Multi-layer flow rate anomaly identification and reconstruction method based on physical constraints and robust regression

Abstract

The present application relates to the technical field of hydrological monitoring and hydrological data processing, and particularly relates to a multilayer flow velocity anomaly identification and reconstruction method based on physical constraints and robust regression. The method comprises the following steps: obtaining synchronous monitoring data of a target hydrological monitoring section; constructing a single-layer flow velocity pure space calibration model based on real-time flow measurement data of each flow layer at the same test time, and deducing the theoretical spatial expected value of the target easily disturbed surface layer; comparing the measured flow velocity of the target easily disturbed surface layer with the theoretical spatial expected value, identifying and marking abnormal outliers; constructing a physical drawdown zone state judgment condition to separate normal exposure off-target zero values and underwater signal failure abnormal zero values; extracting the full-profile active effective data set in the synchronous monitoring data, constructing a cross-coupling prediction matrix, and performing multidimensional targeted joint reconstruction to output flow measurement reconstruction data. The present application helps to improve the physical consistency, anti-interference ability and section flow calculation accuracy of online flow measurement data.

Description

Technical Field

[0001] This invention relates to the field of hydrological monitoring and hydrological data processing technology, and in particular to a method for identifying and reconstructing multi-layer velocity anomalies based on physical constraints and robust regression. Background Technology

[0002] With the advancement of "digital twin watershed" construction, ultrasonic time-of-flight flowmeters are widely used in online flow measurement systems for large and medium-sized rivers due to their non-contact measurement and multi-layer velocity acquisition capabilities. However, under complex hydrological conditions, the monitoring data is prone to problems such as surface velocity jumps, mid-to-bottom signal attenuation, and abnormal zero drops due to water level fluctuations, floating debris interference, and sediment concentration changes. This results in a large number of abnormal data mixed in with the original velocity sequence, which in turn affects the accurate calculation of cross-sectional flow.

[0003] Existing technologies still have certain limitations in handling outlier data: traditional statistical methods are prone to benchmark distortion when outliers are clustered, leading to a decrease in identification ability; time-series filtering methods rely on the assumption of stationarity, which can easily produce lag and tailing errors under continuous abnormal operating conditions; pure data-driven methods or deep learning models lack physical constraints, making it difficult to distinguish between real hydraulic changes and equipment or signal anomalies, and they do not fully consider the hydrodynamic coupling relationship between multiple flow layers, which can easily produce correction results that do not conform to physical laws under complex operating conditions. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a multi-layer velocity anomaly identification and reconstruction method based on physical constraints and robust regression to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression is proposed, comprising the following steps: Step S1: Obtain synchronous monitoring data of the target hydrological monitoring section, including real-time flow measurement data of each flow layer and real-time water level data; Step S2: Based on the real-time flow measurement data of each flow layer at the same test time, construct a single-layer flow velocity pure spatial verification model, and use the Huber robust regression method to map and fit the real-time flow measurement data of the underwater stable flow layer to deduce the theoretical spatial expected value of the target surface layer that is susceptible to disturbance. Step S3: Compare the measured flow velocity of the target's easily disturbed surface with the theoretical spatial expected value, determine the abrupt change characteristics of the surface flow velocity through the dual threshold joint judgment logic, and identify and mark abnormal flying points that deviate from the cross-sectional hydrodynamic distribution law. Step S4: Based on the real-time water level data, the absolute installation elevation of the transducers corresponding to each flow layer, and the physical drawdown zone status judgment conditions constructed by superimposing the preset water level safety buffer zone distance, perform physical status judgment on the zero or null value sequence in the real-time flow measurement data, and separate the normal exposed off-target zero value from the abnormal zero value of underwater signal failure. Step S5: Extract the full profile active effective dataset from the synchronous monitoring data after removing normal exposure miss zero values ​​and underwater signal failure anomaly zero values, and construct a cross-coupled prediction matrix; based on the cross-coupled prediction matrix, use the median voting mechanism to perform multi-dimensional targeted joint reconstruction on the measurement points corresponding to the abnormal flying points and underwater signal failure anomaly zero values, thereby outputting the current measurement reconstruction data.

[0006] This application constructs a spatial mapping relationship between multiple flow layers at the same time and introduces a Huber robust regression mechanism, effectively replacing the traditional processing method that relies on time series smoothing. This significantly improves the anti-interference capability against long-term extreme anomalies and "flying point clusters," achieving high-fidelity reconstruction of the true vertical velocity structure of the cross section. Simultaneously, by introducing a physical drawdown zone determination mechanism based on the coupling of transducer installation elevation and real-time water level, it achieves a refined distinction between "normal off-target zero values" and "underwater signal failure zero values," avoiding misjudgments and non-physical interpolation in zero-value scenarios by traditional data-driven methods. Furthermore, by constructing a cross-coupled prediction matrix across all flow layers and combining it with a high-collapse-point median voting mechanism, it achieves high-fault-tolerant targeted reconstruction through multi-flow layer collaboration, maintaining the stability and consistency of output results even under conditions of multiple anomalies. Finally, this application achieves highly reliable online velocity data quality control and reconstruction in environments with multi-source noise, complex backwater, and unattended operation, providing a stable data foundation for high-precision sensing and accurate calculation of cross-sectional flow in digital twin watersheds. Attached Figure Description

[0007] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps of the multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression of the present invention. Figure 2 This is a schematic diagram illustrating the identification of physical drawdown zone state determination conditions in one embodiment; Figure 3 This is an example of a joint reconstruction of the data flow diagram across the entire flow layer based on a median voting mechanism. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0008] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0009] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0010] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0011] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for identifying and reconstructing multilayer flow velocity anomalies based on physical constraints and robust regression. The method includes the following steps: Step S1: Obtain synchronous monitoring data of the target hydrological monitoring section, including real-time flow measurement data of each flow layer and real-time water level data; In one embodiment of this application, synchronous monitoring data of a target watershed section, such as a complex section affected by the combined effects of cascade hydropower station regulation and downstream backwater, is acquired through a hydrological monitoring platform or station terminal. This data includes real-time velocity sequence data for each layer collected by a multi-layer ultrasonic time-of-flight online flowmeter, and real-time water level data synchronously collected by a high-precision water level gauge. Simultaneously, the absolute installation elevation parameters of each layer of transducers are read, and combined with historical test data of the transducers, the effects of wind and wave run-up and water surface stagnation are assessed, and a safe buffer zone distance for the system water level is preset.

