Rod bundle channel annular flow characteristic prediction method based on disturbance wave dynamic parameters

Abstract

The invention discloses a rod bundle channel annular flow characteristic prediction method based on disturbance wave dynamic parameters, and relates to the technical field of nuclear reactor thermal hydraulic and multiphase flow measurement. The method comprises the following steps: acquiring spatio-temporal evolution data of the surface of an annular flow liquid film in a rod bundle channel by adopting a non-invasive optical measurement system, and preprocessing the spatio-temporal evolution data; extracting a disturbance wave dynamic parameter set from the preprocessed data; constructing a hybrid prediction model by taking the disturbance wave dynamic parameter set as input and taking the target annular flow characteristics as output; and predicting a to-be-predicted rod bundle channel annular flow by using the trained hybrid prediction model to obtain an annular flow characteristic prediction result. The invention can overcome the defects of intrusion interference, incomplete local description or high calculation cost of the existing rod bundle channel annular flow characteristic prediction method, and provides a non-intrusive rod bundle channel annular flow characteristic high-precision prediction method which is based on disturbance wave dynamic parameters and can reflect the global spatio-temporal evolution characteristics of a flow field.

Description

Technical Field

[0001] This invention relates to the field of nuclear reactor thermal-hydraulic and multiphase flow measurement technology, and in particular to a method for predicting the characteristics of annular flow in rod bundle channels based on dynamic parameters of disturbance waves. Background Technology

[0002] In nuclear facilities such as pressurized water reactors, gas-liquid two-phase annular flow is widely present in fuel rod bundle channels under both normal operation and accident conditions. The characteristics of the annular flow (such as liquid film thickness, interface wave statistical characteristics, liquid film velocity, entrainment / deposition rate, etc.) directly determine the cooling efficiency of the fuel rods and the flow resistance of the channels, which is crucial to the safety and economy of the reactor.

[0003] Currently, research on the annular flow characteristics of rod bundle channels mainly relies on empirical relationships, computational fluid dynamics (CFD) simulations, and local probe measurements. Empirical relationships have poor universality and low prediction accuracy in complex rod bundle geometries; CFD simulations are computationally expensive and have limited ability to capture transient details such as interface waves; traditional probe measurements (such as conductivity / capacitance probes and fiber optic probes) are mostly invasive, which interferes with the original flow field and usually only obtains single-point or sparse point information, making it difficult to comprehensively characterize the non-uniform distribution characteristics of the liquid film in the circumferential and axial directions of the rod bundle.

[0004] Disturbance waves are the dominant structure on the surface of annular flow liquid films, and their dynamic parameters (such as wave velocity, frequency, amplitude, and spatial wavelength) are closely related to the average thickness of the liquid film, flow stability, and interphase interaction forces. Therefore, developing a method for predicting the characteristics of annular flow based on non-invasive, global dynamic parameters of disturbance waves has significant engineering application value. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave, aiming to solve or improve at least one of the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave includes: The spatiotemporal evolution data of the annular flow film surface in the rod bundle channel were acquired using a non-invasive optical measurement system and preprocessed to obtain preprocessed data. The dynamic parameter set of the perturbation wave is extracted from the preprocessed data; the dynamic parameter set of the perturbation wave includes wave velocity field, characteristic frequency and energy spectrum, amplitude statistics, spatial wavenumber spectrum, nonlinear interaction parameters and circumferential non-uniformity parameters; A hybrid prediction model is constructed using the set of dynamic parameters of the disturbance wave as input and the target annular flow characteristics as output; the target annular flow characteristics include at least the time-averaged liquid film thickness distribution, the circumferential distribution of liquid film volumetric flow rate, the interfacial shear stress, the liquid film flow pressure drop gradient, the droplet entrainment rate and the deposition rate. For the annular flow in the bar bundle channel to be predicted, the corresponding dynamic parameter set of the disturbance wave is extracted and input into the trained hybrid prediction model to obtain the prediction results of the annular flow characteristics.

