Low-frequency oscillation suppression methods, systems and equipment for generator sets

Abstract

This application provides a method, system, and equipment for suppressing low-frequency oscillations in generator sets, relating to the field of generator technology. The methods include: real-time monitoring of the generator set to obtain a monitoring sequence; low-frequency oscillation detection of the generator set to obtain an oscillation heatmap; adjustment of the generator set control decisions based on the oscillation heatmap; multiple mutation optimizations of the generator set control group based on generator set regulation adaptability evaluation conditions to obtain the optimal regulation result; and low-frequency oscillation suppression management of the generator set based on the optimal regulation result. This application solves the technical problem in existing technologies where oscillation suppression is difficult to balance with power supply tracking and equipment failure risk, leading to insufficient power system stability, power supply quality, and equipment safety. It achieves optimal low-frequency oscillation suppression management that balances oscillation suppression effectiveness, power supply adaptability, and generator set operational safety, ensuring the long-term stable operation of the target generator set.

Description

Technical Field

[0001] This invention relates to the field of generator technology, and more specifically to a method, system, and equipment for suppressing low-frequency oscillations in generator sets. Background Technology

[0002] In power systems, generator sets, as the core units of power generation, directly determine the grid frequency quality and power supply reliability through their operational stability. Low-frequency oscillations are a common stability problem during generator operation, often caused by dynamic imbalances between the generator's power angle, speed, excitation parameters, and grid load. Specifically, they manifest as periodic fluctuations in generator power, speed, or power angle. If not effectively suppressed in a timely manner, low-frequency oscillations can cause the grid frequency to deviate from its rated value, leading to unstable power supply voltage, abnormal shutdowns of user-side electrical equipment, and in severe cases, even generator loss of synchronization and grid disconnection, posing a significant threat to regional power supply security.

[0003] In summary, existing technologies suffer from the difficulty of balancing oscillation suppression with power supply tracking and equipment failure risk, resulting in insufficient power system stability, power supply quality, and equipment safety. Summary of the Invention

[0004] This application provides a method, system, and equipment for suppressing low-frequency oscillations in generator sets, which addresses the technical problem in the prior art where it is difficult to balance oscillation suppression with power supply tracking and equipment failure risk, resulting in insufficient power system stability, power supply quality, and equipment safety.

[0005] In view of the above problems, this application provides a method, system and equipment for suppressing low-frequency oscillations of generator sets.

[0006] A first aspect of this application provides a method for suppressing low-frequency oscillations in a generator set, the method comprising:

[0007] The system performs real-time monitoring of target generator units in the power system to obtain a generator unit monitoring sequence. Based on the monitoring sequence, it performs low-frequency oscillation detection on the target generator units to obtain an oscillation heatmap. Based on the oscillation heatmap, it adjusts the generator unit control decisions for the target generator units to obtain a first generator unit control and regulation group. Based on a low-frequency oscillation risk prediction model, it optimizes the first generator unit control and regulation group for low-frequency oscillation risk to obtain a second generator unit control and regulation group. Based on the load prediction characteristic data of the target generator units, it performs power supply tracking evaluation and optimization on the second generator unit control and regulation group to obtain a third generator unit control and regulation group. Based on the generator unit regulation adaptability evaluation conditions, it performs multiple variation optimizations on the third generator unit control and regulation group to obtain the generator unit regulation optimization result. Finally, it performs low-frequency oscillation suppression management on the target generator units based on the generator unit regulation optimization result.

[0008] A second aspect of this application provides a low-frequency oscillation suppression system for a generator set, the system comprising:

[0009] The system comprises the following modules: a monitoring module for real-time monitoring of target generator sets in the power system to obtain a monitoring sequence; a detection module for detecting low-frequency oscillations in the target generator sets based on the monitoring sequence to obtain an oscillation heatmap; a regulation module for adjusting the generator control decisions of the target generator sets based on the oscillation heatmap to obtain a first generator control regulation group; a first optimization module for optimizing the first generator control regulation group based on a low-frequency oscillation risk prediction model to obtain a second generator control regulation group; a second optimization module for optimizing the second generator control regulation group based on load prediction characteristic data of the target generator sets to obtain a third generator control regulation group; and a management module for performing multiple variation optimizations on the third generator control regulation group based on generator regulation adaptability evaluation conditions to obtain generator regulation optimization results, and performing low-frequency oscillation suppression management on the target generator sets based on the generator regulation optimization results.

[0010] A third aspect of this application provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the low-frequency oscillation suppression method for a generator set provided in this application when executing the executable instructions stored in the memory.

[0011] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0012] This application achieves optimal low-frequency oscillation suppression management that balances oscillation suppression effectiveness, power supply adaptability, and unit operation safety, thereby ensuring the long-term stable operation of the target generator set by real-time monitoring of the target generator set; detecting low-frequency oscillations of the target generator set based on the generator set monitoring sequence; adjusting the generator set control decisions based on the generator set oscillation heatmap; performing multiple mutation optimizations on the control group based on generator set adjustment adaptability evaluation conditions to obtain generator set adjustment optimization results; and implementing low-frequency oscillation suppression management of the target generator set based on the generator set adjustment optimization results. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1This is a schematic flowchart of a low-frequency oscillation suppression method for a generator set provided in an embodiment of this application.

[0015] Figure 2 This is a schematic diagram of the low-frequency oscillation suppression system for a generator set provided in an embodiment of this application.

[0016] Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application.

[0017] Explanation of reference numerals in the attached drawings: Bus 300, Receiver 301, Processor 302, Transmitter 303, Memory 304, Bus Interface 305, Monitoring Module 11, Detection Module 12, Adjustment Module 13, First Optimization Module 14, Second Optimization Module 15, Management Module 16. Detailed Implementation

[0018] This application provides a method, system, and equipment for suppressing low-frequency oscillations in generator sets. It addresses the technical problem in the prior art where it is difficult to balance oscillation suppression with power supply tracking and equipment failure risk, resulting in insufficient power system stability, power supply quality, and equipment safety. The application achieves optimal low-frequency oscillation suppression management that balances oscillation suppression effect, power supply adaptability, and generator set operation safety, thereby ensuring the long-term stable operation of the target generator set.

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0021] Example 1, as Figure 1 As shown, this application provides a method for suppressing low-frequency oscillations in a motor unit, the method comprising:

[0022] S100: Real-time monitoring of target generator sets in the power system to obtain generator set monitoring sequences.

[0023] Specifically, the first step is to monitor the target generator units of the power system in real time to obtain a monitoring sequence. Target generator units include large-capacity units responsible for supplying the main power to the region, units located at weak points in cross-regional power transmission, and units that have historically exhibited low-frequency oscillations or whose current power angle, speed, and other parameters are close to stable critical values. Subsequently, sensors are installed at key locations on the target generator units, and PMUs are configured to achieve high-frequency sampling. Simultaneously, GPS or BeiDou satellite positioning systems are used to synchronize data from each monitoring point. The collected raw data undergoes Kalman filtering or wavelet filtering to remove redundant information such as electromagnetic interference and signal transmission noise. If data loss occurs due to brief communication interruptions, it is supplemented using linear or polynomial interpolation. Finally, a monitoring sequence is constructed, which is a continuous time series with time as the horizontal axis and multi-dimensional operating parameters as the vertical axis. Each timestamp corresponds to a set of data values ​​for power angle, speed, power, excitation parameters, etc. Typically, the monitoring data for the most recent 5-10 minutes is retained in a real-time rolling format to ensure dynamic reflection of the latest operating status of the units.

