Method and system for predicting primary frequency modulation power response performance of thermal power generating unit

Abstract

The invention relates to the technical field of thermal power generating unit state monitoring and performance prediction, and discloses a thermal power generating unit primary frequency modulation power response performance prediction method and system, and the method comprises the steps: monitoring a unit speed regulation execution mechanism physical position feedback signal, and calculating the baseline fluctuation characteristics of the unit speed regulation execution mechanism in a verification time period; judging whether a preset working condition stability condition is met or not; if yes, a time-varying virtual disturbance signal is injected into the speed regulation controller, response is monitored, and the current value of the health indicator is determined based on the corresponding relation between the disturbance signal and response change; repeating the steps according to a preset time period to obtain time sequence data of the health indicator; and adopting a time sequence prediction model to predict a future value of the health indicator based on the data so as to generate prediction information of the primary frequency modulation power response performance of the unit. According to the method, virtual disturbance is actively injected, so that the physical health indicator which is high in signal-to-noise ratio and can be used for reliable prediction can be obtained, and the accuracy and reliability of performance prediction are improved.

Description

A method and system for predicting the primary frequency regulation power response performance of thermal power units Technical Field

[0001] This invention relates to the field of thermal power unit condition monitoring and performance prediction technology, specifically to a method and system for predicting the primary frequency regulation power response performance of thermal power units. Background Technology

[0002] The current physical foundation ensuring this performance—the turbine's speed control system, servo mechanism, and regulating valves—is a complex system comprising precision mechanical and hydraulic components. During long-term operation, it inevitably experiences physical wear, aging, and parameter drift. This gradual performance degradation process is unavoidable. Therefore, accurately predicting this performance degradation trend and conducting predictive maintenance accordingly is a core technical requirement in this field. Currently, the assessment and diagnosis of this performance in this field heavily relies on high-energy disturbance events. One approach is to test the system through offline testing, artificially applying large step disturbances. While accurate, this method is costly, results in power generation loss, and poses a potential impact on the safe operation of the unit. Another approach relies on real frequency disturbance events occurring in the power grid, such as large-capacity units disconnecting from the grid, and then passively controlling the unit. Post-event analysis of response data is a lagging approach, capable of diagnosing problems that have already occurred but failing to meet the needs of prediction. To achieve online prediction, current research attempts to utilize the unavoidable minor frequency fluctuations in the daily operation of the power grid, using complex system identification algorithms such as deep learning models to fit the mathematical model of the system online, and using this as the basis for health status assessment. However, this passive micro-disturbance approach has a fundamental contradiction in engineering principles: the power system is in a stable state most of the time, at which point the energy of the micro-disturbance signal is extremely low, completely submerged in measurement noise and the random fluctuations of the unit itself. Relying on such low signal-to-noise ratio data for complex online model identification faces significant challenges in terms of the stability and reliability of the identification results in practical engineering. Based on this, the reliability of predictions is difficult to meet the requirements of high-reliability industrial applications.

[0003] To address the issue of low-quality passive identification data, some research in this field has shifted towards constructing high-precision mechanistic simulation models, attempting to indirectly evaluate performance through complex model parameter identification. For example, Chinese invention patent CN113031565B discloses a method and system for predicting the primary frequency regulation power response performance of thermal power units. This approach establishes a complex prediction model that includes multiple sub-models such as electro-hydraulic regulation, servo mechanism, turbine, and main steam pressure, and relies on collected operating data to identify model parameters. The fundamental flaw of this approach is that the accuracy of its predictions highly depends on the completeness of the constructed mechanistic model and the precision of parameter identification. However, thermal power units are high-dimensional, nonlinear, and time-varying systems. Their physical characteristics, such as the gradual degradation process of wear and aging, are difficult to accurately characterize using fixed mathematical models. This method attempts to fit the macroscopic response of the entire system rather than measuring the physical root causes of performance degradation, such as the physical dead zone or jamming of the speed regulating actuator. As a result, its prediction results are still susceptible to changes in operating conditions and model mismatch, and its reliability also faces challenges in engineering practice.

[0004] Therefore, the technical problem to be solved by this invention is how to actively and with a high signal-to-noise ratio acquire a simple health indicator that can directly characterize the physical health degradation state of the frequency modulation actuator without interfering with the normal operation of the unit, and to build a robust and reliable performance degradation prediction model based on this indicator. Summary of the Invention

[0005] In order to overcome the defects of the prior art, the purpose of this invention is to provide a method and system for predicting the primary frequency regulation power response performance of thermal power units, so as to solve the technical problem in the prior art that it is impossible to actively obtain a physical health indicator with a high signal-to-noise ratio that can be used for reliable prediction without interfering with the operation of the unit.

[0006] This invention is achieved through the following technical solution: Firstly, this invention provides a method for predicting the primary frequency regulation power response performance of a thermal power unit, comprising: Step 1, monitoring the physical position feedback signal of the unit's speed control actuator, calculating the baseline fluctuation characteristics of the signal during a verification period, and determining whether the baseline fluctuation characteristics meet preset operating condition stability conditions; Step 2, when the baseline fluctuation characteristics meet the operating condition stability conditions, injecting a time-varying virtual disturbance signal into the speed controller and monitoring the signal response, and determining the current value of a health indicator characterizing the physical response characteristics of the actuator based on the correspondence between the disturbance signal and the response change; Step 3, repeating steps 1-2 at a predetermined time period to obtain time series data of the health indicator; Step 4, using a time series prediction model to predict the future value of the health indicator based on the time series data of the health indicator; Step 5, generating prediction information for the unit's primary frequency regulation power response performance based on the future value of the health indicator.

