Hydroelectric generating set PSS performance online evaluation method
By collecting data from the power grid and generating units, and combining simulation and real data, a multi-scenario adaptive evaluation system was constructed. This system solved the problems of real-time updating and multi-dimensional evaluation of PSS performance, and realized real-time quantification and dynamic optimization of PSS performance, thereby improving the dynamic stability of the power grid and generating units.
Patent Information
- Application Number
- CN202511931431.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-03-27
AI Technical Summary
Existing PSS performance evaluation methods rely on offline experiments, making it difficult to achieve real-time parameter updates and multi-dimensional evaluation. Furthermore, they fail to fully consider the coupling effects of multiple parameters under dynamic operating conditions, resulting in evaluation results that do not accurately reflect actual performance.
By collecting input parameter measurement data from the grid side and the generating unit side, and combining the data from the source-grid coordination simulation device and the actual machine SCADA system, a multi-scenario adaptability evaluation system is constructed. The damping ratio, phase compensation degree, regulation speed and robustness index are calculated to achieve real-time quantitative evaluation of PSS performance and dynamic parameter optimization.
It enables real-time quantitative evaluation and dynamic parameter optimization of PSS performance, improves the scientificity and comprehensiveness of the evaluation, adapts to complex dynamic operating conditions, shortens the evaluation time, and improves the dynamic stability of the power grid and generating units.
Smart Images

Figure CN121749513A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydropower unit technology, and in particular to an online evaluation method for the PSS performance of hydropower units. Background Technology
[0002] With the continuous development and optimization of power systems, the role of power system stabilizers (PSS) in improving system stability is becoming increasingly crucial. Traditional PSS performance evaluation relies on field disturbance tests, such as step response and white noise injection, to collect data. This process is cumbersome and complex, and real-time parameter updates are difficult to achieve. Regarding parameter tuning, existing optimization algorithms such as genetic algorithms and particle swarm optimization are prone to getting trapped in local optima and have high computational costs, making them unsuitable for online tuning. Traditional methods such as trial-and-error methods require point-by-point parameter calculations, which are time-consuming and highly dependent on experience. Furthermore, most existing PSS evaluation methods focus on damping ratio or phase compensation, failing to fully consider the coupling effects of multiple parameters under dynamic operating conditions, resulting in evaluation results that do not accurately reflect actual performance. While existing evaluation methods can assess PSS performance to some extent, they still have shortcomings in real-time parameter updates, tuning efficiency, and multi-dimensional evaluation.
[0003] Existing solutions mainly include the WAMS-based online PSS evaluation method and the Gaussian threshold method. The WAMS-based online PSS evaluation detects low-frequency oscillation modes through a wide-area measurement system and calculates the amplitude and phase characteristics of the PSS output signal and the oscillation mode. However, this method fails to address the issues of dynamic operating condition adaptability and multi-index fusion. The Gaussian threshold method uses the statistical distribution of historical health data to set index limits, but this method does not adjust the thresholds in conjunction with real-time grid conditions and unit operating conditions. Summary of the Invention
[0004] This invention provides an online evaluation method for the PSS performance of hydropower units, which integrates simulation data and actual machine operating conditions to construct a multi-scenario adaptive evaluation system. It breaks through the limitations of traditional methods that rely on offline tests, and aims to achieve real-time quantitative evaluation of PSS performance and dynamic optimization of parameters.
[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an online evaluation method for the PSS performance of hydropower units, comprising the following steps: S1. Data Acquisition: Acquire input parameter measurement point data from the grid side and the generator side, and simultaneously acquire the output data of the source-grid coordination simulation device and the real-time operation data of the actual machine SCADA system; wherein, the grid side input parameter measurement points include frequency deviation, active power, and generator terminal voltage, and the generator side input parameter measurement points include speed deviation, head, guide vane opening, vibration in the xyz direction, and peak swing. S2. Performance index evaluation: Based on the data collected in step S1, calculate four core performance indicators: damping ratio, phase compensation degree, adjustment speed, and robustness index. Among them, the damping ratio reflects the oscillation decay rate, the phase compensation degree measures the PSS's ability to compensate for the hysteresis of the excitation system, the adjustment speed characterizes the dynamic response agility of the PSS, and the robustness index evaluates the ability of the PSS parameters to adapt to different working conditions. S3. Online Evaluation: The system sequentially performs data preprocessing and feature extraction, performance index calculation, adaptive threshold dynamic adjustment, fuzzy comprehensive evaluation, parameter optimization feedback, and health report generation to achieve real-time quantitative evaluation and dynamic parameter optimization of PSS performance.
[0006] Preferably, the data preprocessing and feature extraction operation in step S3 specifically involves: performing noise reduction and normalization preprocessing on the collected raw data from the power grid side and the generating unit side, and then extracting key features such as spectral characteristics and dynamic response curves to provide a data foundation for subsequent performance index calculations.
[0007] Preferably, in step S3, the adaptive threshold dynamic adjustment operation specifically involves: using a Gaussian distribution model to dynamically update the index limits, setting the normal threshold to the mean μ ± 1 standard deviation σ, and the warning threshold to the mean μ ± 2 standard deviation σ, and automatically adjusting the threshold range of each index in conjunction with the real-time grid status and unit operating conditions to adapt to dynamic operating condition changes.
[0008] Preferably, the fuzzy comprehensive evaluation operation in step S3 includes two sub-steps: weight allocation and membership degree calculation. (1) Weight allocation: The comprehensive weights of the four indicators, namely damping ratio, phase compensation degree, adjustment speed and robustness index, are determined by combining the CRITIC objective weight method and the AHP subjective weight method. (2) Membership degree calculation: The degree of membership of each indicator in the "good", "normal", "warning" and "abnormal" state intervals is quantified by using the triangular semi-trapezoidal function, and then the comprehensive evaluation result of PSS performance is generated by fuzzy logic operation.
[0009] Preferably, the parameter optimization feedback operation in step S3 is triggered when the comprehensive performance evaluation result of PSS fails to meet the standard; the operation specifically involves: using an improved artificial fish swarm algorithm to optimize the PSS parameters, and feeding the optimized parameters back to the PSS controller to form a closed-loop control.
[0010] Preferably, the improvement of the artificial fish swarm algorithm is the introduction of nonlinear inertial weights and adaptive field of view, achieved through a formula relating the current iteration number t and the maximum iteration number T. Dynamically adjust the search strategy to avoid the algorithm getting stuck in local optima.
[0011] Preferably, in step S3, the trigger condition for generating the health report is that the comprehensive evaluation result of PSS performance meets the standard; the specific operation is: summarizing the status of each performance indicator, the comprehensive evaluation result and parameter optimization suggestions, generating a PSS performance health report, and the report supports historical data backtracking and trend analysis to provide decision-making basis for operation and maintenance personnel.
[0012] Preferably, in step S2, the damping ratio The calculation formula is: In the formula As the attenuation factor, Oscillation frequency; Phase compensation degree The calculation formula is: Adjust speed The calculation formula is: In the formula For power adjustment range, For adjusting time; robustness index The calculation formula is: In the formula This represents the average damping ratio.
[0013] Preferably, the real-time grid status includes grid load changes, and the unit operating conditions include unit head fluctuations.
[0014] Preferably, the output data of the source-grid coordination simulation device in step S1 is the PSS operation correlation data under the simulation condition, and the real machine SCADA system real-time data is the PSS monitoring data during the actual grid-connected operation of the hydropower unit. The two realize the data fusion of source-grid coordination simulation and real machine operation condition.
[0015] This invention provides a method for online evaluation of the PSS performance of hydropower units, which has the following beneficial effects: I. Breaking offline dependence and enabling real-time online evaluation 1. Abandoning the traditional offline disturbance test mode that requires shutdown or load reduction, this system integrates source-grid coordination simulation data with real-time SCADA data from actual machines to construct a dual-track data system of simulation and actual measurement, covering steady-state, transient, and extreme operating conditions. Performance evaluation can be continuously carried out without manual intervention, shortening the evaluation time from the traditional 72 hours to real-time continuous operation. This ensures both grid safety and economical power plant operation, while also solving the problem of insufficient coverage of operating conditions from a single data source.
[0016] 2. The entire process is designed as an online closed loop, with data acquisition, index calculation, threshold adjustment, evaluation optimization, and report generation synchronously following the dynamic evolution of power grid status load changes and unit operating conditions head fluctuations. It adapts to the multi-parameter coupling effects under complex dynamic operating conditions and overcomes the shortcomings of traditional methods in evaluating results that are detached from actual operating conditions.
[0017] II. Multi-dimensional indicator system to enhance the scientific rigor and comprehensiveness of evaluation. 1. Innovatively construct four core indicators: damping ratio, phase compensation degree, adjustment speed, and robustness index. From four physical dimensions, it fully characterizes the performance of PSS from the perspectives of oscillation suppression, phase matching, response time, and operating condition adaptation. This changes the one-sidedness of traditional evaluation that focuses on a single indicator and accurately reflects the true effectiveness of PSS under multi-physics coupling conditions.
[0018] 2. The index calculation adopts a scientific method combining time and frequency domains. The damping ratio is accurately solved by the time domain attenuation rate method or the frequency domain method. The phase compensation degree is quantified based on the relative phase difference between the PSS and AVR links. The adjustment speed is characterized by the ratio of power adjustment amplitude to adjustment time. The robustness index is constructed by the damping ratio dispersion under multiple operating conditions to ensure the reproducibility and physical rationality of the calculation of each index.
[0019] 3. The CRITIC objective weighting method and AHP subjective weighting method are integrated to determine the index weights. The triangular semi-trapezoidal membership function is combined to realize the fuzzy comprehensive evaluation that combines qualitative and quantitative methods, avoiding the misjudgment risk of traditional hard threshold evaluation and improving the fault tolerance and robustness of the evaluation results.
[0020] III. Dynamically adapting to operating conditions enhances evaluation robustness and generalization ability. 1. The threshold values of the indicators are dynamically updated using a Gaussian distribution model. The normal threshold is set to the mean μ ± 1 standard deviation σ, and the warning threshold is set to the mean μ ± 2 standard deviation σ. At the same time, the threshold range is automatically adjusted in combination with the changes in power grid load and the fluctuation of unit head, so as to adapt to dynamic operating conditions such as the switching between wet and dry seasons and large load fluctuations, and solve the problem of frequent false alarms or missed alarms of traditional fixed thresholds.
