficients of each objective. The weight coefficient of the objective of minimizing equipment life loss is w1, and the weight coefficient of the objective of optimizing grid voltage stability is w2. The weight calculation formulas are: w1=HSI, w2=1-HSI. At the same time, during the optimization process, (based on the online life loss calculation model based on the rainflow counting method and Miner's linear cumulative damage theory), the real-time calculated life loss value is used as the constraint condition for the optimization search. S5. The excitation control system adjusts the excitation voltage of the synchronous condenser group according to the load control command, and controls the reactive power output.

[0011] Preferably, the method for monitoring and controlling the condition of synchronous condenser units for reactive power compensation in power grids further includes: S6. After the load control command is executed, the vibration, temperature, and excitation current response data of the synchronous condenser unit, as well as the voltage and frequency recovery data of key nodes in the power grid, are collected in real time. The response data and power grid recovery data are fused using a Kalman filter. Based on the data processed by the Kalman filter, the mapping relationship between the feature vectors and the health status index (HSI) in the health status assessment model is corrected online. At the same time, the power grid reactive power-voltage sensitivity coefficient and equipment life loss model parameters used in the multi-objective particle swarm optimization algorithm are updated with feedback.

[0012] Preferably, in the method for monitoring and controlling the condition of synchronous condenser units for reactive power compensation in the power grid, the specific process of constructing and running the health status assessment model in step S3 includes: Configure a multi-task deep learning network whose input is a feature vector composed of vibration frequency band energy features extracted by wavelet packet decomposition, normalized temperature signal, and normalized excitation current signal. The first task branch of the network outputs a health state vector. Each element of the health state vector corresponds to the quantized health state score of a key component of the synchronous condenser group. The key components include at least the rotor, bearing, stator winding and cooling system. The second task branch of the network outputs a fault propagation probability matrix, which is generated by a parallel graph neural network module. The nodes of the graph neural network are configured to correspond to key components, and the edge weights are dynamically updated based on the time-domain correlation coefficient between the vibration, temperature and excitation current signals collected in real time. The output layer of the second task branch adopts a Bayesian neural network structure, which describes the uncertainty of the network weights through a probability distribution. The multi-task deep learning network is also equipped with a task weight adjuster, which dynamically adjusts the weight ratio of the two tasks in the total loss function based on the gradient magnitude ratio of the health state vector output by the first task branch and the fault propagation probability matrix output by the second task branch during the training process.

[0013] Preferably, the method for monitoring and controlling the status of synchronous condensers for reactive power compensation in power grids includes the following steps in calculating the real-time health status index (HSI): Construct a health status vector containing the key components of the camera shifter group; Based on the characteristics of vibration, temperature and excitation current signals collected in real time, the fault propagation probability matrix that characterizes the fault influence relationship between components is dynamically calculated. Multiply the health state vector by the fault propagation probability matrix to obtain the corrected health state vector; The corrected health state vector is then normalized. The element with the smallest value in the normalized vector is selected as the real-time health status index (HSI) of the entire camera group.

[0014] Preferably, in the method for monitoring and controlling the status of synchronous condenser units for reactive power compensation in the power grid, the adaptive weighting strategy in step S4 is implemented as follows: Multiplying the real-time health status index (HSI) by the correction factor yields the weighting coefficient w1 for minimizing equipment lifespan loss. This correction factor is calculated based on the current load rate of the synchronous condenser unit, with the load rate influence factor ranging from 20% to 50%. The weighting coefficient for the optimal target of grid voltage stability is w2 = 1 - w1; When the detected grid frequency deviation exceeds 0.15 Hz, the weighting coefficient w2 is set to 0.8.

[0015] Preferably, in the method for monitoring and controlling the condition of synchronous condenser units for reactive power compensation in the power grid, the constraint construction of the multi-objective particle swarm optimization algorithm in step S4 includes: The failure risk index of each component is calculated based on the failure propagation probability matrix. This index is the sum of the products of all elements in the corresponding column of the matrix and their health status scores. The component with the highest failure risk index was selected as the critical vulnerable component. A two-tiered lifespan loss constraint mechanism is established. The first tier is a global constraint, requiring that the overall lifespan loss of the unit does not exceed the preset global maximum allowable value L. max The second level is a local constraint, which requires the life loss value L of critical and vulnerable components to be within acceptable limits. a ≤L max ×(1-(α×R I α is the risk adjustment coefficient, with a value ranging from 0.1 to 0.3, R I Risk index; When the continuous operating time of the synchronous condenser exceeds 70% of its overhaul cycle, all life loss constraint thresholds will be reduced by 20%.

[0016] The present invention also provides a synchronous condenser unit status monitoring and load control system for reactive power compensation in power grids that implements the above method, comprising: The data acquisition module is configured to acquire voltage and frequency signals in real time through voltage transformers and current transformers arranged at key nodes of the power grid, and to acquire vibration, temperature and excitation current signals in real time through vibration sensors, temperature sensors and excitation current sensors installed on the synchronous condenser unit body. The reactive power deficit calculation module is configured to calculate the current reactive power deficit of the power grid based on the voltage signal and frequency signal. The health status assessment module is configured to extract frequency band energy features from the vibration signal by wavelet packet decomposition, select matching wavelet basis functions for the current vibration signal using a wavelet basis adaptive selection algorithm based on Shannon entropy, normalize the temperature signal and excitation current signal, combine the extracted frequency band energy features, the processed normalized temperature signal and the normalized excitation current signal into a feature vector, and input the feature vector into the health status assessment model to obtain the real-time health status index (HSI) of the synchronous condenser group. The optimized control command generation module is configured to dynamically calculate and output the optimal load control command to the excitation control system of the synchronous condenser group based on the reactive power deficit and the real-time health status index (HSI) using a multi-objective particle swarm optimization algorithm. The multi-objective particle swarm optimization algorithm takes minimizing the equipment life loss of the synchronous condenser group and optimizing the grid voltage stability as common optimization objectives, and introduces an adaptive weight strategy based on the real-time health status index to dynamically adjust the weight coefficients of each objective. The weight coefficient of the objective of minimizing equipment life loss is w1, and the weight coefficient of the objective of optimizing grid voltage stability is w2. The weight calculation formulas are: w1=HSI, w2=1-HSI. At the same time, during the optimization process, the online life loss calculation model based on the rainflow counting method and Miner's linear cumulative damage theory is used as the constraint condition for optimization search. The excitation control execution module is configured to adjust the excitation voltage of the synchronous condenser group according to the load control command, and control the reactive power output.

[0017] The present invention also provides an electronic device comprising: At least one processor; Memory, which stores computer programs; When the computer program is executed by at least one processor, the at least one processor implements the above-described method for monitoring the status of synchronous condensers and controlling loads for reactive power compensation in a power grid.

