Water level data acquisition and analysis method and system

Abstract

The application discloses a water level data acquisition and analysis method and system, relates to the technical field of hydraulic engineering monitoring, and comprises the following steps: monitoring a water pressure sensitive deformation rod vertically arranged in a section, collecting water level optical fiber data, wherein the water level optical fiber data comprises a first optical fiber sequence and a second optical fiber sequence; performing temperature deviation processing on the water level optical fiber data through a double-mode difference algorithm to obtain a net strain sequence; acquiring historical hydrostatic pressure data, performing spatial derivative analysis mapping on the net strain sequence to obtain actual water level data; performing singular value decomposition on the actual water level data by using a comprehensive reconstruction network model to obtain a water level reference sequence; constructing a hydrodynamic model based on the water level reference sequence to obtain simulated water level errors, constructing a linear regression and scaling correction model, and performing real-time correction on simulated water levels of the whole basin to obtain corrected flow data, and inputting the corrected flow data into a CKLSTM deep prediction network to output reservoir inflow flood flow and risk early warning results, and the early warning response time is significantly shortened.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering monitoring technology, and in particular to a method and system for water level data acquisition and analysis. Background Technology

[0002] Water level data is a core fundamental parameter for hydrological monitoring, flood forecasting, water resource allocation, and dam safety operation. Accurate, continuous, and reliable water level acquisition and analysis methods are of great significance for flood control and disaster reduction, optimal allocation of water resources, and safety management of water conservancy projects.

[0003] Traditional water level monitoring methods are mainly divided into traditional contact sensor monitoring and radar level gauge and remote sensing monitoring. Traditional contact sensor devices such as float-type water level gauges, pressure water level gauges, and ultrasonic water level gauges typically use a single-point measurement method, that is, a sensor is placed at a fixed position on the monitoring section, and the water level at that point is used to represent the water level of the entire section. However, under complex flow conditions such as irregular cross-sectional shape, rapid water flow, and backwater, single-point measurement cannot accurately reflect the average water level of the section, resulting in insufficient representativeness. In addition, pressure water level gauges are susceptible to changes in water temperature, siltation, and sensor drift, resulting in poor long-term stability; ultrasonic water level gauges are easily affected by wind speed, air stratification, and floating debris on the water surface.

[0004] Radar level gauges and satellite altimeters are used for remote sensing monitoring of water levels. Radar level gauges use non-contact measurement, which avoids water corrosion and siltation problems. However, their measurement accuracy is greatly affected by the radar beam angle, water surface waves, and installation height, and they are also costly. Satellite altimeters can obtain water level information over a wide area, but their spatiotemporal resolution is low, making it difficult to meet the high-frequency data requirements of real-time flood forecasting. In addition, existing technologies mainly use hydrodynamic models to simulate and extrapolate water level and flow. Although pure physical models have clear physical mechanisms, their accuracy is heavily dependent on the accuracy of input parameters such as river roughness and cross-sectional geometric parameters. In actual rivers, roughness varies with water level, flow regime, and riverbed evolution. Physical models with fixed parameters often produce systematic biases. On the other hand, pure data-driven deep learning models have strong nonlinear fitting capabilities, but lack physical constraints. Under extreme conditions, they may output prediction results that do not conform to physical laws, posing an extrapolation risk. Summary of the Invention

[0005] The technical problem solved by this invention is that: existing water level sensors mostly adopt single-point measurement methods, which are difficult to characterize the real water level distribution under complex cross-sectional morphology; fiber optic sensing technology is difficult to effectively separate the coupling effect of temperature and strain in water level monitoring; and physical hydrodynamic models are limited by parameter uncertainties and have systematic biases.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a method for acquiring and analyzing water level data includes the following steps: Step S1: Collect water level fiber optic data through a water pressure sensitive deformation rod. The water level fiber optic data includes a first fiber optic sequence and a second fiber optic sequence. Step S2: Temperature deviation processing is performed on the water level fiber optic data using a dual-mode differential algorithm to obtain the net strain sequence; Step S3: Obtain historical hydrostatic pressure data, perform spatial derivative analysis mapping on the net strain sequence, and obtain actual water level data; Step S4: Based on the principle of hydrodynamic phase balance alignment, the actual water level data is reconstructed by singular value decomposition using a comprehensive reconstruction network model to obtain the water level reference sequence. Step S5: Construct a hydrodynamic model based on the water level reference sequence to obtain the simulated water level error. Construct a correction model to perform real-time post-correction on the simulated water level of the entire basin to obtain the corrected flow data. Input the corrected water level data and the corrected flow data of the entire basin into the CKLSTM network to output the inflow flood flow and risk warning results.

[0007] Preferably, step S1 includes the following sub-steps: Step S11: Using the bottom fixed point of the water pressure sensitive deformation rod as the coordinate origin 0, establish a one-dimensional spatial elevation coordinate axis z along the upward elevation direction of the water pressure sensitive deformation rod; let the total length of the water pressure sensitive deformation rod be L, then the monitoring space domain is defined as z∈[0,L]; The first and second optical fibers are simultaneously laid on the water pressure sensitive deformation rod; The first optical fiber is firmly bonded to the pressure-bearing surface of the water pressure-sensitive deformation rod along the direction of the spatial elevation coordinate axis z, so that the deformation of the first optical fiber is completely coupled with the deformation of the water pressure-sensitive deformation rod; the second optical fiber is stress-free inserted into the capillary tube fixed on the surface of the water pressure-sensitive deformation rod, so that the second optical fiber is not constrained by the mechanical deformation of the water pressure-sensitive deformation rod, and the second optical fiber only exchanges heat with the external environment. Step S12: Based on the backscattered Rayleigh signal, obtain the first Rayleigh scattering spectrum frequency shift corresponding to the first optical fiber and the second Rayleigh scattering spectrum frequency shift corresponding to the second optical fiber. The first Rayleigh scattering spectrum frequency shift satisfies the first physical response formula: ; The second Rayleigh scattering spectrum frequency shift satisfies the second physical response formula: ; in, The first Rayleigh scattering spectrum frequency shift, This is the frequency shift of the second Rayleigh scattering spectrum. Let be the coordinates of the i-th discrete spatial sampling point along the elevation direction on the z-axis of the spatial elevation coordinate system, where i = 1, 2, ..., N, N is the total number of discrete spatial sampling points, and t is the current data acquisition time. The strain sensitivity coefficients of the first and second optical fibers are denoted as . The temperature sensitivity coefficients of the first and second optical fibers; The actual mechanical strain transmitted to the first optical fiber is caused by the bending of the water pressure-sensitive deformation rod due to water pressure. This represents the change in ambient temperature at the current data acquisition time t relative to the initial calibration time. Step S13: At the current data acquisition time t, spatially discretize and package all the acquired first Rayleigh scattering spectrum frequency shifts and all the second Rayleigh scattering spectrum frequency shifts to generate a first fiber sequence and a second fiber sequence. By matching the first fiber sequence with the second fiber sequence one-to-one, water level fiber data is obtained.

[0008] Preferably, step S2 includes the following sub-steps: Step S21: Extract the first fiber sequence and the second fiber sequence from the water level fiber data. The first fiber sequence includes the first Rayleigh scattering spectral frequency shift at each discrete spatial sampling point, and the second fiber sequence includes the second Rayleigh scattering spectral frequency shift at each discrete spatial sampling point. Based on the spatial elevation coordinate axis z, according to the division position of the discrete spatial sampling points, match the first Rayleigh scattering spectral frequency shift of the i-th sampling point in the first fiber sequence with the second Rayleigh scattering spectral frequency shift of the i-th sampling point in the second fiber sequence one by one to obtain the discrete spatial sampling point set. Step S22: Subtract the frequency shift of the second Rayleigh scattering spectrum corresponding to the discrete spatial sampling point from the frequency shift of the first Rayleigh scattering spectrum corresponding to the discrete spatial sampling point to obtain the differential frequency shift. Then divide the differential frequency shift by the strain sensitivity coefficient to obtain the true mechanical strain of the water pressure sensitive deformation rod at the discrete spatial sampling point. Step S23: Discretize and arrange all the real mechanical strains according to the elevation direction coordinates of the spatial elevation coordinate axis z to generate the net strain sequence.

[0009] Preferably, step S3 includes the following sub-steps: Step S31: Obtain historical hydrostatic pressure data of the target monitoring section and stiffness gradient compensation factor corresponding to the water pressure sensitive deformation rod with gradient stiffness along the elevation direction from the database; set the derivative mutation threshold for identifying the water surface interface position; calculate the spatial derivative of the net strain sequence along the elevation direction coordinate order of the spatial elevation coordinate axis z to obtain the initial numerical change rate. Step S32: Multiply the stiffness gradient compensation factor by the initial numerical rate of change, and physically amplify the strain mutation signal at the water surface interface in the top low stiffness region to obtain the corrected spatial derivative sequence. Step S33: Traverse the corrected spatial derivative sequence, find the target discrete spatial sampling point where the absolute value of the numerical rate of change first exceeds the derivative mutation threshold, and mark the target discrete spatial sampling point as the critical strain inflection point. Step S34: Extract the specific elevation coordinate values ​​corresponding to the marked critical strain inflection points from the spatial elevation coordinate axis z, and map the specific elevation coordinate values ​​to the actual water surface elevation of the current monitoring section and output them as the actual water level data at the current moment.