[0012] Step S2: Based on the real-time flow measurement data of each flow layer at the same test time, construct a single-layer flow velocity pure spatial verification model, and use the Huber robust regression method to map and fit the real-time flow measurement data of the underwater stable flow layer to deduce the theoretical spatial expected value of the target surface layer that is susceptible to disturbance. In a further embodiment, surface velocity "flypoints" caused by surface waves or multipath effects often have large durations and extreme numerical distortions. In such cases, the time series assumption fails, and the filtering window is easily contaminated by continuous outliers. Therefore, this application completely abandons the time series dimension and establishes a pure spatial mapping mechanism between multiple flow layers at the same moment. Since the transducers in each layer of the online flow measurement system are installed at different elevations, they are all controlled by the hydrodynamic laws of the same cross-section at the same moment, and there is a strong physical linkage between the instantaneous velocities of each water layer. Assuming the station contains multiple velocity measurement devices, for the most easily disturbed target layer, its true hydraulic expectation value is jointly derived from the other relatively stable flow layers at the bottom: In the formula, for The theoretical spatial expectation of the target layer at any given time; The measured velocity is that of a reference flow layer that is simultaneously located in the deep underwater layer and has strong anti-interference capabilities. This is the spatial mapping derivation function form expressed in the constructed single-layer velocity pure spatial verification model. Its essence is used to characterize the target's easily disturbed surface layer (the first layer). The spatial coupling relationship between the hydrosphere and other stable underwater reference hydrospheres under the same time-space conditions, where the function The parameterized mapping operator corresponding to the model is used to map the instantaneous velocity input of the multi-reference flow layer to the theoretically desired output of the target layer. The mapping function is constructed using measured samples. In practice, the target layer training set inevitably contains extreme outliers several times the actual flow rate. Using ordinary least squares would amplify the weights of these outliers, leading to a deviation in the fit. Therefore, this application introduces Huber robust regression, whose loss function combines the advantages of absolute error and squared error, defined as: ; ; In the formula, To predict residuals; A threshold parameter used to define normal fluctuations and outliers. When the absolute value of the residual is less than the threshold... When the model uses squared loss, it ensures the fitting accuracy for normal flow pulsation intervals; when encountering extreme flypoints, the residual error is greater than... At that time, it automatically switches to linear loss and applies a constant penalty to outliers. At the same time, it forces the intercept to be zero to satisfy the physical boundary condition that the entire cross-section of the water flow is reduced to zero when it is at rest.

[0013] It is worth noting that during the training process of the single-layer velocity pure spatial verification model, the measured velocity of the underwater stable flow layer after removing abnormal contamination is used as the training sample input, and the "reference flow layer set" is constructed into the input feature vector. Using the observed flow velocity of the target's easily disturbed surface as the supervision label, parameter fitting is achieved by minimizing the Huber robust regression loss function, thus making the mapping function... While ensuring the accuracy of fitting normal hydrodynamic fluctuations, it maintains the ability to suppress flybys and extreme outliers. The specific training process is as follows: First, a stable underwater reference flow layer is selected from the cross-sectional data at the same time to form training sample pairs (spatial input vector - target output value); second, the samples are input into a regression solution process without intercept constraints, and the intercept term is fixed at 0 to satisfy the still water boundary conditions; then, the mapping parameters are updated through iterative optimization to make the Huber loss function converge to the stable threshold range; finally, the converged function is obtained. This is the core spatial deduction structure of the single-layer velocity pure spatial verification model, used for the real-time calculation of the expected theoretical spatial value of the target surface layer.

[0014] Step S3: Compare the measured flow velocity of the target's easily disturbed surface with the theoretical spatial expected value, determine the abrupt change characteristics of the surface flow velocity through the dual threshold joint judgment logic, and identify and mark abnormal flying points that deviate from the cross-sectional hydrodynamic distribution law. In a further embodiment, to avoid excessive interference with the normal, minute pulsations of water flow, the theoretically expected values ​​for the entire time period are obtained. Then, an absolute tolerance difference was designed. and relative multiples The logic for joint determination of dual thresholds: The first condition is used to filter out normal small-amplitude water wave interference (such as setting...). The second condition confirms whether the jump constitutes an aberrant mutation (e.g., setting...). Only when When true ( =1 Mark it as an anomalous flypoint and force the use of the theoretical space expected value. Replace and reconstruct; the remaining smaller values ​​or normal fluctuation points retain their original appearance.

[0015] Step S4: Based on the real-time water level data, the absolute installation elevation of the transducers corresponding to each flow layer, and the physical drawdown zone status judgment conditions constructed by superimposing the preset water level safety buffer zone distance, perform physical status judgment on the zero or null value sequence in the real-time flow measurement data, and separate the normal exposed off-target zero value from the abnormal zero value of underwater signal failure. In a further embodiment, since a purely mathematical model cannot identify the environmental state of the transducer, a high-precision water level sequence with time synchronization is introduced. As a core physical constraint, firstly, the set of measured points strictly equivalent to 0.0 or marked as missing (NaN) in the real-time flow measurement data of each flow layer was extracted. Then, the absolute installation elevation of the combined transducer was determined. and water level safety buffer zone distance State classification is performed using a logic matrix: 1) Normal exposure judgment (physical miss): when there is no flow velocity at the measuring point and the conditions are met. When zero values ​​are found to be normal zero drops due to physical conditions, this data will be retained and marked as "normal (surface)" in subsequent processing and will not be included in interpolation reconstruction to conform to hydraulic logic. 2) Underwater zero drop detection (signal anomaly): When there is no flow velocity at the measuring point but the conditions are met... If the absence of flow velocity is detected, it is determined to be a false zero caused by equipment signal loss or strong turbulence interference. It is then marked as "abnormal (underwater zero drop)" and added to the reconstructing sequence.

[0016] Figure 2 This is a schematic diagram illustrating the criteria for determining the state of the physical settling zone in one embodiment; as shown below. Figure 2 As shown, when performing layered measurements on the cross-section of the watershed, the vertical profile of the river channel is divided into flow layers 1 to 4. Flow layer 1 is located near the bottom of the riverbed, and flow layer 4 is located near the water surface. Ultrasonic time-of-flight transducers are deployed in each flow layer, with the absolute installation elevation of each transducer serving as the baseline. A real-time water level H is introduced to characterize the current dynamic position of the water surface, and a preset safety buffer zone B is superimposed above the water level to depict the fluctuation margin caused by wind and wave rise and water surface movement. As can be seen from the figure, when the water level H drops below the installation elevation corresponding to flow layer 4 and enters the "normal exposure and off-target" region, the transducer in this layer is exposed to air, and its measurement value shows a stable zero value. However, this zero value is a normal zero output caused by physical off-target. In the vicinity of flow layer 2, if the water level H is still higher than its installation elevation... But at the same time, it exceeds ( If the threshold range is exceeded, the layer, although submerged underwater, may experience abnormal underwater zero-signal drops due to acoustic signal attenuation caused by strong turbulence or sudden changes in sediment content. The "safety buffer zone B" in the diagram is the transitional zone used to distinguish between normal submersion and abnormal signal failure. Overall, this judgment mechanism uses water level H and installation elevation... With buffer zone The spatial superposition relationship enables the zonal identification of measurement points in different flow layers (normal underwater measurement, normal exposure and target failure, and abnormal zero drop underwater), thus providing a physically consistent classification basis for subsequent flow velocity reconstruction. In the figure, water level H represents the raw water level observation value directly output by the water level gauge in real time. This is an engineering measurement without mathematical abstraction, used to reflect the actual spatial position of the water surface relative to the reference elevation at the current moment, primarily serving as the input observation signal. Water level This indicates the theoretical or normalized water level state variable introduced into the judgment model, which is used in conjunction with the transducer installation elevation. and safety buffer zone To provide a unified mathematical relational expression, for example, for constructing or These are the conditions for judgment, meaning they act as the variables for judgment.

