Open channel three-dimensional flow field construction method and device of sensing information fusion system, and medium

Abstract

The invention discloses an open channel three-dimensional flow field construction method and device of a sensing information fusion system, and a medium, and relates to the technical field of open channel three-dimensional flow field construction. The method comprises the following steps: acquiring multi-source measurement data of a water cross section, and performing temperature compensation processing on measurement results of a radar sensor and an ultrasonic transducer array based on temperature to acquire a surface point flow velocity, an underwater layered line flow velocity, a lower boundary of the water cross section and a surface two-dimensional flow field; and performing fusion calibration through a Kalman filtering model to generate an optimized surface flow field. An underwater layered flow field is generated by extracting and optimizing a transverse distribution mode of a surface flow field and applying the transverse distribution mode to each underwater layer. And based on the optimized surface flow field and the underwater layered flow field, reconstructing a three-dimensional flow field of the open channel by adopting a Gaussian process regression model taking a preset fluid mechanics model as a prior mean value function.

Description

Technical Field

[0001] This invention relates to the field of open channel three-dimensional flow field construction technology, and more specifically, to a method, apparatus, and medium for constructing an open channel three-dimensional flow field using a sensor information fusion system. Background Technology

[0002] In the field of open channel flow monitoring, the flow rate is generally calculated by measuring the velocity distribution (i.e., three-dimensional flow field) at a certain cross-section and combining it with the cross-sectional area of ​​the water passage. Currently, common technical methods based on this principle include using radar current meters to measure surface point velocity, using array-type ultrasonic time-of-flight methods to obtain vertical velocity, and using video image analysis to obtain the two-dimensional surface flow field.

[0003] However, all of the aforementioned existing technologies suffer from a lack of comprehensive information dimension, making it difficult to fully reflect the actual flow field characteristics. Point velocity methods can only acquire local information and cannot reflect the lateral distribution differences of surface velocity. Line velocity methods lack the ability to capture changes in horizontal velocity. While surface velocity methods can cover the surface flow field, their inference of underwater velocity relies on empirical models, leading to significant errors under complex flow conditions. Furthermore, various sensors are susceptible to environmental interference and lack cross-validation mechanisms between multi-source data, resulting in insufficient system anti-interference capabilities and unreliable reliability.

[0004] On the other hand, existing methods are usually based on the assumption of steady flow or uniform flow, which makes it difficult to adapt to the complex working conditions such as unsteady flow and non-uniform flow that are common in actual open channels. Summary of the Invention

[0005] This invention provides a method, apparatus, and medium for constructing a three-dimensional flow field in an open channel using a sensor information fusion system, in order to improve at least one of the aforementioned technical problems.

[0006] In a first aspect, the present invention provides a method for constructing a three-dimensional flow field in an open channel using a sensor information fusion system, comprising steps S1 to S4.

[0007] S1. Acquire multi-source measurement data characterizing the open channel flow velocity and boundary conditions collected by radar sensors, image sensors, first ultrasonic transducer array, second ultrasonic transducer array, and temperature sensor group deployed on the same cross-section. Based on the water temperature and air temperature collected by the temperature sensor group, perform temperature compensation processing on the measurement results of radar sensors and ultrasonic transducer array to obtain the compensated surface point velocity, underwater stratification line velocity, lower boundary of the cross-section, and surface two-dimensional flow field.

[0008] S2. Based on the surface point velocity and the surface two-dimensional flow field, a fusion calibration is performed using a Kalman filter model to generate an optimized surface flow field.

[0009] S3. Based on the optimized surface flow field and the underwater stratified linear velocity, the lateral distribution pattern of the optimized surface flow field is extracted and applied to each underwater stratum to generate an underwater stratified flow field.

[0010] S4. Based on the optimized surface flow field and the underwater stratified flow field, a Gaussian process regression model with a preset fluid dynamics model as the prior mean function is used to reconstruct the three-dimensional flow field of the open channel.

[0011] Secondly, the present invention provides a three-dimensional flow field construction device for open channels in a sensor information fusion system, which includes a preprocessing module, a point-surface module, a surface-line module, a layering module, and a calculation module.

[0012] The preprocessing module is used to acquire multi-source measurement data characterizing the open channel flow velocity and boundary conditions collected by radar sensors, image sensors, a first ultrasonic transducer array, a second ultrasonic transducer array, and a temperature sensor group deployed on the same cross-section. Based on the water temperature and air temperature collected by the temperature sensor group, the module performs temperature compensation processing on the measurement results of the radar sensors and ultrasonic transducer array to obtain the compensated surface point velocity, underwater stratification line velocity, lower boundary of the cross-section, and surface two-dimensional flow field.

[0013] The point-to-surface module is used to generate an optimized surface flow field by performing fusion calibration based on the surface point flow velocity and the surface two-dimensional flow field through a Kalman filter model.

[0014] The surface line module is used to generate an underwater stratified flow field by extracting the lateral distribution pattern of the optimized surface flow field and applying it to each underwater layer, based on the optimized surface flow field and the underwater stratified line velocity.

[0015] The layered module is used to reconstruct the three-dimensional flow field of the open channel based on the optimized surface flow field and the underwater layered flow field, using a Gaussian process regression model with a preset fluid dynamics model as the prior mean function.