[0024] S200: Perform low-frequency oscillation detection on the target generator set according to the unit monitoring sequence to obtain the unit oscillation thermal map.

[0025] Specifically, the target generator set is subjected to low-frequency oscillation detection based on the generator set monitoring sequence. Low-frequency oscillation detection refers to identifying whether the target generator set has an operating state that conforms to the characteristics of low-frequency oscillation by analyzing real-time monitoring data, and at the same time judging the intensity, dominant parameters and range of the oscillation. The time-continuous data of each parameter in the real-time acquired generator set monitoring sequence is input into the trained low-frequency oscillation tracing network. The network outputs the generator set oscillation characteristic detection result by comparing the feature differences between the real-time sequence and the historical oscillation path set. The result includes information such as whether oscillation exists, the dominant oscillation parameters, such as speed oscillation or power angle oscillation, oscillation frequency and amplitude, and the duration of oscillation. Finally, based on the unit oscillation characteristic detection results, trend prediction is performed. Combining the coupling relationship between parameters and the historical oscillation development pattern, the oscillation intensity change trend of each monitoring parameter in the near future is predicted, thereby generating a unit oscillation heat map. This heat map is a visual presentation carrier. The horizontal axis is the time series, which is consistent with the time dimension of the unit monitoring sequence. The vertical axis represents the key monitoring parameters of the target generator unit. The color depth intuitively reflects the oscillation intensity of each parameter at different time points. For example, dark red represents the oscillation intensity exceeding the standard, yellow represents the oscillation intensity critical, and green represents no oscillation. At the same time, the oscillation start time, the region corresponding to the dominant parameter, and the color change gradient of the trend prediction can be marked. For example, when it is predicted that the oscillation of a certain parameter will increase, the color of the corresponding time area transitions from yellow to red.

[0026] S300: Adjust the unit control decision of the target generator unit according to the unit oscillation thermogram to obtain the first unit control adjustment group.

[0027] Specifically, the process begins by extracting parameters from the unit's oscillation heatmap that indicate excessive oscillation intensity, clarifying the oscillation duration, peak periods, and the correlation between parameters. Subsequently, unit control decisions are initiated for adjustment. Based on the oscillation characteristics, the operating parameters and strategies of the target generator unit's control module are adjusted to weaken or block the development of low-frequency oscillations. Each adjustment strategy must define a specific range for control parameter adjustments and ensure that the adjusted values ​​are within the unit's safe operating threshold. After formulating multiple differentiated adjustment strategies, a simplified simulation model is used to apply a particular strategy to the target generator unit under the current oscillation state. The model is then used to determine whether the strategy can cause the oscillation intensity level of the corresponding parameter in the heatmap to change towards lower intensity. Finally, all candidate adjustment strategies that have passed preliminary effectiveness verification and target different oscillation characteristics are integrated to form the first unit control and adjustment group—a collection of multiple control and adjustment schemes with the potential to alleviate the current low-frequency oscillations, initially selected based on the unit's oscillation heatmap.

[0028] S400: Based on the low-frequency oscillation risk prediction model, the control and regulation group of the first unit is optimized for low-frequency oscillation risk to obtain the control and regulation group of the second unit.

[0029] Specifically, the low-frequency oscillation risk prediction model is based on historical low-frequency oscillation event data of the target generator unit and past implementation effect data of different control and regulation schemes. It is built using machine learning algorithms, such as gradient boosting trees and neural networks. After receiving a control and regulation scheme for a given unit, the model outputs the probability value that the target generator unit will still experience low-frequency oscillations after the scheme is implemented; this is the low-frequency oscillation risk coefficient. The closer the coefficient is to 0, the lower the risk; the closer it is to 1, the higher the risk. Subsequently, a low-frequency oscillation risk optimization process is initiated. This process aims to select regulation schemes whose low-frequency oscillation risk after implementation is below a safety threshold. Through model quantitative evaluation and threshold comparison, oscillation risk judgment results are formed: compliant (risk coefficient ≤ threshold), non-compliant (risk coefficient > threshold). Finally, based on the judgment results, the first unit control and regulation group is screened, eliminating all non-compliant high-risk schemes and retaining compliant low-risk schemes. These low-risk schemes are then integrated to obtain the second unit control and regulation group, which is a set of schemes deeply optimized based on the risk dimension of the first unit control and regulation group.

[0030] S500: Based on the load prediction characteristic data of the target generator set, the second generator set control and regulation group is evaluated and optimized to obtain the third generator set control and regulation group.

[0031] Specifically, firstly, the control and regulation scheme of the f-th unit is extracted one by one from the second unit control and regulation group, where f is a positive integer representing the verification of all schemes in the second group to ensure no omissions. Then, load forecasting feature data is introduced. This data, obtained through time series forecasting algorithms (such as ARIMA and LSTM), contains key features of the load over a future period, including load peaks, valleys, fluctuation frequency, and duration. Based on the load forecasting feature data, the parameters of the f-th unit control and regulation scheme are input into the unit's three-dimensional model to simulate the power supply operation process of the unit under this scheme in future load change scenarios, outputting the f-th unit control simulation data. Subsequently, a power supply tracking evaluation channel is established. The control simulation data of the f-th unit is input into this channel. The channel calculates and outputs the power supply tracking evaluation coefficient of the f-th unit based on the degree of matching between the simulation data and the load prediction characteristic data. The value of this coefficient is usually in the range of 0-1. The closer it is to 1, the stronger the power supply tracking capability of the scheme and the better it can adapt to future load changes. Then, the coefficient is compared with the preset power supply tracking evaluation threshold. If the coefficient is ≥ the threshold, it means that the control and regulation scheme of the f-th unit not only has low low-frequency oscillation risk, but also can effectively adapt to the power supply demand of future loads, and it is included in the candidate set. After completing the above evaluation process for all schemes in the second unit control and regulation group, all qualified candidate schemes are integrated to obtain the third unit control and regulation group. This group is a set of schemes formed by superimposing high power supply adaptability screening criteria on the low-risk basis of the second unit control and regulation group, which simultaneously meets the dual requirements of safe and oscillation-free risk and power supply matching load demand.

[0032] S600: Based on the unit regulation adaptability evaluation conditions, the control and regulation group of the third unit is subjected to multiple mutation optimizations to obtain the unit regulation optimization results, and the target generator unit is subjected to low-frequency oscillation suppression management based on the unit regulation optimization results.

[0033] Specifically, the evaluation criteria for unit regulation adaptability include key proportions of power supply tracking evaluation weight, low-frequency oscillation risk weight, and unit fault risk weight. These are then weighted to form a quantified unit regulation adaptability value; a higher value indicates better overall performance. The goal is to select and optimize the regulation scheme with the best overall performance. Multiple mutation optimizations are performed on the third unit control and regulation group to obtain the optimal unit regulation result. This mutation optimization is a cyclical process of fine-tuning parameters of the initially compliant scheme and repeatedly verifying and optimizing it to obtain a better scheme. The optimal unit regulation result is selected based on the principle of maximizing fitness, i.e., selecting the 1-2 schemes with the highest fitness values. This result represents the optimal regulation scheme obtained after the entire process optimization, simultaneously meeting the comprehensive requirements of low oscillation risk, high power supply adaptability, and low fault probability. Finally, low-frequency oscillation suppression management is implemented based on the specific parameters in the optimization result. Real-time parameter configuration is performed on the excitation system, speed regulation system, and other control modules of the target generator unit to ensure long-term stable operation of the target generator unit and effective suppression of low-frequency oscillations.