[0017] Step S5: Extract the full profile active effective dataset from the synchronous monitoring data after removing normal exposure miss zero values ​​and underwater signal failure anomaly zero values, and construct a cross-coupled prediction matrix; based on the cross-coupled prediction matrix, use the median voting mechanism to perform multi-dimensional targeted joint reconstruction on the measurement points corresponding to the abnormal flying points and underwater signal failure anomaly zero values, thereby outputting the current measurement reconstruction data.

[0018] In a further embodiment, to achieve effective reconstruction of abnormal measurement points, a full-profile active effective dataset is extracted after removing the aforementioned abnormal points and normal off-target points. Since the flow velocities of any two layers within the cross-section exhibit significant linear coupling within the effective dataset, a Huber regression model without intercept is used to reconstruct the full profile. Each flow layer is subjected to pairwise cross-fitting to generate Scale matrix of flow layers of order : ; matrix elements Representing the Laminar flow velocity on the first Robust extrapolation ratio of layer velocity. To avoid excessive intervention in normal flow measurement data and ensure the accuracy of data quality control, the triggering condition of the reconstruction mechanism is first determined at this point: if and only if the target flow layer... The multi-dimensional targeted joint reconstruction mechanism in this step is only triggered when the current flow measurement data is marked as "spatial scale abnormal flying point" or "abnormal (underwater zero drop)"; if the target flow layer If the current state is marked as normal fluctuation or normal drawdown zone physical off-target exposure, the reconstruction algorithm will not be activated, and its original measured value or normal zero value background will be directly retained. For the target flow layer to be reconstructed that meets the above reconstruction triggering conditions... To obtain other effective reference flow layers within the profile Derived Individual Independent Predictions To maximize the fault tolerance of the reconstruction mechanism and avoid the degradation and failure of mathematical operators due to an insufficient number of reference flow layers, the number of available effective reference flow layers at the current moment is counted. (i.e., the number of normal tropospheric layers that are submerged underwater and have not experienced signal drop-off or flyby), and introduces an adaptive operator switching mechanism based on quantity constraints: (a) When the number of effective reference flow layers At that time, a median voting mechanism is used to perform multidimensional reconstruction: ; The median operator is mathematically highly stable, meaning that even if nearly half of the reference flow layers participating in the voting experience synchronization anomalies, the final generated expected value can still maintain extremely high stability. (ii) When the number of effective reference flow layers When the number of available reference layer samples is too small, the median voting operator will mathematically degenerate or even become ineffective. In this case, the algorithm automatically triggers degradation control, adaptively switching to the weighted average operator to replace the median voting. By calculating the weighted average of each effective reference flow layer as the expected output for reconstruction, the reconstructed flow velocity results are guaranteed to remain hydrodynamically and physically reasonable even under extremely low effective reference layer conditions.

[0019] It is worth noting that, regardless of whether median voting or weighted average calculation is used, the formula ends with... All functions are pre-set with a positive, small baseline in hydrodynamics to prevent the division-by-zero algorithm from failing in subsequent multiplier-based verification logic due to the absolute zero value derived from the still water at the bottom layer. Finally, the joint reconstruction expectation value generated by the adaptive operator is used to accurately target and replace the abnormal measurement points in the target layer, thereby generating high-quality flow velocity results with complete full-profile structure and smooth hydraulic characteristics.

[0020] It is worth noting that the construction method of the intercept-free Huber regression model includes: if there is a stable spatial hydrodynamic coupling relationship between the effective reference flow layer and the target predicted flow layer at the current moment, then the measured velocity data of multiple flow layers at the same test moment are used as the training sample set. The velocity sequence of the reference flow layer with strong underwater anti-interference capability is used as the input independent variable, and the measured velocity of the target easily disturbed flow layer is used as the output dependent variable, constructing a linear mapping relationship without constant bias terms. During the model parameter solution process, to suppress the interference of abnormal flypoints caused by wind and wave disturbances, multipath effects, and instantaneous turbulence on the fitting process, a Huber robust loss function is introduced to perform piecewise constraint processing on the residuals. When the absolute value is within the preset robustness threshold range, a quadratic error function is used to ensure the fitting accuracy of the normal hydrodynamic fluctuation range. When the absolute value of the residual exceeds the threshold, a linear penalty mechanism is automatically switched to reduce the impact of outliers on parameter updates. At the same time, since each flow layer should satisfy the physical boundary characteristic of the flow velocity across the entire cross section approaching zero under still water conditions, the intercept term of the model is forcibly constrained to zero during parameter optimization, so that the mapping function maintains physical consistency and does not produce offset under the condition of no external driving. Finally, the robust regression coefficients of each reference flow layer to the target flow layer are obtained by iterative optimization, thus forming a non-intercept Huber space mapping model that satisfies the ability to resist abnormal interference and hydrodynamic physical constraints.

[0021] Figure 3 To jointly reconstruct the data flow graph across all flow layers based on the median voting mechanism; such as Figure 3 As shown, the entire spherical joint reconstruction process unfolds based on the vertical structure of layers 1 to 4. Layers 1 to 3 are primarily used as underwater stable reference layers, while layer 4, as the target prediction layer, is located in the uppermost region susceptible to wind, waves, and surface turbulence. The data flow in the figure shows that each underwater reference layer first outputs its individual predictions to the upper layers based on its measured current velocity through a robust mapping relationship similar to Huber regression. This process corresponds to the multiple arrow paths from bottom to top in the figure, representing the independent inference contributions of different layers to the target layer. Simultaneously, an anti-abnormal disturbance mechanism is introduced in each cross-layer mapping to weaken local fly-spots or abrupt changes, ensuring that the output of a single path is not dominated by extreme values. After obtaining multiple sets of independent prediction results from layers 1, 2, and 3 to layer 4, the system performs a median voting fusion operation at the target layer convergence node, robustly aggregating the multi-source prediction results to reduce the impact of single-layer error propagation. The "×" marks in the figure indicate data points identified as abnormal interference or dropped zeros, which do not participate in subsequent fusion calculations. Finally, the fused results are backfilled to the corresponding measurement points in flow layer 4 to achieve the reconstruction output of the target layer velocity.