[0016] Thirdly, the present invention provides a multi-sensor information fusion system, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a method for constructing a three-dimensional flow field in an open channel as described in any paragraph of the first aspect.

[0017] Fourthly, the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform a method for constructing a three-dimensional flow field in an open channel as described in any paragraph of the first aspect.

[0018] By adopting the above technical solution, the present invention can achieve the following technical effects: This invention fuses multi-dimensional information from points, lines, and surfaces at the feature layer and reconstructs the three-dimensional flow field using a physical model as a strong constraint. This represents a paradigm shift from inference to reconstruction, fundamentally solving the accuracy bottleneck caused by the single dimension of information, and making the three-dimensional flow field more accurate. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the specific embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some specific embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the component layout of a multi-sensor information fusion system.

[0021] Figure 2 This is a flowchart illustrating the method for constructing a three-dimensional flow field in an open channel.

[0022] The diagram is labeled as follows: 101-Radar sensor, 102-Image sensor, 103-First ultrasonic transducer array, 106-Second ultrasonic transducer array, 107-Temperature sensor group, 104-Data fusion processing unit. Detailed Implementation

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention.

[0024] Example 1, please refer to Figures 1 to 2 The first embodiment of the present invention provides a method for constructing a three-dimensional flow field in an open channel using a sensor information fusion system.

[0025] like Figure 1 and Figure 2 As shown, the multi-sensor information fusion system includes a radar (i.e., radar sensor 101), a camera (i.e., image sensor 102), a first ultrasonic transducer array 103, a second ultrasonic transducer array 106, an underwater temperature sensor, and an air temperature sensor, all mounted on the same open channel cross-section. The radar sensor 101 measures surface point velocity and water level. The image sensor 102 measures the surface flow field. The first ultrasonic transducer array 103 measures underwater stratified linear velocity. The second ultrasonic transducer array 106 measures silt height. The underwater temperature sensor and the air temperature sensor constitute a temperature sensor group 107.

[0026] In this embodiment, radar sensor 101 employs a frequency-modulated continuous wave (FMCW) radar to achieve velocity and distance measurement functions. The velocity measurement function accurately calculates the flow velocity at the water surface directly below the radar by analyzing the phase change (Doppler effect) between the continuously transmitted signal and the echo, providing a high-precision "point velocity" reference for the system. The distance measurement function accurately calculates the distance between the radar and the water surface by measuring the frequency difference between the transmitted signal and the echo, used to obtain water level information.

[0027] In this embodiment, the image sensor 102 uses an industrial-grade camera, deployed at a location that overlooks the entire cross-section of the channel. The function of the image sensor 102 is to continuously acquire multiple frames of images of the water surface and process them using a built-in artificial intelligence-based pure vision flow measurement algorithm (which integrates one or more cutting-edge technologies such as STIV, PIV, LSPIV, OP, and PTV) to obtain a high-resolution two-dimensional surface flow field covering the entire water surface.

[0028] The first ultrasonic transducer array 103 in this embodiment employs a multi-layer distributed ultrasonic transducer array, installed on the channel bottom or sidewall, for measuring underwater flow velocity. The function of the first ultrasonic transducer array 103 is to calculate the water flow velocity along its channel path by accurately measuring the time difference between the downstream and upstream propagation of ultrasonic waves using the ultrasonic time-of-flight method. By deploying multiple channels at different depths, stratified "linear flow velocities" at multiple discrete depths underwater can be collected. Preferably, the first ultrasonic transducer array sets up a pair of ultrasonic transducers at four different depths in the open channel to detect stratified linear flow velocities at different depths.

[0029] The second ultrasonic transducer array 106 in this embodiment employs another set of ultrasonic sensors, or one integrated with the first ultrasonic transducer array 103, to measure the true boundary of the channel. The function of the second ultrasonic transducer array 106 is to accurately calculate the height of the top surface of the silt from the sensor by measuring the time it takes for the emitted ultrasonic pulse to return to the sensor after reaching the channel bottom or silt interface, based on the ultrasonic echo ranging method and combined with the speed of sound propagation in water. This data is used to dynamically determine the lower boundary of the true water-passing cross-section to calculate the accurate water-passing cross-sectional area. Specifically, the second ultrasonic transducer array includes at least one pair of ultrasonic transducers positioned on both sides of the water-passing cross-section and facing the channel bottom.

[0030] The temperature sensor group 107 includes an underwater temperature sensor and an air temperature sensor. The underwater temperature sensor is configured to measure the water temperature and is used to correct the propagation speed of ultrasonic waves in water in real time, which is crucial for ensuring the accuracy of ultrasonic velocity and ranging. The air temperature sensor is configured to measure the air temperature and is used to compensate for the ranging function of the radar sensor 101 to obtain a more accurate water level reading.

[0031] The data fusion processing unit 104, acting as the "brain" of the system, receives raw data or pre-processed feature data from all the aforementioned sensors and processes it to obtain the three-dimensional flow field information of the cross-section. It is understood that the data fusion processing unit or the open channel three-dimensional flow field construction device can be an electronic device with computing power, such as an industrial control computer, a portable laptop computer, a desktop computer, a server, a smartphone, or a tablet computer.

