Hydropower station agc load distribution method and system considering vibration risk

Abstract

This invention belongs to the technical field of information technology, specifically disclosing a method and system for AGC load allocation in hydropower stations considering vibration risk. The method includes: acquiring real-time tilt angle time-series data of the unit bearing housing and guide vane support; processing the real-time tilt angle time-series data using preset high-pass and low-pass filters; the high-pass filter output representing rapid fluctuation components changing from seconds to minutes, and the low-pass filter output representing slow drift components changing from hours to days; calculating the transient vibration impact index and the long-term structural degradation index based on the amplitude characteristics of the rapid fluctuation components and the cumulative trend of the slow drift components, respectively; the transient vibration impact index reflecting the instantaneous mechanical stress level caused by the current load adjustment; the purpose of this invention is to solve the problem in the prior art of how to simultaneously identify and weigh the transient vibration risk caused by rapid tilt angle changes and the long-term structural degradation risk caused by slow tilt angle drift in load allocation decisions.

Description

Technical Field

[0001] This invention relates to the technical field of information technology, specifically to a method and system for AGC load allocation in hydropower stations that takes vibration risk into account. Background Technology

[0002] Automatic generation control (AGC) load allocation in hydropower stations is a core technological link in ensuring the stable operation of the power grid and the efficient utilization of hydropower. Its importance lies in both rapidly responding to changes in grid load demand and ensuring the long-term safe and reliable operation of generating units. With the large-scale integration of new energy sources, the requirements of the power grid for frequency regulation and peak shaving of hydropower stations are becoming increasingly stringent. The accuracy and security of load allocation directly affect the stability and economy of the entire power system.

[0003] Current load allocation methods primarily optimize based on unit output characteristics, head changes, and efficiency curves. However, in actual operation, the dynamic impact of vibration risks is often overlooked. Existing methods often only focus on static or average vibration levels, making it difficult to capture localized abnormal vibrations caused by uneven mechanical structural responses during rapid load adjustments. Furthermore, it's even harder to distinguish whether these vibrations are transient shocks or precursors to continuous, cumulative structural damage. This neglect means that in high-frequency load adjustment scenarios, some units may unknowingly experience stresses exceeding safety thresholds, creating potential equipment failure hazards.

[0004] The core characteristic of vibration risk in hydropower station units lies in the fact that vibration is not a single indicator, but is closely related to the time scale. Inclination changes in critical components such as bearing housings and guide vane supports exhibit both rapid fluctuations and slow drifts: inclination fluctuations on the order of seconds to minutes often correspond to transient vibration shocks caused by sudden load changes, while cumulative inclination drift on the order of hours to days reflects long-term degradation caused by structural fatigue, foundation settlement, or temperature changes. These two types of changes are physically interconnected: rapid fluctuations amplify local stress concentrations, accelerating the occurrence of slow drifts; conversely, slow drifts reduce structural stiffness, causing the same load adjustment to trigger larger transient vibration amplitudes. Existing monitoring methods typically only focus on vibration amplitude, making it difficult to effectively separate and quantify the individual contributions and interactive effects of these two types of changes. This results in an inability to accurately determine whether a load command at a given moment will simultaneously trigger the dual risks of transient failure and long-term damage.

[0005] For example, when the power grid requires the generating unit to rapidly climb from low load to high load within minutes, the guide vane support tilt angle may experience drastic fluctuations on a second-by-second basis. Simultaneously, the bearing housing may have accumulated slow misalignment over several hours. If the load is still allocated in the conventional manner, the generating unit may experience high-vibration shocks in a short period. These shocks further exacerbate the existing misalignment, creating a vicious cycle. How to simultaneously identify and weigh the transient vibration risks from rapid tilt angle changes against the long-term structural degradation risks caused by slow tilt angle drift in load allocation decisions becomes a key issue in achieving safe and efficient AGC load allocation for hydropower stations. Summary of the Invention

[0006] This invention provides a hydropower station AGC load allocation method and system that considers vibration risk. The purpose is to solve the problem in the prior art of how to simultaneously identify and weigh the transient vibration risk caused by rapid tilt angle changes and the long-term structural degradation risk caused by slow tilt angle drift in load allocation decisions.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for AGC load allocation in a hydropower station considering vibration risk includes: acquiring real-time tilt angle time-series data of the unit bearing housing and guide vane support; processing the real-time tilt angle time-series data using preset high-pass and low-pass filters; the high-pass filter output represents the rapid fluctuation component changing from seconds to minutes, and the low-pass filter output represents the slow drift component changing from hours to days; calculating the transient vibration impact index and the long-term structural degradation index based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, respectively; the transient vibration impact index reflects the instantaneous mechanical stress level caused by the current load adjustment, and the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation; establishing a vibration risk coupling model, the input of which is the transient vibration impact index and the long-term structural degradation index, and the output of which is a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at the two time scales of rapid fluctuation and slow drift; and for each unit participating in AGC allocation... The system obtains the current comprehensive vibration risk coefficient of the unit and compares it with the preset safe operation threshold of the unit to determine whether the unit is in a high vibration risk state. If the unit is determined to be in a high vibration risk state, an optimization algorithm based on vibration risk constraints is adopted when the unit participates in the load allocation calculation. This algorithm introduces the comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units under the premise of meeting the total load command of the power grid. The optimization algorithm based on vibration risk constraints is used to solve the problem and obtain a set of output commands for each unit that meet the total load requirements. This set of output commands meets the power grid demand while actively avoiding load allocation schemes that may cause high-risk vibrations. The calculated output commands for each unit are sent to the unit control system for execution, and the tilt angle time series data obtained in the first step is continuously monitored during the execution process, forming a closed-loop control process from vibration monitoring, risk analysis to load allocation decision.

[0008] In one aspect of the invention, the acquisition of real-time tilt angle time-series data of the unit bearing housing and guide vane support involves processing the real-time tilt angle time-series data using preset high-pass and low-pass filters. The high-pass filter output represents rapid fluctuation components changing from the second to the minute level, and the low-pass filter output represents slow drift components changing from the hour to the day level. Acquire real-time tilt angle timing data of the unit bearing housing and guide vane support; Real-time tilt time-series data is processed using a preset high-pass filter to obtain fast fluctuation components; Real-time tilt time-series data is processed using a preset low-pass filter to obtain the slow drift component; The standard deviation of the rapid fluctuation component is calculated for each minute to obtain a minute-level fluctuation intensity sequence; If three consecutive points in the minute-level fluctuation intensity sequence exceed the preset fluctuation threshold, it is judged as a rapid abnormal event, and the time of occurrence of the event is recorded. For the slowly drifting component, a sample value is taken once per hour to obtain an hourly drift sequence; The difference between adjacent sampling points is calculated based on the hourly drift sequence to obtain the drift rate sequence; If the absolute value of a point in the drift rate sequence exceeds the preset drift rate threshold, it is judged as a slow drift anomaly, and the time and direction of the anomaly are recorded. The timing of a rapid anomaly is compared with the timing of a slow drift anomaly. If the time difference between the two is less than a preset time window, it is determined to be a composite anomaly. Output the time, location, and type of all compound exception events.

[0009] In one aspect of the invention, the calculation of the transient vibration impact index and the long-term structural degradation index based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, respectively, wherein the transient vibration impact index reflects the instantaneous mechanical stress level caused by the current load adjustment, and the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation, includes: Obtain the vibration signal sequence; The fast fluctuation component and the slow drift component are separated by the high-pass filter and the low-pass filter, respectively. Calculate the amplitude and frequency sequences for fast fluctuation components; Determine the transient vibration shock index based on the amplitude sequence; If the transient vibration impact index exceeds the preset threshold, a high transient mechanical stress event is marked as occurring at the current moment. Extracting cumulative change trend sequences from slowly drifting components; Calculate the long-term structural degradation index based on the cumulative change trend sequence; By comparing the long-term structural degradation index with a pre-set fatigue risk classification threshold, the current level of structural fatigue damage can be obtained.

