Method for measuring speed of water turbine toothed disc based on adaptive de-noising

Abstract

This application discloses a method for measuring the speed of a turbine gear disk based on adaptive interference cancellation, comprising the following steps: real-time acquisition of speed information of the turbine gear disk using a speed sensor; spectral analysis of the speed information; establishment of an interference signal feature data identification model to identify interference features in the speed information; and elimination of the interference features based on a filtering algorithm to obtain accurate speed information. This application addresses the shortcomings of existing gear disk speed measurement interference processing technologies in the face of the complex operating environment and dynamically changing interference signals of turbines, failing to meet the high-precision speed measurement requirements of modern turbines.

Description

Technical Field

[0001] This application relates to the field of hydropower equipment technology, and in particular to a method for measuring the speed of a turbine gear disc based on adaptive disturbance reduction. Background Technology

[0002] Accurate measurement of the rotational speed of a hydro turbine is crucial for ensuring its stable and efficient operation and the reliable power supply of the power system. Gear disc tachometer, a commonly used method for measuring hydro turbine rotational speed, involves installing a gear disc on the turbine shaft and using a sensor to detect the pulse signals generated by the disc's rotation to obtain the rotational speed information. However, in practical applications, this method faces several significant challenges, primarily concerning the impact of interference signals on measurement accuracy and the limitations of traditional processing methods.

[0003] The operating environment of a hydroelectric turbine is extremely complex, with multiple sources of interference. Firstly, regarding the mechanical structure, the turbine's own mechanical vibration is unavoidable during high-speed rotation. This vibration causes minute displacements and oscillations in the rotating gear disc, resulting in fluctuations in the signals sensed by the sensors and introducing mechanical interference signals. For example, after long-term operation, components such as the turbine runner and shaft system may experience wear and deformation, exacerbating mechanical vibration and making the interference signals more complex and intense. Secondly, regarding the electromagnetic environment, numerous electrical devices surround the turbine, such as generators and excitation systems. The electromagnetic radiation generated by these devices can interfere with the signal transmission of sensors, causing electromagnetic interference signals to be mixed into the speed measurement signals. This problem is particularly pronounced in older power plants where the electromagnetic compatibility of electrical equipment is poor.

[0004] Traditional methods for handling interference in turbine speed measurement primarily employ filters with fixed parameters, such as low-pass and band-pass filters. These filters filter signals based on a preset frequency range, attempting to remove interference components above or below a specific frequency. However, this method has significant limitations. Firstly, due to the complex and variable operating conditions of hydro turbines, the frequency characteristics of interference signals differ greatly under different conditions. For example, during turbine startup and shutdown, the rotational speed changes significantly, and the frequency range of interference signals also changes accordingly. Filters with fixed parameters struggle to adapt to these changes, potentially distorting the true speed information while filtering out interference signals, or failing to effectively filter out interference signals with certain frequency variations. Secondly, traditional filters are prone to phase lag when processing signals. Phase lag causes the measured rotational speed information to be out of sync with the actual rotational speed. For systems requiring real-time and precise turbine control, this can lead to a series of problems, such as untimely control response and reduced adjustment accuracy, severely impacting the turbine's operating efficiency and stability.

[0005] Furthermore, most existing interference processing technologies do not fully consider the real-time operating status of the turbine and the dynamic characteristics of the interference signal. In actual operation, the interference signal is not static but constantly changes with factors such as turbine load variations and equipment aging. Traditional methods lack the ability to monitor and adaptively adjust these changes in real time, and cannot optimize the interference processing strategy in a timely manner according to the actual situation. As a result, under complex and variable operating conditions, the speed measurement accuracy is difficult to guarantee effectively.

[0006] In summary, existing gear-disc speed measurement interference processing technologies have many shortcomings when facing the complex operating environment and dynamically changing interference signals of hydro turbines, and cannot meet the high-precision speed measurement requirements of modern hydro turbines. Therefore, it is urgent to develop an adaptive interference cancellation technology that can adapt to changes in interference signals in real time, effectively avoid phase lag, and has intelligent adjustment capabilities. Summary of the Invention

[0007] The main objective of this application is to provide a turbine gear disc speed measurement method based on adaptive interference cancellation, in order to solve the problem that existing gear disc speed measurement interference processing technology has many shortcomings when facing the complex operating environment and dynamically changing interference signals of turbines, and cannot meet the needs of modern turbines for high-precision speed measurement.

[0008] To achieve the above objectives, this application provides the following technical solution: A method for measuring the speed of a turbine gear disc based on adaptive disturbance cancellation includes the following steps: The speed information of the turbine gear disc is collected in real time by a speed sensor; Perform spectral analysis on the velocity information; An interference signal feature data identification model is established to identify interference features in the speed information; The interference features are eliminated based on the filtering algorithm to obtain accurate speed information.

[0009] As a further improvement to this application, the velocity information is subjected to spectral analysis, including the following steps: The speed information is converted into a digital signal through ADC conversion; The digital signal is subjected to spectral analysis by optimizing the Fast Fourier Transform algorithm.

[0010] As a further improvement to this application, the digital signal is subjected to spectral analysis by optimizing the Fast Fourier Transform algorithm, including the following steps: When optimizing the Fast Fourier Transform algorithm, perform the following steps: The input data address for each butterfly operation is calculated using formula (1), and the data is accessed directly in the required order during the butterfly operation. For the k-th level butterfly operation, formula (1) is expressed as: ,in, The address of the input data for the i-th butterfly operation. Here, i represents the starting address of the data storage, and i represents the sequence number of the butterfly operation. This indicates a bitwise XOR operation. Stride is the interval of the current butterfly operation, and Stride is the step size for data storage. The number of points in the Fast Fourier Transform algorithm is dynamically adjusted based on the speed information characteristics of the turbine at different operating stages.

[0011] As a further improvement to this application, the spectral analysis of the digital signal by optimizing the Fast Fourier Transform algorithm further includes the following steps: Based on the speed information characteristics of the turbine under different operating environments, an appropriate window function is selected to perform windowing processing on the digital signal, including the following steps: The digital signal is subjected to feature analysis, including frequency range, amplitude variation, and noise intensity; Based on the feature analysis results, an appropriate window function selection strategy is formulated; When switching between different window functions, a linear weighted algorithm is used for a smooth transition.

