A flow field self-adaptive calculation method and system based on an acoustic Doppler current profiler

Abstract

The application discloses a flow field self-adaptive calculation method and system based on an acoustic Doppler current profiler. The method comprises the following steps: collecting original spectrum data of a Doppler current meter and synchronous attitude, position and water level information; pre-processing and quality marking the original data; extracting multi-dimensional features, identifying the current dominant flow field type through a pre-trained flow field classification model; adaptively selecting a flow velocity profile reconstruction algorithm according to the flow field type, and dynamically adjusting key algorithm parameters in the reconstruction process according to flow field stability or turbulent intensity parameters; generating a calculation grid for flow integration calculation based on the reconstruction strategy, and comprehensively evaluating data, models and integration errors to quantify the uncertainty of the output flow result; and finally, performing flow field anomaly diagnosis, short-term flow trend prediction and equipment health state early warning. The application solves the problem of insufficient accuracy of traditional fixed algorithms under complex flow fields, and significantly improves the reliability and intelligent level of flow monitoring.

Description

Technical Field

[0001] This invention relates to the field of signal processing, and in particular to an adaptive flow field calculation method and system based on an acoustic Doppler flowmeter. Background Technology

[0002] Accurate flow monitoring is fundamental to scientific water resource management, flood control and disaster reduction decision-making, and optimized scheduling of water conservancy projects. Acoustic Doppler current profilers, as the current mainstream flow monitoring equipment, offer advantages such as non-contact measurement, a wide measurement range, and the ability to acquire velocity profiles. However, in practical applications in complex water bodies, existing technologies still face significant challenges:

[0003] First, traditional ADCP (Acoustic Doppler Current Profiler) data processing often relies on fixed interpolation algorithms, such as triangulation or fixed kernel function interpolation. These methods struggle to strike a balance between smoothness and detail preservation when dealing with complex flow fields involving strong turbulence, vortices, or backflow. Excessive smoothing can lead to the loss of key flow field features, while overemphasizing detail can introduce noise, resulting in distorted reconstruction.

[0004] Secondly, physical information neural networks, as a data-driven method that integrates physical laws, have been introduced to improve the physical rationality of flow field reconstruction. However, under strong turbulent conditions, the strict constraints of the Navier-Stokes equations may suppress the true turbulent fluctuation characteristics, causing the reconstruction results to deviate from reality. How to reasonably balance data-driven and physical-driven approaches in complex flow fields has become a technical challenge.

[0005] Furthermore, existing flow monitoring methods mostly focus on flow velocity calculation itself, lacking a refined grading and evaluation of the quality of raw data, and failing to provide a systematic quantification of uncertainty in the final flow results. This makes it difficult to assess the reliability of the monitoring results, posing risks in practical engineering decision-making.

[0006] Therefore, there is an urgent need for a new generation of flow monitoring technology that can intelligently sense flow field characteristics, adaptively adjust algorithm strategies, and quantify the reliability of results in order to meet the needs of accurate measurement in complex hydrodynamic environments. Summary of the Invention

[0007] To address the aforementioned problems, this invention provides an adaptive flow field calculation method based on an acoustic Doppler flowmeter. This method uses artificial intelligence technology to identify the flow field type in real time and adaptively selects the optimal reconstruction strategy and parameters accordingly. During the reconstruction process, the balance between data and physical constraints is dynamically adjusted based on flow field characteristics (such as turbulence intensity). Finally, the uncertainty of the entire process from data acquisition to flow rate calculation is quantified, outputting reliable flow rate results with confidence intervals. The specific technical solution is as follows:

[0008] S1: Deploy a Doppler current meter at the monitoring section to collect raw Doppler spectrum data, and simultaneously acquire the instrument's three-dimensional attitude information, spatial position coordinates, and real-time water level data of the section. All data are synchronized in time using a precision clock.