[0010] In one aspect of the invention, the vibration risk coupling model is established, the input of which is a transient vibration impact index and a long-term structural degradation index. The model outputs a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift. The transient vibration shock index and long-term structural degradation index at the current moment are obtained, and the Z-score algorithm is used for normalization to obtain standardized features. A baseline risk function containing linear weighting terms and nonlinear interaction terms is constructed, and preliminary risk coefficients are calculated based on standardized features; The standardized features are input into a support vector regression model that has been trained using a dataset of historical operation and fault records to obtain a comprehensive vibration risk coefficient that couples vibration risks at two time scales.

[0011] In one aspect of the invention, the step of obtaining the current comprehensive vibration risk coefficient of each unit participating in AGC allocation and comparing it with a preset safe operation threshold for the unit to determine whether the unit is in a high vibration risk state includes: The current comprehensive vibration risk coefficient data is obtained from each unit participating in automatic power generation control through the data acquisition system; Based on the obtained comprehensive vibration risk coefficient, it is compared with the preset safe operation threshold, and the difference between the two is calculated. If the calculated difference exceeds the preset safety range, the unit is determined to be in a high vibration risk state. If the difference is within the safe range, it is considered to be in a normal state; For units identified as being in a high vibration risk state, obtain their historical operating data records of vibration risk state changes and conduct trend analysis; Based on the results of trend analysis, the duration and magnitude of the high vibration risk state are determined, and dynamic output constraints for the unit are generated. The generated unit output constraints are transmitted to the optimization algorithm module as safety boundary limits for subsequent objective function optimization calculations.

[0012] In one aspect of the invention, if a generating unit is determined to be in a high vibration risk state, then when the unit participates in load allocation calculation, an optimization algorithm based on vibration risk constraints is adopted. This algorithm introduces a comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficients and the total water consumption of all participating generating units, while satisfying the total load command of the power grid. This includes: Obtain real-time total load commands from the power grid, as well as operating status data, operating head, and current comprehensive vibration risk coefficient of each generating unit; A mathematical model for an optimization algorithm based on vibration risk constraints is constructed, which includes constraints and an objective function; The constraints include power balance equation constraints, upper and lower limit inequality constraints for the output of each unit, unit ramp rate limitation constraints, and vibration risk threshold constraints for high-risk units. The objective function is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units. In the objective function, the comprehensive vibration risk prediction value of each unit's output is multiplied by the risk penalty weight as a safety penalty item. By using the constructed mathematical model to limit the feasible domain space of load allocation, the optimal boundary is set for the comprehensive vibration risk of each unit in different output ranges.

[0013] In one aspect of the invention, the optimization algorithm based on vibration risk constraints is used to obtain a set of unit output commands that meet the total load requirements. This set of output commands, while meeting grid demands, proactively avoids load allocation schemes that may induce high-risk vibrations, including: The mean values ​​of rapid fluctuation components and slow drift components in different steady-state load ranges are extracted from the historical operating data of the unit and input into the vibration risk coupling model. An interpolation algorithm is used to generate a continuous risk prediction curve, thereby obtaining the risk prediction value corresponding to any output in real time during the optimization solution. The mathematical model is solved iteratively using a sequential quadratic programming algorithm; In each iteration, the objective function value and gradient are calculated, and the output vector of each unit is updated until the convergence condition is met. The converged optimal unit output vector is output as the set of output commands to meet the total load requirements; This set of commands is used as the final output command scheme and transmitted to each unit for execution through the system interface.

[0014] In one aspect of the invention, the step of issuing the calculated output commands of each unit to the unit control system for execution, and continuously monitoring the tilt angle timing data obtained in the first step during execution, forming a closed-loop control process from vibration monitoring and risk analysis to load allocation decision-making, includes: Acquire unit tilt angle timing data; The acquired unit tilt angle time series data is input into the pre-established vibration risk coupling model to obtain the current comprehensive vibration risk coefficient; the comprehensive vibration risk coefficient is compared with the preset risk classification standard to determine whether the current vibration risk is low risk, medium risk or high risk; If the vibration risk is high, obtain historical and current power output data to determine the maximum load reduction range for each unit. If the vibration risk is classified as medium risk, historical and current power output data should be obtained to determine the appropriate load reduction range for each unit. If the vibration risk is low, then the current output data should be maintained as the baseline. Using the sequential quadratic programming method, with the goal of minimizing the overall plant vibration risk and water consumption, the maximum load reduction or moderate load reduction is used as the downward dynamic boundary of the current unit output. The safe allowable output range of each unit is updated as a constraint condition, and the target output value of each unit is calculated. The target output value of each unit is converted into an output command and sent to the unit control system through the communication interface; After the command is issued, continue to acquire a new round of tilt timing data; The new round of tilt angle time series data is input into the pre-established vibration risk coupling model to obtain a new comprehensive vibration risk coefficient; Based on the new comprehensive vibration risk coefficient, the risk assessment, output range determination, and load distribution calculation are repeatedly performed to form a continuous closed-loop adjustment process.

[0015] In another aspect, the present invention also relates to an AGC load distribution system for a hydroelectric power station that takes into account vibration risk, the system comprising: The data acquisition and preprocessing module is used to acquire real-time tilt angle time series data of the unit bearing housing and guide vane support. The real-time tilt angle time series data is processed by preset high-pass and low-pass filters. The high-pass filter output represents the rapid fluctuation component that changes from the second to the minute level, and the low-pass filter output represents the slow drift component that changes from the hour to the day level. The vibration feature extraction module is used to calculate the transient vibration impact index and the long-term structural degradation index based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, respectively. The transient vibration impact index reflects the instantaneous mechanical stress level caused by the current load adjustment, and the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation. The risk coupling modeling module is used to establish a vibration risk coupling model. The input of the model is the transient vibration impact index and the long-term structural degradation index. The model outputs a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift. The risk status judgment module is used to obtain the current comprehensive vibration risk coefficient of each unit participating in AGC allocation, and compare it with the preset safe operation threshold of the unit to determine whether the unit is in a high vibration risk state. The optimization algorithm module is used to use an optimization algorithm based on vibration risk constraints when a unit is judged to be in a high vibration risk state and participates in the load allocation calculation. The algorithm introduces a comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units under the premise of satisfying the total load command of the power grid. The load allocation solution module is used to solve the problem using an optimization algorithm based on vibration risk constraints, and obtain a set of output commands for each unit that meet the total load requirements. This set of output commands not only meets the grid demand, but also actively avoids load allocation schemes that may cause high-risk vibrations. The closed-loop control execution module is used to send the calculated output commands of each unit to the unit control system for execution, and continuously monitor the tilt angle timing data obtained in the first step during the execution process, forming a closed-loop control process from vibration monitoring, risk analysis to load allocation decision-making.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention acquires real-time tilt angle time-series data of the unit's bearing housing and guide vane support. High-pass and low-pass filters are used to separate components representing rapid fluctuations at the second to minute level and slow drifts at the hour to day level, respectively. Based on this, a transient vibration impact index reflecting instantaneous mechanical stress levels and a long-term structural degradation index reflecting the cumulative risk of structural fatigue are calculated. A vibration risk coupling model is constructed to generate a comprehensive vibration risk coefficient, thereby quantifying the superposition effect of vibration risks at the two time scales. For each unit participating in AGC allocation, this invention compares the comprehensive vibration risk coefficient with a safety threshold to determine if it is in a high-risk vibration state. If it is in a high-risk state, an optimization algorithm with the comprehensive vibration risk coefficient as a penalty term is introduced into the load allocation calculation. Under the premise of strictly meeting the total grid load command, the sum of the comprehensive vibration risk coefficients of all units is minimized to obtain an optimized output command to avoid high-risk vibrations, which is then issued and executed. Simultaneously, tilt angle data is continuously monitored to form a closed-loop control. This invention effectively solves the risk of excessive unit vibration and structural damage caused by the combined effects of transient shocks and long-term cumulative degradation during the AGC process of hydropower stations due to frequent and drastic load changes, and achieves coordinated optimization of safety, economy and power grid demand. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an AGC load allocation method for a hydropower station that takes vibration risk into account, according to the present invention.