[0007] Preferably, in step 1, when the baseline fluctuation feature is the variance value of the physical location feedback signal during the verification period, the preset operating condition stability condition is that the variance value is lower than a preset fluctuation threshold; when the baseline fluctuation feature is the peak-to-peak value of the physical location feedback signal during the verification period, the preset operating condition stability condition is that the peak-to-peak value is lower than a preset fluctuation threshold.

[0008] Preferably, in step 2, the health indicator is the physical dead zone width. The specific way to determine the current value of the health indicator is as follows: when the physical position feedback signal undergoes a preset response change, the amplitude of the virtual disturbance signal that triggers the response change is recorded, and the physical dead zone width is determined by the amplitude.

[0009] Furthermore, after the physical position feedback signal undergoes a preset response change, the signal is continuously monitored within a preset transient capture window, and a second health indicator characterizing the dynamic response characteristics of the actuator is determined based on the signal within the window; the time series data is dual time series data of the health indicator and the second health indicator, and a multivariate time series prediction model is used to jointly predict the dual time series data.

[0010] Furthermore, while determining the current value of the health indicator that characterizes the physical response characteristics of the actuator, the operating condition reference parameters that characterize the current operating conditions of the unit are simultaneously acquired.

[0011] Furthermore, based on the historical correspondence between the operating condition baseline parameters and the current value of the health indicator, a regression fitting method is used to establish an operating condition-health indicator baseline mapping model; based on this mapping model and the current operating condition baseline parameters, the baseline value corresponding to the current value of the health indicator is calculated; the residual value between the current value of the health indicator and the baseline value is determined, and the residual value is used as the current data point of the time series data; wherein the time series operation is a prediction of the future value of the residual value time series data; wherein the formula for calculating the residual value is as follows:

[0012] in, The residual value, The current value of the health indicator. This is the baseline value for health indicators.

[0013] Preferably, before injecting the time-varying virtual disturbance signal into the speed controller, the baseline fluctuation characteristics calculated in step 1 and the current degradation state value of the health indicator output by the time series prediction model in step 4 are obtained; based on the baseline fluctuation characteristics and the current degradation state value, the rate of change of the virtual disturbance signal is determined by a preset decision rule table; the injected virtual disturbance signal is time-varying according to the rate of change.

[0014] Preferably, when determining whether the baseline fluctuation characteristics meet the preset operating condition stability conditions, if it is determined that the baseline fluctuation characteristics do not meet the preset operating condition stability conditions, then step 2 is terminated and step 1 is re-executed after a preset waiting period.

[0015] Preferably, in step 5, in generating prediction information of the unit's primary frequency regulation power response performance based on the future value of the health indicator, the future value of the health indicator is compared with a preset performance failure threshold. When the future value reaches the failure threshold, the remaining performance life is determined and an early warning message is generated.

[0016] Secondly, the present invention also provides a primary frequency regulation power response performance prediction system for thermal power units, comprising: an operating condition module for monitoring the physical position feedback signal of the unit's speed regulation actuator, calculating the baseline fluctuation characteristics of the signal during a verification period, and determining whether the baseline fluctuation characteristics meet preset operating condition stability conditions; a health indicator module for injecting a time-varying virtual disturbance signal into the speed controller and monitoring the signal response when the baseline fluctuation characteristics meet the operating condition stability conditions, and determining the current value of a health indicator characterizing the physical response characteristics of the actuator based on the correspondence between the disturbance signal and the response change; a time series construction module for repeatedly executing the above steps according to a predetermined time period to obtain time series data of the health indicator; a prediction calculation module for using a time series prediction model to predict the future value of the health indicator based on the time series data of the health indicator; and a performance generation module for generating prediction information of the unit's primary frequency regulation power response performance based on the future value of the health indicator.

[0017] Compared with existing technologies, this invention has the following beneficial technical effects: Firstly, this invention provides a method for predicting the primary frequency regulation power response performance of thermal power units. By monitoring the mid-baseline fluctuation characteristics and judging the stable operating conditions, the virtual disturbance signal injection operation is only performed when the unit is in a stable operating condition. This avoids the interference that signal interaction may cause to the normal operation of the unit under fluctuating operating conditions. Simultaneously, the signal environment under stable operating conditions significantly reduces the impact of external interference on the feedback signal, laying a high signal-to-noise ratio signal foundation for the accurate acquisition of subsequent health indicators. Secondly, this invention adopts an active detection method of injecting time-varying virtual disturbance signals. Compared with the passive data collection mode in existing technologies, this method can more directly trigger the response of the actuator, thereby accurately capturing the current value of the health indicator that characterizes the physical response characteristics of the actuator. This achieves active acquisition of the health indicator and solves the problem of unclear health status characterization signals and difficulty in extracting effective indicators under passive acquisition methods.

[0018] Furthermore, regarding the acquisition and prediction process of health indicators, various optimization designs have improved the reliability of the indicators and the accuracy of predictions: On the one hand, by selecting two baseline fluctuation characteristics, variance and peak-to-peak value, it can adapt to the operating characteristics of different units, ensuring the accuracy of stable operating condition judgment. Combined with the termination and retry mechanism when the operating condition is unstable, it ensures the effectiveness of the acquired signals. On the other hand, by adding a second health indicator and jointly predicting with dual time-series data, introducing operating condition benchmark parameters and residual value time-series analysis, the interference of operating condition coupling factors on the health indicator is effectively eliminated, making the health indicator more realistically reflect the inherent health status of the actuator. The design of dynamically determining the change rate of the virtual disturbance signal based on the baseline fluctuation characteristics and the current degradation state value of the health indicator can not only ensure the effective triggering of the response signal, but also avoid the impact of excessive disturbance amplitude on the unit operation, strengthening the technical advantages of "no interference" and "high signal-to-noise ratio".