[0012] As a further improvement to this application, an interference signal feature number identification model is established to identify interference features of the speed information, including the following steps: An interference feature database was constructed based on the interference signals generated during the operation of the water turbine. The support vector machine algorithm was optimized and trained based on the aforementioned interference feature database; The artificial neural network algorithm is optimized and trained based on the aforementioned interference feature database; The recognition results of the support vector machine algorithm and the artificial neural network algorithm are fused to obtain the final interference feature recognition result.

[0013] As a further improvement to this application, when optimizing the support vector machine algorithm, the kernel function parameters of the support vector machine algorithm are dynamically adjusted based on cross-validation, including the following steps: The training dataset is randomly divided into k disjoint subsets using k-fold cross-validation. Set an initial range of kernel function parameters, and gradually increase it from a small kernel function parameter value to a larger kernel function parameter value with a set first step size to form multiple candidate kernel function parameter values; For each candidate value of the kernel function parameter, K-fold cross-validation is performed to select the optimal solution for the kernel function parameter.

[0014] As a further improvement to this application, when optimizing the artificial neural network algorithm, the Leaky ReLU function is selected, and the α parameter is optimized, including the following steps: Set an initial range for the α parameter, and gradually increase it to a larger α parameter value starting from a smaller α parameter value with a set second step size, thus forming multiple candidate α parameter values; For each α parameter value, a neural network with the same structure is constructed, and the same perturbation feature dataset is used for training and testing. During training, the optimal solution for the α parameter is selected by using convergence speed and classification accuracy as evaluation metrics.

[0015] As a further improvement to this application, the recognition results of the support vector machine algorithm and the artificial neural network algorithm are fused to obtain the final interference feature recognition result, including the following steps: Based on the historical accuracy of the support vector machine algorithm and the artificial neural network algorithm in the identification of different types of interference feature signals, different weights are assigned to each algorithm. Based on the recognition results and corresponding weights of the support vector machine algorithm and the artificial neural network algorithm, the corresponding basic probability allocation function is recalculated. Based on the DS evidence theory fusion method, the basic probability allocation functions of the support vector machine algorithm and the artificial neural network algorithm are fused, and the final interference feature identification result is obtained according to the fused basic probability allocation function.

[0016] As a further improvement to this application, the interference features are eliminated based on a filtering algorithm to obtain accurate speed information, including the following steps: In the adaptive minimum mean square error filtering algorithm, a variable step size factor is introduced. The variable step size factor is dynamically adjusted according to the statistical characteristics of the error signal and the input signal at the current time. The input signal is defined as the speed information after interference feature identification, and the error signal is defined as the difference between the expected output signal and the output signal after filtering algorithm processing. The filter coefficients are continuously updated according to the adaptive minimum mean square error filtering algorithm until they converge to the optimal value. The update range of the filter coefficients is set, and the filter output signal is monitored and analyzed in real time. When abnormal fluctuations are detected in the filter output signal, the initial filter coefficients are automatically adjusted.

[0017] As a further improvement to this application, based on the adaptive minimum mean square error filtering algorithm, the filter coefficients are continuously updated and iterated until they converge to the optimal value, including the following steps: In each iteration, based on the input signal at the current moment... Sum of error signals According to formula (2): Calculate the variable step size factor ,in and These are the maximum and minimum values ​​of the variable step size factor, respectively. It is a constant; Based on the calculated step size factor Update the filter coefficients.

[0018] The beneficial effects of this application are as follows: (1) The acquired speed information electrical signals are digitally processed and subjected to spectrum analysis. After obtaining the signal frequency components using the Fast Fourier Transform algorithm, an interference signal feature data identification model is established. Based on the differences in frequency characteristics between the real speed information and the interference signal, a reasonable frequency threshold range is set to accurately identify the interference signal characteristics. Subsequently, an adaptive filtering algorithm is used to automatically adjust the filter parameters according to the real-time signal characteristics, dynamically reducing interference features. This ensures that the real speed information can be restored to the greatest extent in various complex interference environments, avoiding the distortion of measurement results by interference signals and improving measurement accuracy.

[0019] (2) The variable step-size factor mechanism enables the filter to dynamically adjust the step-size factor based on the statistical characteristics of the current error signal and the input signal. During turbine startup, shutdown, or rapid load changes, the interference signal is strong and fluctuates dramatically. In these conditions, the step-size factor automatically increases, accelerating the convergence speed of the filter coefficients, quickly adapting to changes in the interference signal, and effectively reducing interference characteristics. Conversely, under stable operating conditions, the interference signal is relatively weak, and the step-size factor automatically decreases, improving filter stability and avoiding over-adjustment of the actual speed information. This adaptive capability ensures stable and accurate speed information output under various operating conditions, maintaining the stable operation of the turbine.

[0020] (3) The adaptive filtering algorithm can monitor signal changes in real time and ensure that the phase of the true speed information is not changed or only has a very small and negligible effect during the process of eliminating interference features. This avoids the phase lag problem that traditional fixed parameter filters are prone to when processing signals, which leads to the measured speed and the actual speed being out of sync in time, affecting the timeliness and accuracy of control. This allows the measured speed information to reflect the actual operating state of the turbine in real time and accurately. For subsystems such as the turbine speed regulation system and excitation system that rely on real-time speed information for precise control, this greatly improves the control accuracy and response speed, and effectively avoids control errors and operational instability caused by phase lag. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the method steps for measuring the speed of a turbine gear disk based on adaptive disturbance reduction in this application. Figure 2 This is a flowchart illustrating the method steps for performing spectral analysis on the speed information in step S2 of this application. Figure 3 This is a flowchart illustrating the method steps for performing spectral analysis on the digital signal by optimizing the Fast Fourier Transform algorithm in step S22 of this application. Figure 4 This is a flowchart illustrating the method of windowing the digital signal by selecting an appropriate window function based on the speed information characteristics of different operating environments of the water turbine in step S22 of this application. Figure 5 This is a flowchart illustrating the method steps for establishing an interference signal feature number identification model and identifying interference features of the speed information in step S3 of this application. Figure 6 This is a flowchart illustrating the method steps in step S32 of this application for dynamically adjusting the kernel function parameters of the support vector machine algorithm based on cross-validation when optimizing the support vector machine algorithm. Figure 7 This is a flowchart illustrating the method steps in step S33 of this application, which involves selecting the Leaky ReLU function and optimizing the α parameter when optimizing the artificial neural network algorithm. Figure 8 This is a flowchart illustrating the method steps in step S34 of this application for fusing the recognition results of the support vector machine algorithm and the artificial neural network algorithm to obtain the final interference feature recognition result. Figure 9 This is a flowchart illustrating the method steps in step S4 of this application for eliminating the interference features based on a filtering algorithm to obtain accurate speed information. Figure 10This is a flowchart illustrating the method steps in step S42 of this application, which involves continuously updating and iterating the filter coefficients according to the adaptive minimum mean square error filtering algorithm until they converge to the optimal value. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0023] The adaptive disturbance cancellation-based turbine gear disc speed measurement method of this application is mainly applied to... like Figure 1 As shown, this application provides a method for measuring the speed of a turbine gear disk based on adaptive disturbance cancellation, which includes the following steps: S1. The speed information of the turbine gear disc is collected in real time through a speed sensor.