[0019] Figure 2 This is a schematic diagram of an AGC load allocation method for hydropower stations that takes vibration risk into account, according to the present invention.

[0020] Figure 3 This is another schematic diagram of an AGC load allocation method for hydropower stations that takes vibration risk into account, according to the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0022] Please see Figures 1-3 As shown, this embodiment discloses a method and system for AGC load allocation in a hydropower station that considers vibration risk. The method may specifically include: S101. Acquire real-time tilt angle time-series data of the unit bearing housing and guide vane support. Process the real-time tilt angle time-series data using preset high-pass and low-pass filters. The high-pass filter output represents the rapid fluctuation component changing from the second to the minute level, and the low-pass filter output represents the slow drift component changing from the hour to the day level.

[0023] Acquire real-time tilt angle time-series data of the unit bearing housing and guide vane support. Process the real-time tilt angle time-series data using a preset high-pass filter to obtain the fast fluctuation component. Process the real-time tilt angle time-series data using a preset low-pass filter to obtain the slow drift component. Calculate the standard deviation for each minute of the fast fluctuation component to obtain a minute-level fluctuation intensity sequence. If three consecutive points in the minute-level fluctuation intensity sequence exceed a preset fluctuation threshold, it is identified as a fast anomaly event, and the occurrence time of this event is recorded. For the slow drift component, sample values ​​are taken once per hour to obtain an hourly drift sequence. Calculate the difference between adjacent sampling points based on the hourly drift sequence to obtain a drift change rate sequence. If the absolute value of a point in the drift change rate sequence exceeds a preset drift rate threshold, it is identified as a slow drift anomaly, and the occurrence time and direction of change of this anomaly are recorded. Compare the occurrence time of the fast anomaly event with the occurrence time of the slow drift anomaly. If the time difference between the two is less than a preset time window, it is identified as a composite anomaly event. Output the time, location, and type of all composite anomaly events.

[0024] Specifically, in the hydro-generator unit operation monitoring system, real-time tilt angle time-series data is first acquired using high-precision MEMS tilt sensors (sampling rate 100Hz, range ±5.000°, resolution 0.001°) installed on the bearing housing and guide vane support. Then, digital filtering is applied to the original sequences: a high-pass Butterworth filter (4th order) with a cutoff frequency of 0.008Hz (corresponding to a period of approximately 125 seconds) is used, and zero-phase bidirectional filtering is performed using the `filtfilt` function to obtain the fast fluctuation component θ_fast(t), reflecting transient vibrations from the second to the minute level; a low-pass Butterworth filter (4th order) with a cutoff frequency of 0.000278Hz (corresponding to a period of approximately 1 hour) is also used, with zero-phase processing, to obtain the slow drift component θ_slow(t), reflecting the trend drift from the hour to the day level.

[0025] The method for determining the cutoff frequencies of the high-pass and low-pass filters is as follows: Historical tilt time-series data of similar generating units at the target power plant under different operating conditions are retrieved. Fast Fourier Transform (FFT) is used to obtain the overall power spectral density distribution of the signal. Analysis of the power spectral density reveals that the transient impact energy caused by sudden load changes is mainly concentrated in the high-frequency region, while the drift energy caused by the slow release of structural stress is concentrated in the extremely low-frequency region. A bimodal threshold segmentation algorithm is used to calculate the energy minimum point interval between the two peaks of the power spectrum. The lower boundary frequency of the high-frequency peak is set as the cutoff frequency of the high-pass filter, and the upper boundary frequency of the extremely low-frequency peak is set as the cutoff frequency of the low-pass filter. For example, through FFT spectrum analysis of 1000 hours of historical data, the high-frequency impact energy is mainly concentrated above 0.01Hz, and the low-frequency energy is concentrated below 0.0002Hz. Through optimization of the energy minimum, the high-pass cutoff frequency is determined to be 0.008Hz, and the low-pass cutoff frequency is 0.000278Hz.

[0026] For the fast fluctuation component θ_fast(t), the standard deviation σ_min within each minute is calculated to obtain a minute-level fluctuation intensity sequence. A fluctuation threshold is set based on historical normal operation data. =0.02° (corresponding to 3 times the upper limit of normal fluctuation), if σ_min exceeds for three consecutive minutes If the event occurs, it is determined to be a fast abnormal event, and the time of the event is recorded as t_fast.

[0027] The above-mentioned preset fluctuation threshold The method for determining the values ​​is as follows: Based on the tilt angle time series dataset of the target hydropower station units operating under rated conditions without vibration alarm records for the past 36 months, the sliding window method is used to calculate the rapid fluctuation standard deviation of each minute-level window. A Gaussian mixture model (GMM) is applied to fit the distribution of all standard deviation data, and the mean of the fitted curve is extracted. with standard deviation ,use The theoretical upper limit of normal fluctuations is calculated in principle. And this theoretical upper limit is used as the preset fluctuation threshold. For example, the mean standard deviation of minute-level fluctuations can be obtained by fitting historical fault-free data. for Standard deviation for Then the preset fluctuation threshold is calculated. .

[0028] For the slow drift component θ_slow(t), a sample value is taken every hour (at the top of the hour) to obtain the hourly drift sequence θ_hour[k]. The difference between adjacent sample points Δθ_hour[k] = θ_hour[k] - θ_hour[k-1] is calculated to obtain the drift rate sequence (unit: ° / h). A drift rate threshold is set according to the equipment installation specifications. =0.01° / h. If |Δθ_hour[k]|>Δθ_th at a certain point, it is judged as a slow drift anomaly. Record the time t_slow when the anomaly occurs and the direction of change.

[0029] The aforementioned preset drift rate threshold The method for determining the value is as follows: retrieve the hourly drift sequence dataset of normal operation during the historical life cycle of the same type of unit, calculate the absolute value of the difference between all adjacent sampling points, fit the probability density of this absolute value set using a Weibull distribution, select the value corresponding to the cumulative probability reaching 99% as the theoretical drift upper limit, and set this theoretical drift upper limit as the preset drift rate threshold. For example, through Weibull fitting analysis of 50,000 hours of historical normal operation data, the drift rate corresponding to a cumulative probability of 99% was determined to be... If so, then this value will be used as the threshold.

[0030] The time of the fast anomaly event (t_fast) is compared with the time of the slow drift anomaly (t_slow). If the time difference between the two is less than a preset time window (e.g., 1 hour), it is determined to be a composite anomaly event, indicating that there is coupling between transient impact and long-term drift. The time, location, and type of all composite anomalies are output.

[0031] The method for determining the preset time window is as follows: retrieve data on all composite fault cases caused by transient impacts leading to long-term structural deformation that have occurred throughout the unit's historical life cycle, statistically analyze the time interval between the rapid abnormal event and the subsequently monitored slow drift abnormality in each case, fit the probability distribution curve of the time interval using the kernel density estimation (KDE) method, and select the time length corresponding to a cumulative probability of 95% as the preset time window; for example, by analyzing 50 historical fault data, the kernel density estimation results show that 95% of the fault cascade reaction occurs within 50 minutes after the median time, add a 10-minute safety assessment margin, and finally calculate the preset time window to be 60 minutes.