[0019] Furthermore, by constructing time-series data and predicting time series data, combined with failure threshold judgment and early warning information generation, it is possible to predict the primary frequency regulation power response performance in advance based on highly reliable health indicator time-series data. This provides accurate decision-making basis for unit operation and maintenance, effectively avoiding grid frequency regulation risks caused by primary frequency regulation performance failure. Compared with the existing technology that relies on experience judgment or post-event maintenance, this significantly improves the foresight and economy of operation and maintenance. The quantitative design of residual value calculation further improves the accuracy of prediction, making it easier to accurately capture the degradation trend of health status, providing solid support for reliable prediction. Attached Figure Description

[0020] Figure 1 is a logical architecture diagram of the primary frequency regulation power response performance prediction method for thermal power units according to the present invention; Figure 2 is an example diagram of baseline fluctuation characteristics under pseudo-steady operating conditions according to the present invention; Figure 3 is an analysis diagram of the key technical pillars for achieving reliable prediction according to the present invention; Figure 4 is a closed-loop execution logic state diagram of the prediction method according to the present invention; Figure 5 is a schematic diagram of the primary frequency regulation power response performance prediction system for thermal power units according to the present invention; In the figures: 1. Operating condition project module; 2. Health index module; 3. Time series construction module; 4. Prediction calculation module; 5. Performance generation module. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] The purpose of this invention is to provide a method and system for predicting the primary frequency regulation power response performance of thermal power units, so as to solve the technical problem in the prior art that it is impossible to actively obtain physical health indicators with high signal-to-noise ratio that can be used for reliable prediction without interfering with the operation of the unit.

[0024] The invention will now be described in further detail with reference to the accompanying drawings: The invention provides a method for predicting the primary frequency regulation power response performance of thermal power units. The specific execution process of this method is as follows: Steps 101 and 102 are executed by the operating condition verification unit, namely, performing a pre-detection operating condition purity verification. In actual unit operation, the physical position feedback signal of the speed regulation actuator (the core basis for subsequent monitoring) may be contaminated by minor fluctuations in the power grid or the unit's own oscillations. If a disturbance is injected into this pseudo-stable baseline, it will be impossible to distinguish whether the response is triggered by the disturbance or by the superposition of background noise, leading to inaccurate measurements of subsequent health indicators. To avoid this problem, the operating condition verification unit performs steps 101 and 102, respectively. The condition verification unit continuously monitors a physical position feedback signal of the unit's speed control actuator in a silent manner, i.e., without injecting any signal, within a preset verification period, which can be set to 10 seconds. This signal is typically the position feedback signal of the regulating valve in engineering practice. The unit calculates a baseline fluctuation characteristic of the physical position feedback signal within the verification period. This characteristic is a statistical index used to quantify baseline stability. In one embodiment, the baseline fluctuation characteristic is the variance of the physical position feedback signal within the verification period; in another embodiment, the baseline fluctuation characteristic is the peak value of the physical position feedback signal within the verification period.

[0025] The operating condition verification unit then executes step 102 to determine whether the calculated baseline fluctuation characteristics meet a preset operating condition stability condition. Specifically, this condition requires the baseline fluctuation characteristics to be below a preset fluctuation threshold. Determining this threshold is an engineering calibration process to ensure measurement reliability. During unit commissioning, under confirmed absolutely stable operating conditions, the background noise data of the aforementioned position feedback signal is collected, and the threshold is set based on the statistical distribution of this background noise. A specific method is to take three times the standard deviation of its variance to ensure that the threshold can effectively filter out background oscillation interference while avoiding frequent detection interruptions due to excessive sensitivity. The system continues to step 103 only when the baseline fluctuation characteristics meet the operating condition stability condition. If the condition is not met, the execution of subsequent steps 103 is stopped, and after a preset waiting period (e.g., 15 minutes), the system returns to re-execute the operating condition verification of step 101. This mechanism ensures the purity and reliability of subsequent measurement data from the source. It should be noted that steps 101, 102, and... A crucial execution logic loop exists between steps 103 to address the engineering reality that the unit may not meet the stable operating conditions for an extended period. When the baseline fluctuation characteristics are determined not to meet the preset stable operating conditions in step 102, the system will execute a stop command, preventing the disturbance injection unit from executing step 103, thus avoiding invalid measurements on the contaminated baseline. After the stop is executed, the system will start an internal timer and enter a preset waiting period. The duration of this waiting period can be set according to the operating characteristics of the unit, with a specific duration of 15 minutes being acceptable. After the preset waiting period ends, the system will automatically return and re-execute step 101, i.e., restart a verification period to reassess the purity of the current operating conditions. This verification-stop-wait-retry loop ensures that the method can automatically and continuously search for measurement windows until the operating conditions meet the requirements and a high-reliability health indicator is successfully obtained, thereby guaranteeing the long-term availability and robustness of the entire prediction system in a real operating environment.