[0010] Preferably, step S4 specifically includes: Obtain the original water level time series set from multiple cross-sections, calculate the water level fluctuation variance, and determine that the original water level time series set from multiple cross-sections is in a period of severe fluctuation when the water level fluctuation variance is greater than or equal to a preset fluctuation variance threshold, and trigger a singular value decomposition reconstruction mechanism based on hydrodynamic phase translation alignment. When the original water level time series set of multiple cross sections is in a period of violent fluctuation, the spatial channel distance between two adjacent monitoring sections is obtained, and the average water depth and average flow velocity of the channel corresponding to the spatial channel distance are obtained. According to the theory of shallow water wave dynamics, the average water depth of the river channel is multiplied by the gravitational acceleration and the square root is taken. The result of the square root calculation is added to the average flow velocity to calculate the propagation speed of shallow water waves. Divide the spatial river channel distance by the shallow water wave propagation speed to obtain the propagation time delay corresponding to the evolution of flood waves between two adjacent monitoring sections; Using the time coordinates of the upstream monitoring section in the multi-section original water level time series set as a reference, and utilizing the propagation time delay, the data sequences corresponding to the downstream monitoring sections in the multi-section original water level time series set are shifted forward in time phase to obtain the aligned multi-section water level time series set.

[0011] Preferably, step S4 further includes: The aligned multi-section water level time series set is transformed into a two-dimensional matrix form to construct a topological fluctuation matrix M. This topological fluctuation matrix is ​​then input into the singular value decomposition (SVD) reconstruction model to perform matrix decomposition, yielding the singular value decomposition formula: ; in, It is a left singular vector matrix. It is the transpose of the right singular vector matrix. This is a diagonal matrix containing multiple singular values, where the singular values ​​are arranged in descending order of their numerical values ​​along the main diagonal of the matrix. A singular value truncation threshold is preset. Multiple singular values ​​on the main diagonal of the diagonal matrix are traversed. Each singular value is compared with the singular value truncation threshold. Singular values ​​that are greater than or equal to the singular value truncation threshold are retained as primary singular values. Singular values ​​smaller than the singular value truncation threshold are replaced with zero; a denoising diagonal matrix is ​​constructed using the retained principal singular values ​​and the singular values ​​replaced with zero. The denoising reconstruction matrix is ​​obtained by performing matrix multiplication using the transpose of the left singular vector matrix, the denoising diagonal matrix, and the right singular vector matrix. Satisfying the reduction formula ; Extract the data sequence from the noise reduction and reconstruction matrix to complete the missing time point data; Based on the propagation time delay, the extracted data sequence is reverse-shifted backward in time phase to restore the time phase of the data sequence corresponding to the downstream monitoring section to the original real physical time axis, thus obtaining the wave reconstructed water level sequence. The fluctuating reconstructed water level sequence is output as the water level reference sequence.

[0012] Preferably, step S5 specifically includes: Obtain basic geographic parameters for the entire watershed and measured upstream flow data; Among them, the basic geographic parameters of the entire basin include channel roughness, cross-sectional geometry, riverbed slope and channel connectivity topology, and the water level reference sequence is a high-precision water level sequence at one or more control monitoring sections; The basic geographic parameters of the entire basin, upstream measured flow data, and boundary verification information corresponding to the water level benchmark sequence are input into a pre-set one-dimensional hydrodynamic simulation model for numerical solution, outputting simulated water level data for the entire basin. and simulated traffic data ; A set of control sections with water level reference sequences The simulated water level data for the entire basin is subtracted from the water level reference sequence to obtain the simulated water level error at the control section. : ; in, For water level reference sequence, To simulate water level error, at the upstream flow observation section At this point, the upstream measured flow rate data is subtracted from the simulated flow rate at the upstream cross-section output by the model to obtain the upstream flow rate error. The upstream flow error is mapped and correlated with the simulated water level error at different control sections to construct a linear regression correction model that considers propagation time delay across different sections. Specifically, the propagation time delay is calculated based on the river hydrodynamic propagation characteristics. Establish a linear regression correction model : ; in, and To control the cross-section The corresponding error correction coefficients, the output of the linear regression correction model The predicted water level error is used for subsequent additive compensation of the simulated water level. The error correction coefficients obtained from multiple control sections are interpolated or piecewise assigned along the spatial direction of the river channel to obtain the error correction coefficient field for the entire watershed. And obtain the water level error estimate for the entire basin. : ; A scaling correction model is constructed to perform amplitude consistency correction on the simulated flow across the entire watershed, and the flow scaling factor is calculated. : ; Where ϵ is a very small positive number to prevent the denominator from being zero, and the corrected flow data is obtained by proportionally correcting the simulated flow based on the flow scaling factor: ; To correct the flow rate data, real-time post-correction compensation was performed on the simulated water level data of the entire basin using error correction coefficients. The corrected water level data of the entire basin was obtained by employing an additive error stripping method. ; in, This provides water level data for the entire basin. Historical meteorological data of the target water area is obtained. The corrected whole basin water level data, corrected flow data and historical meteorological data are input into the pre-trained CKLSTM deep prediction network. Multi-dimensional time series feature extraction and inference are performed through the CKLSTM deep prediction network to predict and output the inflow flood flow data and water level evolution trend data for future periods. By using inflow flood data and water level evolution trend data for future periods, we can construct and solve the non-steady flow surface differential equation to calculate the dynamic additional reservoir capacity data generated by the slope of the reservoir water surface. The dynamic additional storage capacity data is accumulated and calculated with the static storage capacity curve data to generate real-time dynamic storage capacity data; Real-time dynamic reservoir capacity data is substituted into the reservoir water balance equation to perform flood regulation calculations and estimate the highest water level in front of the dam. Based on the highest water level in front of the dam and the dam crest elevation data, the probability of dam flood overtopping risk is calculated. The probability of dam flood overtopping risk is input into the risk rule engine for interval matching and comparison to determine the early warning response level and generate risk early warning results.

[0013] Preferably, the processing logic of the CKLSTM deep prediction network that incorporates physical priors specifically includes: Following a unified time step, the acquired corrected whole-basin water level data, corrected flow data, and historical meteorological data were aligned and normalized according to the time dimension. The aligned and normalized corrected whole-basin water level data, corrected flow data and historical meteorological data are spliced ​​together to construct a multi-dimensional time-series input matrix. The multidimensional temporal input matrix is ​​input into the CKLSTM deep prediction network. The sliding convolution kernel is used to perform sliding window convolution operation on the time step dimension of the multidimensional temporal input matrix to extract the local spatial cross features between the corrected whole basin water level data, the corrected flow data and historical meteorological data, and output the local cross feature sequence. The local cross-feature sequence is input into the Kolmogorov-Arnold network of the CKLSTM deep prediction network. The learnable univariate activation function deployed on the network edge is used to perform high-dimensional nonlinear feature space transformation on the local cross-feature sequence to eliminate non-stationary noise interference in hydrological and meteorological data and output nonlinear mapping feature sequence. The nonlinear mapping feature sequence is input into the long short-term memory network of the CKLSTM deep prediction network. The historical hydrological evolution pattern in the nonlinear mapping feature sequence is calculated by long-term memory and short-term update, global temporal dependency features are extracted, and temporal hidden state vector is output. The temporal hidden state vector is input into the fully connected decoding layer of the CKLSTM deep prediction network; the fully connected decoding layer inversely maps the temporal hidden state vector to the hydrological physical dimension space, and predicts and outputs the inflow flood flow data and water level evolution trend data for future periods.

[0014] Preferably, the processing logic of the risk rule engine is as follows: The preset risk probability interval thresholds include a first threshold, a second threshold, and a third threshold, wherein the first threshold is less than the second threshold and less than the third threshold, and the warning response levels include blue warning, yellow warning, orange warning, and red warning. The calculated probability of dam flood overtopping is compared with the risk probability interval threshold: The probability of dam overflowing by floodwaters When the first threshold is reached, it is determined that the current state is under blue alert; When the first threshold dam flood overtopping risk probability When the second threshold is reached, it is determined that the current situation is under a yellow alert; When the second threshold dam flood overtopping risk probability When the third threshold is reached, it is determined that the current state is under orange alert; The probability of dam overflowing by floodwaters When the third threshold is reached, it is determined that the current situation is under a red alert.

