An adaptive algorithm and system for improving the frequency measurement accuracy and response speed of a speed regulator

Abstract

The application provides a kind of adaptive algorithm and system for improving the frequency measurement accuracy and response speed of governor, comprising: Step1, signal acquisition and pretreatment;Step2, dynamically adjust filter parameters to suppress power grid signal noise;Step3, real-time extraction of frequency characteristics to improve frequency resolution;Step4, use synchronous clock as reference frequency source, periodically correct clock reference value automatically, eliminate long-term drift;Step5, construct a nonlinear dynamic model of hydroelectric generating set, real-time compensate the nonlinear relationship between governor valve opening and output, improve frequency regulation accuracy;Step6, output instantaneous frequency.The scheme constructs a multi-dimensional optimization system through adaptive frequency measurement algorithm, high-precision clock calibration, hardware acceleration mechanism and nonlinear compensation strategy, performs high-precision frequency measurement, can quickly effect, so that the hydroelectric system has long-term stability and high adaptability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydropower station modulation, and particularly relates to a self-adaptive algorithm and system for improving the frequency measurement accuracy and response speed of a speed regulator. BACKGROUND

[0002] The speed regulator is the core equipment for peak regulation and frequency modulation of a hydropower station, which responds to frequency fluctuations of a power grid by regulating the output of a hydroelectric generator unit to ensure stable operation of the power grid. However, the existing speed regulator control system has the following problems:

[0003] 1. Nonlinearity and time-varying characteristics: The dynamic characteristics of a hydroelectric generator unit are complex, and the traditional speed regulator control method often uses linearization processing. This control method is difficult to adapt to the nonlinear changes of the system, resulting in insufficient frequency modulation accuracy.

[0004] 2. Poor anti-interference ability of existing frequency measurement means: The existing frequency measurement channel is easily disturbed by noise, and cannot accurately capture small changes (such as ±0.01 Hz) in the frequency of the power grid.

[0005] 3. Response delay: The traditional algorithm has low calculation efficiency, and the hardware processing capacity is insufficient, resulting in a frequency measurement response time of more than 200 ms, which is difficult to meet the demand for fast frequency modulation.

[0006] 4. Time reference value drift: In long-term operation, the clock source has poor stability, and the reference frequency drift error accumulates. After the time reference value drifts, the regulation reliability will be affected. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a self-adaptive algorithm and system for improving the frequency measurement accuracy and response speed of a speed regulator, which solves the technical defects of the speed regulator frequency modulation of a hydropower station.

[0008] To solve the above technical problems, the technical solution adopted by the present application is:

[0009] A self-adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed regulator, comprising the following steps:

[0010] Step 1, signal acquisition and preprocessing;

[0011] Step 2, using a self-adaptive frequency measurement algorithm; using an improved least mean square error LMS adaptive filter to dynamically adjust the filter parameters and suppress the noise of the power grid signal;

[0012] Step 3, combining short-time Fourier transform STFT and sliding window technology to extract frequency characteristics in real time and improve the frequency measurement resolution;

[0013] Step 4, using a synchronous clock as a reference frequency source, embedding a periodic calibration algorithm, and periodically correcting the clock reference value to eliminate long-term drift.

[0014] Step5、Constructing the nonlinear dynamic model of hydroelectric generating set, designing adaptive control law through Lyapunov stability theory, compensating the nonlinear relationship between the valve opening and output of the governor in real time, and improving the frequency modulation accuracy;

[0015] Step6、Outputting the instantaneous frequency.

[0016] The specific steps of signal acquisition and preprocessing in Step1 are as follows:

[0017] Acquiring the three-phase voltage signal of the power grid, the output current signal of the hydroelectric generating set, and the feedback signal of the governor valve opening;

[0018] Signal preprocessing:

[0019] First, filter, then eliminate jump data, then suppress noise, and finally synchronize data.

[0020] In the jump data elimination, set the difference threshold of adjacent sampling points. If the difference between the current sampling value and the previous value exceeds the threshold, it is determined as jump interference and is directly eliminated and interpolated.