[0032] It should be noted that the composite abnormal event information is transmitted to the risk status assessment module in real time. If a unit experiences a composite abnormal event within the current preset time window, the risk status assessment module will forcibly set the comprehensive vibration risk coefficient of the unit to be no less than the high-risk judgment threshold, thereby immediately triggering preventive load constraints.

[0033] S102. Based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, calculate the transient vibration impact index and the long-term structural degradation index respectively.

[0034] The specific steps of S102 are as follows: The vibration signal sequence is acquired. The fast fluctuation component and the slow drift component are separated using the high-pass and low-pass filters, respectively. The amplitude and frequency sequences are calculated for the fast fluctuation component. The transient vibration impact index is determined based on the amplitude sequence. If the transient vibration impact index exceeds a preset threshold, a high instantaneous mechanical stress event is identified at the current moment. The cumulative trend sequence is extracted for the slow drift component. The long-term structural degradation index is calculated based on the cumulative trend sequence. The long-term structural degradation index is compared with a preset fatigue risk classification threshold to determine the current level of structural fatigue damage. The method for determining the aforementioned preset fatigue risk classification thresholds is as follows: Based on the annual maintenance logs and metal flaw detection reports of similar hydro-generator units, historical values ​​of the actual long-term structural degradation index of different units within the maintenance cycle are collected and calculated. The 99th percentile of the index of the sample group with no abnormalities detected by flaw detection is set as the upper limit threshold for low risk; the median of the index of the sample group with minor cracks but allowed to continue operation is set as the upper limit threshold for medium risk; and the lower boundary of the index of the sample group with severe damage requiring immediate shutdown and component replacement is set as the starting threshold for high risk. For example, statistical analysis shows that the historical degradation index extreme value of the unit with no abnormalities is... The median index of the microcrack unit is The lower limit of the index for severely damaged units is Therefore, the fatigue risk classification threshold is extracted and set accordingly. , as well as The three gradient parameters.

[0035] Specifically, firstly, the fast fluctuation component θ_fast(t) output by the high-pass filter is extracted, and its amplitude range is typically between 0.02° and 0.15°. The instantaneous amplitude A(t) of θ_fast(t) is obtained using the Hilbert transform, and the peak value A_peak and effective value A_rms over the most recent 5 minutes are calculated. Then, the transient vibration shock index I_imp is calculated using the following formula:

[0036] in, The reference amplitude is used to make the second term dimensionless. Its value is determined based on the statistical average value during the unit's historical normal operation. For example, in this embodiment, it is taken as 0.05° based on historical data.

[0037] The above reference amplitude The method for determining the value is as follows: Extract the instantaneous amplitude sequence of rapid fluctuation components under all steady-state full-load operating conditions of the hydropower station unit in the past year. After removing transient shock outliers, use the kernel density estimation (KDE) method to calculate the probability distribution density function of the remaining steady-state amplitude data. Extract the mode amplitude corresponding to the point with the highest probability density as the reference amplitude. For example, after kernel density estimation, the probability density distribution curve of historical steady-state amplitude data is... If the peak value is reached, then the reference amplitude is determined. for .

[0038] For example, when =0.12° When the angle is 0.04°, the calculation yields:

[0039] If I_imp exceeds the preset threshold I_th=15.0, then a high instantaneous mechanical stress event is marked at the current moment.

[0040] The above-mentioned preset threshold The method for determining the value is as follows: Vibration data from the 10 minutes prior to a known transient mechanical overload fault such as shear pin shearing or guide vane arm deformation during the unit's historical lifecycle is retrieved as the fault sample set. Simultaneously, vibration data from the safe operation period is retrieved as the safe sample set. The transient vibration impact index of the two types of samples is calculated separately. The principle of maximum information entropy is used to find the optimal split point that maximizes the classification information gain between the safe and fault samples. The value of this split point is shifted 10% towards the safe zone and set as the preset threshold. For example, the exponential distribution range of historical failure samples is 17.2 to 25.8, and the distribution range of safe samples is 3.1 to 14.5. The calculated maximum information entropy segmentation point is 16.7. After adjusting it by 10%, the preset threshold is obtained. It is 15.0.

[0041] For the slow drift component θ_slow(t), its cumulative change over the past 24 hours is extracted as Δθ = θ_slow(t) - θ_slow(t-24h). Using the structural mechanics model, the tilt angle change is converted into an equivalent strain ε = α·Δθ, where α is the structural conversion coefficient (taken based on the finite element analysis and unit geometry). Based on the fatigue characteristics of materials, the equivalent cyclic stress amplitude Δσ = E·ε, where E is the elastic modulus of the material (2.1 × 10⁻⁶ for steel). 5 MPa). Using Miner's linear cumulative damage theory, the long-term structural degradation index D is calculated as follows:

[0042] in, For the first The actual number of cycles at the stress level. This represents the fatigue life limit number of cycles at this stress level. m and C are material fatigue characteristic constants, derived from the fitting of fatigue test data of bearing housing materials; n is the equivalent stress cycle number in 24 hours (usually taken as 1, representing daily cumulative damage), which is determined by fitting the SN fatigue curve test results of bearing housing materials. , For example, when hour, , ,but The current level of fatigue damage in the structure is obtained by comparing D with the preset fatigue risk classification thresholds (such as D_low=1e-5, D_mid=1e-4, D_high=1e-3).

[0043] The above structural conversion coefficients The method for determining the values ​​is as follows: Based on the 3D CAD drawings of the unit's guide vane support and bearing housing, a finite element simulation model of the unit entity with fine meshing is constructed. Ten sets of stepped virtual angular displacement constraints with different gradients (from 0.01° to 0.1°) are applied to the model to simulate the actual physical tilt angle deformation. The maximum principal strain value of the key stress section of the bearing housing under each virtual tilt angle is calculated using a finite element structural statics solver. The least squares method is used to perform linear regression fitting on the tilt angle values ​​and strain results of all sets, and the slope value of the fitted line is extracted as the structural conversion coefficient. For example, finite element simulation results show that at a tilt angle of 0.05°, the strain at the critical section is 1.74 × 10⁻⁶. -6 After regression fitting of multiple sets of data, a slope of 0.002 was obtained, and the structural transformation coefficient was determined. 0.002 rad -1 .

[0044] S103. Establish a vibration risk coupling model. The inputs of the model are the transient vibration impact index and the long-term structural degradation index. The model outputs a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift.

[0045] The transient vibration shock index and long-term structural degradation index at the current moment are obtained, and the Z-score algorithm is used for normalization to obtain standardized features. A baseline risk function containing linear weighting terms and nonlinear interaction terms is constructed, and preliminary risk coefficients are calculated based on standardized features; The standardized features are input into a support vector regression (SVR) model that has been trained using a dataset of historical operation and failure records to obtain a comprehensive vibration risk coefficient that couples the vibration risks of the two time scales.

[0046] Specifically, the raw data pair (I_imp, D) is obtained by real-time acquisition of the transient vibration impact index I_imp and the long-term structural degradation index D. Z-score standardization is used to normalize both, yielding standardized features I' and D'. A preliminary risk coefficient R0, combining linear weighting and nonlinear interaction terms, is then established.

