River and lake water level and flow prediction method and system

Abstract

The invention discloses a river and lake water level and flow prediction method and system in the field of remote sensing hydrology and hydraulic engineering, and the method comprises the steps: obtaining satellite radar altimeter data, river and lake water system data, rainfall inversion data and surface temperature inversion data based on a target river, a target lake and a watershed range file; taking the cross interval as a virtual hydrometric station observation point, extracting and processing waveform data, calculating water surface elevation information through a wavelet tracking algorithm, constructing an elevation profile group to establish an initial water level time sequence, and generating a river and lake water level inversion data set through fitting and elevation reselection. Meanwhile, NDWI data are obtained from satellite radar altimeter data, water surface width data are extracted, an inversion data set is formed by combining actually-measured river channel section data and a Manning formula inversion river channel flow rate, and finally various kinds of data are input into a pre-constructed intelligent prediction model to obtain future river and lake water level and flow rate change conditions. According to the invention, the precision of river and lake water level and flow prediction in areas lacking data is effectively improved.

Description

Technical Field

[0001] The present invention relates to a method and system for predicting river and lake water levels and flows, and belongs to the technical field of remote sensing hydrology and water conservancy engineering. Background Art

[0002] River and lake water levels and flows are key indicators of the water cycle and are crucial for the development and utilization of water resources in river basins and the protection of river ecosystems. However, in data-deficient and data-deficient areas of the plateau cold region, a severe lack of hydrological monitoring and technical limitations result in a lack of hydrological data and underlying surface parameters, hindering the identification of hydrological response mechanisms in the plateau cold region as a globally sensitive climate zone.

[0003] Furthermore, the plateau's complex topography and unpredictable climate make traditional hydrological monitoring methods difficult to implement effectively, further exacerbating data acquisition challenges. For example, in data-scarce areas, particularly in river and lake basins in the cold regions of the plateau, monitoring water levels and flows presents numerous challenges. On the one hand, a shortage of ground-based monitoring equipment makes it difficult to obtain continuous and accurate hydrological data. On the other hand, existing hydrological models have limitations in depicting the complex hydrological processes of plateau lakes, resulting in predictions that fall short of practical requirements. Against this backdrop, there is an urgent need to develop cost-effective and efficient methods for accurately acquiring water level and flow information for plateau rivers and lakes. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for predicting river and lake water levels and flows, which can solve the problems of low hydrological prediction accuracy in areas with insufficient hydrological monitoring data, insufficient adaptability of traditional models, and complex terrain and climatic conditions.

[0005] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.

[0006] In one aspect, the present invention provides a method for predicting river and lake water levels and flows, comprising: Obtain satellite radar altimeter data, river and lake system data, precipitation inversion data, and surface temperature inversion data based on target rivers, lakes, and their basin range files; The intersection of the satellite radar altimeter data and the river and lake system data is used as a virtual hydrological station observation point, and the waveform data in the satellite radar altimeter data is extracted using the virtual hydrological observation point; Processing the waveform data and decomposing the processed waveform data by a wavelet tracking algorithm to extract water surface elevation information, constructing an elevation profile group based on the water surface elevation information, and establishing an initial water level time series; Fitting the initial water level time series, performing elevation reselection in combination with the elevation profile group, and generating a river and lake water level inversion dataset; Acquiring normalized difference water index data based on the satellite radar altimeter data, extracting water surface width data from the normalized difference water index data, and inverting river flow using the measured river cross-section data and the Manning formula to form a river flow inversion data set; The river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation inversion data and surface temperature inversion data are input into the pre-built river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

[0007] In combination with the first aspect, further, using the virtual hydrological observation point to extract waveform data from satellite radar altimeter data includes: Using a water body frequency analysis method, the water surface range of the virtual hydrological observation point is determined, and a buffer zone is set; Using the satellite radar altimeter data, screening valid virtual hydrological observation points within the buffer zone; 1 Hz waveform data and 20 Hz auxiliary data are extracted from the effective virtual hydrological observation points.

[0008] In combination with the first aspect, further, constructing the elevation profile group and establishing the initial water level time series includes: Selecting a quasi-rectangular area containing the valid virtual hydrological observation point, and in the quasi-rectangular area, sequentially performing waveform resampling, average waveform synthesis, peak detection, and filtering on the extracted 1 Hz waveform data to obtain a filtered average waveform; Performing wavelet decomposition on the average waveform to obtain a plurality of wavelets, and matching the decomposed wavelets with the effective peak values ​​of the average waveform; The OCOG algorithm is used to reshape the waveform of the decomposed wavelet, and then the wavelet tracking algorithm is used to calculate the elevation of the decomposed wavelet to extract the water surface elevation value. Based on the matching result of the decomposed wavelet and the effective peak value of the average waveform, a multi-objective optimization function is constructed with the standard deviation of the water surface elevation value and the peak value of the wavelet as constraints to screen out the effective water surface elevation value; According to the effective water surface elevation value, an elevation profile group is constructed, and the elevation profiles at different time points are arranged in chronological order to generate an initial water level time series.