[0009] S2: Perform system background noise elimination and environmental noise suppression preprocessing on the raw spectrum data, and perform quality assessment and classification labeling on each data unit;

[0010] S3: Extract time-domain, frequency-domain, and spatial features from the preprocessed data to construct feature vectors, and input them into a pre-trained flow field classification model to identify the current dominant flow field type;

[0011] S4: Select the corresponding reconstruction strategy based on the identified flow field type, and dynamically adjust the key algorithm parameters according to the flow field characteristics during the reconstruction process;

[0012] S5: Based on the reconstruction strategy, the flow field dynamically generates a computational grid, performs flow integral calculation, and comprehensively evaluates data quality, model error and integral error to quantify the uncertainty of the output flow result;

[0013] S6: Performs anomaly diagnosis on velocity distribution and flow rate results, performs trend prediction based on historical data, and provides real-time monitoring and early warning of equipment status.

[0014] Preferably, step S2 includes:

[0015] S2.1: Collect spectral data in a still water environment to generate a baseline noise template spectrum, and dynamically subtract the noise template from the measured spectrum in subsequent measurements;

[0016] S2.2: Adaptive wavelet transform method is used to decompose and threshold the spectral data to separate noise from useful signals;

[0017] S2.3: Each data unit shall be marked with a quality rating based on signal strength, signal-to-noise ratio and spectral continuity. The marking categories shall include at least reliable, doubtful, discontinuous and invalid.

[0018] Preferably, in step S3:

[0019] The flow field classification model adopts a convolutional neural network architecture, takes a regularized two-dimensional time-frequency spectrum as input, and outputs a probability distribution of multiple preset flow field types. The preset flow field types include at least laminar flow, steady turbulent flow, strong turbulent flow, and vortex flow.

[0020] Preferably, the adaptive velocity field reconstruction and dynamic parameter adjustment in step S4 specifically includes:

[0021] When the flow field type is identified as laminar or steady turbulent, a thin plate spline interpolation algorithm based on physical constraints is adopted, and the bandwidth of the interpolation kernel function is dynamically adjusted according to the flow field stability parameters.

[0022] When the flow field type is identified as strong turbulence or vortex flow, a reconstruction algorithm based on physical information neural network is adopted, and the physical constraint weights are dynamically adjusted according to the turbulence intensity parameters.

[0023] Preferably, the kernel function bandwidth adjustment rule of the thin-plate spline interpolation algorithm is as follows:

[0024] h = h_base × exp(-k × σ);

[0025] Where h is the adjusted bandwidth, h_base is the baseline bandwidth, k is the sensitivity coefficient, and σ is the flow field stability parameter calculated based on the historical flow velocity standard deviation.

[0026] Preferably, the physical constraint weights λ of the physical information neural network are dynamically adjusted based on the turbulence intensity parameter TI:

[0027] λ=λ_max-(λ_max-λ_min)×sigmoid(α×(TI-TI0));

[0028] Where λ_max and λ_min are the upper and lower limits of the weight, respectively, α is the adjustment rate, and TI0 is the preset turbulence intensity threshold.

[0029] Preferably, it also includes a data recovery step:

[0030] Missing regions are identified based on data quality markers. When the area ratio of the missing region is less than the first threshold, interpolation based on surrounding reliable data combined with a generative adversarial network is used to perform local completion.

[0031] When the proportion of the missing region is greater than or equal to the first threshold, global reconstruction is performed using ensemble Kalman filtering data assimilation technology.

[0032] Preferably, the uncertainty quantification in step S5 includes:

[0033] Calculate the data measurement uncertainty components based on data quality markers;

[0034] The model reconstruction uncertainty components are calculated by comparing the differences in the outputs of different reconstruction models.

[0035] The numerical integral uncertainty components are calculated using grid convergence analysis.

[0036] The total composite standard uncertainty of the flow rate is obtained by combining the various uncertainty components, and the confidence interval of the flow rate is output at a preset confidence level.

[0037] Preferably, step S6 specifically includes:

[0038] S6.1: Extract key parameters of the current flow velocity distribution and compare them with a historical database of similar flow fields. If the parameter deviation exceeds the dynamic threshold, a graded alarm will be triggered.

[0039] S6.2: Based on the Long Short-Term Memory Network model, using historical flow velocity, flow rate and external hydrological data as input, short-term flow trend prediction and early warning are issued;

[0040] S6.3: Use control charts to monitor key performance indicators of equipment, and generate maintenance warnings when indicators show a degradation trend or exceed control limits.

