Hydraulic turbine unit operation system and method based on digital twinning

Abstract

The present application relates to a digital twin-based water turbine unit operation system and method, comprising a water turbine unit top cover, a plurality of soundprint collection arrays are arranged on the water turbine unit top cover, each soundprint collection array comprises a plurality of soundprint collectors arranged in a vertical direction and having a fixed interval. The soundprint collector is used to collect original soundprint data at different positions on the water turbine unit top cover, and send the collected original soundprint data to a data center. After receiving the original soundprint data, the data center pre-processes the original soundprint data, establishes a digital twin benchmark model based on the mapping relationship between real-time working condition parameters and soundprint signal characteristics through digital twin technology, constructs an expected feature vector based on a soundprint collector position compensation vector and a mapping matrix, and adjusts the difference vector using a correction factor. The system uses multiple orientations and multiple soundprint collectors configured in each orientation, and the soundprint collection accuracy is high under extreme water flow conditions.

Description

Technical Field

[0001] This invention relates to the field of digital twin and turbine control technology, specifically to a turbine operation system and method based on digital twin. Background Technology

[0002] Acoustic fingerprint acquisition under extreme water flow conditions during turbine operation remains challenging. In extremely low flow conditions, the turbine's overall operation is relatively stable, with weaker vibrations and sound wave intensity. The acquired acoustic fingerprint signals may be low-amplitude, resulting in a low signal-to-noise ratio (SNR) that is easily masked by background noise. Therefore, cavitation and other anomalies are often accompanied by high-frequency pulse signals, but at low flow velocities, these signals may struggle to elicit typical high-frequency responses, leading to insufficient energy in characteristic frequency bands and unclear spectral characteristics, hindering subsequent identification. In extremely high flow conditions, equipment vibration and liquid impact generate very strong noise, potentially causing excessively high signal amplitudes received by the acoustic fingerprint sensor, leading to sensor saturation and distortion. High-speed flow introduces significant turbulence noise and random noise from water collisions, potentially filling the spectrum with various interference components, making it difficult to separate cavitation and other specific anomaly signals.

[0003] Furthermore, the attenuation, propagation time, and phase shift of the high-frequency pulse noise generated during cavitation are closely related to the distance between the sensor and the cavitation source. At closer distances, signal attenuation is small, amplitude is high, and details are richer; therefore, at greater distances, there will be some attenuation and phase delay. The sound waves generated by cavitation may have a certain directionality; the signal amplitude, spectral distribution, and phase information captured by sensors installed in different directions will vary. For example, sensors located in the main impact direction of cavitation are more likely to capture high-energy signals, while sensors positioned to the side or away may capture relatively weaker signals.

[0004] In summary, the existing technology for acoustic signature acquisition under extreme water flow conditions needs further improvement. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to resolve the aforementioned deficiencies and propose a digital twin-based turbine unit operation system and method.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of the present invention discloses a turbine generator unit operation system based on digital twins. The system includes a turbine generator unit top cover, and multiple sets of acoustic fingerprint acquisition arrays are provided on the turbine generator unit top cover. Each set of acoustic fingerprint acquisition arrays includes multiple acoustic fingerprint collectors arranged in a vertical direction and with a fixed spacing.

[0008] The acoustic fingerprint collector is used to collect raw acoustic fingerprint data from different locations on the top cover of the turbine unit and send the collected raw acoustic fingerprint data to the data center.

[0009] After receiving the original voiceprint data, the data center preprocesses the original voiceprint data and establishes a digital twin benchmark model based on the mapping relationship between real-time operating parameters and voiceprint signal features using digital twin technology. It also constructs an expected feature vector based on the voiceprint collector position compensation vector and mapping matrix, and adjusts the difference vector using a correction factor.

[0010] The actual operating parameters include real-time water flow rate, water flow pressure, and water flow temperature. The difference vector is calculated by preprocessing the actual feature vector of the original voiceprint data actually collected by the voiceprint collector and the expected feature vector.

[0011] The second aspect of this invention discloses a method for operating a hydro-turbine unit based on digital twins, implemented through the hydro-turbine unit operation system based on digital twins described in the first aspect, the method comprising:

[0012] Acquire raw voiceprint data from multiple perspectives and preprocess the raw voiceprint data;

[0013] A real-time operation model of the turbine unit was established using digital twin technology, and a expected spectrum model was generated based on the turbine unit's structure, fluid dynamics, and historical acoustic data.

[0014] Based on the acquisition device location compensation vector and mapping matrix, an expected feature vector based on the voiceprint acquisition device location is constructed, and the difference vector is adjusted by a correction factor.

[0015] The turbine generator set has multiple acoustic signature acquisition arrays on its top cover. Each acoustic signature acquisition array contains multiple acoustic signature acquisition devices arranged vertically with a fixed spacing. The preprocessing includes differential preprocessing, time-frequency domain decomposition, adaptive filtering, and dynamic range adjustment.

[0016] Furthermore, the acquisition of multi-dimensional raw voiceprint data and the preprocessing of the raw voiceprint data include:

[0017] Based on the preprocessed raw acoustic fingerprint data and the parameters of the digital twin benchmark model, a preliminary installation strategy for the acoustic fingerprint collectors on the top cover of the turbine unit is determined. The preliminary installation strategy includes multiple acoustic fingerprint collection arrays and the number, arrangement direction and arrangement spacing of the acoustic fingerprint collectors on each acoustic fingerprint collection array.

[0018] The acoustic fingerprint scanner installed according to the preliminary installation strategy is calibrated, and the sampling frequency, sensitivity and dynamic range are set. The amplitude of the acoustic fingerprint scanner is adjusted by the real-time water flow rate to obtain the calibration parameters.

[0019] Based on the installation position and orientation of the acoustic fingerprint collector on the top cover of the turbine unit, and in conjunction with the aforementioned correction parameters, different compensation strategies are adopted to adjust the amplitude normalization of the acoustic fingerprint collector signal based on the real-time water flow rate. The acoustic fingerprint signal is then subjected to noise reduction and dynamic range expansion processing to obtain the original acoustic fingerprint data after differentiated preprocessing.

[0020] The original acoustic signature data after differential preprocessing is synchronized with real-time water flow data and other operating parameters in time and fused together to form a joint data matrix.

[0021] Other operating parameters include water flow pressure and water flow temperature.

[0022] Furthermore, the process of establishing a real-time operation model of the turbine unit using digital twin technology, and generating a expected spectrum model based on the turbine unit structure, fluid dynamics, and historical acoustic signature data, includes:

[0023] The continuous acoustic signals from each acoustic sensor are segmented according to a fixed time window to form discrete acoustic signal segments. An adjustment factor for each acoustic signal segment is calculated based on the real-time water flow rate. The gain of each acoustic signal segment is then adjusted based on the adjustment factor to obtain a new acoustic signal.

[0024] The new voiceprint signal of each voiceprint collector is decomposed in the time-frequency domain to extract the spectral features of the new voiceprint signal, and differential noise filtering is performed according to the location of the voiceprint collector.

[0025] The time-frequency matrix is ​​extracted by short-time Fourier transform, and the local features of the new voiceprint signal are obtained by combining wavelet transform.

[0026] Furthermore, the step of establishing a real-time operation model of the turbine unit using digital twin technology, and generating a expected spectrum model based on the turbine unit structure, fluid dynamics, and historical acoustic signature data, also includes:

[0027] Key features are extracted from the preprocessed video data of each acoustic signature collector, and the key features are differentially fused. The energy difference of the acoustic signature signals at different locations is corrected by using the compensation information of real-time water flow rate, so as to obtain the fused feature vector.