[0007] Optionally, the spatiotemporal evolution data of the annular flow film surface within the rod bundle channel is acquired using a non-invasive optical measurement system and preprocessed to obtain preprocessed data, specifically including: For the experimental section of the target rod bundle channel or the simulated channel with a transparent observation window, a non-invasive optical measurement system is built, and the spatiotemporal evolution data of the liquid film are collected based on the non-invasive optical measurement system. The spatiotemporal evolution data of the liquid film is subjected to noise reduction filtering, background removal, outlier correction, and data alignment to obtain preprocessed data. The noise reduction filtering adopts wavelet threshold noise reduction or three-dimensional Gaussian filtering. The background removal is obtained by subtracting the time-averaged thickness at each spatial point from the original signal to obtain the pulsating thickness field. The outlier correction is used to correct abnormal data points caused by bubbles and droplets. The data alignment uses channel geometry to perform coordinate alignment and distortion correction on the image data.

[0008] Optionally, the method for acquiring the spatiotemporal evolution data of the liquid film is as follows: Based on the non-invasive optical measurement system, a specific concentration of fluorescent dye is uniformly mixed in the liquid working medium. A sheet light source is used to illuminate a cross-section or oblique section containing the gap between the rod bundles along the channel axis. A high-speed camera is arranged perpendicular to the sheet light plane to synchronously acquire a sequence of fluorescence intensity images on the surface of the liquid film. By using a pre-performed static thickness-fluorescence intensity calibration curve, the fluorescence intensity image sequence is converted into an instantaneous local thickness field of the liquid film surface relative to the rod wall, thus obtaining the spatiotemporal evolution matrix of the liquid film.

[0009] Optionally, extracting the dynamic parameter set of the disturbance wave from the preprocessed data specifically includes: The pulsating thickness field in the preprocessed data is processed using the intrinsic orthogonal decomposition method to obtain multi-order decomposed modes. The spatiotemporal data of each mode obtained by decomposition are reorganized into a two-dimensional matrix and singular value decomposition is performed to obtain a series of spatiotemporal mode pairs arranged in descending order of energy. The first M spatiotemporal modes are then identified as the dominant modes of large-scale perturbation waves. Based on the dominant mode of the large-scale disturbance wave and the pulsating thickness field, multi-dimensional dynamic parameters are extracted, and a comprehensive set of dynamic parameters of the disturbance wave is constructed.

[0010] Optionally, the construction process of the hybrid prediction model includes: A deep neural network is used as the core regressor, and a physical information neural network is constructed. The physical conservation equation derived from interface wave dynamics is introduced into the loss function as a constraint term for model training.

[0011] Optionally, the intrinsic orthogonal decomposition method is to expand the pulsating thickness field into a two-dimensional matrix along the time dimension and then perform snapshot POD analysis.

[0012] Optionally, the deep neural network contains three or more hidden layers, each hidden layer having no fewer nodes than the dimension of the input feature vector X, and the activation function is either ReLU or Tanh.

[0013] Optionally, it also includes: online model updates; when a set amount of reliable experimental data for new operating conditions is accumulated, the deployed hybrid prediction model is adjusted and updated using transfer learning or incremental learning techniques.

[0014] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention discloses a method for predicting the characteristics of annular flow in a rod bundle channel based on perturbation wave dynamic parameters. The method includes acquiring spatiotemporal evolution data of the annular flow liquid film surface within the rod bundle channel using a non-invasive optical measurement system, and preprocessing the data to obtain preprocessed data; extracting a set of perturbation wave dynamic parameters from the preprocessed data; the set of perturbation wave dynamic parameters includes wave velocity field, characteristic frequency and energy spectrum, amplitude statistics, spatial wavenumber spectrum, nonlinear interaction parameters, and circumferential non-uniformity parameters; constructing a hybrid prediction model using the set of perturbation wave dynamic parameters as input and the target annular flow characteristics as output; the target annular flow characteristics include at least the time-averaged liquid film thickness distribution, the circumferential distribution of liquid film volumetric flow rate, interfacial shear stress, liquid film flow pressure drop gradient, droplet entrainment rate, and deposition rate; for the annular flow in the rod bundle channel to be predicted, extracting the corresponding set of perturbation wave dynamic parameters and inputting it into the trained hybrid prediction model to obtain the predicted annular flow characteristics. This invention overcomes the shortcomings of existing rod bundle channel annular flow characteristic prediction methods, such as intrusive interference, incomplete local description, or high computational cost. It provides a non-intrusive, high-precision prediction method for rod bundle channel annular flow characteristics based on perturbation wave dynamic parameters, which can reflect the spatiotemporal evolution characteristics of the entire flow field. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the method for predicting the characteristics of annular flow in a rod bundle channel based on the dynamic parameters of a disturbance wave, as described in this invention. Figure 2 This is a schematic diagram of the overall prediction framework in this embodiment. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The purpose of this invention is to provide a method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave, aiming to solve or improve at least one of the above-mentioned technical problems.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figures 1-2 As shown, this invention provides a method for predicting the characteristics of annular flow in a rod bundle channel based on the dynamic parameters of a disturbance wave, comprising: Step 100: Acquire spatiotemporal evolution data of the annular flow film surface within the rod bundle channel using a non-invasive optical measurement system, and preprocess the data to obtain preprocessed data. Specifically, this includes: For the experimental section of the target rod bundle channel or the simulated channel with a transparent observation window, a non-invasive optical measurement system is built, and liquid film spatiotemporal evolution data is collected based on the non-invasive optical measurement system; the liquid film spatiotemporal evolution data is subjected to noise reduction filtering, background removal, outlier correction and data alignment to obtain preprocessed data.