[0021] The calculation formula of the improved least mean square error LMS adaptive filter in Step2 is as follows:

[0022] ;

[0023] Wherein, is the weight coefficient of the filter at the nth iteration, dimensionless; is the step factor, that is, the learning rate, which controls the convergence speed and stability, and the value range is usually , is the maximum eigenvalue of the autocorrelation matrix of the input signal; is the error signal, that is, the difference between the expected signal and the actual output, unit: same as the input signal; is the nth sampling value of the input signal, that is, the value of the power grid voltage or current signal at the nth sampling time, unit: V or A; The above formula is used for weight update of the adaptive filter, dynamically adjusts the filter parameters to suppress noise and improve frequency measurement accuracy. In Step2, the output signal is forced to approach the expected signal by continuously adjusting the filter weight, minimizing the error to filter out the noise interference in the power grid signal.

[0024] In Step2, the output signal is forced to approach the expected signal by continuously adjusting the filter weight, minimizing the error to filter out the noise interference in the power grid signal.

[0025] In Step2, the output signal is forced to approach the expected signal by continuously adjusting the filter weight, minimizing the error to filter out the noise interference in the power grid signal.

[0026] ​​The frequency estimation formula of the short-time Fourier transform STFT in Step 3 is:

[0027] ;

[0028] Wherein, is the estimated grid frequency, unit: Hz; is the number of sampling points in the sliding window, which determines the frequency resolution; is the grid voltage or current signal amplitude of the th sampling point, unit: V or A; is the fundamental frequency; is the sampling period, unit: seconds;

[0029] The above formula is used for time-frequency analysis of the grid signal to extract the instantaneous frequency value at the current time and improve the frequency resolution.

[0030] In Step 3, the frequency estimation formula of the short-time Fourier transform STFT is used to extract signal segments through a sliding window, and the Fourier transform is performed on the signal in each window to calculate the amplitude of the fundamental frequency in the frequency spectrum, and finally the instantaneous frequency estimation value is obtained by weighted averaging.

[0031] The periodic calibration formula in Step 4 is:

[0032] ;

[0033] Wherein, is the calibrated reference frequency, unit: Hz; is the original reference frequency output by the synchronous clock, unit: Hz; is the frequency offset, unit: ppm, one millionth, caused by temperature or aging.

[0034] The control equation for non-linear dynamic model compensation of hydroelectric generating units in Step 5 is:

[0035] ;

[0036] Wherein, is the control output of the governor; , : proportional, integral, and derivative gain coefficients, used to adjust the control response speed and stability, dimensionless; is the frequency deviation, i.e. the difference between the target frequency and the actual frequency, unit: Hz; is the non-linear compensation amount, calculated in real time according to the unit dynamic model, unit: %.

[0037] ​The system for improving the frequency measurement accuracy and response speed of a speed regulator by using the adaptive algorithm comprises an adaptive frequency measurement algorithm module, a clock calibration module, a nonlinear dynamic compensation module and a sampling sensor group.

[0038] The adaptive algorithm and system for improving the frequency measurement accuracy and response speed of a speed regulator have the following beneficial effects:

[0039] 1. High-precision frequency measurement: the adaptive algorithm improves the frequency measurement resolution to 0.01 Hz, significantly enhancing the ability to capture small frequency fluctuations in the power grid.

[0040] 2. Fast response: hardware acceleration shortens the frequency measurement response time to 80 ms, meeting the power grid second-level frequency modulation requirements.

[0041] 3. Long-term stability: the combination of GPS clock and calibration algorithm reduces the reference drift rate by 90%.

[0042] 4. Strong adaptability: the nonlinear compensation strategy effectively deals with the dynamic characteristics of hydroelectric generating units, and the frequency modulation accuracy is improved by 13%. BRIEF DESCRIPTION OF DRAWINGS

[0043] The application will be further described below in conjunction with the drawings and examples:

[0044] Figure 1 It is the adaptive frequency measurement algorithm flowchart of the application. DETAILED DESCRIPTION

[0045] The technical solutions of the application will be described in detail below in conjunction with the drawings and examples.