[0024] S2. Perform spectral analysis on the speed information.

[0025] S3. Establish an interference signal feature data identification model to identify interference features in the speed information.

[0026] S4. Eliminate the interference features based on the filtering algorithm to obtain accurate speed information.

[0027] This application digitizes and performs spectral analysis on the acquired speed information electrical signal to obtain its frequency components. Then, by establishing an interference signal characteristic data identification model, and based on the differences in frequency characteristics between the actual speed information and the interference signal, the interference signal components are accurately identified. Subsequently, a filtering algorithm is used to dynamically reduce interference features. This ensures that the true speed information can be restored to the greatest extent possible under various complex interference environments, avoiding distortion of the measurement results by the interference signal and improving measurement accuracy.

[0028] Specifically, in step S1, a high-sensitivity, high-reliability electromagnetic induction speed sensor is selected. This type of sensor is based on the principle of electromagnetic induction and can convert the magnetic field changes generated when the gear disk rotates into electrical signals. To improve the accuracy and stability of signal acquisition, two electromagnetic induction speed sensors are symmetrically arranged on the radial outer side of the gear disk. The gap between the two electromagnetic induction speed sensors and the gear disk is adjusted by a precision adjustment mechanism to ensure that the gap is uniform and maintained at approximately 1.5 mm. At the same time, the installation position of the electromagnetic induction speed sensors must avoid areas that may generate strong electromagnetic interference, such as near the excitation winding of a generator. To further enhance anti-interference capabilities, the sensor housing is shielded with a soft magnetic material with high magnetic permeability, such as permalloy, effectively blocking the influence of external stray magnetic fields on the sensor.

[0029] like Figure 2 As shown, in step S2, the velocity information is subjected to spectral analysis, which includes the following steps: S21. The speed information is converted into a digital signal by ADC conversion.

[0030] S22. Perform spectrum analysis on the digital signal by optimizing the Fast Fourier Transform algorithm.

[0031] Specifically, in step S21, a high-performance DSP chip is used, such as the TMS320C6678 chip from TI. The ADC module of this chip supports sampling frequencies up to 250kHz and a sampling accuracy of 24 bits. This allows for extremely high-speed sampling of the analog electrical signal from the electromagnetic induction speed sensor and conversion into a digital signal with 24-bit precision. This captures extremely subtle changes in the signal, providing a detailed data foundation for subsequent precise spectral analysis of the digital signal.

[0032] like Figure 3 As shown, in step S22, the digital signal is subjected to spectral analysis by optimizing the Fast Fourier Transform algorithm, including the following steps: When optimizing the Fast Fourier Transform algorithm, perform the following steps: S221. Calculate the input data address for each butterfly operation using formula (1), and access the data directly in the required order during the butterfly operation. For the k-th level butterfly operation, formula (1) is expressed as: ,in, The address of the input data for the i-th butterfly operation. Here, i represents the starting address of the data storage, and i represents the sequence number of the butterfly operation. This indicates a bitwise XOR operation. Stride is the interval of the current butterfly operation, and Stride is the step size for data storage.

[0033] S222. Based on the speed information characteristics of the turbine at different operating stages, dynamically adjust the number of points in the Fast Fourier Transform algorithm.

[0034] Existing Fast Fourier Transform (FFT) algorithms for spectral analysis typically employ radix-2 decimation-time (DRC), where data needs to be rearranged before and after the butterfly operation. This is because the FFT algorithm's computational structure requires data to be processed in a specific order, while the actual data storage order is often different. For example, for an N-point FFT operation, assuming the input data is stored in array x[n], the data needs to be read in bit-reversed order during the butterfly operation. The traditional approach is to first re-store the data in bit-reversed order into another array before performing the butterfly operation. This data rearrangement not only occupies additional memory space but also increases data transmission time, significantly impacting the FFT algorithm's efficiency, especially when N is large.

[0035] To overcome the shortcomings of traditional data rearrangement, in step S221, a special address generation logic is formed using formula (1), which can be implemented using the address calculation unit and index register inside the DSP chip. This allows data to be accessed directly in the required order during butterfly operations without additional data rearrangement. Taking the Fast Fourier Transform of a point as an example, for the k-th level butterfly operation, formula (1) is expressed as: ,in, The address of the input data for the i-th butterfly operation. Here, i represents the starting address of the data storage, and i represents the sequence number of the butterfly operation. This indicates a bitwise XOR operation. The interval for the current butterfly operation is defined by `Stride`, which is the step size for data storage, typically the number of bytes in the data type. This address generation method allows data to be read directly from the original data storage location in the required order during butterfly operations, significantly reducing the number of data storage and transmission operations and improving the algorithm's execution efficiency.

[0036] This optimized data rearrangement technique significantly reduces memory accesses and data transfer time when processing large amounts of Fast Fourier Transform (FFT) data. Experiments show that for an N=4096-point FFT operation, the optimized address generation method reduces the algorithm's execution time by approximately 30% compared to traditional data rearrangement methods, effectively improving the real-time performance of the FFT algorithm for spectral analysis.