[0022] In another embodiment, a hydrological station in a certain area is equipped with a four-layer ultrasonic time-of-flight online flow meter. The absolute installation elevations of the four pairs of transducers from bottom to top are 173.90m (velocity 1), 175.40m (velocity 2), 177.10m (velocity 3), and 178.60m (velocity 4), respectively. Due to the interference of floating debris and wind waves, the uppermost transducer (velocity 4) is prone to long-term, high-multiplier abnormal velocity abrupt changes. In this embodiment, the model in steps S2 and S3 is called, and the absolute tolerance difference is set to 0.4m / s, and the relative abrupt change factor is 1.5. When disturbed by shallow water waves, velocity 4 will suddenly jump to over 1.0m / s and exhibit a "continuous high-value plateau". Traditional time-series filtering will inevitably be "lifted" by this batch of continuous flying points, resulting in obvious tailing errors. The algorithm of this invention directly blocks the interference of the time dimension. It uses the relatively stable underwater flow velocities 1 to 3 at the same moment as the independent variable matrix, and forcibly fits the spatial physical expectation of the flow velocity 4 at the current moment through the Huber operator. The experimental correction results show that the algorithm accurately identifies and reconstructs the abrupt anomaly using the theoretical expected value, completely eliminates the contamination of "continuous flying points", and at the same time fully retains the normal water surface pulsations that do not exceed the absolute tolerance of 0.4 m / s, demonstrating an excellent balance between "eliminating large values" and "fidelity".

[0023] In another embodiment, for the case of occasional zero values ​​and complex water level fluctuations across the entire cross-section, the system executes the joint reconstruction algorithm of the entire flow layer in steps S4 and S5, with the elevation buffer zone set at 0.15m. During the typical drawdown period of water level fluctuations, the system performs refined discrimination of the physical drawdown zone state: when the water level drop causes the transducer to actually be in the water-discharging state, the algorithm accurately strips the physical zero value without performing any unnecessary correction calculations, retaining the zero value background; when the water level is significantly higher than the transducer installation elevation, but a certain flow layer still outputs a zero value, the algorithm captures it and marks it as "abnormal (underwater zero drop)". For this abnormal point, the system aggregates the individual predicted values ​​of the remaining active flow layers within the profile and takes the median as the final reconstruction output. The results show that even with extremely low flow velocity disturbances of 0.01~0.03m / s in the lower layer, the median voting mechanism still ensures the stability of the reconstruction expectation and forcibly lowers the output baseline to 0.01m / s, maintaining the completeness of the mathematical calculation.

[0024] Optionally, the construction of a single-layer velocity pure spatial verification model in step S1 includes: The measured velocity data of several flow layers at the same test time are extracted from the real-time flow measurement data. Based on the measured velocity data of the cross sections, the flow layer with the most severe surface turbulence interference is selected as the target prediction layer. Combined with the measured velocity data of the reference flow layers in other flow layers whose underwater anti-interference ability is higher than the preset anti-interference ability threshold, the theoretical spatial expectation derivation function of the target prediction layer is constructed. In training the theoretical spatial expectation extrapolation function using measured velocity data from the reference flow layer, the loss function during the training process is set to the Huber robust regression loss function, which includes the absolute error norm and the squared error norm. The intercept term is forced to be zero to satisfy the physical boundary condition that the still water velocity is reduced to zero, thereby constructing a single-layer velocity pure spatial verification model.

[0025] In this embodiment, the instantaneous velocity sequence of each flow layer in the cross-section is synchronously read from the measured velocity data at the same test time, and a current profile vector is formed according to the flow layer number, for example, including data from 5-12 ultrasonic time-of-flight measurement points. Then, based on the short-term fluctuation amplitude (e.g., 5-minute sliding standard deviation)... The system uses historical stability scores (≥0.8) to select relatively stable underwater reference flow layers. Simultaneously, surface flow layers, strongly influenced by surface waves with fluctuations exceeding 0.35 m / s, are identified as target prediction layers for subsequent modeling and extrapolation. Throughout this process, all flow layer data are aligned to the same timestamp to avoid structural biases caused by time misalignment. In the model building phase, the velocity sequences of the selected reference flow layers are used as input features, and the velocity of the target prediction layer is used as the output response. A spatial mapping function structure without constant bias terms is established, meaning the output is entirely determined by a linear combination of the reference flow layers. During training, the Huber loss function is used to constrain the error, with the threshold parameter... The velocity is set between 0.2 and 0.5 m / s to distinguish between normal flow pulsations and anomalous flying point disturbances. When the error is within the threshold, a squared penalty is used to ensure the smoothness of the fit; when it exceeds the threshold, a linear penalty is switched to suppress the influence of extreme outliers. At the same time, by constraining the intercept term to always be 0, the model will not produce spurious offsets in the ideal state of no flow, thus ensuring that the spatial extrapolation results meet the physical consistency requirement of zero still water velocity.

[0026] It is worth noting that the target vulnerable surface layer and the target prediction layer are functionally layered under the same conditions at the same time in the same cross-sectional flow field structure. The target prediction layer serves as the benchmark for spatial theoretical deduction, while the target vulnerable surface layer reflects the observation layer most significantly affected by surface disturbances or multipath effects. In practice, the system synchronously samples the flow velocity of each flow layer at the same test time. Based on short-term fluctuation intensity indicators (e.g., flow velocity standard deviation σ ≥ 0.3 m / s within a 3-10 minute window) and signal stability scores (e.g., signal-to-noise ratio below 0.75), each layer is ranked and determined. The flow layer with the most intense fluctuations and the highest anomaly density is identified as the target vulnerable surface layer, while the layer below it, which is attenuated by surface disturbances and meets stability thresholds (e.g., [missing information]). The underwater flow layers with velocity ≤0.15 m / s and a continuous effective sampling rate ≥95% are defined as the reference flow layer set. Further, the target prediction layer is the modeling object selected from the target's vulnerable surface layer. Its values ​​do not directly participate in the training input but serve as the output of the derivation, used to receive the theoretical velocity mapping results constructed by the reference flow layers through spatial coupling relationships. This achieves a structural prediction relationship of "stable layer deducing unstable layer." In most implementation scenarios, the target's vulnerable surface layer and the target prediction layer can be the same physical flow layer, distinguished only by their observed and derivation attributes in modeling semantics.

[0027] Optionally, the calculation logic of the Huber robust regression loss function includes: Calculate the absolute value of the residual between the measured value and the predicted value output by the theoretical spatial expectation extrapolation function in the measured flow velocity data; When the absolute value of the residual is less than or equal to the preset robustness threshold parameter, the squared loss function is used: ; When the absolute value of the residual is greater than the robustness threshold parameter, a linear penalty loss function is used: ; In the formula It is the absolute value of the residual between the measured value and the predicted value. This is a robust threshold parameter.