[0009] In combination with the first aspect, further, generating a river and lake water level inversion dataset includes: Performing time series fitting on the initial water level time series using the SG function to obtain first fitting water level data; Performing sliding median fitting on the first fitting water level data to generate second fitting water level data, and identifying abnormal noise point data that deviates from the median in the second fitting water level data; Combined with the constraints of the multi-objective optimization function, the abnormal noise point data is subjected to elevation reselection and discrimination. If the elevation value of the abnormal noise point data exceeds the constraint range of the multi-objective optimization function, the abnormal noise point data is subjected to noise revision to obtain a revised water level time series; if the abnormal noise point data does not exceed the constraint range, the original data is retained and directly output as a valid water level time series; Combining the revised water level time series with the effective water level time series to obtain the latest water level time series; The latest water level time series is integrated with the effective water surface elevation value to generate a river and lake water level inversion dataset.

[0010] In combination with the first aspect, further, forming a river flow inversion dataset includes: Conduct field measurements at preset cross-section locations of target rivers and lakes to obtain river cross-section data, and simultaneously record the width of the measured river cross-section; The normalized difference water index data is inverted using a hybrid pixel decomposition algorithm to obtain the remote sensing water surface width of the target rivers and lakes; Comparing the measured width of the river section with the remotely sensed water surface width, adjusting the threshold of the normalized difference water index data according to the comparison result, and extracting water surface width data in batches from the normalized difference water index data based on the adjusted threshold; The batch-extracted water surface width data are input into the pre-built river flow inversion model to generate a river flow inversion dataset.

[0011] In combination with the first aspect, further, based on the adjusted threshold, the expression for batch extracting water surface width data from the normalized difference water index data is: ; Where W represents the water surface width data; PA represents the actual area of ​​the pixel; VL represents the length of the valley area of ​​interest; The value of the normalized difference water index of the pixel; represents the land NDWI threshold; represents the NDWI threshold of the water area; x represents the total number of pixels in the valley area of ​​interest; j represents the index value of the pixel currently being processed.

[0012] In combination with the first aspect, further, the process of constructing the river flow inversion model includes: Calculating the water-passing cross-sectional area, wetted perimeter, and hydraulic radius of the measured river channel section based on the geometric features of the measured river channel section; wherein the geometric features include the shape of the river channel section, the width of the water surface, and the average water depth of the section; Based on the cross-sectional area and wetted perimeter, the initial flow rate is calculated using the Manning formula. ; Based on hydraulic geometry theory, a power function relationship between water surface width w, average cross-section water depth d, average flow velocity v and river discharge Q is constructed; The initial water surface width measured by , the calculated initial flow and the corresponding water depth The nonlinear least squares method is used to fit and determine the water surface width benchmark coefficient a, the cross-section average water depth benchmark coefficient c, the average flow velocity benchmark coefficient k, the water surface width flow index b, the water depth flow index f, and the flow velocity flow index m, and verify whether the benchmark coefficient and flow index meet the constraints b + f + m = 1, and a * c * k = 1; if not, readjust the parameters of the river flow inversion model until the benchmark coefficient and flow index meet the constraints, and finally complete the construction of the river flow inversion model.

[0013] In combination with the first aspect, further, the initial flow The expression is: ; in, represents the initial flow rate; Indicates the cross-sectional area of ​​water flow; represents the wetted perimeter; n represents the roughness; Indicates the slope.

[0014] In combination with the first aspect, further, obtaining the changes in river and lake water levels and flows in the corresponding future time period includes: Normalizing the river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation inversion data, and surface temperature inversion data using a feature scaling method; An intelligent model for river and lake water levels and flows was constructed based on a variety of machine learning and deep learning models. The normalized data was divided into training and test sets, and 10-fold cross-validation was introduced to train and validate the selected models. The machine learning models included support vector regression, gradient boosting trees, and random forests; the deep learning models included long-term memory networks and BP neural network models. The mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R) between the predicted and actual values ​​of the water level and flow of the river and lake are calculated on each fold validation set. 2, summarize the average and standard deviation of the 10 validation results, and compare the mean square error MSE, mean absolute error MAE and determination coefficient R of each model on the validation set 2 , select the optimal model as the intelligent prediction model for river and lake water level and flow; Compare and analyze the river and lake water level inversion data set, river flow inversion data set, water surface width data, precipitation inversion data, and surface temperature inversion data with the historical forecast data for the same period, and calculate the average value of the water surface width data, precipitation inversion data, and surface temperature inversion data in the current period and the average value of the corresponding historical forecast data for the same period; The correction coefficient is obtained based on the ratio of the average value of the water surface width data, precipitation inversion data and surface temperature inversion data set in the current period to the average value of the corresponding historical forecast data in the same period; Multiplying the correction coefficient by the forecast data for a predetermined future time period to obtain revised water surface width data, precipitation inversion data, and surface temperature inversion data; The revised water surface width data, precipitation inversion data and surface temperature inversion data set are input into the optimal river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