[0028] A digital twin baseline model is constructed based on the fused feature vector and real-time operating parameters to establish the mapping relationship between the real-time operating parameters and the voiceprint signal. A joint preprocessing dataset is constructed by integrating the preprocessed original voiceprint data and model parameters.

[0029] The key features include energy in each frequency band and local high-frequency pulse indicators.

[0030] Furthermore, the step of constructing an expected feature vector based on the location of the voiceprint collector using the collector location compensation vector and mapping matrix, and adjusting the difference vector using a correction factor, includes:

[0031] Based on the digital twin benchmark model, the expected feature vector of each acoustic fingerprint collector is constructed by combining the real-time operating parameters with the location parameters of the acoustic fingerprint collector, and the real-time water flow rate is incorporated into the digital twin benchmark model as a mapping parameter to construct a location-dependent mapping formula.

[0032] The actual feature vector corresponding to the preprocessed voiceprint data actually collected by each voiceprint collector is compared with the expected feature vector to calculate the difference vector between the actual feature vector and the expected feature vector, and the difference vector is corrected according to the change of the real-time water flow rate.

[0033] Furthermore, the step of constructing the expected feature vector based on the location of the voiceprint collector based on the collector location compensation vector and the mapping matrix, and adjusting the difference vector through a correction factor, also includes:

[0034] The L2 norm is calculated based on the corrected difference vector as the anomaly score of the acoustic fingerprint collector. Combined with the location and height of the acoustic fingerprint collector array, a weighted average is used to determine whether there is cavitation anomaly in the turbine unit.

[0035] Based on the preset judgment rules and the anomaly score, a cavitation anomaly judgment report is generated and fed back to the monitoring system. At the same time, the parameters of the digital twin benchmark model are updated based on the abnormal voiceprint data.

[0036] Furthermore, the method also includes:

[0037] Based on the anomaly score and the spatial location information of each voiceprint collector, a basic data set is generated. The basic data set is then weighted and averaged to obtain the first cavitation location. The components of the first cavitation location in each direction are then corrected using auxiliary physical parameters to obtain the second cavitation location.

[0038] The actual cavitation location is obtained, and when the actual cavitation location is inconsistent with the second cavitation location, a first deviation vector between the actual cavitation location and the second cavitation location is calculated to correct the second cavitation location and obtain the third cavitation location.

[0039] Calculate the second deviation vector between the third cavitation location and the corresponding acoustic pattern sensor, and adjust the position of each acoustic pattern strip sensor in the horizontal and vertical directions in combination with the anomaly score and the current spatial position information of each acoustic pattern sensor.

[0040] The basic dataset is used to locate cavitation sites and includes anomaly scoring datasets and sensor location datasets. The auxiliary physical parameters include vibration anomaly indicators and temperature anomaly indicators for the area where the acoustic signature collector is located.

[0041] A third aspect of the present invention discloses a terminal, including a processor and a storage medium; characterized in that:

[0042] The storage medium is used to store instructions;

[0043] The processor is configured to operate according to the instructions to perform the steps of the method described in the second aspect.

[0044] The fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the method described in the second aspect.

[0045] The beneficial effects of this invention are:

[0046] Multiple acoustic signature collectors, positioned at various locations and configured at each location, were used to achieve multi-directional coverage and multi-layered acoustic signature data acquisition for the turbine unit. This ensured that acoustic signature collectors at different directions and heights could capture acoustic signature signals from the turbine unit. Furthermore, due to the different positions of each acoustic signature collector, the amplitude of the received acoustic signature signal varied with distance and direction. If a particular location was closer to the main cavitation area, characteristics such as high-frequency pulses and phase shifts would be more pronounced. Because of the different relative positions between the acoustic signature collectors and the cavitation source, different propagation delays and phase differences occurred, resulting in higher accuracy of acoustic signature acquisition under extreme water flow conditions. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the structure of a digital twin-based hydro turbine operating system;

[0048] Figure 2 This is a flowchart of a method for operating a hydro turbine unit based on digital twins. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] like Figure 1 As shown, in one embodiment, a digital twin-based turbine generator operating system includes a turbine generator top cover, on which multiple sets of acoustic signature acquisition arrays are installed. Each acoustic signature acquisition array contains multiple acoustic signature collectors arranged vertically with a fixed spacing.

[0051] The voiceprint collector is used to collect raw voiceprint data from different locations on the top cover of the turbine unit and send the collected raw voiceprint data to the data center.

[0052] After receiving the raw voiceprint data, the data center preprocesses the raw voiceprint data and establishes a digital twin benchmark model based on the mapping relationship between real-time operating parameters and voiceprint signal features using digital twin technology. It also constructs the expected feature vector based on the voiceprint collector location compensation vector and mapping matrix, and adjusts the difference vector using correction factors.

[0053] It should be noted that the expression for the digital twin baseline model is:

[0054] F=M*X+∈

[0055] In the formula, F is the fused preprocessed feature vector, which contains energy information of each frequency band; M is the model parameter matrix, where each element represents the influence coefficient of any operating condition parameter on the preprocessed feature vector; X is the operating condition parameter vector, which is composed of water flow rate, water flow pressure and water flow temperature; ∈ is the model error vector, where each element represents the prediction residual.

[0056] Each voiceprint collector is equipped with a location compensation vector, which includes the location of the corresponding voiceprint collector (encoded value, such as East = 1, South = 2), vertical position (lower layer, middle layer, upper layer), and local environmental factors (such as structural occlusion and openness, such as values ​​ranging from 0 to 1, where 1 indicates no occlusion, 0 indicates severe occlusion, and the middle value indicates partial occlusion).

[0057] The correction factor adjusts the difference vector based on the change in the current real-time water flow rate, and its expression is:

[0058] Δ_i_corrected=Δ_i / (1+α*(Q-Q_ref))

[0059] In the formula, Δ_i_corrected is the correction factor, Δ_i is the difference vector, Q is the current real-time water flow rate, Q_ref is the reference water flow rate, and α is the water flow influence coefficient, ranging from 0.1 to 0.5.

[0060] The actual operating parameters include real-time water flow rate, water flow pressure, and water flow temperature. The difference vector is calculated by preprocessing the actual feature vector and the expected feature vector based on the original voiceprint data actually collected by the voiceprint collector.

[0061] like Figure 2 As shown, in one embodiment, a method for operating a hydroelectric turbine unit based on digital twins includes the following steps:

[0062] Step S110: Obtain raw voiceprint data from multiple perspectives and preprocess the raw voiceprint data.

[0063] The turbine generator set has multiple acoustic signature acquisition arrays installed on its top cover. Each acoustic signature acquisition array contains multiple acoustic signature acquisition devices arranged vertically with a fixed spacing. The preprocessing includes differential preprocessing, time-frequency domain decomposition, adaptive filtering, and dynamic range adjustment.

[0064] In some embodiments, the turbine unit operation method based on digital twins provided by the present invention includes the following steps in step S110:

[0065] Step S111: Based on the preprocessed raw acoustic fingerprint data and the parameters of the digital twin benchmark model, determine the preliminary installation strategy of the acoustic fingerprint collector on the top cover of the turbine unit. The preliminary installation strategy includes multiple acoustic fingerprint collection arrays and the number, arrangement direction and arrangement spacing of the acoustic fingerprint collectors on each acoustic fingerprint collection array.