[0018] The present invention has at least the following beneficial effects: 1. This invention achieves coordinated consideration of grid demand and equipment health status during load control command generation by real-time acquisition of multi-source data from the power grid and generating units, and by introducing the Health Status Index (HSI) as a key decision variable. Utilizing a multi-objective particle swarm optimization algorithm, and embedding an HSI-based adaptive weighting strategy and online lifetime loss constraints, it dynamically and intelligently seeks the optimal balance between ensuring grid voltage stability and minimizing equipment lifetime loss. This effectively avoids the problem of excessively sacrificing long-term equipment lifespan to meet short-term grid demand, significantly improving the economy and safety of synchronous condenser unit operation, and providing core technical support for predictive maintenance and the stable operation of smart grids.

[0019] 2. This invention introduces a post-execution feedback and online correction mechanism, forming a complete "perception-decision-execution-learning" closed loop. A Kalman filter is used to fuse the post-execution response data, improving the accuracy of state perception. Based on this, the mapping relationship of the health status assessment model and key parameters in the optimization algorithm are corrected and updated online, enabling the entire system to possess self-learning and adaptive capabilities. This allows the system to track the slow degradation of equipment performance and the time-varying characteristics of power grid parameters, effectively overcoming initial model errors and time-varying drift problems. This ensures the accuracy and robustness of state assessment and control decisions during long-term operation, extending the system's effective service life.

[0020] 3. This invention employs a multi-task deep learning network architecture, simultaneously performing health status assessment and fault propagation modeling. It fully leverages the inherent correlations between multi-source monitoring data. The health status vector provides a local health profile of each component, while the fault propagation probability matrix dynamically generated by the graph neural network clearly quantifies the fault impact relationships between components, enabling the prediction of potential fault chain reactions. The application of Bayesian neural networks expresses the uncertainty of model predictions, enhancing the reliability of the results. The task weight adaptive mechanism ensures that the two tasks develop in a balanced manner during model training, avoiding the problem of one task dominating while neglecting the other. This significantly improves the depth and breadth of status assessment, laying the foundation for more refined fault early warning and risk prevention.

[0021] 4. The HSI calculation method provided by this invention not only considers the independent states of each component, but more importantly, it corrects the original health state vector through a fault propagation probability matrix, reflecting the risk that the weakest component may be further deteriorated due to the involvement of other component failures. Finally, the minimum value in the normalized vector is selected as the overall HSI, adhering to the "barrel principle" to ensure that the index can sensitively and conservatively reflect the overall health level of the unit, especially the currently highest-risk and most critical aspect. This provides a crucial and reliable input for subsequent optimized control, enabling the control strategy to always focus on the most severe current risk point.

[0022] 5. The adaptive weighting strategy of this invention enables dynamic and refined adjustment of the target weights. The weight coefficients w1 and w2 are no longer fixed, but are closely related to the real-time health status (HSI) and current load rate of the synchronous condenser unit. When the unit's health status is poor or the load is high, the weight (w1) of the equipment life loss target is automatically increased, and the control strategy becomes more conservative to protect the equipment. When there are emergencies such as severe frequency deviations in the power grid, priority is given to ensuring grid stability (assigning a higher fixed value to w2). This dynamic weighting mechanism enables the control strategy to have situational awareness, allowing it to make the most reasonable trade-offs under different operating conditions. This ensures the safety of the power grid in extreme situations and maximizes the extension of equipment life under normal circumstances.

[0023] 6. The dual-level lifespan loss constraint mechanism of this invention represents a leap from "extensive" overall constraints to "refined" local constraints. It not only constrains the overall lifespan loss of the machine, but more importantly, identifies key vulnerable components and applies stricter local constraints (L) related to the risk index to them. a ≤L max ×(1-(α×R I This ensures that the optimization algorithm actively avoids operation points that could cause further serious damage to components already in a high-risk state when searching for solutions. Furthermore, the constraint threshold is adaptively adjusted according to the overhaul cycle progress, reflecting a preventative maintenance approach based on runtime. This constraint mechanism greatly enhances the protection of weak points within the equipment, effectively preventing premature local failures that could lead to system downtime and improving the overall reliability of the system.

[0024] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to embodiments, so that those skilled in the art can implement it based on the description.

[0026] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0027] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0028] In the description of this invention, the terms "lateral", "longitudinal", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0029] In one embodiment of the present invention, a method for monitoring the status of synchronous condenser units and controlling loads for reactive power compensation in a power grid includes the following steps: S1. Voltage and frequency signals are collected in real time by voltage transformers and current transformers arranged at key nodes of the power grid, and vibration, temperature and excitation current signals are collected in real time by vibration sensors, temperature sensors and excitation current sensors installed on the main body of the synchronous condenser. S2. Calculate the current reactive power deficit of the power grid based on voltage and frequency signals; S3. Wavelet packet decomposition is performed on the vibration signal to extract the frequency band energy features. The wavelet basis adaptive selection algorithm based on Shannon entropy is used to select the matching wavelet basis function for the current vibration signal. The vibration signal, temperature signal and excitation current signal are normalized. The extracted frequency band energy features, the processed normalized temperature signal and the normalized excitation current signal are combined into a feature vector. The feature vector is then input into the health status assessment model to obtain the real-time health status index HSI of the synchronous condenser group. S4. Based on reactive power deficit and real-time health status index (HSI), the optimal load control command is dynamically calculated and output to the excitation control system of the synchronous condenser group through a multi-objective particle swarm optimization algorithm. The multi-objective particle swarm optimization algorithm takes minimizing the equipment life loss of the synchronous condenser group and optimizing the grid voltage stability as common optimization objectives. An adaptive weight strategy based on the real-time health status index is introduced to dynamically adjust the weight coefficients of each objective. The weight coefficient of the objective of minimizing equipment life loss is w1, and the weight coefficient of the objective of optimizing grid voltage stability is w2. The weight calculation formulas are: w1=HSI, w2=1-HSI. At the same time, during the optimization process, (based on the online life loss calculation model based on the rainflow counting method and Miner's linear cumulative damage theory), the real-time calculated life loss value is used as the constraint condition for the optimization search. S5. The excitation control system adjusts the excitation voltage of the synchronous condenser group according to the load control command, and controls the reactive power output.

[0030] This embodiment relates to a method for monitoring the status and controlling the load of synchronous condensers for reactive power compensation in power grids, aiming to solve the technical problem of the disconnect between the control strategy of synchronous condensers and the actual health status of the equipment in existing technologies. In power system operation, synchronous condensers bear the important responsibility of providing reactive power and stabilizing grid voltage. Traditional control methods mainly rely on real-time grid parameters (such as voltage deviation and frequency deviation) to generate control commands, prioritizing grid stability. While this method can quickly respond to system demands, it neglects the health status of the synchronous condenser itself as a complex electromechanical device. During operation, key components such as rotors, bearings, and windings gradually age due to electrical, thermal, and mechanical stresses. Traditional control methods cannot detect this gradual change in internal state, and may still command the equipment to bear high loads even when potential hazards exist or the equipment is in a sub-healthy state, thereby accelerating equipment wear and even inducing faults, ultimately endangering grid safety. This is a long-standing and unresolved contradiction: how to meet the dynamic demands of the grid while maximizing the service life of the equipment. The difficulty lies in the lack of an indicator that can quantify the health status of the equipment and deeply integrate it with grid control objectives.