[0032] The method for constructing a three-dimensional flow field in an open channel using a sensor information fusion system can be executed by an open channel three-dimensional flow field construction device within the sensor information fusion system, or by a data fusion processing unit within a multi-sensor information fusion system (hereinafter referred to as: open channel three-dimensional flow field construction device). Specifically, it is executed by one or more processors within the open channel three-dimensional flow field construction device to implement steps S1 to S4.

[0033] S1. Acquire multi-source measurement data characterizing the open channel flow velocity and boundary conditions collected by radar sensors, image sensors, first ultrasonic transducer array, second ultrasonic transducer array, and temperature sensor group deployed on the same cross-section. Based on the water temperature and air temperature collected by the temperature sensor group, perform temperature compensation processing on the measurement results of radar sensors and ultrasonic transducer array to obtain the compensated surface point velocity, underwater stratification line velocity, lower boundary of the cross-section, and surface two-dimensional flow field.

[0034] Acquiring multi-source measurement data specifically includes: The radar sensor is a frequency-modulated continuous wave radar, which is deployed above the cross-section of the open channel to measure the surface point velocity of the water body based on the Doppler effect, and to measure the distance from the radar to the water surface based on the ranging function to obtain the water level.

[0035] An image sensor is positioned above the cross-section of the water body to continuously acquire image sequences of the water surface and generate a two-dimensional flow field covering the cross-section of the water body using one or more image flow measurement algorithms based on STIV, PIV, LSPIV, OP, and PTV.

[0036] The first ultrasonic transducer array is deployed at multiple different depths on the channel bottom and / or sidewalls to measure the underwater stratification velocity at multiple depths based on the ultrasonic time-of-flight method.

[0037] The second ultrasonic transducer array is arranged vertically towards the bottom of the channel to obtain the height of the top surface of the silt through ultrasonic echo ranging, so as to determine the lower boundary of the water passage section.

[0038] The temperature sensor group includes an underwater temperature sensor and an air temperature sensor. The underwater temperature sensor is used to perform temperature compensation on the measurement results of the first ultrasonic transducer array and the second ultrasonic transducer array, and the air temperature sensor is used to perform temperature compensation on the measurement results of the radar sensor.

[0039] like Figure 2 As shown, the data fusion processing unit first performs velocity and distance compensation on the measurement data of the radar sensor, the first ultrasonic transducer array, and the second ultrasonic transducer array based on the real-time air and water temperatures collected by the temperature sensor group, respectively, to eliminate measurement errors caused by temperature changes, and obtains the following results: 1. At a fixed point Measuring high-precision, high-frequency flow velocities over time series. .

[0040] 2. Precise linear flow rates at multiple different depths, after temperature compensation.

[0041] 3. The lower boundary of the water-passing section.

[0042] Furthermore, data collected by an image sensor is used to obtain a global, high spatial resolution two-dimensional surface flow field. .

[0043] Specifically, the acquisition of surface point velocity, underwater stratification line velocity, lower boundary, and surface two-dimensional flow field are all existing technologies, and will not be elaborated upon in this invention. For example, based on data collected by an image sensor, a global, high spatial resolution two-dimensional surface flow field can be obtained using existing surface velocity methods, such as LSPIV technology. Alternatively, based on data collected by radar sensors, the point velocity method using existing technology, based on the Doppler effect, can be used at a fixed point. Measuring high-precision, high-frequency flow velocities over time series. .

[0044] S2. Based on the surface point velocity and the surface two-dimensional flow field, a fusion calibration is performed using a Kalman filter model to generate an optimized surface flow field.

[0045] This invention constructs a Kalman filter model that incorporates global flow field information and high-precision point observation information into a unified state space framework. Preferably, step S2 specifically includes steps S21 to S25.

[0046] S21. Define a state vector, the state vector... For a moment The global scale factor of the image-measured flow field relative to the real flow field Additive bias .Right now This indicates transpose. For time indexing.

[0047] Specifically, the image sensor at any time The key feature of the global surface flow field is defined as the system's "state". The state vector is: The global scaling factor is The global scale factor (core calibration parameter) of the flow field measured in the time-lapse image relative to the true flow field. The additive bias is... The additive bias of the flow field measured by the time image (to compensate for non-multiplicative systematic errors, which can be simplified to a zero vector).

[0048] S22. Establish the state transition equation: ,in The current state vector, State transition matrix, The state vector of the previous time step, The process noise follows a zero-mean Gaussian distribution.

[0049] Based on the optimal estimate of the previous time step To obtain the prior predicted state at the current moment. and prediction error covariance .in The error covariance of the previous time step, Let be the covariance matrix of the process noise.

[0050] Specifically, assuming that the scaling factor and bias are stable or exhibit a slow random walk over a short period of time, the state transition model is as follows: The state transition matrix simplifies to the identity matrix. Assume the state is constant. The process noise follows a zero-mean Gaussian distribution. ;in For process noise, The distribution is a zero-mean Gaussian distribution. Indicates natural distribution.

[0051] Then, based on the optimal estimate from the previous moment... ,get Time-ahead prediction.

[0052] Predicted status: .

[0053] Prediction error covariance: .

[0054] The prior predicted state is the predicted value of the current scale factor and bias.

[0055] S23. Establish the observation equation: .in For radar sensors at any time The observed values ​​(i.e., surface point velocity) ), For the observation matrix, The observation noise follows a zero-mean Gaussian distribution.