[0015] In a second aspect, a river and lake water level and flow prediction system includes: The river and lake water level remote sensing inversion module is used to obtain satellite radar altimeter data, river and lake water system data, precipitation, and surface temperature inversion data based on target rivers, lakes, and their basin range files; use the intersection of the satellite radar altimeter data and the river and lake water system data as a virtual hydrological station observation point, and use the virtual hydrological observation point to extract waveform data from the satellite radar altimeter data; process the waveform data and decompose the processed waveform data using a wavelet tracking algorithm to extract water surface elevation information; construct an elevation profile group based on the water surface elevation information, and establish an initial water level time series; fit the initial water level time series, perform elevation reselection in combination with the elevation profile group, and generate a river and lake water level inversion dataset; a river cross-sectional flow estimation module, configured to obtain normalized difference water index data based on the satellite radar altimeter data, extract water surface width data from the normalized difference water index data, and invert the river flow using the measured river cross-sectional data and the Manning formula to form a river flow inversion dataset; The river and lake water level and flow prediction module is used to input the river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation and surface temperature inversion data into the pre-built river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention acquires satellite radar altimeter data, river and lake drainage data, precipitation inversion data, and land surface temperature inversion data based on target river, lake, and basin boundary files. The method uses the intersection of satellite radar altimeter data and river and lake drainage data as virtual hydrological station observation points and extracts waveform data from them. By processing the waveform data and calculating water surface elevation information using a wavelet tracking algorithm, an elevation profile set is constructed to establish an initial water level time series. After fitting and elevation reselection, a river and lake water level inversion dataset is generated. Simultaneously, normalized difference water index data is obtained from the satellite radar altimeter data, and water surface width data is extracted. This inversion dataset is then formed by combining measured river cross-section data with the Manning formula to invert river flow. Finally, these data are input into a pre-built intelligent prediction model to determine future changes in river and lake water levels and flows. This method can effectively address the challenges of hydrological data monitoring in data-deficient regions such as plateaus and cold regions, improving prediction accuracy and providing strong support for water resource management and flood prevention and disaster reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 FIG2 is a flow chart of a method for predicting river and lake water level and flow provided by an embodiment of the present invention; Figure 2 It is a schematic diagram showing the parameter calculation of the water flow cross-sectional area, the water flow cross-sectional wetted perimeter and the hydraulic radius provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0019] The term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " generally indicates an "or" relationship between the related objects. Example 1

[0020] See also Figure 1 This embodiment introduces a method for predicting river and lake water levels and flows, including: Obtain satellite radar altimeter data, river and lake system data, precipitation inversion, and surface temperature inversion data based on target rivers, lakes, and their basin range files; The intersection of satellite radar altimeter data and river and lake system data is used as a virtual hydrological station observation point, and the waveform data in the satellite radar altimeter data is extracted using the virtual hydrological observation point. Processing the waveform data and decomposing the processed waveform data by a wavelet tracking algorithm to extract water surface elevation information, constructing an elevation profile group based on the water surface elevation information, and establishing an initial water level time series; Fit the initial water level time series and perform elevation reselection based on the elevation profile group to generate a river and lake water level inversion dataset; Normalized Difference Water Index (NDWI) data is obtained based on satellite radar altimeter data. Water surface width data is extracted from the NDWI data. River flow is inverted using the Manning formula in combination with measured river cross-section data to form a river flow inversion dataset. The river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation, and surface temperature inversion data are input into the pre-built river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

[0021] Among them, the model adopts a multi-model fusion framework including support vector regression, random forest, long short-term memory network, etc., and selects the optimal prediction model through 10-fold cross-validation. At the same time, based on the deviation between historical data and forecast data, it generates correction coefficients and dynamically calibrates the input parameters of future water surface width data, precipitation inversion data and surface temperature inversion data, and finally outputs high-precision prediction results of river and lake water levels and flows in future periods in areas with insufficient data.