[0015] Secondly, a water level data acquisition and analysis system includes a data acquisition module, a net strain sequence extraction module, a water level inversion module, a water level data reconstruction module, and an error correction and risk prediction module. The data acquisition module is used to collect water level optical fiber data by vertically deploying water pressure sensitive deformation rods at the monitoring section. The water level optical fiber data includes a first optical fiber sequence and a second optical fiber sequence. The net strain sequence extraction module is used to process the temperature deviation of the water level fiber optic data using a dual-mode differential algorithm to obtain the net strain sequence. The water level inversion module is used to acquire historical hydrostatic pressure data, perform spatial derivative analysis and mapping on the net strain sequence, and obtain actual water level data. The water level data reconstruction module is used to reconstruct the actual water level data by singular value decomposition using a comprehensive reconstruction network model to obtain a water level benchmark sequence. The error correction and risk prediction module is used to construct a hydrodynamic model based on the water level reference sequence, obtain the simulated water level error, construct a linear regression and scaling correction model, perform real-time post-correction on the simulated water level of the entire basin, obtain the corrected flow data, and input the corrected flow data into the CKLSTM deep prediction network to predict and output the inflow flood flow and risk warning results.

[0016] The beneficial effects of the present invention are as follows: By vertically arranging water pressure sensitive deformation rods at the monitoring section and continuously arranging distributed optical fibers along the elevation direction, the present invention can obtain the continuous strain distribution within the entire water depth range. By identifying the critical strain inflection point through spatial derivative analysis, the true water surface elevation can be directly derived. This method is not limited by the cross-sectional shape and can accurately reflect the actual water level under complex flow conditions, significantly improving the representativeness of water level data. Secondly, the dual-mode differential algorithm completely eliminates temperature interference and improves long-term measurement stability. The invention simultaneously deploys a first optical fiber that is fully coupled to the pole and a second optical fiber that is not mechanically constrained. The first optical fiber senses the coupling response of strain and temperature, while the second optical fiber independently senses changes in ambient temperature. Through the dual-mode differential algorithm, the first optical fiber sequence and the second optical fiber sequence are subtracted point by point to accurately separate the real mechanical strain caused by water pressure, eliminating the influence of temperature drift on measurement accuracy. This allows the system to maintain high-precision and stable operation even in environments with large day-night temperature differences and significant seasonal changes. Thirdly, an integrated reconstruction network was used to achieve adaptive data denoising and repair, improving data quality. Based on the fluctuation characteristics of water level time series, this invention dynamically selects either a deep temporal reconstruction network or a singular value decomposition (SVD) reconstruction model for data processing. During stable periods, a deep temporal reconstruction network is used for feature extraction and missing data generation, maintaining data smoothness. During fluctuating periods, singular value decomposition and hard thresholding effectively filter out high-frequency noise while preserving the steep characteristics of flood rise and fall. This adaptive reconstruction strategy takes into account the data quality requirements under different hydrological conditions. Fourthly, this invention uses the high-precision water level reference sequence retrieved from optical fiber as the constraint condition for the physical model input. The one-dimensional hydrodynamic simulation results are post-corrected in real time through linear regression and scaling correction models, effectively eliminating the systematic biases introduced by factors such as fixed roughness and equation discretization. On this basis, the corrected high-fidelity data is input into the CKLSTM deep prediction network, making full use of the ability of deep learning to extract nonlinear time series features. This not only preserves the physical interpretability of the hydrodynamic model, but also leverages the high-precision fitting advantage of the data-driven model. Finally, the CKLSTM deep prediction network in this invention integrates a one-dimensional convolutional feature extraction module, KAN, and a long short-term memory network LSTM. It can simultaneously extract local spatial cross features, high-dimensional nonlinear features, and global temporal dependency features between water level, flow rate, and meteorological data, achieving deep fusion of multi-dimensional information. The prediction results are further input into a dam flood overtopping risk extrapolation model that considers dynamic reservoir capacity. Combined with a multi-level risk rule engine, it automatically generates graded early warning instructions. This process realizes full-chain automation from data collection to risk early warning, significantly shortening the early warning response time. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the steps of a water level data acquisition and analysis method according to an embodiment of the present invention; Figure 2 This is a basic flowchart of a water level data acquisition and analysis system provided in one embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1 This paper provides a method for water level data acquisition and analysis, including the following steps: Step S1: Collect water level fiber optic data through the water pressure sensitive deformation rod. The water level fiber optic data includes a first fiber optic sequence and a second fiber optic sequence. Step S2: Temperature deviation processing is performed on the water level fiber optic data using a dual-mode differential algorithm to obtain the net strain sequence; Step S3: Obtain historical hydrostatic pressure data, perform spatial derivative analysis mapping on the net strain sequence, and obtain actual water level data; Step S4: Based on the principle of hydrodynamic phase balance alignment, the actual water level data is reconstructed by singular value decomposition using a comprehensive reconstruction network model to obtain the water level reference sequence. Step S5: Construct a hydrodynamic model based on the water level reference sequence to obtain the simulated water level error. Construct a correction model to perform real-time post-correction on the simulated water level of the entire basin to obtain the corrected flow data. Input the corrected water level data and the corrected flow data of the entire basin into the CKLSTM network to output the inflow flood flow and risk warning results.

[0020] This embodiment achieves continuous distributed water level sensing across the entire cross-section through vertically deployed water pressure-sensitive deformation rods and a dual-fiber structure, overcoming the problem of insufficient representativeness of traditional single-point sensors. The dual-mode differential algorithm effectively eliminates the impact of temperature drift on measurement accuracy, improving data reliability. Furthermore, by combining spatial derivative analysis mapping and a comprehensive reconstruction network, automatic conversion from raw strain signals to high-precision water level reference sequences is achieved. Finally, through a cascaded architecture of a hydrodynamic model and a CKLSTM deep prediction network, under the dual constraints of physical mechanisms and data-driven approaches, the accuracy of inflow flood flow prediction and the timeliness of risk warning are significantly improved, providing reliable technical support for flood control scheduling in water conservancy projects.

[0021] Step S1 includes the following sub-steps: Step S11: Take the bottom fixed point of the water pressure sensitive deformation rod as the coordinate origin 0, and establish a one-dimensional spatial elevation coordinate axis z along the upward elevation direction of the water pressure sensitive deformation rod; let the total length of the water pressure sensitive deformation rod be L, then the monitoring space domain is defined as z∈[0,L]; A first optical fiber and a second optical fiber are simultaneously laid on the water pressure sensitive deformation rod; the first optical fiber is firmly attached to the pressure-bearing surface of the water pressure sensitive deformation rod along the direction of the spatial elevation coordinate axis z, so that the deformation of the first optical fiber is completely coupled with the deformation of the water pressure sensitive deformation rod; the second optical fiber is stress-free inserted into a capillary tube fixed on the surface of the water pressure sensitive deformation rod, so that the second optical fiber is not constrained by the mechanical deformation of the water pressure sensitive deformation rod, and the second optical fiber only exchanges heat with the external environment; Step S12: Based on the backscattered Rayleigh signal, obtain the first Rayleigh scattering spectrum frequency shift corresponding to the first optical fiber and the second Rayleigh scattering spectrum frequency shift corresponding to the second optical fiber. The frequency shift of the first Rayleigh scattering spectrum satisfies the first physical response formula: ; The frequency shift of the second Rayleigh scattering spectrum satisfies the second physical response formula: ; in, The first Rayleigh scattering spectrum frequency shift, This is the frequency shift of the second Rayleigh scattering spectrum. Let be the coordinates of the i-th discrete spatial sampling point along the elevation direction on the z-axis, i = 1, 2, ..., N, where N is the total number of discrete spatial sampling points, and t is the current data acquisition time. This is the strain sensitivity coefficient; This is the temperature sensitivity coefficient; The actual mechanical strain transmitted to the first optical fiber is caused by the bending of the water pressure-sensitive deformation rod due to water pressure. This represents the change in ambient temperature at the current data acquisition time t relative to the initial calibration time. Step S13: At the current data acquisition time t, spatially discretize and package all the acquired first Rayleigh scattering spectrum frequency shifts and all the second Rayleigh scattering spectrum frequency shifts to generate a first fiber sequence and a second fiber sequence. By matching the first fiber sequence with the second fiber sequence one-to-one, water level fiber data is obtained.