[0026] After the operating condition verification is passed, the disturbance injection unit and the health indicator determination unit work together to execute step 103, namely the active micro-disturbance detection process. The disturbance injection unit injects a time-varying virtual disturbance signal into the unit's speed controller. This signal is not a real disturbance to the power grid, but is injected into the speed controller through the DCS system, such as a software signal for the speed / frequency setpoint. In a preferred embodiment, this signal is a virtual frequency offset signal. To achieve zero impact on unit operation, the rate of change of this signal is strictly limited to below a preset micro-disturbance threshold. This rate can be set to an extremely slow micro-ramp, such as... This rate is far lower than the normal frequency regulation response rate of the unit, causing the power change to be spread below the noise level, ensuring the safety and concealment of the detection process. Simultaneously with the injection of the virtual disturbance signal, the health indicator determination unit continuously monitors the response of the physical position feedback signal, such as a 10ms periodicity signal, at high frequency. Based on the correspondence between the virtual disturbance signal and the physical position feedback signal's response changes, it determines the current value of a health indicator (HI) characterizing the physical response characteristics of the actuator. In one specific implementation, the health indicator is determined as the physical dead zone width characterizing the mechanical jamming and static friction characteristics of the actuator. The specific procedure for determining the current value of the health indicator is as follows: when the physical position feedback signal undergoes a preset response change, this response change is quantified as the signal's movement amplitude explicitly exceeding the fluctuation threshold specified in step 102 for the first time. The system immediately records the current amplitude of the virtual disturbance signal that triggered this response change as the upward dead zone boundary. The disturbance injection unit reverses the virtual disturbance signal (e.g., using...). The health indicator determination unit monitors and records the signal amplitude when a reverse response change occurs in the same way, which is recorded as the downlink dead zone boundary. Ultimately, the current value of the health indicator was determined to be: This value directly and with a high signal-to-noise ratio quantifies the physical dead zone of the actuator. The system executes step 104, that is, according to a predetermined time period, such as setting it to repeat steps 101 to 103 every day at 3:00 AM, thereby obtaining a time series of the health indicator, denoted as... Next, the performance prediction unit executes step 105, which uses a time series prediction model to predict a future value of the health indicator based on the acquired time series of the health indicator. Since this technical solution ensures that the HI time series is low-frequency, slowly changing and has a high signal-to-noise ratio through active detection and operating condition verification, the model selection does not need to rely on complex models designed for processing low-quality data. Instead, a simple, transparent and computationally robust classical statistical model can be used. In a preferred embodiment, the time series prediction model is an autoregressive integral moving average (ARIMA) model or a Kalman filter model.

[0027] Finally, the information generation unit executes step 106, generating predictive information on the unit's primary frequency regulation power response performance based on the future value of the health indicator. In one specific implementation, step 106 specifically includes: comparing the future value of the health indicator or its extrapolated trend with a preset performance failure threshold, such as one set according to grid assessment standards. Dead zone thresholds are compared; when it is predicted that the future value of the health indicator will reach the performance failure threshold, the system determines a remaining performance lifetime (RUL) and generates corresponding early warning information, providing operators with actionable predictive maintenance decision-making basis; to further improve the performance and diagnostic dimensions of the predictive system, this invention also provides a series of in-depth technical paths. In one in-depth implementation, it aims to solve the problem of a single dimension of PFR performance degradation, and the health of the actuator includes not only mechanical jamming (… That is, the physical dead zone), and also includes the dynamic response characteristics of the servo system ( Therefore, this solution determines the dead zone in step 103. During the process, the detection information is reused. Specifically, in step 103 (corresponding to step 301), after the physical position feedback signal changes in response and overcomes the dead zone, the system does not immediately stop monitoring. Instead, it continues to monitor the dynamic waveform of the physical position feedback signal within a transient capture window (such as an additional 1-2 seconds). In step 302, based on the physical position feedback signal within this transient capture window (such as calculating the maximum slope of its response curve), a second health indicator characterizing the dynamic response characteristics of the actuator is determined. Accordingly, step 104 is upgraded to obtaining... and The two-dimensional time series; step 105 then upgrades to using a multivariate time series prediction model (such as a vector autoregressive (VAR) model) to jointly predict the two time series. This method obtains orthogonal degradation dimensions without adding any new test actions, extending the system's capabilities from failure prediction to fault mode attribution to distinguish between mechanical and hydraulic problems; in another in-depth implementation, the aim is to solve the problem of operating condition coupling, that is, the difficulty in distinguishing whether changes in HI are due to physical degradation or operating condition drift. To this end, this scheme introduces an operating condition decoupling mechanism. Specifically, in step 103, the operating condition decoupling mechanism is determined... While obtaining the current value, a reference parameter characterizing the current operating condition of the unit is simultaneously acquired, such as the steady-state average value of the feedback of the regulating valve position. In step 104, the following sub-steps are performed: Step 401, based on the operating condition reference parameters. With health indicators The historical correspondence of the current value, that is, the data accumulated in the early stage. A baseline mapping model of working condition-health indicator is established using regression fitting methods (such as polynomial fitting). This model characterizes the inherent relationship between HI and operating conditions under a healthy state; step 402, based on this benchmark mapping model and the current operating condition benchmark parameters... ,calculate The corresponding health indicator baseline value Step 403, confirm and A residual value between The residual value is used as the current data point of the HI time series; the calculation of the residual value is limited by the following formula: In the formula, The residual value, The current value of the health indicator. This serves as a baseline value for the health indicator; correspondingly, step 105 specifically involves predicting the time series of this residual value. The future value of this mechanism ensures that the tracked sequence purely reflects the actual wear process of the physical components.

[0028] In another, more refined implementation, the aim is to address the trade-off between detection rate and speed. A fixed detection rate cannot simultaneously satisfy the need for slow quasi-static measurement accuracy and the need for fast detection to avoid operational disturbances. Therefore, this solution introduces an adaptive compensation mechanism for the detection rate. Specifically, before injecting the virtual disturbance signal in step 103, the following step is added: Step 601, obtaining the baseline fluctuation characteristics calculated in step 101 to characterize the noise level of the current operating condition. Step 602: Obtain the current degradation state value of the health indicator output by the time series prediction model in step 105. Step 603, based on baseline fluctuation characteristics ( ) and the current degradation status value of the health indicator ( The rate of change of the virtual disturbance signal is determined by a pre-defined decision rule table, such as a two-dimensional lookup table. ); an example of this decision rule is: if High indicates a noisy working condition. Take a faster rate to avoid contamination; if High indicates a sluggish system. A slower rate is adopted to ensure quasi-static measurement; and the virtual disturbance signal injected in step 103 is based on the dynamically determined rate of change. By implementing time-varying mechanisms, the detection system is upgraded from an open-loop to a closed-loop system, improving the effectiveness of measurements under real-world conditions.