[0034] Furthermore, based on the unit monitoring sequence, low-frequency oscillation detection is performed on the target generator unit to obtain a unit oscillation thermal map, including:

[0035] Accident path tracing is performed based on the low-frequency oscillation event set of the target generator set to obtain a low-frequency oscillation tracing path set; a graph structure training is performed on the low-frequency oscillation tracing path set using a graph neural network to obtain a low-frequency oscillation tracing network; low-frequency oscillation detection is performed on the unit monitoring sequence based on the low-frequency oscillation tracing network to obtain unit oscillation feature detection results; trend prediction is performed based on the unit oscillation feature detection results to generate the unit oscillation heat map.

[0036] Specifically, the accident path tracing is first carried out based on the low-frequency oscillation event set of the target generator unit. This set integrates complete operating data of the unit when low-frequency oscillations occurred in the past, covering unit monitoring parameters such as power angle, speed, and power at the time of oscillation triggering, oscillation duration, peak intensity, inducing factors, and quelling measures. For each low-frequency oscillation event in the event set, the fluctuation sequence and transmission logic of each key parameter during the oscillation process are reversed, and finally a low-frequency oscillation tracing path set is formed. This set records the dynamic correlation and change law of each parameter under different oscillation scenarios with time sequence as the clue. Each path corresponds to a set of parameter coupling evolution characteristics over time, providing feature samples for model training.

[0037] Next, a graph neural network is introduced. This model can accurately capture the complex relationships between data and is suitable for analyzing the oscillation characteristics of multi-parameter coupling in generator units. The low-frequency oscillation tracing path set is trained with a graph structure. Key operating parameters of the target generator unit, such as power angle, speed, stator power, and excitation voltage, are defined as nodes of the graph. The influence relationships between parameters, such as the adjustment effect of speed change on power angle and the correlation effect of excitation voltage on power output, are defined as edges of the graph. The fluctuation amplitude and rate of change of each parameter in the tracing path set are used as attribute values ​​of nodes and edges. Through multiple rounds of iterative training, the graph neural network learns the inherent logic of abnormal changes in node attributes → adjustment of edge correlation strength → occurrence of low-frequency oscillation. Finally, a low-frequency oscillation tracing network is obtained. This network has the ability to identify oscillation characteristics based on real-time parameters and can quickly match the similarity between real-time data and historical oscillation paths.

[0038] Subsequently, the real-time collected unit monitoring sequence, i.e., the continuous time dimension of the operating data containing each parameter, is input into the trained low-frequency oscillation tracing network to start low-frequency oscillation detection. By comparing the feature differences between the real-time monitoring sequence and the historical oscillation path set, the network determines whether the target generator unit has periodic parameter fluctuations at low frequency (and identifies the core operation of the oscillation state). Then, the unit oscillation feature detection result is output, which includes key information such as whether low-frequency oscillation exists, the dominant oscillation parameter, such as power angle or speed, the current oscillation frequency and amplitude, the duration of the oscillation, and the stage of the oscillation, such as nascent / developing / stabilized.

[0039] Finally, trend prediction is performed based on the unit oscillation characteristic detection results. This involves combining historical oscillation development patterns with current oscillation intensity, dominant parameters, and other characteristics, and using time series prediction algorithms to infer the oscillation trend of each parameter within the next 5-10 minutes. For example, it determines whether the oscillation will intensify, weaken, or remain. Based on the prediction results, a unit oscillation heat map is generated. This heat map is a visual chart with time as the horizontal axis, consistent with the time dimension of the unit monitoring sequence, and key operating parameters as the vertical axis. The color depth intuitively reflects the oscillation intensity of each parameter at different time points: dark red represents oscillation exceeding the standard, yellow represents the critical point, and green represents normal. At the same time, the oscillation starting point, high-risk areas of dominant parameters, and future trend color gradients are marked, clearly presenting the spatiotemporal distribution characteristics of the oscillation.

[0040] Furthermore, based on the low-frequency oscillation risk prediction model, the control and regulation group of the first unit is optimized for low-frequency oscillation risk to obtain the control and regulation group of the second unit, including:

[0041] Each unit control and regulation scheme within the first unit control and regulation group is input into the low-frequency oscillation risk prediction model to obtain multiple low-frequency oscillation risk coefficients; it is determined whether the multiple low-frequency oscillation risk coefficients are less than or equal to the low-frequency oscillation risk threshold to obtain multiple oscillation risk judgment results; based on the multiple oscillation risk judgment results, the first unit control and regulation group is optimized for low-frequency oscillation risk to generate the second unit control and regulation group.

[0042] Specifically, the process begins by addressing the control and regulation schemes for each unit within the first unit control and regulation group. These schemes include specific operational instructions such as adjusting the excitation system parameters of the target generator unit and optimizing the turbine speed regulation system parameters. Each unit control and regulation scheme is decomposed into a parameter vector recognizable by the model and then input into a low-frequency oscillation risk prediction model. This model, trained based on historical low-frequency oscillation event data of the target generator unit, past risk records of different regulation schemes, and power system safety operation thresholds, possesses the ability to quantitatively assess the risk of regulation schemes. It analyzes the impact of the regulation scheme on key parameters such as the unit's power angle and speed, outputting the probability value of low-frequency oscillations occurring after the scheme's implementation. Through feature matching and parameter influence calculation between the scheme and historical risk cases, the model ultimately outputs a corresponding low-frequency oscillation risk coefficient for each scheme. This coefficient quantifies the risk of low-frequency oscillations occurring after the scheme's implementation, typically ranging from 0 to 1. A coefficient closer to 0 indicates a lower risk, while a coefficient closer to 1 indicates a higher risk. For example, a coefficient of 0.1 represents a 10% probability of low-frequency oscillations occurring after implementation, while a coefficient of 0.3 represents a 30% probability.

[0043] Subsequently, an oscillation risk assessment is conducted. First, based on the oscillation tolerance limits of the target generator unit, such as the unit's power angle stability critical value, allowable speed fluctuation range, power industry safety operation standards, and historical safe operation data of the unit, a low-frequency oscillation risk threshold is set. For example, a value of 0.2 is taken based on the actual situation of the unit, meaning that schemes with a risk coefficient ≤ 0.2 are judged to meet the safety operation requirements. Then, the low-frequency oscillation risk coefficient of each unit's control and regulation scheme is compared with this threshold one by one to obtain the oscillation risk assessment result. The result is divided into only two categories: one is compliant, that is, the risk coefficient ≤ the threshold, which means that the low-frequency oscillation risk is within a controllable and safe range after the scheme is implemented; the other is non-compliant, that is, the risk coefficient > the threshold, which means that there is a high probability of low-frequency oscillation being triggered or remaining after the scheme is implemented, and it needs to be excluded from the candidate set.

[0044] Finally, low-frequency oscillation risk optimization is performed, aiming to eliminate high-risk schemes and retain low-risk schemes. A standardized process is used to precisely filter the initial control scheme cluster, ensuring the safety of the retained schemes. Based on all oscillation risk assessment results, substandard high-risk schemes are completely eliminated from the first unit control control group, retaining only compliant low-risk schemes. These low-risk schemes are then categorized and integrated according to control parameter type to form the second unit control control group. This group is a set of schemes from the first unit control control group after in-depth optimization based on risk dimensions. All schemes in this group have passed risk threshold verification and will not trigger new low-frequency oscillation risks after implementation.