[0022] At the upstream monitoring section of a reservoir dam, a water pressure sensitive deformation rod with a total length of L=15 meters is installed; with the fixed point at the bottom of the rod as the coordinate origin 0, a spatial elevation coordinate axis z is established along the rod upwards, and the monitoring spatial domain is defined as z∈[0,15]. The first optical fiber is firmly attached to the pressure surface of the rod along the z direction (closely attached), and a capillary tube is fixed on the surface of the rod. The second optical fiber is stress-free inserted into the capillary tube so that it is only affected by the ambient temperature and not by the deformation of the rod. Using a sampling period of 1 second and a spatial sampling interval of 0.1 meters along the elevation direction, the first and second Rayleigh scattering spectral frequency shifts at the i-th discrete spatial sampling point are obtained. The strain sensitivity coefficients of the first and second optical fibers are 0.78 GHz / At 1.2 GHz / ℃, the mechanical strain and temperature change at each sampling point are calculated in real time; Pack the first Rayleigh scattering spectral frequency shifts at all discrete sampling points into a first fiber sequence, pack the second Rayleigh scattering spectral frequency shifts into a second fiber sequence, and then match the two sets of sequences one-to-one according to spatial coordinates to generate complete water level fiber data, which is then stored in the local cache. This embodiment achieves physical separation of strain and temperature signals by simultaneously deploying a first optical fiber and a second optical fiber on a water pressure-sensitive deformation rod. The first optical fiber is fully coupled to the rod body to sense the strain caused by water pressure, while the second optical fiber is deployed without stress to independently sense the ambient temperature. Combined with the physical response formula of the backscattering Rayleigh spectral frequency shift, it can accurately characterize the contribution of strain and temperature to the frequency shift at each spatial sampling point, providing an accurate physical basis for subsequent temperature compensation. This deployment structure and signal acquisition method are simple and reliable, and are suitable for large-scale continuous water level monitoring in long-term field environments.

[0023] Step S2 includes the following sub-steps: Step S21: Extract the first fiber sequence and the second fiber sequence from the water level fiber data. The first fiber sequence includes the first Rayleigh scattering spectral frequency shift at each discrete spatial sampling point, and the second fiber sequence includes the second Rayleigh scattering spectral frequency shift at each discrete spatial sampling point. Based on the spatial elevation coordinate axis z, according to the division position of the discrete spatial sampling points, match the first Rayleigh scattering spectral frequency shift of the i-th sampling point in the first fiber sequence with the second Rayleigh scattering spectral frequency shift of the i-th sampling point in the second fiber sequence one by one to obtain the discrete spatial sampling point set. Step S22: Subtract the frequency shift of the second Rayleigh scattering spectrum corresponding to the discrete spatial sampling point from the frequency shift of the first Rayleigh scattering spectrum corresponding to the discrete spatial sampling point to obtain the differential frequency shift. Then divide the differential frequency shift by the strain sensitivity coefficient to obtain the true mechanical strain of the water pressure sensitive deformation rod at the discrete spatial sampling point. Step S23: Discretize and arrange all the real mechanical strains according to the elevation direction coordinates of the spatial elevation coordinate axis z to generate the net strain sequence.

[0024] The first and second fiber sequences are extracted from the water level fiber data. Using the spatial elevation coordinate axis z as a reference, and with a discrete sampling interval of 0.1 meters, the Rayleigh scattering spectrum frequency shift of the i-th sampling point in the first fiber sequence is paired with the Rayleigh scattering spectrum frequency shift of the same sampling point in the second fiber sequence. The calculated real mechanical strains at all discrete sampling points are arranged in ascending order of elevation coordinates to generate a net strain sequence, which is then stored in a database for subsequent analysis. This embodiment employs a dual-mode differential algorithm to perform point-by-point difference calculations between the first and second fiber sequences in the spatial domain, and combines this with the strain sensitivity coefficient. The algorithm completes dimensionally consistent strain transformation calculations, directly eliminating the influence of temperature changes on strain measurement and obtaining a net strain sequence that is only related to water pressure. This algorithm does not require a complex temperature compensation model, has high computational efficiency and strong real-time performance, and can be run directly on field monitoring terminals. At the same time, the differential operation preserves the spatial distribution characteristics of strain along the elevation direction, providing a high-quality data foundation for the subsequent accurate identification of water surface location.

[0025] Step S3 includes the following sub-steps: Step S31: Obtain historical hydrostatic pressure data of the target monitoring section and stiffness gradient compensation factor corresponding to the water pressure sensitive deformation rod with gradient stiffness along the elevation direction from the database; set the derivative mutation threshold for identifying the water surface interface position; calculate the spatial derivative of the net strain sequence along the elevation direction coordinate order of the spatial elevation coordinate axis z to obtain the initial numerical change rate. Step S32: Multiply the stiffness gradient compensation factor by the initial numerical rate of change, and physically amplify the strain mutation signal at the water surface interface in the top low stiffness region to obtain the corrected spatial derivative sequence. Step S33: Along the spatial elevation coordinate axis z, traverse the corrected spatial derivative sequence from the top to the bottom of the rod, find the target discrete spatial sampling point where the numerical rate of change suddenly changes from zero to non-zero and the absolute value exceeds the derivative sudden change threshold for the first time, and mark the target discrete spatial sampling point as the critical strain inflection point. Step S34: Extract the specific elevation coordinate values ​​corresponding to the marked critical strain inflection points from the spatial elevation coordinate axis z, and map the specific elevation coordinate values ​​to the actual water surface elevation of the current monitoring section and output them as the actual water level data at the current moment.

[0026] Retrieve historical hydrostatic pressure data of the target monitoring section from the database for the past three years, and obtain the stiffness gradient compensation factor corresponding to the water pressure sensitive deformation rod with gradient stiffness along the elevation direction. Based on the historical fluctuation patterns of hydrostatic pressure data, a threshold for identifying derivative abrupt changes at the water surface interface is set as follows: Along the elevation direction coordinate sequence of the spatial elevation coordinate axis z, and according to the spatial sampling interval of 0.1 meters, the spatial derivative of the real mechanical strain between adjacent discrete spatial sampling points in the net strain sequence is calculated to obtain the initial numerical rate of change corresponding to each discrete spatial sampling point. Multiply the stiffness gradient compensation factor by the initial numerical rate of change. For example, at a discrete spatial sampling point of z = 8.5 meters, since it is in the top low-stiffness region, the stiffness gradient compensation factor is preset to 2.5. The calculated initial numerical rate of change (assuming it is...) Multiply by the stiffness gradient compensation factor of 2.5, and physically amplify the strain abrupt change signal at the water surface interface in the top low stiffness region through the multiplication operation to obtain the corrected spatial derivative sequence. Traverse the corrected spatial derivative sequence, compare the absolute value of the numerical rate of change in the corrected spatial derivative sequence with the derivative mutation threshold one by one, and find the target discrete spatial sampling point where the absolute value of the numerical rate of change first exceeds the derivative mutation threshold. When the absolute value of the numerical rate of change corresponding to the target discrete spatial sampling point at z=8.5 meters first exceeds the derivative mutation threshold, mark the target discrete spatial sampling point at z=8.5 meters as the critical strain inflection point. From the spatial elevation coordinate axis z, the specific elevation coordinate value of 8.5 meters corresponding to the marked critical strain inflection point is extracted, and the specific elevation coordinate value of 8.5 meters is equivalently mapped to the real water surface elevation of the current monitoring section. The real water surface elevation of 8.5 meters is then output to the system server as the actual water level data at the current moment for storage and subsequent retrieval. This embodiment completely breaks through the technical bottleneck of low signal-to-noise ratio caused by the extremely small force on traditional constant cross-section measuring rods in low water levels or shallow water areas by introducing a water pressure sensitive deformation rod with gradient variable stiffness design and stiffness gradient compensation factor. By physically amplifying the tiny strain change signal at the water surface interface through multiplication operation, and combining the spatial derivative analysis mapping of derivative change threshold, the complex physical force conversion problem is cleverly transformed into a highly robust boundary feature identification problem. The inversion mechanism is extremely immune to long-term aging and stiffness degradation of water pressure sensitive deformation rod materials, effectively avoiding the systematic cumulative error caused by traditional water level calculation based on absolute strain values.

[0027] Step S4 specifically includes: Obtain the original water level time series set from multiple cross-sections, calculate the water level fluctuation variance, and determine that the original water level time series set from multiple cross-sections is in a period of severe fluctuation when the water level fluctuation variance is greater than or equal to a preset fluctuation variance threshold, and trigger a singular value decomposition reconstruction mechanism based on hydrodynamic phase translation alignment. When the original water level time series set of multiple cross sections is in a period of violent fluctuation, the spatial channel distance between two adjacent monitoring sections is obtained, and the average water depth and average flow velocity of the channel corresponding to the spatial channel distance are obtained. According to the theory of shallow water wave dynamics, the average water depth of the river channel is multiplied by the gravitational acceleration and the square root is taken. The result of the square root calculation is added to the average flow velocity to calculate the propagation speed of shallow water waves. Divide the spatial river channel distance by the shallow water wave propagation speed to obtain the propagation time delay corresponding to the evolution of flood waves between two adjacent monitoring sections; Using the time coordinates of the upstream monitoring section in the multi-section original water level time series set as a reference, and utilizing the propagation time delay, the data sequences corresponding to the downstream monitoring sections in the multi-section original water level time series set are shifted forward in time phase to obtain the aligned multi-section water level time series set.