[0056] The definition of the observation model (update phase) is as follows: using the high-precision point flow velocity of the radar sensor as the "observation value", the "predicted state" is corrected.

[0057] Radar fixed to At this coordinate point, there are two flow velocities: the image sensor flow velocity. and radar sensor flow rate The flow velocity from the radar sensor is used as a truth reference.

[0058] The two satisfy the following relationship: .

[0059] Standardized to observation equations: .

[0060] in, .

[0061] The observation noise follows a zero-mean Gaussian distribution. The covariance matrix of the observed noise can be determined through radar technical manuals or calibration experiments. Let be the observation matrix, representing the mapping of the state vector to the observation space. It is an identity matrix with dimensions consistent with the bias.

[0062] S24. Calculate the Kalman gain. Update state estimation and update error covariance .in To measure the error covariance of the observed noise, It is an identity matrix. Includes optimal scaling factor With optimal bias .

[0063] State update (fusion) specifically involves combining observations. Calculate the residual and pass it through Kalman gain. By correcting the prior prediction, we obtain the posterior optimal state estimate. .

[0064] First, calculate the Kalman gain: The role of Kalman gain is to balance the credibility of "predictions" and "observations"; minor Larger, more reliable in observation. minor They are young and believe more in predictions.

[0065] Then update the state estimate: Core formula: Final estimate = Predicted value + Gain × (Observed value - Predicted observed value), Output Optimal scaling factor at time With optimal bias .

[0066] Last updated error covariance: .

[0067] S25, Based on the optimal scaling factor With optimal bias Generate optimized surface flow field .

[0068] .

[0069] In the formula For a moment The original two-dimensional flow field on the surface.

[0070] Specifically, using the optimal scaling factor With optimal bias The entire image flow field is calibrated to obtain the optimized surface flow field: .

[0071] This step leverages the high-precision local "observations" of the radar to dynamically calibrate the global "predicted state" of the image sensor, achieving a complementary advantage. Radar accuracy is achieved through... It is applied to the entire surface flow field, improving the accuracy of the absolute velocity values ​​at all points. Filtering smooths short-term, drastic fluctuations in image flow measurement (such as those caused by sudden changes in illumination or water surface reflection), improving the stability and reliability of the flow field.

[0072] S3. Based on the optimized surface flow field and the underwater stratified linear velocity, the lateral distribution pattern of the optimized surface flow field is extracted and applied to each underwater stratum to generate an underwater stratified flow field.

[0073] Specifically, in step S1, the data collected by the ultrasonic transducer array is used to obtain precise "linear velocities" at multiple different depths after temperature compensation through the linear velocity method. This represents the average flow velocity along the ultrasonic channel path, and it is essentially a one-dimensional scalar. It possesses precise depth information and flow velocity values, but completely lacks information on the lateral (width) flow velocity distribution at that depth level. Directly using this one-dimensional linear flow velocity to represent the flow velocity across the entire two-dimensional horizontal plane is a typical example of the "generalization" of existing technology.

[0074] Step S3 aims to address the issue of insufficient "linear velocity" information. Its core idea is to utilize the optimized surface flow field generated in step S2, which already contains a precise lateral distribution pattern. As a "template" or "prior knowledge", the one-dimensional linear velocity is... Intelligently extended into a two-dimensional layered flow field .

[0075] Preferably, step S3 includes steps S31 to S32.

[0076] S31. Based on the optimized surface flow field, the transverse velocity distribution at the cross-section of the water passage is... Calculate the normalized lateral distribution pattern function : .

[0077] In the formula, To optimize the integral average value of the surface flow field along the ultrasonic channel path.

[0078] S32, for each depth measured by the first ultrasonic transducer array Underwater stratification velocity The two-dimensional layered flow field at this depth is generated using the lateral distribution mode function. : .

[0079] Specifically, inputting S1 yields the underwater stratification velocity at M depths. ,in, =1,2,...M, Here, M represents the depth layer index, and M represents the number of depth layers. The optimized surface flow field is obtained from input S2. It is a two-dimensional vector field that contains a precise lateral velocity distribution pattern.

[0080] First, from Extract the surface velocity distribution on the cross section of the ultrasonic array. Define a normalized transverse distribution mode function. It refers to the lateral velocity distribution, specifically optimizing the surface flow field in the ultrasonic cross-section and lateral position. The magnitude of the flow velocity at that location. It is the path integral average of the optimized surface flow field along the ultrasonic channel path L. This value represents "what value would be obtained if the surface flow field were measured using ultrasound". Therefore, it becomes a dimensionless function that describes the surface flow field at various points laterally. The relationship between the flow velocity and its "line average velocity".

[0081] In most open channel flows, although the flow velocities vary at different depths, the shape (pattern) of their lateral distribution is highly similar. This similarity stems from the uniform boundary condition that dominates the flow (friction between the channel walls). Therefore, using surface distribution patterns to guide the expansion of underwater distribution has a sound hydrodynamic basis.

[0082] Then, the mode function is applied to each underwater layer. This applies to each depth measured by the ultrasonic array. Assuming its lateral velocity distribution It follows a similar pattern to the surface, but its "line-average flow velocity" is determined by ultrasonic measurements at that depth. Decide.