[0022] In summary, this method effectively solves the problem of lack of hydrological monitoring data in plateau cold regions through the deep integration of remote sensing data and machine learning, and achieves improved accuracy of water level and flow inversion. Example 2

[0023] See also Figure 1 Based on Example 1, this embodiment further optimizes the remote sensing inversion step of river and lake water levels in the river and lake water level and flow prediction method to improve the accuracy and reliability of water level inversion, as follows: Step 1: Remote sensing inversion of river and lake water levels: 1.1 Extraction of waveform data Based on the target rivers, lakes and their basin range files, high-precision satellites such as Jason / Sentinel and Landsat are used to obtain radar altimeter data and river and lake system data. The radar altimeter data and river and lake system data are combined to determine the intersection interval, which is used as the observation point of the virtual hydrological station. The NDWI and OSTU algorithms (water body frequency analysis method) are used to extract the water surface range of the target river and lake, that is, to determine the water surface range of the virtual hydrological observation point. The calculation expression of its NDWI is: (1) Wherein, NDWI represents the normalized difference water index; Gb represents the surface reflectance in the green band; Indicates the surface reflectance in the near-infrared band.

[0024] Taking into account the fluctuations and changes of the water surface and possible data errors, a buffer zone is set around the water surface boundary to ensure that all possible water surface areas related to the virtual observation point are covered while reducing interference from non-water bodies; Then, the satellite radar altimeter data is used to screen out the valid virtual hydrological observation points in the buffer zone, and 1Hz waveform data and 20Hz auxiliary data are extracted from the valid virtual hydrological observation points.

[0025] 1.2 Waveform Analysis and Wavelet Decomposition In order to determine a relatively stable analysis area, a quasi-rectangular area containing effective virtual hydrological observation points is selected as the central footpoint set to determine the key area of ​​analysis. Compared with the traditional method, the footpoint selection range is expanded and the traditional waveform selection criteria are improved.

[0026] In terms of waveform analysis, the extracted 1Hz waveform data is resampled (upsampling and downsampling) within a quasi-rectangular area (i.e., the key area of ​​analysis) to eliminate sampling rate differences. The resampled waveform data is then averaged (average waveform synthesis) to generate an average waveform for eliminating noise interference. Peaks are then detected in the average waveform to filter out interference peaks caused by noise or non-water surface reflections (peak detection and filtering processing), thereby extracting the effective peak waveform. The filtered average waveform can more clearly reflect the water surface characteristics.

[0027] Then, the filtered average waveform is decomposed into multiple wavelets with different frequency and time characteristics. The decomposed wavelets are matched with the effective peaks of the average waveform to determine which wavelets correspond to the water surface reflection signals.

[0028] The decomposed wavelet can be a single frequency or a harmonic with a certain frequency range, thereby achieving accurate analysis in the time and frequency domains and extracting useful information.

[0029] Before performing elevation calculations, the OCOG algorithm is first used to reshape the waveform of the decomposed wavelet to eliminate the influence of factors such as terrain and reflection differences, thereby extracting wave height information that can reflect the true elevation information of the earth's surface. The decomposed wavelet is then subjected to elevation calculations using the wavelet tracking algorithm to extract the water surface elevation value. That is, based on the extracted wavelet parameters and valid wave height information, an elevation calculation model such as a geometric model or a physical model is selected for calculation to obtain the water surface elevation information. Based on the peak position of the average waveform and the wavelet pairing results, a multi-objective optimization function is constructed with the standard deviation of the water surface elevation information and the wavelet peak as constraints. According to the multi-objective optimization function, abnormal data with excessive fluctuations in elevation values ​​are identified, and the elevation values ​​corresponding to the wavelets with significant capabilities are retained, thereby screening out valid elevation values. The filtered effective elevation values ​​are integrated according to spatial positions (such as along the river profile) to form an elevation profile group, and the elevation profiles at different time points are arranged in chronological order to generate the initial water level time series.

[0030] 1.3 Noise filtering and elevation reselection For the time series of the initial water level, calculate the mean and standard deviation of the water level within the time series to understand the long-term trend and volatility of the water level; According to the long-term trend and volatility of the water level, the SG function (Savitzky-Colay filter) is used to perform time series fitting on the generated initial water level time series to obtain the first fitted water level data; Noise detection is performed on the first fitting water level data by sliding median fitting to generate second fitting water level data, thereby identifying abnormal noise point data that deviates from the median in the second fitting water level data; Combined with the constraints of the multi-objective optimization function, it is determined whether the abnormal noise point data needs to be revised. If the abnormal noise point data exceeds the constraint range of the multi-objective optimization function, the abnormal noise point data is revised or eliminated; if the abnormal noise point data does not exceed the constraint range, the original data is retained and directly output as a valid water level time series; Combining the revised water level time series with the effective water level time series to obtain the latest water level time series; The latest water level time series is integrated with the effective water surface elevation value to generate a river and lake water level inversion dataset.