[0066] Step S112: The acoustic fingerprint collector installed according to the preliminary installation strategy is calibrated, and the sampling frequency, sensitivity and dynamic range are set. The amplitude of the acoustic fingerprint collector is adjusted by the real-time water flow rate to obtain the calibration parameters.

[0067] In some embodiments, the turbine unit operation method based on digital twins provided by the present invention further includes the following steps in step S110:

[0068] Step S113: Based on the installation position and orientation of the acoustic fingerprint collector on the turbine unit top cover, and combined with the correction parameters, different compensation strategies are used to adjust the amplitude normalization of the acoustic fingerprint collector signal based on the real-time water flow rate, and the acoustic fingerprint signal is subjected to noise reduction and dynamic range expansion processing to obtain the original acoustic fingerprint data after differential preprocessing.

[0069] Step S114: The original acoustic signature data after differential preprocessing is synchronized with real-time water flow data and other operating parameters in time and fused together to form a joint data matrix.

[0070] Other operating parameters include water flow pressure and water flow temperature.

[0071] Step S120: Establish a real-time operation model of the turbine unit using digital twin technology, and generate a expected spectrum model based on the turbine unit structure, fluid dynamics, and historical acoustic data.

[0072] It should be noted that fluid dynamics in this example focuses on studying the laws governing water flow and its interaction with the surrounding turbine structure, adhering to the laws of conservation of mass, momentum, and energy. Historical acoustic signature data refers to the acoustic signature data recorded and stored prior to the current raw acoustic signature data collected by the acoustic signature collector. The expected spectrum model is used to characterize the mapping relationship between real-time operating parameters and acoustic signature signals; its expression is:

[0073] F_expected,i=M_i*X+B_i*L_i

[0074] In the formula, F_expected,i is the expected feature vector of the voiceprint collector i in n frequency bands; M_i is the mapping matrix related to the operating parameter X, corresponding to the voiceprint collector i, and each element of it represents the influence coefficient of the operating parameter on the feature; B_i is the position compensation matrix, which is composed of the position compensation vector L_i, corresponding to the voiceprint collector i; X and L_i are both represented by column vectors.

[0075] In some embodiments, the turbine unit operation method based on digital twins provided by the present invention includes the following steps in step S120:

[0076] Step S121: The continuous acoustic signals from each acoustic sensor are segmented according to a fixed time window to form discrete acoustic signal segments. An adjustment factor for each acoustic signal segment is calculated based on the real-time water flow rate. The gain of each acoustic signal segment is adjusted based on the adjustment factor to obtain a new acoustic signal.

[0077] It should be noted that the algorithm expression for calculating the adjustment factor for each acoustic signal segment based on the real-time water flow rate is as follows:

[0078] G = {

[0079] G_low, if Q≤Q_low

[0080] 1. if Q_low < Q < Q_high

[0081] G_high,Q≥Q_high

[0082] }

[0083] Where Q represents the real-time flow rate, Q_low represents the low flow rate threshold, Q_high represents the high flow rate threshold, G_low represents the gain factor used for extremely low flow rates, and G_high represents the attenuation factor used for extremely high flow rates.

[0084] Combining the above expressions, the expression for the new voiceprint signal obtained after gain adjustment is:

[0085] x_adj(n) = G*x(n)

[0086] In the formula, x_adj(n) is the new voiceprint signal after gain adjustment, x(n) is the original voiceprint signal segment of the nth segment before gain adjustment, and G is the adjustment factor.

[0087] Step S122: Perform time-frequency domain decomposition on the new voiceprint signal of each voiceprint collector to extract the spectral features of the new voiceprint signal, and perform differential noise filtering according to the location of the voiceprint collector.

[0088] Step S123: Extract the time-frequency matrix through short-time Fourier transform and obtain the local features of the new voiceprint signal by combining wavelet transform.

[0089] In some embodiments, the turbine unit operation method based on digital twins provided by the present invention further includes the following steps in step S120:

[0090] Step S124: Extract key features based on the preprocessed video data from each acoustic fingerprint collector, perform differential fusion on the key features, and use the compensation information of real-time water flow rate to correct the energy difference of acoustic fingerprint signals at different locations to obtain a fused feature vector.

[0091] Step S125: Construct a digital twin benchmark model based on the fused feature vector and real-time operating parameters to establish the mapping relationship between real-time operating parameters and voiceprint signals, and construct a joint preprocessing dataset by integrating the preprocessed original voiceprint data and model parameters.

[0092] Key features include energy in each frequency band and local high-frequency pulse indicators.

[0093] It should be noted that the expression for the digital twin baseline model is:

[0094] F=M*X+∈

[0095] In the formula, F is the fused preprocessed feature vector, which contains energy information of each frequency band; M is the model parameter matrix, where each element represents the influence coefficient of any operating condition parameter on the preprocessed feature vector; X is the operating condition parameter vector, which is composed of water flow rate, water flow pressure and water flow temperature; ∈ is the model error vector, where each element represents the prediction residual.

[0096] The mapping relationship between real-time operating parameters and acoustic signature signals is expressed as follows:

[0097] F_expected,i=M_i*X+B_i*L_i

[0098] In the formula, F_expected,i is the expected feature vector of the voiceprint collector i in n frequency bands; M_i is the mapping matrix related to the operating parameter X, corresponding to the voiceprint collector i, and each element of it represents the influence coefficient of the operating parameter on the feature; B_i is the position compensation matrix, which is composed of the position compensation vector L_i, corresponding to the voiceprint collector i; X and L_i are both represented by column vectors.

[0099] It should be further noted that the energy of each frequency band refers to the signal energy intensity of each segment of the acoustic signature signal after segmentation, and the local high-frequency pulse index refers to the preset typical cavitation signal frequency band. The joint preprocessing dataset S is represented as follows:

[0100] S={F_preprocessed,M,∈}

[0101] In the formula, F_preprocessed represents the set of preprocessed feature vectors after joint fusion of all orientations, and M and ∈ are the parameters and error vectors of the digital twin benchmark model, respectively.

[0102] Step S130: Construct the expected feature vector based on the location of the voiceprint collector based on the collector location compensation vector and the mapping matrix, and adjust the difference vector by a correction factor.

[0103] In some embodiments, the digital twin-based turbine unit operation method provided by the present invention includes the following steps in step S130:

[0104] Step S131: Based on the digital twin benchmark model, the expected feature vector of each acoustic sensor is constructed by combining the real-time operating parameters with the position parameters of the acoustic sensor, and the real-time water flow rate is incorporated into the digital twin benchmark model as a mapping parameter to construct a position-dependent mapping formula.

[0105] Step S132: Compare the actual feature vector corresponding to the preprocessed voiceprint data actually collected by each voiceprint collector with the expected feature vector to calculate the difference vector between the actual feature vector and the expected feature vector, and correct the difference vector according to the change of real-time water flow rate.

[0106] It should be noted that each element in the actual feature vector represents the preprocessed voiceprint data obtained from the actual voiceprint data collected by each voiceprint collector. The expression for correcting the difference vector based on the real-time water flow rate change is as follows:

[0107] Δ_i_corrected=Δ_i / (1+α*(Q-Q_ref))

[0108] In the formula, Δ_i_corrected is the correction factor, Δ_i is the difference vector of the voiceprint collector i, Q is the current real-time water flow rate, Q_ref is the preset reference water flow rate, and α is the water flow influence coefficient, ranging from 0.1 to 0.5.