[0028] In this embodiment, the actual water level data of multiple monitoring sections within the target watershed during the historical monitoring period are obtained, and the data are combined to generate a multi-section original water level time series set. The water level fluctuation variance of the multi-section original water level time series set is calculated, and a preset fluctuation variance threshold of 1.0 square meters is set. The water level fluctuation variance is then compared with the preset fluctuation variance threshold of 1.0 square meters. When encountering extreme conditions such as torrential rains during the flood season, if the variance of water level fluctuation is greater than or equal to the preset fluctuation variance threshold of 1.0 square meters, the original water level time series set of multiple sections is determined to be in a period of severe fluctuation, and a singular value decomposition reconstruction mechanism based on hydrodynamic phase translation alignment is automatically triggered: The spatial distance between two adjacent monitoring sections is obtained among multiple monitoring sections, and the average water depth and average flow velocity of the river corresponding to the spatial water distance are obtained simultaneously. According to the shallow water wave dynamics theory, the average water depth of the river is multiplied by the gravitational acceleration and the square root is calculated to obtain the result of the square root calculation. The result of the square root calculation is added to the average flow velocity to calculate the shallow water wave propagation speed. Dividing the spatial river channel distance by the shallow water wave propagation speed yields the propagation time delay corresponding to the evolution of flood waves between two adjacent monitoring sections. Using the time coordinate of the upstream monitoring section in the multi-section original water level time series set as the benchmark, the data sequence corresponding to the downstream monitoring section in the multi-section original water level time series set is forward-shifted in time phase using the propagation time delay. That is, the data sequence corresponding to the downstream monitoring section is propagated forward by the time delay on the time axis. Through the forward-shifting operation of time phase, the temporal misalignment deviation caused by the evolution of flood waves from the upstream monitoring section to the downstream monitoring section in the spatial river channel is eliminated, so that the flood peak characteristics corresponding to all monitoring sections are physically aligned in a unified time dimension. Finally, the aligned multi-section water level time series set is obtained, and the aligned multi-section water level time series set is output to the integrated reconstruction network model for subsequent matrix construction. This embodiment overcomes the blind spots of traditional pure mathematical interpolation algorithms or static matrix reconstruction methods when processing spatial multi-node hydrological time series data. By introducing a preset fluctuation variance threshold for state determination, it can adaptively identify extreme flood season conditions. By introducing shallow water wave dynamics theory to calculate the propagation speed and propagation time delay of shallow water waves, and using the propagation time delay to perform a forward translation operation on the time phase of the data sequence corresponding to the monitoring section downstream, it completely eliminates the time series misalignment deviation caused by flood waves propagating along the river channel from the perspective of fluid dynamics physical evolution mechanism. Compared with conventional data sequences that have not undergone phase alignment processing, the aligned multi-section water level time series set perfectly aggregates the spatially dispersed main features of flood peaks at the physical level in a unified time dimension, which greatly improves the accuracy of hydrological data reconstruction and waveform restoration in high missing rate datasets.

[0029] Step S4 also includes: The aligned multi-section water level time series set is transformed into a two-dimensional matrix form to construct the topological fluctuation matrix M. This topological fluctuation matrix is ​​then input into the singular value decomposition (SVD) reconstruction model to perform matrix decomposition, yielding the SVD formula: ; in, It is a left singular vector matrix. It is the transpose of the right singular vector matrix. This is a diagonal matrix containing multiple singular values, where the singular values ​​are arranged in descending order of their numerical values ​​along the main diagonal of the matrix. A singular value truncation threshold is preset. Multiple singular values ​​on the main diagonal of the diagonal matrix are traversed. Each singular value is compared with the singular value truncation threshold. Singular values ​​that are greater than or equal to the singular value truncation threshold are retained as primary singular values. Singular values ​​smaller than the singular value truncation threshold are replaced with zero; a denoising diagonal matrix is ​​constructed using the retained principal singular values ​​and the singular values ​​replaced with zero. The denoising reconstruction matrix is ​​obtained by performing matrix multiplication using the transpose of the left singular vector matrix, the denoising diagonal matrix, and the right singular vector matrix. Satisfying the reduction formula ; Extract the data sequence from the noise reduction and reconstruction matrix to complete the missing time point data; Based on the propagation time delay, the extracted data sequence is reverse-shifted backward in time phase to restore the time phase of the data sequence corresponding to the downstream monitoring section to the original real physical time axis, thus obtaining the wave reconstructed water level sequence. The fluctuating reconstructed water level sequence is output as the water level reference sequence.

[0030] In this embodiment, the aligned multi-section water level time series set after the time phase forward translation operation is obtained, and the aligned multi-section water level time series set is converted into a two-dimensional matrix form to construct the topological wave matrix M. The correct data in the topological fluctuation matrix are labeled, a mask matrix for missing locations is generated, and linear interpolation at adjacent time points is used to initially fill the correct locations, resulting in an initial filling matrix. The non-missing values ​​in the initial filling matrix are consistent with those in the topological fluctuation matrix, while the missing locations are filled with the results of linear interpolation. The initial filling matrix is ​​used as the starting point for the iteration in the singular value decomposition reconstruction model to perform an iterative low-rank approximation process. In each iteration, singular value decomposition is performed on the current matrix. The missing positions of the current matrix are updated with the values ​​corresponding to the missing positions in the reconstructed matrix, while keeping the original observations of the non-missing positions unchanged. The iteration is repeated until the norm of the difference between the reconstructed matrices of two adjacent iterations is less than the singular value truncation threshold, thus obtaining the singular value decomposition formula. A singular value truncation threshold of 2.5 is preset. Multiple singular values ​​on the main diagonal of the diagonal matrix are traversed, and each singular value is compared with the preset singular value truncation threshold. Singular values ​​that are greater than or equal to the singular value truncation threshold of 2.5 are retained as the principal singular values ​​representing the physical characteristics of the true flood peak principal components. Singular values ​​that are less than the singular value truncation threshold of 2.5 are directly replaced with zero. The singular values ​​replaced with zero represent the abnormal noise characteristics caused by missing data. The denoising diagonal matrix is ​​reconstructed using the retained principal singular values ​​and the singular values ​​replaced with zeros. Matrix multiplication is then performed using the transposes of the left singular vector matrix, the denoising diagonal matrix, and the right singular vector matrix to obtain the denoising reconstruction matrix, which satisfies the restoration formula. ; The data sequence with missing time points and abnormal noise filtered out is extracted from the noise reduction and reconstruction matrix; based on the previously calculated propagation time delay, the extracted data sequence is shifted backward in reverse time phase to accurately restore the time phase of the data sequence corresponding to the downstream monitoring section to the original real physical time axis, thus obtaining the fluctuating reconstructed water level sequence; the fluctuating reconstructed water level sequence is output as the final high-precision water level reference sequence and transmitted to the error correction and risk prediction module; This embodiment overcomes the technical bottleneck of traditional data interpolation methods, which tend to smooth out extreme flood peaks when dealing with extreme flood conditions. By constructing a topological fluctuation matrix for singular value decomposition and introducing a preset singular value truncation threshold of 2.5 for hard threshold judgment, it can accurately remove high-frequency noise and abnormal mutations caused by missing data. At the same time, it perfectly preserves the main singular values ​​on the main diagonal that represent the principal components of the true flood peak. In addition, after the matrix reconstruction calculation is completed, the data sequence is reverse-shifted backward in time phase, strictly following the spatiotemporal conservation law of hydrodynamic evolution. This avoids the time coordinate distortion caused by the previous time phase shift alignment, and achieves accurate noise reduction and completion of complex hydrological time series data under high missing rate environment, ensuring the absolute correctness of the reconstructed water level benchmark sequence in physical space and time dimension.