[0083] Therefore, the two-dimensional layered flow field at this depth is generated by the following formula.

[0084] .

[0085] in A "baseline" value for the flow rate at this depth level is provided (precisely measured by the ultrasonic array). This provides an indication of how the flow rate should be transverse based on this baseline value. "Shaping" in direction.

[0086] By analyzing all M depth levels ( , ,... Repeating this step yields a series of discrete underwater stratified flow fields with two-dimensional spatial distribution. .

[0087] The innovation of step S3 lies in its enhancement of information dimension. It doesn't create information out of thin air, but cleverly grafts the surface "spatial information" verified in S1 onto the underwater "depth information" in S2, thus expanding the one-dimensional "line" data into two-dimensional "surface" data. This step completely solves the fundamental problem of the "line velocity method" completely lacking lateral velocity information. Without increasing any underwater hardware costs, it generates high-value two-dimensional underwater flow field data with far more information than the original measurements, thus adding value to the data. This provides crucial input data with spatial distribution characteristics for the final three-dimensional reconstruction in step S4, rather than just a few isolated points or lines.

[0088] S4. Based on the optimized surface flow field and the underwater stratified flow field, a Gaussian process regression model with a preset fluid dynamics model as the prior mean function is used to reconstruct the three-dimensional flow field of the open channel.

[0089] Existing technologies, after obtaining discrete measurement data, typically employ simple linear interpolation or fit vertical velocity profiles based on empirical formulas (such as the logarithmic law). This inference method has two major drawbacks: 1. It cannot integrate lateral and longitudinal distribution information. 2. In sparse areas not covered by sensors, the inference results are highly unreliable, and no reliability assessment can be provided.

[0090] Step S4 aims to completely abandon the inference model based on limited information and shift to a more scientific and reliable reconstruction model. The goal is to fuse all discrete data points obtained in S1 and S2 to generate a continuous, complete, and high-fidelity three-dimensional flow field that conforms to the laws of fluid dynamics. .

[0091] It should be noted that the input data for step S4 includes: the optimized surface flow field from S1. A series of characteristic points in the data. M underwater stratified flow fields from S2. A series of feature points in the dataset. These feature points together constitute the dataset. ,in These are spatial coordinates. It is the flow velocity at that point. For feature point indexes.

[0092] The core idea of ​​the Gaussian process regression (GPR) model is that it assumes the flow velocity at any given set of points follows a joint Gaussian distribution. It utilizes a known dataset... To predict any unknown point flow rate The posterior distribution. In the formula It is a normal Gaussian distribution. This indicates a normal distribution. That is, the predicted result follows a Gaussian distribution, including the predicted mean. (Most likely flow rate) and prediction variance (Uncertainty of the prediction results).

[0093] Preferably, step S4 includes steps S41 to S43.

[0094] S41. Construct the prior mean function of the Gaussian process regression model based on the preset fluid dynamics model.

[0095] Specifically, a classical physical model describing the vertical velocity distribution in an open channel is used as the prior mean function of the GPR, thereby realizing the physical prior.

[0096] Preferably, the fluid dynamics model is a log-law or power-law flow model.

[0097] When the exponential law is used, the prior mean function is: .

[0098] The prior mean function using the exponential law is transformed as follows: .

[0099] In the formula, For fluid dynamics model (representing points) (prior mean vector function at the location) It is surface flow rate, It is the height from the open channel. Is it water depth? It is the roughness correlation coefficient. yes Surface velocity directly above the point.

[0100] The innovation and safeguard of this embodiment lies in constructing the fluid dynamics model as a prior mean function of a Gaussian process regression model. This embodiment does not use the default zero-mean function of GPR, but instead uses the aforementioned fluid dynamics model... As a priori mean, and further improvements were made to this fluid dynamics, using replace This allows for the acquisition of a more accurate three-dimensional flow field.

[0101] The improved prior mean function tells the GPR model that it should assume the water flow follows physical laws before seeing any actual data, giving the model an intuitive and more logical understanding. It also improves the model's predictive ability in sparse regions: in unknown areas far from any measured data points, the model's predictions will naturally revert to this physical prior, rather than reverting to an unreasonable zero value. This makes the reconstruction results more realistic and smooth globally.

[0102] S42. Construct the covariance function (i.e., kernel function) of the Gaussian process regression model.

[0103] .

[0104] In the formula For covariance, and Represent any two points in space, For natural exponential function, This is the length scale parameter.

[0105] Specifically, a kernel function defines any two points in space. and The correlation or similarity of flow velocities between them is considered. A squared exponential kernel (RBF kernel) is typically chosen: the length scale parameter represents the range of influence of the flow velocity, which can be learned from the data by maximizing the marginal likelihood function.

[0106] S43, Based on a sample set containing samples from the optimized surface flow field and the underwater stratified flow field. Perform Gaussian process regression prediction for any location within the open channel. The flow velocity is predicted to reconstruct the three-dimensional flow field of the open channel.

[0107] For any point to be predicted, its predicted mean and variance for: .

[0108] .

[0109] In the formula: For point Its own covariance, The covariance vector between the test point and all training points. For the covariance matrix between training data points, It is a vector of flow velocity observations at data points. It is the prior mean vector at the data points. For fluid dynamics model (representing points) (prior mean vector function at the location) It is the variance of observation noise. It is an identity matrix.