[0109] In some embodiments, the turbine unit operation method based on digital twins provided by the present invention further includes the following steps in step S130:

[0110] Step S133: Calculate the L2 norm based on the corrected difference vector as the anomaly score of the acoustic fingerprint collector, and combine the location and height of the acoustic fingerprint collector array to determine whether there is cavitation anomaly in the turbine unit using a weighted average.

[0111] For example, for i data collectors (i=3) in the same location, the weighted average is expressed as:

[0112] S_d=(w_1*S_{1,\text{norm}}+w_2*S_{2,\text{norm}}+w_3*S_{3,\text{norm}})

[0113] In the formula, S_d represents the result after weighted averaging, w_i represents the weight of the voiceprint collector i in this location, which is determined according to the installation position of the voiceprint collector (upper, middle, or lower layer). For example, the upper layer signal may be more sensitive, so it can be given a higher weight; S_{i, \text{norm}} represents the abnormal score of the voiceprint collector i after standardization.

[0114] Step S134: Generate a cavitation anomaly judgment report based on the anomaly score according to the preset judgment rules, and feed the cavitation anomaly judgment report back to the monitoring system. At the same time, update the parameters of the digital twin benchmark model based on the abnormal voiceprint data.

[0115] In some embodiments, the digital twin-based turbine unit operation method provided by the present invention further includes the following steps:

[0116] Step S210: Generate a basic data set based on the anomaly score and the spatial location information of each voiceprint collector, perform a weighted average on the basic data set to obtain the first cavitation location, and combine auxiliary physical parameters to correct the components of the first cavitation location in each direction to obtain the second cavitation location.

[0117] Step S220: Obtain the actual cavitation location, and when the actual cavitation location is inconsistent with the second cavitation location, calculate the first deviation vector between the actual cavitation location and the second cavitation location to correct the second cavitation location and obtain the third cavitation location.

[0118] Step S230: Calculate the second deviation vector between the third cavitation location and the corresponding acoustic pattern sensor, and adjust the position of each acoustic pattern sensor in the horizontal and vertical directions in combination with the anomaly score and the current spatial position information of each acoustic pattern sensor.

[0119] It should be noted that the first cavitation position is the preliminary cavitation position obtained by weighted averaging of the basic data set, the second cavitation position is the cavitation position obtained after correcting each component of the preliminary cavitation position by combining auxiliary physical parameters, and the third cavitation position is the cavitation position obtained after correcting the second cavitation position based on the deviation vector between the actual cavitation position and the second cavitation position.

[0120] The basic dataset is used to locate cavitation sites and includes anomaly scoring datasets and sensor location datasets. Auxiliary physical parameters include vibration anomaly indicators and temperature anomaly indicators in the area where the acoustic signature collector is located.

[0121] The aforementioned digital twin-based turbine operation method employs multiple acoustic signature collectors at various locations to achieve multi-directional coverage and multi-layered acoustic signature data acquisition. This ensures that acoustic signature collectors at different directions and heights can capture acoustic signature signals from the turbine. Furthermore, due to the different positions of each acoustic signature collector, the amplitude of the received acoustic signature signal varies with distance and direction. If a particular location is closer to the main cavitation area, features such as high-frequency pulses and phase shifts will be more pronounced. Because the relative positions of the acoustic signature collectors and the cavitation source differ, varying propagation delays and phase differences occur, resulting in higher accuracy of acoustic signature acquisition under extreme water flow conditions.

[0122] In a specific embodiment, the turbine unit operation method based on digital twin provided by the present invention includes steps 1 to 5:

[0123] Step 1: Multi-directional voiceprint data acquisition.

[0124] At least four sets of acoustic signature collection arrays are arranged in the four main directions (east, south, west, and north) of the turbine generator top cover. Each array contains three acoustic signature collectors (SC1, SC2, and SC3), arranged in sequence according to the vertical direction, with a spacing of d1 (e.g., 0.5 to 1.0 meters).

[0125] Targeted preprocessing reduces the risk of misjudgment: Based on the characteristics of signal amplitude variation under different water flow conditions, the raw acoustic data undergoes differentiated preprocessing, enabling the differentiation between normal transient signals and abnormal cavitation signals during drastic changes in operating conditions (such as startup, shutdown, and sudden load changes), thus reducing the misjudgment rate. Furthermore, adaptive filtering and dynamic range adjustment are used to mitigate the impact of multi-source noise and signal interference on spectral characteristics, ensuring that subsequent analysis relies on high-quality data input.

[0126] Specifically, this includes steps 1.1 to 1.4:

[0127] Step 1.1, array layout and data acquisition unit installation.

[0128] Specifically, using the preprocessed raw acoustic signature data and digital twin baseline model parameters as the expected signal reference, the installation strategy for the acoustic signature collectors on the turbine generator top cover was determined. This requires the placement of one acoustic signature collector array in each of the four main directions (east, south, west, and north) of the turbine generator top cover. Each array contains three acoustic signature collectors, installed vertically from top to bottom, with equal spacing between each array. Furthermore, considering that the acoustic signature signals received by collectors at different locations may vary under the influence of water flow, it is necessary to ensure that the installation locations of the acoustic signature collectors cover areas prone to cavitation and other anomalies.

[0129] For example, if the vertical length of the collection area on the top cover of the turbine unit is 1.2 meters, then the installation spacing between each row of acoustic fingerprint collectors is 0.6 meters. In the eastern array, if acoustic fingerprint collector 1 in the first array is installed at the lowest position, i.e., 0 meters, then the acoustic fingerprint collector 2 above it is installed at 0.6 meters, and the uppermost acoustic fingerprint collector 3 is installed at 1.2 meters.

[0130] Step 1.2: Calibration of the acoustic signature collector and initial acquisition of real-time water flow data.

[0131] Specifically, each installed acoustic fingerprint sensor is calibrated, with the sampling frequency, sensitivity, and dynamic range set. First, the amplitude of the sensor is initially adjusted using real-time water flow rate to ensure that the acoustic fingerprint signal measured under extremely low and high water flow conditions is not distorted, providing accurate correction parameters for subsequent preprocessing.

[0132] The original voiceprint signal can be represented as:

[0133] Signal(t) = A * cos(2πft + φ)

[0134] In the formula, Signal(t) represents the original acoustic signature signal at time t, A is the amplitude of the acoustic signature signal, which is affected by the water flow state, f is the signal frequency, t is the time, and φ is the initial phase.

[0135] It should be noted that the real-time water flow rate is provided by the flow meter, and its impact on the signal amplitude is related to the acoustic signature acquisition sensitivity compensation coefficient.

[0136] Step 1.3: Differentiated preprocessing based on the location and orientation of the voiceprint collector.

[0137] Due to differences in the location (height, orientation) of the acoustic fingerprint collector, the raw acoustic fingerprint signals it collects will vary when affected by water flow conditions:

[0138] The acoustic signature collector located on the upper layer may receive a stronger water flow impact signal, with a higher signal amplitude and more prominent high-frequency components in the spectrum.

[0139] The voiceprint collector located on the lower layer may be affected by local structure and reflection, resulting in significant signal attenuation.