[0031] Step S5 includes the following sub-steps: Obtain basic geographic parameters for the entire watershed and measured upstream flow data; Among them, the basic geographic parameters of the entire basin include channel roughness, cross-sectional geometry, riverbed slope and channel connectivity topology, and the water level reference sequence is a high-precision water level sequence at one or more control monitoring sections; The basic geographic parameters of the entire basin, upstream measured flow data, and boundary verification information corresponding to the water level benchmark sequence are input into a pre-set one-dimensional hydrodynamic simulation model for numerical solution, outputting simulated water level data for the entire basin. and simulated traffic data ; A set of control sections with water level reference sequences The simulated water level data for the entire basin is subtracted from the water level reference sequence to obtain the simulated water level error at the control section. : ; in, For water level reference sequence, To simulate water level error, at the upstream flow observation section At this point, the upstream measured flow rate data is subtracted from the simulated flow rate at the upstream cross-section output by the model to obtain the upstream flow rate error. The upstream flow error is mapped and correlated with the simulated water level error at different control sections to construct a linear regression correction model that considers propagation time delay across different sections. Specifically, the propagation time delay is calculated based on the river hydrodynamic propagation characteristics. Establish a linear regression correction model : ; in, and To control the cross-section The corresponding error correction coefficients, the output of the linear regression correction model The predicted water level error is used for subsequent additive compensation of the simulated water level. The error correction coefficients obtained from multiple control sections are interpolated or piecewise assigned along the spatial direction of the river channel to obtain the error correction coefficient field for the entire watershed. And obtain the water level error estimate for the entire basin. : ; A scaling correction model is constructed to perform amplitude consistency correction on the simulated flow across the entire watershed, and the flow scaling factor is calculated. : ; Where ϵ is a very small positive number to prevent the denominator from being zero, and the corrected flow data is obtained by proportionally correcting the simulated flow based on the flow scaling factor: ; To correct the flow rate data, real-time post-correction compensation was performed on the simulated water level data of the entire basin using error correction coefficients. The corrected water level data of the entire basin was obtained by employing an additive error stripping method. ; in, This provides water level data for the entire basin. Historical meteorological data of the target water area is obtained. The corrected whole basin water level data, corrected flow data and historical meteorological data are input into the pre-trained CKLSTM deep prediction network. Multi-dimensional time series feature extraction and inference are performed through the CKLSTM deep prediction network to predict and output the inflow flood flow data and water level evolution trend data for future periods. By using inflow flood data and water level evolution trend data for future periods, we can construct and solve the non-steady flow surface differential equation to calculate the dynamic additional reservoir capacity data generated by the slope of the reservoir water surface. The dynamic additional storage capacity data is accumulated and calculated with the static storage capacity curve data to generate real-time dynamic storage capacity data; Real-time dynamic reservoir capacity data is substituted into the reservoir water balance equation to perform flood regulation calculations and estimate the highest water level in front of the dam. Based on the highest water level in front of the dam and the dam crest elevation data, the probability of dam flood overtopping risk is calculated. The probability of dam flood overtopping risk is input into the risk rule engine for interval matching and comparison to determine the early warning response level and generate risk early warning results.

[0032] This embodiment deeply integrates the high-precision true values ​​(water level reference sequence) obtained by fiber optic inversion with a one-dimensional hydrodynamic simulation model. First, a linear regression correction model and a scaling correction model are used to perform real-time post-correction on the physical simulation results of hydrodynamics, completely eliminating the systematic bias caused by the fixed roughness and discretized equations of the pure physical model. Then, the high-quality data corrected by the physical mechanism is input into the CKLSTM deep prediction network, and the powerful nonlinear feature extraction capability of the deep learning model is used to predict future flow. Finally, the dynamic reservoir capacity physical model is combined to deduce the risk of overflow.

[0033] The processing logic of the CKLSTM deep prediction network that incorporates physical priors specifically includes: Following a unified time step, the acquired corrected whole-basin water level data, corrected flow data, and historical meteorological data were aligned and normalized according to the time dimension. The aligned and normalized corrected whole-basin water level data, corrected flow data and historical meteorological data are spliced ​​together to construct a multi-dimensional time-series input matrix. The multidimensional temporal input matrix is ​​input into the CKLSTM deep prediction network. The sliding convolution kernel is used to perform sliding window convolution operation on the time step dimension of the multidimensional temporal input matrix to extract the local spatial cross features between the corrected whole basin water level data, the corrected flow data and historical meteorological data, and output the local cross feature sequence. The local cross-feature sequence is input into the Kolmogorov-Arnold network of the CKLSTM deep prediction network. The learnable univariate activation function deployed on the network edge is used to perform high-dimensional nonlinear feature space transformation on the local cross-feature sequence to eliminate non-stationary noise interference in hydrological and meteorological data and output nonlinear mapping feature sequence. The nonlinear mapping feature sequence is input into the long short-term memory network of the CKLSTM deep prediction network. The historical hydrological evolution pattern in the nonlinear mapping feature sequence is calculated by long-term memory and short-term update, global temporal dependency features are extracted, and temporal hidden state vector is output. The temporal hidden state vector is input into the fully connected decoding layer of the CKLSTM deep prediction network; the fully connected decoding layer inversely maps the temporal hidden state vector to the hydrological physical dimension space, and predicts and outputs the inflow flood flow data and water level evolution trend data for future periods.

[0034] In this embodiment, during the preprocessing and computation of the CKLSTM deep prediction network, the corrected whole-basin water level data, corrected flow data, and historical meteorological data (including rainfall, temperature, and wind speed) are aligned in the time dimension according to a uniform time step (1 hour). For missing values, linear interpolation is used to fill them. Then, Min-Max normalization is performed on each feature dimension to map all data to the [0,1] interval. The three types of data after alignment and normalization are concatenated in the channel dimension. Assuming the time step is T=72 hours and the feature dimensions include: water level data (152 cross sections), flow data (152 cross sections), and meteorological data (8 rain gauges × 3 meteorological elements), the dimension of the concatenated multidimensional time series input matrix is ​​T×(152+152+24)=72×328. The multidimensional temporal input matrix is ​​input into the one-dimensional convolutional feature extraction module of the CKLSTM deep prediction network. The module uses three one-dimensional convolutional layers with kernel sizes of 3, 5, and 3, and the number of kernels in each layer is 64, 128, and 64, respectively. The stride is 1, and the activation function is ReLU. The convolutional kernels slide along the time step dimension to extract local spatial cross-features between water level, flow rate, and meteorological data. For example, the convolutional kernels simultaneously capture the correlation between the rise in water level at a certain cross section and the increase in rainfall at the upstream rain gauge station in adjacent time steps, and output a local cross-feature sequence with a dimension of 72×64. The local cross-feature sequence is input into the Kolmogorov-Arnold (KAN) network of the CKLSTM deep prediction network. The KAN network contains a two-layer structure, with each layer deploying a learnable univariate activation function (represented by B-spline basis functions). The network performs high-dimensional nonlinear feature space transformation on the local cross-feature sequence, effectively eliminating non-stationary noise interference in hydrological and meteorological data, and outputs a nonlinear mapping feature sequence with the dimension maintained at 72×64. The nonlinear mapping feature sequence is input into the Long Short-Term Memory (LSTM) network of the CKLSTM deep prediction network. The LSTM has 128 units and two layers. Dropout=0.3 is used to prevent overfitting. The network performs long-term memory and short-term update calculations on the historical hydrological evolution patterns, extracts global temporal dependency features, and outputs a temporal hidden state vector with a dimension of 1×128. The temporal hidden state vector is input into the fully connected decoding layer of the CKLSTM deep prediction network. The decoding layer contains two fully connected sublayers with 64 and 1 neurons (predicting flow) or 64 and 152 neurons (predicting water level). The fully connected decoding layer inversely maps the hidden state vector to the hydrological physical dimension space (recovering the actual dimensions through inverse normalization) and predicts the inflow flood flow data and water level evolution trend data for future periods.

[0035] The processing logic of the risk rules engine is as follows: The preset risk probability interval thresholds include a first threshold, a second threshold, and a third threshold. The first threshold is less than the second threshold and less than the third threshold. The warning response levels include blue warning, yellow warning, orange warning, and red warning. The calculated probability of dam flood overtopping is compared with the risk probability interval threshold: The probability of dam overflowing by floodwaters When the first threshold is reached, it is determined that the current state is under blue alert; When the first threshold dam flood overtopping risk probability When the second threshold is reached, it is determined that the current situation is under a yellow alert; When the second threshold dam flood overtopping risk probability When the third threshold is reached, it is determined that the current state is under orange alert; The probability of dam overflowing by floodwaters When the third threshold is reached, it is determined that the current situation is under a red alert.

[0036] In this embodiment, the preset risk probability interval thresholds are: the first threshold is set to 0.1, the second threshold is set to 0.3, and the third threshold is set to 0.6. The warning response levels are blue warning, yellow warning, orange warning, and red warning in order from low to high. According to the preset rules, when the risk probability is between the second threshold and the third threshold, it is determined that the current state is orange warning, and then the orange warning response process is triggered: send a warning notification to the dam management unit, start emergency communication broadcast, notify the downstream potentially affected areas to prepare for evacuation, and increase the monitoring frequency. If the probability of subsequent risk rises to 0.65 (exceeding the third threshold), it will be automatically upgraded to a red alert, and the highest level of emergency response will be activated, including notifying all downstream areas to evacuate immediately, activating all flood discharge facilities, and requesting support from higher-level emergency forces.