[0041] An adaptive velocity field reconstruction and flow monitoring system based on intelligent flow field recognition, used to implement the method described in any of the above-mentioned embodiments, characterized in that it includes:

[0042] The data acquisition unit, including a Doppler current meter, an attitude sensor, and a positioning module, is used to perform step S1;

[0043] An edge processing unit, deployed at a field terminal, is used to execute steps S2 and S3;

[0044] The cloud-based intelligent analysis platform includes a flow field classification model library, an adaptive reconstruction engine, an uncertainty assessment module, and an early warning analysis module, which are used to execute steps S4, S5, and S6.

[0045] A visual interactive terminal is used to display monitoring results, uncertainty assessment information, and early warning notifications.

[0046] The beneficial effects of this invention are as follows:

[0047] Significantly improves measurement accuracy under complex flow fields: Through intelligent identification of flow field type, the system automatically matches the optimal reconstruction algorithm and parameters for flow fields with different characteristics (such as laminar flow, turbulent flow, and vortex flow), overcoming the shortcomings of poor adaptability of traditional single algorithms. The accuracy improvement is more obvious, especially under complex working conditions such as strong turbulence and vortex flow.

[0048] Adaptive optimization of algorithm parameters: An innovative dynamic parameter adjustment mechanism based on real-time flow field characteristics (such as stability and turbulence intensity) is introduced. For example, in PINN, the physical constraint weights are adaptively adjusted according to the turbulence intensity. In strong turbulence, the physical constraints are appropriately relaxed to avoid excessive suppression of real fluctuations, while in steady flow, the constraints are strengthened to ensure physical rationality, thereby achieving an intelligent balance between data fitting and physical laws.

[0049] It provides quantifiable credibility assessment across the entire value chain: a complete uncertainty assessment chain has been established, from raw data quality assessment to model reconstruction error analysis and numerical integration error quantification, ultimately outputting traffic results with confidence intervals. This makes the monitoring results no longer just single numerical values, but also includes clear credibility information, greatly enhancing the scientific value and decision support capabilities of the data.

[0050] Enhance the system's intelligence and automation: It integrates automatic diagnosis of flow field anomalies, intelligent prediction of short-term flow, and early warning of equipment health status, transforming the monitoring system from a passive data collection tool into a proactive intelligent monitoring and early warning platform, thereby reducing operation and maintenance costs and reliance on professional personnel.

[0051] It possesses excellent engineering applicability and scalability: The system adopts a cloud-edge-device collaborative architecture, which takes into account both real-time requirements and complex computing capabilities; the method design is compatible with various types of Doppler current meters and different installation methods, and has good universality; the algorithm model supports online updates and incremental learning, and can adapt to changes in the environment and equipment during long-term monitoring. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0053] The technical solutions in the embodiments of the present invention are clearly and completely described in the following description. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0054] In the description of the embodiments of the present invention, it should be noted that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is conventionally placed during use, or the orientation or positional relationship in which those skilled in the art conventionally understand it during use. This is only for the convenience of describing the present invention and simplifying the description, and is not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of the present invention. Furthermore, the terms "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0055] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0056] Embodiment 1 of the present invention discloses an adaptive flow field calculation method based on an acoustic Doppler flowmeter, such as... Figure 1 As shown, the specific method is as follows:

[0057] S1: Multi-dimensional data synchronous collection.

[0058] The data acquisition program in the edge gateway triggers a complete acquisition cycle at a frequency of 1Hz:

[0059] Send an acquisition command to ADCP to obtain the raw spectrum data matrix Spectrum(t0) at the current time t0, with dimensions of [4 beams × 30 depth cells × 128 frequency points].

[0060] Read the attitude data packet with the timestamp closest to t0 from the IMU buffer to obtain the roll angle φ, pitch angle θ, and heading angle ψ.

[0061] Obtain the location data with the closest timestamp t0 from the GPS receiver, including latitude L, longitude B, and elevation H.

[0062] Read the current value h_w from the ultrasonic water level gauge.

[0063] All data is packaged with a unified timestamp t0 to form a complete record, which is then uploaded to the cloud data receiving service via a 4G network.