[0045] Furthermore, based on the load prediction characteristic data of the target generator set, the second unit control and regulation group is evaluated and optimized for power supply tracking to obtain a third unit control and regulation group, including:

[0046] According to the second unit control and regulation group, the control and regulation scheme of the f-th unit is extracted, where f is a positive integer; a three-dimensional model of the target generator unit is constructed; based on the load prediction feature data, the three-dimensional model of the unit is simulated and controlled according to the control and regulation scheme of the f-th unit to obtain the control simulation data of the f-th unit; a power supply tracking evaluation channel is established according to the power supply tracking evaluation record set; the control simulation data of the f-th unit is input into the power supply tracking evaluation channel to obtain the power supply tracking evaluation coefficient of the f-th unit; if the power supply tracking evaluation coefficient of the f-th unit is greater than or equal to the power supply tracking evaluation threshold, the control and regulation scheme of the f-th unit is added to the third unit control and regulation group.

[0047] Specifically, the control and regulation scheme for unit f is first extracted sequentially from the second unit control and regulation group, where f is a positive integer representing the processing of each scheme within the group. For example, f=1 corresponds to the first scheme, f=2 corresponds to the second scheme, and each unit f control and regulation scheme contains specific control parameter adjustment instructions for the target generator unit. These include quantitative parameters that can directly drive the unit's operation, such as the excitation system voltage adjustment range, the turbine valve opening adjustment rate, and the reactive power compensation device's operational capacity. The target generator unit here refers to the core unit in the power system that plays a crucial role in frequency stability. This includes large-capacity units undertaking regional backbone power supply tasks, units located at weak points in inter-regional power transmission, or units with a history of low-frequency oscillations. Their operating status directly affects the overall system stability and is the object of power supply tracking and evaluation. Next, a three-dimensional model of the target generator set is constructed. This model is based on the actual physical structure of the target generator set, such as the rotor, stator winding, excitation control cabinet, turbine regulation system, and equipment parameters, such as rated power, speed range, excitation peak value multiple, and operating logic, such as the matching relationship between power output and load, and the influence of regulation parameters on power supply characteristics. The digital model constructed through three-dimensional modeling technology and simulation algorithms can accurately map the actual operating state of the unit under different control parameters and load conditions.

[0048] Subsequently, based on preset load prediction characteristic data, which reflects the load change trend of the target generator set over a future period, such as load peak, valley, and fluctuation frequency, the specific control parameters in the control and regulation scheme of the f-th unit are input into the three-dimensional model of the unit. The three-dimensional model of the unit is simulated by controlling the simulation, that is, the power supply operation process of the unit is simulated according to the scheme parameters. For example, the excitation system output is simulated according to the excitation voltage adjustment value, and the turbine power input is simulated according to the valve opening adjustment rate. At the same time, the impact of actual load changes on the unit is simulated by combining the load prediction characteristic data. Finally, the control simulation data of the f-th unit is output. This data is the key power supply indicators of the unit recorded during the simulation operation, covering the stability of the power supply voltage, frequency deviation, active power output fluctuation amplitude, reactive power compensation accuracy, etc., which directly reflects the power supply performance of the f-th scheme under future load scenarios.

[0049] Subsequently, a power supply tracking evaluation channel is established based on the power supply tracking evaluation record set. This record set integrates implementation data of different control and regulation schemes during the historical operation of the target generator unit, corresponding load scenario data, and power supply tracking effect data, such as the matching degree between power supply parameters and load demand, and user-side power supply quality feedback datasets. The power supply tracking evaluation channel can receive generator control simulation data and output quantitative evaluation results of power supply tracking capability.

[0050] After the control simulation data of the f-th unit is input into the power supply tracking evaluation channel, the channel will calculate and output the power supply tracking evaluation coefficient of the f-th unit based on the degree of matching between the simulation data and the load prediction characteristic data. The value of this coefficient is usually in the range of 0-1. The closer the coefficient is to 1, the stronger the power supply tracking capability of the f-th scheme and the better it can adapt to future load changes. The closer the coefficient is to 0, the weaker the tracking capability, and there may be a problem of power supply and load demand being out of sync.

[0051] Subsequently, the power supply tracking evaluation coefficient of the f-th unit is compared with the preset power supply tracking evaluation threshold. This threshold is set according to the power system power supply quality standard. For example, a value of 0.8 means that a coefficient ≥ 0.8 is considered to meet the power supply tracking capability standard. If the coefficient meets the condition of ≥ threshold, it means that the control and regulation scheme of the f-th unit not only has low low-frequency oscillation risk, but can also effectively adapt to the power supply demand of future loads. It is then added to the third unit control and regulation group. Finally, all compliant schemes are integrated, that is, the schemes corresponding to all positive integers of f are taken to form the third unit control and regulation group. This group is a set of schemes with high power supply adaptability screening standards superimposed on the low-risk basis of the second unit control and regulation group, which simultaneously meets the dual requirements of safety and practicality.

[0052] Furthermore, based on the power supply tracking and evaluation record set, a power supply tracking and evaluation channel is established, including:

[0053] Multiple learners are cross-supervised and trained based on the power supply tracking evaluation record set to obtain multiple power supply tracking evaluation models; the multiple power supply tracking evaluation models are federated and aggregated to obtain a power supply tracking evaluator; the power supply tracking evaluation record set is perturbed and injected using an adversarial example generator to obtain a power supply tracking evaluation perturbation set; the power supply tracking evaluator is robustly validated and trained based on the power supply tracking evaluation perturbation set to obtain a power supply tracking evaluation channel.

[0054] Specifically, the first step is to use a power supply tracking evaluation record set as a foundation. This record set integrates the implementation parameters of different unit control and regulation schemes during the historical operation of the target generator units, the load characteristic data of the corresponding time periods, and the power supply tracking effect data, which is the key basis for training the evaluation model. Based on this record set, cross-supervised training is conducted on multiple learners. A learner refers to an algorithm module with data learning and feature fitting capabilities. Here, differentiated algorithms such as decision trees, random forests, and neural networks are selected to cover the learning needs of different types of data relationships and avoid the limitations of a single algorithm. Cross-supervised training involves randomly dividing the power supply tracking evaluation record set into a training set at an 8:2 ratio. This training set is used for the learners to initially learn the mapping relationship between regulation schemes, loads, and power supply effects, and serves as a validation set to generate supervision signals and verify the learner's accuracy. The training set is input into each learner (e.g., a decision tree), and through algorithmic iteration, the learner learns the correlation between parameters such as excitation voltage and valve opening in the regulation scheme and load peak and fluctuation frequency, as well as the impact of these correlations on the power supply tracking effect, thus initially generating a supervised learning model with basic evaluation capabilities. The validation set is then input into all supervised learning models that have completed the first round of training. Each model outputs an evaluation value of the power supply tracking performance for each data point in the validation set, such as a power supply matching score. The deviation between any two model evaluation results is then calculated. If the deviation exceeds a preset threshold, the deviation data and the majority consensus evaluation value are marked as supervised samples. The majority consensus evaluation value serves as the supervision signal, representing a more reliable evaluation result for that data point. For each supervised learning model, the marked supervised samples are re-inputted into the model, and the key parameters of the model are adjusted in reverse, aiming to make the model output value approach the supervision signal. After each iteration, the evaluation deviation of all learners on the validation set is recalculated. If the overall average deviation drops below a preset accuracy threshold, training stops. If it does not reach the threshold, the training and validation sets are re-split to ensure data randomness, and the above steps are repeated until the deviation between the evaluation result of each learner and the majority consensus is consistently below the threshold. After training, each learner becomes a power supply tracking evaluation model.

