LSTM-based water level flow boundary prediction method and device, and computer storage medium

Abstract

The invention discloses a water level flow boundary prediction method and device based on LSTM and a computer storage medium, and relates to the technical field of hydrological simulation and prediction, and the method comprises the steps: obtaining historical topographic feature data, water level data and flow data of a target river channel, and carrying out the data cleaning and standardization; taking the upstream flow, the water level data and the topographic features of the riverway as input, taking the downstream outlet boundary water level of the riverway as output, and building an LSTM model comprising at least two LSTM layers; data sets corresponding to different historical terrains of the river channel are used for training and verification, and the prediction precision of the LSTM model is evaluated; the downstream water level is predicted based on future flow process data under the past terrain condition of the river channel; on the basis of the current terrain condition of the river channel, the downstream water level is restored based on past flow process data, and the problem that in the prior art, when the terrain changes drastically, the influence of the terrain change is not fully considered in water level flow boundary prediction, and consequently the simulation result is distorted can be solved.

Description

Technical Field

[0001] This invention relates to the field of hydrological simulation and prediction technology, specifically to a method, device, and computer storage medium for predicting water level and discharge boundaries based on LSTM. Background Technology

[0002] In hydrological research and numerical simulation, the water level-discharge relationship is a crucial element, and its accuracy directly affects the reliability of the simulation results. For example, if a river channel has undergone long-term, long-distance scouring, the scouring depth in some areas will increase, causing drastic changes in the river channel topography, which in turn leads to a significant alteration in the water level-discharge relationship.

[0003] Currently, when using numerical simulations to invert the impact of river channel changes on the hydrodynamic performance of the river, the common practice is to use water level-discharge process curves based on the current or past conditions, without considering changes in the water flow-discharge relationship under different river topography. However, changes in the water level-discharge relationship caused by topographic changes mean that using the same water flow-discharge relationship curve under different erosion and deposition terrains will inevitably lead to distorted simulation results, greatly interfere with hydrodynamic analysis, and even directly cause erroneous results.

[0004] Taking the Jiujiang section of the Yangtze River as an example, the water level-discharge relationship varies significantly under different topographical conditions (e.g., the topography in 1998 before the Three Gorges Dam impoundment and the topography in 2021 after impoundment). Furthermore, topographic erosion has caused an overall drop in the water level of the Jiujiang section, significantly impacting its discharge. In numerical simulations of river flow, it is typically necessary to specify the upstream discharge boundary conditions and the downstream water level outlet boundary conditions. When researchers want to analyze changes in hydrodynamic performance such as river discharge caused by topographical changes, they need to compare the changes in river hydrodynamic performance under the same discharge process under different topographical conditions. However, existing methods use the same downstream outlet water level boundary conditions under different topographical conditions. In reality, after the Three Gorges Dam impoundment, significant erosion occurred in the Yangtze River channel, resulting in completely different downstream water levels under the same upstream discharge under different topographical conditions. Calculating using the same water level would lead to significant calculation errors, or even completely distorted results.

[0005] Existing methods fail to fully consider the key factor of topographic change, making it difficult to accurately provide the adaptive water level process at the downstream outlet boundary when the upstream flow process is the same. They cannot truly reflect the impact of topographic change on water level and flow. Therefore, there is an urgent need for a prediction method that can solve the above problems. Summary of the Invention

[0006] This invention provides a method, device, and computer storage medium for predicting water level and discharge boundaries based on LSTM, which can overcome the problem in the prior art that the water level and discharge boundary prediction does not fully consider the influence of terrain changes when the terrain changes drastically, resulting in distorted simulation results.

[0007] LSTM-based water level-discharge boundary prediction methods include: Acquire historical topographic features, water level, and flow data of the target river channel, and perform data cleaning and standardization; Using upstream flow, water level data, and topographic features of the river as input, and downstream outlet boundary water level of the river as output, an LSTM model containing at least two LSTM layers is constructed. The prediction accuracy of the LSTM model was evaluated by training and validation using datasets corresponding to different historical topography of the river channel. Based on the past topography of the river channel, the downstream water level is predicted based on future flow process data; based on the current topography of the river channel, the downstream water level is reconstructed based on past flow process data.