[0064] The system employs the IEEE 1588-based PTP protocol. A PTP master clock (Grandmaster) is deployed in the field equipment room and connected to the edge gateway via Ethernet. The ADCP, IMU, and GPS devices act as PTP slave devices, and their internal clocks synchronize with the master clock by exchanging synchronization messages, ensuring that the timestamp deviation of all data is less than 10 microseconds.

[0065] S2: Data preprocessing and quality labeling.

[0066] S2.1 System background noise elimination

[0067] After the Doppler current meter was installed, 40 minutes of still water spectrum were collected during the nighttime period when there were no boats. For each frequency point f (f=1,2,...,128), the mean μ(f) and standard deviation σ(f) of the amplitude of all still water samples were calculated. The noise template function was defined as N(f)=μ(f)+3σ(f) and persistently stored. A frequency-amplitude noise template was generated through statistical analysis. During actual measurement, the template was dynamically subtracted using frequency domain subtraction. For each real-time spectrum data S_raw(f), the dynamic subtraction function S_clean(f)=max( S_raw(f)-N(f),0) was executed, where N(f) is the noise template function. Negative values ​​were set to zero to eliminate the inherent electronic noise of the system.

[0068] S2.2 Environmental noise suppression (adaptive wavelet denoising)

[0069] Wavelet transform is a powerful signal processing tool capable of decomposing signals into subbands of different frequencies. The Daubechies db6 wavelet basis is used to suppress noise in the spectral data after noise reduction in step S2.1. Specifically, the noise-reduced data is decomposed into five levels, and an improved semi-soft thresholding function is used to process the detail coefficients of each level. The threshold can be adaptively determined based on the median of the coefficients in that level. The final denoised spectrum is obtained after noise reduction using wavelet transform.

[0070] S2.3 Data Quality Grading Markers

[0071] For each data cell (corresponding to a specific beam b and depth cell d), its signal-to-noise ratio (SNR) is calculated. Quality is assessed and graded based on the SNR: SNR ≥ 10 dB is marked as "reliable"; 5 dB ≤ SNR < 10 dB is marked as "questionable"; and SNR < 5 dB is marked as "invalid". Furthermore, the velocity difference between adjacent depth cells is calculated; if the difference is greater than 0.3 m / s, it is marked as "discontinuous".

[0072] S3: Flow field feature extraction and intelligent recognition.

[0073] Features are extracted from the preprocessed, high-quality "reliable" data: time domain—mean signal strength; frequency domain—spectral width, peak frequency, spectral entropy; spatial domain—vertical velocity gradient, lateral correlation. These features are regularized into a 128×128 two-dimensional matrix and input into a pre-trained ResNet-18 model. This model was trained using 5000 sets of measured data and tens of thousands of sets of CFD simulation data (simulating laminar flow, turbulent flow, vortex shedding, etc.) covering various flow regimes. The network outputs a 4-dimensional vector P = [p1, p2, p3, p4], representing the probabilities of belonging to "laminar flow," "steady turbulent flow," "strong turbulent flow," and "vortex flow," respectively. argmax(P) is taken as the dominant flow field type at the current moment. For example, an output of [0.05, 0.10, 0.80, 0.05] indicates "strong turbulent flow."

[0074] S4: Adaptive flow velocity field reconstruction and dynamic parameter adjustment.

[0075] Based on the S3 identification results, the system automatically enters the corresponding reconstruction branch:

[0076] Scenario A: Identified as "laminar flow" or "steady turbulent flow"

[0077] Algorithm selection: Physical constraint reconstruction based on thin-plate spline interpolation is adopted. This method minimizes the bending energy (physical smoothness constraint) of the interpolated surface while requiring the interpolation result to be as close as possible to the observed value at the measurement point.

[0078] Dynamic parameter adjustment – ​​bandwidth h:

[0079] Calculate the flow field "stability parameter" σ: take the average of the standard deviations of the flow velocity at each point over the past 10 measurement cycles (10 seconds).

[0080] Bandwidth adjustment formula: h=h_base*exp(-k*σ), where h_base=2.0m is the base bandwidth and k=2.0 is the sensitivity coefficient.