[0021] 2. Achieve deep integration of source-network coordination simulation data and actual machine SCADA data. Simulation data compensates for the lack of extreme operating condition data of actual machines, and actual machine data corrects the parameter deviations of nonlinear links in the simulation model, expanding the coverage of evaluation conditions. This allows the same set of evaluation rules to be adapted to different head sections and different types of hydropower units, improving the generalization ability of the evaluation system.
[0022] IV. Closed-loop optimization and intelligent operation and maintenance to enhance practical value and engineering efficiency 1. Establish a parameter optimization feedback mechanism. When the evaluation results do not meet the standards, the improved artificial fish swarm algorithm with nonlinear inertial weight and adaptive field of view is adopted to optimize the PSS parameters. The parameters are fed back to the controller through a standardized protocol to achieve hot update. The parameter tuning can be completed without stopping the machine, and the tuning cycle is compressed from the traditional several hours to minutes.
[0023] 2. When the target status is met, a PSS performance health report is automatically generated, summarizing the status of indicators, comprehensive evaluation results and parameter optimization suggestions. It supports historical data backtracking and trend analysis, and predicts future indicator changes through a triple exponential smoothing algorithm, upgrading experience-driven periodic maintenance to data-driven accurate predictive maintenance, reducing the professional threshold and workload of operation and maintenance.
[0024] V. The project has yielded significant results, ensuring the stable operation of the power grid and generating units. 1. In actual testing, this method shortened the low-frequency oscillation suppression time by 37% and reduced the power angle swing amplitude of the unit by 29% under cross-regional DC blocking faults, effectively improving the dynamic stability of hydropower units in the face of complex power grid disturbances.
[0025] 2. In the data preprocessing stage, through denoising, normalization, and feature extraction, the damping ratio calculation error is reduced to within ±0.02, and the false alarm rate of health reports is reduced from 12.7% of the traditional method to 2.1%, providing high-quality data support for evaluation and optimization and ensuring the engineering reliability of the overall technical solution. Attached Figure Description
[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0027] Traditional PSS performance evaluation has long relied on offline disturbance tests, such as manually applying step signals or white noise excitation and then collecting response curves, followed by time-frequency analysis to estimate damping ratio and phase characteristics. This approach has three inherent drawbacks: First, the test requires shutdown or load reduction, affecting the economic operation of the power plant and the safety margin of the power grid; second, a single test only covers specific operating points (such as rated head and full load), making it difficult to reflect the true adaptability of hydropower units under dynamic scenarios such as daily load fluctuations, water level changes, and multiple units operating in parallel; third, the evaluation dimensions are singular, with most methods focusing only on whether the damping ratio threshold is met, neglecting the timeliness of PSS's phase lag compensation for the excitation system, the speed of power regulation response, and the robust stability of parameters under different head-guide vane opening combinations. For example, in the actual operation of a 300MW mixed-flow turbine unit at the Three Gorges Power Station, when the head suddenly dropped from 85m to 72m and the load suddenly increased from 240MW to 280MW, although the original PSS setting met the offline acceptance requirement of "damping ratio ≥ 0.03" in DL / T650—2017, the on-site waveform recording showed that the low-frequency oscillation decay time was extended by 42% and the overshoot of the generator terminal voltage increased by 1.8 times, exposing the problem that the traditional evaluation system is insensitive to multi-physics coupled operating conditions. The technical problem to be solved in this embodiment is to overcome the problems of evaluation distortion and parameter lag caused by the existing technology of PSS performance evaluation being detached from the real-time operating environment, fragmented index dimensions, and lack of closed-loop feedback, and to realize an online quantitative evaluation and dynamic optimization mechanism that does not require artificial disturbance and can adaptively evolve with the source and network conditions.
[0028] Based on this, this application provides a method for online evaluation of the PSS performance of hydropower units, including the following steps: S1. Data Acquisition: Collect input parameter measurement data from the grid side and the generator side, and simultaneously acquire the output data of the source-grid coordination simulation device and the real-time operation data of the actual machine SCADA system; among which, the grid side input parameter measurement points include frequency deviation, active power, and generator terminal voltage, and the generator side input parameter measurement points include speed deviation, head, guide vane opening, vibration in the xyz direction, and peak swing. S2. Performance index evaluation: Based on the data collected in step S1, calculate four core performance indicators: damping ratio, phase compensation degree, adjustment speed, and robustness index. Among them, the damping ratio reflects the oscillation decay rate, the phase compensation degree measures the PSS's ability to compensate for the hysteresis of the excitation system, the adjustment speed characterizes the dynamic response agility of the PSS, and the robustness index evaluates the ability of the PSS parameters to adapt to different working conditions. S3. Online Evaluation: The system sequentially performs data preprocessing and feature extraction, performance index calculation, adaptive threshold dynamic adjustment, fuzzy comprehensive evaluation, parameter optimization feedback, and health report generation to achieve real-time quantitative evaluation and dynamic parameter optimization of PSS performance.
[0029] Step S1: Data Acquisition: Acquire input parameter measurement point data from the grid side and the generator side, and simultaneously acquire the output data of the source-grid coordination simulation device and the real-time operation data of the actual SCADA system; among which, the grid side input parameter measurement points include frequency deviation (Δf), in Hz, which is captured in real time by the station PMU (Phasor Measurement Unit) at a sampling rate of 50Hz to capture the instantaneous offset of the system frequency relative to the nominal value of 50Hz; active power (Pe), in MW, taken from the analog quantity output by the power transmitter from the secondary side of the generator outlet CT (Current Transformer) and PT (Potential Transformer), or directly read from the active power value in the GOOSE message parsed by the IEC61850 protocol in the SCADA system; generator terminal voltage (Vt), in kV, obtained by sampling the 100V signal from the secondary side of the generator outlet PT through a high-precision AD converter, with a sampling resolution of not less than 16 bits. The unit-side input parameter measurement points include: rotational speed deviation (Δω), in rad / s, obtained by calculating the rate of change of angular velocity from the pulse signal output by the photoelectric encoder mounted on the main shaft through the FPGA (Field-Programmable Gate Array) high-speed counting module; head (H), in m, obtained by measuring the difference in water level between upstream and downstream through the pressure transmitters in front of the dam and the tailrace channel, and outputting a 4–20mA standard signal after temperature compensation and zero-point calibration; guide vane opening (α), in %, provided in real time by the guide vane relay piston rod displacement sensor (LVDT, Linear Variable Differential Transformer), with a linearity error ≤ ±0.5%FS; and peak values of vibration and sway in the xyz directions, in μm, collected by the X / Y / Z triaxial acceleration sensor (ICP type, sensitivity 100mV / g) and the non-contact eddy current sensor (probe diameter φ8mm, range ±200μm), respectively, and outputting the maximum amplitude per second after anti-aliasing filtering (cutoff frequency 1kHz) and peak detection algorithm.The output data of the source-grid coordination simulation device refers to the PSS input signal (ΔVref), output signal (ΔVpss), excitation system response (ΔEfd) and time sequence waveform data output by the closed-loop simulation platform based on PSASP (Power System Analysis Software Package) or RTDS (Real-Time Digital Simulator), which includes a detailed model of the hydropower station, load dynamic characteristics and regional power grid topology. The sampling interval is 1ms. The real-time data of the real machine SCADA system refers to the PSS operation monitoring data uploaded by the local control unit (LCU) of the Three Gorges Hydropower Plant to the central monitoring system through the Modbus TCP protocol. This includes the PSS on / off status, gain coefficient Kpss, time constant T1 / T2 of the lead and lag elements, and 100ms periodic sampling sequences of the measured ΔVpss and Δω. The two types of data mentioned above are spatiotemporally aligned using the OPCUA (Open Platform Communications Unified Architecture) protocol under a unified timestamp (IEEE 1588v2 precision clock synchronization, synchronization error <100ns), forming a source-network collaborative "simulation-measurement" dual-track dataset. This data acquisition architecture supports multiple alternatives: for example, frequency deviation can be replaced with regional inertia center frequency deviation based on WAMS wide-area measurement; head measurement can use a redundant configuration of ultrasonic level gauge + pressure level gauge; vibration sensors can be MEMS accelerometers to reduce installation costs; and the simulation data source can also be expanded to a multi-condition Monte Carlo simulation sample library generated by the digital twin platform.
[0030] Step S2: Performance index evaluation: Based on the data collected in step S1, calculate four core performance indicators: damping ratio, phase compensation degree, adjustment speed, and robustness index. Among them, the damping ratio reflects the oscillation decay rate, the phase compensation degree measures the PSS's ability to compensate for the hysteresis of the excitation system, the adjustment speed characterizes the dynamic response agility of the PSS, and the robustness index evaluates the ability of the PSS parameters to adapt to different working conditions.
[0031] The damping ratio (ξ) is a dimensionless parameter characterizing the attenuation characteristics of electromechanical oscillation modes. Its physical essence is the ratio of the real part of the system's characteristic root to its magnitude. In this embodiment, the time-domain attenuation rate method is used for calculation: the Δω or ΔPe signal acquired in step S1 is bandpass filtered (center frequency 0.2–2.5Hz, Chebyshev type II, order 8) to extract the dominant low-frequency oscillation component. The instantaneous amplitude envelope is obtained using Hilbert transform, and the linear relationship between the natural logarithm of the envelope ln(A(t)) and time t is fitted. The slope is the attenuation factor σ. Then, the oscillation frequency fd corresponding to the envelope is identified by FFT (Fast Fourier Transform), and ξ is obtained by substituting it into the formula ξ=σ / √(σ²+(2πfd)²). This calculation can be replaced by the frequency domain method, which performs Welch power spectrum estimation on the Δω signal and calculates ξ≈Δf / (2fd) using the -3dB bandwidth Δf on both sides of the peak and the center frequency fd. The phase compensation degree (θ) is used to quantify the compensation effect of the PSS output signal ΔVpss on the inherent phase lag ∠Havr(jω) of the excitation system AVR (Automatic Voltage Regulator). In this embodiment, the frequency response identification method is adopted: when the PSS is in operation, a pseudo-random binary sequence (PRBS) excitation signal is injected into the PSS input terminal, and ΔVpss and ΔEfd are collected synchronously. The transfer function Hps(jω)=Gvpe(jω) / Gee(jω) is calculated by the cross-spectral density Gvpe(jω) and the self-spectral density Gee(jω). Then, the phase difference is calculated by θ=∠Hps(jω)-∠Havr(jω) with the AVR open-loop frequency response Havr(jω) (obtained by modeling the unit's factory parameters). This method can be replaced by an online identification method, which uses the least squares recursive algorithm to update the equivalent second-order model parameters of PSS and AVR in real time and dynamically calculate the phase difference. The regulation speed (K1) is defined as the ratio of the active power regulation amplitude ΔPc caused by the PSS action to the regulation time Tsettle required for the system to reach steady state. ΔPc is the peak-to-peak value of the ΔPe signal within a 1–3 s window after excitation, and Tsettle is defined as the time point at which ΔPe enters the ±2% steady-state bandwidth and remains within the limit for 1 s. This index can be replaced by the rise time Tr (10%→90% response interval) or the reciprocal of the peak time Tp. The robustness index (R) is used to evaluate the stability margin of the PSS parameters at different operating points. In this embodiment, the operating condition clustering statistical method is used: the continuous 24-hour data collected in step S1 is divided into n operating condition clusters according to the head H and guide vane opening α (e.g., H∈[65,75)m and α∈[40%,60%] is cluster 1). The damping ratio ξ is calculated independently for the data within each cluster. i , and then with all ξ i The variance is used as the robustness index R = 1 / n∑(ξ) i -uξ)²(uξ represents all ξ) i(mean); this definition can be replaced by the condition number method to construct the state matrix A of the PSS closed-loop system under various operating conditions. i Calculate κ i =‖A i ‖·‖A i -1 ‖, take κ i The mean value serves as a robustness measure. The four types of indicators are interrelated: the damping ratio and phase compensation degree jointly determine the low-frequency oscillation suppression effectiveness, and there is a coupling constraint between the two. Excessive phase lead will cause high-frequency resonance, and even if the damping ratio value meets the standard, the system may still become unstable. The adjustment speed and robustness index form a dynamic-static trade-off relationship. Increasing K1 often requires increasing the PSS gain, but it will reduce the R value. Therefore, the four indicators must be evaluated in a coordinated manner and cannot be judged in isolation.