[0008] Furthermore, the data cleaning process includes the following steps: Identify missing values ​​in water level and / or flow rate data; If the missing amount is determined to be within the normal range, linear interpolation is used to supplement it based on adjacent data. Identify outliers in water level and / or flow rate data; If an outlier is determined to be outside the normal range and is caused by a malfunction in the monitoring equipment, it should be deleted and replaced by interpolation.

[0009] Furthermore, data standardization includes the following processes: Convert paper maps into raster data format using a digitizer; Extract terrain feature information, including elevation and slope features; The terrain features, water level data, and flow data are standardized so that their values ​​range from 0 to 1.

[0010] Furthermore, the topographic feature data includes topographic elevation data obtained through topographic surveying and river width data obtained through analysis using a geographic information system; the water level and flow data come from measured data from one or more hydrological stations above the river section, and the time span covers the time period corresponding to the collected topographic feature data.

[0011] Furthermore, the first LSTM layer in the LSTM model contains 64 neurons to initially extract temporal features and nonlinear relationships from the data; the second LSTM layer contains 32 neurons to further process and filter the features; and the output layer contains 1 neuron to output the predicted water level at the downstream outlet boundary.

[0012] Furthermore, the LSTM model uses the sigmoid activation function for the input gate, forget gate, and output gate, and the tanh activation function for the cell state update part; the Adam optimization algorithm is selected, and the learning rate is set to 0.001.

[0013] Furthermore, the process of predicting downstream water level based on past river topographic conditions and future flow process data includes: inputting the upstream water level and flow process data for the future period into an LSTM model trained with the past topographic and corresponding water level data, and the output of the LSTM model is the predicted value of the downstream outlet boundary water level for the future period under the past topographic conditions.

[0014] Furthermore, the process of reconstructing the downstream water level based on the current topographic conditions of the river channel and past flow data includes: inputting the upstream water level and flow data of the past period into an LSTM model trained with the current topography and corresponding water level data; the output of the LSTM model is the predicted value of the downstream outlet boundary water level under the current topography for the past period.

[0015] A computer device includes a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described above.

[0016] A computer-readable storage medium storing computer instructions for causing a computer to perform the above-described method.

[0017] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following: 1. By constructing an LSTM model, it is possible to accurately predict the adaptive water level process at the downstream outlet boundary under different terrain conditions and with the same upstream flow process. This effectively avoids the problem of simulation result distortion caused by using the same water level-flow process curve, and improves the accuracy of hydrodynamic analysis.

[0018] 2. It enables water level prediction based on past topography and water level reconstruction based on past topography, making hydrological data under different topographic conditions comparable and providing reliable data support for related research on the impact of topographic changes on river hydrodynamic performance, discharge capacity and hydrological system.

[0019] 3. The LSTM network is adopted, which has a powerful ability to process time series data and capture long-term dependencies. It can better fit the complex nonlinear relationship between water level, flow rate and topography, thus improving the prediction accuracy.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a data flow diagram disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the LSTM model structure disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the training error curve and training gradient changes disclosed in an embodiment of the present invention; Figure 4 This is a histogram of the model training prediction error distribution disclosed in an embodiment of the present invention; Figure 5 This is a schematic diagram comparing the model training prediction values ​​and actual values ​​disclosed in an embodiment of the present invention. Figure 6 This is a scatter plot of linear comparison between model training predictions and actual values ​​disclosed in an embodiment of the present invention. Figure 7 This is the first comparison result between the actual value and the predicted value disclosed in the embodiments of the present invention; Figure 8 This is a second comparison result between the actual value and the predicted value disclosed in the embodiments of the present invention; Figure 9 This is the third comparison result between the actual value and the predicted value disclosed in the embodiments of the present invention. Detailed Implementation

[0023] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0024] like Figure 1 As shown, the present invention proposes an LSTM-based method for predicting water level and discharge boundaries. This method is applicable when river topography undergoes drastic changes due to natural or human factors, enabling accurate prediction of future and past water level and discharge boundaries. It provides reliable data support for research such as river hydrodynamic performance analysis, and includes the following steps: S1, Data Collection and Preprocessing S11, Data Collection: Collect topographic feature data of the target river channel at different times, as well as corresponding water level and flow data. Topographic feature data includes topographic elevation data obtained through topographic surveying and river width data obtained through analysis using a geographic information system (GIS). Water level and flow data are obtained from measured data from one or more hydrological stations upstream of the river section, and the time span should cover the time period corresponding to the collected topographic feature data.