[0031] The specific implementation method is as follows: First, system voltage and frequency signals are acquired through instrument transformers installed at key nodes of the power grid. Simultaneously, vibration, temperature, and excitation current signals are acquired in real time through various sensors installed on the synchronous condenser unit. Based on the power grid signals, the current reactive power deficit of the system is calculated, which is the reactive power that needs to be compensated by the synchronous condenser unit. On the other hand, the acquired unit body signals are subjected to in-depth processing. The vibration signal is decomposed by wavelet packet to extract its frequency band energy features, and an information entropy-based algorithm is used to adaptively select the most suitable wavelet basis function to ensure the accuracy of feature extraction. The temperature and excitation current signals are normalized. These processed features are combined into a comprehensive feature vector and input into a pre-trained health status assessment model. This model outputs a quantified real-time health status index (HSI), which can comprehensively reflect the current operating status of the synchronous condenser unit.

[0032] Subsequently, the control core will simultaneously consider the reactive power deficit of the power grid and the real-time health status index (HSI) of the generating units. A multi-objective particle swarm optimization algorithm is employed to dynamically calculate the optimal load control command. This algorithm explicitly sets two optimization objectives: minimizing the lifetime loss of synchronous condenser units and optimizing grid voltage stability. These two objectives are inherently competitive. To resolve this contradiction, the algorithm introduces an adaptive weighting strategy based on HSI. Specifically, when the HSI value is high, indicating good equipment health, the algorithm assigns a higher weight to the "minimize lifetime loss" objective, allowing the generating units to operate under better equipment loss conditions. Conversely, when the HSI value is low, indicating poor equipment condition, a higher weight is assigned to the "optimal voltage stability" objective, prioritizing grid safety while imposing protective limitations on the equipment. The weight allocation follows an intuitive linear relationship: the weight of the lifetime loss objective equals the HSI value, and the weight of the voltage stability objective equals one minus the HSI value. Furthermore, the optimization process is not an unlimited search; its constraints come from an online lifetime loss calculation model. This model, based on rainflow counting and cumulative damage theory, estimates in real time the potential lifetime loss to the equipment caused by the current control command, ensuring that the optimal solution found is within the equipment's tolerance range. Finally, the calculated optimal load control command is sent to the excitation control system of the synchronous condenser, which precisely controls its reactive power output by adjusting the excitation voltage.

[0033] Compared to the closest existing technologies, the biggest difference in this implementation lies in breaking down the information barrier between grid control and equipment condition monitoring. Existing technologies typically treat these two as relatively independent issues, or only employ passive protection methods such as simple threshold over-limit alarms. This invention creatively uses the HSI index, which characterizes the internal health status of equipment, as one of the core decision variables, directly participating in the optimization and generation process of control commands. This transforms the control strategy from open-loop or semi-open-loop to closed-loop decision-making, making reactive power compensation control no longer a simple response to grid demand, but a complex decision-making process capable of intelligently balancing and adaptively adjusting between grid security and equipment durability.

[0034] This implementation method shifts from passive response to proactive prevention. By introducing health status feedback into the control loop, it intervenes by adjusting load strategies before potential equipment problems escalate into failures, effectively extending the service life of key components and even the entire synchronous condenser unit. Simultaneously, with quantified awareness of equipment status, grid dispatchers can more clearly understand the equipment's capacity boundaries, enabling them to formulate more refined and scientific operating procedures while ensuring grid voltage stability, thus improving the overall system's operational economy and reliability. Ultimately, this method provides an effective technical path for achieving intelligent operation and maintenance and full lifecycle management of critical power system equipment.

[0035] According to another embodiment of the present invention, the method for monitoring the status of synchronous condenser units and controlling loads for reactive power compensation in power grids further includes: S6. After the load control command is executed, the vibration, temperature, and excitation current response data of the synchronous condenser unit, as well as the voltage and frequency recovery data of key nodes in the power grid, are collected in real time. The response data and power grid recovery data are fused using a Kalman filter. Based on the data processed by the Kalman filter, the mapping relationship between the feature vectors and the health status index (HSI) in the health status assessment model is corrected online. At the same time, the power grid reactive power-voltage sensitivity coefficient and equipment life loss model parameters used in the multi-objective particle swarm optimization algorithm are updated with feedback.

[0036] This implementation aims to address the problems of model inaccuracy and performance degradation in control systems during long-term operation. In complex industrial environments, the initial parameters of both assessment models reflecting the health status of the unit and power grid parameter and lifespan loss models used for optimization control gradually drift due to equipment aging, changes in operating conditions, and external environmental disturbances, resulting in deviations from the actual situation. The accumulation of these deviations leads to inaccurate state assessment results and unreliable model foundations for optimization algorithms, causing the final generated control commands to gradually deviate from the true optimal solution and potentially harbor hidden risks. Existing technologies typically use periodic shutdowns for maintenance to calibrate and correct the model. This method cannot achieve real-time performance and heavily relies on the experience of professionals. During the interval between maintenance periods, the system is actually in an open-loop or degrading operating state. How to construct a mechanism that can adaptively and automatically maintain model accuracy online is a key challenge in improving the long-term reliability and autonomy of the system.

[0037] The specific implementation method is as follows: After adjusting the excitation voltage of the synchronous condenser unit according to the load control command, thereby changing its reactive power output, the system does not stop working, but initiates a feedback correction process. The system continuously collects two types of data in real time: first, the response data of the synchronous condenser unit itself, including vibration, temperature, and excitation current signals, which reflect the changes in the unit's operating status after executing the new load command; second, the voltage and frequency recovery data of key nodes in the power grid, which directly reflects the adjustment effect of the control command on the stability of the power grid. These newly collected data constitute a dynamic data stream. Subsequently, a Kalman filter is used to fuse this data stream. This algorithm can optimally estimate the true state of the system and effectively filter out interference caused by measurement noise. The filtered clean data was used for two aspects of online calibration: first, to calibrate the health status assessment model, that is, to fine-tune the mapping relationship between the feature vector and the health status index (HSI) based on the latest input feature vector and the corresponding overall system response, so that the assessment model can track the gradual changes in equipment status; second, to update the key parameters relied upon by the multi-objective particle swarm optimization algorithm, mainly including the reactive power-voltage sensitivity coefficient of the power grid (which reflects the degree of influence of reactive power changes on voltage) and relevant parameters in the equipment life loss model, to ensure that the optimization algorithm always makes decisions based on the most realistic system model at present.

[0038] Compared to the closest existing technology, the core improvement of this implementation lies in the addition of closed-loop feedback and online learning capabilities. Existing technologies, including the systems they construct, typically have fixed model parameters or can only be set manually offline, making system operation essentially a "static optimization" process. This implementation, however, introduces response monitoring and data fusion after control command execution, and uses the results to back-correct the model, constructing a complete "monitoring-control-evaluation-correction" dynamic closed loop. This transforms the entire system from a mere automated program executing preset rules into an intelligent system with continuous learning and self-optimization capabilities.