[0032] Step S3: Online Evaluation: The following steps are performed sequentially: data preprocessing and feature extraction, performance index calculation, adaptive threshold dynamic adjustment, fuzzy comprehensive evaluation, parameter optimization feedback, and health report generation, to achieve real-time quantitative evaluation and dynamic parameter optimization of PSS performance.
[0033] The data preprocessing and feature extraction process enhances the reliability and condenses the information of the original measurement data. First, a sliding window median filter (11 points) is used to eliminate impulse noise, followed by Z-score normalization (x′=(x−μ) / σ) to eliminate dimensional differences. Then, two key features are extracted: spectral features (using STFT (Short-Time Fourier Transform) on the Δω signal to obtain the energy entropy, dominant frequency amplitude ratio, and harmonic distortion rate in the 0.1–3Hz band) and dynamic response features (using cross-correlation analysis on ΔVpss and Δω to obtain the time delay τ, and calculating the initial slope d(ΔVpss) / dt| under a unit step input). t=0 This process can be replaced by wavelet packet decomposition to extract node energy features, or by using an LSTM (Long Short-Term Memory) network to automatically learn time series feature representations. Performance index calculation involves executing the four-dimensional index algorithm defined in step S2. Its input data source is the preprocessed feature vector, the calculation period is 10 seconds (consistent with the SCADA refresh cycle), and the results are stored in the time series database in structured JSON format. Adaptive threshold dynamic adjustment establishes dynamic limits for the four indicators based on the Gaussian distribution assumption: using the historical evaluation results of the past 30 days as the sample set, the mean μ and standard deviation σ of each indicator are calculated, and the normal threshold range is set as [μ-σ, μ+σ], and the warning threshold is set as [μ-2σ, μ+2σ]. When the real-time grid status (e.g., load change rate > 5MW / min) or unit operating condition (e.g., head fluctuation rate > 0.3m / min) triggers the preset conditions, the system calls the operating condition mapping table (the lookup basis is the μ of the current H and α clusters). i σi The threshold is updated. For example, when the head is in the low head region (H<70m), the lower limit of the normal threshold for the damping ratio is adjusted from μ-σ to μ-0.8σ to adapt to the physical characteristics of the unit's naturally weak damping under low head conditions. This mechanism can be replaced by a Bayesian update model, which uses real-time observation data to correct the prior distribution parameters online. The fuzzy comprehensive evaluation adopts a two-layer decision structure: the first layer is weight allocation, which integrates the CRITIC method (which calculates objective weights based on the conflict and contrast strength between indicators) and the AHP method (which derives subjective weights by having 5 relay protection experts compare the importance of indicators pairwise), and then weights them together using the entropy weight method to obtain the final weight vector w=[w1, w2, w3, w4]; the second layer is membership degree calculation, which designs a triangular semi-trapezoidal membership function for each indicator i according to four levels: "good" (ξ>0.08), "normal" (0.05<ξ≤0.08), "warning" (0.03<ξ≤0.05), and "abnormal" (ξ≤0.03); finally, the comprehensive evaluation value r∈[0,1] is output through the weighted average fuzzy synthesis operator R=w·M (M is the membership degree matrix). Parameter optimization feedback is initiated when r is below a set threshold (e.g., r < 0.6): the improved artificial fish swarm algorithm (IAFSA) is called to optimize the three core parameters of PSS, Kpss, T1, and T2, with the objective function J = -(w1ξ + w2cos|θ| + w3K1 - λR), where λ is the robustness penalty coefficient; the algorithm introduces a nonlinear inertial weight w(t) = 0.9 - (0.9 - 0.4) × t / Tmax (Tmax = 200) to regulate the search step size, and dynamically adjusts the visual range according to the neighborhood density of the current optimal solution to avoid premature convergence; the optimization results are written to the PSS controller register via OPCUA to form a closed loop of "perception, evaluation, decision-making, and execution"; this step can be replaced by a reinforcement learning framework, using the PSS parameters as the action space and the increase in system damping ratio as the reward signal to train the DQN (DeepQ-Network) policy network. Health report generation is triggered when r ≥ 0.85: A PDF report is automatically generated, including indicator trend charts (time series curves of ξ, θ, K1, and R for the past 7 days), operating condition matching analysis (marking the current position of H / α in the historical cluster), parameter optimization suggestions (such as "It is recommended to adjust T1 from 0.5s to 0.35s to improve phase compensation"), and a historical backtracking interface (supporting retrieval by date and operating condition tag). This report can be expanded into a web visualization dashboard and integrated into the power plant's smart operation and maintenance platform.
[0034] Through the above steps, this application achieves the following: By acquiring multi-source heterogeneous data in S1, the three types of data—wide-area power grid measurement, unit-based sensing, and source-grid simulation output—are integrated under a unified spatiotemporal benchmark, solving the problem of insufficient operating condition coverage caused by traditional evaluation relying on a single data source; By constructing a four-dimensional index system of damping ratio, phase compensation degree, regulation speed, and robustness index in S2, the PSS performance is fully characterized from four physical dimensions: oscillation suppression, phase matching, response time, and operating condition adaptation, overcoming the one-sidedness of single-index evaluation; By establishing a full-link online closed loop of "preprocessing, calculation, threshold adjustment, fuzzy evaluation, optimization feedback, and report generation" in S3, the PSS parameter tuning is upgraded from "offline trial and error" to "online self-optimization," significantly improving the dynamic stability of hydropower units in response to complex power grid disturbances. Specifically, in the actual test of Unit #1 at Gezhouba Power Plant, this method shortened the PSS performance evaluation time from the traditional 72 hours to real-time continuous, reduced the low-frequency oscillation suppression time by 37%, and reduced the unit's power angle swing amplitude by 29% under cross-regional DC blocking faults, verifying the engineering effectiveness of this technical solution in ensuring the grid connection safety of large hydropower bases.
[0035] The data preprocessing and feature extraction operations in step S3 are as follows: the collected raw data from the power grid side and the generating unit side are preprocessed by denoising and normalization, and then key features such as spectral characteristics and dynamic response curves are extracted to provide a data foundation for subsequent performance index calculations.
[0036] Among them, the raw data on the grid side refers to the time-series signals such as frequency deviation Δf, active power Pe, and turbine terminal voltage Vt synchronously collected from the Wide Area Measurement System (WAMS) or the dispatch automation system; the raw data on the generator side refers to the speed deviation Δw, head H, guide vane opening α, and three-dimensional (x, y, z) vibration acceleration signals and swing displacement signals arranged at key positions such as the upper guide bearing, lower guide bearing, and turbine top cover, with a sampling frequency of not less than 1kHz and a time resolution better than 10ms, in order to meet the requirements for low-frequency oscillation (0.1–2.5Hz) mode identification.
[0037] The denoising process involves using an adaptive wavelet thresholding method or an improved empirical mode decomposition (IEMD) to suppress noise in the original signal. The wavelet basis function can be Daubechies4 (db4), and the number of decomposition layers is set to 5 based on the Nyquist–Shannon sampling theorem. The soft thresholding function automatically selects the optimal threshold using the Stein unbiased risk estimation (SURE) criterion. IEMD adds paired positive and negative white noise to assist in mode separation and suppress end-point effects and mode aliasing. This denoising process effectively filters out high-frequency electromagnetic interference, mechanical shock pulses, and quantization noise introduced by SCADA communication, improving the signal-to-noise ratio (SNR) by ≥15dB. Alternatively, moving average filtering (window length of 128 points), median filtering (3×3 neighborhood), or a deep learning-based 1D-CNN denoising model can be used, all achieving the same level of noise suppression.
[0038] Normalization refers to mapping the denoised measurement data to the interval [0, 1] using the Min-Max linear normalization formula: ; in These are the original signal sample values. , These are the minimum and maximum values statistically obtained within the historical operating cycle (no less than 72 hours) of the measuring point, respectively. For parameters with physical boundaries (such as guide vane opening α∈[0%, 100%], speed deviation Δw∈[-5%, +5%]), the nominal range of the equipment is directly used as the normalization benchmark. For dynamic signals without clear boundaries (such as vibration peak value, swing peak value), a rolling window (window length 2 hours) is used to update the extreme values in real time to adapt to the drift of the operating conditions. This normalization strategy ensures that data with different dimensions (Hz, MW, V, mm / s, μm) and different amplitude ranges (mV-level sensor signals vs. V-level voltage signals) have comparability and numerical stability in subsequent joint modeling. As an optional implementation method, Z-score normalization can also be used. Both LogisticSigmoid nonlinear normalization and LogisticSigmoid nonlinear normalization are applicable to the multi-source heterogeneous data fusion scenario involved in this embodiment.