[0140] Therefore, during preprocessing, different compensation strategies need to be adopted according to the installation location and orientation of the acoustic fingerprint collectors. The amplitude normalization of the signals from each acoustic fingerprint collector is adjusted based on the real-time water flow rate, and the acoustic fingerprint signals are processed for noise reduction and dynamic range expansion.

[0141] For the i-th voiceprint collector, the preprocessed amplitude A_i is defined as:

[0142] A_i=k_i*Q+C_i

[0143] In the formula, k_i is the sensitivity compensation coefficient of the i-th acoustic fingerprint collector, which is obtained by calibration, Q is the real-time water flow rate, and C_i is the position compensation constant, which is used to reflect the influence of structural obstruction and multipath effect on the acoustic fingerprint collector, and its value is usually in the range of 0 to 0.1V.

[0144] Step 1.4, Data synchronization and construction of the joint data matrix.

[0145] Specifically, the raw acoustic data of each acoustic data collector after differentiated preprocessing is synchronized with the real-time water flow rate and other operating parameters (water flow pressure, water flow temperature) and fused together to form a joint data matrix, providing high-precision input for subsequent digital twin modeling and anomaly detection.

[0146] The joint data matrix is ​​composed of the preprocessed signal, water flow rate, water flow pressure, water flow temperature, and discrete sampling time collected by the acoustic fingerprint collector at any discrete time point.

[0147] Step 2, Signal preprocessing and digital twin environment modeling.

[0148] The collected raw acoustic signature data was decomposed in the time and frequency domain using Short-Time Fourier Transform (STFT) and Wavelet Transform. Adaptive noise reduction and amplitude normalization were then performed in conjunction with water flow state parameters (extremely low and extremely high flow rates). Simultaneously, a real-time operating model of the turbine unit was established using digital twin technology. Based on the turbine unit's structure, fluid dynamics, and historical data, a predicted spectrum model was generated to provide a benchmark for subsequent anomaly identification.

[0149] Different window lengths and filtering strategies were employed to process the upper, middle, and lower data acquisition units separately, compensating for signal characteristic differences caused by installation location and orientation, thus improving data consistency and accuracy. Furthermore, a mapping relationship was established between real-time operating parameters and signal characteristics. By introducing a model parameter matrix and a position compensation matrix, the digital twin model could adapt to signal variations under different acquisition unit locations and operating conditions.

[0150] Specifically, this includes steps 2.1 to 2.4:

[0151] Step 2.1: Dynamic signal segmentation and gain adjustment based on the influence of real-time water flow rate.

[0152] Specifically, the amplitude of the original acoustic signature signal collected by the acoustic signature collectors at different locations will vary significantly under different water flow rates. In order to avoid insufficient amplitude or unsaturated amplitude of the acoustic signature signal due to extreme water flow conditions, each acoustic signature collector is dynamically segmented and its gain is adjusted based on the real-time water flow rate.

[0153] In the process of acoustic signature signal segmentation, the continuous acoustic signature signal collected by each acoustic signature collector is segmented according to a fixed time window to form discrete signal segments. Then, an adjustment factor for each acoustic signature signal segment is calculated based on the real-time water flow rate. This adjustment factor is jointly determined by the real-time water flow rate, low flow rate threshold, high flow rate threshold, a gain factor for extremely low flow rates, and an attenuation factor for extremely high flow rates. After adjusting the gain of each acoustic signature signal segment, a new acoustic signature signal is obtained.

[0154] For example, when a certain voiceprint collector is in a water flow rate of 0.2m... 3 At an extremely low water flow rate of / s, with a gain factor of 4, the average amplitude of the original voiceprint signal segment is 0.05V, and the adjusted new voiceprint signal is approximately 0.2V, ensuring that the low-amplitude signal is amplified.

[0155] Step 2.2, Time-frequency domain decomposition and differential noise filtering.

[0156] Specifically, the voiceprint signals of each voiceprint collector are decomposed in the time-frequency domain to extract their spectral features, and differentiated noise filtering strategies are adopted according to the location of the voiceprint collector (upper layer, middle layer, lower layer). Among them, the upper layer voiceprint collectors have richer high-frequency information due to the impact of direct current water, while the lower layer voiceprint collectors may be affected by structural damping, resulting in greater noise interference. Therefore, short-time Fourier transform (STFT) is used to extract the time-frequency matrix, and then wavelet transform is combined to obtain local detailed features.

[0157] During the short-time Fourier transform (SFT) process, the window length is adjusted according to the installation position of the acoustic signature collector. Specifically, the upper-layer acoustic signature collector uses a short window length to improve temporal resolution, while the lower-layer acoustic signature collector uses a long window length to enhance spectral resolution. A SFT is applied to each new acoustic signature signal segment to obtain the time-frequency matrix. Then, discrete wavelet transform is used to extract local features, and low-energy noise is filtered out based on a set noise threshold (which can be determined based on pre-measured background noise energy).

[0158] Step 2.3: Extraction and fusion of location-specific features based on voiceprint collector.

[0159] Specifically, the location and orientation differences caused by different acoustic fingerprint collectors result in variations in their time-frequency characteristics. Key features (such as energy in each frequency band and local high-frequency pulse indicators) are extracted from the preprocessed time-frequency data of each acoustic fingerprint collector and then fused differentially. Using real-time water flow rate compensation information, the energy differences between acoustic fingerprint signals at different locations are further corrected, laying the foundation for constructing a unified digital twin benchmark model.

[0160] In this regard, considering location compensation, a normalized vector needs to be constructed, which is obtained by combining historical data statistics with location-specific energy bias compensation.

[0161] In the process of feature extraction and fusion, for each acoustic fingerprint collector, different frequency band energies are extracted from the time-frequency matrix and local features to form a preliminary feature vector. Combined with the real-time water flow rate and the calibration parameters of each acoustic fingerprint collector, a normalized feature vector is calculated. For multiple acoustic fingerprint collectors in the same location, the weighted average and principal component analysis methods are used to fuse their respective normalized feature vectors to form a unified feature vector for that location.

[0162] Step 2.4: Construct a digital twin benchmark model and output a joint preprocessed dataset.

[0163] Specifically, based on the feature vectors extracted and fused in step 2.3, combined with real-time operating parameters (water flow rate, water flow pressure, and water flow temperature) and a joint preprocessed dataset consisting of a joint data matrix, a digital twin baseline model is constructed to propose a mapping relationship between operating parameters and acoustic signature signal features. Subsequently, the preprocessed raw acoustic signature data and model parameters are integrated to construct a joint preprocessed dataset, which serves as a high-precision input for subsequent anomaly detection and operating condition adjustment.

[0164] The operating condition parameter vector is composed of real-time water flow rate, water flow pressure, and water flow temperature. The constructed digital twin benchmark model is composed of the fused preprocessed feature vector (energy information for each frequency band), the model parameter matrix (influence coefficient of each time-frequency feature element on local feature elements), and the model error vector (prediction residual represented by each element). The joint preprocessed dataset is composed of the set of fused preprocessed feature vectors from all directions, as well as the parameters and error vectors of the digital twin benchmark model.

[0165] In the process of constructing the digital twin benchmark model and the joint preprocessing dataset, historical data and real-time data are used. The model parameter matrix is ​​solved by the least squares method and updated in real time. Sub-models are established for different orientations and integrated. Finally, the joint preprocessing dataset is output to provide a unified high-precision data input for subsequent anomaly detection and operating condition optimization.