[0039] This implementation significantly enhances the robustness and long-term stability of the system, effectively overcoming the negative impacts of time-varying model parameters and inaccurate initial settings. As a result, the health status assessment results become increasingly consistent with the actual condition of the equipment over time, and the load control commands generated by the optimized algorithm are better adapted to the current operating characteristics of the system. This reduces the over-reliance on the accuracy of the initial model, decreases the frequency of manual intervention and professional maintenance, and enables the system's autonomous and intelligent operation and maintenance. It fundamentally ensures the accuracy and effectiveness of the status monitoring and load control strategies throughout the entire equipment lifecycle.

[0040] According to another embodiment of the present invention, the specific process of constructing and running the health status assessment model in step S3 of the method for monitoring and controlling the condition of synchronous condenser units for reactive power compensation in power grids includes: Configure a multi-task deep learning network whose input is a feature vector composed of vibration frequency band energy features extracted by wavelet packet decomposition, normalized temperature signal, and normalized excitation current signal. The first task branch of the network outputs a health state vector. Each element of the health state vector corresponds to the quantized health state score of a key component of the synchronous condenser group. The key components include at least the rotor, bearing, stator winding and cooling system. The second task branch of the network outputs a fault propagation probability matrix, which is generated by a parallel graph neural network module. The nodes of the graph neural network are configured to correspond to key components, and the edge weights are dynamically updated based on the time-domain correlation coefficient between the vibration, temperature and excitation current signals collected in real time. The output layer of the second task branch adopts a Bayesian neural network structure, which describes the uncertainty of the network weights through a probability distribution. The multi-task deep learning network is also equipped with a task weight adjuster, which dynamically adjusts the weight ratio of the two tasks in the total loss function based on the gradient magnitude ratio of the health state vector output by the first task branch and the fault propagation probability matrix output by the second task branch during the training process.

[0041] This implementation aims to address the technical challenges of existing health status assessment methods being too general and static. Traditional assessment methods often treat the generator set as a whole, outputting a single, global health index. This simplistic approach fails to reveal the individual state differences of key components within the unit (such as rotors, bearings, and windings), let alone depict how a failure in one component affects and propagates to other components. Due to this lack of refined internal state insight, control strategies struggle to implement differentiated and predictive protection, failing to suppress the spread of localized faults in their early stages. A minor degradation in one component may ultimately trigger a chain reaction, leading to the shutdown of the entire unit. Furthermore, traditional neural network models are typically deterministic, outputting a fixed value and failing to express the degree of confidence in their own assessments—a significant drawback in complex and ever-changing industrial environments.

[0042] The specific implementation is as follows: The health status assessment model is constructed as a multi-task deep learning network. The input to this network is a comprehensive feature vector composed of vibration frequency band energy, normalized temperature, and normalized excitation current. The network structure is designed to simultaneously complete two closely related but different tasks. The first task branch is responsible for outputting a health status vector, where each element directly corresponds to the quantified health score of a specific key component of the synchronous condenser group. For example, the rotor, bearing, stator winding, and cooling system each have their own independent scores, thus achieving component-level status visualization. The second task branch goes deeper, aiming to output a fault propagation probability matrix. This matrix is ​​generated by a parallel graph neural network module, where the nodes of the graph are defined as various key components, and the weights of the edges connecting these nodes are not fixed but dynamically calculated and updated based on the temporal correlation between real-time acquired multi-source signals. This allows the model to learn and reflect the dynamic interaction relationships between components. Crucially, the output layer of this task branch adopts a Bayesian neural network structure. By introducing a probability distribution into the network weights, its output is no longer a deterministic matrix value but a probability distribution, thereby quantifying the uncertainty of the model's predictions. Finally, the entire multi-task network also incorporates a task weight adjuster, which automatically and dynamically adjusts the weight ratio of the two tasks in the overall learning objective based on the gradient magnitude ratio of the loss function during training, ensuring that the two tasks can be optimized in a balanced and collaborative manner.

[0043] Compared to the closest existing technologies, the core difference of this implementation lies in the depth, dimension, and dynamism of its assessment. Existing technologies typically provide a "black box" overall health assessment, while this solution offers a "white box" insight, not only providing independent status reports for each component but also revealing the interrelationships between them through dynamic graph neural networks. Its assessment results are upgraded from a single scalar to a multi-dimensional information system encompassing vectors, matrices, and uncertainties. The application of Bayesian neural networks further elevates the assessment from deterministic judgment to probabilistic inference.

[0044] This implementation significantly enhances the richness, interpretability, and reliability of condition assessment. By obtaining component-level condition decomposition, maintenance personnel can accurately pinpoint the specific components whose health is beginning to decline, thereby enabling predictive maintenance. The fault propagation model provides the ability to anticipate potential risks, allowing the control system to intervene at the nascent stage of a fault to prevent its spread. Probabilistic outputs increase the credibility of the results; when the model's predictions have high uncertainty, the system can adopt more conservative control strategies, improving the safety of decision-making. This deep state awareness lays a solid data foundation for the subsequent realization of truly intelligent adaptive control.

[0045] According to another embodiment of the present invention, the method for monitoring and controlling the status of synchronous condensers for reactive power compensation in power grids includes the following steps in calculating the real-time health status index (HSI): Construct a health status vector containing the key components of the camera shifter group; Based on the characteristics of vibration, temperature and excitation current signals acquired in real time, the fault propagation probability matrix that characterizes the fault influence relationship between components is dynamically calculated. Multiply the health state vector by the fault propagation probability matrix to obtain the corrected health state vector; The corrected health state vector is then normalized. The element with the smallest value in the normalized vector is selected as the real-time health status index (HSI) of the entire camera group.

[0046] This implementation aims to address the technical challenge of synthesizing a comprehensive index from component-level health status information that can both represent the overall health status and sensitively reflect the weakest link in the system. Simply calculating the average or maximum value as the overall health index after obtaining the independent status scores of each key component and the fault impact relationships between them is unscientific. The average value can mask serious defects in specific components, while the maximum value cannot reflect the risk of fault propagation. An ideal comprehensive index should be able to dynamically identify the component with the worst current health status and the most likely to trigger a chain reaction of failures, using the state of that component as the key bottleneck restricting the overall system's operational capability. Existing technologies lack a mechanism to effectively integrate the independent status of components with the system's interconnected structure, making it difficult to generate a conservative health index that can reliably guide high-risk operational decisions.

[0047] The specific implementation method is as follows: The calculation of the real-time health status index (HSI) is a multi-step information fusion process. First, based on the output of a multi-task deep learning network, a health status vector clearly depicting the independent operating states of each key component of the synchronous condenser group is obtained. Simultaneously, a fault propagation probability matrix is ​​dynamically calculated based on real-time sensor data. This matrix quantitatively describes the probability that a fault in any component will affect other components. Then, the health status vector is multiplied by the fault propagation probability matrix. The essence of this operation is to correct the original independent state scores using fault propagation relationships. The final score of a component depends not only on its own state but also on the drag from the states of other components that will affect it after a fault. After matrix correction, a corrected health status vector that better reflects the true risk status of the system is obtained. After normalizing this corrected vector, instead of using an arithmetic mean or weighted average, the value of the smallest element is selected as the real-time health status index (HSI) for the entire synchronous condenser group. This means that the overall health level of the system is determined by the most vulnerable and highest-risk component at present.