[0046] An adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed regulator comprises the following steps:

[0047] Step 1, signal acquisition and preprocessing;

[0048] Step 2, using an adaptive frequency measurement algorithm; using an improved least mean square (LMS) adaptive filter to dynamically adjust the filter parameters and suppress power grid signal noise;

[0049] Step 3, combining short-time Fourier transform (STFT) and sliding window technology to extract frequency characteristics in real time and improve the frequency measurement resolution;

[0050] Step 4, using a GPS synchronous clock as a reference frequency source, embedding a periodic calibration algorithm, periodically correcting the clock reference value automatically, and eliminating long-term drift;

[0051] Step 5, construct the nonlinear dynamic model of hydroelectric generating set, design adaptive control law through Lyapunov stability theory, compensate the nonlinear relationship between the valve opening and output of the governor in real time, and improve the frequency modulation accuracy;

[0052] Step 6, output the instantaneous frequency.

[0053] The specific steps of signal acquisition and preprocessing in Step 1 above are as follows:

[0054] This step completes the acquisition and preliminary processing of power grid signals through high-precision sensors and preprocessing modules. The main process is as follows: first, acquire signals, and then preprocess signals. The preprocessing process is divided into filtering, noise suppression, and data synchronization. The specific process is as follows:

[0055] 1. Signal acquisition type and acquisition equipment:

[0056] Acquire signals:

[0057] Three-phase voltage signal of power grid (unit: V);

[0058] Hydroelectric generator output current signal (unit: A);

[0059] Speed governor valve opening feedback signal (unit: %).

[0060] Hardware configuration:

[0061] Voltage transformer (PT): transformation ratio 100V / 5V, acquire three-phase voltage signal of power grid;

[0062] Current transformer (CT): transformation ratio 1000A / 5A, acquire generator current signal;

[0063] 16-bit high-precision AD converter, acquire valve signal.

[0064] 2. Preprocessing steps:

[0065] 2.1. Filtering:

[0066] Perform filter filtering with a cutoff frequency of 500Hz to eliminate high-frequency noise interference.

[0067] 2.2. Jump data rejection:

[0068] Set the threshold value of adjacent sampling points (such as ±10% range). If the difference between the current sampling value and the previous value exceeds the threshold value, it is determined to be jump interference, and direct rejection and interpolation completion (such as linear interpolation) are performed.

[0069] 3. Noise suppression:

[0070] Set the digital filter threshold (such as ±0.5V), and set the signal segment below the threshold to zero.

[0071] 4. Data synchronization:

[0072] Synchronize the AD sampling clock with the GPS reference clock to ensure sampling interval accuracy.

[0073] The formula for calculating the improved minimum mean square error (LMS) adaptive filter in Step 2 above is:

[0074] ;

[0075] in, For the first The weight coefficients of the filter in the next iteration are dimensionless. The step size factor, also known as the learning rate, controls the convergence speed and stability, and its value typically ranges from [value range missing]. , The largest eigenvalue of the autocorrelation matrix of the input signal; The error signal is the difference between the expected signal and the actual output, measured in terms of the input signal. Consistent; The first input signal The sampled value, i.e. the th sampled value The grid voltage or current signal value at each sampling time, in V or A;

[0076] The above formula is used for weight updates of adaptive filters, dynamically adjusting filter parameters to suppress noise and improve frequency measurement accuracy.

[0077] In Step 2 above, by continuously adjusting the filter weights, the output signal is made to approximate the desired signal, and the error is minimized, thereby filtering out noise interference in the power grid signal.

[0078] The frequency estimation formula for the Short-Time Fourier Transform (STFT) in Step 3 above is as follows:

[0079] ;

[0080] in, For a moment Estimated power grid frequency, in Hz; The number of sampling points within the sliding window determines the frequency resolution (e.g., =1024); For the first The amplitude of the grid voltage or current signal at each sampling point, in V or A; The base frequency (the nominal frequency of the power grid, such as 50Hz or 60Hz); Sampling period, in seconds. The sampling frequency, such as 1000Hz;

[0081] The above formula is used for time-frequency analysis of the power grid signal, extracts the instantaneous frequency value at the current time, and improves the frequency measurement resolution.