[0110] It should be noted that the predicted mean for all points Forming a continuous and complete three-dimensional flow field The predicted standard deviation of all points Form a corresponding three-dimensional uncertainty field This was used to quantify the reliability of the reconstruction results at each point.

[0111] The principle of Gaussian process regression prediction in this embodiment is: predict the mean. It is the sum of two parts: one part is the prior guess from the physical model (i.e., the fluid dynamics model). The other part involves correcting this conjecture based on measured data. The magnitude of the correction depends on the distance between the measured point and the known points (defined by the kernel function).

[0112] Deep Integration of Data-Driven and Physical Model-Driven Approaches: This embodiment innovatively combines the flexible nonparametric modeling capabilities of GPR with the deterministic physical laws of fluid mechanics for the construction of three-dimensional flow fields, achieving a deep integration of data-driven and physical model-driven approaches. Compared to existing technologies that can only provide a single predicted value, this invention improves upon the predicted value by providing a predicted distribution, which can give a complete probability distribution including uncertainty. This represents a revolutionary improvement in the understanding of the reliability of measurement results.

[0113] The reconstructed three-dimensional flow field, achieved through the above steps, not only fits all sensor data but also conforms to physical laws in data-sparse regions, exhibiting extremely high fidelity. Furthermore, it can clearly identify which regions have highly reliable velocity estimates (close to the sensor) and which regions have a greater element of speculation (far from the sensor), providing a basis for decision-making.

[0114] Through the detailed explanations of S1 to S4 above, a complete process is presented, demonstrating how a three-dimensional flow field, closest to physical reality, is reconstructed by using a progressively rigorous, logically sound, and physically meaningful fusion algorithm, starting from multi-source, heterogeneous, and multi-dimensional sensor data. Each step builds upon the optimized output of the previous step and provides higher-value input for the next, fully demonstrating the invention's systematic nature, rationality, and high degree of creativity.

[0115] Compared to the various single measurement methods described in the background art, the present invention, through its unique system structure and hierarchical progressive data fusion method, has produced unexpected technical effects and brought about the following significant and beneficial advancements.

[0116] Overcoming the inherent flaw of overgeneralization, this invention achieves a qualitative leap in measurement accuracy: Existing technologies, whether for point, line, or surface flow measurement, essentially infer the whole from local information. This invention completely abandons this approach by fusing multi-dimensional information from points, lines, and surfaces at the feature layer and reconstructing the three-dimensional flow field using a physical model as a strong constraint. This represents a paradigm shift from inference to reconstruction, fundamentally solving the accuracy bottleneck caused by the single dimension of information, resulting in an order-of-magnitude improvement in the accuracy of the three-dimensional flow field.

[0117] This invention solves the problem of information silos between heterogeneous data, significantly improving system robustness: In existing technologies, measurement results from different sensors cannot be cross-validated. This invention, through point-to-surface calibration and surface-to-line extension, enables sensor data from different sources and dimensions to mutually verify and reinforce each other. For example, the high-precision point velocity from radar provides an absolute reference for the surface flow field in video, while the surface flow field in video provides a basis for the lateral distribution of the linear velocity in ultrasonic waves. This synergistic effect effectively suppresses the random errors of individual sensors and environmental interference, making the measurement results of the entire system stable and reliable.

[0118] This invention enhances adaptability to complex operating conditions and broadens application scenarios: Existing technologies rely on simplified flow regime assumptions, which can lead to significant errors under complex flow regimes such as flood season and gate regulation. This invention, by integrating multi-dimensional field measurement data that reflects the true flow regime (especially the global surface flow field and stratified underwater velocity), reconstructs a three-dimensional flow field that accurately reflects the non-uniform and non-constant complex flow structure. Therefore, this invention can adapt to a wider range of more challenging actual operating conditions, with an application scope far exceeding that of existing technologies.

[0119] This invention provides a quantitative uncertainty assessment, enhancing its application value: The Gaussian process regression model used in this invention, while outputting the predicted flow velocity value, can also provide the prediction uncertainty for each spatial point. This is something that existing technologies cannot provide at all. This function enables users to quantitatively assess the reliability of the data, providing crucial confidence information for advanced applications such as water resource scheduling decisions and hydraulic engineering model verification.

[0120] Example 2: The present invention provides a three-dimensional flow field construction device for open channel in a sensor information fusion system, which includes a preprocessing module, a point-surface module, a surface-line module, a layering module, and a calculation module.

[0121] The preprocessing module is used to acquire multi-source measurement data characterizing the open channel flow velocity and boundary conditions collected by radar sensors, image sensors, a first ultrasonic transducer array, a second ultrasonic transducer array, and a temperature sensor group deployed on the same cross-section. Based on the water temperature and air temperature collected by the temperature sensor group, the module performs temperature compensation processing on the measurement results of the radar sensors and ultrasonic transducer array to obtain the compensated surface point velocity, underwater stratification line velocity, lower boundary of the cross-section, and surface two-dimensional flow field.

[0122] The point-to-surface module is used to generate an optimized surface flow field by performing fusion calibration based on the surface point flow velocity and the surface two-dimensional flow field through a Kalman filter model.