[0048] Compared to the closest existing technology, the core difference of this implementation lies in the logical depth of its comprehensive evaluation. Existing technologies provide two parallel types of information: component status and interrelationships, but do not specify how to use them for final decision-making. This solution provides an innovative information fusion algorithm that does not simply present the two types of information side-by-side, but rather deeply integrates "individual status" and "interrelated effects" through mathematical operations, ultimately extracting a single indicator with significant engineering conservatism and safety guidance. The formation mechanism of this indicator has been upgraded from "reporting status" to "assessing risk."

[0049] This implementation generates a highly reliable and safety-oriented comprehensive health index. By selecting the worst value as the final output, the index is inherently conservative and can always remind the control system to pay attention to the unit's maximum risk points. This avoids the possibility of ignoring serious local defects due to the average effect, making subsequent load control decisions based on this HSI value necessarily safer. It can proactively avoid further impacts on the most vulnerable components, thereby suppressing the risk of local faults escalating into global faults at the root, and greatly enhancing the robustness and safety of the entire system operation.

[0050] According to another embodiment of the present invention, in the method for monitoring and controlling the status of synchronous condenser units for reactive power compensation in a power grid, the adaptive weighting strategy in step S4 is implemented as follows: Multiplying the real-time health status index (HSI) by the correction factor yields the weighting coefficient w1 for minimizing equipment lifespan loss. This correction factor is calculated based on the current load rate of the synchronous condenser unit, with the load rate influence factor ranging from 20% to 50%. The weighting coefficient for the optimal target of grid voltage stability is w2 = 1 - w1; When the power grid frequency deviation is detected to exceed 0.15 Hz, the weighting coefficient w2 is set to 0.8.

[0051] This implementation aims to address the technical challenge of insufficient flexibility and robustness of adaptive weighting strategies under different operating conditions. While linear weight allocation based on the Health Status Index (HSI) achieves a basic trade-off, its adjustment logic remains relatively static. It fails to fully consider the direct impact of instantaneous load pressure on synchronous condenser units on equipment capacity, nor does it provide a special response to extreme emergency situations in the power grid. For example, when units are already operating at high load rates, the rate of equipment wear and tear naturally accelerates. In this case, allocating weights solely based on HSI values ​​may still allow for significant lifespan loss, failing to provide sufficient protection. On the other hand, when there is a significant deviation in the power grid frequency, this is usually a sign that system stability is under serious threat. In this situation, the highest priority control objective should be to restore power grid stability to the best of its ability, and any optimization based on equipment lifespan should temporarily take a backseat. How to further integrate information on two key operating conditions—current load level and power grid emergency states—on top of conventional adaptive strategies, enabling the weighting strategy to possess multi-level judgment capabilities, is a problem that requires in-depth solutions.

[0052] The specific implementation method is as follows: The core lies in introducing a correction mechanism into the calculation of the weighting coefficient w1 for the equipment lifespan loss target. This coefficient is no longer simply equal to HSI, but rather HSI is multiplied by a correction coefficient calculated from the current load rate of the synchronous condenser unit. The introduction of the load rate as a direct influencing factor allows the weight allocation to be sensitive to the current operating pressure level. When the unit is operating at a high load rate, this correction coefficient decreases, thereby reducing the value of w1. This means that in the optimization objectives, the importance of minimizing equipment lifespan loss is relatively reduced, reflecting respect for the equipment's capacity limits under high loads and strengthening its protection. Correspondingly, the weighting coefficient w2 for the optimal grid voltage stability target is always maintained at one minus w1 to ensure that the sum of the weights of the two objectives is a constant value. Furthermore, this strategy sets a highest-priority override rule: when the grid frequency deviation is detected to exceed a specific threshold of 0.15 Hz, the system will ignore other calculations and directly set the weighting coefficient w2 to a higher fixed value of 0.8. This clearly indicates that when the power grid shows serious signs of instability, the optimization algorithm must allocate the vast majority of its optimization weights to the voltage stability objective and concentrate all resources to prioritize the response to the power grid crisis.

[0053] Compared to the closest existing technology, the core improvement of this implementation lies in its multi-dimensional and hierarchical response mechanism for its weighting strategy. Existing technologies have a single adaptive dimension, relying solely on the slowly varying parameter of the Health Status Index (HSI). This solution, however, adds a response to the rapidly changing operating condition parameter of instantaneous load factor, and a highest-priority response to the special event of extreme grid frequency deviation. This transforms the weighting allocation strategy from a single-input, single-output regulator into an intelligent decision-maker with both fine-tuning capabilities in normal conditions and decisive handling in emergencies.

[0054] This implementation significantly improves the adaptability and decision-making rationality of the control strategy under different operating scenarios. By introducing load factor correction, the strategy automatically favors equipment protection when the unit is operating under high load, compensating for the potential lag in single HSI assessment and enhancing the timeliness of protection. Furthermore, the mandatory weight setting for large frequency deviations provides the system with clear action guidelines that do not require complex logical judgments in response to extreme emergencies, ensuring that the control system reacts quickly, decisively, and optimally when grid security is most severely threatened, greatly improving the overall system's operational robustness and reliability.

[0055] According to another embodiment of the present invention, in the method for monitoring and controlling the condition of synchronous condenser units for reactive power compensation in power grids, the constraint construction of the multi-objective particle swarm optimization algorithm in step S4 includes: The failure risk index of each component is calculated based on the failure propagation probability matrix. This index is the sum of the products of all elements in the corresponding column of the matrix and their health status scores. The component with the highest failure risk index was selected as the critical vulnerable component. A two-tiered lifespan loss constraint mechanism is established. The first tier is a global constraint, requiring that the overall lifespan loss of the unit does not exceed the preset global maximum allowable value L. max The second level is a local constraint, which requires the life loss value L of critical and vulnerable components to be within acceptable limits. a ≤L max ×(1-(α×R I α is the risk adjustment coefficient, with a value ranging from 0.1 to 0.3, R I Risk index; When the continuous operating time of the synchronous condenser exceeds 70% of its overhaul cycle, all life loss constraint thresholds will be reduced by 20%.

[0056] This implementation aims to address the technical challenges of existing lifespan loss constraint methods being overly broad and egalitarian. Traditional constraint methods typically set a uniform, global upper limit for lifespan loss for the entire synchronous condenser unit. This method implicitly assumes that the health status and importance of all components within the unit are the same. As a result, control commands generated by optimization algorithms may apply excessive stress to a critical, vulnerable component that is already in a sub-optimal state, even if the overall loss does not exceed the limit, thereby accelerating the component's failure process. Premature failure of a component can easily trigger a chain reaction, leading to unplanned downtime, with the final economic losses and safety risks far exceeding the uniform loss of the overall lifespan. Therefore, identifying critical, vulnerable components within the unit and tailoring stricter, more protective constraints for them is the core challenge in achieving refined control and differentiated protection.