[0021] The method for acquiring the spatiotemporal evolution data of the liquid film is as follows: Based on the non-invasive optical measurement system, a specific concentration of fluorescent dye is uniformly mixed in the liquid working fluid. A sheet light source is used to illuminate a cross-section or oblique section containing the gap between the rod bundles along the channel axis. A high-speed camera is arranged perpendicular to the sheet light plane to synchronously acquire a sequence of fluorescence intensity images on the surface of the liquid film. Through a pre-performed static thickness-fluorescence intensity calibration curve, the sequence of fluorescence intensity images is converted into an instantaneous local thickness field of the liquid film surface relative to the rod wall, thus obtaining the spatiotemporal evolution matrix of the liquid film.

[0022] Step 200: Extract the dynamic parameter set of the perturbation wave from the preprocessed data; the dynamic parameter set of the perturbation wave includes wave velocity field, characteristic frequency and energy spectrum, amplitude statistics, spatial wavenumber spectrum, nonlinear interaction parameters, and circumferential inhomogeneity parameters. Specifically, it includes: The pulsating thickness field in the preprocessed data is processed using an intrinsic orthogonal decomposition method. The spatiotemporal data is reorganized into a two-dimensional matrix and subjected to singular value decomposition. The top M spatiotemporal modes, arranged in descending energy order, are then identified as the dominant modes of large-scale perturbation waves. Based on the dominant modes of large-scale perturbation waves and the pulsating thickness field, multi-dimensional dynamic parameters are extracted, and a comprehensive set of perturbation wave dynamic parameters is constructed. Specifically, the intrinsic orthogonal decomposition method involves expanding the pulsating thickness field along the time dimension into a two-dimensional matrix and then performing snapshot POD analysis.

[0023] Step 300: Construct a hybrid prediction model using the dynamic parameter set of the disturbance wave as input and the target annular flow characteristics as output; the target annular flow characteristics include at least the time-averaged liquid film thickness distribution, the circumferential distribution of liquid film volumetric flow rate, interfacial shear stress, liquid film flow pressure drop gradient, droplet entrainment rate, and deposition rate. The construction process of the hybrid prediction model includes: A deep neural network is used as the core regressor, and a physical information neural network is constructed. The physical conservation equations derived from interface wave dynamics are introduced as constraints into the loss function for model training. The deep neural network contains at least three hidden layers, with each hidden layer having at least as many nodes as the dimension of the input feature vector X. ReLU or Tanh activation functions are selected.

[0024] Step 400: For the annular flow in the bar bundle channel to be predicted, extract the corresponding dynamic parameter set of the disturbance wave and input it into the trained hybrid prediction model to obtain the prediction result of the annular flow characteristics.

[0025] Furthermore, as a further implementation method, this method also includes online model updates; when a set amount of reliable experimental data for new operating conditions is accumulated, transfer learning or incremental learning techniques are used to adjust and update the deployed hybrid prediction model.

[0026] Based on the above technical solution, the following embodiments are provided.

[0027] S100: Non-invasive acquisition and refined preprocessing of spatiotemporal evolution data of liquid film interfaces. A high-frequency synchronous laser and a high-speed camera are deployed outside the transparent outer casing of the rod bundle channel, employing laser-induced fluorescence. A trace amount of fluorescent dye is added to the liquid phase, and an axial-circumferential plane of the rod bundle channel is illuminated by a sheet laser. The high-speed camera records the fluorescence intensity distribution on the liquid film surface (which is related to the liquid film thickness), and after calibration, it is converted into a spatiotemporal matrix of liquid film surface elevation.