[0055] Subsequently, multiple power supply tracking evaluation models are federated and aggregated. First, the evaluation accuracy of each model on the validation set is calculated, and weights are allocated according to the accuracy ratio. Then, the output results of multiple models for the same input data are merged into a single evaluation value through a weighted summation formula, thereby obtaining a power supply tracking evaluator. This evaluator combines the advantages of different algorithms, such as the nonlinear fitting ability of neural networks and the interpretability of decision trees, resulting in more accurate and stable evaluation results.

[0056] Then, an adversarial example generator is introduced. This generator needs to classify three types of perturbations: First, sensor error perturbation, simulating measurement deviations of speed and power sensors. For continuous numerical parameters, a Gaussian perturbation algorithm is used to inject a small deviation near the mean of the original data, while ensuring that the perturbed value does not exceed the safety boundary. Second, load prediction deviation perturbation, simulating errors in the load prediction algorithm. For load characteristic data, interval perturbation is used, injecting a deviation based on the original load peak value. The deviation is set according to the regional load prediction accuracy, and it must ensure that the load change trend after perturbation is consistent with the original time series pattern to avoid disrupting the time series correlation of the load data. Third, signal transmission fluctuation perturbation, simulating the data transmission process... For minor fluctuations, an absolute deviation perturbation is applied to the power supply performance data. A fixed deviation is injected into the original power matching rate to ensure that the perturbation amplitude meets the noise level of signal transmission. Subsequently, perturbation injection is performed on the power supply tracking evaluation record set. This involves superimposing the minor perturbations generated by the adversarial sample generator onto the original data of the record set. This process removes samples where the data exceeds the safety boundary after perturbation, samples where the perturbation causes logical inconsistencies in the data, and samples where the perturbation amplitude is within the preset range and conforms to physical logic. Finally, a power supply tracking evaluation perturbation set is formed. This set contains the original data and the corresponding perturbation data. Each sample can simulate a real interference scenario and can be used to test the evaluation stability of the power supply tracking evaluator when there are minor errors in the data.

[0057] Finally, robustness verification training of the power supply tracking evaluator is carried out based on the power supply tracking evaluation disturbance set. First, the disturbance set is input into the evaluator, and the deviation between its output evaluation value and the original evaluation result of the record set is calculated. If the deviation exceeds the threshold, the weight allocation rules or model parameters of the evaluator are adjusted. For example, if the load data after the disturbance causes the evaluation deviation to be too large, the model weight sensitive to load characteristic data is increased. This disturbance test-parameter adjustment process is iterated repeatedly until the evaluation deviation of the evaluator on the disturbance set is stably lower than the threshold. Finally, the power supply tracking evaluation channel is obtained. This channel can stably receive unit control simulation data and output accurate and interference-resistant power supply tracking evaluation coefficients.

[0058] Furthermore, based on the unit regulation adaptability evaluation conditions, the control and regulation group of the third unit is subjected to multiple variation optimizations to obtain the unit regulation optimization results, including:

[0059] The third unit control and regulation group is used to predict the fault risk of the target generator unit to obtain a unit fault risk set. Based on the unit fault risk set, the third unit control and regulation group is optimized and screened according to the unit fault risk threshold to obtain a fourth unit control and regulation group. The fitness of the fourth unit control and regulation group is calculated according to the unit regulation fitness evaluation conditions to obtain a unit regulation fitness set. Based on the unit regulation fitness set, the fourth unit control and regulation group is filtered according to the unit regulation fitness threshold to obtain a first domain of unit regulation optimization. The fourth unit control and regulation group is subjected to multiple mutation optimizations according to the first domain of unit regulation optimization to obtain multiple unit regulation optimization domains. The unit regulation fitness maximization optimization is performed according to the multiple unit regulation optimization domains to generate the unit regulation optimization result.

[0060] Specifically, firstly, based on the historical fault database of the target generator set, which includes past mechanical faults caused by improper adjustment parameters, such as bearing wear and excessive turbine vibration, as well as electrical faults, such as winding overheating and excitation system faults, and combined with the specific control parameters of each scheme within the third unit control and regulation group, fault risk prediction is carried out through fault tree analysis or machine learning prediction models, such as logistic regression and random forest. That is, the probability of equipment failure of the target generator set within a preset period after the implementation of each regulation scheme is quantitatively assessed. The fault risk probabilities corresponding to all schemes are integrated to form a unit fault risk set. Each element in this set corresponds one-to-one with the regulation scheme in the third group, and the closer the value is to 0, the lower the fault risk.

[0061] Subsequently, based on the unit failure risk set and the preset unit failure risk threshold, optimization screening is carried out to eliminate schemes whose probability value in the unit failure risk set exceeds the threshold, and only schemes that meet the risk standards are retained. These compliant schemes are integrated to form the fourth unit control and regulation group. The fourth unit control and regulation group can ensure that the subsequent optimized schemes have low oscillation risk and basic power supply adaptability, while further avoiding potential equipment failure hazards.

[0062] Next, based on the unit's adaptability evaluation criteria, which include power supply tracking evaluation weights: Low-frequency oscillation risk weight: Weighting of unit failure risk: The three weights need to be set in conjunction with the characteristics of the target generator set, and their sum should be 1, in order to balance power supply adaptation, oscillation safety, and equipment safety. Fitness calculations are performed for each scheme within the control and regulation group of the fourth unit, specifically:

[0063] From the control and regulation group of the fourth unit, a specific scheme is selected, and three types of core basic data for that scheme are extracted: power supply tracking evaluation coefficient: The value ranges from 0 to 1, with higher coefficients indicating stronger adaptability; Low-frequency oscillation risk coefficient: The value ranges from 0 to 1, with lower coefficients indicating lower risk; Unit failure risk probability: The value ranges from 0 to 1, with lower probabilities indicating stronger safety. Since the low-frequency oscillation risk coefficient and the probability of unit failure are negative indicators, lower values ​​indicate better performance. Therefore, it needs to be converted to a positive indicator to avoid logical conflicts in calculations. The conversion formula is 1 - the negative indicator value, which is then substituted into the weighted summation fitness calculation formula:

[0064] Unit adjustment adaptability value = (1 - )× +(1- )× +(1- )×

[0065] Following this logic, the fitness value is calculated for each scheme in the control and regulation group of the fourth unit, and the correspondence between scheme ID and fitness value is finally organized into a structured set, which forms the unit regulation fitness set.

[0066] Based on the unit regulation fitness set, a unit regulation fitness threshold filter is adopted. That is, according to the system's requirements for the unit's comprehensive performance, the comprehensive performance of the schemes below the threshold does not meet the dual standards of safety and practicality. Schemes with fitness values ​​below the threshold are eliminated, and schemes that meet the standards are retained, forming the first domain of unit regulation optimization. This domain is the core subset of schemes in the fourth unit control and regulation group that have initially met the comprehensive performance standards, providing high-quality basic samples for subsequent in-depth optimization.

[0067] Then, based on the scheme of the first domain of unit regulation optimization, the key control parameters are slightly varied to generate variant schemes. The variant schemes are then sequentially input into the low-frequency oscillation risk prediction model to verify whether the oscillation risk still meets the standard, the fault risk prediction model to verify whether the fault risk still meets the standard, and the power supply tracking evaluation channel to verify whether the power supply adaptability still meets the standard. Variant schemes that still meet all standards are selected to form a new unit regulation optimization domain. This process of parameter variation-multi-dimensional verification-screening of compliant schemes is repeated until the preset number of variations is reached, finally resulting in multiple unit regulation optimization domains to ensure coverage of more potential high-quality schemes.