[0057] The specific implementation method is as follows: The constraint construction method described is a dynamic and refined process. First, using the calculated fault propagation probability matrix and health state vector, a fault risk index is calculated for each component. This index is calculated by multiplying the component's corresponding column vector in the fault propagation matrix (representing the probability of faults from other components propagating to this component) with the health state vector (representing the current state of each component). This result comprehensively reflects the overall risk faced by the component due to its own poor condition and the influence of other components. The system then selects the component with the highest fault risk index and identifies it as the most critical vulnerable component that requires the most attention and protection. Based on this, the algorithm establishes a two-level lifetime loss constraint mechanism. The first level is a traditional global constraint, requiring that the overall lifetime loss value of the unit must not exceed a preset maximum allowable value. The second level is an innovative local constraint, specifically targeting the identified critical vulnerable components and setting a more stringent, individualized loss limit than the global constraint. This new limit is based on the global maximum value multiplied by a discount coefficient based on the component's risk index. The higher the risk, the larger the discount, the lower the allowable loss limit, and the stronger the protection. In addition, the constraint mechanism also has the ability to adapt in the time dimension. When the continuous running time of the synchronous condenser group exceeds a certain proportion of its overhaul cycle, the system will automatically lower all constraint thresholds further, reflecting the operating logic that the equipment nearing its overhaul period needs extra care.

[0058] Compared to the closest existing technology, the significant difference of this implementation lies in the precision and dynamism of its constraint strategy. Existing technologies and the preceding claims of this invention employ a "one-size-fits-all" approach to constraint, treating all components equally. This solution, however, achieves a shift from "equal constraint" to "precise constraint," identifying the weakest link in the system and implementing targeted protection. Its constraints are no longer static values ​​but dynamically adjusted variables based on real-time risk assessments and equipment operating cycles, making the constraint itself a proactive risk management tool.

[0059] This implementation significantly enhances the security of the control strategy and the foresight of equipment protection. By implementing differentiated local constraints, it effectively avoids secondary damage to the component with the worst health condition caused by control commands, significantly reducing the probability of unplanned shutdown of the entire unit due to premature failure of a single component. This dynamic constraint mechanism based on real-time risk guides the optimization search to a safer and more balanced solution space for the equipment as a whole. Thus, while ensuring the function of the power grid, it achieves targeted protection of critical vulnerable points of the equipment and extends the safe operating cycle of the unit.

[0060] The present invention also provides a synchronous condenser unit status monitoring and load control system for reactive power compensation in power grids that implements the above method, comprising: The data acquisition module is configured to acquire voltage and frequency signals in real time through voltage transformers and current transformers arranged at key nodes of the power grid, and to acquire vibration, temperature and excitation current signals in real time through vibration sensors, temperature sensors and excitation current sensors installed on the synchronous condenser unit body. The reactive power deficit calculation module is configured to calculate the current reactive power deficit of the power grid based on the voltage signal and frequency signal. The health status assessment module is configured to extract frequency band energy features from the vibration signal by wavelet packet decomposition, select matching wavelet basis functions for the current vibration signal using a wavelet basis adaptive selection algorithm based on Shannon entropy, normalize the temperature signal and excitation current signal, combine the extracted frequency band energy features, the processed normalized temperature signal and the normalized excitation current signal into a feature vector, and input the feature vector into the health status assessment model to obtain the real-time health status index (HSI) of the synchronous condenser group. The optimized control command generation module is configured to dynamically calculate and output the optimal load control command to the excitation control system of the synchronous condenser group based on the reactive power deficit and the real-time health status index (HSI) using a multi-objective particle swarm optimization algorithm. The multi-objective particle swarm optimization algorithm takes minimizing the equipment life loss of the synchronous condenser group and optimizing the grid voltage stability as common optimization objectives, and introduces an adaptive weight strategy based on the real-time health status index to dynamically adjust the weight coefficients of each objective. The weight coefficient of the objective of minimizing equipment life loss is w1, and the weight coefficient of the objective of optimizing grid voltage stability is w2. The weight calculation formulas are: w1=HSI, w2=1-HSI. At the same time, during the optimization process, the online life loss calculation model based on the rainflow counting method and Miner's linear cumulative damage theory is used as the constraint condition for optimization search. The excitation control execution module is configured to adjust the excitation voltage of the synchronous condenser group according to the load control command, and control the reactive power output.

[0061] This implementation provides a system for realizing the above method, specifically through the collaborative work of five core modules. The data acquisition module is responsible for interfacing and communicating with all sensors deployed at grid nodes and on the generator unit itself, ensuring real-time and synchronous acquisition of raw data such as voltage, frequency, vibration, temperature, and excitation current. The reactive power deficit calculation module receives grid data and executes a specific algorithm to quickly calculate the reactive power deficit that the current grid needs to compensate, providing target input for control. The health status assessment module is one of the intelligent cores of the entire system. It receives sensor data streams from the generator unit, initiates a series of calculation processes such as wavelet packet decomposition, feature extraction, and normalization, and calls the built-in health status assessment model for inference, ultimately outputting a quantified real-time health status index (HSI). The optimized control command generation module is another intelligent core. It comprehensively receives two key inputs from the reactive power deficit calculation module and the health status assessment module, runs a multi-objective particle swarm optimization algorithm, and dynamically solves for the current optimal load control command while considering online lifetime loss constraints and adaptive weighting strategies. The excitation control execution module is responsible for converting the digital command into actual physical actions. It sends signals to the excitation control system of the synchronous condenser group through an interface to adjust the excitation voltage, thereby precisely controlling the output of reactive power. These five modules are connected in sequence to form a complete closed loop from data sensing to control execution.

[0062] The present invention also provides an electronic device comprising: At least one processor; Memory, which stores computer programs; When the computer program is executed by at least one processor, the at least one processor implements the above-described method for monitoring the status of synchronous condensers and controlling loads for reactive power compensation in a power grid.

[0063] The electronic device, with its core structure consistent with that of a general-purpose industrial computer or server architecture, mainly includes at least one processor and memory. The memory stores computer program instructions that implement the methods described above. These instructions constitute the complete operational logic, covering all steps from data acquisition interface calls, reactive power deficit calculation, Health Status Index (HSI) assessment, multi-objective optimization algorithm execution, to the generation of final control commands. When the electronic device is powered on, the processor reads and executes these computer program instructions from memory, transforming this general-purpose computing device into a dedicated device for intelligent monitoring and control of synchronous condenser groups. Its operation involves the processor continuously executing instruction loops, processing input data in real time, running built-in algorithm models, and ultimately outputting control commands.

[0064] An embodiment of intelligent reactive power control for synchronous condenser units in a converter station (1) Background: At a ±800kV UHVDC converter station in a power grid, a momentary short-circuit fault occurred in the remote sending-end power grid, causing a brief fluctuation in the DC power supplied to the grid. The AC bus voltage of the converter station dropped, and the frequency shifted slightly (from 50.00Hz to 49.82Hz). A 300Mvar synchronous condenser unit in the station needed to quickly increase its reactive power output to support the grid voltage to stabilize. However, it was also necessary to ensure that the unit itself (especially the slightly aged bearing components) would not suffer excessive lifespan loss due to this emergency response.