[0031] Step 2: Estimation of river cross-section flow 2.1 Data measurement of river sections After selecting the cross-section location, drones are used to conduct field measurements of the river channel cross-section to obtain drone-based impact data for the monitored section. This data then provides the geometric characteristics of the river section, including its shape, water surface width, and average water depth. Simultaneously, through field data collection and field measurements, key hydrological parameters such as velocity, water level, flow rate, slope, and roughness are accurately determined. The geometric characteristics of the river section and the hydrological parameters together constitute the measured river channel cross-section data.

[0032] For example, a single-beam echo sounder is used to emit a single sound wave pulse underwater, and the water level is calculated by calculating the round-trip time and sound speed of the sound wave; based on the surface flow velocity measurement method, a rotor flow meter is selected, starting from the starting point of the section, the flow velocity parameters are collected at the same distance as the interval; a single-station RTK satellite positioning measurement method is adopted, and a single mobile station RTK is combined with satellite signals to obtain slope data; the determination of roughness should be based on the n-value table made from people's long-term engineering practice experience and experimental data, and priority should be given to the measured data and operation conditions of local and foreign channels of the same type to ensure that the selection of n value is consistent with the actual situation.

[0033] 2.2 Batch extraction of water surface width from satellite remote sensing NDWI data is obtained based on high-precision Jason / Sentinel / Landsat satellite radar altimeter data, and the hybrid pixel decomposition algorithm is used to invert the NDWI data to obtain the remote sensing water surface width of the target rivers and lakes; In order to distinguish different categories such as water bodies and non-water bodies during the inversion process, the width of the river section obtained by field measurement is compared with the remotely sensed water surface width of the target rivers and lakes. The NDWI threshold is adjusted according to the comparison results, and the water surface width data is batch extracted from the NDWI data based on the adjusted threshold. Finally, the batch-extracted water surface width data are input into the pre-built river flow inversion model to generate a river flow inversion dataset; Specifically, satellite radar altimeter data was used to establish four training zones: land, water, river valley, and river length. These zones were used to determine the NDWI thresholds for land and water, the water surface area in river valleys, and the average water surface width. After the training zones were established, the land and water zones were checked to ensure accurate coverage of the land and water surfaces, and the river length zone was verified for length consistency.

[0034] The NDWI data on the date when the river width is known, the NDWI data on the date when the river width needs to be inverted, and the data of the training area are used to calculate the land NDWI threshold, water NDWI threshold, and water surface area in the valley training area. The river water surface width is calculated using the following formula, which is expressed as: (2) Where W represents the width of the water surface; PA represents the actual area of ​​the pixel; VL represents the length of the valley area of ​​interest; The value of the normalized difference water index of the pixel; represents the land NDWI threshold; represents the NDWI threshold of the water area; x represents the total number of pixels in the valley area of ​​interest; j represents the index value of the pixel currently being processed.

[0035] 2.3 Constructing a river and lake flow inversion model Based on the geometric features of the measured river section, the cross-section surface, wetted perimeter and hydraulic radius are calculated; Based on the water flow cross section, wetted perimeter and Manning formula, the initial flow Q0 is calculated, and its expression includes: (3) in, represents the initial flow rate; Indicates the cross-sectional area of ​​water flow; represents the wetted perimeter, based on the cross-sectional shape (parabolic, trapezoidal or triangular); n represents the roughness; Indicates the slope.

[0036] It should be noted that the wetted perimeter varies depending on the cross-sectional shape (parabola, trapezoid or triangle), resulting in different cross-sectional areas, wetted perimeters and hydraulic radii. For parameter calculation, see Figure 2 .

[0037] According to the hydraulic geometry theory proposed by Leopold and Maddock, a power function relationship is constructed between the water surface width w, the average water depth d of the section, the average flow velocity v and the river flow Q. The relationship is: (4) Where, w represents the water surface width; a represents the water surface width reference coefficient; Q represents the river channel flow; b represents the water surface width flow index; d represents the cross-sectional average water depth; c represents the cross-sectional average water depth reference coefficient; f represents the water depth flow index, which characterizes the response degree of the flow change to the average water depth, and satisfies 0 < f < 1; v represents the average flow velocity; k represents the average flow velocity reference coefficient; m represents the flow velocity flow index, which characterizes the influence of the flow change on the flow velocity, and satisfies 0 < m < 1.

[0038] Through the measured initial water surface width , the calculated initial flow and the corresponding water depth , flow velocity , the water surface width reference coefficient a, the cross-sectional average water depth reference coefficient c, the average flow velocity reference coefficient k, and the water surface width flow index b, the water depth flow index f, and the flow velocity flow index m are determined by fitting with the non-linear least squares method, and it is verified whether the parameters satisfy the constraint conditions b + f + m = 1 and a * c * k = 1; if not satisfied, the parameters of the river channel flow inversion model are re-adjusted for fitting until the reference coefficients and flow indices satisfy the constraint conditions, and finally the construction of the river channel flow inversion model is completed.