[0037] The frequency characteristics of the turbine's speed signal vary significantly across different operating phases. During startup, the turbine speed gradually increases from zero, and the signal frequency gradually increases from a low frequency, with relatively slow frequency changes. During shutdown, the speed gradually decreases, and the signal frequency also decreases, exhibiting the same low-frequency, slow-changing characteristics. However, during normal, stable operation, the turbine speed remains near its rated speed, and the signal frequency is relatively high and stable.

[0038] Therefore, in step S222, the number of points in the Fast Fourier Transform (FFT) algorithm is dynamically adjusted based on the speed information characteristics of the turbine at different operating stages. Specifically, during the startup and shutdown phases, a larger number of FFT points is selected to improve the frequency resolution in the low-frequency band. For example, the FFT point count is set to 8192 points. A larger FFT point count means higher resolution in the frequency domain, enabling clearer differentiation of the frequency components of low-frequency interference signals.

[0039] According to the frequency resolution formula , where f s Where f is the sampling frequency, and N is the number of points in the Fast Fourier Transform. s At 1000Hz, an 8192-point Fast Fourier Transform (FFT) operation provides a frequency resolution of approximately 0.122Hz. Compared to a smaller FFT point count, this allows for more accurate analysis of low-frequency signals. During normal, stable operation, a smaller FFT point count, such as 512 points, is recommended. This ensures sufficient frequency resolution for high-frequency analysis while reducing computational load and improving the algorithm's real-time performance. This is because small changes in frequency resolution have minimal impact on the analysis results at high frequencies, and reducing the number of FFT points significantly reduces the number of computations, thus increasing the algorithm's execution speed.

[0040] When performing spectral analysis on digital signals, since the actual acquired signal is of finite length, this is equivalent to truncating the infinitely long original signal. According to signal processing theory, truncation in the time domain is equivalent to multiplying the frequency domain signal by a rectangular window function. The spectrum of the rectangular window function in the frequency domain is a sinc function with a main lobe and side lobes. The width of the main lobe determines the main energy distribution range of the signal spectrum, while the side lobes cause spectral leakage.

[0041] Spectral leakage can severely impact the accuracy of spectral analysis, blurring previously clearly distinguishable frequency components and making it difficult to accurately determine the true frequency information in a signal. This is particularly true in hydroelectric turbine signal analysis, where the frequency components of interfering signals and the true rotational speed signal may overlap. Spectral leakage further exacerbates this confusion, making it impossible to accurately identify interfering signals and thus affecting the effectiveness of adaptive interference cancellation techniques.

[0042] like Figure 4 As shown, to address the issue that spectral leakage can severely impact the accuracy of spectral analysis, in step S22, by optimizing the Fast Fourier Transform algorithm, when performing spectral analysis on the digital signal, an appropriate window function is selected to window the digital signal based on the speed information characteristics of different operating environments of the turbine. This includes the following steps: S223. Perform feature analysis on the digital signal, including frequency range, amplitude variation and noise intensity.

[0043] S224. Based on the feature analysis results, formulate an appropriate window function selection strategy.

[0044] S225. When switching between different window functions, a smooth transition is achieved through a linear weighting algorithm.

[0045] Specifically, in step S233, a dedicated signal characteristic monitoring module is set in the DSP chip to perform real-time analysis of the acquired signal. The monitored signal characteristics include the signal's frequency range, amplitude variation, and noise intensity. For example, noise intensity can be estimated through statistical analysis of the signal amplitude over a period of time; and the main frequency range of the signal can be determined through preliminary analysis of the signal spectrum.

[0046] In step S224, during feature analysis, if the monitored signal frequency is mainly concentrated in the low-frequency band, for example, when the turbine speed signal contains many low-frequency components and has little noise interference, the Hanning window function is selected to window the acquired signal. The Hanning window function is a commonly used weighted window function, and its expression is: in Let be the signal length. The Hanning window function is characterized by its smooth weighting of the signal in the time domain, with smaller weights at the ends and larger weights in the middle. This weighting method results in a relatively narrow main lobe width and relatively low side lobe amplitudes in the frequency domain. Compared to the rectangular window function, the Hanning window function has approximately twice the main lobe width, but significantly reduced side lobe amplitudes, effectively suppressing spectral leakage.

[0047] In feature analysis, if a large amount of high-frequency noise is detected in the signal, or if the frequency components are complex and the interference is severe, such as when there is strong electromagnetic interference or mechanical vibration interference in the operating environment of a water turbine, the Blackman window function is selected, and its expression is: , in The signal length is given. Compared to the Hanning window function, the Blackman window function has a more complex weighting of the signal in the time domain, resulting in a lower sidelobe level in the frequency domain. This better suppresses spectral leakage of interference signals, making the spectrum of the true speed signal more prominent in complex interference environments. By applying the Blackman window to the actual acquired turbine speed signal subjected to strong interference, followed by a Fast Fourier Transform, the spectral leakage of the interference signal on the spectrum diagram is greatly suppressed, allowing the frequency components of the true speed signal to be clearly distinguished. This provides a more reliable spectral analysis basis for accurately identifying and reducing interference signals.

[0048] In step S225, when the system determines that a window function switch is needed, it first buffers the currently acquired signal. Then, it performs windowing processing and Fast Fourier Transform (FFT) on the buffered signal using both the window function to be switched and the current window function. By comparing the spectral results after processing with the two window functions, a smooth transition method is used to switch the spectral analysis results from the current window function to the new window function. For example, a linear weighting method can be used to gradually increase the weight of the new window function processing result and decrease the weight of the current window function processing result over a certain period of time, ensuring a smooth transition of the spectral analysis results. This guarantees that the FFT can run continuously and stably during the window function switching process, accurately analyze the frequency components in the signal, and provide a reliable basis for subsequent interference reduction.

[0049] like Figure 5 As shown, in step S3, an interference signal feature number identification model is established to identify interference features of the speed information, including the following steps: S31. Construct an interference feature database based on the interference signals generated during the operation of the water turbine.

[0050] S32. Optimize the support vector machine algorithm and train it based on the interference feature database.

[0051] S33. Optimize the artificial neural network algorithm and train it based on the interference feature database.

[0052] S34. The recognition results of the support vector machine algorithm and the artificial neural network algorithm are fused to obtain the final interference feature recognition result.