[0068] Finally, the fitness of unit regulation is maximized for all schemes in multiple unit regulation optimization domains. That is, the fitness value of each scheme is extracted, and the 1-2 schemes with the highest fitness values ​​are selected by comparison. If there are multiple optimal schemes with the same fitness value, the scheme with lower energy consumption can be selected by combining the unit operation energy consumption cost. This scheme is determined as the unit regulation optimization result. This result is the optimal regulation scheme after the whole process optimization, which simultaneously meets the requirements of low low-frequency oscillation risk, high power supply adaptability, low equipment failure risk, and high comprehensive fitness. It can be directly used for subsequent unit control.

[0069] Furthermore, based on the first domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to multiple variation optimizations to obtain multiple unit regulation optimization domains, including:

[0070] Based on the first domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to difference detection to obtain a set of unit regulation difference features. Then, based on the set of unit regulation difference features, the control and regulation group of the fourth unit is subjected to variational regulation to obtain a first group of unit regulation variations. Based on the low-frequency oscillation risk prediction model, the first group of unit regulation variations is subjected to low-frequency oscillation risk optimization to obtain a second group of unit regulation variations. Based on the power supply tracking evaluation channel, the second group of unit regulation variations is subjected to power supply tracking evaluation optimization to obtain a third group of unit regulation variations. Based on the unit fault risk threshold, the third group of unit regulation variations is subjected to fault risk optimization to obtain a fourth group of unit regulation variations. Based on the unit regulation fitness evaluation conditions, the fourth group of unit regulation variations is subjected to fitness evaluation optimization to obtain a second domain of unit regulation optimization. Based on the second domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to variational optimization until multiple unit regulation optimization domains that satisfy a predetermined number of variational optimizations are obtained.

[0071] Specifically, firstly, taking the first domain of unit regulation optimization as the benchmark, difference detection is carried out on the control and regulation group of the fourth unit. That is, through parameter comparison algorithm, the numerical differences, range deviations and coupling relationship differences of each scheme in the fourth group and the scheme in the first domain of optimization are identified in terms of key control parameters. The range difference is the detection object. These differences are integrated in the form of structured data to form a set of unit regulation difference features. The set contains information such as parameter type, difference value, difference range, and influence weight.

[0072] Subsequently, based on the unit regulation difference feature set, variation regulation is implemented on the fourth unit control and regulation group. According to the parameter difference direction and safety range in the difference feature set, the key control parameters of the schemes in the fourth group are slightly and directionally adjusted. This operation generates the first group of unit regulation variation. This group is the initial set of the fourth unit control and regulation group after parameter variation. Only parameter fine-tuning has been completed. It has not yet undergone multi-dimensional screening. There may be some schemes that do not meet safety or performance requirements.

[0073] Subsequently, the previously constructed low-frequency oscillation risk prediction model is invoked to optimize the low-frequency oscillation risk. This process involves inputting the parameters of each scheme within the first group of unit regulation variations into the model, calculating the low-frequency oscillation risk coefficient after the implementation of each scheme, eliminating schemes with coefficients exceeding the preset risk threshold, and retaining schemes that meet the risk standards to form the second group of unit regulation variations. This group ensures that the mutated schemes still have low-frequency oscillation risk and avoids the introduction of new oscillation hazards due to parameter variations.

[0074] Next, relying on the established power supply tracking evaluation channel, the power supply tracking evaluation optimization is carried out. The parameters of the schemes in the second group of unit regulation variation are input into the three-dimensional model of the unit. Combined with the load prediction characteristic data, simulated control data is generated. Then, the simulated data is input into the power supply tracking evaluation channel to calculate the power supply tracking evaluation coefficient of each scheme. Schemes with coefficients ≥ the power supply tracking evaluation threshold are selected to obtain the third group of unit regulation variation. This group ensures that the modified schemes can still adapt to future load changes and maintain good power supply tracking capability.

[0075] Subsequently, based on the preset unit failure risk threshold, failure risk optimization is carried out. The failure risk prediction model is used to evaluate the equipment failure probability after the implementation of each scheme in the third group of unit regulation variations. Schemes with a probability exceeding the failure risk threshold are eliminated, and schemes with failure risk meeting the standard are retained to form the fourth group of unit regulation variations, which takes into account both performance and equipment safety.

[0076] Finally, based on the unit regulation fitness evaluation conditions, fitness evaluation and optimization are carried out. The comprehensive fitness value of each scheme in the fourth group of unit regulation variation is calculated, and the scheme with a value ≥ fitness threshold is selected and integrated to form the second domain of unit regulation optimization. This domain is a subset of schemes that meet the comprehensive performance standard after variation. Compared with the first domain of unit regulation optimization, the parameters are more in line with the current optimization requirements and the comprehensive performance is better.

[0077] Finally, using the second domain of unit regulation optimization as a new benchmark, the complete process of difference detection → variation regulation → low-frequency oscillation risk optimization → power supply tracking evaluation optimization → fault risk optimization → fitness evaluation optimization is repeated to continuously perform variation optimization on the fourth unit control and regulation group. Each variation is based on the difference between the previous optimization domain and the fourth group, ensuring that the parameter adjustment direction is always toward improving comprehensive performance. This process continues until the predetermined number of variation optimizations is completed. The unit regulation optimization domains obtained in each cycle are then integrated, such as optimizing the first domain, the second domain, and the third domain, to finally obtain multiple unit regulation optimization domains that meet the predetermined number of variation optimizations. This set contains a subset of schemes that meet the criteria after multiple rounds of optimization, covering high-quality schemes under different variation directions, providing sufficient samples for subsequent fitness maximization optimization.

[0078] Furthermore, the unit regulation adaptability evaluation conditions include power supply tracking evaluation weight, low-frequency oscillation risk weight, and unit fault risk weight.

[0079] Specifically, the evaluation conditions for unit regulation adaptability include power supply tracking evaluation weight, low-frequency oscillation risk weight, and unit failure risk weight. The power supply tracking evaluation weight is a quantitative indicator that measures the importance of the regulation scheme's ability to adapt to future load changes of the target generator unit in the comprehensive evaluation. Its setting must be linked to the practical goal of safe power supply in the power system: if the target generator unit's area experiences frequent load fluctuations, this weight needs to be appropriately increased to avoid the scheme only suppressing oscillations but failing to match load demand. The low-frequency oscillation risk weight is a quantitative indicator representing the priority of the regulation scheme in avoiding low-frequency oscillation risks. Since low-frequency oscillation suppression is a core objective of the scheme, this weight must maintain the highest proportion regardless of the unit's operating scenario to prevent neglecting the fundamental requirement of oscillation prevention due to excessive emphasis on other dimensions. The unit failure risk weight is a quantitative indicator reflecting the importance of the regulation scheme in preventing equipment failures. It is mainly related to the equipment status of the target generator unit. For example, if the unit has been in service for many years or core components have historical failure records, the weight needs to be increased to avoid equipment damage such as bearing wear and winding overheating due to improper regulation parameters.