[0123] The surface line module is used to generate an underwater stratified flow field by extracting the lateral distribution pattern of the optimized surface flow field and applying it to each underwater layer, based on the optimized surface flow field and the underwater stratified line velocity.

[0124] The layered module is used to reconstruct the three-dimensional flow field of the open channel based on the optimized surface flow field and the underwater layered flow field, using a Gaussian process regression model with a preset fluid dynamics model as the prior mean function.

[0125] Example 3: This invention provides a multi-sensor information fusion system, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a method for constructing a three-dimensional flow field in an open channel using a sensor information fusion system as described in any paragraph of Example 1.

[0126] Example 4: This invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform a method for constructing a three-dimensional flow field in an open channel using a sensor information fusion system, as described in any paragraph of Example 1.

[0127] Obviously, the embodiments described above are only some embodiments of the present invention, and not all embodiments. 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.

[0128] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0129] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0130] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0131] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention are also intended to include the plural forms unless the context clearly indicates otherwise.

[0132] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0133] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0134] The terms "first" and "second" used in the embodiments are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0135] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An open channel three-dimensional flow field construction method of a sensor information fusion system, characterized by, The application relates to a method for reconstructing a three-dimensional flow field of an open channel. The method comprises the following steps: acquiring multi-source measurement data representing flow velocity and boundary conditions of the open channel, and performing temperature compensation on measurement results of the radar sensor and the ultrasonic transducer array based on water temperature and air temperature acquired by the temperature sensor group, so as to obtain compensated surface point flow velocity, underwater layer flow velocity, a lower boundary of the water section and a surface two-dimensional flow field; fusing and calibrating through a Kalman filtering model based on the surface point flow velocity and the surface two-dimensional flow field, so as to generate an optimized surface flow field; extracting a transverse distribution mode of the optimized surface flow field and applying the transverse distribution mode to each underwater layer, so as to generate an underwater layer flow field based on the optimized surface flow field and the underwater layer flow velocity; 2. The open channel three-dimensional flow field construction method of a sensor information fusion system according to claim 1, characterized in that, reconstructing a three-dimensional flow field of the open channel by adopting a Gaussian process regression model taking a preset fluid mechanics model as a prior mean function based on the optimized surface flow field and the underwater layer flow field. define a state vector, said state vector is the global scale factor of the image-measured flow field relative to the real flow field at time and an additive bias ; i.e. ; denotes the transpose;​ The state transition equation is established as: wherein is a state vector at a current time, is a state transition matrix, is a state vector at a previous time, is process noise following a zero-mean Gaussian distribution; based on the optimal estimation at the last time to obtain the prior prediction state at the current time and the prediction error covariance ; wherein is the error covariance at the last time, is the covariance matrix of the process noise; Establish the observation equation: ; wherein is an observation value of the radar sensor at time , is an observation matrix, is an observation noise subject to a zero-mean Gaussian distribution; computing a Kalman gain updating a state estimate and updating an error covariance ; wherein is an error covariance of the observation noise; is an identity matrix; includes an optimal scale factor and an optimal bias ; based on with , generating an optimized surface flow field: ; where is the original surface two-dimensional flow field at time .

3. The open channel three-dimensional flow field construction method of claim 1, wherein, The fusing and calibrating through the Kalman filtering model specifically comprises: According to the optimization of the transverse velocity distribution of the surface flow field at the water cross section , the normalized transverse distribution mode function is calculated : ； wherein is the integrated average of the optimized surface flow field along the path of the ultrasound wave channel. for each depth measured by the first ultrasonic transducer array :​​ 。 4. The open channel three-dimensional flow field construction method of claim 1, wherein, extracting a transverse distribution mode of the optimized surface flow field and applying the transverse distribution mode to each underwater layer, specifically comprising: reconstructing a three-dimensional flow field of the open channel by adopting a Gaussian process regression model taking a preset fluid mechanics model as a prior mean function, specifically comprising: Constructing the covariance function of Gaussian process regression model; where is the covariance, and denote any two points in the space, is the natural exponential function, is the length scale parameter; performing a Gaussian process regression prediction based on a sample set comprising samples from the optimized surface flow field and the underwater layered flow field, to predict flow velocities at any location within the open channel, reconstructing a three-dimensional flow field of the open channel ; ； ； where: is the point its own covariance, is the covariance vector between the point to be tested and all training points, is the covariance matrix between training data points, is the flow velocity observation vector of data points, is the fluid dynamics model, is the prior mean vector at data points, is the observation noise variance; is the identity matrix.

5. The open channel three-dimensional flow field construction method of claim 4, wherein, constructing a prior mean function of the Gaussian process regression model according to the preset fluid mechanics model; When an exponential law is used, the prior mean function is: ; The prior mean function when using the exponential law is deformed as: ; wherein is a fluid mechanics model, is a surface flow velocity, is a distance from the channel height, is a water depth, is a roughness correlation coefficient, is a point directly above the surface flow velocity.