[0053] Specifically, in step S31, the construction of the interference feature database may include the following two parts: For mechanical vibration interference signals, historical data detailing their characteristics under different operating conditions is recorded. Beyond frequency range, amplitude variation patterns, phase characteristics, and waveform features, the correlation between vibration interference and turbine structural components is further analyzed. For example, through dynamic modeling and simulation of key components such as the turbine runner and shaft system, combined with actual operational monitoring data, the unique characteristics of vibration interference caused by faults in different components in terms of frequency and amplitude are determined. When cracks appear in the runner blades, the vibration interference signal may experience a sudden amplitude change at a specific frequency, and this frequency is related to the blade's natural frequency. Through long-term accumulation and analysis of a large amount of such data, the description of mechanical vibration interference characteristics in the database is continuously refined and improved, making it more accurate and practical.

[0054] For electromagnetic interference signals, in addition to recording and analyzing frequency distribution characteristics, the relationship between amplitude and electromagnetic source strength, and signal modulation methods, the propagation path and coupling mode of electromagnetic interference are also analyzed. Through electromagnetic compatibility testing and theoretical analysis, the propagation characteristics of interference signals generated by different electromagnetic sources, such as generator excitation systems and power electronic equipment, in the turbine speed measurement system are determined, including the distribution of electric and magnetic fields and the way they couple to the signal lines. For example, high-frequency electromagnetic interference may enter the signal lines of the speed sensor through capacitive or inductive coupling, causing specific signal distortions. Simultaneously, the joint characteristics of different types of electromagnetic interference in the time and frequency domains are analyzed, such as the pulse width and interval of pulse-type electromagnetic interference in the time domain and the harmonic distribution in the frequency domain. These detailed electromagnetic interference characteristics are incorporated into a database to provide richer evidence for accurate identification of electromagnetic interference.

[0055] like Figure 6 As shown, in step S32, when optimizing the support vector machine algorithm, the kernel function parameters of the support vector machine algorithm are dynamically adjusted based on cross-validation, including the following steps: S321. The training dataset is randomly divided into k disjoint subsets using the k-fold cross-validation method.

[0056] S322. Set an initial range of kernel function parameters, starting from a smaller kernel function parameter value and gradually increasing it to a larger kernel function parameter value with a set first step size, to form multiple candidate kernel function parameter values.

[0057] S323. For each candidate value of the kernel function parameter, perform K-fold cross-validation to select the optimal solution for the kernel function parameter.

[0058] Specifically, in step S321, since the interference feature database contains various types of interference feature signals and the data distribution may be uneven, k-fold cross-validation helps overcome the bias caused by data partitioning. Therefore, when dynamically adjusting the kernel function parameters, the entire training dataset is randomly divided into k disjoint subsets, each with approximately equal data volume, where k typically ranges from 5 to 10. This partitioning method can comprehensively evaluate the model's performance on different data subsets while ensuring computational efficiency. For example, in a certain partition, one subset may contain more electromagnetic interference signals, while another subset may have a larger proportion of mechanical vibration interference signals. Cross-validation allows the model to be trained and validated on subsets with different types of interference signal distributions, thereby more accurately evaluating the adaptability of the kernel function parameters to the overall data.

[0059] In step S322, the commonly used Gaussian kernel function for the support vector machine algorithm is expressed as follows: Its kernel function parameters It plays a crucial role in the classification performance of the support vector machine algorithm. When the value is small, the Gaussian kernel function has strong locality and can more sensitively capture the local features of the data, but it may lead to an overly complex classification hyperplane, which is prone to overfitting. When the value is large, the globality of the Gaussian kernel function is enhanced, and the classification hyperplane is relatively smooth, but it may ignore some important local features, leading to underfitting.

[0060] In order to determine the optimal kernel function parameters The value needs to be defined within a comprehensive and detailed range of initial parameter candidates. Starting with a relatively small value, such as... Start with a value of 0.01, gradually increasing to a larger value, such as 0.01. =10. This creates a rich pool of candidates. value: .

[0061] In step S323, for each candidate All candidate values ​​must undergo rigorous k-fold cross-validation. For example, in each k-fold cross-validation with K=5, four subsets are used as the training set to train a support vector machine (SVM) algorithm model based on a Gaussian kernel function. The specific training process is existing technology and will not be elaborated here. The remaining subset is used as the validation set to evaluate the model's performance. Evaluation metrics cover multiple key indicators such as classification accuracy, recall, and F1 score. For example, in a certain cross-validation, after training the SVM algorithm model using the training set, the classification accuracy on the validation set was 80%, the recall was 75%, and the F1 score was 77.5%. After all values ​​have undergone k-fold cross-validation, record each value in detail. The average evaluation metric corresponding to the value. After comprehensive cross-validation, it was found that when When the accuracy is 1.2, the average classification accuracy reaches its highest value of 93%, at which point it can be preliminarily determined that... =1.2 is the optimal parameter for the Gaussian kernel function in the current dataset.

[0062] like Figure 7 As shown, in step S33, when optimizing the artificial neural network algorithm, the Leaky ReLU function is selected, and the α parameter is optimized, including the following steps: S331. Set an initial range for the α parameter, and gradually increase it to a larger α parameter value starting from a smaller α parameter value with a set second step size to form multiple candidate α parameter values; S332. For each α parameter value, construct a neural network with the same structure and use the same perturbation feature dataset for training and testing. S333. During the training process, the optimal solution for the α parameter is selected from the convergence speed and classification accuracy as evaluation indicators.

[0063] In artificial neural network algorithms, the activation function determines the output characteristics of neurons and plays a crucial role in the nonlinear mapping capability of the neural network. For the task of identifying interference signals from water turbines, the ReLU function is a commonly used and effective activation function. Its expression is: f(x) = max(0,x). The ReLU function can effectively solve the gradient vanishing problem and accelerate the network convergence speed.