[0166] Step 3: Digital twin comparison and cavitation anomaly determination

[0167] The preprocessed raw acoustic signature data is compared in real time with a digital twin benchmark model, and typical cavitation signal frequency bands (high-frequency impulse noise) are identified using dynamically adjusted thresholds. Based on the expected characteristics under extreme water flow conditions (extremely low and extremely high flow rates), the discrimination algorithm is adjusted to distinguish between startup, load abrupt changes, and cavitation impact signals, avoiding misjudgments. By introducing a collector location compensation vector and mapping matrix, expected feature vectors based on the location of each acoustic signature collector are constructed, enabling the model to provide personalized expected values ​​for acoustic signature collectors in different installation locations. A correction factor is used to adjust the difference vector, ensuring the comparability of acoustic signature signal deviations under extremely low and extremely high flow conditions, thereby improving the accuracy of anomaly scoring.

[0168] Specifically, this includes steps 3.1 to 3.4:

[0169] Step 3.1: Construct the position-dependent expected feature vector.

[0170] Specifically, due to differences in installation height and orientation, the acoustic signature characteristics (such as frequency band energy distribution) expected to be collected by acoustic signature collectors at different locations will vary. Based on the digital twin benchmark model, the expected feature vector of each acoustic signature collector is constructed by combining real-time operating parameters with acoustic signature collector location parameters. At the same time, the water flow rate has a direct impact on the output of the acoustic signature collector, so the water flow rate needs to be incorporated into the model as a mapping parameter to construct a location-dependent mapping formula.

[0171] In this embodiment, a compensation vector is set for each voiceprint collector, which is composed of the location, vertical position, and local environmental factors (such as structural occlusion and openness) of the voiceprint collector. The expression for the location-dependent digital twin expected feature vector is:

[0172] F_expected,i=M_i*X+B_i*L_i

[0173] In the formula, F_expected,i represents the expected feature vector of the acoustic fingerprint collector i in n frequency bands; M_i represents the mapping matrix related to the operating parameters X, corresponding to the acoustic fingerprint collector i, and each element of M_i represents the influence coefficient of the operating parameters on the expected signal features; B_i is the position compensation matrix, and each element of B_i represents the contribution of the position compensation factor to the expected signal features; L_i represents the pre-set position parameters of the acoustic fingerprint collector, and both the operating parameters X and L_i are represented by column vectors.

[0174] In the process of constructing the position-dependent expected feature vector, for each voiceprint collector, based on the historical data and preprocessed features obtained above, the least squares method is used to fit and obtain the mapping matrix and position compensation matrix. The expected feature vector is calculated based on the real-time collected operating parameters and the pre-set voiceprint collector position parameters.

[0175] Step 3.2, Calculation of difference vector and correction of water flow state.

[0176] Specifically, the preprocessed feature vector actually collected by each acoustic signature collector is compared with the corresponding expected feature vector, and the difference vector between the two is calculated. Simultaneously, the calculated difference vector is corrected based on the real-time changes in water flow rate to ensure the comparability of acoustic signature signal deviations under low and high flow conditions. Furthermore, to correct for the influence of water flow, a correction factor can be defined to adjust the difference vector according to changes in water flow rate.

[0177] The difference vector is the difference between the preprocessed feature vector actually collected by the acoustic signature collector and the corresponding expected feature vector. The correction factor is composed of the current real-time water flow rate, the reference water flow rate, and the water flow influence coefficient, which is determined from historical data.

[0178] In this embodiment, the original difference vector of each acoustic fingerprint collector needs to be calculated, and the correction factor is calculated based on the real-time water flow rate and the reference water flow rate. The original difference vector is then adjusted to the corrected difference vector, and the corrected difference vector is used as the basic indicator for the anomaly determination of each acoustic fingerprint collector.

[0179] Step 3.3, Anomaly score calculation and voiceprint collector anomaly determination.

[0180] Specifically, based on the corrected difference vectors of each acoustic signature collector obtained above, anomaly scores are calculated using the L2 norm, and a comprehensive judgment is made by combining the location and altitude information of the acoustic signature collector array. This involves not only calculating the anomaly score of each individual acoustic signature collector, but also using a weighted average method based on its different locations to determine whether there is overall cavitation anomaly.

[0181] The anomaly score is determined by the difference vector after correction of the voiceprint collector in the feature dimension and the number of feature dimensions.

[0182] In the process of anomaly score calculation and comprehensive anomaly determination of the acoustic fingerprint collectors, anomaly scores are calculated for each acoustic fingerprint collector. These scores are then standardized based on a preset judgment threshold determined from historical data, resulting in a standardized anomaly score. Next, a weighted average of the standardized anomaly scores for each group of acoustic fingerprint collectors is calculated based on the collectors' directional information, yielding a comprehensive anomaly score for that directional location. Finally, according to preset judgment rules, if the comprehensive anomaly score exceeds a set threshold of 1, an anomaly is determined to exist in that directional location; if more than 50% of the locations show anomalies, the entire location is judged as experiencing cavitation anomalies.

[0183] Step 3.4, anomaly detection feedback and adaptive update of digital twin benchmark model.

[0184] Specifically, after the anomaly score calculation is completed, a cavitation anomaly judgment report is generated according to the preset judgment rules and fed back to the monitoring system. At the same time, the digital twin benchmark model is updated based on the anomaly data to achieve adaptive learning, so as to ensure that the model can continuously correct the deviation caused by changes in the location of the acoustic fingerprint collector and the water flow state.

[0185] In this embodiment, for each acoustic fingerprint sensor, the update amount is calculated based on the deviation between its actual measurement result and the model prediction result to update the model parameters corresponding to all acoustic fingerprint sensors. The generated cavitation anomaly judgment report includes the anomaly score of each acoustic fingerprint sensor and its location, the judgment conclusion, real-time operating parameters, and suggested measures (such as adjusting the load and water flow velocity). The updated model parameters are saved for use in the next training cycle.

[0186] Step 4: Further determine the location of cavitation.

[0187] Specifically, this includes steps 4.1 to 4.4:

[0188] Step 4.1: Collect voiceprint anomaly data and sensor location information.

[0189] Based on the anomaly scores of the noise-reduced voiceprint data, combined with the spatial location information of each voiceprint collector, a basic data set for localization is generated, including anomaly score set and sensor location set.

[0190] Specifically, firstly, an anomaly score set is constructed based on the comprehensive anomaly score calculated in step 3.3. This anomaly score set is composed of the anomaly score values ​​of each voiceprint collector. Secondly, according to the spatial position of each voiceprint collector in the digital twin baseline model, a corresponding sensor position set is constructed. This position set is composed of the position vector corresponding to each voiceprint collector, and the position vector of each voiceprint collector is composed of the X-axis, Y-axis, and Z-axis coordinates of that voiceprint collector.

[0191] In this example, it is necessary to ensure that all abnormal scores from the voiceprint collectors have undergone depth-adaptive noise reduction, and that all location information contained in the location set has been measured and corrected on-site and converted to a unified coordinate system.

[0192] Step 4.2, preliminary calculation of the location of the abnormal center of gravity.

[0193] Specifically, based on the anomaly score set and sensor location set, the centroid position of the anomalous acoustic signature is calculated using a weighted average as a preliminary estimate of the cavitation location. First, the preliminary anomaly centroid position is defined using a weighted average formula, and then the preliminary anomaly centroid position is expanded to obtain the corresponding components, i.e., the mean values ​​of the anomaly centroid position on the X, Y, and Z axes.