[0029] Example 1: In a specific application scenario, a 600MW thermal power unit faces the peak summer electricity demand period two months later, and the power grid has strict requirements for its primary frequency regulation performance. The unit's operation and maintenance team suspects that the unit's speed regulation actuator has progressive physical wear, and its primary frequency regulation performance may be on the verge of being acceptable. However, existing technologies in this field face an engineering trade-off: performing an offline, active high-disturbance test, while accurately calibrating the current performance, requires shutting down the unit before the peak period and incurring test costs and power generation losses; while relying on passive micro-disturbance data from daily operation for online identification yields data with too low a signal-to-noise ratio to provide a reliable degradation conclusion for decision-making, making it impossible for the operation and maintenance team to formulate a scientific maintenance plan. The unit has deployed a predictive system; after the system is put into operation, its operating condition verification unit automatically triggers according to preset logic, taking 3:00 AM every day as an example; in the first execution cycle, the unit executes step 101, entering a 10-second silent monitoring phase, sampling the position feedback signal of the regulating valve, and calculating its baseline fluctuation characteristics. The obtained variance value is In step 102, the system compares the variance value with a preset stable operating condition, i.e., a... The fluctuation threshold is compared to determine the current feature. If the value is below this threshold, the purity of the operating condition passes the verification.

[0030] After obtaining permission from the operating condition verification unit, the system immediately executes step 103, and the disturbance injection unit then injects a disturbance with a change rate of [missing value] into the speed controller. The virtual frequency offset signal; simultaneously with signal injection, the unit's DCS system displays that its actual active power fluctuation is less than 0.1MW, which is completely submerged in the unit's normal random fluctuations and has no perceptible impact on unit operation and grid stability; during this process, the health indicator determination unit continuously monitors the position feedback signal of the regulating valve until it detects a clear response change in the signal exceeding the aforementioned fluctuation threshold, at which point the upward dead zone boundary is recorded. for The signal is injected in reverse, and the downlink dead zone boundary is measured in the same way. for The system then determined the date. The health indicator, namely the physical dead zone width, has a current value of Following step 104, the system automatically repeats steps 101 to 103 daily for the next 30 days, using a predetermined time period of one day, thereby obtaining a time series of a health indicator. The numerical values ​​of this sequence are presented as The sequence is sent to the performance prediction unit to execute step 105; the performance prediction unit uses an autoregressive integral moving average model, i.e., the ARIMA model, to fit and extrapolate the time series of the health indicator to predict the future value of the HI sequence; the information generation unit executes step 106 to compare the predicted trend with a built-in... The performance failure threshold is compared, and a performance prediction information is finally output: the physical dead zone index of the unit's primary frequency regulation performance is expected to be within the next 45 days ( The system reached its failure threshold on the 30th of the summer peak season, with a remaining performance lifetime (RUL) of 45 days. This forecast provided the operations team with a decision window, allowing them to schedule preventative maintenance on the speed control actuators 30 days before the peak season, thus avoiding the risk of performance failures and penalties during peak periods.

[0031] Example 2: To objectively verify the necessity of steps 101 and 102 of the present invention for actively detecting the high-confidence health indicator (HI) in step 103, and to verify the robustness of the method under background noise interference, this example constructs a digital simulation test platform; the platform is based on a thermal power unit simulation model that includes the nonlinear characteristics of the speed regulating actuator, and the model can set a real physical dead zone width that slowly degrades over time ( ), and can superimpose a low-frequency background oscillation signal simulating a pseudo-steady state in real working conditions ( The experiment setup is as follows: (Settings...) Day 0 at the start of the experiment was and daily The rate increases linearly, simulating the slow physical wear of components; [Settings are missing from the original text] For a peak value is A periodic oscillation signal is continuously applied to the position feedback signal of the regulating valve during the test to simulate micro-amplitude oscillation contamination under operating conditions; the fluctuation threshold in step 102 of the operating condition stability condition of the method of the present invention is set as... In terms of variance, this threshold is higher than the simulation background noise but lower than... The resulting fluctuations; the rate of change of the virtual disturbance signal in the active detection step 103 is set to... The experiment was conducted in three groups, with measurements taken once daily for 10 consecutive days: Control group A (simulating existing passive micro-perturbation methods): This group did not inject any signal, but only attempted to obtain signals from the superimposed... In the daily operation signals, the online system identification algorithm is used to fit HI; control group B simulates active detection without operating condition verification: this group does not execute steps 101 and 102, but directly executes the active detection in step 103 at a fixed time every day. At this time, the detection signal will be superimposed on the unknown Above; the sample group of this invention adopts the complete method of this invention: the group strictly follows steps 101 to 103, that is, it must first pass the working condition verification, in Active detection was triggered only during the trough of oscillations, i.e., when the baseline oscillation characteristic fell below the oscillation threshold. The experiment recorded the daily HI values ​​measured by the three methods over 10 days and compared them with the simulation settings. For comparison, see Table 1 for key data.

[0032] Table 1: Comparison of HI measurement values ​​between the experimental group and the control group

[0033] The data analysis is as follows: the measurement results of passive identification in control group A (RMSE is...). It exhibits violent, irregular fluctuations, completely submerged in In this context, the data sequence cannot reflect... The true degradation trend; although the measurement results of control group B (active probe without verification) were stable, they systematically and severely deviated from the actual degradation trend. (RMSE is) Its measured value is approximately Far below the true value This is because The peak value of the oscillation wave is approximately The superposition of the actively injected signal causes the system to misjudge a response change when the injected signal has not yet reached the true dead zone boundary, thus giving a false and lower measurement result; in contrast, steps 101 and 102 of the working condition verification unit of the present invention detect... If the baseline fluctuation characteristics exceed the limit, the detection mission will be suspended until a window of clean operating conditions is captured before proceeding to step 103; its measurement results (RMSE are only...) )and The data accurately and consistently tracked the true degradation trajectory of the physical dead zone. This experimental data shows that the operating condition verification step is a necessary prerequisite for obtaining a high-reliability HI. The HI data obtained by the active detection control group B, which lacks operating condition verification, under pseudo-steady operating conditions is distorted. If a false low dead zone value is used to make a prediction based on this data, an incorrect conclusion about the system's health will be drawn, thus missing the maintenance window and causing risks in health management decisions.