[0025] S12, Data Cleaning: For missing values ​​in water level and flow data, if the number of missing values ​​is small, linear interpolation is used to supplement them based on adjacent data. If obvious abnormal values ​​are found (far exceeding the normal range of historical data for that period), and it is determined that they are caused by a monitoring equipment malfunction, they are deleted and supplemented by interpolation.

[0026] The missing data for water level and flow rate constitutes the overall missing percentage, where the total sample size for the entire period is defined as the missing percentage. The number of missing points is defined as missing proportion ,exist (Direct linear interpolation).

[0027] The method for identifying obvious outliers is as follows: ,in, This is the current water level / flow rate data. This represents the historical average of water level / flow rate data.

[0028] S13, Data Standardization: The terrain feature data is digitized by converting paper maps into raster data format using a digitizer. GIS software is used to extract terrain features such as elevation and slope. The terrain feature information, as well as water level and flow data, are standardized so that their values ​​are between 0 and 1 for subsequent model processing. The standardization process is called normalization.

[0029] For example, taking the Jiujiang section of the Yangtze River as an example, actual topographic feature data were collected for 1998 (before the Three Gorges Dam impoundment) and 2021 (after the Three Gorges Dam impoundment), including topographic elevation data obtained through topographic surveying and river width data obtained through GIS analysis. Simultaneously, measured water level and flow data from multiple hydrological stations along this section were collected from 1996 to 2023, covering the time period corresponding to the collected topographic feature data.

[0030] During the cleaning of water level and flow data, it was discovered that water level data for two consecutive days in March 2005 was missing. Linear interpolation was used to supplement the missing values ​​based on the water level values ​​of adjacent dates. During the data review, a set of flow data for July 2010 was found to be significantly abnormal (far exceeding the normal range of historical data for that period). It was verified that this was due to a malfunction in the monitoring equipment; the abnormal value was deleted, and the data was supplemented using interpolation.

[0031] For terrain feature data, the paper map obtained from terrain surveying is digitized using a digitizer and converted into raster data format. Then, GIS software is used to extract terrain features such as elevation and slope, and the data is standardized so that the values ​​are between 0 and 1, which facilitates subsequent model processing.

[0032] S2, Construct an LSTM model and train it. S21, Determine the input and output: The input data includes the flow and water level process of the upstream station (one or more) (selecting daily flow data over a continuous period of time) and topographic feature data; the output data is the water level value at the downstream outlet boundary.

[0033] S22, Design the network structure: Build a network containing two LSTM layers. The number of neurons in the input layer is determined by the number of features in the input data. For example, if the input features are the water level and flow rate of a station, then the number of neurons in the input layer is 2. The first LSTM layer has 64 neurons, and the second LSTM layer has 32 neurons. The output layer has 1 neuron, corresponding to the predicted water level for the next day. The LSTM model structure is as follows: Figure 2 As shown.

[0034] S23, Selecting the activation function and optimization algorithm: In the LSTM unit, the input gate, forget gate, and output gate all use the sigmoid activation function, and the cell state update part uses the tanh activation function; the Adam optimization algorithm is selected, and the learning rate is set to 0.001.

[0035] For example, the input and output are determined as follows: the input data are the flow and water level processes of the upstream station (daily flow data for 2-3 consecutive years) and the topographic feature data of 1998 and 2021; the output data is the water level value at the downstream outlet boundary.

[0036] Network architecture design: Construct a network containing two LSTM layers. The input layer has 2 neurons (corresponding to the water level and flow characteristics of a station); the first LSTM layer has 64 neurons, and the second LSTM layer has 32 neurons; the output layer has 1 neuron (corresponding to the predicted water level for the next day).