[0047] The weighting coefficients w1, w2, and w3 were obtained by fitting historical sample data using the least squares method, with the constraints w1 + w2 + w3 = 1 and w_i ≥ 0. The aforementioned weighting coefficients... , and The method for determining the value is as follows: retrieve the operational characteristic records of vibration risk events of different degrees that have been clearly identified in the historical life cycle of the unit, and extract the standardized transient impact characteristics corresponding to each event. and structural degradation characteristics As the independent variable matrix, and simultaneously using the baseline risk coefficient value, assessed by domain experts based on the degree of equipment damage after the corresponding event, as the dependent variable, a multiple linear regression equation is established. The model parameters are fitted using the least squares method with non-negativity constraints. The combination of model coefficients that minimizes the sum of squared prediction errors is solved, and then normalized to obtain the final weighted coefficients. For example, extracting 150 sets of historical risk event data for least squares fitting yields original coefficient combinations of 0.52, 0.31, and 0.21, which are then normalized to determine the final weighted coefficients. , , .

[0048] To capture the complex coupling effect of vibration risk at two time scales, this embodiment uses a support vector regression (SVR) model to directly establish a mapping from standardized features (I', D') to the comprehensive vibration risk coefficient R.

[0049]

[0050] in, This is the trained SVR model. The training dataset for the model comes from historical operation records: the input is the standardized features (I',D') at each historical moment, and the output is the risk level label (within the range of [0,1], where 0 represents no risk and 1 represents the highest risk) marked by experts based on the post-incident inspection results and vibration event records. The SVR model uses a radial basis function kernel. The method for determining the penalty parameter C and kernel parameter γ in the aforementioned Support Vector Regression (SVR) model is as follows: The dataset containing historical standardized features and corresponding risk level labels is divided into training and validation sets in an 8:2 ratio. The search interval for the penalty parameter C is set to [0.1, 100] and the search interval for the kernel parameter γ is set to [0.01, 10]. A ten-fold cross-validation method is used to perform a grid search within the set search intervals, calculating the mean squared error (MSE) of the validation set for each parameter combination. The parameter combination that minimizes the MSE of the validation set is selected as the final hyperparameters of the model. For example, the grid search results show that when parameter C=10 and γ=0.5, the MSE of the validation set reaches the global minimum value of 0.024. Therefore, the penalty parameter C=10 and the kernel parameter γ=0.5 are determined for this SVR model. The final output R is the quantified comprehensive vibration risk coefficient; a larger value indicates a higher vibration risk.

[0051] For example, at a certain moment, after standardization, I'=0.75 and D'=0.42. Substituting these values, we get R0=0.5×0.75+0.3×0.42+0.2×0.75×0.42=0.375+0.126+0.063=0.564. After SVR calibration, R=0.592, indicating that the risk is at a moderately high level.

[0052] S104. For each unit participating in AGC allocation, obtain its current comprehensive vibration risk coefficient and compare it with the preset safe operation threshold of the unit to determine whether the unit is in a high vibration risk state.

[0053] The data acquisition system obtains the current comprehensive vibration risk coefficient data from each unit participating in automatic power generation control. Based on the obtained comprehensive vibration risk coefficient, it is compared with a preset safe operating threshold, and the difference between the two is calculated. If the calculated difference exceeds the preset safe range, the unit is determined to be in a high vibration risk state; if the difference is within the safe range, it is determined to be in a normal state. For units determined to be in a high vibration risk state, the vibration risk state change records in their historical operating data are obtained, and trend analysis is performed. Based on the results of the trend analysis, the duration and magnitude of the high vibration risk state are determined, and dynamic output constraints for the unit are generated. The generated unit output constraints are transmitted to the optimization algorithm module as safety boundary limits for subsequent objective function optimization calculations.

[0054] Specifically, the comprehensive vibration risk coefficient R is obtained in real time from each unit participating in Automatic Generation Control (AGC) through the hydropower station's computer monitoring system. Preset safe operation thresholds are divided into three levels: R < 0.3 indicates low risk, 0.3 ≤ R < 0.6 indicates medium risk, and R ≥ 0.6 indicates high risk. The current R value is compared with the threshold; if R ≥ 0.6, the unit is determined to be in a high vibration risk state; otherwise, it is in a medium or low risk state.

[0055] The method for determining the aforementioned preset safe operation threshold is as follows: extract historical time-series data of the comprehensive vibration risk coefficient of all units in the power plant over the past five years, and simultaneously match the current operating conditions and equipment life loss rate. Use the K-Means++ clustering algorithm to perform unsupervised clustering of the two-dimensional data points of the risk coefficient and the corresponding loss rate, set the number of clusters to three, calculate the optimal decision boundary between adjacent clusters as the support vector classification surface, and use the risk coefficient value corresponding to the classification surface as the dividing point of the safe operation threshold. For example, the clustering algorithm results show that the upper boundary of the risk coefficient of the first cluster representing the low-risk state is 0.3, the data of the second cluster representing the medium-risk state is distributed between 0.3 and 0.6, and the lower boundary of the data of the third cluster representing the high-risk state is 0.6. Thus, 0.3 and 0.6 are determined as the critical points for determining the safe operation threshold.

[0056] For example, a 100MW mixed-flow turbine generator unit currently has an R=0.475, classifying it as a medium-risk unit. The system automatically retrieves the unit's historical vibration risk data from the past 24 hours and analyzes its trend: the R value gradually increased from 0.32 to 0.475 over the past 6 hours, a change of 0.155 over a duration of 6 hours. Based on the trend analysis, its risk level is classified as medium-risk. For different levels of high-risk units, corresponding operational adjustment strategy suggestions are generated, such as "limiting the load change rate to no more than 5% / min" or "recommending checking the guide vane bearing clearance during the next maintenance." The suggested data is transmitted to the AGC control system via a communication interface, enabling dynamic monitoring and response to the unit's status.

[0057] The method for determining the load change rate limit threshold is as follows: retrieve the historical AGC regulation test dataset of the target power station under different reservoir water level conditions, calculate and extract the load change rate of the unit and the corresponding transient vibration impact index during each command regulation process, apply the support vector regression algorithm to construct a nonlinear mapping model between the load change rate and the transient vibration increment, and under the boundary condition that the additional vibration increment does not trigger the moderate risk warning, back-calculate the maximum load change rate that the model can allow and set it as the load change rate limit threshold; for example, the model calculation shows that when the load change rate reaches 6% / min, the predicted transient vibration is very likely to exceed the 0.3 concern limit, while the vibration increment is stable when limited to 5% / min. Therefore, the load change rate limit threshold is determined to be 5% / min.

[0058] S105. If a generating unit is determined to be in a high vibration risk state, then when the unit participates in the load allocation calculation, an optimization algorithm based on vibration risk constraints is adopted. This algorithm introduces a comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating generating units while satisfying the total load command of the power grid. The specific model is as follows:

[0059] in, β is the water consumption rate coefficient (m³ / kWh) of the i-th unit, which is determined by the current head and the unit efficiency curve. This item represents the power generation cost; β is the risk penalty weight, which is set by the operators according to safety requirements (10 in this example) to balance economy and safety. This represents the comprehensive vibration risk coefficient when the unit output is Pi.

[0060] The above risk penalty weights The method for determining the values ​​is as follows: Based on Pareto front analysis of multi-objective optimization of the units, a series of test weight values ​​of different levels are set. Each test weight value is substituted into the optimization objective function for multiple rounds of offline simulation. The total water consumption of the power plant and the total comprehensive vibration risk of all units in the plant are recorded in each solution. A Pareto curve of water consumption and total risk is plotted. The weight value corresponding to the inflection point on the curve that maximizes the rate of decrease in total risk and keeps the rate of increase in water consumption below 2% is selected as the risk penalty weight. For example, offline simulations show that when the test weight value increases from 5 to 10, the total risk decreases by 35% while water consumption only increases by 0.8%. When the weight continues to increase to 15, water consumption increases sharply by 4%. Therefore, the inflection point value of 10 is selected as the risk penalty weight. .