[0034] Example 3: To objectively verify the necessity of the technical solution of the present invention, this example sets up two sets of comparative experiments. The first set of comparative experiments is used to compare the stability and reliability of the active-perturbation detection combined with simplified prediction model (steps 103-105) of the present invention with the passive perturbation identification combined with complex prediction model in the background technology in terms of remaining performance lifetime (RUL) prediction. This experiment uses the same digital simulation test platform as Example 2, and sets... From day 0 Starting daily The rate of increase is linear, i.e., the actual RUL is 130 days on day 0. Experimental group 1 of the present invention sample group is set up: the high signal-to-noise ratio, low frequency once-daily health indicator (HI) time series obtained by the sample group of the present invention in Example 2 is used. This series is fed into a simplified time series prediction model ARIMA model, and RUL prediction is performed in step 105. Experimental group 2 is set up as a comparative sample group. The background technology is simulated: the high frequency (10ms sampling), low signal-to-noise ratio regulating valve position feedback signal raw data stream obtained by control group A in Example 2 is used. This data stream is fed into a complex prediction model for processing time series data, a long short-term memory neural network, LSTM model. This model is trained to directly identify the system state from high frequency noise data and predict RUL. Both groups perform RUL prediction on days 30, 40, 50 and 60 of the experiment. The comparison between the prediction results and the actual RUL is shown in Table 2.

[0035] Table 2: Comparison of the stability of prediction methods

[0036] Experimental results show that, due to the low signal-to-noise ratio of the passive perturbation in its input data source, the degradation trend of the HI (Health Management Scale) in experimental group 2 was completely submerged in measurement noise and operating condition fluctuations. Despite employing a complex LSTM model to attempt to extract features from the noise, the predicted RUL values ​​for 185.3 days, 72.1 days, 143.7 days, and 95.2 days exhibited drastic and irregular jumps, rendering the results unreliable and unable to provide effective maintenance decision-making support for G05B23 (Health Management). In contrast, experimental group 1, the present invention sample, obtained high signal-to-noise ratio from the source through the active perturbation detection step 103. The noise ratio (HI) sequence clearly reflects the physical degradation trend; therefore, by using only a simplified ARIMA model step 105, stable and high-precision RUL predictions can be made at different time points for 100.8 days, 90.5 days, 80.9 days, and 69.7 days. The prediction results are in high agreement with the actual RUL. This set of comparative data confirms that the technical path that relies on passive observation data and attempts to process data through complex models has inherent defects in prediction reliability. The technical path of actively acquiring high-quality data sources combined with a simplified model adopted in this invention is a prerequisite for achieving reliable RUL prediction.

[0037] The second set of comparative experiments was used to verify the necessity of steps 101 and 102 of the present invention for the prediction results. This experiment used the same simulation platform and parameter settings as Example 2, and specifically observed a peak-to-peak value of... Low-frequency background oscillation signal ( A comparative sample group is set up, whose technical solution removes steps 101 and 102 (i.e., the operating condition verification unit) of the present invention; this sample group exists Under interference conditions, the active perturbation detection in step 103 is directly executed to obtain the HI time series; this series corresponds to the distorted data measured by control group B in Example 2, i.e., a mean of... For nearby HI sequences with no obvious trend, this comparative sample group uses the same steps 105 (ARIMA model) and 106 (RUL calculation) as the sample group of this invention to extrapolate the trend of the distorted HI sequence. The experimental results show that, due to the input HI sequence (mean... ) and the true value ( The above (the data) contains a significant deviation, and its subtle internal degradation trend has been... Due to contamination by random disturbances, the ARIMA model cannot fit the correct degradation rate. When the model performs prediction on day 30, it calculates the remaining performance lifetime (RUL) as 1342 days. This 1342-day prediction result deviates from the actual RUL of 100 days by more than 1200%, which is a completely wrong prediction conclusion. This conclusion will lead to maintenance personnel misjudging the health status of the unit, thus missing maintenance opportunities and causing the risk of PFR performance failure. The comparative data confirms that the active-disturbance detection step 103 must work in conjunction with the operating condition verification steps 101 and 102. If the operating condition verification step is missing, the HI data obtained by the active detection itself under pseudo-steady operating conditions will be distorted. Based on this distorted data, the prediction steps 105 and 106 will lead to serious distortion of the prediction results and cause decision-making errors. This, in turn, proves the necessity of the complete verification-detection-prediction technology chain proposed in this invention.

[0038] Example 4: This example, in conjunction with Figures 1 to 4, describes a method and system for predicting the primary frequency regulation power response performance of a thermal power unit. As shown in Figure 1, this architecture relies on an operating condition verification unit to monitor the input physical location feedback signal of the unit to calculate baseline fluctuation characteristics and determine whether the operating condition stability condition is met. If the condition is met, the disturbance injection unit injects a time-varying virtual disturbance signal into the speed controller. The health indicator determination unit then determines the current value of the health indicator HI and obtains its time series based on the disturbance and the unit response. This determination process can also receive the health indicator benchmark value calculated from the operating condition health indicator benchmark mapping model to achieve operating condition decoupling. The performance prediction unit uses a time series prediction model to predict the future value of HI. Finally, the information generation unit generates performance prediction information including the remaining lifetime (RUL) based on the future value of HI. If the operating condition verification does not meet the condition, the process is redirected to stop execution and retry after waiting for a cycle.