[0194] Step 4.3: Introduce auxiliary physical parameters for correction.

[0195] Specifically, the preliminary abnormal centroid position obtained in step 4.2 is corrected by combining auxiliary physical parameters such as vibration and temperature to obtain a more refined cavitation position. Then, the contribution weight of each acoustic fingerprint collector is corrected by using a comprehensive abnormality index to enhance the accuracy of cavitation positioning.

[0196] First, for each voiceprint sensor, in addition to the voiceprint anomaly score, it is also necessary to obtain a vibration anomaly index reflecting vibration anomalies in the area where the voiceprint sensor is located, and a temperature anomaly index reflecting temperature anomalies in the area where the voiceprint sensor is located. The aforementioned voiceprint anomaly score, vibration anomaly index, and temperature anomaly index are all monitored by vibration sensors and infrared temperature cameras installed at corresponding locations, and are converted into standard dimensionless indices after preprocessing.

[0197] Secondly, in defining the comprehensive anomaly index, weighting coefficients are introduced for the voiceprint anomaly score, vibration anomaly index, and temperature anomaly index, with the sum of these weighting coefficients being 1. Therefore, the comprehensive anomaly index is the sum of the products of the voiceprint anomaly score, vibration anomaly index, and temperature anomaly index and their respective weighting coefficients, used to reflect the overall detection anomaly situation.

[0198] Finally, the abnormal locations of each acoustic fingerprint collector are weighted using the aforementioned comprehensive anomaly index to obtain the corrected cavitation locations.

[0199] Step 4.4, on-site verification and feedback correction.

[0200] Specifically, the corrected cavitation location is verified using actual on-site monitoring data, and further corrected based on on-site feedback to obtain the final determined cavitation location.

[0201] First, real-time data collected through on-site inspections, fixed monitoring cameras, and high-precision sensors is used to verify whether the cavitation anomaly areas displayed on the digital twin platform are consistent with the actual situation. Next, a correction deviation vector is calculated by defining a reference position obtained from on-site detection; this vector represents the difference between the actual cavitation location and the cavitation location on the digital twin platform that needs correction. The correction deviation vector is then added to the cavitation location obtained in step 4.3 for further correction, yielding the final cavitation location.

[0202] This invention corrects the calculation results by using on-site feedback data, similar to differential GPS technology. By fine-tuning the measurement results using known reference points, it can ensure that the final determined cavitation location is consistent with the actual situation.

[0203] Step 5: Based on the finally determined cavitation location, readjust the positions of the 12 sensors.

[0204] The location of cavitation is closely related to the equipment design and installation status, and therefore is usually relatively fixed in the short term. However, with long-term operation, wear, maintenance, and changes in environmental conditions, the cavitation area may experience slight displacement and expansion. Therefore, it is necessary to readjust the position of the acoustic signature sensor to more closely monitor the cavitation status and obtain more accurate results.

[0205] Specifically, this includes steps 5.1 to 5.4:

[0206] Step 5.1: Data preprocessing and voiceprint anomaly distribution statistics.

[0207] Based on the acoustic anomaly score obtained in step 3 and the cavitation location finally determined in step 4, the relative distribution of each acoustic fingerprint collector and its sensor location is statistically analyzed to provide a basis for subsequent calculation of the acoustic fingerprint collector spacing.

[0208] Specifically, firstly, based on the aforementioned set of anomaly scores, sensor locations, and the finally determined cavitation locations, the deviation vector between each acoustic signature collector and the finally determined cavitation location is calculated. Then, using the acoustic signature anomaly scores as weights, the weighted standard deviations at each location and direction (X-axis, Y-axis, Z-axis) are calculated as an indicator of the distribution width of the anomaly data. In calculating the weighted standard deviations of the acoustic signature anomaly scores at each location and direction, the weighted average deviation at each location and direction is calculated first, and then the weighted standard deviation is calculated based on this weighted average deviation. These weighted standard deviations at each location and direction (X-axis, Y-axis, Z-axis) reflect the degree of distribution and diffusion of acoustic signature anomalies in each direction; the more dispersed the acoustic signature anomaly data distribution, the higher the demand for acoustic signature monitoring in that direction.

[0209] Step 5.2: Calculate the vertical spacing within the voiceprint collector group.

[0210] Based on the standard deviation of the abnormal distribution of acoustic signatures in the vertical direction obtained in step 5.1 and the on-site monitoring requirements, the optimal vertical spacing within the acoustic signature collector group is determined.

[0211] Specifically, firstly, the vertical spacing of adjacent acoustic fingerprint collectors in the same group is obtained. Then, a vertical spacing adjustment formula is constructed based on the vertical diffusion standard deviation of acoustic fingerprint anomalies and the required on-site monitoring accuracy. The adjusted vertical spacing is the product of the vertical adjustment coefficient (obtained from historical data and on-site preset adjustments) and the standard deviation of the vertical anomaly distribution. If the acoustic fingerprint anomalies are relatively dispersed vertically (i.e., the standard deviation is large), a larger vertical spacing is needed to ensure that each layer of acoustic fingerprint collectors covers different heights; conversely, the vertical spacing of the acoustic fingerprint collectors is reduced to improve local acoustic fingerprint monitoring accuracy.

[0212] Step 5.3: Determine the lateral spacing between the voiceprint collector groups.

[0213] Based on the standard deviation of the abnormal distribution of the acoustic fingerprint collectors in the horizontal direction obtained in step 5.1, and combined with the characteristics of the cavitation area, the optimal lateral spacing between each acoustic fingerprint collector group (at the same height) is determined.

[0214] Specifically, first, the initial horizontal spacing between different acoustic signature acquisition groups is determined. Then, considering the abnormal distribution of acoustic signatures in the horizontal direction, a comprehensive horizontal diffusion index is defined by combining the weighted standard deviations in the X and Y axes. Therefore, the adjustment formula for the horizontal spacing is the product of the comprehensive horizontal diffusion index and the horizontal adjustment coefficient (preset according to the site conditions).

[0215] Step 5.4, Comprehensive scheme verification and final parameter output.

[0216] The optimal vertical spacing and optimal horizontal spacing obtained from steps 5.2 and 5.3 are integrated to obtain the overall calculation result. The overall calculation result is then verified to ensure that it meets the requirements for uniform coverage on site. Finally, the optimized horizontal and vertical adjustment spacings are output.

[0217] Specifically, the lateral and vertical adjustment spacings are compared with the final determined cavitation locations and the standard deviation of acoustic signature anomalies to check whether each group of acoustic signature collectors can cover key areas under the theoretical arrangement. For example, it is checked whether the lateral distance between the centers of each acoustic signature collector group is theoretically satisfied, that is, whether the distance between the acoustic signature collectors at the same height and their respective centers is half of the spacing between them.

[0218] In this example, the newly arranged spacing adjustment scheme can be simulated using the horizontal and vertical adjustment spacings calculated above to ensure that each group of acoustic fingerprint collectors is evenly distributed in the vertical and horizontal directions, and that the positions of the acoustic fingerprint collectors meet the design requirements. During the simulation, the new positions can be calculated (i.e., step 5.3), and the distances between each calculated new position and the cavitation center can be checked to see if they are within the preset range. After verification, the final horizontal and vertical adjustment spacings can be set as the final optimization parameters and recorded as the basis for subsequent acoustic fingerprint collector reinstallation and system updates.