[0082] In Step 3 above, the frequency estimation formula of the short-time Fourier transform STFT is used to intercept the signal segment through a sliding window, and the Fourier transform is performed on the signal in each window to calculate the fundamental frequency The corresponding amplitude, and finally the weighted average of the instantaneous frequency estimation value.

[0083] The periodic calibration formula in Step 4 above is:

[0084] ;

[0085] Where, is the calibrated reference frequency, unit: Hz; is the original reference frequency output by the synchronous clock, unit: Hz; is the frequency offset, unit: ppm, one millionth, caused by temperature or aging (e.g. ±0.1 ppm).

[0086] The control equation for the non-linear dynamic model compensation of the hydroelectric generator set in Step 5 above is:

[0087] ;

[0088] Where, is the control output of the governor (such as valve opening command, unit: %); 、 : proportional, integral, and differential gain coefficients, used to adjust the control response speed and stability, dimensionless; is the frequency deviation, i.e. the difference between the target frequency and the actual frequency, unit: Hz; is the non-linear compensation amount, calculated in real time according to the dynamic model of the unit, unit: %.

[0089] The system using the above adaptive algorithm for improving the frequency measurement accuracy and response speed of the governor, the system includes an adaptive frequency measurement algorithm module, a clock calibration module, a non-linear dynamic compensation module, and a sampling sensor group.

[0090] Embodiment:

[0091] I. System hardware configuration and initialization

[0092] 1. FPGA hardware acceleration platform

[0093] Hardware selection: use a certain domestic self-controllable chip, configure 8 parallel DSP modules, each module supports 18x25-bit multiplier, total logic unit 326,080.

[0094] Architecture design:

[0095] Adaptive filter module: occupies 15% of FPGA resources, adopts pipeline structure, and completes weight update in a single cycle ).

[0096] STFT calculation unit: 4 parallel FFT cores are configured, and the window length N=1024.

[0097] Nonlinear compensation module: based on lookup table method (LUT) to realize real-time calculation of Lyapunov compensation amount, storage depth 1024x32bit.

[0098] Clock configuration:

[0099] FPGA main frequency 200MHz, AD sampling clock is synchronized to GPS reference clock, jitter≤1ps.

[0100] 2. High-precision clock source integration

[0101] Installation and calibration:

[0102] Deploy GPS module, output 1PPS (second pulse) signal, synchronized to FPGA clock management unit.

[0103] Initialize the reference frequency =50.000000Hz, calibration period 24 hours, calibration program triggers GPS frequency deviation (accuracy ±0.1ppm) reading.

[0104] Hardware interface: GPS module communicates with FPGA through RS-422 interface and performs parity check.

[0105] II. Implementation of adaptive frequency measurement algorithm

[0106] 1. Signal acquisition and preprocessing

[0107] 1. Signal acquisition and preprocessing

[0108] ① Sensor configuration and signal type:

[0109] Acquisition signal:

[0110] Three-phase voltage signal of power grid (unit: V), collected through voltage transformer (PT), transformation ratio 100V / 5V, accuracy 0.2 level;

[0111] Hydro-generator output current signal (unit: A), collected through current transformer (CT), transformation ratio 1000A / 5A, bandwidth 0~1kHz;

[0112] The governor valve opening feedback signal (unit: %) is collected by a 16-bit high-precision AD converter, with a range of ±10V and a resolution of 16 bits.

[0113] Sampling parameters:

[0114] The sampling frequency is 1000Hz, which is an integer multiple of the power grid fundamental frequency (50Hz).

[0115] Anti-aliasing filter, cutoff frequency 500Hz.

[0116] ② Pretreatment step:

[0117] Jump data rejection:

[0118] Set the adjacent sampling point difference threshold to ±10% range (e.g. voltage signal range ±10V, threshold ±1V).

[0119] If the difference between the current sampling value and the previous value exceeds the threshold, it is determined to be a jump disturbance, and the data is directly rejected and completed by linear interpolation.

[0120] Noise suppression:

[0121] The front-end analog filter (cutoff frequency 60Hz) eliminates high-frequency noise.