[0028] S110: Data Acquisition: For the experimental section of the target rod bundle channel or a simulated channel with a transparent observation window, a non-invasive optical measurement system is constructed. Planar laser-induced fluorescence (PLIF) or white light interferometry is preferred. For the PLIF method, a specific concentration of fluorescent dye is uniformly mixed in the liquid working fluid. A sheet light source is used to illuminate a cross-section or oblique section containing the rod bundle gaps along the channel axis. A high-speed camera is positioned perpendicular to the sheet light plane to simultaneously acquire the fluorescence intensity image sequence I(x,y,t) of the liquid film surface. Using a pre-performed static thickness-fluorescence intensity calibration curve, I(x,y,t) is converted into the instantaneous local thickness field of the liquid film surface relative to the rod wall, i.e., the spatiotemporal evolution matrix η(x,y,t) of the liquid film elevation, where x is the axial coordinate along the main flow direction, y is the arc length coordinate along the circumference of the rod bundle, and t is time.

[0029] S120: Data Preprocessing: The raw η(x,y,t) data is processed as follows to improve the signal-to-noise ratio and extract effective fluctuation information: Noise reduction filtering: Wavelet threshold denoising or three-dimensional Gaussian filtering is used to suppress high-frequency random noise introduced by camera noise, light source fluctuations, etc.

[0030] Background Removal: The time-averaged thickness bar{η}(x,y) at each spatial point (x,y) is calculated and subtracted from the original signal to obtain the pulsating thickness field η'(x,y,t)=η(x,y,t)-\bar{η}(x,y). This pulsating field mainly contains information about disturbance waves.

[0031] Outlier correction and data alignment: Corrects outlier data points caused by bubbles and droplets, and performs coordinate alignment and distortion correction on image data based on channel geometry.

[0032] S200: Dominant Mode Identification and Quantitative Extraction of Multi-Dimensional Dynamic Parameters for Disturbance Waves. Intrinsic orthogonal decomposition is performed on the η(x,y,t) data to obtain multiple decomposed modes. From the time coefficient sequence of the first-order spatiotemporal mode obtained from the decomposition (representing the most energetic disturbance wave structure), the propagation velocity c of the disturbance wave is calculated using the cross-correlation method. w The dominant frequency f is obtained through power spectrum analysis. d The spatial structure of the modes provides wavelength information. Simultaneously, the amplitude A of the perturbation wave is statistically analyzed from the original signal. w These parameters are divided into sections according to the bar bundle gap and the circumferential position of the bar surface to obtain Φ(y).

[0033] S210: Dominant Spatiotemporal Mode Decomposition: The pulsating thickness field η'(x,y,t) is processed using the intrinsic orthogonal decomposition method. The spatiotemporal data is reorganized into a two-dimensional matrix and subjected to singular value decomposition, yielding a series of spatiotemporal mode pairs arranged in descending energy order. The first M-order spatiotemporal modes (typically the first 1-4 orders, with a cumulative energy contribution >70%) are identified as the dominant modes of the large-scale perturbation wave. The spatial components of these modes reveal the typical structure of the perturbation wave (such as wave number and propagation direction), while the time coefficient sequence reflects its dynamic evolution.

[0034] S220: Construction of a Multi-Dimensional Dynamic Parameter Set: Based on the dominant mode and the original pulsating field, a comprehensive dynamic parameter set Φ is extracted, which includes: Wave velocity field (c w ): Perform cross-correlation analysis on the time coefficient sequence of the dominant mode, or perform a two-dimensional spatiotemporal Fourier transform on η' along the x-axis, and obtain the frequency ω and wavenumber k corresponding to the peak of the energy spectrum. x Calculate phase velocity c p =ω / k x The average wave velocity and wave velocity distribution can be statistically obtained.

[0035] Characteristic frequencies and energy spectrum (f d P(f): Perform power spectral density analysis on the POD time coefficient or the η' time series at a fixed point to extract the dominant frequency f. d And the characteristics of the energy spectrum, such as spectral width and slope.

[0036] Amplitude statistics (A) w In a pulsating field η', identify and track individual perturbation wave events, statistically analyze the probability density distribution of their peak amplitudes, and extract the root mean square amplitude A. rms Average peak amplitude A mean wait.