[0039] Among them, spectral characteristics refer to key frequency domain indicators reflecting the system's resonance characteristics extracted through Fast Fourier Transform (FFT) or Short-Time Fourier Transform (STFT), including: the frequency of the dominant oscillation mode. (Corresponding to the energy peak value in the low-frequency band of 0.2–1.5Hz), damping ratio related spectral width (half-power bandwidth Δf), phase difference spectrum between PSS output signal and terminal voltage, and the resonance peak order of each vibration channel (such as 1×, 2×, 5 / 6× frequency components). This spectral characteristic is used to support the calculation of damping ratio ξ and phase compensation degree θ in step S2, with a resolution of not less than 0.02Hz. The Hanning window is selected as the window function to balance spectral leakage and frequency resolution. As an optional implementation, Welch power spectrum estimation, wavelet packet energy spectrum, or instantaneous frequency spectrum of Hilbert-Huang transform (HHT) can also be used instead, all of which can effectively characterize the frequency domain response evolution law of hydropower units in transient processes such as load change and head fluctuation.
[0040] Among them, the key features of the dynamic response curve are the structured feature vectors extracted from the time-domain response that characterize the dynamic quality of the PSS closed-loop control, including: the overshoot σ% and rise time of the power regulation curve under a step disturbance. (10%→90%) Adjustment time (Time required to enter the ±2% steady-state band), number of oscillations N; and the attenuation factor σ of the speed deviation attenuation envelope under zero input response (through linear fitting) (Obtained); also includes kurtosis, impulse factor, and Hurst index (a measure of long-term memory) of the oscillation / swing peak sequence. These features directly support the regulation velocity. With robustness index The quantitative calculation is performed, and it is sensitive to typical faults such as PSS parameter mismatch, excitation system hysteresis, and hydraulic inertial coupling. As an optional implementation, a unit-grid topology response map constructed by dynamic time warping (DTW) distance, LSTM encoder hidden state sequence, or graph neural network (GNN) can also be introduced as a higher-order dynamic feature representation.
[0041] There are deterministic logical dependencies among the above-mentioned technical features: denoising is a prerequisite for normalization—data without eliminating abnormal impulses will lead to extreme value statistical distortion, which in turn will cause the normalization result to deviate from the true dynamic range; normalization is the basis of feature extraction, and multi-source signals that have not been scaled uniformly cannot participate in joint spectrum analysis or time-domain curve alignment under the same weight; and the two types of features, spectrum characteristics and dynamic response curves, complement each other: the former describes the steady-state oscillation mode, and the latter reflects the transient transition process. Together, they constitute the complete dynamic behavior profile required for PSS performance evaluation.
[0042] Through the above steps, this application achieves high-quality structured reconstruction of the original collected data: denoising significantly improves the detectability of low-amplitude oscillation signals, reducing the damping ratio calculation error to within ±0.02; normalization eliminates the dimensional mismatch problem caused by differences in sensor accuracy and installation position deviations, ensuring the objectivity of CRITIC weight calculation; the collaborative extraction of spectral characteristics and dynamic response curves compresses the original massive time-series data into a feature set with controllable dimensions (≤20 dimensions), clear physical meaning, and strong correlation with PSS performance, reducing the inference time of the subsequent fuzzy comprehensive evaluation model by 63%, while reducing the false alarm rate of health reports from 12.7% of the traditional method to 2.1%. This data preprocessing and feature extraction stage, as the front-end hub of the online evaluation process, fundamentally solves the technical problems of high noise, complex dimensions, and redundant information in the original monitoring data, providing a solid data foundation for real-time, accurate, and robust evaluation of PSS performance.
[0043] In step S3, the adaptive threshold dynamic adjustment operation is specifically as follows: the index limits are dynamically updated using a Gaussian distribution model, the normal threshold is set to mean μ ± 1 standard deviation σ, the warning threshold is set to mean μ ± 2 standard deviation σ, and the threshold range of each index is automatically adjusted in combination with the real-time grid status and unit operating conditions to adapt to dynamic operating condition changes.
[0044] The Gaussian distribution model is based on long-term observation data of four performance indicators—damping ratio ξ, phase compensation degree θ, regulation speed K1, and robustness index R—collected under historical healthy operating conditions. A univariate normal distribution model is independently constructed for each indicator. This model calculates the sample mean μ and sample standard deviation σ of each indicator under steady-state operating conditions using maximum likelihood estimation or moment estimation methods, thereby determining its natural fluctuation range. This model does not rely on specific disturbance conditions or artificially set boundaries, but rather extracts statistical features from historical data of typical scenarios such as continuous operation of the actual SCADA system for no less than 30 days, covering high / low water periods, full / low load, and different head ranges (e.g., H∈[45m, 95m]). As a variant implementation, kernel density estimation (KDE) can also be used instead of the Gaussian distribution to more accurately fit asymmetric or bimodal indicator distributions, especially suitable for operating conditions where severe head fluctuations cause the damping ratio to exhibit multimodal characteristics.
[0045] The normal threshold, μ ± 1 standard deviation σ, represents the theoretical boundary of the healthy operating range of the equipment at a 99.7% confidence level (derived according to the 3σ principle of normal distribution). In practical applications, it corresponds to the state where the PSS function is intact and no intervention is required. This range is used to trigger daily monitoring alarm shielding and status green light indicators. As an optional implementation, when the historical data of a certain indicator (such as the robustness index R) is significantly skewed (absolute value of skewness > 0.8), the normal threshold can be adjusted to μ ± 0.8σ to improve sensitivity; or a sliding time window mechanism (window length set to 72 hours) can be introduced to make μ and σ dynamically updated with the most recent operating data, thereby enhancing the response capability to short-term trend drift.
[0046] The warning threshold, defined as the mean μ ± 2 standard deviations σ corresponding to approximately 95.4% confidence level, is a critical region where performance degradation begins to show signs. At this point, the system automatically initiates a secondary diagnostic process, including retrieving the vibration spectrum of the past hour, comparing it with the baseline curve of the same operating condition output by the simulation device, and checking whether the deviation between the guide vane opening command and feedback exceeds the limit (>±1.5%). This threshold is used to drive the setting of the left / right fulcrum of the membership function of the "warning" state in the fuzzy comprehensive evaluation model. As a variant embodiment, the warning threshold can be coupled and corrected with the real-time amplitude of the grid frequency deviation Δf according to the IEC61000-4-30 power quality standard: when |Δf|>0.1Hz, the upper limit of the warning for the phase compensation degree θ is relaxed to μ+2.2σ to avoid false warnings caused by transient disturbances.
[0047] Among them, automatically adjusting the threshold range of each indicator by combining real-time grid status and unit operating conditions refers to establishing a threshold dynamic correction coefficient matrix Γ=diag(γ1, γ2, γ3, γ4), where γ i ∈[0.8, 1.2] represents the working condition sensitivity coefficient of the i-th index (i=1~4 corresponding to ξ, θ, K1, R respectively); γ i Determined by a two-layer mapping relationship: the first layer is the grid-side input, with the real-time load factor L (per unit value, L=P). e / P enom With L as the variable, when L < 0.4, γ1 (damping ratio correction coefficient) is reduced to 0.9, reflecting the natural weakening of damping in the oscillation mode under low load; the second layer is the unit-side input, using the measured value of head H and the rated head H nom The ratio r = H / H nomAs a variable, when r < 0.7, γ3 (adjustment speed correction coefficient) is increased to 1.15 to compensate for the response hysteresis caused by the increase in hydraulic inertia under low head. This two-dimensional correction mechanism is implemented by lookup table method or linear interpolation. The lookup table dimension is a two-dimensional grid of (L, r) with grid point intervals of 0.1 and 0.05, respectively. As an alternative, a lightweight BP neural network (hidden layer node number = 6, activation function is tanh) can be used to learn the mapping relationship between Γ and the four-dimensional input of (L, r, Δf, α) online. The model training data comes from the closed-loop test results of the source network coordination simulation device under 1000 sets of working conditions.
[0048] Among them, the adaptation to dynamic operating condition changes emphasizes that the threshold adjustment has a hybrid feedforward-feedback characteristic: the feedforward part obtains the initial correction coefficient based on the measured operating condition parameters (L, r, Δf, α) at the current moment by looking up a table; the feedback part monitors five consecutive sampling periods (period T). s =100ms) cumulative deviation of the indicator from the mean Σ|X i When the value of −μ| exceeds 1.5 times the threshold dynamic bandwidth, it automatically triggers a re-estimation of μ and σ and resets the Γ matrix; this mechanism ensures that during rapid load increases and decreases (dP) of the unit... e Even under strong non-stationary processes such as ΔH > 2m / 10s (dt > 50MW / min) or sudden head changes (ΔH > 2m / 10s), the threshold can still maintain statistical validity. As a structured implementation, a real-time stream processing module can be deployed in an embedded evaluation terminal (such as Advantech UNO-2484G industrial computer), using a sliding window + incremental Welford algorithm to synchronously update μ and σ, with a calculation latency of less than 15ms.
[0049] Through the above steps, this application achieves the following: a natural fluctuation benchmark for indicators is constructed based on a Gaussian distribution model; health levels are classified using a dual-layer threshold structure of μ±σ and μ±2σ; and a dual operating condition sensing mechanism of grid load factor L and head ratio r is superimposed to dynamically scale the judgment boundaries of each performance indicator. Since the thresholds are no longer fixed but evolve in real time with the operating environment, the technical problem of frequent false alarms or missed alarms in scenarios such as the switching between wet and dry seasons and large load fluctuations caused by traditional fixed thresholds is solved. Furthermore, the introduction of statistical modeling and operating condition coupling for dual calibration allows the same set of threshold rules to cover the Three Gorges Dam (H... nom ≈80m), Xiangjiaba Hydropower Station (H nom By using units with different head sections (≈65m), the generalization ability and engineering applicability of the evaluation system were significantly improved; ultimately, a high degree of consistency between the online performance evaluation results of PSS and the actual physical state was achieved, providing reliable criterion support for subsequent fuzzy comprehensive evaluation and parameter optimization feedback.
[0050] The fuzzy comprehensive evaluation operation includes two sub-steps: weight allocation and membership degree calculation. (1) Weight allocation: The comprehensive weights of the four indicators, namely damping ratio, phase compensation degree, adjustment speed and robustness index, are determined by combining the CRITIC objective weight method and the AHP subjective weight method. (2) Membership degree calculation: The degree of membership of each indicator in the "good", "normal", "warning" and "abnormal" state intervals is quantified by using the triangular semi-trapezoidal function, and then the comprehensive evaluation result of PSS performance is generated by fuzzy logic operation.