[0039] Input the water surface width data after threshold calibration into the constructed river and lake flow inversion model to obtain the flow at the target time , and its expression is: (5) Where, represents the flow at the target time; represents the calibrated water surface width data; represents the initial water surface width; represents the initial flow.

[0040] Step Three: Prediction of River and Lake Water Levels and Flows 3.1 Preprocessing of Prediction Model Data According to the target river, lake and their water area range files, download multiple publicly available satellite remote sensing precipitation products such as GPCC and CPM, and satellite remote sensing surface temperature products such as GLDAS and Landsat-8, and obtain the corresponding precipitation inversion data and surface temperature inversion data from them; Use the feature scaling method to normalize the river and lake water level inversion data set in Step One, the river channel flow inversion data set in Step Two, the water surface width data, and the precipitation inversion data and surface temperature inversion data in Step Three to ensure the comparability between different models and the accuracy of prediction.

[0041] The expression of the normalization process is: (6) in, represents the normalized data; X represents the original data; Indicates the minimum value of the original data; Indicates the maximum value of the original data.

[0042] 3.2 Construction of intelligent prediction model for river and lake water level and flow Machine learning models such as support vector regression, gradient boosting trees, and random forests, as well as deep learning models such as long-short-term memory networks and BP neural networks, were selected as prediction models for river and lake water levels and flows. In hydrological simulation and prediction, machine learning models such as support vector regression, gradient boosting trees, and random forests can effectively handle noise and outliers that may exist in the data, as well as capture complex nonlinear relationships in the data. Deep learning models such as long-short-term memory networks and BP neural networks can automatically learn and model these complex relationships when processing hydrological data involving a large number of nonlinear relationships and time series characteristics, providing accurate prediction results for the model.

[0043] Based on a variety of machine learning models and deep learning models, an intelligent model of river and lake water level and flow is constructed. From the perspectives of internal prediction and extrapolation prediction, the normalized data is divided into a training set and a test set. At the same time, a 10-fold cross-validation is introduced to train and validate the selected models (support vector regression, gradient boosting tree, random forest, long-term memory network, and BP neural network model). The mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R) between the predicted and actual values ​​of river and lake water level and flow are calculated on each fold validation set. 2 ; The mean and standard deviation of the 10 validation results are summarized, and the mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R) of each model on the validation set are compared. 2 , select the optimal model as the lake level and flow intelligent prediction model.

[0044] The mean square error MSE, mean absolute error MAE and determination coefficient R 2 The expressions are: (7) (8) (9) Among them, MSE represents mean square error; MAE represents mean absolute error; represents the coefficient of determination; represents the actual value of the i-th sample; represents the predicted value of the i-th sample; n represents the number of samples; Represents the average value of the actual sample value; i represents the index value of the current sample.

[0045] 3.3 Forecast of future river and lake water levels and flows The river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation inversion data, and surface temperature inversion dataset are aligned at the same time, and compared with the forecast data in the same time period, that is, the forecast data in the same historical period, and the average values ​​of the water surface width data, precipitation inversion data, and surface temperature inversion dataset in the current period are calculated. And the average value of the corresponding historical forecast data for the same period; Based on the average value of water surface width data, precipitation inversion data and surface temperature inversion data set in the current period The average value of the forecast data for the corresponding period of the previous history The correction coefficient is obtained by the ratio of The expression of the correction coefficient is: (10) in, represents the correction factor; Represents the average value of the water surface width data, precipitation inversion data, and surface temperature inversion data set in the current period; It represents the average value of the water surface width forecast data, precipitation inversion forecast data, and surface temperature inversion forecast data set in the current period.

[0046] The correction factor is compared with the forecast data for the future predetermined period of time. Multiply them together to obtain the corrected water surface width data, precipitation inversion data and surface temperature inversion data; The revised water surface width data, precipitation inversion data and surface temperature inversion data set are input into the optimal lake level and flow intelligent prediction model to obtain the changes in river and lake water levels and flows in the corresponding period in the future. Example 3

[0047] A river and lake water level and flow prediction system, comprising: The river and lake water level remote sensing inversion module is used to obtain satellite radar altimeter data, river and lake water system data, precipitation, and surface temperature inversion data based on target rivers, lakes, and their basin range files; use the intersection of the satellite radar altimeter data and the river and lake water system data as a virtual hydrological station observation point, and use the virtual hydrological observation point to extract waveform data from the satellite radar altimeter data; process the waveform data and decompose the processed waveform data using a wavelet tracking algorithm to extract water surface elevation information; construct an elevation profile group based on the water surface elevation information, and establish an initial water level time series; fit the initial water level time series, perform elevation reselection in combination with the elevation profile group, and generate a river and lake water level inversion dataset; a river cross-sectional flow estimation module, configured to obtain normalized difference water index data based on the satellite radar altimeter data, extract water surface width data from the normalized difference water index data, and invert the river flow using the measured river cross-sectional data and the Manning formula to form a river flow inversion dataset; The river and lake water level and flow prediction module is used to input the river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation and surface temperature inversion data into the pre-built river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