[0039] As shown in Figure 2, this chart uses time (in seconds) as the horizontal axis and signal amplitude as the vertical axis to show the relationship between the physical location feedback signal and a fluctuation threshold. The physical location feedback signal shown in the figure exhibits periodic oscillations, with its peaks and troughs repeatedly crossing the set fluctuation threshold. This situation corresponds to pseudo-stable operating conditions that need to be identified and filtered out by the operating condition verification unit to ensure the accuracy of subsequent detection. As shown in Figure 3, achieving the goal of reliable prediction of the primary frequency regulation performance of thermal power units relies on technological innovation in four dimensions: first, the operating condition verification mechanism to ensure data purity; second, active detection data source, which achieves zero-impact safety detection and obtains high signal-to-noise ratio (HI) by injecting time-varying virtual disturbances; third, multi-dimensional health indicators to distinguish between physical dead zone width (HI1) and dynamic response characteristics (HI2) to achieve fault mode attribution; and fourth, advanced model algorithms, including operating condition decoupling residual prediction and the use of simplified models such as ARIMA and multivariate joint prediction.

[0040] As shown in Figure 4, the flowchart illustrates the closed-loop flow of the system between three states: operating condition verification, suspension waiting, and active detection. The system starts from the operating condition verification state. If the operating condition is not met, it transitions to the suspension waiting state and returns to the operating condition verification state after a preset waiting period. If the operating condition meets the conditions, it transitions to the active detection state. In this state, the system will fully perform core tasks such as injecting virtual disturbance signals, acquiring HI time series, and using models to predict future values. After detection and prediction are completed, the system returns to the operating condition verification state to wait for the next predetermined period.

[0041] Example 5: This example describes the specific calibration process for several key parameters in the aforementioned method for predicting the primary frequency regulation power response performance of a thermal power unit. These parameters include the fluctuation threshold in step 102, the operating condition-health indicator benchmark mapping model in step 401, the decision rule table in step 603, and the rate threshold used for fault mode classification. The calibration process provides reproducible engineering basis for deploying this method on specific units. To determine the fluctuation threshold in step 102, i.e., the basis for distinguishing between a pure operating condition and a pseudo-stable operating condition, a standardized calibration procedure is performed. This procedure is performed during the unit's recognized stable period, such as 2:00 AM to 4:00 AM at night, when the grid frequency fluctuation is less than [a certain value]. Under the specified conditions, the operating condition verification unit is triggered to continuously perform 100 silent monitoring cycles. Each monitoring cycle, or verification period, lasts for 10 seconds and is used to collect the position feedback signal of the regulating valve. The variance values ​​of these 100 sets of signals are calculated to form a variance sample set containing 100 data points. Statistical analysis is performed on this sample set to obtain its mean. The standard deviation is To avoid misclassifying normal background noise as operational instability at a 99.7% confidence level (i.e., the 3-sigma principle), the fluctuation threshold was determined to be: A value is obtained through calculation. This value serves as the basis for determining the stability condition in step 102.

[0042] To establish the operating condition-health indicator baseline mapping model in step 401, the system utilizes the aforementioned [method / method] during the first month of operation. The fluctuation threshold was determined, and a total of 90 active-disturbance detection tests that passed the operating condition verification were performed, i.e., the execution of step 103; the system recorded the operating condition reference parameters corresponding to each detection, i.e., the average valve position. Its measurement range covers 32.5% to 91.3%, and it measures health indicators, namely physical dead zones. These 90 sets of data are... Used as the training set, the system employs a third-order polynomial regression fitting method, which is a specific implementation of step 401. The model coefficients are calculated using the least squares method, ultimately establishing a work condition-health indicator baseline mapping model with definite coefficients, in the form of: This model is embedded in the health indicator determination unit and is used to calculate the relationship with the current operating condition in each subsequent measurement. Corresponding health indicator baseline values Then, in step 403, the residual value used for prediction is calculated. Its calculation follows: In the formula, The residual value, The current value of the health indicator. As a baseline value for health indicators, Using the current operating condition baseline parameters, and to construct the decision rule table in step 603 to achieve adaptive detection rate, the system is based on the baseline fluctuation characteristics (variance value) obtained in step 601 and the current degradation state value of HI obtained in step 602. The current value defines the decision logic; this logic divides the baseline fluctuation characteristics into two levels: less than The state is defined as quiet. to The state is defined as a perturbation, while the state greater than 1 is a perturbation. Then the detection is stopped; at the same time, the current degradation state value of HI is divided into three levels: less than Defined as health to Defined as attention, greater than Defined as an early warning; based on this, the system establishes a 2x3 decision table to dynamically determine the rate of change of the virtual disturbance signal in step 103. If the operating conditions are quiet and the patient is in good health, then Set to standard rate If the operating condition is quiet but the status is "attention" or "warning," it indicates that the system is sluggish. Set to a slower rate To ensure quasi-static measurement; if the operating condition is a slight disturbance, it indicates a risk of contamination. All are set to a relatively fast rate To expedite detection and avoid interference; and to determine the rate threshold for fault mode classification, after one year of stable operation, the performance prediction unit has acquired a time series of HI residuals containing 300 valid data points. The system calculates the first derivative of the sequence, i.e., the daily degradation rate, and performs histogram statistical analysis on the rate dataset, showing that its distribution is highly concentrated near zero; the system sets a normal wear threshold. The 95th quantile of this rate distribution was calculated. Simultaneously set a threshold for sudden failures. A value based on engineering margins and much larger than normal wear is set as follows: This value is close to The two thresholds are fixed in the information generation unit and used to compare the current degradation rate output by the prediction model in step 105 when performing step 106, so as to achieve automatic classification of normal wear and sudden failure.