[0065] (2) Control method flow Step 1: Data Acquisition and Preprocessing 1.1 Data Acquisition: Grid Side: Voltage transformers recorded an AC bus voltage drop from 525kV to 518kV. Frequency measurement unit recorded a system frequency of 49.82Hz. Unit Side: Vibration sensor: Recorded vertical vibration signal of the bearing housing; the time-domain waveform amplitude slightly increased. Temperature sensor: Measured bearing temperature stable at 65℃; stator winding hot spot temperature at 82℃. Excitation System: Recorded current excitation current as 2500A.

[0066] 1.2 Pretreatment: Wavelet packet decomposition was performed on the vibration signal to extract energy characteristics from multiple frequency bands, including 0-200Hz and 200-500Hz. It was found that the energy in the 200-500Hz band increased significantly compared to the baseline, which is typically associated with early defects in the bearing.

[0067] The Shannon entropy algorithm was used, and the 'db4' wavelet basis was adaptively selected for decomposition to best match the characteristics of the current vibration signal.

[0068] The temperature signal (65, 82) and the excitation current signal (2500) are normalized and combined with the extracted vibration frequency band energy features to form a feature vector F.

[0069] Step 2: Health Status Index (HSI) Calculation 2.1 Reasoning for the Health Status Assessment Model: The feature vector F is input into the pre-trained multi-task deep learning model.

[0070] Task 1 outputs a health status vector H: [Rotor: 0.85, Bearing: 0.35, Stator Winding: 0.90, Cooling System: 0.95]. This indicates that the bearing's health status score is low, making it the weakest link.

[0071] In Task 2, the graph neural network dynamically updates the influence relationships between components based on real-time signal correlation, outputting a fault propagation probability matrix P. The matrix shows that if a bearing fails, there is a high probability that it will affect the rotor.

[0072] The model calculates a corrected health state vector H' = H × P. Calculations show that the bearing's score is further reduced after correction due to its poor inherent condition and susceptibility to influence from other components.

[0073] After normalizing H', the minimum value is taken to obtain the real-time health status index of the whole machine, HSI=0.32 (mainly determined by the bearing condition).

[0074] Step 3: Calculation of reactive power deficit 3.1 Calculation: Based on the voltage drop (525kV->518kV) and frequency deviation (-0.18Hz), the control system quickly calculates the current reactive power deficit of the power grid, ΔQ=180Mvar.

[0075] Step 4: Multi-objective optimization and control instruction generation 4.1 Optimize issue settings: Objective 1: Minimize unit life loss (especially bearings).

[0076] Objective 2: Maximize voltage stability (rapidly compensate for 180Mvar shortfall).

[0077] Adaptive weights: The current HSI is 0.32, and the unit load factor is 85%. The calculated weights are w1 = 0.32(1 - 0.85 × 0.3) ≈ 0.24 (load factor correction factor is 0.3). w2 = 1 - 0.24 = 0.76.

[0078] However, the system detected a grid frequency deviation |-0.18Hz|>0.15Hz threshold, triggering an emergency rule and forcibly setting w2=0.8 and w1=0.2, indicating that the system prioritizes grid stability.

[0079] Constraints: Identifying critical vulnerable components: Based on the failure risk index, the bearing was identified as a critical vulnerable component.

[0080] Global constraint: The overall lifespan loss during this action must not exceed L. max (Default value).

[0081] Local constraint: The bearing's life loss must not exceed L. max ×(1-(0.2×0.85)) (Risk adjustment coefficient α is taken as 0.2, bearing risk index R I (It is 0.85), which is a more stringent constraint.

[0082] 4.2 MOPSO Optimization Solution: The multi-objective particle swarm optimization algorithm searches under strict local constraints and a weighting system that prioritizes voltage stability.

[0083] The algorithm ultimately finds a trade-off solution: the output command is to increase the excitation current to 2900A, which is expected to output 170Mvar of reactive power (slightly lower than the deficit, but can significantly alleviate voltage drop), and the bearing life loss is predicted to be within the safe constraint under this command.

[0084] Step 5: Control Execution and Feedback 5.1 Instruction Execution: The optimized control command is sent to the excitation system. The excitation control system smoothly increases the excitation current from 2500A to 2900A.

[0085] The reactive power output of the synchronous condenser increased accordingly, the grid voltage began to rise, and the frequency gradually recovered.

[0086] 5.2 Feedback and Learning: The system collects data after the command is executed: Unit response: The bearing vibration value increased to a certain extent after the excitation current was increased (within the predicted range), and the temperature slowly rose to 68℃.

[0087] Grid response: Voltage restored to 522kV, frequency restored to 49.92Hz.

[0088] The Kalman filter performs fusion filtering on these data.

[0089] Online calibration: Based on the new (vibration, temperature, excitation current) data and the final HSI performance, the mapping relationship between features and HSI in the health status assessment model is fine-tuned to make its assessment of bearing condition more accurate.

[0090] The reactive power-voltage sensitivity coefficient is updated based on the ratio of the actual voltage recovery value (522kV) to the reactive power output increment (~120Mvar).

[0091] Based on the actual load change process, the parameters of the life loss model were corrected.

[0092] (3) In this example, the invention method demonstrates its core advantages: 1. Collaborative decision-making: Instead of blindly compensating for the 180Mvar shortfall in full, the decision-making process took into account both the grid's urgent needs (large frequency deviation) and the equipment's health status (low bearing score).

[0093] 2. Adaptive trade-off: The weight strategy is dynamically adjusted, with more emphasis on equipment lifespan during normal times (w1 is larger), and priority given to grid stability during emergencies (w2 is forced to 0.8).

[0094] 3. Refined protection: By using local constraints, the most vulnerable bearing components are given key protection, avoiding secondary damage that may be caused by "one-size-fits-all" control.

[0095] 4. Continuous learning: The system performs online corrections based on feedback data after execution, enabling its model to make more accurate assessments and decisions in the future.

[0096] Ultimately, the technical solution provided by this invention ensures rapid restoration of grid voltage and prevents the escalation of accidents, while maximizing the protection of key equipment in synchronous condenser units, extending their service life, and achieving the optimal balance between grid safety and equipment safety.

[0097] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0098] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details.