[0037] Example 2, refer to Figure 2 A water level data acquisition and analysis system is provided, including a data acquisition module, a net strain sequence extraction module, a water level inversion module, a water level data reconstruction module, and an error correction and risk prediction module; The data acquisition module is used to collect water level fiber optic data by vertically deploying water pressure-sensitive deformation rods at the monitoring section. The water level fiber optic data includes a first fiber optic sequence and a second fiber optic sequence. The net strain sequence extraction module is used to process the temperature deviation of the water level fiber optic data using a dual-mode differential algorithm to obtain the net strain sequence. The water level inversion module is used to acquire historical hydrostatic pressure data, perform spatial derivative analysis and mapping on the net strain sequence, and obtain actual water level data. The water level data reconstruction module is used to reconstruct the actual water level data by singular value decomposition using a comprehensive reconstruction network model to obtain the water level benchmark sequence. The error correction and risk prediction module is used to construct a hydrodynamic model based on the water level benchmark sequence, obtain the simulated water level error, construct a linear regression and scaling correction model, perform real-time post-correction on the simulated water level of the entire basin, obtain the corrected flow data, and input the corrected flow data into the CKLSTM deep prediction network to predict the inflow flood flow and risk warning results.

[0038] In this embodiment, the data acquisition module acquires physical signals under the monitoring environment by vertically deploying water pressure-sensitive deformation rods at the monitoring section of the target water area; the module also acquires water level fiber optic data, which includes a first fiber sequence generated by a first fiber firmly attached to the pressure-bearing surface of the water pressure-sensitive deformation rod, and a second fiber sequence generated by a second fiber threaded in a capillary tube fixed to the surface of the water pressure-sensitive deformation rod. The acquired water level fiber optic data is then transmitted in real time to the net strain sequence extraction module. The net strain sequence extraction module receives water level fiber optic data and performs point-by-point matching and subtraction operations on the first and second fiber optic sequences on a unified spatial elevation coordinate axis using a dual-mode differential algorithm. The module also processes the water level fiber optic data for temperature deviation to offset temperature interference caused by ambient temperature changes. Finally, the module acquires the true mechanical strain after eliminating temperature interference and discretizes, packages, and arranges it according to the elevation coordinate order to obtain the net strain sequence. The net strain sequence extraction module then sends the net strain sequence to the water level inversion module. The water level inversion module receives the net strain sequence, retrieves historical hydrostatic pressure data from the system database, and sets the derivative mutation threshold based on the historical hydrostatic pressure data. The water level inversion module calculates the spatial derivative of the net strain sequence to generate an initial numerical rate of change, and amplifies and corrects the initial numerical rate of change by combining a preset stiffness gradient compensation factor; it performs spatial derivative analysis and mapping on the corrected numerical rate of change, finds the critical strain inflection point on the spatial elevation coordinate axis where the absolute value of the numerical rate of change first exceeds the derivative mutation threshold, and directly maps the elevation coordinates corresponding to the critical strain inflection point to the actual water level data of the current monitoring section; and sends the actual water level data to the water level data reconstruction module. The water level data reconstruction module receives actual water level data from multiple monitoring sections and combines them to generate a multi-section original water level time series set. Using a comprehensive reconstruction network model, when the water level fluctuation variance of the multi-section original water level time series set is determined to be greater than or equal to a preset fluctuation variance threshold of 1.0 square meters, a forward translation and alignment operation is performed on the time phase of the multi-section original water level time series set based on the shallow water wave propagation velocity. A topological fluctuation matrix is ​​constructed using the translated and aligned multi-section original water level time series set. Singular value decomposition is performed on the topological fluctuation matrix, and a hard thresholding process is applied using a set singular value truncation threshold of 2.5 to remove small singular values ​​representing abnormal noise. Matrix multiplication reconstruction is performed on the retained main singular values, and a reverse backward translation operation is performed on the reconstructed data sequence to obtain a continuous and complete water level reference sequence. The water level reference sequence is then sent to the error correction and risk prediction module. The error correction and risk prediction module receives the water level reference sequence, constructs a hydrodynamic model based on it, and calculates the difference between the simulated water level data of the entire basin output by the hydrodynamic model and the water level reference sequence to obtain the simulated water level error. It then uses this simulated water level error to construct a linear regression and scaling correction model, extracting error correction coefficients. These coefficients are used to perform real-time post-correction on the simulated water level data of the entire basin, removing the discretization bias of the hydrodynamic model to obtain corrected water level data and corrected flow data for the entire basin. The entire basin's water level data and corrected flow data are input into a pre-trained CKLSTM deep prediction network. Under the constraint of the law of conservation of physical mass, the network predicts and outputs inflow flood flow data and water level evolution trend data for future periods. The inflow flood flow data and water level evolution trend data are then substituted into a dam flood overtopping risk extrapolation model that considers dynamic reservoir capacity. The probability of dam flood overtopping risk is calculated, and based on a pre-set risk probability interval threshold, the risk warning results (blue, yellow, orange, or red) are output to the large display screen of the flood control command center.

[0039] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for acquiring and analyzing water level data, characterized in that, Includes the following steps: Step S1: Collect water level fiber optic data through a water pressure sensitive deformation rod. The water level fiber optic data includes a first fiber optic sequence and a second fiber optic sequence. Step S2: Temperature deviation processing is performed on the water level fiber optic data using a dual-mode differential algorithm to obtain the net strain sequence; Step S3: Obtain historical hydrostatic pressure data, perform spatial derivative analysis mapping on the net strain sequence, and obtain actual water level data; Step S4: Based on the principle of hydrodynamic phase balance alignment, the actual water level data is reconstructed by singular value decomposition using a comprehensive reconstruction network model to obtain the water level reference sequence. Step S5: Construct a hydrodynamic model based on the water level reference sequence to obtain the simulated water level error. Construct a correction model to perform real-time post-correction on the simulated water level of the entire basin to obtain the corrected flow data. Input the corrected water level data and the corrected flow data of the entire basin into the CKLSTM network to output the inflow flood flow and risk warning results.

2. The water level data acquisition and analysis method as described in claim 1, characterized in that, Step S1 specifically includes: With the bottom fixed point of the water pressure sensitive deformation rod as the coordinate origin 0, a one-dimensional spatial elevation coordinate axis z is established along the upward elevation direction of the water pressure sensitive deformation rod; The first and second optical fibers are simultaneously laid on the water pressure sensitive deformation rod; The first optical fiber is installed on the pressure-bearing surface of the water pressure-sensitive deformation rod along the direction of the spatial elevation coordinate axis z, and the second optical fiber is installed in the capillary tube on the surface of the water pressure-sensitive deformation rod. Based on the backscattering Rayleigh signal, the frequency shift of the first Rayleigh scattering spectrum corresponding to the first optical fiber and the frequency shift of the second Rayleigh scattering spectrum corresponding to the second optical fiber are obtained, and then packaged to generate the first optical fiber sequence and the second optical fiber sequence respectively. By matching the first fiber sequence with the second fiber sequence one-to-one, water level fiber data is obtained.

3. The water level data acquisition and analysis method as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S21: Extract the first fiber sequence and the second fiber sequence from the water level fiber data. The first fiber sequence includes the first Rayleigh scattering spectral frequency shift at each discrete spatial sampling point, and the second fiber sequence includes the second Rayleigh scattering spectral frequency shift at each discrete spatial sampling point. Based on the spatial elevation coordinate axis z, according to the division position of the discrete spatial sampling points, match the first Rayleigh scattering spectral frequency shift of the i-th sampling point in the first fiber sequence with the second Rayleigh scattering spectral frequency shift of the i-th sampling point in the second fiber sequence one by one to obtain the discrete spatial sampling point set. Step S22: Subtract the frequency shift of the second Rayleigh scattering spectrum corresponding to the discrete spatial sampling point from the frequency shift of the first Rayleigh scattering spectrum corresponding to the discrete spatial sampling point to obtain the differential frequency shift. Then divide the differential frequency shift by the strain sensitivity coefficient to obtain the true mechanical strain. Step S23: Discretize and arrange all the real mechanical strains according to the elevation direction coordinates of the spatial elevation coordinate axis z to generate the net strain sequence.

4. The water level data acquisition and analysis method as described in claim 3, characterized in that, Step S3 includes the following sub-steps: Step S31: Obtain historical hydrostatic pressure data of the target monitoring section and stiffness gradient compensation factor corresponding to the water pressure sensitive deformation rod with gradient stiffness along the elevation direction from the database; set the derivative mutation threshold for identifying the water surface interface position; calculate the spatial derivative of the net strain sequence along the elevation direction coordinate order of the spatial elevation coordinate axis z to obtain the initial numerical change rate. Step S32: Multiply the stiffness gradient compensation factor by the initial numerical rate of change, and physically amplify the strain mutation signal at the water surface interface in the top low stiffness region to obtain the corrected spatial derivative sequence. Step S33: Traverse the corrected spatial derivative sequence, find the target discrete spatial sampling point where the absolute value of the numerical rate of change first exceeds the derivative mutation threshold, and mark the target discrete spatial sampling point as the critical strain inflection point. Step S34: Extract the specific elevation coordinate values ​​corresponding to the marked critical strain inflection points from the spatial elevation coordinate axis z, and map the specific elevation coordinate values ​​to the actual water surface elevation of the current monitoring section and output them as the actual water level data at the current moment.