[0037] Spatial wavenumber spectrum (Λ(k) x k y ): Perform Fourier transform on the two-dimensional field in η' space at a specific time, analyze its wavenumber spectrum, and obtain the dominant wavelengths in the axial and circumferential directions and their anisotropic characteristics.

[0038] Nonlinear interaction parameters: By analyzing biphase coherence spectra, nonlinear resonant coupling is detected between perturbation waves of different frequencies / wavenumbers.

[0039] Circumferential non-uniformity parameter (Φ(y)): The above parameters are statistically analyzed by dividing the circumferential position of the rod bundle channel (such as the center of the sub-channel, the gap region, and the surface of the rod) to form a parameter distribution vector that varies with the circumferential position y, so as to characterize the strong three-dimensional effect caused by the rod bundle geometry.

[0040] S300: Construction, training, and validation of a data-physical fusion prediction model. Construct a model with Φ(y) (containing c) w f d A w A gradient boosting decision tree model is constructed, using Φ(y) (the time-averaged liquid film thickness circumferential distribution) as the input feature and δ(y) as the output target. Multiple benchmark experiments were conducted by varying the gas and liquid phase flow rates to obtain a training dataset [Φ(y), δ(y)] covering a wide range of operating conditions. 80% of the data was used for training, and 20% for testing.

[0041] S310: Model Architecture Design: Construct a hybrid prediction model that uses the multi-dimensional perturbation wave dynamic parameter set Φ (or Φ(y)) extracted in step S2 as the input feature vector X. The model output is the target annular flow characteristic set Y, including but not limited to: Time-averaged liquid film thickness distribution δ(x,y); circumferential distribution of liquid film volumetric flow rate Q f(y) Interfacial shear stress τ i Droplet entrainment rate E and deposition rate D; liquid film flow pressure drop gradient (dp / dz) film .

[0042] S320: Model Implementation I. Data-Driven Core: Gradient Boosting Decision Trees (GTB) or Deep Neural Networks (DNNs) are used as the core regressors. GTB models can effectively handle nonlinear features, while DNNs (such as fully connected networks) are suitable for learning high-dimensional complex mappings.

[0043] II. Enhanced Physical Constraints: A physical information neural network is constructed, and physical conservation equations derived from interface wave dynamics (such as simplified thin film equations) are introduced into the loss function as constraint terms, so that the prediction results not only fit the data, but also obey physical laws, thereby improving the extrapolation ability.

[0044] III. Multi-task learning: Design a multi-task learning model to predict multiple related annular flow characteristics Y simultaneously, and improve the overall prediction accuracy and model efficiency by utilizing the physical correlation between tasks.

[0045] S330: Model Training and Validation: Under a wide range of operating conditions (different gas / liquid apparent velocities, pressures, subcooling) and geometries (different rod spacing, positioning grids), sufficient pairs of [X] models are generated through precise benchmark experiments or validated high-fidelity two-phase flow CFD simulations. i Y i The dataset is divided into training, validation, and test sets. The training set is used to optimize model parameters, the validation set is used for hyperparameter tuning and to prevent overfitting, and finally, the model's prediction accuracy (e.g., root mean square error, coefficient of determination R0) is evaluated on an independent test set. 2 ).

[0046] S400: Online / Offline Prediction and Application of Target Channel Circulating Flow Characteristics. For an unknown new operating condition, repeat steps one and two to obtain its Φ. new(y) Input the trained GBDT model to directly predict δ under this working condition. pred(y) .

[0047] S410: Characteristic Prediction: For any new rod bundle channel scenario (new operating condition or new geometry) where the characteristics of the annular flow need to be predicted, first implement steps S1 and S2 to non-invasively acquire its liquid film dynamic data and extract the corresponding disturbance wave dynamic parameter feature vector X. target .

[0048] S420: Model Inference: X target The data is input into the optimal prediction model that has been trained and solidified in step S3. The model automatically and quickly outputs high-precision predicted values ​​Y for the annular flow characteristics of all targets in this scenario. pred .

[0049] S430: Engineering Application Feedback: The predicted Y pred (Especially the liquid film thickness distribution and entrainment rate) can be directly input into the reactor thermal-hydraulic system analysis program or sub-channel analysis program to more accurately calculate the temperature field of the fuel rods, critical heat flux margin, and channel pressure drop, thereby providing key inputs for reactor safety analysis and optimized operation.