[0048] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A method for predicting river and lake water levels and flows, characterized in that: include: Obtain satellite radar altimeter data, river and lake system data, precipitation inversion data, and surface temperature inversion data based on target rivers, lakes, and their basin range files; The intersection of the satellite radar altimeter data and the river and lake system data is used as a virtual hydrological station observation point, and the waveform data in the satellite radar altimeter data is extracted using the virtual hydrological observation point; Processing the waveform data and decomposing the processed waveform data by a wavelet tracking algorithm to extract water surface elevation information, constructing an elevation profile group based on the water surface elevation information, and establishing an initial water level time series; Fitting the initial water level time series, performing elevation reselection in combination with the elevation profile group, and generating a river and lake water level inversion dataset; Acquiring normalized difference water index data based on the satellite radar altimeter data, extracting water surface width data from the normalized difference water index data, and inverting river flow using the measured river cross-section data and the Manning formula to form a river flow inversion data set; The river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation inversion data and surface temperature inversion data are input into the pre-built river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

2. The method for predicting river and lake water levels and flows according to claim 1, characterized in that: Extracting waveform data from satellite radar altimeter data using the virtual hydrological observation point includes: Using a water body frequency analysis method to determine the water surface range of the virtual hydrological observation point and set a buffer zone; Using the satellite radar altimeter data, screening valid virtual hydrological observation points within the buffer zone; 1 Hz waveform data and 20 Hz auxiliary data are extracted from the effective virtual hydrological observation points.

3. The method for predicting river and lake water levels and flows according to claim 2, characterized in that: The step of constructing an elevation profile group and establishing an initial water level time series includes: Selecting a quasi-rectangular area containing the valid virtual hydrological observation point, and in the quasi-rectangular area, sequentially performing waveform resampling, average waveform synthesis, peak detection, and filtering on the extracted 1 Hz waveform data to obtain a filtered average waveform; Performing wavelet decomposition on the average waveform to obtain a plurality of wavelets, and matching the decomposed wavelets with effective peak values ​​of the average waveform; The OCOG algorithm is used to reshape the waveform of the decomposed wavelet, and then the wavelet tracking algorithm is used to calculate the elevation of the decomposed wavelet to extract the water surface elevation value. Based on the matching result of the decomposed wavelet and the effective peak value of the average waveform, a multi-objective optimization function is constructed with the standard deviation of the water surface elevation value and the peak value of the wavelet as constraints to screen out the effective water surface elevation value; According to the effective water surface elevation value, an elevation profile group is constructed, and the elevation profiles at different time points are arranged in chronological order to generate an initial water level time series.

4. The method for predicting river and lake water levels and flows according to claim 3, characterized in that: The generation of river and lake water level inversion datasets includes: Performing time series fitting on the initial water level time series using the SG function to obtain first fitting water level data; Performing sliding median fitting on the first fitting water level data to generate second fitting water level data, and identifying abnormal noise point data that deviates from the median in the second fitting water level data; Combined with the constraints of the multi-objective optimization function, the abnormal noise point data is subjected to elevation reselection and discrimination. If the elevation value of the abnormal noise point data exceeds the constraint range of the multi-objective optimization function, the abnormal noise point data is subjected to noise revision to obtain a revised water level time series; if the abnormal noise point data does not exceed the constraint range, the original data is retained and directly output as a valid water level time series; Combining the revised water level time series with the effective water level time series to obtain the latest water level time series; The latest water level time series is integrated with the effective water surface elevation value to generate a river and lake water level inversion dataset.

5. The method for predicting river and lake water levels and flows according to claim 1, characterized in that: The river flow inversion dataset is formed, including: Conduct field measurements at preset cross-section locations of target rivers and lakes to obtain river cross-section data, and simultaneously record the width of the measured river cross-section; The normalized difference water index data is inverted using a hybrid pixel decomposition algorithm to obtain the remote sensing water surface width of the target rivers and lakes; Comparing the measured width of the river section with the remotely sensed water surface width, adjusting the threshold of the normalized difference water index data according to the comparison result, and extracting water surface width data in batches from the normalized difference water index data based on the adjusted threshold; The batch-extracted water surface width data are input into the pre-built river flow inversion model to generate a river flow inversion dataset.