[0061] The system acquires real-time total load commands from the power grid, as well as operating status data, operating head, and current comprehensive vibration risk coefficients for each generating unit. It then constructs a mathematical model for an optimization algorithm based on vibration risk constraints, including constraints and an objective function. These constraints include power balance equation constraints, inequality constraints on the upper and lower limits of each generating unit's output, unit ramp rate limits, and vibration risk threshold constraints for high-risk units. The objective function minimizes the sum of the comprehensive vibration risk coefficients and total water consumption of all participating generating units. The predicted comprehensive vibration risk value of each generating unit's output is multiplied by a risk penalty weight as a safety penalty term. The constructed mathematical model is used to define the feasible domain space for load allocation, and optimization boundaries are set for the comprehensive vibration risk of each generating unit in different output ranges.

[0062] Specifically, data such as the current output P_i, comprehensive vibration risk coefficient R_i, and operating head H_i of each of the N generating units participating in load sharing are obtained. An optimization model is constructed with the objective of minimizing the total vibration risk while also considering the power generation water consumption rate.

[0063] Where α_i is the water consumption rate coefficient (m³ / kWh) of the i-th unit, determined by the current head and unit efficiency curve; β is the risk penalty weight, set by the operator according to safety requirements (10 in this example); R_i(P_i) represents the comprehensive vibration risk coefficient when the unit output is P_i (obtainable through interpolation of the risk curve). Constraints include: 1. Power balance: , This is the total load command for the power grid; 2. Upper and lower limits of output: ; 3. Gradient Limitation: ; in, and These are the lower and upper limits of the safe permissible output of the i-th generating unit, respectively; This represents the actual output of the i-th unit in the previous control cycle; This represents the maximum allowable gradient for the unit within a single control cycle.

[0064] Extract the mean values ​​of rapid fluctuation components and slow drift components from the historical operating data of the unit in different steady-state load ranges (with a step size of 5MW); input the historical mean values ​​of each load range into the trained SVR vibration risk coupling model to calculate the corresponding benchmark comprehensive risk coefficient; and use the target output as the benchmark. With the independent variable being the baseline comprehensive risk coefficient and the dependent variable being the benchmark comprehensive risk coefficient, a continuous risk prediction curve is generated using a cubic spline interpolation algorithm, thereby enabling the real-time acquisition of arbitrary output commands during optimization iterations. corresponding Predicted value.

[0065] The above nonlinear programming problem is solved using the Sequential Quadratic Programming (SQP) algorithm. Taking a total load demand increase of 50MW as an example, the participating units are A (risk coefficient 0.2), B (0.3), and C (0.65, high risk). After optimization, the high-risk unit C only bears a small portion of the load increase (e.g., 5MW), with the remainder shared by A and B. This reduces the sum of the total risk coefficients from 1.15 (unconstrained) to 0.85, while simultaneously meeting the total load requirement.

[0066] S106. An optimization algorithm based on vibration risk constraints is used to solve the problem and obtain a set of output commands for each unit that meet the total load requirements. This set of output commands not only meets the grid demand but also actively avoids load allocation schemes that may cause high-risk vibrations.

[0067] The mean values ​​of rapid fluctuation components and slow drift components within different steady-state load ranges are extracted from the historical operating data of the generating units and input into the vibration risk coupling model. An interpolation algorithm is used to generate continuous risk prediction curves, thereby obtaining the risk prediction value corresponding to any output in real time during the optimization solution. A sequential quadratic programming algorithm is used to iteratively solve the mathematical model. In each iteration, the objective function value and gradient are calculated, and the output vector of each unit is updated until the convergence condition is met. The optimal unit output vector after convergence is output as the set of output commands that meet the total load requirements. This set is used as the final output command scheme and transmitted to each unit for execution through the system interface.

[0068] Specifically, based on the optimization model established in step S105, the real-time total load command P_total of the power grid and the current operating status of each unit are obtained, and the SQP algorithm is called to iteratively solve the problem. Algorithm initialization: Assume that the initial output of each unit is the current actual output, the iteration step size is 0.1MW, and the convergence accuracy is that the change in the objective function is less than 10. -4 In each iteration, the objective function value and gradient are calculated, and the output force vector is updated until the KKT conditions are met.

[0069] For example, a hydropower station has 4 generating units with a total load command P_total = 320MW. The upper and lower limits of the output of each unit are [40,100]MW, [50,110]MW, [30,90]MW, and [60,120]MW, respectively, and the current actual output is [80,75,85,80]MW. According to step S104, the current risk coefficient R3 of unit 3 is 0.68 (high risk), and the others are all less than 0.4. Optimization process: First, the risk threshold constraint is checked, and the upper limit of the output of unit 3 is dynamically adjusted to 75MW (to forcibly suppress the expansion of risk); after 15 iterations, the optimal output command is obtained as [85,88,65,82]MW, and the total is still 320MW. Under this scheme, the risk coefficients for each unit are 0.24, 0.31, 0.52, and 0.29, respectively. The output of high-risk unit 3 is significantly reduced (from 85MW to 65MW), successfully removing it from the high-risk zone, and the overall risk is optimally controllable. The instructions are issued to the unit's LCU for execution via the IEC104 protocol.

[0070] S107. The calculated output commands of each unit are sent to the unit control system for execution, and the tilt angle timing data obtained in the first step is continuously monitored during the execution process to form a closed-loop control process from vibration monitoring and risk analysis to load allocation decision.

[0071] Acquire unit tilt angle time-series data. Input the acquired unit tilt angle time-series data into the pre-established vibration risk coupling model to obtain the current comprehensive vibration risk coefficient. Compare the comprehensive vibration risk coefficient with the preset risk classification standard to determine whether the current vibration risk is low, medium, or high. If the vibration risk is high, acquire historical output data and current output data to determine the maximum load reduction range for each unit. If the vibration risk is medium, acquire historical output data and current output data to determine the appropriate load reduction range for each unit. If the vibration risk is low, maintain the current output data as the baseline. Using a sequential quadratic programming method, with the minimization of the overall plant vibration risk and water consumption as the objective function, use the maximum load reduction range or appropriate load reduction range as the downward dynamic boundary of the current unit output, update the safe allowable output range of each unit as a constraint, and calculate the target output value for each unit. Convert the target output value of each unit into an output command and send it to the unit control system through the communication interface. After the command is sent, continue to acquire a new round of tilt angle time-series data. The new round of tilt angle time series data is input into the pre-established vibration risk coupling model to obtain a new comprehensive vibration risk coefficient. Based on the new comprehensive vibration risk coefficient, the risk assessment, output range determination, and load distribution calculation are repeatedly performed to form a continuous closed-loop adjustment process.

[0072] The method for determining the appropriate load reduction range for each of the above-mentioned generating units is as follows: retrieve historical operating data of similar hydro-generator units during previous entries into the moderate vibration risk range, extract effective load reduction operation records that enable the units to smoothly leave the resonance zone or abnormal fluctuation zone within 15 minutes, and use a Gaussian mixture model (GMM) to fit the probability distribution of all effective load reduction percentages to rated output, selecting the mean of the probability density curve as the appropriate load reduction range for each generating unit; for example, by analyzing 200 historical effective load reduction records, the fitting yields an effective load reduction percentage of 15% with a standard deviation of 2%, then the appropriate load reduction range for each generating unit is set to 15% of the rated output.