[0028] In this embodiment, the measured velocity of the target flow layer at the same time is aligned point by point with the predicted velocity output by the spatial extrapolation function, for example, by forming a paired sequence synchronously at a sampling frequency of 1 Hz. Then, the difference between the two is calculated and the absolute value is taken as the residual index. This residual is used to reflect the degree to which the current measuring point deviates from the theoretical hydrodynamic structure. In the error processing stage, a robust threshold parameter is introduced. The robustness threshold parameter is set at approximately 0.3 m / s based on historical cross-sectional fluctuation statistics. When the absolute value of the residual is less than or equal to this threshold, it indicates that the deviation falls within the normal flow pulsation range. In this case, a quadratic penalty method is used for updating, causing the loss function to... The formal constraints on small errors are enhanced, thus ensuring that the fitted curve remains smooth and continuous within the normal flow range. When the absolute value of the residual exceeds... When the system determines that the measurement point may be affected by flying points or signal abrupt changes, it switches to a linear penalty mechanism to mitigate these interferences. The form linearly suppresses the error while subtracting a constant correction term. This is used to maintain the continuity of the loss function, so that outliers do not dominate the parameter update process, thereby avoiding excessive bias of the spatial extrapolation function structure by extreme noise.

[0029] Optionally, the abrupt change characteristics of surface flow velocity determined by the dual-threshold joint judgment logic in step S3 include: The joint logic is satisfied if and only if the absolute difference between the measured flow velocity and the theoretical spatial expected value of the target's vulnerable surface is greater than a preset absolute tolerance difference threshold, and the measured flow velocity of the target's vulnerable surface is greater than the theoretical spatial expected value multiplied by a preset relative multiple threshold. ; The measurement points on the target's easily disturbed surface are marked as abnormal flying points with abnormally large spatial proportions, and reconstructed and replaced using theoretical spatial expected values; In the formula, To measure the flow rate, This is the expected value in the theoretical space. This is the absolute tolerance threshold. The relative multiple threshold, This is an indicator for identifying abnormal flying points.

[0030] In this embodiment, the measured flow velocity on the target's vulnerable surface is compared point-by-point with the theoretical value output by the spatial extrapolation model at the same time. Typically, a 1Hz sampling frequency is used to form synchronous data pairs, and the difference and ratio between the two are calculated to determine whether the point deviates from the normal hydrodynamic distribution range. The absolute difference reflects the instantaneous offset amplitude, and the relative ratio reflects whether the deviation exhibits a proportional amplification characteristic. During the judgment and processing phase, the system introduces two empirical threshold parameters for joint constraint, such as the absolute tolerance difference. Set at The range is used to filter out small fluctuations caused by normal wave disturbances, while the relative multiple... Set as Between these parameters, anomalies exceeding the average structural proportion of the cross-section are identified. When a measuring point simultaneously meets both the "absolute deviation exceeding the limit and the relative multiple exceeding the limit", the measuring point will be directly marked as an anomalous flypoint, and its original observation value will be removed from the flow field sequence. Instead, the theoretical spatial expected value at the corresponding time will be used to backfill the value, thereby ensuring that the cross-sectional velocity distribution will not be structurally distorted due to local sudden increases.

[0031] Optionally, the method for constructing the physical drawdown zone state determination conditions in step S4 includes: Set the absolute installation elevation of each stratospheric transducer to assess the wind and wave rise and water surface lingering effect, and determine the safe buffer zone distance of the system water level; In this embodiment, the absolute installation elevation of each transducer layer is made into a fixed spatial reference, such as Ei obtained by station calibration (e.g., layer heights of 1.2m, 2.8m, 4.5m, etc.). At the same time, a safe buffer zone distance B is estimated by combining the short-term rise in water level in historical wind and wave records (e.g., maximum rise of 0.25m, wave period of 2-5s). It is generally taken as 0.2-0.4m to cover the impact of water surface lingering and short-term wave crests. This step is equivalent to pre-defining the physical boundary of whether the equipment may be exposed above the water surface.

[0032] Real-time water level data collected synchronously is extracted from synchronous monitoring data, and the real-time water level data, absolute installation elevation, and safety buffer zone distance of the system water level are superimposed to construct the physical drawdown zone status determination conditions.

[0033] In a further embodiment, real-time water level data output by the synchronous water level gauge is continuously read (e.g., updated at 1 Hz or 5 Hz), and this water level value is compared with the absolute installation elevation of the corresponding flow layer. Perform layer-by-layer matching, and then superimpose the previously obtained safety buffer band distance. This forms a dynamic judgment range. If the real-time water level is lower than " If the water level is above a certain threshold, the troposphere is considered to have entered a drawdown or is close to being exposed; if it is above this threshold, it is considered to still be under stable underwater observation conditions. In this way, water level, structure height, and wind and wave corrections are uniformly mapped to the same judgment scale, providing a basic constraint for subsequent differentiation between physical misses and signal anomalies.

[0034] Optionally, separating the normal exposure zero value from the underwater signal failure zero value in step S4 includes: Extract the set of measured points from real-time flow measurement data whose measured values ​​are strictly equivalent to 0.0 or marked as missing. In this embodiment, the real-time flow measurement data is first cleaned, and the data with a measured value of 0.0 or marked as NaN are extracted to form a set to be judged, i.e. the set of measured points. At the same time, the corresponding measurement points, flow layer numbers and timestamps are retained for subsequent physical state backtracking.

[0035] When there is no flow velocity at any of the measured points in the set of measured points and the following conditions are met: When the flow velocity phenomenon at the measuring point is determined to be caused by the physical removal of the transducer from the water surface due to the receding water in the environment, it is marked and retained as normal off-target data. At the same time, the zero value corresponding to the measuring point is classified as normal exposure off-target zero value. In a further embodiment, the water level gauge output at each velocity-free measuring point in the set of measured points is simultaneously read at that moment. and the absolute installation elevation of the corresponding flow layer. Compare the samples and then add a pre-set safety buffer. (Generally take) (used to cover short-term wave rise error), if the following conditions are met If the point is considered to be in a state of probe water discharge caused by water level drop or receding, it is only marked and retained, does not participate in subsequent reconstruction calculations, and is included in the normal exposure zero value sequence.

[0036] When there is no flow velocity at any of the measured points in the set of measured points, but the following condition is met: When the phenomenon of no flow velocity at the measuring point is determined to be caused by a surge in sediment content or strong turbulence leading to the attenuation and loss of underwater signal, the underwater abnormal zero data is marked and included in the sequence to be reconstructed. At the same time, the zero value corresponding to the measuring point is classified as an abnormal zero value of underwater signal failure. In the formula, For real-time water level data, For the first Absolute installation elevation of the tropospheric transducer This is the safety buffer zone distance for the system water level.

[0037] In another embodiment, if the opposite is satisfied If the flow layer is theoretically still underwater, it indicates that there is zero flow velocity in reality. This situation is usually related to a sudden increase in sediment content or signal attenuation caused by turbulence. It is marked as an abnormal drop in the underwater signal and placed in the queue to be reconstructed. At the same time, its flow layer index and timestamp are retained for subsequent backfilling based on the spatial coupling model.