[0081] Physical meaning: When σ is small (flow field is stable), h is larger, and the interpolation result is smoother; when σ is large (flow field fluctuates), h automatically decreases to retain more details of spatial changes.

[0082] Case B: Identified as "strong turbulence" or "vortex-containing flow"

[0083] Algorithm selection: Reconstruction based on Physical Information Neural Network (PINN) is adopted.

[0084] Network structure: A fully connected deep neural network, with spatial coordinates (x, y, z) as input and flow velocity components (u, v, w) at that point as output.

[0085] Loss function: L = L_data + λ * L_pde + L_bc;

[0086] L_data: The mean square error of the network output at the effective measurement point compared to the flow velocity retrieved by ADCP.

[0087] L_pde: The sum of squared residuals of the network output (u,v,w) at a large number of randomly selected "matching points" within the computational domain, which does not satisfy the simplified NS equation (such as the RANS equation).

[0088] L_bc: The error between the network output and the boundary conditions on known boundaries (such as riverbed or water surface).

[0089] λ: Physical constraint weight, which is a key dynamic parameter of this invention.

[0090] Dynamic parameter adjustment – ​​physical constraint weight λ:

[0091] Calculate the "turbulence intensity parameter" TI of the current flow field: TI = u_rms / u_mean, where u_rms is the root mean square of the velocity fluctuation and u_mean is the cross-sectional average velocity.

[0092] The sigmoid function is used to dynamically adjust λ: λ = λ_max - (λ_max - λ_min) * sigmoid(α * (TI – TI0)).

[0093] λ_max is the preset maximum value, λ_min is the preset minimum value, α is the adjustment rate, and TI0 is the preset initial turbulence intensity parameter.

[0094] Set λ_max=1.0, λ_min=0.1, α=10, TI0=0.25.

[0095] Physical meaning: When TI < 0.25 (weak turbulence), λ is close to 1.0, emphasizing the constraints of the physical equations and ensuring a reasonable flow field structure; when TI > 0.25 (strong turbulence), λ approaches 0.1, weakening the constraints of the Navier-Stokes equations, which may not be fully applicable, making the network more dependent on measured data, thereby preserving the real turbulent fluctuation structure.

[0096] Missing data repair:

[0097] For the "invalid" region marked in S2.3:

[0098] If the area of ​​the invalid region (as a percentage of the entire profile) is less than 15% (a set threshold), local completion is performed using Kriging interpolation combined with a Generative Adversarial Network (GAN). The GAN generator generates flow velocities in the missing region based on reliable surrounding data, and the discriminator judges the reasonableness of the generated results.

[0099] If the invalid region area is ≥ 15%, then ensemble Kalman filter (EnKF) data assimilation is used. A fast CFD model is used as the background field, and sparse ADCP observation data is assimilated into it. Iterative optimization yields a physically consistent global flow field.

[0100] S5: Flow calculation and uncertainty assessment.

[0101] S51. Based on the cross-sectional topography (pre-measured) and water level, generate a two-dimensional triangular mesh. Then, generate an adaptive computational mesh (basic resolution 0.5m × 0.5m) based on the reconstructed velocity field, and perform numerical integration.

[0102] Instantaneous flow rate:

[0103] Vi: The flow velocity at the center (or representative point) of the i-th grid cell;

[0104] Ai: The area of ​​the i-th grid cell;

[0105] n: The total number of effective mesh elements within the cross-section

[0106] S52. Uncertainty Measurement

[0107] Measurement uncertainty: calculated based on a weighted average of "reliable" and "doubtful" data.

[0108] Model uncertainty: Compare the differences between thin-plate spline and Kriging interpolation results.

[0109] Integral uncertainty: Convergence analysis was performed by refining the mesh to 0.25m.

[0110] Combined standard uncertainty:

[0111]

[0112] Expanded uncertainty: Take the coverage factor k1=2,

[0113] Instantaneous flow rate Q = 86.4, expanded uncertainty U = 5.2%.

[0114] Q*U≈4.5

[0115] Final result: (95% confidence level)

[0116] S6: Comprehensive intelligent diagnosis and early warning.

[0117] It performs anomaly diagnosis on flow velocity distribution and flow rate results, makes trend predictions based on historical data, and monitors and provides early warnings for equipment status in real time.