[0219] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. "A plurality of" means two or more, unless otherwise explicitly specified.

[0220] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0221] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0222] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0223] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0224] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0225] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0226] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0227] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0228] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for operating a hydro-turbine unit based on digital twins, characterized in that, The method includes: Acquire raw voiceprint data from multiple perspectives and preprocess the raw voiceprint data; A real-time operation model of the turbine unit was established using digital twin technology, and a expected spectrum model was generated based on the turbine unit's structure, fluid dynamics, and historical acoustic data. Based on the acquisition device location compensation vector and mapping matrix, an expected feature vector based on the voiceprint acquisition device location is constructed, and the difference vector is adjusted by a correction factor. The turbine generator set has multiple acoustic fingerprint acquisition arrays installed on its top cover. Each acoustic fingerprint acquisition array contains multiple acoustic fingerprint collectors arranged vertically with a fixed spacing. The preprocessing includes differential preprocessing, time-frequency domain decomposition, adaptive filtering, and dynamic range adjustment. The method is used to perform operation in a digital twin-based turbine generator set operating system, the system including a turbine generator set top cover, on which multiple sets of acoustic fingerprint acquisition arrays are provided, each set of acoustic fingerprint acquisition arrays containing multiple acoustic fingerprint collectors arranged vertically and with a fixed spacing. The acoustic fingerprint collector is used to collect raw acoustic fingerprint data from different locations on the top cover of the turbine unit and send the collected raw acoustic fingerprint data to the data center. After receiving the original voiceprint data, the data center preprocesses the original voiceprint data and establishes a digital twin benchmark model based on the mapping relationship between real-time operating parameters and voiceprint signal features using digital twin technology. It also constructs an expected feature vector based on the voiceprint collector position compensation vector and mapping matrix, and adjusts the difference vector using a correction factor. The real-time operating parameters include real-time water flow rate, water flow pressure, and water flow temperature. The difference vector is calculated by combining the actual feature vector after preprocessing the original voiceprint data actually collected by the voiceprint collector with the expected feature vector. The acquisition of raw voiceprint data from multiple perspectives and the preprocessing of the raw voiceprint data include: Based on the preprocessed raw acoustic fingerprint data and the parameters of the digital twin benchmark model, a preliminary installation strategy for the acoustic fingerprint collectors on the top cover of the turbine unit is determined. The preliminary installation strategy includes multiple acoustic fingerprint collection arrays and the number, arrangement direction and arrangement spacing of the acoustic fingerprint collectors on each acoustic fingerprint collection array. The acoustic fingerprint scanner installed according to the preliminary installation strategy is calibrated, and the sampling frequency, sensitivity and dynamic range are set. The amplitude of the acoustic fingerprint scanner is adjusted by the real-time water flow rate to obtain the calibration parameters. Based on the installation position and orientation of the acoustic fingerprint collector on the top cover of the turbine unit, and in conjunction with the aforementioned correction parameters, different compensation strategies are adopted to adjust the amplitude normalization of the acoustic fingerprint collector signal based on the real-time water flow rate. The acoustic fingerprint signal is then subjected to noise reduction and dynamic range expansion processing to obtain the original acoustic fingerprint data after differentiated preprocessing. The original acoustic signature data after differential preprocessing is synchronized with real-time water flow data and other operating parameters in time and fused together to form a joint data matrix. Other operating parameters include water flow pressure and water flow temperature; The process of establishing a real-time operation model of the turbine unit using digital twin technology, and generating a expected spectrum model based on the turbine unit structure, fluid dynamics, and historical acoustic signature data, includes: The continuous acoustic signals from each acoustic sensor are segmented according to a fixed time window to form discrete acoustic signal segments. An adjustment factor for each acoustic signal segment is calculated based on the real-time water flow rate. The gain of each acoustic signal segment is then adjusted based on the adjustment factor to obtain a new acoustic signal. The new voiceprint signal of each voiceprint collector is decomposed in the time-frequency domain to extract the spectral features of the new voiceprint signal, and differential noise filtering is performed according to the location of the voiceprint collector. The time-frequency matrix is ​​extracted by short-time Fourier transform, and the local features of the new voiceprint signal are obtained by combining wavelet transform. The process of establishing a real-time operation model of the turbine unit using digital twin technology, and generating a expected spectrum model based on the turbine unit structure, fluid dynamics, and historical acoustic signature data, also includes: Key features are extracted from the preprocessed video data of each acoustic signature collector, and the key features are differentially fused. The energy difference of the acoustic signature signals at different locations is corrected by using the compensation information of real-time water flow rate, so as to obtain the fused feature vector. A digital twin baseline model is constructed based on the fused feature vector and real-time operating parameters to establish the mapping relationship between the real-time operating parameters and the voiceprint signal. A joint preprocessing dataset is constructed by integrating the preprocessed original voiceprint data and model parameters. The key features include energy in each frequency band and local high-frequency pulse indicators; The process of constructing an expected feature vector based on the location of the voiceprint collector using the collector location compensation vector and mapping matrix, and adjusting the difference vector using a correction factor, includes: Based on the digital twin benchmark model, the expected feature vector of each acoustic fingerprint collector is constructed by combining the real-time operating parameters with the location parameters of the acoustic fingerprint collector, and the real-time water flow rate is incorporated into the digital twin benchmark model as a mapping parameter to construct a location-dependent mapping formula. The actual feature vector corresponding to the preprocessed voiceprint data actually collected by each voiceprint collector is compared with the expected feature vector to calculate the difference vector between the actual feature vector and the expected feature vector, and the difference vector is corrected according to the change of the real-time water flow rate.

2. The method for operating a hydro-turbine unit based on digital twins according to claim 1, characterized in that, The process of constructing an expected feature vector based on the location of the voiceprint collector using the collector location compensation vector and mapping matrix, and adjusting the difference vector using a correction factor, further includes: The L2 norm is calculated based on the corrected difference vector as the anomaly score of the acoustic fingerprint collector. Combined with the location and height of the acoustic fingerprint collector array, a weighted average is used to determine whether there is cavitation anomaly in the turbine unit. Based on the preset judgment rules and the anomaly score, a cavitation anomaly judgment report is generated and fed back to the monitoring system. At the same time, the parameters of the digital twin benchmark model are updated based on the abnormal voiceprint data.

3. The method for operating a hydro-turbine unit based on digital twins according to claim 2, characterized in that, The method further includes: Based on the anomaly score and the spatial location information of each voiceprint collector, a basic data set is generated. The basic data set is then weighted and averaged to obtain the first cavitation location. The components of the first cavitation location in each direction are then corrected using auxiliary physical parameters to obtain the second cavitation location. The actual cavitation location is obtained, and when the actual cavitation location is inconsistent with the second cavitation location, a first deviation vector between the actual cavitation location and the second cavitation location is calculated to correct the second cavitation location and obtain the third cavitation location. Calculate the second deviation vector between the third cavitation location and the corresponding acoustic pattern sensor, and adjust the position of each acoustic pattern strip sensor in the horizontal and vertical directions in combination with the anomaly score and the current spatial position information of each acoustic pattern sensor. The basic dataset is used to locate cavitation sites and includes anomaly scoring datasets and sensor location datasets. The auxiliary physical parameters include vibration anomaly indicators and temperature anomaly indicators for the area where the acoustic signature collector is located.

4. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-3.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-3.

Images