[0038] Optionally, constructing the cross-coupling prediction matrix in step S5 includes: After removing the sets of data marked as normal off-target data and underwater abnormal zero-drop data from the real-time flow measurement data, the remaining dataset is extracted as the full profile active effective dataset. In this embodiment, data points that have been marked as normal zero values ​​and abnormal zero values ​​underwater are first removed from the real-time flow measurement data. Only the flow velocity observations that are still continuously valid are retained. They are then aligned and recombined at the same time to form a full-profile active valid dataset that contains only the underwater valid measurement status. This type of data generally requires a continuous sampling rate of not less than 95%. Otherwise, the data at that time will be downweighted or delayed in participating in the modeling.

[0039] Using a pre-defined Huber regression model without intercept, cross-fitting is performed on any pair of active flow layers in the full-profile active effective dataset to construct a cross-coupling prediction matrix, where any element in the cross-coupling prediction matrix... The representative stated that after eliminating the abnormal contamination, the first Effective flow velocity of the troposphere on the first Robust extrapolation of the target flow velocity in the troposphere with proportional weights.

[0040] In a further embodiment, each of the remaining tropospheric combinations is paired up, for example, the first... Layer as input, the first The layer is used as output to construct the mapping relationship between the flow layers. Here, a Huber regression structure without intercept is used, meaning that no constant offset term is introduced; it only relies on the proportional relationship between the flow layers to fit the data, avoiding overall drift. The Huber threshold is usually set at... This is used to distinguish between normal flow fluctuations and the effects of residual anomalies. The calculations obtained in this way... Essentially, this means that after removing abnormal contamination, the first... Laminar flow velocity on the first The stability of laminar flow velocity affects the intensity, all When combined, they form an N×N cross-coupling prediction matrix, which is used to describe the spatial transmission relationship between flow layers throughout the entire cross section.

[0041] Optionally, performing multidimensional targeted joint reconstruction in step S5 includes: Based on the cross-coupling prediction matrix, obtain the individual independent prediction values ​​of the target layer to be reconstructed in each active flow layer; In this embodiment, the integer row coefficients corresponding to the target layer to be reconstructed are first extracted from the cross-coupling prediction matrix, and then multiplied item by item with the current velocity value of each active flow layer to obtain the individual prediction result of each reference flow layer for the target flow layer. This step is performed in real time with a time step of 1 Hz, and the input velocity is processed. The slight smoothing avoids amplifying instantaneous jitter at the prediction end.

[0042] It is worth noting that the target layer to be reconstructed refers to the flow layer in which, at the same test time, there are abnormal measurement points or abnormal flying points with abnormally large spatial proportions whose corresponding data are marked as the sequence to be reconstructed.

[0043] Aggregate the individual independent prediction values ​​of all non-target layers to be reconstructed, call the preset median voting mechanism to perform unified calculation on the aggregation results, and thus derive the joint reconstruction expected value of the target layers to be reconstructed; In a further embodiment, during the result fusion stage, instead of directly averaging, all individual prediction values ​​from different reference flow layers are gathered into the same set, and then robust fusion is performed using the median method. That is, the median value after sorting is selected as the target flow layer output at the current moment. This can avoid individual flow layer outliers from causing deviations in the results, and the effect is more stable, especially when the number of reference flow layers is ≥3. This is equivalent to performing an anomaly-resistant voting screening on the multi-source prediction results.

[0044] The joint reconstruction expected value is used to target and replace the measurement points corresponding to the abnormal flying points and the abnormal zero value of the underwater signal failure, thereby obtaining the current measurement reconstruction data.

[0045] In a further embodiment, during the final replacement stage, the obtained joint reconstruction value is written back to the anomaly point location. If the point was previously identified as a fly point, the original mutation value is directly overwritten. If it belongs to the zero value of underwater signal failure, it is located and backfilled by combining its timestamp and flow layer number to ensure that the replacement occurs at the correct spatial-temporal index position, thereby forming a continuous cross-sectional velocity sequence and restoring the integrity of the overall hydrodynamic structure.

[0046] Of particular importance, targeted replacements include: Establish a mapping constraint relationship between the measurement points corresponding to abnormal flying points and abnormal zero values ​​of underwater signal failure and the expected value of joint reconstruction; In this embodiment, the identified abnormal flying points and the zero values ​​of underwater signal failure are first indexed and bound together. That is, each abnormal measurement point is combined and encoded with its own flow layer number and sampling timestamp, for example, forming an (i,t) structure. Then, a one-to-one correspondence is established with the joint reconstruction expected value calculated at the current time. This is equivalent to locking the "abnormal point location" and the "replaceable value source" into the same spatial and temporal coordinate system to avoid misalignment in subsequent backfilling.

[0047] Referring to the mapping constraint relationship, for the measurement points marked as anomalous flying points, the joint reconstruction expected value is used as a spatial consistency substitute for the flow layer to which the measurement point belongs and the corresponding timestamp, and a fixed value is backfilled. In a further embodiment, when repairing flying points, fixed-point replacement is performed directly according to the flow layer dimension. For example, if a certain flow layer experiences a spatial abrupt increase anomaly at t=120s, the original observation value is no longer used. Instead, the expected value of joint reconstruction at the corresponding time is used to cover it. This value comes from the output result after the fusion of the cross-coupling matrix and the median. The update frequency is 1Hz, thereby achieving spatial consistency correction for local anomalous abrupt changes and keeping the cross-sectional profile a continuous and smooth structure.

[0048] Referring to the mapping constraint relationship, for the measurement point marked as the zero value of the underwater signal failure, the transducer installation elevation and real-time water level data of the flow layer where the measurement point is located are combined to determine whether the measurement point is within the range of the effective underwater measurement conditions. If it is within the range of effective underwater measurement conditions, the joint reconstruction expected value will be replaced as the physical consistency recovery value. If the measurement point is not within the effective underwater measurement range, it will be marked as non-reconfigurable and the original observation value will be maintained.

[0049] In a further embodiment, when processing the abnormal zero value of underwater signal failure, an additional physical condition verification mechanism is introduced. This involves comparing the transducer installation elevation at the measuring point with the real-time water level, and using the safety buffer zone B (0.2–0.4 m) to determine whether it is still within the effective underwater observation range. If the effective underwater conditions are met, the zero value is considered to be a signal anomaly rather than a physical miss, and the joint reconstruction value is used to replace and restore it. If it is determined that the water level is exposed or receding, numerical reconstruction is not performed, only the original zero value or missing marker is retained and marked as unreconstructable, in order to avoid incorrect correction of the actual physical receding process.