5. The water level data acquisition and analysis method as described in claim 4, characterized in that, Step S4 specifically includes: Obtain the original water level time series set from multiple cross-sections, calculate the water level fluctuation variance, and when the water level fluctuation variance is greater than or equal to a preset fluctuation variance threshold, determine that the original water level time series set from multiple cross-sections is in a period of severe fluctuation, and trigger the singular value decomposition reconstruction mechanism: When the original water level time series set of multiple cross sections is in a period of violent fluctuation, the spatial channel distance between two adjacent monitoring sections is obtained, and the average water depth and average flow velocity of the channel corresponding to the spatial channel distance are obtained. According to the theory of shallow water wave dynamics, the average water depth of the river channel is multiplied by the gravitational acceleration and the square root is taken. The result of the square root calculation is added to the average flow velocity to calculate the propagation speed of shallow water waves. Divide the spatial river channel distance by the shallow water wave propagation speed to obtain the propagation time delay corresponding to the evolution of flood waves between two adjacent monitoring sections; By utilizing the propagation time delay, the data sequence corresponding to the downstream monitoring section in the original multi-section water level time series is shifted forward in time phase to obtain the aligned multi-section water level time series.

6. The water level data acquisition and analysis method as described in claim 5, characterized in that, Step S4 further includes: The aligned multi-section water level time series set is transformed into a two-dimensional matrix form to construct a topological fluctuation matrix M. This topological fluctuation matrix is ​​then input into the singular value decomposition (SVD) reconstruction model to perform matrix decomposition, yielding the singular value decomposition formula: ； in, It is a left singular vector matrix. It is the transpose of the right singular vector matrix. This is a diagonal matrix containing multiple singular values, where the singular values ​​are arranged in descending order of their numerical values ​​along the main diagonal of the matrix. A singular value truncation threshold is preset. Multiple singular values ​​on the main diagonal of the diagonal matrix are traversed. Each singular value is compared with the singular value truncation threshold. Singular values ​​that are greater than or equal to the singular value truncation threshold are retained as primary singular values. Singular values ​​smaller than the singular value truncation threshold are replaced with zero. A denoising diagonal matrix is ​​constructed using the retained principal singular values ​​and the singular values ​​replaced with zero. The denoising reconstruction matrix is ​​obtained by performing matrix multiplication using the transpose of the left singular vector matrix, the denoising diagonal matrix, and the right singular vector matrix. Satisfying the reduction formula ; Extract the data sequence from the noise reduction and reconstruction matrix to complete the missing time point data; Based on the propagation time delay, the extracted data sequence is reverse-shifted backward in time phase to restore the time phase of the data sequence corresponding to the downstream monitoring section to the original real physical time axis, thus obtaining the wave reconstructed water level sequence. The fluctuating reconstructed water level sequence is output as the water level reference sequence.

7. The water level data acquisition and analysis method as described in claim 6, characterized in that, Step S5 specifically includes: Acquire basic geographic parameters of the entire basin and measured upstream flow data. The basic geographic parameters of the entire basin include channel roughness, cross-sectional geometry, riverbed slope, and channel connectivity topology. The basic geographic parameters of the entire basin, the measured upstream flow data, and the boundary verification information corresponding to the water level benchmark sequence are input into a preset one-dimensional hydrodynamic simulation model for numerical solution, and the simulated water level data and simulated flow data of the entire basin are output. The simulated water level data of the entire basin is subtracted from the water level reference sequence to obtain the simulated water level error at the control section. The measured upstream flow data is subtracted from the simulated upstream flow output by the model to obtain the upstream flow error. The upstream flow error is mapped and correlated with the simulated water level errors at different control sections to construct a linear regression correction model and output the predicted value of the water level error. The error correction coefficients obtained from multiple control sections are interpolated and assigned along the spatial direction of the river channel to obtain the error correction coefficient field of the entire basin and the water level error estimate of the entire basin. A scaling correction model is constructed, the flow scaling factor is calculated, and the simulated flow is proportionally corrected based on the flow scaling factor to obtain the corrected flow data. The error correction factor is used to perform real-time post-correction compensation on the simulated water level data of the whole basin. The corrected water level data of the whole basin is obtained by using the additive error stripping method. Historical meteorological data of the target water area is obtained. The corrected whole basin water level data, corrected flow data and historical meteorological data are input into the pre-trained CKLSTM deep prediction network. Multi-dimensional time series feature extraction and inference are performed through the CKLSTM deep prediction network to predict and output the inflow flood flow data and water level evolution trend data for future periods. By using inflow flood data and water level evolution trend data for future periods, we can construct and solve the non-steady flow surface differential equation to calculate the dynamic additional reservoir capacity data generated by the slope of the reservoir water surface. The dynamic additional storage capacity data is accumulated and calculated with the static storage capacity curve data to generate real-time dynamic storage capacity data; Real-time dynamic reservoir capacity data is substituted into the reservoir water balance equation to perform flood regulation calculations, and the highest water level in front of the dam is estimated. Based on the highest water level in front of the dam and the dam crest elevation data, the probability of dam flood overtopping risk is calculated. The probability of dam flood overtopping risk is input into the risk rule engine for interval matching and comparison to obtain the early warning response level and generate risk warning results.

8. The water level data acquisition and analysis method as described in claim 7, characterized in that, The processing logic of the CKLSTM deep prediction network that incorporates physical priors specifically includes: Following a unified time step, the acquired corrected whole-basin water level data, corrected flow data, and historical meteorological data were aligned and normalized according to the time dimension. The aligned and normalized corrected whole-basin water level data, corrected flow data and historical meteorological data are spliced ​​together to construct a multi-dimensional time-series input matrix. The multidimensional temporal input matrix is ​​input into the CKLSTM deep prediction network. The sliding convolution kernel is used to perform sliding window convolution operation on the time step dimension of the multidimensional temporal input matrix to extract the local spatial cross features between the corrected whole basin water level data, the corrected flow data and historical meteorological data, and output the local cross feature sequence. The local cross-feature sequence is input into the Kolmogorov-Arnold network of the CKLSTM deep prediction network. The learnable univariate activation function deployed on the network edge is used to perform high-dimensional nonlinear feature space transformation on the local cross-feature sequence, and output nonlinear mapping feature sequence. The nonlinear mapping feature sequence is input into the long short-term memory network of the CKLSTM deep prediction network, and the output is a temporal hidden state vector. The temporal hidden state vector is input into the fully connected decoding layer of the CKLSTM deep prediction network to predict and output inflow flood data and water level evolution trend data for future periods.

9. The water level data acquisition and analysis method as described in claim 8, characterized in that, The processing logic of the risk rules engine is as follows: The preset risk probability interval thresholds include a first threshold, a second threshold, and a third threshold, wherein the first threshold is less than the second threshold and less than the third threshold, and the warning response levels include blue warning, yellow warning, orange warning, and red warning. The calculated probability of dam flood overtopping is compared with the risk probability interval threshold: The probability of dam overflowing by floodwaters When the first threshold is reached, it is determined that the current state is under blue alert; When the first threshold dam flood overtopping risk probability When the second threshold is reached, it is determined that the current situation is under a yellow alert; When the second threshold dam flood overtopping risk probability When the third threshold is reached, it is determined that the current state is under orange alert; The probability of dam overflowing by floodwaters When the third threshold is reached, it is determined that the current situation is under a red alert.

10. A water level data acquisition and analysis system, applied in a water level data acquisition and analysis method as described in any one of claims 1-9, characterized in that, It includes a data acquisition module, a net strain sequence extraction module, a water level inversion module, a water level data reconstruction module, and an error correction and risk prediction module; The data acquisition module is used to collect water level optical fiber data by vertically deploying water pressure sensitive deformation rods at the monitoring section. The water level optical fiber data includes a first optical fiber sequence and a second optical fiber sequence. The net strain sequence extraction module is used to process the temperature deviation of the water level fiber optic data using a dual-mode differential algorithm to obtain the net strain sequence. The water level inversion module is used to acquire historical hydrostatic pressure data, perform spatial derivative analysis and mapping on the net strain sequence, and obtain actual water level data. The water level data reconstruction module is used to reconstruct the actual water level data by singular value decomposition using a comprehensive reconstruction network model to obtain a water level benchmark sequence. The error correction and risk prediction module is used to construct a hydrodynamic model based on the water level reference sequence, obtain the simulated water level error, construct a linear regression and scaling correction model, perform real-time post-correction on the simulated water level of the entire basin, obtain the corrected flow data, and input the corrected flow data into the CKLSTM deep prediction network to predict and output the inflow flood flow and risk warning results.

Images