[0051] The weighting process involves assigning scientifically reasonable relative importance coefficients to the four core performance indicators—damping ratio (ξ), phase compensation degree (θ), regulation speed (K1), and robustness index (R)—to support subsequent multi-objective fusion evaluation. This step employs a fusion of the CRITIC (Criteria Importance Through Intercriteria Correlation) and AHP (Analytic Hierarchy Process) methods: The CRITIC method automatically calculates objective weights based on the comparative strength (standard deviation) and conflict (correlation coefficient between indicators) of each indicator in historical operating data, reflecting the variation and independence of the data itself; the AHP method involves at least three senior engineers with over ten years of experience in excitation control and PSS setting of hydropower units, using a case library of PSS commissioning evaluations from the Three Gorges cascade hydropower stations over the past five years, constructing pairwise judgment matrices, and deriving subjective weight vectors after consistency testing (CR < 0.1). The fusion process uses a linear weighted synthesis formula: The value of α ranges from 0.4 to 0.7, with a selectable value of 0.6. This parameter can be dynamically updated based on the power plant operation and maintenance knowledge graph. For example, if a unit experiences head surges for more than 35% of the time for three consecutive months, the system automatically lowers α to 0.5, enhancing the guiding role of expert experience in weight allocation. This weight allocation mechanism can be replaced by a combination of entropy weighting and Delphi methods, or by using a TOPSIS-grey relational hybrid weighting model, ensuring the independence of indicators while also considering engineering interpretability.
[0052] The membership calculation refers to mapping the normalized measured values of the four indicators to the fuzzy sets corresponding to the four categories of linguistic variables: "Good," "Normal," "Warning," and "Abnormal," thus achieving a semantic transformation from precise numerical values to qualitative states. The triangular semi-trapezoidal membership function used combines mathematical simplicity with engineering practicality: a left semi-trapezoidal shape (a straight line at the top and a flat top at the bottom) is used for the "Good" interval, a right semi-trapezoidal shape (a flat bottom at the bottom and a straight line at the bottom) is used for the "Abnormal" interval, and a standard triangle is used for "Normal" and "Warning." The inflection point coordinates of each function are set according to the industry standard GB / T36572—2018 Power System Stabilizer Performance Test Specification and the statistical distribution of historical health data from the actual SCADA system. For example, the "Good" membership function for the damping ratio ξ is defined as: when ξ∈[0.08, 0.12], μ... lin (ξ)=1; linearly increases to 1 when ξ∈[0.06, 0.08); remains 1 when ξ>0.12; and is 0 when ξ<0.06. This function structure can be replaced with Gaussian, S-type, or Γ-type membership functions, and can also be configured differently according to the unit type: for pumped storage units, the lower limit of its "warning" interval is 15% higher than that of conventional mixed-flow units to adapt to its faster power response characteristics.
[0053] In this process, weight allocation and membership degree calculation form a closed-loop synergistic relationship: the weight vector determines the contribution intensity of each index in fuzzy synthesis, while the shape and boundary of the membership degree function directly affect the confidence output of each index at different state levels; both are input into the fuzzy inference engine, and the weighted average fuzzy synthesis (i.e., ∑w) is completed by the Zadeh operator. i ·μ i (x)) Finally, a comprehensive membership value in the interval [0, 1] is output, and the current performance level of PSS is determined according to the principle of maximum membership. This collaborative mechanism avoids the risk of misjudgment in the traditional hard threshold evaluation. For example, when the damping ratio is slightly lower than the "normal" lower limit but the phase compensation degree is significantly better than the benchmark value, the weighted membership can still be maintained in the "normal" range, reflecting the fault tolerance and robustness of the system evaluation.
[0054] Through the above-described steps, a multi-dimensional, qualitative and quantitative integrated online evaluation of the PSS performance of hydropower units was achieved. By employing a dual-source weighting approach combining CRITIC and AHP, the objective laws inherent in massive amounts of real-machine operation data are preserved, while also incorporating the domain knowledge of senior engineers regarding hydropower speed regulation characteristics, excitation inertia, and grid disturbance modes, thus resolving the distortion of indicator importance caused by single-weighting methods. The use of a triangular semi-trapezoidal membership function creates a smooth transition zone between adjacent state levels, avoiding the drastic jumps in evaluation results at threshold critical points seen with traditional step-type membership functions. Furthermore, fuzzy logic operations allow for compensatory evaluation through other advantageous indicators even when some indicators are in non-ideal states, thus more realistically reflecting the actual adaptability of the PSS under complex hydro-grid coupling conditions. This implementation method constructs a comprehensive evaluation model with strong interpretability, high adaptability, and good human-computer interaction characteristics, improving the accuracy of PSS health status identification and the effectiveness of operation and maintenance decision support.
[0055] The parameter optimization feedback operation is triggered when the comprehensive performance evaluation result of PSS fails to meet the standard. Specifically, the operation involves using an improved artificial fish swarm algorithm to optimize the PSS parameters and feeding the optimized parameters back to the PSS controller to form a closed-loop control.
[0056] The "PSS performance comprehensive evaluation result not meeting the standard" refers to the final evaluation level output by the fuzzy comprehensive evaluation model in step S3 being lower than the preset qualified threshold. Specifically, this manifests as: at least one of the four core performance indicators falling into the "abnormal" range, or the comprehensive membership weighted score being lower than the set benchmark value of 0.75, or any indicator exceeding the warning threshold range three times consecutively without automatic recovery. This judgment logic is executed in real time by the rule engine built into the online evaluation module, supporting dynamic threshold updates and multi-level alarm linkage to ensure that the triggering conditions are both sensitive and robust. As an optional implementation, this judgment can also introduce a time weight factor to give higher confidence to recent evaluation results and avoid historical outliers interfering with real-time decision-making; it can also be extended to multi-condition weighted judgment, that is, in specific operating scenarios such as sudden changes in head and sudden increases in load, the qualification judgment standard is automatically tightened to improve the response reliability under critical operating conditions.
[0057] The "Improved Artificial Fish Swarm Algorithm" is a global optimization method based on swarm intelligence. Its core lies in simulating fish foraging, swarming, and tail-chasing behaviors to construct a search mechanism. It enhances convergence stability and the ability to escape local optima through two key improvements: nonlinear inertial weights and adaptive field of view. This algorithm does not rely on the gradient information of the objective function and is suitable for complex optimization problems with non-convexity, non-smoothness, and multimodality in the PSS parameter space (typically five-dimensional continuous variables Kp, T1, T2, T3, and T4). As an optional implementation, this algorithm can be replaced by Differential Evolution (DE) or an improved particle swarm optimization algorithm (PSO-ELIT) with an elite-preserving strategy, both of which possess equivalent global exploration capabilities. Alternatively, the artificial fish swarm algorithm can be coupled with a Gaussian Process Regression (GPR) surrogate model, using the surrogate model to predict the objective function value in each iteration, significantly reducing the frequency of real-machine simulation calls and improving online optimization efficiency.
[0058] The "PSS parameters" specifically refer to the adjustable control parameters of the Power System Stabilizer (PSS), including but not limited to: proportional gain Kp, time constants T1 to T4 of the lead-lag elements, filter cutoff frequency fc, and limiting threshold Vlim. These parameters collectively determine the amplitude, phase, and dynamic response characteristics of the PSS output signal, directly affecting its effectiveness in compensating for phase lag in the excitation system and its ability to suppress low-frequency oscillations. As an optional implementation, the parameter optimization can be extended to a dual-input PSS structure with a wide-area signal input. In this case, the optimization variables include the wide-area signal weighting coefficient αw and the communication delay compensation term τd. Alternatively, the optimization can be limited to the two dominant phase compensation parameters T1 and T2, with the remaining parameters frozen to engineering recommended values to balance optimization accuracy and computational real-time performance.
[0059] The phrase "feeding back the optimized parameters to the PSS controller to form a closed-loop control" refers to the following: After optimization, the online evaluation system sends the new parameter set in the form of structured data frames to the PSS functional module within the Automatic Voltage Regulator (AVR) of the hydropower unit via the IEC61850 MMS service or Modbus TCP protocol. Upon receiving the data, the controller immediately performs a hot update of the parameters without requiring shutdown or switching of operating modes. Simultaneously, a lightweight closed-loop verification is triggered during the update process, which involves injecting a small-amplitude pseudo-random disturbance signal to monitor whether the phase-frequency characteristics between the PSS output response and the turbine terminal voltage / speed signal meet the preset margin requirements. If the verification fails, the system rolls back to the previous valid parameter set and records an alarm. As an optional implementation, the feedback channel can be configured with redundant paths: the main channel uses fiber optic Ethernet to achieve millisecond-level transmission, while the backup channel transmits key parameters (such as Kp and T1) through a hard-wired DI / DO interface to ensure basic functional availability under extreme communication failures. Alternatively, a digital signature mechanism can be introduced to authenticate the transmitted parameter packets using the SM2 national cryptographic algorithm, preventing malicious tampering.
[0060] The various technical features form a strict logical closed loop relationship: "The PSS performance comprehensive evaluation result does not meet the standard" serves as the starting condition, driving the "improved artificial fish swarm algorithm" to enter the optimization state; the algorithm uses the weighted comprehensive score of four performance indicators as the fitness function to search for the PSS parameter combination that maximizes the score; after the optimized parameters are verified for security, they are "feedback to the PSS controller" through a standardized communication protocol, completing the physical closed loop from evaluation to execution; after the controller executes the parameter update, the system enters a new round of data collection and evaluation cycle, thus forming an autonomous progressive cycle of "evaluation, diagnosis, optimization, execution, and re-evaluation".
[0061] Through the above-described steps, this application achieves a fundamental shift in PSS parameter tuning from an offline, manual, experience-based model to an online, autonomous, closed-loop model. By employing an improved artificial fish swarm algorithm instead of the traditional trial-and-error method or the prematurely convergent genetic algorithm, the global search efficiency and solution stability in the multi-peak parameter space are improved. The deep coupling of optimization triggering and fuzzy comprehensive evaluation avoids control disturbances caused by indiscriminate and frequent parameter adjustments. Furthermore, the parameter feedback channel supports hot updates and security verification, ensuring zero-downtime implementation of optimization actions during grid-connected operation. Therefore, this scheme effectively solves the problem in the background technology of "lack of automatic optimization mechanism and low efficiency due to reliance on manual intervention when PSS performance is substandard," significantly shortening the parameter tuning cycle (from several hours to minutes), and improving the low-frequency oscillation suppression capability and grid support resilience of hydropower units under dynamic operating conditions such as load fluctuations and head changes.
[0062] The improvement to the artificial fish swarm algorithm lies in introducing nonlinear inertial weights and adaptive field of view, which are correlated with the current iteration number. and maximum number of iterations formula
[0063] Dynamically adjust the search strategy to avoid the algorithm getting stuck in local optima.