[0064] However, the traditional ReLU function always outputs 0 when the input value is less than 0, which can lead to the "death" of neurons during training, meaning that some neurons stop updating during training. To overcome this problem, the Leaky ReLU function is used as an improvement, and its expression is: ,in It is a small constant, typically taking a value between 0 and 1. This small non-zero slope This ensures that even when the input is negative, the neuron can still update with some gradient. For example, when... When x = 0.01, if the input x of a neuron = 1, its output is At a gradient of 0.01, during backpropagation, the neuron will still have a small gradient, allowing its weights to be updated and maintaining its contribution to the neural network output. In this way, the Leaky ReLU function effectively avoids the complete "death" of neurons, improves the network's ability to learn complex interference signal features, and enhances the network's training performance.

[0065] In step S331, during the training process, different methods are compared experimentally. The impact of the value on neural network performance. Similar to step S322, in order to determine the optimal... The value needs to be defined within a comprehensive and detailed range of initial parameter candidates. Starting with a relatively small value, such as... Start with a value of 0.001, and gradually increase it to a larger value, such as 0.001 in the second step. =0.1. This creates a rich pool of candidate variables. value: .

[0066] In step S332, for each We construct a neural network with an identical structure, including the previously determined number of hidden layers and nodes. Then, we train and test it using the same turbine interference signal dataset. The specific training process is existing technology and will not be elaborated here. The dataset is divided into a training set, a validation set, and a test set. The training set is used to train the neural network, and the validation set is used to evaluate different... To assess the performance of the model under different values, so as to select the optimal one. The test set is used to ultimately evaluate the model's generalization ability. To ensure the reliability of the experimental results, each... The experiments were repeated multiple times, for example, 5 times, and the results were statistically analyzed.

[0067] In step S333, during training, the main focus is on two key performance indicators: convergence speed and classification accuracy. Convergence speed is measured by observing how the loss function changes with the number of training epochs. If the loss function decreases rapidly and stabilizes within a relatively small number of training epochs, it indicates that the network converges quickly. For example, using different... When training a neural network, the number of training epochs required for the loss function to decrease to a certain threshold, such as 0.1, is recorded. Classification accuracy is a direct metric for model performance, calculated by classifying interfering signals on the validation and test sets and determining the proportion of correctly classified samples out of the total number of samples. For example, if there are 1000 interfering signal samples in the validation set, and the neural network model correctly classifies 900, then the classification accuracy is 90%.

[0068] After conducting multiple experiments for each α value, the results were comprehensively analyzed. First, the convergence of the loss function and classification accuracy were recorded and statistically analyzed for each experiment. Then, by comparing the various indicators under different α values, the α value that resulted in the fastest convergence speed and the highest classification accuracy was identified. For example, after a series of experiments, it was found that when α=0.01, the loss function tended to stabilize after approximately 100 training rounds, achieving a classification accuracy of 92% on the validation set and 90% on the test set, showing the best performance compared to other α values. Therefore, α=0.01 was determined as the optimal parameter for the Leaky ReLU function in the turbine interference signal recognition neural network, effectively avoiding the "death" problem of neurons while ensuring the training effect and classification performance of the neural network. In practical applications, the α value can be dynamically adjusted according to changes in the turbine's operating conditions to further optimize model performance. For example, when the turbine's operating conditions change and the characteristics of the interference signal change accordingly, experiments can be repeated to determine a new optimal α value to adapt to the new signal characteristics.

[0069] like Figure 8 As shown, in step S34, the recognition results of the support vector machine algorithm and the artificial neural network algorithm are fused to obtain the final interference feature recognition result, including the following steps: S341. Based on the historical accuracy of the support vector machine algorithm and the artificial neural network algorithm in the identification of different types of interference feature signals, assign different weights to each algorithm.

[0070] S342. Based on the recognition results and corresponding weights of the support vector machine algorithm and the artificial neural network algorithm, recalculate the corresponding basic probability allocation function.

[0071] S343. Based on the DS evidence theory fusion method, the basic probability allocation functions of the support vector machine algorithm and the artificial neural network algorithm are fused, and the final interference feature identification result is obtained according to the fused basic probability allocation function.

[0072] In step S341, when training the support vector machine algorithm and the artificial neural network algorithm using the interference feature database, different weights are assigned to each algorithm based on their historical accuracy in identifying different types of interference signals.

[0073] Specifically, for the task of identifying interference signals from water turbines, the accuracy rates of support vector machine (SVM) and artificial neural network (ANN) algorithms in identifying various interference signals, such as mechanical vibration interference, electromagnetic interference, and water flow interference, were statistically analyzed through the study of a large amount of historical data. For example, the statistics showed that the SVM algorithm had a higher accuracy rate of 90% in identifying electromagnetic interference signals, while the ANN algorithm performed better in identifying mechanical vibration interference signals, with an accuracy rate of 85%.

[0074] Based on these historical accuracy rates, weights are assigned to the Support Vector Machine (SVM) and Artificial Neural Network (ANN) algorithms. For example, the accuracy rates are normalized to obtain the weights. Here, it is assumed that the historical average accuracy of the SVM algorithm is 1.5%. The historical average accuracy of artificial neural network algorithms is The weights of the support vector machine algorithm are: The weights of the artificial neural network algorithm are In the example above, , These weights reflect the relative reliability of each algorithm in identifying different interference signals. During the fusion process, the algorithm with higher accuracy will have a greater impact on the final result.

[0075] In step S342, the basic probability allocation function is recalculated based on the recognition results and corresponding weights of the Support Vector Machine (SVM) and Artificial Neural Network (ANN) algorithms. Specifically, when the SVM algorithm classifies an interference signal sample, it outputs the probability value of the sample belonging to each interference signal category. These probability values ​​are adjusted to satisfy the conditions of the basic probability allocation function, i.e., the probability values ​​of all categories are normalized. Here, it is assumed that the SVM algorithm outputs that the interference signal sample belongs to category [category missing]. The probability is ( If ), then the adjusted basic probability assignment function is expressed as: .

[0076] Similarly, the interference signal samples output by the artificial neural network algorithm belong to the category The probability is ( If ), then the adjusted basic probability assignment function is expressed as: .

[0077] In step S343, during the identification of turbine interference signals, the identification results of the Support Vector Machine (SVM) algorithm and the Artificial Neural Network (ANN) algorithm are considered as different sources of evidence. Each source of evidence classifies and judges the interference signal based on its own algorithm and model. However, due to the complexity of turbine interference signals and the limitations of the algorithms themselves, the judgment of a single source of evidence may be uncertain. Therefore, the DS evidence theory fusion method is a method for handling uncertain reasoning, providing an effective framework for multi-source information fusion.