6. The method for predicting river and lake water levels and flows according to claim 5, characterized in that: Based on the adjusted threshold, the expression for batch extracting water surface width data from the normalized difference water index data is: ； Where W represents the water surface width data; PA represents the actual area of ​​the pixel; VL represents the length of the valley area of ​​interest; The value of the normalized difference water index of the pixel; represents the land NDWI threshold; represents the NDWI threshold of the water area; x represents the total number of pixels in the valley area of ​​interest; j represents the index value of the pixel currently being processed.

7. The method for predicting river and lake water levels and flows according to claim 5, characterized in that: The construction process of the river flow inversion model includes: Calculating the cross-sectional area, wetted perimeter, and hydraulic radius of the measured river channel section based on the geometric features of the measured river channel section; wherein the geometric features include the shape of the river channel section, the width of the water surface, and the average water depth of the section; Based on the cross-sectional area and wetted perimeter, the initial flow rate is calculated using the Manning formula. ; Based on hydraulic geometry theory, a power function relationship between water surface width w, average cross-section water depth d, average flow velocity v and river discharge Q is constructed; The initial water surface width measured by , the calculated initial flow and the corresponding water depth The nonlinear least squares method is used to fit and determine the water surface width benchmark coefficient a, the cross-section average water depth benchmark coefficient c, the average flow velocity benchmark coefficient k, the water surface width flow index b, the water depth flow index f, and the flow velocity flow index m, and verify whether the benchmark coefficient and flow index meet the constraints b + f + m = 1, and a * c * k = 1; if not, readjust the parameters of the river flow inversion model until the benchmark coefficient and flow index meet the constraints, and finally complete the construction of the river flow inversion model.

8. The method for predicting river and lake water levels and flows according to claim 7, characterized in that: The initial flow The expression is: ； in, represents the initial flow rate; Indicates the cross-sectional area of ​​water flow; represents the wetted perimeter; n represents the roughness; Indicates the slope.

9. The method for predicting river and lake water levels and flows according to claim 1, characterized in that: The changes in river and lake water levels and flows in the corresponding future periods include: Normalizing the river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation inversion data, and surface temperature inversion data using a feature scaling method; An intelligent model for river and lake water levels and flows was constructed based on a variety of machine learning and deep learning models. The normalized data was divided into training and test sets, and 10-fold cross-validation was introduced to train and validate the selected models. The machine learning models included support vector regression, gradient boosting trees, and random forests; the deep learning models included long-term memory networks and BP neural network models. The mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R) between the predicted and actual values ​​of the water level and flow of the river and lake are calculated on each fold validation set. 2 , summarize the average and standard deviation of the 10 validation results, and compare the mean square error MSE, mean absolute error MAE and determination coefficient R of each model on the validation set 2 , select the optimal model as the intelligent prediction model for river and lake water level and flow; Compare and analyze the river and lake water level inversion data set, river flow inversion data set, water surface width data, precipitation inversion data, and surface temperature inversion data with the historical forecast data for the same period, and calculate the average value of the water surface width data, precipitation inversion data, and surface temperature inversion data in the current period and the average value of the corresponding historical forecast data for the same period; The correction coefficient is obtained based on the ratio of the average value of the water surface width data, precipitation inversion data and surface temperature inversion data set in the current period to the average value of the corresponding historical forecast data in the same period; Multiplying the correction coefficient by the forecast data for a predetermined future time period to obtain revised water surface width data, precipitation inversion data, and surface temperature inversion data; The revised water surface width data, precipitation inversion data and surface temperature inversion data set are input into the optimal river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

10. A river and lake water level and flow prediction system, characterized in that: include: The river and lake water level remote sensing inversion module is used to obtain satellite radar altimeter data, river and lake water system data, precipitation, and surface temperature inversion data based on target rivers, lakes, and their basin range files; use the intersection of the satellite radar altimeter data and the river and lake water system data as a virtual hydrological station observation point, and use the virtual hydrological observation point to extract waveform data from the satellite radar altimeter data; process the waveform data and decompose the processed waveform data using a wavelet tracking algorithm to extract water surface elevation information; construct an elevation profile group based on the water surface elevation information, and establish an initial water level time series; fit the initial water level time series, perform elevation reselection in combination with the elevation profile group, and generate a river and lake water level inversion dataset; a river cross-sectional flow estimation module, configured to obtain normalized difference water index data based on the satellite radar altimeter data, extract water surface width data from the normalized difference water index data, and invert the river flow using the measured river cross-sectional data and the Manning formula to form a river flow inversion dataset; The river and lake water level and flow prediction module is used to input the river and lake water level inversion dataset, river flow inversion dataset, water surface width data, precipitation and surface temperature inversion data into the pre-built river and lake water level and flow intelligent prediction model to obtain the changes in river and lake water level and flow in the corresponding period in the future.

Images