[0080] In the specific setup, the core characteristic data of the target generator set, the threshold requirements for low-frequency oscillation risk in the safety operation standards of the power system to which it belongs, the allowable range of power supply quality deviation, and typical operating scenario data are first collected. These data serve as the objective basis for weight setting, avoiding weight imbalance caused by subjective experience. Subsequently, the weights are quantified by combining the analytic hierarchy process (AHP) with data verification. First, the target layer is constructed using AHP as the unit's regulation adaptability, and the criterion layer is the evaluation dimensions corresponding to the three types of weights. Experts in power system dispatching, unit operation and maintenance, and equipment manufacturing are invited to score the relative importance of the three types of dimensions to initially determine the weight allocation. Then, historical high-quality regulation schemes of the target generator set over the past three years are retrieved, i.e., schemes that effectively suppress oscillations, adapt to the load, and do not cause equipment failures after implementation. The power supply tracking evaluation coefficient, low-frequency oscillation risk coefficient, and unit failure risk probability of these schemes are substituted into the initial weights to calculate the adaptability value. If the matching degree between the calculated result and the actual scheme effect is lower than the preset threshold, the weights are adjusted in reverse. This process is repeated until the matching degree between the weights and the actual effect meets the requirements, and finally, the specific values ​​of the three types of weights are determined.

[0081] In summary, the embodiments of this application have at least the following technical effects:

[0082] This application achieves optimal low-frequency oscillation suppression management that balances oscillation suppression effectiveness, power supply adaptability, and unit operation safety, thereby ensuring the long-term stable operation of the target generator set by real-time monitoring of the target generator set; detecting low-frequency oscillations of the target generator set based on the generator set monitoring sequence; adjusting the generator set control decisions based on the generator set oscillation heatmap; performing multiple mutation optimizations on the generator set control adjustment group based on the generator set adjustment adaptability evaluation conditions to obtain the generator set adjustment optimization results; and implementing low-frequency oscillation suppression management of the target generator set based on the generator set adjustment optimization results.

[0083] Example 2, based on the same inventive concept as the low-frequency oscillation suppression method for the motor set in the previous examples, such as... Figure 2 As shown, this application provides a low-frequency oscillation suppression system for motor sets. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0084] The monitoring module 11 is used to monitor the target generator set of the power system in real time and obtain the generator set monitoring sequence; the detection module 12 is used to detect low-frequency oscillations of the target generator set according to the generator set monitoring sequence and obtain the generator set oscillation heat map; the adjustment module 13 is used to adjust the generator set control decision of the target generator set according to the generator set oscillation heat map and obtain a first generator set control adjustment group; the first optimization module 14 is used to optimize the first generator set control adjustment group for low-frequency oscillation risk according to the low-frequency oscillation risk prediction model and obtain a second generator set control adjustment group; the second optimization module 15 is used to optimize the second generator set control adjustment group for power supply tracking evaluation according to the load prediction characteristic data of the target generator set and obtain a third generator set control adjustment group; the management module 16 is used to perform multiple variation optimizations on the third generator set control adjustment group according to the generator set adjustment adaptability evaluation conditions, obtain the generator set adjustment optimization result, and perform low-frequency oscillation suppression management on the target generator set according to the generator set adjustment optimization result.

[0085] Furthermore, the detection module 12 in the low-frequency oscillation suppression system of the motor set is also used for:

[0086] Accident path tracing is performed based on the low-frequency oscillation event set of the target generator set to obtain a low-frequency oscillation tracing path set; a graph structure training is performed on the low-frequency oscillation tracing path set using a graph neural network to obtain a low-frequency oscillation tracing network; low-frequency oscillation detection is performed on the unit monitoring sequence based on the low-frequency oscillation tracing network to obtain unit oscillation feature detection results; trend prediction is performed based on the unit oscillation feature detection results to generate the unit oscillation heat map.

[0087] Furthermore, the first optimization module 14 in the low-frequency oscillation suppression system of the motor set is also used for:

[0088] Each unit control and regulation scheme within the first unit control and regulation group is input into the low-frequency oscillation risk prediction model to obtain multiple low-frequency oscillation risk coefficients; it is determined whether the multiple low-frequency oscillation risk coefficients are less than or equal to the low-frequency oscillation risk threshold to obtain multiple oscillation risk judgment results; based on the multiple oscillation risk judgment results, the first unit control and regulation group is optimized for low-frequency oscillation risk to generate the second unit control and regulation group.

[0089] Furthermore, the second optimization module 15 in the low-frequency oscillation suppression system of the motor set is also used for:

[0090] According to the second unit control and regulation group, the control and regulation scheme of the f-th unit is extracted, where f is a positive integer; a three-dimensional model of the target generator unit is constructed; based on the load prediction feature data, the three-dimensional model of the unit is simulated and controlled according to the control and regulation scheme of the f-th unit to obtain the control simulation data of the f-th unit; a power supply tracking evaluation channel is established according to the power supply tracking evaluation record set; the control simulation data of the f-th unit is input into the power supply tracking evaluation channel to obtain the power supply tracking evaluation coefficient of the f-th unit; if the power supply tracking evaluation coefficient of the f-th unit is greater than or equal to the power supply tracking evaluation threshold, the control and regulation scheme of the f-th unit is added to the third unit control and regulation group.

[0091] Furthermore, the second optimization module 15 in the low-frequency oscillation suppression system of the motor set is also used for:

[0092] Multiple learners are cross-supervised and trained based on the power supply tracking evaluation record set to obtain multiple power supply tracking evaluation models; the multiple power supply tracking evaluation models are federated and aggregated to obtain a power supply tracking evaluator; the power supply tracking evaluation record set is perturbed and injected using an adversarial example generator to obtain a power supply tracking evaluation perturbation set; the power supply tracking evaluator is robustly validated and trained based on the power supply tracking evaluation perturbation set to obtain a power supply tracking evaluation channel.

[0093] Furthermore, the management module 16 in the low-frequency oscillation suppression system of the motor set is also used for:

[0094] The third unit control and regulation group is used to predict the fault risk of the target generator unit to obtain a unit fault risk set. Based on the unit fault risk set, the third unit control and regulation group is optimized and screened according to the unit fault risk threshold to obtain a fourth unit control and regulation group. The fitness of the fourth unit control and regulation group is calculated according to the unit regulation fitness evaluation conditions to obtain a unit regulation fitness set. Based on the unit regulation fitness set, the fourth unit control and regulation group is filtered according to the unit regulation fitness threshold to obtain a first domain of unit regulation optimization. The fourth unit control and regulation group is subjected to multiple mutation optimizations according to the first domain of unit regulation optimization to obtain multiple unit regulation optimization domains. The unit regulation fitness maximization optimization is performed according to the multiple unit regulation optimization domains to generate the unit regulation optimization result.

[0095] Furthermore, the management module 16 in the low-frequency oscillation suppression system of the motor set is also used for:

[0096] Based on the first domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to difference detection to obtain a set of unit regulation difference features. Then, based on the set of unit regulation difference features, the control and regulation group of the fourth unit is subjected to variational regulation to obtain a first group of unit regulation variations. Based on the low-frequency oscillation risk prediction model, the first group of unit regulation variations is subjected to low-frequency oscillation risk optimization to obtain a second group of unit regulation variations. Based on the power supply tracking evaluation channel, the second group of unit regulation variations is subjected to power supply tracking evaluation optimization to obtain a third group of unit regulation variations. Based on the unit fault risk threshold, the third group of unit regulation variations is subjected to fault risk optimization to obtain a fourth group of unit regulation variations. Based on the unit regulation fitness evaluation conditions, the fourth group of unit regulation variations is subjected to fitness evaluation optimization to obtain a second domain of unit regulation optimization. Based on the second domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to variational optimization until multiple unit regulation optimization domains that satisfy a predetermined number of variational optimizations are obtained.