[0118] Specifically, it includes:

[0119] Flow field anomaly diagnosis: Calculate key parameters of the current velocity distribution, such as the cross-sectional average velocity V_avg, maximum velocity V_max, and velocity distribution non-uniformity coefficient C_v. Compare these parameters with the normal range (e.g., mean ± 3 standard deviation) of the corresponding flow field type (e.g., "strong turbulence") in the "similar flow field history database". If multiple parameters exceed the threshold simultaneously, a graded alarm is triggered (e.g., yellow alert - single parameter anomaly, red alert - multiple key parameters anomaly), and possible causes are indicated (e.g., "increased upstream water flow", "local blockage", etc.).

[0120] Short-term flow trend prediction: Deploy an LSTM prediction model with a Seq2Seq structure. Input features include: the average cross-sectional velocity sequence, flow rate sequence, rainfall sequence from the local station and upstream rain gauges, and water level sequence over the past 72 hours. The model outputs the flow rate prediction for the next 24 hours. If the predicted flow rate is likely to exceed the preset warning flow rate within the next 6 hours, the system automatically issues a warning to management personnel via SMS, platform messages, etc.

[0121] Equipment health status early warning: Establish time series of key performance indicators (KPIs) for the equipment, including transducer transmit voltage, average received signal strength, signal-to-noise ratio trend line, and attitude angle stability (variance). These KPIs are continuously monitored using an exponentially weighted moving average (EWMA) control chart. When a certain indicator exceeds its control limit for three consecutive monitoring periods, or shows a clear monotonically decreasing / increasing trend, a preventative maintenance early warning work order is generated, indicating possible faulty components (e.g., "There may be deposits on the transducer surface," "The IMU may need calibration").

[0122] Example 2: System Architecture and Workflow

[0123] This embodiment provides a monitoring system for implementing the above method, including:

[0124] Data acquisition unit: integrates ADI's ADCP core module, SBG Ellipse series IMU, and Huace Navigation RTK-GPS, and is powered and communicated through a customized data acquisition box.

[0125] Edge processing unit: Employs NVIDIA Jetson AGX edge computing devices, deploying lightweight preprocessing and feature extraction programs to achieve real-time data cleaning and compressed transmission.

[0126] Cloud-based intelligent analytics platform: Built on Alibaba Cloud ECS, using a microservice architecture. Includes:

[0127] Visual interactive terminal: Develop web applications and WeChat mini-programs based on Vue.js. Real-time display of flow velocity cloud maps, flow process lines (with confidence intervals), flow field type evolution diagrams, warning lists, etc. Supports multi-user access and automatic report generation.

[0128] Each unit is networked via a hybrid 4 / 5G or fiber optic network. The edge processing unit processes the data from the data acquisition unit and sends it to the cloud for analysis. Finally, the analysis results are displayed through a visual interactive terminal.

[0129] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. An adaptive flow field calculation method based on an acoustic Doppler flowmeter, characterized in that, include: S1: Deploy a Doppler current meter at the monitoring section to collect raw Doppler spectrum data, and simultaneously acquire the instrument's three-dimensional attitude information, spatial position coordinates, and real-time water level data of the section. All data are synchronized in time using a precision clock. S2: Perform system background noise elimination and environmental noise suppression preprocessing on the raw spectrum data, and perform quality assessment and classification labeling on each data unit; S3: Extract time-domain, frequency-domain, and spatial features from the preprocessed data to construct feature vectors, and input them into a pre-trained flow field classification model to identify the current dominant flow field type; S4: Select the corresponding reconstruction strategy based on the identified flow field type, and dynamically adjust key algorithm parameters according to the flow field characteristics during the reconstruction process, specifically including: When the flow field type is identified as laminar or steady turbulent, a thin plate spline interpolation algorithm based on physical constraints is adopted, and the bandwidth of the interpolation kernel function is dynamically adjusted according to the flow field stability parameters. The kernel function bandwidth adjustment rule for the thin-plate spline interpolation algorithm is as follows: h = h_base × exp(-k × σ) Where h is the adjusted bandwidth, h_base is the baseline bandwidth, k is the sensitivity coefficient, and σ is the flow field stability parameter calculated based on the historical flow velocity standard deviation. When the flow field type is identified as strong turbulence or vortex flow, a reconstruction algorithm based on physical information neural network is adopted, and the physical constraint weights are dynamically adjusted according to the turbulence intensity parameters. The physical constraint weights λ of the physical information neural network are dynamically adjusted based on the turbulence intensity parameter TI: λ=λ_max-(λ_max-λ_min)×sigmoid(α×(TI-TI0)) Where λ_max and λ_min are the upper and lower limits of the weight, respectively, α is the adjustment rate, and TI0 is the preset turbulence intensity threshold; S5: Based on the reconstruction strategy, the flow field dynamically generates a computational grid, performs flow integral calculation, and comprehensively evaluates data quality, model error and integral error to quantify the uncertainty of the output flow result; S6: Performs anomaly diagnosis on velocity distribution and flow rate results, performs trend prediction based on historical data, and provides real-time monitoring and early warning of equipment status.