[0078] The core of the Dempster evidence fusion method is the Dempster synthesis rule, which is used to fuse the basic probability assignment functions of multiple evidence sources. Here, it is assumed that the support vector machine algorithm and the artificial neural network algorithm are evidence sources m respectively. S and m A Their basic probability assignment functions are m S (B) and m A (C), which is equivalent to the above-mentioned and The fused basic probability assignment function m(A) is calculated by the following formula: ,in, This indicates the degree of conflict between two sources of evidence. If K=1, it means the two sources of evidence are completely conflicting and cannot be directly fused; if K≠1, the information from the two sources is merged using the above formula to obtain the fused basic probability allocation function. The category of the interference signal is determined based on the fused basic probability allocation function, and the signal with the largest m( The category of the fused m() value is used as the final recognition result. For example, if the fused m( The largest value indicates that the interference signal sample belongs to category [ ]. This approach fully leverages the advantages of support vector machine and artificial neural network algorithms, improving the accuracy and reliability of interference signal identification.

[0079] like Figure 9 As shown, in step S4, the interference features are eliminated based on a filtering algorithm to obtain accurate speed information, including the following steps: S41. In the adaptive minimum mean square error filtering algorithm, a variable step size factor is introduced. The variable step size factor is dynamically adjusted according to the statistical characteristics of the error signal and the input signal at the current time. The input signal is defined as the speed information after interference feature identification, and the error signal is defined as the difference between the expected output signal and the output signal after filtering algorithm processing.

[0080] S42. Based on the adaptive minimum mean square error filtering algorithm, continuously update the filter coefficients until they converge to the optimal value.

[0081] S43. Set the update range of the filter coefficients. By monitoring and analyzing the filter output signal in real time, when abnormal fluctuations are found in the filter output signal, the initial filter coefficients of the filter are automatically adjusted.

[0082] Existing adaptive minimum mean square error filtering algorithms aim to minimize the mean square error between the filter output signal and the desired output signal by continuously adjusting the filter coefficients, thereby eliminating interference characteristics. In the scenario of turbine gear disk speed signal processing, assuming the input signal is x(n), and the filter coefficient vector is... , where M is the order of the filter. The output y(n) of the filter is obtained by convolving the input signal with the filter coefficients, i.e. The desired output signal d(n) can be understood as the relatively clean rotational speed signal obtained after processing at the previous moment. Initially, the acquired raw signal can be used as the desired output signal. The error signal e(n) is defined as the difference between the desired output signal and the filter output signal, i.e., e(n) = d(n). y(n).

[0083] The adaptive minimum mean square error filtering algorithm updates the filter coefficients using the following iterative formula: ,in, This is the step size factor, which controls the size of the step size for updating the filter coefficients. The choice of [value] plays a crucial role in the convergence speed and stability of the algorithm. A larger [value] A smaller value can speed up convergence, but may lead to algorithm instability, oscillations, or even divergence; While this value can ensure the stability of the algorithm, the convergence speed will become very slow.

[0084] Therefore, in step S41, in order to avoid step size factor To mitigate the negative impact on the algorithm's convergence speed and stability and improve algorithm performance, a variable step size factor mechanism is introduced. The filter is dynamically adjusted based on the statistical characteristics of the current error signal e(n) and the input signal x(n). Specifically, when the error signal e(n) is large, it indicates a significant deviation between the current output signal and the desired output signal; in this case, the variable step size factor is increased. This accelerates the convergence speed of the filter coefficients, enabling rapid tracking of changes in the interference signal. When the error signal e(n) is small, it indicates that the current filter's output signal is close to the desired output signal; at this point, the variable step size factor should be reduced. This improves the stability of the filter and avoids distortion of the real speed signal caused by excessive adjustment of the filter coefficients.

[0085] Formula (2) can be used: Calculate the variable step size factor ,in and These are the maximum and minimum values ​​of the variable step size factor, respectively. It is a constant used to adjust the rate of change of the step size factor.

[0086] like Figure 10 As shown, in step S42, the filter coefficients are continuously updated and iterated according to the adaptive minimum mean square error filtering algorithm until they converge to the optimal value, including the following steps: S421. In each iteration, based on the input signal at the current moment... Sum of error signals According to formula (2): Calculate the variable step size factor ,in and These are the maximum and minimum values ​​of the variable step size factor, respectively. This constant is used to adjust the rate of change of the step size factor; for example, it is set according to the actual operating conditions of the turbine. =0.1, =0.001, =0.01.

[0087] Through this variable step size mechanism, the optimized adaptive minimum mean square error filtering algorithm can maintain good performance under different interference intensities. For example, during the turbine startup phase, the interference signal is strong and varies greatly, and the error signal e(n) is large. At this time, the variable step size factor is calculated using formula (2). The adaptive increase allows the filter to quickly adapt to changes in the interference signal and effectively reduce interference. During the stable operation phase of the turbine, the interference signal is relatively weak and the error signal e(n) is small. At this time, the variable step size factor is calculated using formula (2). Adaptive reduction ensures the stability of the filter and avoids interference with the actual rotational speed signal.

[0088] S422. Based on the calculated variable step size factor The filter coefficients are updated. During the process of eliminating interference features, the filter coefficients are continuously updated according to the optimized adaptive minimum mean square error algorithm. Since the filter adopts a finite impulse response structure, its filter coefficients are adjusted through iterative calculation. In each iteration, a new step size factor is calculated according to the variable step size factor formula based on the current input signal x(n) and error signal e(n). Then, the filter coefficients are updated. Taking a second-order FIR filter as an example, the coefficient update process is as follows: Assumption and For the two coefficients of the filter, the formula is updated according to the optimized adaptive minimum mean square error algorithm: ,in, These are the filter coefficients at the current time. These are the updated filter coefficients.

[0089] ,in, These are the filter coefficients at the current time. Here are the updated filter coefficients, and x(n-1) is the input signal from the previous time step.

[0090] As the iteration proceeds, the filter coefficients gradually converge to the optimal value, enabling the filter to accurately subtract interference components from the acquired velocity information.