Claims

1. A method for condition monitoring and load control of synchronous condenser units for reactive power compensation in power grids, characterized in that, Includes the following steps: S1. Voltage and frequency signals are collected in real time by voltage transformers and current transformers arranged at key nodes of the power grid, and vibration, temperature and excitation current signals are collected in real time by vibration sensors, temperature sensors and excitation current sensors installed on the main body of the synchronous condenser. S2. Calculate the current reactive power deficit of the power grid based on voltage and frequency signals; S3. Wavelet packet decomposition is performed on the vibration signal to extract the frequency band energy features. The wavelet basis adaptive selection algorithm based on Shannon entropy is used to select the matching wavelet basis function for the current vibration signal. The vibration signal, temperature signal and excitation current signal are normalized. The extracted frequency band energy features, the processed normalized temperature signal and the normalized excitation current signal are combined into a feature vector. The feature vector is then input into the health status assessment model to obtain the real-time health status index HSI of the synchronous condenser group. S4. Based on reactive power deficit and real-time health status index (HSI), a multi-objective particle swarm optimization algorithm is used to dynamically calculate and output the optimal load control command to the excitation control system of the synchronous condenser. The multi-objective particle swarm optimization algorithm takes minimizing the lifetime loss of the synchronous condenser equipment and optimizing the grid voltage stability as common optimization objectives. An adaptive weight strategy based on the real-time health status index is introduced to dynamically adjust the weight coefficients of each objective. The weight coefficient of the objective of minimizing equipment lifetime loss is w1, and the weight coefficient of the objective of optimizing grid voltage stability is w2. The weight calculation formulas are: w1 = HSI, w2 = 1 - HSI. At the same time, during the optimization process, the real-time calculated lifetime loss value is used as the constraint condition for the optimization search. S5. The excitation control system adjusts the excitation voltage of the synchronous condenser group according to the load control command to control the reactive power output. The adaptive weighting strategy in step S4 is implemented as follows: Multiplying the real-time health status index (HSI) by the correction factor yields the weighting coefficient w1 for minimizing equipment lifespan loss. This correction factor is calculated based on the current load rate of the synchronous condenser unit, with the load rate influence factor ranging from 20% to 50%. The weighting coefficient for the optimal target of grid voltage stability is w2 = 1 - w1; When the power grid frequency deviation is detected to exceed 0.15 Hz, the weighting coefficient w2 is set to 0.

8.

2. The method for condition monitoring and load control of synchronous condenser units for reactive power compensation in power grids as described in claim 1, characterized in that, Also includes: S6. After the load control command is executed, the vibration, temperature, and excitation current response data of the synchronous condenser unit, as well as the voltage and frequency recovery data of key nodes in the power grid, are collected in real time. The response data and power grid recovery data are fused using a Kalman filter. Based on the data processed by the Kalman filter, the mapping relationship between the feature vectors and the health status index (HSI) in the health status assessment model is corrected online. At the same time, the power grid reactive power-voltage sensitivity coefficient and equipment life loss model parameters used in the multi-objective particle swarm optimization algorithm are updated with feedback.

3. The method for condition monitoring and load control of synchronous condenser units for reactive power compensation in power grids as described in claim 2, characterized in that, The specific process of building and running the health status assessment model in step S3 includes: Configure a multi-task deep learning network whose input is a feature vector composed of vibration frequency band energy features extracted by wavelet packet decomposition, normalized temperature signal, and normalized excitation current signal. The first task branch of the network outputs a health state vector. Each element of the health state vector corresponds to the quantized health state score of a key component of the synchronous condenser group. The key components include at least the rotor, bearing, stator winding and cooling system. The second task branch of the network outputs a fault propagation probability matrix, which is generated by a parallel graph neural network module. The nodes of the graph neural network are configured to correspond to key components, and the edge weights are dynamically updated based on the time-domain correlation coefficient between the vibration, temperature and excitation current signals collected in real time. The output layer of the second task branch adopts a Bayesian neural network structure, which describes the uncertainty of the network weights through a probability distribution. The multi-task deep learning network is also equipped with a task weight adjuster, which dynamically adjusts the weight ratio of the two tasks in the total loss function based on the gradient magnitude ratio of the health state vector output by the first task branch and the fault propagation probability matrix output by the second task branch during the training process.

4. The method for status monitoring and load control of synchronous condenser units for reactive power compensation in power grids as described in claim 3, characterized in that, The calculation of the real-time health status index (HSI) includes the following steps: Construct a health status vector containing the key components of the camera shifter group; Based on the characteristics of vibration, temperature and excitation current signals collected in real time, the fault propagation probability matrix that characterizes the fault influence relationship between components is dynamically calculated. Multiply the health state vector by the fault propagation probability matrix to obtain the corrected health state vector; The corrected health state vector is then normalized. The element with the smallest value in the normalized vector is selected as the real-time health status index (HSI) of the entire camera group.

5. The method for condition monitoring and load control of synchronous condenser units for reactive power compensation in power grids as described in claim 4, characterized in that, The constraint construction for the multi-objective particle swarm optimization algorithm in step S4 includes: The failure risk index of each component is calculated based on the failure propagation probability matrix. This index is the sum of the products of all elements in the corresponding column of the matrix and their health status scores. The component with the highest failure risk index was selected as the critical vulnerable component. A two-tiered lifespan loss constraint mechanism is established. The first tier is a global constraint, requiring that the overall lifespan loss of the unit does not exceed the preset global maximum allowable value L. max The second level is a local constraint, which requires the life loss value L of critical and vulnerable components to be within acceptable limits. a ≤L max ×(1-(α×R I α is the risk adjustment coefficient, with a value ranging from 0.1 to 0.3, R I Risk index; When the continuous operating time of the synchronous condenser exceeds 70% of its overhaul cycle, all life loss constraint thresholds will be reduced by 20%.

6. A synchronous condenser unit status monitoring and load control system for reactive power compensation in a power grid, implementing the method as described in any one of claims 1-5, characterized in that, include: The data acquisition module is configured to acquire voltage and frequency signals in real time through voltage transformers and current transformers arranged at key nodes of the power grid, and to acquire vibration, temperature and excitation current signals in real time through vibration sensors, temperature sensors and excitation current sensors installed on the synchronous condenser unit body. The reactive power deficit calculation module is configured to calculate the current reactive power deficit of the power grid based on the voltage signal and frequency signal. The health status assessment module is configured to extract frequency band energy features from the vibration signal by wavelet packet decomposition, select matching wavelet basis functions for the current vibration signal using a wavelet basis adaptive selection algorithm based on Shannon entropy, normalize the temperature signal and excitation current signal, combine the extracted frequency band energy features, the processed normalized temperature signal and the normalized excitation current signal into a feature vector, and input the feature vector into the health status assessment model to obtain the real-time health status index (HSI) of the synchronous condenser group. The optimized control command generation module is configured to dynamically calculate and output the optimal load control command to the excitation control system of the synchronous condenser group based on the reactive power deficit and the real-time health status index (HSI) using a multi-objective particle swarm optimization algorithm. The multi-objective particle swarm optimization algorithm takes minimizing the equipment life loss of the synchronous condenser group and optimizing the grid voltage stability as common optimization objectives, and introduces an adaptive weight strategy based on the real-time health status index to dynamically adjust the weight coefficients of each objective. The weight coefficient of the objective of minimizing equipment life loss is w1, and the weight coefficient of the objective of optimizing grid voltage stability is w2. The weight calculation formulas are: w1=HSI, w2=1-HSI. At the same time, during the optimization process, the online life loss calculation model based on the rainflow counting method and Miner's linear cumulative damage theory is used as the constraint condition for optimization search. The excitation control execution module is configured to adjust the excitation voltage of the synchronous condenser group according to the load control command, and control the reactive power output.

7. An electronic device, characterized in that, include: At least one processor; Memory, which stores computer programs; When the computer program is executed by at least one processor, the at least one processor implements the synchronous condenser group status monitoring and load control method for reactive power compensation in power grids as described in claims 1-5.