[0122] The digital filter threshold is set to ±0.5V, and the signal segment below the threshold is set to zero.

[0123] Sliding average filter (window length 5 points) is used to further smooth the signal.

[0124] Data synchronization:

[0125] The AD sampling clock is synchronized to the GPS reference clock (1PPS signal), ensuring the sampling interval accuracy (jitter ≤1ps).

[0126] The clock synchronization check between the GPS module and the FPGA is realized through the RS-422 interface.

[0127] ② Noise suppression:

[0128] The front-end analog filter (cutoff frequency 60Hz) eliminates high-frequency noise, and the digital filter threshold is set to ±0.5V, and the signal segment below the threshold is directly set to zero.

[0129] 2. Improved LMS adaptive filter

[0130] Parameter setting:

[0131] Filter order L=64, Step factor (learning rate) =0.02, initial weight .

[0132] Desired signal Error signal = where

[0133] Performance verification:

[0134] When input SNR=30dB, output SNR is improved to 60dB(noise power is reduced to original 1%).

[0135] 3. Short-time Fourier transform (STFT) frequency estimation

[0136] Window design:

[0137] Hanning window (Hanning Window), window length N=1024, overlap rate 50%(sliding step 512 points).

[0138] Frequency resolution , improved to 0.01Hz by interpolation algorithm.

[0139] Real-time calculation:

[0140] Output instantaneous frequency every 10ms , calculation formula:

[0141] ;

[0142] Fundamental frequency =50Hz, sampling period

[0143] The measured data are as follows:

[0144]

[0145] Three, high-precision clock calibration

[0146] 1. Reference frequency correction process

[0147] Calibration trigger: automatically start calibration program at 00:00 every day, read the frequency deviation output by GPS module .

[0148] Calculation formula:

[0149] ;

[0150] Example: if , then:

[0151] Write FPGA: update FPGA internal reference frequency register through SPI interface, write precision 0.001Hz.

[0152] 2. Long-term stability test

[0153] The comparative experiments are shown in the table below:

[0154]

[0155] IV. Nonlinear Dynamic Compensation Control

[0156] 1. Construction of hydropower unit model

[0157] Actual data collection:

[0158] Guide vane opening-output curve: Dead zone ±2%, output linearity decreases by 30% when the opening is >90% in the saturation zone.

[0159] Dynamic response delay: The valve action to output response delay is 80ms (measured average).

[0160] Mathematical model:

[0161] ;

[0162] in =80ms This is a nonlinear compensation term.

[0163] 2. Adaptive Compensation Strategy

[0164] Control law design:

[0165] Traditional PID parameters: .

[0166] Compensation Real-time computation based on Lyapunov functions

[0167] Note: The Lyapunov stability condition formula is as follows:

[0168] ;

[0169] Formula meaning:

[0170] Through Lyapunov functions derivative To determine system stability, if If so, the system is stable.

[0171] Parameter description:

[0172] : Rate of change of frequency deviation (unit: Hz / s).

[0173] Physical meaning: Required frequency deviation Rate of change The sign of the error is reversed to ensure convergence.

[0174] Example:

[0175] If And =−0.05Hz / s, then:

[0176] x(-0.05)

[0177] Real-time adjustment:

[0178] The compensation amount is updated every 20ms, and the FPGA lookup table method achieves a response in microseconds.

[0179] Performance comparison is shown in the following table:

[0180]

[0181] Five, system debugging and full working condition test

[0182] 1. Full working condition verification

[0183] Case 1 (steady-state frequency modulation):

[0184] Simulation: When the frequency fluctuation is ±0.01Hz, the speed regulator output error is ≤0.005Hz;

[0185] Actual measurement: Under the same conditions, the error is ≤0.007Hz.

[0186] Case 2 (noise environment):

[0187] Simulation: When SNR=25dB, the frequency measurement fluctuation is ≤0.012Hz;

[0188] Actual measurement: Under the same conditions, the fluctuation is ≤0.015Hz.

[0189] Case 3 (long-term operation):

[0190] Simulation: 72-hour reference drift ≤0.08ppm;

[0191] Actual measurement: Drift ≤0.12ppm.