[0064] Among them, nonlinear inertia weight It is a key parameter in the artificial fish swarm algorithm that controls the direction and step size of individual movement. Its physical meaning is: in each iteration, it determines the weighted fusion ratio of the artificial fish's own historical best position (i.e., "memory term") and the global best position (i.e., "social term") when updating the current position. This indicates that individuals are endowed with a strong memory retention capacity in the initial stage in order to maintain population diversity and expand the search space; This indicates a trend towards reducing memory dependence and increasing clustering towards the neighborhood of known high-quality solutions during the later convergence phase, thereby improving the accuracy of local development. This is the current iteration number (a positive integer, starting from 1). For a preset maximum number of iterations (ranging from 200 to 500), this formula achieves a non-linear, monotonically decreasing weight as the iteration progresses, unlike traditional linear decreasing strategies (such as...). The arithmetic progression of the weights (with a gradual decrease in rate, a steeper decrease in rate, and a more stable decrease in rate) better aligns with the optimization process's principle of "first broad-area exploration, then focused fine-tuning." Alternatively, the nonlinear inertial weights can also be implemented using an exponential decay mechanism. ( The value range is 0.5–2.0), or it can be in piecewise function form (e.g., maintaining the same value within the first 30% of iterations). The middle 40% is linearly reduced to 0.55, and the last 30% is reduced to 0.4 using a quadratic function, both of which can achieve differentiated control over search behavior.
[0065] Among them, the adaptive field of view range refers to the radius of the perceived neighborhood of an artificial fish when performing basic behaviors such as "foraging", "grouping", and "tail chasing". It is not fixed, but dynamically adjusted based on the current population distribution density, individual convergence status, or the rate of change of the objective function gradient: when the population is highly clustered in a certain region of the solution space and the fitness difference is less than 1%, it is considered to be a population with a high degree of aggregation in the solution space and a fitness difference of less than 1%. When the distance variance between individuals exceeds a set upper limit (e.g., more than 1.8 times the average distance) or the global optimum has not been updated for several consecutive generations, the field of view is expanded to stimulate the ability to escape local traps; the adjustment of the field of view can be based on the population entropy value calculated in real time. ( For the first Mapping the ratio of individual fitness to total fitness, for example, setting... ,in The baseline field of view (valued at 10%–30% of the solution space dimension) is used as the baseline field of view. This is the adjustment coefficient (ranging from 0.5 to 2.0). As an optional embodiment, this adaptive mechanism can also be combined with the PSS parameter sensitivity analysis results to adjust the damping ratio. Parameters that have a significant impact (such as the time constant of the lead and lag elements) , This provides a broader initial perspective, while also considering the robustness index. Parameters with low sensitivity (such as gain) The conservative vision strategy is then adopted to achieve targeted allocation of computing resources.
[0066] Among them, the nonlinear inertial weight and the adaptive field of view work together: the inertial weight dominates the evolution law of the individual's motion direction in the "time dimension", while the field of view dominates the dynamic adaptation of the individual's perception ability in the "spatial dimension". The former ensures that the algorithm has a matching exploration / development balance capability in different iteration stages, while the latter ensures that the algorithm has a matching perception resolution in different solution space regions. The two together constitute a two-degree-of-freedom dynamic control mechanism, which enables the artificial fish swarm to effectively avoid false extreme value traps caused by strong nonlinearity of the excitation system, time delay of the turbine governor and changes in the power grid topology in the complex, multi-peak, and nonlinear PSS parameter optimization problem, and to quickly converge to the Pareto optimal solution set that takes into account the improvement of damping ratio, accurate phase compensation and strong robustness of operating conditions.
[0067] Through the above scheme, this application achieves the following: in the online optimization of PSS parameters, using a nonlinear inertia weight formula... By dynamically constraining individual memory strength and adjusting the perception granularity in real time with an adaptive field of view, the algorithm significantly improves the global optimization capability and convergence stability of the optimization process when facing the technical problems of drastic fluctuations in the PSS performance evaluation objective function and the coexistence of multiple local minima caused by frequent switching of hydropower unit operating conditions (such as load changes and head fluctuations) and time-varying grid oscillation modes (such as frequency drift of low-frequency oscillations between regions). It also solves the defects of traditional artificial fish swarm algorithms, such as premature convergence, susceptibility to local optima, and low parameter optimization success rate caused by fixed or linearly decreasing inertial weights and static setting of the field of view. Ultimately, it achieves high reliability, high precision, and adaptive retuning of PSS controller parameters, supporting hydropower units to continuously provide high-quality subsynchronous oscillation suppression capability in source-grid coordination scenarios.
[0068] The health report generation operation is triggered when the PSS performance comprehensive evaluation result meets the standard. Specifically, the operation involves summarizing the status of each performance indicator, the comprehensive evaluation result, and parameter optimization suggestions to generate a PSS performance health report. The report supports historical data backtracking and trend analysis, providing decision-making basis for operation and maintenance personnel.
[0069] Among them, "PSS performance comprehensive evaluation result meets the standard" means that the final evaluation level output by the fuzzy comprehensive evaluation model in step S3 is judged to be "good" or "normal". That is, the four core performance indicators - damping ratio ξ, phase compensation degree θ, adjustment speed K1, and robustness index R - are all within their respective normal threshold ranges after dynamic adjustment, and the membership weighted comprehensive evaluation value is not lower than the preset threshold (0.85, which can be set to any value in the range of 0.80 to 0.90 according to the power plant operation regulations). This judgment process is executed in real time by the logic judgment unit in the online evaluation module. Its input is the normalized evaluation vector after fuzzy synthesis operation, and the output is a binary status signal ("meets the standard" / "does not meet the standard"), which is used to drive the subsequent health report generation or parameter optimization feedback branch path.
[0070] The "summarizing the status of each performance indicator" refers to the structured organization of the original values of the four indicators calculated within the current evaluation period, their corresponding membership values, their status intervals (e.g., "damping ratio ξ=0.32, belonging to the 'good' interval [0.28, 0.35], membership degree 0.91"), and their deviations from the historical mean μ and standard deviation σ (e.g., "compared to the 30-day mean +2.3%σ") into a unified data frame. This data frame uses JSONSchema format to define the field structure, including required fields such as timestamp, metric_id, raw_value, state_label, membership_degree, deviation_from_mean, and confidence_level (based on the data quality assessment module output), and is compatible with extended fields to adapt to future newly added indicators. The summarization operation is completed by the report generation engine calling the indicator status aggregation service, supporting aggregation by single unit, multi-unit group, or plant-wide dimension.
[0071] The "Comprehensive Evaluation Result" refers to the final evaluation level and its confidence interval obtained after CRITIC-AHP fusion weighting, triangular semi-trapezoidal membership function mapping, and fuzzy synthesis operation. It is presented as a quintuple: (evaluation level, confidence level, dominant influence indicator, state transition indicator, stability score). For example: "Evaluation level = 'Good', confidence level = 0.93, dominant influence indicator = 'phase compensation degree θ', state transition indicator = 'evaluation level has not decreased for 3 consecutive times', stability score = 87.6 / 100". This result is embedded in the report text as structured text, and a corresponding visualization graph (such as a radar chart overlaid with confidence bands and an indicator heat map time series matrix) is generated simultaneously and pushed through both the web interface and mobile terminals.
[0072] Among them, "parameter optimization suggestions" are forward-looking maintenance suggestions provided under the condition of meeting the standard, including three types of content: (1) parameter margin hints, such as "current PSS gain K SSS =12.5, distance from robustness inflection point threshold K SSS =14.2 There is still a 13.6% adjustment margin. It is recommended to check once every quarter. (2) Operating condition adaptation suggestions, such as "In the range of head H∈45–52m, the sensitivity of guide vane opening α to phase compensation degree θ increases by 18%. It is recommended to enable the α feedforward compensation module". (3) Long-term degradation warning, such as "The sliding average of damping ratio ξ in the past 90 days shows a linear decay trend of -0.0015 / day (p<0.01). It is recommended to arrange aging test of excitation system within 6 months". This suggestion is generated by the knowledge graph reasoning engine. Its rule base integrates the PSS operation failure case library of the Three Gorges Power Station in the past ten years and the component life model provided by the unit manufacturer.
[0073] The "PSS Performance Health Report" is encapsulated using Extensible Markup Language (XML) and conforms to a subset of the IEC 61850-6SCL (Substation Configuration Language) specification, ensuring seamless integration with the power plant's existing intelligent operation and maintenance platform (such as iPowerOMS). The report file contains five logical sections: ReportHeader (including report number, generation time, unit ID, PSS model, and SCADA system version number), PerformanceSummary, DetailedMetrics, Recommendations, and Historical Reference. Each section has a digital signature field, signed by the Key Management Center (KMC) using the SM2 national cryptographic algorithm, ensuring the report's immutability and traceability.
[0074] Among them, "support for historical data backtracking" refers to the report's built-in timeline navigation control, which allows users to select any historical evaluation moment (minimum time granularity is 1 minute) as a baseline point, automatically loading the complete indicator sequence of N periods before and after that moment (N defaults to 168, which can backtrack for 7×24 hours, and can be configured as an integer in the range of 1 to 336), and presenting it in a dual Y-axis line chart (left axis: original values of four indicators; right axis: membership degree of comprehensive evaluation level); the backtracking data source comes from a time series database (TimescaleDB cluster), and its storage strategy adopts hierarchical compression: the data of the last 7 days retains the original sampling points (1Hz), the data of 7–30 days is downsampled to a 5-second mean, and the data of more than 30 days is aggregated into hourly statistics (including mean, standard deviation, and extreme values). All data are accompanied by data quality tags (DQTag), indicating the source (simulation / real machine), completeness (Complete / Partial), and credibility (High / Medium / Low).
[0075] Among them, the "Trend Analysis" integrates an improved Holt-Winters triple exponential smoothing algorithm, which models the horizontal, trend, and seasonal components of the indicators respectively; for the unique daily load periodicity and seasonal head characteristics of hydropower units during the flood and dry seasons, the algorithm automatically identifies and locks the period length (T). p =24h is used for load-related indicators, T p =168h is used for head-related indicators), outputting the prediction range for the next 72 hours (confidence level 95%); for example, for the robustness index R, the system can warn that "R is expected to drop to 0.26 in 72 hours (currently 0.29), with an 82% probability of falling below the warning threshold of 0.27". This prediction result is directly linked to the automatic maintenance work order generation module, triggering the preventive maintenance process.