[0097] Furthermore, the unit adjustment adaptability evaluation conditions in the management module 16 of the low-frequency oscillation suppression system of the generator set include power supply tracking evaluation weight, low-frequency oscillation risk weight, and unit fault risk weight.

[0098] Example 3: Based on the inventive concept of the low-frequency oscillation suppression method for generator sets in the foregoing embodiments, this application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of any of the methods described in Example 1 above.

[0099] Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 3 In this document, the bus architecture is represented by bus 300. Bus 300 may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 during operation.

[0100] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0101] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0102] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for suppressing low-frequency oscillations in generator sets, characterized in that, The method includes: Real-time monitoring of target generator units in the power system to obtain unit monitoring sequences; The process of detecting low-frequency oscillations in the target generator unit based on the unit monitoring sequence to obtain a unit oscillation heatmap includes: tracing accident paths based on the low-frequency oscillation event set of the target generator unit to obtain a low-frequency oscillation tracing path set; training the low-frequency oscillation tracing path set with a graph neural network to obtain a low-frequency oscillation tracing network; detecting low-frequency oscillations in the unit monitoring sequence based on the low-frequency oscillation tracing network to obtain unit oscillation feature detection results; and performing trend prediction based on the unit oscillation feature detection results to generate the unit oscillation heatmap. The unit control decision of the target generator unit is adjusted according to the unit oscillation thermogram to obtain the first unit control adjustment group; Based on the low-frequency oscillation risk prediction model, the control and regulation group of the first unit is optimized for low-frequency oscillation risk to obtain the control and regulation group of the second unit. Based on the load prediction feature data of the target generator set, the second unit control and regulation group is evaluated and optimized for power supply tracking to obtain a third unit control and regulation group. This includes: extracting the control and regulation scheme for the f-th unit (f is a positive integer) from the second unit control and regulation group; constructing a three-dimensional model of the target generator set; simulating the control and regulation of the three-dimensional model based on the load prediction feature data and the f-th unit control and regulation scheme to obtain f-th unit control simulation data; and establishing a power supply tracking evaluation channel based on the power supply tracking evaluation record set, including: performing cross-supervised training on multiple learners based on the power supply tracking evaluation record set to obtain... Multiple power supply tracking evaluation models are obtained, and the multiple power supply tracking evaluation models are federated and aggregated to obtain a power supply tracking evaluator. The power supply tracking evaluation record set is perturbed and injected according to the adversarial example generator to obtain a power supply tracking evaluation perturbation set. The power supply tracking evaluator is robustly verified and trained according to the power supply tracking evaluation perturbation set to obtain a power supply tracking evaluation channel. The control simulation data of the f-th unit is input into the power supply tracking evaluation channel to obtain the f-th power supply tracking evaluation coefficient. If the f-th power supply tracking evaluation coefficient is greater than or equal to the power supply tracking evaluation threshold, the control and regulation scheme of the f-th unit is added to the third unit control and regulation group. Based on the unit regulation adaptability evaluation conditions, the control and regulation group of the third unit is subjected to multiple mutation optimizations to obtain the unit regulation optimization results, and the target generator unit is subjected to low-frequency oscillation suppression management based on the unit regulation optimization results.

2. The low-frequency oscillation suppression method for generator sets as described in claim 1, characterized in that, Based on the low-frequency oscillation risk prediction model, the control and regulation group of the first unit is optimized for low-frequency oscillation risk to obtain the control and regulation group of the second unit, including: Each control and regulation scheme of each unit in the first unit control and regulation group is input into the low-frequency oscillation risk prediction model to obtain multiple low-frequency oscillation risk coefficients. Determine whether the multiple low-frequency oscillation risk coefficients are less than or equal to the low-frequency oscillation risk threshold to obtain multiple oscillation risk judgment results; Based on the multiple oscillation risk assessment results, the first unit control and regulation group is optimized for low-frequency oscillation risk, and the second unit control and regulation group is generated.

3. The low-frequency oscillation suppression method for generator sets as described in claim 1, characterized in that, Based on the unit regulation adaptability evaluation conditions, the control and regulation group of the third unit is subjected to multiple variation optimizations to obtain the unit regulation optimization results, including: Based on the third unit control and regulation group, the target generator unit is predicted for fault risk, and a set of unit fault risks is obtained. Based on the unit failure risk set, the third unit control and regulation group is optimized and screened according to the unit failure risk threshold to obtain the fourth unit control and regulation group. Based on the unit regulation fitness evaluation conditions, the fitness of the control and regulation group of the fourth unit is calculated to obtain the unit regulation fitness set. Based on the unit regulation fitness set, the fourth unit control and regulation group is filtered according to the unit regulation fitness threshold to obtain the first domain of unit regulation optimization. Based on the first domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to multiple variation optimizations to obtain multiple unit regulation optimization domains. The unit regulation fitness is maximized based on the multiple unit regulation optimization domains to generate the unit regulation optimization result.

4. The low-frequency oscillation suppression method for generator sets as described in claim 3, characterized in that, Based on the first domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to multiple variation optimizations to obtain multiple unit regulation optimization domains, including: Based on the first domain of unit regulation optimization, the control and regulation group of the fourth unit is subjected to difference detection to obtain a set of unit regulation difference features. Based on the set of unit regulation difference features, the control and regulation group of the fourth unit is subjected to variation regulation to obtain a first group of unit regulation variation. Based on the low-frequency oscillation risk prediction model, the first group of unit regulation variations is optimized for low-frequency oscillation risk to obtain the second group of unit regulation variations. Based on the power supply tracking evaluation channel, the second group of unit regulation variations is evaluated for optimization to obtain the third group of unit regulation variations. Based on the unit failure risk threshold, the third group of unit regulation variations is optimized for failure risk to obtain the fourth group of unit regulation variations. The fourth group of unit regulation variations is evaluated and optimized based on the unit regulation fitness evaluation conditions to obtain the second domain of unit regulation optimization. Based on the second domain of unit regulation optimization, the fourth unit control and regulation group continues to perform variation optimization until the multiple unit regulation optimization domains that satisfy the predetermined number of variation optimizations are obtained.

5. The low-frequency oscillation suppression method for generator sets as described in claim 1, characterized in that, The evaluation criteria for unit regulation adaptability include power supply tracking evaluation weight, low-frequency oscillation risk weight, and unit failure risk weight.

6. A low-frequency oscillation suppression system for a generator set, characterized in that, The system is used to implement the low-frequency oscillation suppression method for a generator set according to any one of claims 1 to 5, and the system comprises: The monitoring module is used to monitor the target generator sets of the power system in real time and obtain the generator set monitoring sequence; The detection module is used to perform low-frequency oscillation detection on the target generator set according to the unit monitoring sequence, and obtain the unit oscillation thermal map; The adjustment module is used to adjust the unit control decision of the target generator unit according to the unit oscillation thermogram to obtain the first unit control adjustment group; The first optimization module is used to perform low-frequency oscillation risk optimization on the first unit control and regulation group according to the low-frequency oscillation risk prediction model, and obtain the second unit control and regulation group. The second optimization module is used to perform power supply tracking evaluation and optimization of the second unit control and regulation group based on the load prediction characteristic data of the target generator set, and to obtain the third unit control and regulation group. The management module is used to perform multiple variation optimizations on the control and regulation group of the third unit according to the unit regulation adaptability evaluation conditions, obtain the unit regulation optimization results, and perform low-frequency oscillation suppression management on the target generator unit according to the unit regulation optimization results.

7. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the low-frequency oscillation suppression method for the generator set according to any one of claims 1-5.

Images