[0192] 2. Comparison experiment summary table

[0193]

[0194] Through the cooperative design of FPGA hardware acceleration, GPS clock calibration, adaptive filtering, and non-linear compensation, this scheme has shown significant advantages in simulation and real machine testing.

Claims

1. An adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller, characterized in that, Includes the following steps: Step 1: Signal Acquisition and Preprocessing; Step 2: Use an adaptive frequency measurement algorithm; utilize the improved minimum mean square error (LMS) adaptive filter to dynamically adjust the filtering parameters and suppress power grid signal noise; Step 3: Combine Short Time Fourier Transform (STFT) with sliding window technology to extract frequency features in real time and improve frequency measurement resolution; Step 4: Use a synchronous clock as the reference frequency source, embed a periodic calibration algorithm, and periodically and automatically correct the clock reference value to eliminate long-term drift. Step 5: Construct a nonlinear dynamic model of the hydropower unit, design an adaptive control law using Lyapunov stability theory, and compensate for the nonlinear relationship between the governor valve opening and the output in real time to improve frequency regulation accuracy. The control equations for nonlinear dynamic model compensation of hydropower units are: ； in, This is the control output of the speed controller; , Proportional, integral, and derivative gain coefficients are used to adjust the control response speed and stability, and are dimensionless. Frequency deviation, i.e., the difference between the target frequency and the actual frequency, is expressed in Hz. This is a nonlinear compensation quantity, calculated in real time based on the unit's dynamic model, in units of % %. Step 6: Output instantaneous frequency.

2. The adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller according to claim 1, characterized in that, The specific steps for signal acquisition and preprocessing in Step 1 are as follows: Collect the three-phase voltage signal of the power grid, the output current signal of the hydro-generator unit, and the valve opening feedback signal of the speed governor; Signal preprocessing: First, filtering is performed, then abrupt data is removed, then noise is suppressed, and finally data synchronization is performed.

3. The adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller according to claim 2, characterized in that, In the aforementioned data skipping process, a threshold for the difference between adjacent sampling points is set. If the difference between the current sampling value and the previous value exceeds the threshold, it is determined to be skipping interference, which is then directly skipped and interpolated to complete the data.

4. The adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller according to claim 3, characterized in that, The formula for calculating the improved minimum mean square error (LMS) adaptive filter in Step 2 is as follows: ； in, For the first The weight coefficients of the filter in the next iteration are dimensionless. The step size factor, also known as the learning rate, controls the convergence speed and stability, and its value typically ranges from [value range missing]. , The largest eigenvalue of the autocorrelation matrix of the input signal; The error signal is the difference between the expected signal and the actual output, measured in terms of the input signal. Consistent; The first input signal The sampled value, i.e. the th sampled value The grid voltage or current signal value at each sampling time, in V or A; The above formula is used for weight updates of adaptive filters, dynamically adjusting filter parameters to suppress noise and improve frequency measurement accuracy.

5. The adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller according to claim 4, characterized in that, In Step 2, by continuously adjusting the filter weights, the output signal is made to approximate the desired signal, and the error is minimized, thereby filtering out noise interference in the power grid signal.

6. The adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller according to claim 5, characterized in that, In Step 3, the frequency estimation formula for the Short-Time Fourier Transform (STFT) involves using a sliding window to extract signal segments, performing a Fourier transform on the signal within each window, and calculating the fundamental frequency in the spectrum. The corresponding amplitudes are then weighted and averaged to obtain the instantaneous frequency estimate.

7. The adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller according to claim 6, characterized in that, The periodic calibration formula in Step 4 is as follows: ； in, The calibrated reference frequency, in Hz; The original reference frequency for the synchronous clock output, in Hz; Frequency offset, unit: ppm, part per million, deviation caused by temperature or aging.

8. A system using the adaptive algorithm for improving the frequency measurement accuracy and response speed of a speed controller as described in claim 7, characterized in that, The system includes an adaptive frequency measurement algorithm module, a clock calibration module, a nonlinear dynamic compensation module, and a sampling sensor group.

Images