[0091] In step S43, to ensure the stability and convergence of the filter coefficients, the update range of the filter coefficients is limited. By setting a maximum change in the filter coefficients, sudden changes in the coefficients under certain circumstances can be prevented, which could lead to a deterioration in filter performance. For example, it is stipulated that the change in each filter coefficient during each update cannot exceed ±0.1. If the update amount of the filter coefficients calculated according to the adaptive minimum mean square error algorithm exceeds this range, the update amount is limited to within ±0.1.

[0092] Meanwhile, by monitoring and analyzing the filter output signal in real time, when abnormal fluctuations are detected in the filter output signal, the initial coefficients of the filter are automatically adjusted or the adaptive filtering process is restarted. For example, if the amplitude of the filter output signal suddenly increases or periodic oscillations are detected, it indicates that the filter may have fallen into a local optimum or been affected by sudden interference. In this case, the filter coefficients can be reset to their initial values ​​and the adaptive filtering process can be restarted to ensure that the filter can continuously and accurately reduce interference signals.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0094] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0095] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for measuring the speed of a turbine gear disc based on adaptive disturbance cancellation, characterized in that, Includes the following steps: The speed information of the turbine gear disc is collected in real time by a speed sensor; Perform spectral analysis on the velocity information; An interference signal feature data identification model is established to identify interference features in the speed information; The interference features are eliminated based on the filtering algorithm to obtain accurate speed information.

2. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 1, characterized in that, The spectral analysis of the velocity information includes the following steps: The speed information is converted into a digital signal through ADC conversion; The digital signal is subjected to spectral analysis by optimizing the Fast Fourier Transform algorithm.

3. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 2, characterized in that, The digital signal is subjected to spectral analysis by optimizing the Fast Fourier Transform algorithm, including the following steps: When optimizing the Fast Fourier Transform algorithm, perform the following steps: The input data address for each butterfly operation is calculated using formula (1), and the data is accessed directly in the required order during the butterfly operation. For the k-th level butterfly operation, formula (1) is expressed as: ,in, The address of the input data for the i-th butterfly operation. Here, i represents the starting address of the data storage, and i represents the sequence number of the butterfly operation. This indicates a bitwise XOR operation. Stride is the interval of the current butterfly operation, and Stride is the step size for data storage. The number of points in the Fast Fourier Transform algorithm is dynamically adjusted based on the speed information characteristics of the turbine at different operating stages.

4. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 2, characterized in that, Spectral analysis of the digital signal by optimizing the Fast Fourier Transform algorithm also includes the following steps: Based on the speed information characteristics of the turbine under different operating environments, an appropriate window function is selected to perform windowing processing on the digital signal, including the following steps: The digital signal is subjected to feature analysis, including frequency range, amplitude variation, and noise intensity; Based on the feature analysis results, an appropriate window function selection strategy is formulated; When switching between different window functions, a linear weighted algorithm is used for a smooth transition.

5. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 1, characterized in that, An interference signal feature identification model is established, and interference feature identification is performed on the speed information, including the following steps: An interference feature database was constructed based on the interference signals generated during the operation of the water turbine. The support vector machine algorithm was optimized and trained based on the aforementioned interference feature database; The artificial neural network algorithm is optimized and trained based on the aforementioned interference feature database; The recognition results of the support vector machine algorithm and the artificial neural network algorithm are fused to obtain the final interference feature recognition result.

6. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 5, characterized in that, When optimizing the support vector machine algorithm, the kernel function parameters of the support vector machine algorithm are dynamically adjusted based on cross-validation, including the following steps: The training dataset is randomly divided into k disjoint subsets using k-fold cross-validation. Set an initial range of kernel function parameters, and gradually increase it from a small kernel function parameter value to a larger kernel function parameter value with a set first step size to form multiple candidate kernel function parameter values; For each candidate value of the kernel function parameter, K-fold cross-validation is performed to select the optimal solution for the kernel function parameter.

7. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 5, characterized in that, When optimizing artificial neural network algorithms, the Leaky ReLU function is selected, and the α parameter is optimized, including the following steps: Set an initial range for the α parameter, and gradually increase it to a larger α parameter value starting from a smaller α parameter value with a set second step size to form multiple candidate α parameter values; For each α parameter value, a neural network with the same structure is constructed, and the same perturbation feature dataset is used for training and testing. During training, the optimal solution for the α parameter is selected by using convergence speed and classification accuracy as evaluation metrics.

8. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 5, characterized in that, The recognition results of the support vector machine algorithm and the artificial neural network algorithm are fused to obtain the final interference feature recognition result, including the following steps: Based on the historical accuracy of the support vector machine algorithm and the artificial neural network algorithm in the identification of different types of interference feature signals, different weights are assigned to each algorithm. Based on the recognition results and corresponding weights of the support vector machine algorithm and the artificial neural network algorithm, the corresponding basic probability allocation function is recalculated. Based on the DS evidence theory fusion method, the basic probability allocation functions of the support vector machine algorithm and the artificial neural network algorithm are fused, and the final interference feature identification result is obtained according to the fused basic probability allocation function.

9. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 1, characterized in that, To eliminate the interference features based on the filtering algorithm and obtain accurate speed information, the following steps are included: In the adaptive minimum mean square error filtering algorithm, a variable step size factor is introduced. The variable step size factor is dynamically adjusted according to the statistical characteristics of the error signal and the input signal at the current time. The input signal is defined as the speed information after interference feature identification, and the error signal is defined as the difference between the expected output signal and the output signal after filtering algorithm processing. The filter coefficients are continuously updated according to the adaptive minimum mean square error filtering algorithm until they converge to the optimal value. The update range of the filter coefficients is set, and the filter output signal is monitored and analyzed in real time. When abnormal fluctuations are detected in the filter output signal, the initial filter coefficients are automatically adjusted.

10. The turbine gear disc speed measurement method based on adaptive disturbance cancellation according to claim 9, characterized in that, The adaptive minimum mean square error filtering algorithm continuously updates and iterates the filter coefficients until it converges to the optimal value, including the following steps: In each iteration, based on the input signal at the current moment... Sum of error signals According to formula (2): Calculate the variable step size factor ,in and These are the maximum and minimum values ​​of the variable step size factor, respectively. It is a constant; Based on the calculated step size factor Update the filter coefficients.

Images