[0050] Optionally, step S5, in the multidimensional targeted joint reconstruction process, also includes: When the number of valid reference layers is greater than 2, a preset median voting mechanism is invoked to uniformly calculate the aggregation results; the form of the median voting mechanism is as follows: ; In the formula, For the target layer to be reconstructed In time The expected value of joint reconstruction , For active flow layer For the target layer to be reconstructed Individual independent predicted values, For active flow layer Effective flow rate for the target layer to be reconstructed The robust derivation of proportions and weights, For active flow layer The measured flow velocity, For active flow layer The corresponding set of measured points; When the number of effective reference flow layers is less than or equal to 2, the weighted average operator is adaptively switched to replace the median voting mechanism. That is, the weighted average of the robust inference ratio weights of each effective reference flow layer is used as the joint reconstruction expectation.

[0051] In this embodiment, the flow segmentation is first determined based on the number of valid reference flow layers selected at the current time. For example, the number of normal flow layers M after removing zero values, missing values, and flypoints is counted, and each reference flow layer is read synchronously at a time step of 1Hz. At the same time, combined with the corresponding scaling factor in the cross-coupling matrix For each reference flow layer, first calculate its target layer to be reconstructed. Single-unit predicted value This involves spatially mapping and transforming the flow velocities of each layer according to a preset proportional relationship, making data from different flow layers comparable at the same scale. When M is greater than 2, averaging is not performed directly; instead, all... Sort by numerical value and then take the median value as the joint reconstruction output at the current moment. This median value is derived from the median voting mechanism, which essentially uses sorting statistics to suppress interference from local abnormal reference flow layers. A base value constraint of 0.01 is added during the calculation to prevent the result from approaching zero in extreme cases, which could lead to instability in subsequent proportional calculations. When M is less than or equal to 2, the median sorting is no longer robust due to the limited number of available reference layers. In this case, a weighted average mode is switched to directly apply the median to each... according to Normalized weighted summation is performed, for example, the weights are controlled to fluctuate within the range of 1 ± 0.05, so as to obtain the joint reconstruction expected value, so as to ensure that the target flow layer estimation results can still be output under low sample conditions.

[0052] Of particular importance are the effective methods for screening reference flow layers, including: If the conditions for multidimensional targeted joint reconstruction are triggered at the current moment, then based on the real-time flow measurement data of the reference flow layer synchronously measured at that moment, the measurement point data with a measurement value strictly equal to 0.0 or marked as missing are stripped, and the reference flow layer where the stripped data is located is used as the candidate flow layer set. Based on the comparison between the absolute installation elevation of the transducer corresponding to each flow layer and the real-time water level data, it is determined that the flow layers in the candidate flow layer set whose real-time water level data is higher than the absolute installation elevation but does not exceed the safe buffer zone distance of the system water level meet the underwater effective measurement conditions. For flow layers that meet the conditions for effective underwater measurement, a signal integrity check is performed. Flow layers with persistent flying point anomalies or sudden changes from zero are eliminated, and the remaining flow layers are the effective reference flow layers.

[0053] In this embodiment, the original velocity sequences of each reference flow layer at the same time are first cleaned. For example, under the conditions of a sampling interval of 1 second and a sliding buffer window of 5 points, invalid measurement points with values ​​of 0.0 or marked as NaN are removed. After the measurement point removal is completed, if the reference flow layer still retains valid data sequences of time-continuous valid observation sequence segments, the entire reference flow layer is included in the candidate flow layer set for subsequent screening. Based on the candidate set, the transducer installation elevation is introduced. With real-time water level The system filters based on the comparison relationship, for example, by pre-configuring a safety buffer zone. ,when Between and If the current layer is in a stable submerged state and still possesses effective current measurement conditions, it is considered an effective underwater measurement layer and retained for the next stage. Subsequently, these effective underwater measurement layers undergo a continuity check, generally based on the stability assessment of the most recent 30-second data segment. If consecutive flying points occur (e.g., three consecutive sampling points deviate from the local median by more than 1.5 times the standard deviation) or frequently jump from non-zero to zero, the layer is considered to have signal distortion issues and is discarded. The remaining layers ultimately constitute the set of effective reference layers for the current moment.

[0054] Optionally, step S5 may include the following steps before performing multidimensional targeted joint reconstruction: Multidimensional targeted joint reconstruction is triggered if and only if the measurement points in the real-time flow measurement data of the target's vulnerable surface are marked as abnormal flying points with abnormally large spatial proportions, or the corresponding measurement values ​​are marked as underwater abnormal zero data. If the target's vulnerable surface is currently marked as being in a normal fluctuation state or a normal drawdown zone physical off-target exposure state, then multidimensional target joint reconstruction will not be activated, and only the original measured value or normal zero value background color will be retained.

[0055] In this embodiment, the velocity sequence of the target's vulnerable surface layer at the current moment is rapidly labeled. For example, using a sampling frequency of 1 second and the most recent 10 time points as the judgment window, if some measuring points are marked as abnormal flying points with abnormal spatial proportions, or if the corresponding measured values ​​are marked as underwater abnormal zero-drop data, these measuring points will be identified as abnormal flying points or underwater abnormal zero-drop data. When abnormal markings of abnormal flying points or underwater abnormal zero-drop data appear, the system considers that the target's vulnerable surface layer has deviated from the normal hydrodynamic distribution structure, thus entering the multi-dimensional targeted joint reconstruction process and calling the subsequent cross-coupling matrix for correction calculation. Conversely, if the measuring points of the surface layer only show small fluctuations, such as fluctuating within 0.2 m / s around the local median, and combined with the water level... With installation elevation It is determined that the area is in a normal drawdown zone exposure state (e.g.) If there are no abnormal jumps, it will be retained as the original observation state without triggering any reconstruction calculations. Only the original flow velocity or natural zero value output will be maintained to ensure the consistency of the underlying record of the real hydrological process.

[0056] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0057] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for identifying and reconstructing multilayer velocity anomalies based on physical constraints and robust regression, characterized in that, Includes the following steps: Step S1: Obtain synchronous monitoring data of the target hydrological monitoring section, including real-time flow measurement data of each flow layer and real-time water level data; Step S2: Based on the real-time flow measurement data of each flow layer at the same test time, construct a single-layer flow velocity pure spatial verification model, and use the Huber robust regression method to map and fit the real-time flow measurement data of the underwater stable flow layer to deduce the theoretical spatial expected value of the target surface layer that is susceptible to disturbance. Step S3: Compare the measured flow velocity of the target's easily disturbed surface with the theoretical spatial expected value, determine the abrupt change characteristics of the surface flow velocity through the dual threshold joint judgment logic, and identify and mark abnormal flying points that deviate from the cross-sectional hydrodynamic distribution law. Step S4: Based on the real-time water level data, the absolute installation elevation of the transducers corresponding to each flow layer, and the physical drawdown zone status judgment conditions constructed by superimposing the preset water level safety buffer zone distance, perform physical status judgment on the zero or null value sequence in the real-time flow measurement data, and separate the normal exposed off-target zero value from the abnormal zero value of underwater signal failure. Step S5: Extract the full profile active effective dataset from the synchronous monitoring data after removing normal exposure miss zero values ​​and underwater signal failure anomaly zero values, and construct a cross-coupled prediction matrix; based on the cross-coupled prediction matrix, use the median voting mechanism to perform multi-dimensional targeted joint reconstruction on the measurement points corresponding to the abnormal flying points and underwater signal failure anomaly zero values, thereby outputting the current measurement reconstruction data.

2. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 1, characterized in that, Step S1 involves constructing a single-layer velocity pure space verification model, including: The measured velocity data of several flow layers at the same test time are extracted from the real-time flow measurement data. Based on the measured velocity data of the cross sections, the flow layer with the most severe surface turbulence interference is selected as the target prediction layer. Combined with the measured velocity data of the reference flow layers in other flow layers whose underwater anti-interference ability is higher than the preset anti-interference ability threshold, the theoretical spatial expectation derivation function of the target prediction layer is constructed. In training the theoretical spatial expectation extrapolation function using measured velocity data from the reference flow layer, the loss function during the training process is set to the Huber robust regression loss function, which includes the absolute error norm and the squared error norm. The intercept term is forced to be zero to satisfy the physical boundary condition that the still water velocity is reduced to zero, thereby constructing a single-layer velocity pure spatial verification model.

3. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 2, characterized in that, The calculation logic of the Huber robust regression loss function includes: Calculate the absolute value of the residual between the measured value and the predicted value output by the theoretical spatial expectation extrapolation function in the measured flow velocity data; When the absolute value of the residual is less than or equal to the preset robustness threshold parameter, the squared loss function is used: ; When the absolute value of the residual is greater than the robustness threshold parameter, a linear penalty loss function is used: ; In the formula It is the absolute value of the residual between the measured value and the predicted value. This is a robust threshold parameter.

4. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 1, characterized in that, Step S3, which uses a dual-threshold joint judgment logic to determine the abrupt change characteristics of surface flow velocity, includes: The joint logic is satisfied if and only if the absolute difference between the measured flow velocity and the theoretical spatial expected value of the target's vulnerable surface is greater than a preset absolute tolerance difference threshold, and the measured flow velocity of the target's vulnerable surface is greater than the theoretical spatial expected value multiplied by a preset relative multiple threshold. ； The measurement points on the target's easily disturbed surface are marked as abnormal flying points with abnormally large spatial proportions, and reconstructed and replaced using theoretical spatial expected values; In the formula, To measure the flow rate, This is the expected value in the theoretical space. This is the absolute tolerance threshold. The relative multiple threshold, This is an indicator for identifying abnormal flying points.

5. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 1, characterized in that, The method for constructing the physical drawdown zone state determination conditions in step S4 includes: Set the absolute installation elevation of each stratospheric transducer to assess the wind and wave rise and water surface lingering effect, and determine the safe buffer zone distance of the system water level; Real-time water level data collected synchronously is extracted from synchronous monitoring data, and the real-time water level data, absolute installation elevation, and safety buffer zone distance of the system water level are superimposed to construct the physical drawdown zone status determination conditions.

6. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 5, characterized in that, Step S4 involves separating the normal exposed zero value from the abnormal zero value of underwater signal failure, including: Extract the set of measured points from real-time flow measurement data whose measured values ​​are strictly equivalent to 0.0 or marked as missing. When there is no flow velocity at any of the measured points in the set of measured points and the following conditions are met: When the flow velocity phenomenon at the measuring point is determined to be caused by the physical removal of the transducer from the water surface due to the receding water in the environment, it is marked and retained as normal off-target data. At the same time, the zero value corresponding to the measuring point is classified as normal exposure off-target zero value. When there is no flow velocity at any of the measured points in the set of measured points, but the following condition is met: When the phenomenon of no flow velocity at the measuring point is determined to be caused by a surge in sediment content or strong turbulence leading to the attenuation and loss of underwater signal, the underwater abnormal zero data is marked and included in the sequence to be reconstructed. At the same time, the zero value corresponding to the measuring point is classified as an abnormal zero value of underwater signal failure. In the formula, For real-time water level data, For the first Absolute installation elevation of the tropospheric transducer This is the safety buffer zone distance for the system water level.

7. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 1, characterized in that, Step S5 involves constructing the cross-coupling prediction matrix, which includes: After removing the sets of data marked as normal off-target data and underwater abnormal zero-drop data from the real-time flow measurement data, the remaining dataset is extracted as the full profile active effective dataset. Using a pre-defined Huber regression model without intercept, cross-fitting is performed on any pair of active flow layers in the full-profile active effective dataset to construct a cross-coupling prediction matrix, where any element in the cross-coupling prediction matrix... The representative stated that after eliminating the abnormal contamination, the first Effective flow velocity in the troposphere for the first Robust extrapolation of the target flow velocity in the troposphere with proportional weights.

8. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 1, characterized in that, Step S5 involves performing multidimensional targeted joint reconstruction, including: Based on the cross-coupling prediction matrix, obtain the individual independent prediction values ​​of the target layer to be reconstructed in each active flow layer; Aggregate the individual independent prediction values ​​of all non-target layers to be reconstructed, call the preset median voting mechanism to perform unified calculation on the aggregation results, and thus derive the joint reconstruction expected value of the target layers to be reconstructed; The joint reconstruction expected value is used to target and replace the measurement points corresponding to the abnormal flying points and the abnormal zero value of the underwater signal failure, thereby obtaining the current measurement reconstruction data.

9. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 8, characterized in that, Step S5, the multidimensional targeted joint reconstruction process, also includes: When the number of valid reference layers is greater than 2, a preset median voting mechanism is invoked to uniformly calculate the aggregation results; the form of the median voting mechanism is as follows: ； In the formula, For the target layer to be reconstructed In time The expected value of joint reconstruction , For active flow layer For the target layer to be reconstructed Individual independent predicted values, For active flow layer Effective flow rate for the target layer to be reconstructed The robust derivation of proportions and weights, For active flow layer The measured flow velocity, For active flow layer The corresponding set of measured points; When the number of effective reference flow layers is less than or equal to 2, the weighted average operator is adaptively switched to replace the median voting mechanism. That is, the weighted average of the robust inference ratio weights of each effective reference flow layer is used as the joint reconstruction expectation.

10. The multilayer velocity anomaly identification and reconstruction method based on physical constraints and robust regression according to claim 1, characterized in that, Step S5, prior to performing multidimensional targeted joint reconstruction, also includes: Multidimensional targeted joint reconstruction is triggered if and only if the measurement points in the real-time flow measurement data of the target's vulnerable surface are marked as abnormal flying points with abnormally large spatial proportions, or the corresponding measurement values ​​are marked as underwater abnormal zero data. If the target's vulnerable surface is currently marked as being in a normal fluctuation state or a normal drawdown zone physical off-target exposure state, then multidimensional target joint reconstruction will not be activated, and only the original measured value or normal zero value background color will be retained.

Images