[0076] The various technical features work together as follows: the trigger condition "PSS performance comprehensive evaluation result meets the standard" serves as a logical gating signal to start the report generation engine; the engine calls the indicator status aggregation service, comprehensive evaluation parsing service, knowledge graph reasoning service, XML report template engine, and historical time series access interface based on this signal to form a closed-loop data flow; among them, structured data frames ensure lossless information transmission, XML standardized encapsulation ensures cross-platform compatibility, digital signature mechanism meets the security protection requirements of power monitoring system, hierarchical time series storage strategy balances storage cost and analysis accuracy, and Holt-Winters trend model upgrades static snapshots to dynamic prediction capabilities. Through the above-described steps, this application achieves the proactive construction of an operation and maintenance support system oriented towards human factors engineering, under the premise that the PSS is in a continuously good operating state. On the one hand, the structured health report transforms abstract performance parameters into a status profile and action instructions that operation and maintenance personnel can understand, significantly reducing the professional threshold. On the other hand, the historical backtracking function makes implicit degradation trends explicit, and the trend analysis model upgrades experience-driven periodic maintenance to data-driven precise predictive maintenance. This effectively solves the technical problems in the background technology where operation and maintenance personnel have difficulty intuitively grasping the long-term operating status of the PSS and lack quantitative decision-making basis, thereby improving the autonomous controllability level and intelligent operation and maintenance maturity of the hydropower station's secondary system.
[0077] In step S2, the damping ratio The calculation formula is: In the formula As the attenuation factor, Oscillation frequency; Phase compensation degree The calculation formula is: Adjust speed The calculation formula is: In the formula For power adjustment range, For adjusting time; robustness index The calculation formula is: In the formula This represents the average damping ratio.
[0078] Among them, damping ratio It is a core quantitative indicator characterizing the low-frequency oscillation mode decay characteristics of hydropower units, and its mathematical expression is... Derived from the standard response model of a second-order linear system, where the attenuation factor Using time-domain fitting algorithms (such as the Prony method or ERA algorithm) to obtain the unit speed deviation Give or active power Extracted from the measured oscillation decay curve, unit is rad / s; oscillation frequency Synchronization is identified through the same fitting process, with units of Hz; this formula strictly satisfies the geometric relationship between the damping ratio and the real and imaginary parts of the poles in classical control theory, ensuring that the calculation results under different operating conditions are physically comparable. As a variant example, s and It can also be obtained through frequency domain methods, such as... After the signal undergoes a Fourier transform, the damping ratio is inversely calculated within the dominant oscillation frequency band using a peak search combined with the bandwidth half-width method. Its equivalent accuracy is no less than that of the time-domain fitting method, making it suitable for field scenarios where high-frequency sampling is limited or the data signal-to-noise ratio is low.
[0079] Among them, phase compensation degree It is a key indicator for measuring the ability of the PSS output signal to dynamically correct the inherent phase lag of the Automatic Voltage Regulator (AVR); This indicates that the PSS element is at angular frequency. oh The phase angle of the complex transfer function at that point. This represents the phase angle of the AVR components at the same frequency. Both are calculated based on the device's factory parameters or small-signal models obtained from field identification; this difference reflects the PSS at the target oscillation frequency. The net phase lead that can be provided should ideally approach 1 / 2. To achieve the overall open-loop phase frequency characteristic of the excitation system maintained at approximately [value missing] at the crossover frequency. Phase margin. As a variant implementation, when uncertainties exist in the AVR model, the measured closed-loop response can be used to replace the theoretical model: that is, under a set disturbance excitation, the PSS input signal and the generator terminal voltage signal are simultaneously acquired, and the phase difference between the two is directly estimated through cross-spectrum analysis, and then reconstructed. i The engineering definition enhances model independence and field applicability.
[0080] Among them, adjusting speed Characterizes the ability to rapidly suppress power oscillations after PSS triggering. Defined as active power after PSS action The absolute value of the difference between the first peak value reached by the response curve during the adjustment process and the steady-state value, expressed in MW; Defined as from the start time of the PSS action to The response enters and remains at a steady state value. The time required within the error band, in seconds; the larger this ratio, the higher the power regulation efficiency per unit time, and the more reasonable the PSS gain and time constant configuration. As a variant embodiment, Can be replaced by speed deviation Peak value variation correspond Enter The time of the rated speed error band is extended. This variant is suitable for older units with strong coupling in the speed control system and significant power measurement delay, and enhances the ability of the index to cover the mechanical inertia response characteristics.
[0081] Among them, robustness index It is a statistical index for evaluating the adaptability and stability of PSS parameters under multiple operating conditions; This indicates the number of typical operating condition samples collected continuously, not less than 30 sets, with each set corresponding to one independent load change, head fluctuation, or grid topology change event; For the first The damping ratio calculated using the aforementioned formula under the combined working conditions; This is the arithmetic mean of the damping ratios of all samples; the formula essentially represents the variance of the damping ratio sequence. A smaller value indicates that the PSS parameter is less sensitive to operating condition disturbances and has stronger robustness. As a variant embodiment, Weighted variance can be used: The weight By assigning probability density values to each operating condition within the annual operational distribution (e.g., a weight of 0.35 for high-head conditions during the wet season and a weight of 0.25 for low-head conditions during the dry season), the robustness evaluation is made more closely aligned with actual dispatch patterns; alternatively, it can be... Replace with preset baseline value (e.g., 0.05), rewritten as This is to strengthen the assessment of absolute performance thresholds.
[0082] The variables in the four formulas above have a clear data link dependency: damping ratio. With adjustment speed Share the same group and Time-series data, after unified preprocessing (denoising, synchronization alignment, and removal of steady-state segments), features are extracted separately; phase compensation degree. Required and The parameters come from a standardized model library output by the source-network coordination simulation device, which supports the use of head data reported in real time by SCADA. Guide vane opening Dynamically calling matching sub-models; robustness index The calculation takes the historical calculation results of the aforementioned three indicators within a rolling time window (such as 72 hours) as input, forming a closed-loop data flow of "single-point evaluation - time-series aggregation - statistical modeling".
[0083] Through the above-described steps, this application achieves rigorous mathematical definitions and reproducible calculations for the four core performance indicators of PSS: By clearly defining a unified time-frequency domain calculation path for the damping ratio, it solves the problem of oscillation suppression effects being unquantifiable due to reliance on empirical thresholds in traditional methods; by anchoring the phase compensation degree to the relative phase difference between the PSS and AVR stages, it overcomes the technical defects of single-stage phase testing being susceptible to noise interference and failing to reflect system-level compensation effectiveness; by characterizing the response speed with the ratio of power adjustment amplitude to adjustment time, it avoids dynamic performance misjudgments caused by focusing only on one-sided indicators such as overshoot or rise time; and by innovatively using damping ratio dispersion to construct a robustness index, it breaks through the limitation of verifying parameter effectiveness solely through fixed-condition tests in existing technologies, establishing for the first time a quantitative scale for PSS parameter stability covering the entire operating domain. Therefore, this embodiment provides a solid, consistent, and engineering-verifiable mathematical foundation for the online evaluation system constructed in claim 1, ensuring logical consistency and credible results throughout the entire process from data acquisition and indicator calculation to fuzzy evaluation.
[0084] Real-time grid status includes changes in grid load, and unit operating conditions include fluctuations in unit head.
[0085] Real-time grid status refers to the measurable, observable, and quantifiable time-series state variables characterizing the external dynamics of the power system during the grid-connected operation of hydropower units. Its core function is to reflect the time-varying characteristics of system inertia level, power balance margin, and low-frequency oscillation modes, and it is a key external boundary condition affecting the PSS phase compensation requirement and damping effectiveness. Grid load changes specifically refer to the rate of increase (ΔP) of the total regional load power on a time scale ranging from minutes to seconds. L / Δt), step amplitude (|ΔP) L |) Duration and trend (rising / falling / oscillating), this parameter can be obtained in real time through the bus active power measurement point in the dispatch automation system (EMS) or wide area measurement system (WAMS); as an alternative embodiment, the equivalent load inertia time constant T can also be used. J,eq =2H sys ·S base / P L Perform normalized characterization, where H sys S is the system's equivalent inertial time constant. base As the baseline capacity, P L As the real-time load power, this parameter is more directly related to the low-frequency oscillation attenuation characteristics and is suitable for threshold adaptive adjustment in scenarios with a high proportion of new energy access.
[0086] The operating conditions of a turbine-generator unit refer to the set of key physical parameters that affect the hydraulic-mechanical-electrical coupling response characteristics of the unit in a real operating environment. Its core function is to determine the dynamic response bandwidth of the speed control system, the intensity of the water hammer effect, and the distribution of torsional vibration modes in the shaft system, thereby changing the phase lag that the power supply system (PSS) needs to compensate for and the target range of the optimal damping ratio. Specifically, the head fluctuation of the unit refers to the standard deviation σ of the difference between the upstream water level and the downstream tailrace water level (i.e., net head H) within a time window of 5 seconds to 30 minutes. H Range ΔH = H max -H min , rate of change dH / dt and steady-state offset (HH) rated This parameter can be obtained by jointly calculating the pressure transmitter at the inlet of the unit's pressure steel pipe and the tailrace level gauge; as an alternative embodiment, the dimensionless head fluctuation coefficient K can also be used. H =σ H / H rated (H) rated Using the rated head as the basis for threshold adjustment, this coefficient can eliminate dimensional interference caused by differences in unit models and improve the generalization ability of the evaluation model for cross-basin power stations.
[0087] There is a clear physical coupling between the two technical features mentioned above: a sudden increase in grid load will cause a drop in system frequency, triggering a primary frequency regulation action and increasing the guide vane opening; however, changes in guide vane opening under non-constant head conditions will cause water hammer pressure fluctuations, further exacerbating instantaneous head deviations. This chain reaction of "load change → frequency regulation response → head disturbance" necessitates that the PSS simultaneously adapt to the dual constraints of increased electrical-side damping requirements and increased hydraulic-side response delays. Therefore, in the adaptive threshold dynamic adjustment stage, using both as joint input variables can drive the mean μ and standard deviation σ of the Gaussian distribution model to be corrected in the following way: μ adj =μ hist +k1·|ΔP L / P L,nom |+k2·|ΔH / H rated |, σ adj =σ hist ·(1+k3·σ H / H rated ), Where k1, k2, and k3 are the coupling weight coefficients calibrated on-site, with values ranging from [0.05, 0.15], [0.1, 0.3], and [0.2, 0.5], respectively; this correction mechanism ensures that the normal threshold range [μ adj -σ adj μ adj +σ adj ] and the warning threshold range [μ adj -2σadj μ adj +2σ adj It always keeps pace with the current source network collaborative operation status to avoid false alarms or missed judgments caused by static thresholds.