[0050] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0051] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave, characterized in that, include: The spatiotemporal evolution data of the annular flow film surface in the rod bundle channel were acquired using a non-invasive optical measurement system and preprocessed to obtain preprocessed data. The dynamic parameter set of the perturbation wave is extracted from the preprocessed data; the dynamic parameter set of the perturbation wave includes wave velocity field, characteristic frequency and energy spectrum, amplitude statistics, spatial wavenumber spectrum, nonlinear interaction parameters and circumferential non-uniformity parameters; A hybrid prediction model is constructed using the set of dynamic parameters of the disturbance wave as input and the target annular flow characteristics as output; the target annular flow characteristics include at least the time-averaged liquid film thickness distribution, the circumferential distribution of liquid film volumetric flow rate, the interfacial shear stress, the liquid film flow pressure drop gradient, the droplet entrainment rate and the deposition rate. For the annular flow in the bar bundle channel to be predicted, the corresponding dynamic parameter set of the disturbance wave is extracted and input into the trained hybrid prediction model to obtain the prediction results of the annular flow characteristics.

2. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 1, characterized in that, The method employs a non-invasive optical measurement system to acquire spatiotemporal evolution data of the annular flow film surface within the rod bundle channel, and performs preprocessing to obtain preprocessed data, specifically including: For the experimental section of the target rod bundle channel or the simulated channel with a transparent observation window, a non-invasive optical measurement system is built, and the spatiotemporal evolution data of the liquid film are collected based on the non-invasive optical measurement system. The spatiotemporal evolution data of the liquid film is subjected to noise reduction filtering, background removal, outlier correction, and data alignment to obtain preprocessed data. The noise reduction filtering adopts wavelet threshold noise reduction or three-dimensional Gaussian filtering. The background removal is obtained by subtracting the time-averaged thickness at each spatial point from the original signal to obtain the pulsating thickness field. The outlier correction is used to correct abnormal data points caused by bubbles and droplets. The data alignment uses channel geometry to perform coordinate alignment and distortion correction on the image data.

3. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 2, characterized in that, The method for acquiring the spatiotemporal evolution data of the liquid film is as follows: Based on the non-invasive optical measurement system, a specific concentration of fluorescent dye is uniformly mixed in the liquid working medium. A sheet light source is used to illuminate a cross-section or oblique section containing the gap between the rod bundles along the channel axis. A high-speed camera is arranged perpendicular to the sheet light plane to synchronously acquire a sequence of fluorescence intensity images on the surface of the liquid film. By using a pre-performed static thickness-fluorescence intensity calibration curve, the fluorescence intensity image sequence is converted into an instantaneous local thickness field of the liquid film surface relative to the rod wall, thus obtaining the spatiotemporal evolution matrix of the liquid film.

4. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 1, characterized in that, The extraction of the dynamic parameter set of the disturbance wave from the preprocessed data specifically includes: The pulsating thickness field in the preprocessed data is processed using the intrinsic orthogonal decomposition method to obtain multi-order decomposed modes. The spatiotemporal data of each mode obtained by decomposition are reorganized into a two-dimensional matrix and singular value decomposition is performed to obtain a series of spatiotemporal mode pairs arranged in descending order of energy. The first M spatiotemporal modes are then identified as the dominant modes of large-scale perturbation waves. Based on the dominant mode of the large-scale disturbance wave and the pulsating thickness field, multi-dimensional dynamic parameters are extracted, and a comprehensive set of dynamic parameters of the disturbance wave is constructed.

5. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 1, characterized in that, The construction process of the hybrid prediction model includes: A deep neural network is used as the core regressor, and a physical information neural network is constructed. The physical conservation equation derived from interface wave dynamics is introduced into the loss function as a constraint term for model training.

6. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 4, characterized in that, The intrinsic orthogonal decomposition method is as follows: after expanding the pulsating thickness field into a two-dimensional matrix along the time dimension, a snapshot POD analysis is performed.

7. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 5, characterized in that, The deep neural network contains three or more hidden layers, each with no fewer than the dimension of the input feature vector X, and the activation function is either ReLU or Tanh.

8. The method for predicting the characteristics of annular flow in a rod bundle channel based on dynamic parameters of a disturbance wave according to claim 1, characterized in that, It also includes: online model updates; when a set amount of reliable experimental data for new operating conditions is accumulated, transfer learning or incremental learning techniques are used to adjust and update the deployed hybrid prediction model.

Images