[0073] Specifically, the tilt angle time-series data of the bearing housings and guide vane supports of each unit are acquired in real time using a high-speed data acquisition module at a sampling frequency of 100Hz. Data windows every 30 seconds are input into the processing flow of steps S102-S103 to obtain the latest comprehensive vibration risk coefficient R_new. Based on R_new and the preset grading standard, the current vibration risk level is determined. If R_new≥0.6 (high risk), then an emergency adjustment is initiated: based on the current output and historical data, the maximum allowable load reduction for each unit is determined (e.g., a 30% reduction in rated output), and the optimization algorithm is re-initiated; The method for determining the maximum load reduction range of the above-mentioned units is as follows: retrieve the simulation data of the hydraulic transient process during the historical load shedding test and emergency shutdown test of the units, extract the extreme value data of the spiral casing water hammer pressure and tailrace vacuum under different load reduction ranges, and establish a mapping model between the load reduction range and the extreme water pressure using multivariate polynomial regression. Under the premise of ensuring that the extreme water pressure does not exceed the safety limit threshold of the water intake system, solve for the maximum load adjustment percentage that the mapping model can output, and use this as the maximum load reduction range. For example, the regression model calculation shows that when the load reduction range reaches 30% of the rated output, the tailrace vacuum is just close to the lower boundary of the allowable limit. Therefore, the maximum load reduction range is safely set to 30%.

[0074] If 0.3≤R_new<0.6 (medium risk), then preventive adjustments will be initiated: the load change rate of high-risk units will be appropriately limited, and their output will be reduced in the next round of AGC instructions. If R_new < 0.3 (low risk), then maintain the existing output allocation scheme.

[0075] A sequential quadratic programming method is employed, aiming to minimize the overall plant vibration risk and total water consumption. The determined maximum or moderate load reduction range is used as the downward dynamic boundary for the current output. The safe allowable output range of each unit is updated in real time as a constraint, and the target output value for each unit is recalculated, generating a new output command. After the command is issued via the communication interface, tilt angle data for the next cycle is collected, and the above steps are repeated, forming a continuous closed-loop adjustment process.

[0076] This invention provides an AGC load distribution system for hydropower stations that considers vibration risk, mainly comprising: The data acquisition and preprocessing module is used to acquire real-time tilt angle time series data of the unit bearing housing and guide vane support. The real-time tilt angle time series data is processed by preset high-pass and low-pass filters. The high-pass filter output represents the rapid fluctuation component that changes from the second to the minute level, and the low-pass filter output represents the slow drift component that changes from the hour to the day level. The vibration feature extraction module is used to calculate the transient vibration impact index and the long-term structural degradation index based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, respectively. The transient vibration impact index reflects the instantaneous mechanical stress level caused by the current load adjustment, and the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation. The risk coupling modeling module is used to establish a vibration risk coupling model. The input of the model is the transient vibration impact index and the long-term structural degradation index. The model outputs a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift. The risk status judgment module is used to obtain the current comprehensive vibration risk coefficient of each unit participating in AGC allocation, and compare it with the preset safe operation threshold of the unit to determine whether the unit is in a high vibration risk state. The optimization algorithm module is used to use an optimization algorithm based on vibration risk constraints when a unit is judged to be in a high vibration risk state and participates in the load allocation calculation. The algorithm introduces a comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units under the premise of satisfying the total load command of the power grid. The load allocation solution module is used to solve the problem using an optimization algorithm based on vibration risk constraints, and obtain a set of output commands for each unit that meet the total load requirements. This set of output commands not only meets the grid demand, but also actively avoids load allocation schemes that may cause high-risk vibrations. The closed-loop control execution module is used to send the calculated output commands of each unit to the unit control system for execution, and continuously monitor the tilt angle timing data obtained in the first step during the execution process, forming a closed-loop control process from vibration monitoring, risk analysis to load allocation decision-making.

[0077] If the technical solution of this application involves the collection, processing, or application of personal information, the relevant products have, in accordance with applicable laws and regulations, fully and clearly informed individuals of the processing rules and obtained their voluntary and explicit consent before implementing any personal information processing activities. If sensitive personal information is involved, the product has obtained the individual's separate consent before processing, and this consent is given in an explicit manner. For example, prominent signs are placed in the area where information collection devices such as cameras are located, clearly indicating "Entering is considered consent to the collection of personal information"; or through pop-ups, checkboxes, user-initiated uploads, etc., the user is required to actively complete the authorization operation with a clear list of the processor's identity, processing purpose, processing method, and information type. The above mechanisms ensure that all personal information processing activities are based on legal authorization and fully comply with national compliance requirements regarding personal information protection.

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for AGC load allocation in a hydropower station considering vibration risk, characterized in that, include: The real-time tilt angle time series data of the unit bearing housing and guide vane support are acquired. The real-time tilt angle time series data are processed by preset high-pass and low-pass filters to obtain the fast fluctuation component and the slow drift component, respectively. Based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, the transient vibration shock index and the long-term structural degradation index are calculated respectively. The transient vibration shock index reflects the instantaneous mechanical stress level caused by the current load adjustment, while the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation. A vibration risk coupling model is established. The inputs of the model are the transient vibration impact index and the long-term structural degradation index. The model outputs a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift. For each unit participating in AGC allocation, its current comprehensive vibration risk coefficient is obtained and compared with the preset safe operation threshold of the unit to determine whether the unit is in a high vibration risk state. If a unit is judged to be in a high vibration risk state, then when the unit participates in the load allocation calculation, an optimization algorithm based on vibration risk constraints is adopted. This algorithm introduces a comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units under the premise of meeting the total load command of the power grid. An optimization algorithm based on vibration risk constraints was used to solve the problem and obtain a set of power output commands for each unit that meet the total load requirements. This set of power output commands not only meets the grid demand but also actively avoids load allocation schemes that may cause high-risk vibrations. The calculated output commands for each unit are sent to the unit control system for execution, and the tilt angle timing data obtained in the first step is continuously monitored during the execution process, forming a closed-loop control process from vibration monitoring and risk analysis to load allocation decision-making.

2. The AGC load allocation method for hydropower stations considering vibration risk according to claim 1, characterized in that, The process of acquiring real-time tilt angle time-series data of the unit bearing housing and guide vane support involves processing the real-time tilt angle time-series data using preset high-pass and low-pass filters to obtain rapid fluctuation components and slow drift components, including: Acquire real-time tilt angle timing data of the unit bearing housing and guide vane support; Real-time tilt time-series data is processed using a preset high-pass filter to obtain fast fluctuation components; Real-time tilt time-series data is processed using a preset low-pass filter to obtain the slow drift component; The standard deviation of the rapid fluctuation component is calculated for each minute to obtain a minute-level fluctuation intensity sequence; If three consecutive points in the minute-level fluctuation intensity sequence exceed the preset fluctuation threshold, it is judged as a rapid abnormal event, and the time of occurrence of the event is recorded. For the slowly drifting component, a sample value is taken once per hour to obtain an hourly drift sequence; The difference between adjacent sampling points is calculated based on the hourly drift sequence to obtain the drift rate sequence; If the absolute value of a point in the drift rate sequence exceeds the preset drift rate threshold, it is judged as a slow drift anomaly, and the time and direction of the anomaly are recorded. The timing of a rapid anomaly is compared with the timing of a slow drift anomaly. If the time difference between the two is less than a preset time window, it is determined to be a composite anomaly. Output the time, location, and type of all compound exception events.

3. The AGC load allocation method for hydropower stations considering vibration risk according to claim 1, characterized in that, Based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, the transient vibration impact index and the long-term structural degradation index are calculated respectively. The transient vibration impact index reflects the instantaneous mechanical stress level caused by the current load adjustment, and the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation, including: Obtain the vibration signal sequence; The fast fluctuation component and the slow drift component are separated by the high-pass filter and the low-pass filter, respectively. Calculate the amplitude and frequency sequences for fast fluctuation components; Determine the transient vibration shock index based on the amplitude sequence; If the transient vibration impact index exceeds the preset threshold, a high transient mechanical stress event is marked as occurring at the current moment. Extracting cumulative change trend sequences from slowly drifting components; Calculate the long-term structural degradation index based on the cumulative change trend sequence; By comparing the long-term structural degradation index with a pre-set fatigue risk classification threshold, the current level of structural fatigue damage can be obtained.