[0088] Through the above-described steps, this application achieves the engineering implementation of the technical feature in claim 3, "automatically adjusting the threshold range of various indicators by combining real-time grid status and unit operating conditions." On the one hand, by using grid load changes and unit head fluctuations as concrete inputs, the abstract real-time grid status and unit operating conditions are transformed into standardized signals that can be directly acquired by the existing SCADA system of the power plant, significantly reducing the on-site deployment threshold. On the other hand, these two factors respectively characterize the external electrical disturbance source and the internal hydraulic inertial source that affect PSS performance. Their combined introduction makes the threshold adjustment physically interpretable, overcoming the failure problem caused by simply relying on historical statistical distributions. Ultimately, without increasing hardware costs, the robustness and accuracy of the adaptive mechanism under complex operating conditions such as frequent load fluctuations and seasonal head changes are improved, solving the technical problem described in the background art that "existing evaluation methods do not fully consider the multi-parameter coupling effects under dynamic operating conditions, and the evaluation results are difficult to reflect actual performance."
[0089] The output data of the source-grid coordination simulation device in step S1 is the PSS operation correlation data under the simulation condition, and the real-time data of the real machine SCADA system is the PSS monitoring data during the actual grid-connected operation of the hydropower unit. The two realize the data fusion of source-grid coordination simulation and real machine operation condition.
[0090] The “source-grid coordination simulation device” refers to a closed-loop simulation system built on power system electromagnetic transient or electromechanical transient modeling tools (such as PSCAD / EMTDC, MATLAB / Simulink, PSASP or self-developed digital twin simulation platform) that can reproduce the coupling effect of grid-side disturbances (such as short-circuit faults, load changes, tie-line power oscillations) and unit-side dynamic responses (such as water hammer effect, speed regulation system delay, water diversion system inertia). The device drives the PSS controller model to run by preset typical operating condition sets (including small disturbance stable operating conditions, large disturbance extreme operating conditions, islanded grid operating conditions, low head high load operating conditions, etc.), and outputs its input-output response sequence, internal state variable trajectory and key performance intermediate quantities (such as PSS output signal amplitude, phase, spectral energy distribution, excitation voltage response hysteresis angle) under different source-grid interaction scenarios, i.e., “PSS operation correlation data under simulation conditions”. This type of data is controllable, repeatable, and boundary complete. In particular, it can cover extreme operating scenarios that are difficult to trigger on-site or pose safety risks (e.g., 0.2Hz~2.5Hz wideband oscillation caused by a three-phase short circuit on a 500kV bus, and low-damped oscillation mode when the head drops to 60% of the rated value), thus providing a high-confidence "virtual benchmark sample" for the evaluation model.
[0091] Among them, the "real-machine SCADA system" refers to the real-time monitoring data platform that collects and uploads data to the dispatch master station during the grid-connected operation of the hydropower unit by the distributed control system (DCS) and the plant-level monitoring system (such as NARI NS3000, XJ WYK-800, and Sifang CSC-2000); the "PSS monitoring data" collected includes the actual excitation regulation command (i.e., PSS output signal U) after processing by the PSS link. PSS The data includes synchronous generator rotor current I_f, terminal voltage U_t, active power P_e, frequency f, speed ω, guide vane opening α, head H, and the original waveforms of vibration / sway after filtering (x, y, z triaxial acceleration and displacement peak values). This type of data is authentic, timely, and continuous in operating conditions, fully reflecting the actual PSS effect of the unit in the real power grid environment (including AGC / AVC command tracking, cross-regional power support, and fluctuations in the penetration rate of new energy sources). However, due to the constraints of field tests and safety regulations, it is not possible to actively apply strong disturbances to stimulate full-mode response, and it is also difficult to obtain high signal-to-noise ratio wideband dynamic characteristics.
[0092] The "PSS operation correlation data under simulation conditions" and the "real-time data of the actual SCADA system" are not simply superimposed, but deeply integrated through spatiotemporal alignment, feature mapping, and confidence weighting mechanisms. First, in the time dimension, based on the SCADA system's 100ms~1s sampling period, the simulation data is interpolated and resampled at the same time resolution, and microsecond-level synchronization is achieved using GPS timing signals or IEC61850-9-2 sampling value message timestamps. Second, in the spatial dimension, a physical quantity mapping table is established between the two sets of data. For example, the "PSS output signal" from the simulation corresponds to the "excitation" in the actual SCADA system. The "controller PSS channel output value" in the simulation and the "terminal voltage phase angle deviation" in the simulation correspond to the "dominant oscillation mode phase angle extracted by FFT from the actual measured "ΔU_t". The "head H" parameter in the simulation is calibrated and corrected with the equivalent net head calculated by the "pressure steel pipe inlet pressure sensor + tailwater level sensor" of the actual machine. Finally, in terms of data quality, a confidence factor γ∈[0,1] is introduced to assign weights to two types of data: for steady-state or small disturbance conditions, γ is 0.3~0.5, focusing on the weight of the actual machine data; for large disturbance or critical instability conditions, γ is 0.7~0.9, improving the contribution of simulation data and ensuring that the evaluation model still has reliable criteria in the data sparse region.
[0093] Through the above-described steps, this application achieves cross-domain collaborative modeling of source-grid coordination simulation data and actual machine operation data. Since simulation data compensates for the lack of actual machine data under extreme operating conditions, while actual machine data corrects parameter deviations in nonlinear components of the simulation model (such as governor dead zone, guide vane mechanical hysteresis, and turbine S-characteristics), the resulting hybrid dataset significantly expands the operating condition coverage and dynamic response identification accuracy of PSS performance evaluation. This supports robust calculations of indicators such as damping ratio ξ and phase compensation degree θ in subsequent steps, avoiding evaluation bias caused by a single data source. Ultimately, this enables the online evaluation system to provide reliable, timely, and traceable PSS health status judgments at critical junctures such as the initial commissioning of new units (without sufficient historical disturbance data), the parameter reset phase after maintenance, or significant adjustments to the power grid structure (such as the addition of DC feeds or reinforcement of inter-regional interconnections).
[0094] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.
Claims
1. A method for online evaluation of PSS performance of hydropower units, characterized in that, Includes the following steps: S1. Data Acquisition: Acquire input parameter measurement point data from the grid side and the generator side, and simultaneously acquire the output data of the source-grid coordination simulation device and the real-time operation data of the actual machine SCADA system; wherein, the grid side input parameter measurement points include frequency deviation, active power, and generator terminal voltage, and the generator side input parameter measurement points include speed deviation, head, guide vane opening, vibration in the xyz direction, and peak swing. S2. Performance index evaluation: Based on the data collected in step S1, calculate four core performance indicators: damping ratio, phase compensation degree, adjustment speed, and robustness index. Among them, the damping ratio reflects the oscillation decay rate, the phase compensation degree measures the PSS's ability to compensate for the hysteresis of the excitation system, the adjustment speed characterizes the dynamic response agility of the PSS, and the robustness index evaluates the ability of the PSS parameters to adapt to different working conditions. S3. Online Evaluation: The system sequentially performs data preprocessing and feature extraction, performance index calculation, adaptive threshold dynamic adjustment, fuzzy comprehensive evaluation, parameter optimization feedback, and health report generation to achieve real-time quantitative evaluation and dynamic parameter optimization of PSS performance.
2. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, The data preprocessing and feature extraction operations in step S3 are as follows: the collected raw data from the power grid side and the generating unit side are preprocessed by denoising and normalization, and then key features such as spectral characteristics and dynamic response curves are extracted to provide a data foundation for subsequent performance index calculations.
3. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, In step S3, the adaptive threshold dynamic adjustment operation specifically involves: using a Gaussian distribution model to dynamically update the index limits, setting the normal threshold to the mean μ ± 1 standard deviation σ, and the warning threshold to the mean μ ± 2 standard deviation σ, and automatically adjusting the threshold range of each index in conjunction with the real-time grid status and unit operating conditions to adapt to dynamic operating condition changes.
4. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, The fuzzy comprehensive evaluation operation in step S3 includes two sub-steps: weight allocation and membership degree calculation. (1) Weight allocation: The comprehensive weights of the four indicators, namely damping ratio, phase compensation degree, adjustment speed and robustness index, are determined by combining the CRITIC objective weight method and the AHP subjective weight method. (2) Membership degree calculation: The degree of membership of each indicator in the "good", "normal", "warning" and "abnormal" state intervals is quantified by using the triangular semi-trapezoidal function, and then the comprehensive evaluation result of PSS performance is generated by fuzzy logic operation.
5. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, The parameter optimization feedback operation in step S3 is triggered when the comprehensive performance evaluation result of PSS fails to meet the standard. Specifically, this operation involves using an improved artificial fish swarm algorithm to optimize the PSS parameters and feeding the optimized parameters back to the PSS controller to form a closed-loop control.
6. The method for online evaluation of PSS performance of hydropower units according to claim 5, characterized in that, The improvement to the artificial fish swarm algorithm lies in the introduction of nonlinear inertial weights and adaptive field of view, achieved through a formula relating the current iteration number t and the maximum iteration number T. Dynamically adjust the search strategy to avoid the algorithm getting stuck in local optima.
7. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, In step S3, the trigger condition for generating the health report is that the comprehensive evaluation result of PSS performance meets the standard. Specifically, this operation involves summarizing the status of each performance indicator, the comprehensive evaluation result, and parameter optimization suggestions to generate a PSS performance health report. The report supports historical data backtracking and trend analysis, providing decision-making basis for operation and maintenance personnel.
8. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, In step S2, the damping ratio The calculation formula is In the formula As the attenuation factor, Oscillation frequency; Phase compensation degree The calculation formula is Adjust speed The calculation formula is In the formula For power adjustment range, To adjust the time; Robustness Index The calculation formula is In the formula This represents the average damping ratio.
9. The method for online evaluation of PSS performance of hydropower units according to claim 3, characterized in that, The real-time grid status includes changes in grid load, and the unit operating conditions include fluctuations in unit head.
10. The method for online evaluation of PSS performance of hydropower units according to claim 1, characterized in that, The output data of the source-grid coordination simulation device in step S1 is the PSS operation correlation data under the simulation condition, and the real machine SCADA system real-time data is the PSS monitoring data during the actual grid-connected operation of the hydropower unit. The two realize the data fusion of source-grid coordination simulation and real machine operation condition.
Citation Information
Cited By
Hydropower station governor primary frequency modulation intelligent cooperative control method and system
CN122052038A