[0043] Example 6, as shown in Figure 5, also provides a primary frequency regulation power response performance prediction system for thermal power units. This system can be deployed in the distributed control system (DCS) of the thermal power unit or in an online monitoring server connected to it. Functionally, the system includes an operating condition module 1, a health indicator module 2, a timing construction module 3, a prediction calculation module 4, and a performance generation module 5. Specifically, it includes: the operating condition module 1, used to monitor the physical position feedback signal of the unit's speed regulation actuator, calculate the baseline fluctuation characteristics of the signal during the verification period, and determine whether the baseline fluctuation characteristics meet the preset operating condition stability conditions; and the health indicator module 2, which, when the baseline fluctuation characteristics meet the operating condition stability conditions, injects a time-varying virtual disturbance signal into the speed controller and monitors the signal response, and determines the current value of the health indicator characterizing the physical response characteristics of the actuator based on the correspondence between the disturbance signal and the response change.

[0044] The time series construction module 3 repeats the above steps according to a predetermined time period to obtain the time series data of the health indicator; the prediction calculation module 4 is used to use a time series prediction model to predict the future value of the health indicator based on the time series data of the health indicator; the performance generation module 5 is used to generate prediction information of the unit's primary frequency regulation power response performance based on the future value of the health indicator.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the primary frequency regulation power response performance of a thermal power unit, characterized in that, include: Step 1: Monitor the physical position feedback signal of the unit's speed regulation actuator, calculate the baseline fluctuation characteristics of the signal during the verification period, and determine whether the baseline fluctuation characteristics meet the preset stable operating conditions. Step 2: When the baseline fluctuation characteristics meet the stable operating conditions, inject a time-varying virtual disturbance signal into the speed controller and monitor the signal response. Based on the correspondence between the disturbance signal and the response change, determine the current value of the health indicator that characterizes the physical response characteristics of the actuator. Step 3: Repeat steps 1-2 at a predetermined time period to obtain the time series data of the health indicator. Step 4: Use a time series prediction model to predict the future value of the health indicator based on the time series data of the health indicator. Step 5: Based on the future value of the health indicator, generate prediction information for the primary frequency regulation power response performance of the unit.

2. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 1, characterized in that, In step 1, when the baseline fluctuation feature is the variance value of the physical location feedback signal during the verification period, the preset operating condition stability condition is that the variance value is lower than a preset fluctuation threshold; when the baseline fluctuation feature is the peak-to-peak value of the physical location feedback signal during the verification period, the preset operating condition stability condition is that the peak-to-peak value is lower than a preset fluctuation threshold.

3. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 1, characterized in that, In step 2, the health indicator is the physical dead zone width. The specific way to determine the current value of the health indicator is as follows: when the physical position feedback signal undergoes a preset response change, the amplitude of the virtual disturbance signal that triggers the response change is recorded, and the physical dead zone width is determined by the amplitude.

4. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 3, characterized in that, After the physical position feedback signal undergoes a preset response change, the signal is monitored again within a preset transient capture window. Based on the signal within the window, a second health indicator characterizing the dynamic response of the actuator is determined. The time series data consists of dual time series data of the health indicator and the second health indicator. A multivariate time series prediction model is used to jointly predict the dual time series data.

5. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 3, characterized in that, While determining the current value of the health indicator that characterizes the physical response characteristics of the actuator, the operating condition reference parameters that characterize the current operating conditions of the unit are simultaneously acquired.

6. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 5, characterized in that, Based on the historical correspondence between the benchmark parameters of the operating condition and the current value of the health indicator, a benchmark mapping model of the operating condition and health indicator is established by using a regression fitting method. Based on this mapping model and the current operating condition baseline parameters, the baseline value corresponding to the current value of the health indicator is calculated; The residual value between the current value of the health indicator and the baseline value is determined, and the residual value is used as the current data point of the time series data; wherein the time series operation is to predict the future value of the time series data of the residual value; wherein the formula for calculating the residual value is as follows: in, The residual value, The current value of the health indicator. This is the baseline value for the health indicator.

7. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 1, characterized in that, Before injecting the time-varying virtual disturbance signal into the speed controller, the baseline fluctuation characteristics calculated in step 1 and the current degradation state value of the health indicator output by the time series prediction model in step 4 are obtained; based on the baseline fluctuation characteristics and the current degradation state value, the rate of change of the virtual disturbance signal is determined by a preset decision rule table; The injected virtual disturbance signal is time-varying according to the stated rate of change.

8. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 1, characterized in that, When determining whether the baseline fluctuation characteristics meet the preset operating condition stability conditions, if it is determined that the baseline fluctuation characteristics do not meet the preset operating condition stability conditions, then step 2 is terminated and step 1 is re-executed after a preset waiting period.

9. The method for predicting the primary frequency regulation power response performance of a thermal power unit according to claim 1, characterized in that, In step 5, based on the future value of the health indicator, the predicted information of the unit's primary frequency regulation power response performance is generated. The future value of the health indicator is compared with the preset performance failure threshold. When the future value reaches the failure threshold, the remaining performance life is determined and an early warning information is generated.

10. A system for predicting the primary frequency regulation power response performance of a thermal power unit, characterized in that, include: The operating condition module is used to monitor the physical position feedback signal of the unit's speed regulation actuator, calculate the baseline fluctuation characteristics of the signal during the verification period, and determine whether the baseline fluctuation characteristics meet the preset operating condition stability conditions. The health indicator module injects a time-varying virtual disturbance signal into the speed controller and monitors the signal response when the baseline fluctuation characteristics meet the stable operating conditions. Based on the correspondence between the disturbance signal and the response change, it determines the current value of the health indicator that characterizes the physical response characteristics of the actuator. The time series construction module repeats the above steps at a predetermined time period to obtain the time series data of the health indicator. The prediction calculation module is used to predict the future value of the health indicator based on the time series data of the health indicator using a time series prediction model. The performance generation module is used to generate predictive information on the primary frequency regulation power response performance of the unit based on future values ​​of health indicators.

Images