6. The open channel three-dimensional flow field construction method of claim 1, wherein, the fluid mechanics model is a flow velocity logarithmic law or an exponential law; the acquiring multi-source measurement data specifically comprises: the radar sensor is a frequency-modulated continuous wave radar and is arranged above the water section of the open channel, so as to measure surface point flow velocity of a water surface based on the Doppler effect and measure a distance from the radar to the water surface based on a ranging function to acquire a water level; the image sensor is arranged at a position above the water section and can overlook the water section, so as to continuously acquire an image sequence of the water surface and generate a surface two-dimensional flow field covering the water section by adopting one or more image flow measurement algorithms such as STIV, PIV, LSPIV, OP and PTV; the first ultrasonic transducer array is arranged at multiple different depth positions of the channel bottom and / or side wall, so as to measure underwater layer flow velocity at multiple depths based on an ultrasonic time-of-flight difference method; the second ultrasonic transducer array is arranged towards the channel bottom, so as to acquire a silt top surface height through ultrasonic echo ranging to determine the lower boundary of the water section; 7. An open channel three-dimensional flow field construction device of a sensor information fusion system, characterized by, the temperature sensor group comprises an underwater temperature sensor and an air temperature sensor, the underwater temperature sensor is used for temperature compensation on measurement results of the first ultrasonic transducer array and the second ultrasonic transducer array, and the air temperature sensor is used for temperature compensation on measurement results of the radar sensor. The application relates to a method for reconstructing a three-dimensional flow field of an open channel. The method comprises the following steps: acquiring multi-source measurement data representing flow velocity and boundary conditions of the open channel, and performing temperature compensation on measurement results of the radar sensor and the ultrasonic transducer array based on water temperature and air temperature acquired by the temperature sensor group, so as to obtain compensated surface point flow velocity, underwater layer flow velocity, a lower boundary of the water section and a surface two-dimensional flow field; fusing and calibrating through a Kalman filtering model based on the surface point flow velocity and the surface two-dimensional flow field, so as to generate an optimized surface flow field; extracting a transverse distribution mode of the optimized surface flow field and applying the transverse distribution mode to each underwater layer, so as to generate an underwater layer flow field based on the optimized surface flow field and the underwater layer flow velocity; reconstructing a three-dimensional flow field of the open channel by adopting a Gaussian process regression model taking a preset fluid mechanics model as a prior mean function based on the optimized surface flow field and the underwater layer flow field. The fusing and calibrating through the Kalman filtering model specifically comprises: extracting a transverse distribution mode of the optimized surface flow field and applying the transverse distribution mode to each underwater layer, specifically comprising: reconstructing a three-dimensional flow field of the open channel by adopting a Gaussian process regression model taking a preset fluid mechanics model as a prior mean function, specifically comprising: constructing a prior mean function of the Gaussian process regression model according to the preset fluid mechanics model; the fluid mechanics model is a flow velocity logarithmic law or an exponential law; the acquiring multi-source measurement data specifically comprises: the radar sensor is a frequency-modulated continuous wave radar and is arranged above the water section of the open channel, so as to measure surface point flow velocity of a water surface based on the Doppler effect and measure a distance from the radar to the water surface based on a ranging function to acquire a water level; the image sensor is arranged at a position above the water section and can overlook the water section, so as to continuously acquire an image sequence of the water surface and generate a surface two-dimensional flow field covering the water section by adopting one or more image flow measurement algorithms such as STIV, PIV, LSPIV, OP and PTV; the first ultrasonic transducer array is arranged at multiple different depth positions of the channel bottom and / or side wall, so as to measure underwater layer flow velocity at multiple depths based on an ultrasonic time-of-flight difference method; the second ultrasonic transducer array is arranged towards the channel bottom, so as to acquire a silt top surface height through ultrasonic echo ranging to determine the lower boundary of the water section; the temperature sensor group comprises an underwater temperature sensor and an air temperature sensor, the underwater temperature sensor is used for temperature compensation on measurement results of the first ultrasonic transducer array and the second ultrasonic transducer array, and the air temperature sensor is used for temperature compensation on measurement results of the radar sensor. The preprocessing module is configured to acquire multi-source measurement data representing open channel flow velocity and boundary conditions collected by a radar sensor, an image sensor, a first ultrasonic transducer array, a second ultrasonic transducer array and a temperature sensor group arranged at the same water section, and perform temperature compensation on measurement results of the radar sensor and the ultrasonic transducer array based on water temperature and air temperature collected by the temperature sensor group, to obtain compensated surface point flow velocity, underwater layered line flow velocity, a lower boundary of the water section and a surface two-dimensional flow field. The point-surface module is configured to generate an optimized surface flow field by performing fusion calibration through a Kalman filtering model based on the surface point flow velocity and the surface two-dimensional flow field. The surface-line module is configured to generate an underwater layered flow field by extracting a transverse distribution mode of the optimized surface flow field and applying the transverse distribution mode to each underwater layer based on the optimized surface flow field and the underwater layered line flow velocity. The layered module is configured to reconstruct a three-dimensional flow field of the open channel by adopting a Gaussian process regression model with a preset fluid mechanics model as a prior mean function based on the optimized surface flow field and the underwater layered flow field.

8. A multi-sensor information fusion system characterized by, The computer readable storage medium comprises a stored computer program, wherein the computer program controls a device where the computer readable storage medium is located to perform the open channel three-dimensional flow field construction method of the sensing information fusion system according to any one of claims 1 to 6 when the computer program is running.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer program controls a device where the computer readable storage medium is located to perform the open channel three-dimensional flow field construction method of the sensing information fusion system according to any one of claims 1 to 6 when the computer program is running.

Images