2. The adaptive flow field calculation method based on an acoustic Doppler flowmeter according to claim 1, characterized in that, Step S2 includes: S2.1: Collect spectral data in a still water environment to generate a baseline noise template spectrum, and dynamically subtract the noise template from the measured spectrum in subsequent measurements; S2.2: Adaptive wavelet transform method is used to decompose and threshold the spectral data to separate noise from useful signals; S2.3: Each data unit shall be marked with a quality rating based on signal strength, signal-to-noise ratio and spectral continuity. The marking categories shall include at least reliable, doubtful, discontinuous and invalid.

3. The adaptive flow field calculation method based on an acoustic Doppler flowmeter according to claim 1, characterized in that, In step S3: The flow field classification model adopts a convolutional neural network architecture, takes a regularized two-dimensional time-frequency spectrum as input, and outputs a probability distribution of multiple preset flow field types. The preset flow field types include at least laminar flow, steady turbulent flow, strong turbulent flow, and vortex flow.

4. The adaptive flow field calculation method based on an acoustic Doppler flowmeter according to claim 1, characterized in that, It also includes data recovery steps: Missing regions are identified based on data quality markers. When the area ratio of the missing region is less than the first threshold, interpolation based on surrounding reliable data combined with a generative adversarial network is used to perform local completion. When the proportion of the missing region is greater than or equal to the first threshold, global reconstruction is performed using ensemble Kalman filtering data assimilation technology.

5. The adaptive flow field calculation method based on an acoustic Doppler flowmeter according to claim 1, characterized in that, The uncertainty quantification in step S5 includes: Calculate the data measurement uncertainty components based on data quality markers; The model reconstruction uncertainty components are calculated by comparing the differences in the outputs of different reconstruction models. The numerical integral uncertainty components are calculated using grid convergence analysis. The total composite standard uncertainty of the flow rate is obtained by combining the various uncertainty components, and the confidence interval of the flow rate is output at a preset confidence level.

6. The adaptive flow field calculation method based on an acoustic Doppler flowmeter according to claim 1, characterized in that, Step S6 specifically includes: S6.1: Extract key parameters of the current flow velocity distribution and compare them with a historical database of similar flow fields. If the parameter deviation exceeds the dynamic threshold, a graded alarm will be triggered. S6.2: Based on the Long Short-Term Memory Network model, using historical flow velocity, flow rate and external hydrological data as input, short-term flow trend prediction and early warning are issued; S6.3: Use control charts to monitor key performance indicators of equipment, and generate maintenance warnings when indicators show a degradation trend or exceed control limits.

7. A flow field adaptive calculation system based on an acoustic Doppler flowmeter, used to implement the method according to any one of claims 1-6, characterized in that, include: The data acquisition unit, including a Doppler current meter, an attitude sensor, and a positioning module, is used to perform step S1; An edge processing unit, deployed at a field terminal, is used to execute steps S2 and S3; The cloud-based intelligent analysis platform includes a flow field classification model library, an adaptive reconstruction engine, an uncertainty assessment module, and an early warning analysis module, which are used to execute steps S4, S5, and S6. A visual interactive terminal is used to display monitoring results, uncertainty assessment information, and early warning notifications.

Images