4. The AGC load allocation method for hydropower stations considering vibration risk according to claim 1, characterized in that, The vibration risk coupling model is established, with transient vibration impact index and long-term structural degradation index as inputs. The model outputs a comprehensive vibration risk coefficient, which quantifies the interaction and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift. The transient vibration shock index and long-term structural degradation index at the current moment are obtained, and the Z-score algorithm is used for normalization to obtain standardized features. A baseline risk function containing linear weighting terms and nonlinear interaction terms is constructed, and preliminary risk coefficients are calculated based on standardized features; The standardized features are input into a support vector regression model that has been trained using a dataset of historical operation and fault records to obtain a comprehensive vibration risk coefficient that couples vibration risks at two time scales.

5. The AGC load allocation method for a hydropower station considering vibration risk according to claim 1, characterized in that, For each unit participating in AGC allocation, its current comprehensive vibration risk coefficient is obtained and compared with a preset safe operation threshold for the unit to determine whether the unit is in a high vibration risk state, including: The current comprehensive vibration risk coefficient data is obtained from each unit participating in automatic power generation control through the data acquisition system; Based on the obtained comprehensive vibration risk coefficient, it is compared with the preset safe operation threshold, and the difference between the two is calculated. If the calculated difference exceeds the preset safety range, the unit is determined to be in a high vibration risk state. If the difference is within the safe range, it is considered to be in a normal state; For units identified as being in a high vibration risk state, obtain their historical operating data records of vibration risk state changes and conduct trend analysis; Based on the results of trend analysis, the duration and magnitude of the high vibration risk state are determined, and dynamic output constraints for the unit are generated. The generated unit output constraints are transmitted to the optimization algorithm module as safety boundary limits for subsequent objective function optimization calculations.

6. The AGC load allocation method for a hydropower station considering vibration risk according to claim 1, characterized in that, If a generating unit is determined to be in a high vibration risk state, then when that unit participates in load allocation calculations, an optimization algorithm based on vibration risk constraints is adopted. This algorithm introduces a comprehensive vibration risk coefficient as a penalty term into the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficients and total water consumption of all participating generating units, while satisfying the total load command of the power grid. This includes: Obtain real-time total load commands from the power grid, as well as operating status data, operating head, and current comprehensive vibration risk coefficient of each generating unit; A mathematical model for an optimization algorithm based on vibration risk constraints is constructed, which includes constraints and an objective function; The constraints include power balance equation constraints, upper and lower limit inequality constraints for the output of each unit, unit ramp rate limitation constraints, and vibration risk threshold constraints for high-risk units. The objective function is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units. In the objective function, the comprehensive vibration risk prediction value of each unit's output is multiplied by the risk penalty weight as a safety penalty item. By using the constructed mathematical model to limit the feasible domain space of load allocation, the optimal boundary is set for the comprehensive vibration risk of each unit in different output ranges.

7. The AGC load allocation method for a hydropower station considering vibration risk according to claim 1, characterized in that, The optimization algorithm based on vibration risk constraints is used to solve the problem, resulting in a set of unit output commands that meet the total load requirements. This set of output commands, while satisfying grid demand, proactively avoids load allocation schemes that may trigger high-risk vibrations, including: The mean values ​​of rapid fluctuation components and slow drift components in different steady-state load ranges are extracted from the historical operating data of the unit and input into the vibration risk coupling model. An interpolation algorithm is used to generate a continuous risk prediction curve, thereby obtaining the risk prediction value corresponding to any output in real time during the optimization solution. The mathematical model is solved iteratively using a sequential quadratic programming algorithm; In each iteration, the objective function value and gradient are calculated, and the output vector of each unit is updated until the convergence condition is met. The converged optimal unit output vector is output as the set of output commands to meet the total load requirements; This set of commands is used as the final output command scheme and transmitted to each unit for execution through the system interface.

8. The AGC load allocation method for a hydropower station considering vibration risk according to claim 1, characterized in that, The calculated output commands for each unit are sent to the unit control system for execution, and the tilt angle time series data obtained in the first step is continuously monitored during the execution process, forming a closed-loop control process from vibration monitoring and risk analysis to load allocation decision-making, including: Acquire unit tilt angle timing data; The acquired unit tilt angle time series data is input into the pre-established vibration risk coupling model to obtain the current comprehensive vibration risk coefficient; Based on the comparison between the comprehensive vibration risk coefficient and the preset risk classification standard, the current vibration risk is determined to be low risk, medium risk or high risk. If the vibration risk is high, obtain historical and current power output data to determine the maximum load reduction range for each unit. If the vibration risk is classified as medium risk, historical and current power output data should be obtained to determine the appropriate load reduction range for each unit. If the vibration risk is low, then the current output data should be maintained as the baseline. Using the sequential quadratic programming method, with the goal of minimizing the overall plant vibration risk and water consumption, the maximum load reduction or moderate load reduction is used as the downward dynamic boundary of the current unit output. The safe allowable output range of each unit is updated as a constraint condition, and the target output value of each unit is calculated. The target output value of each unit is converted into an output command and sent to the unit control system through the communication interface; After the command is issued, continue to acquire a new round of tilt timing data; The new round of tilt angle time series data is input into the pre-established vibration risk coupling model to obtain a new comprehensive vibration risk coefficient; Based on the new comprehensive vibration risk coefficient, the risk assessment, output range determination, and load distribution calculation are repeatedly performed to form a continuous closed-loop adjustment process.

9. A hydropower station AGC load distribution system considering vibration risk, characterized in that, The system includes: The data acquisition and preprocessing module is used to acquire real-time tilt angle time series data of the unit bearing housing and guide vane support. It uses preset high-pass and low-pass filters to process the real-time tilt angle time series data to obtain fast fluctuation components and slow drift components respectively. The vibration feature extraction module is used to calculate the transient vibration impact index and the long-term structural degradation index based on the amplitude characteristics of the rapid fluctuation component and the cumulative trend of the slow drift component, respectively. The transient vibration impact index reflects the instantaneous mechanical stress level caused by the current load adjustment, and the long-term structural degradation index reflects the potential risk level of structural fatigue accumulation. The risk coupling modeling module is used to establish a vibration risk coupling model. The input of the model is the transient vibration impact index and the long-term structural degradation index. The model outputs a comprehensive vibration risk coefficient, which quantifies the mutual influence and superposition effect of vibration risks at two time scales: rapid fluctuation and slow drift. The risk status judgment module is used to obtain the current comprehensive vibration risk coefficient of each unit participating in AGC allocation, and compare it with the preset safe operation threshold of the unit to determine whether the unit is in a high vibration risk state. The optimization algorithm module is used to use an optimization algorithm based on vibration risk constraints when a unit is judged to be in a high vibration risk state and participates in the load allocation calculation. The algorithm introduces a comprehensive vibration risk coefficient as a penalty term in the objective function. The optimization objective is to minimize the sum of the comprehensive vibration risk coefficient and the total water consumption of all participating units under the premise of satisfying the total load command of the power grid. The load allocation solution module is used to solve the problem using an optimization algorithm based on vibration risk constraints, and obtain a set of output commands for each unit that meet the total load requirements. This set of output commands not only meets the grid demand, but also actively avoids load allocation schemes that may cause high-risk vibrations. The closed-loop control execution module is used to send the calculated output commands of each unit to the unit control system for execution, and continuously monitor the tilt angle timing data obtained in the first step during the execution process, forming a closed-loop control process from vibration monitoring, risk analysis to load allocation decision-making.

Images