[0037] Activation function and optimization algorithm selection: In the LSTM unit, the input gate, forget gate and output gate all use the sigmoid activation function, and the cell state update part uses the tanh activation function; the Adam optimization algorithm is selected, and the learning rate is set to 0.001.

[0038] Model training S24, Data Partitioning: The collected data containing different terrain features is divided into different parts, each corresponding to the water level-discharge relationship under different terrain features. For each terrain feature's dataset, it is divided into three parts: 70% as the training set, 15% as the validation set, and 15% as the test set.

[0039] S25, Training Process: The training set data is input into the model in batches, with each batch containing 32 samples. In each training epoch, the loss values ​​(using the mean squared error loss function) on both the training and validation sets are calculated. If the loss value on the validation set begins to show a fluctuating upward trend after a certain number of epochs, it indicates that the model may be overfitting, and training is terminated prematurely. Figure 3 The diagram illustrates the training error curve and gradient changes of the LSTM model. Figure 4 The histogram of the model training prediction error distribution is shown.

[0040] For example, the collected data from 1996 to 2023 is divided into two parts. The first part is the data from 1996 to 2000, which represents the water level-discharge relationship corresponding to the topography in 1998. The second part is the data from 2018 to 2023, which represents the water level-discharge relationship corresponding to the topography in 2021.

[0041] For the 2021 terrain dataset, 70% of the data from 2018 to 2023 was selected as the training set, 15% as the validation set, and 15% as the test set.

[0042] During training, the training set data was input into the model with 32 samples per batch. In each training epoch, the loss values ​​(using the mean squared error loss function) on both the training and validation sets were calculated. After 50 epochs of training, it was observed that the loss value on the validation set began to fluctuate and increase after the 30th epoch, indicating that the model might be overfitting. Therefore, training was terminated early at the 30th epoch.

[0043] S3, Model Validation and Evaluation The trained model was validated using a test set, and metrics such as root mean square error (RMSE) and mean absolute error (MAE) were calculated to evaluate the model's prediction accuracy. Compared to traditional water level and discharge prediction models that do not consider topographic changes, the RMSE and MAE of the model in this invention should be significantly reduced. Figure 5 This diagram illustrates the comparison between the model's training predictions and the actual values. Figure 6 This diagram illustrates a scatter plot showing a linear comparison between the model's training predictions and the actual values.

[0044] For example, the trained model was validated using a test set (data from 2022), and the calculated root mean square error (RMSE) was 0.25m and the mean absolute error (MAE) was 0.18m. Compared with traditional water level and discharge prediction models that do not consider topographic changes, the RMSE was reduced by 0.4m and the MAE was reduced by 0.3m, indicating that the model of the present invention has higher prediction accuracy.

[0045] S4, Future and Past Water Level-Discharge Boundary Prediction, such as Figures 7-9 As shown: Future prediction: To predict the downstream outlet boundary water level of a certain terrain in the future at a certain time, the upstream water level and flow process data for that future time are input into the model trained by the past terrain and the corresponding water level data. The output of the model is the predicted value of the downstream outlet boundary water level for that future time under the past terrain.

[0046] Past Reconstruction: To predict the downstream outlet boundary water level of a certain terrain over a certain period of time in the past, the upstream water level and flow process data of that period of time are input into the model trained by the terrain and the corresponding water level data. The output of the model is the predicted value of the downstream outlet boundary water level of that terrain over that period of time.

[0047] For example, future forecasting: Suppose we want to predict the water level at the downstream outlet boundary of the Jiujiang River in 2024, based on the topography of 1998, and we have known the upstream water level and flow process data for 2024. We input this data into a model trained with the 1998 topography and water level data from 1996 to 2000. The model can then output the predicted downstream outlet boundary water level for 2024. This result represents the change in downstream water level under the influence of the upstream flow process in 2024, given the 1998 topography.

[0048] Past Reconstruction: Suppose we want to predict the water level at the downstream outlet boundary of the Jiujiang River in 1998, based on the topography in 2021, and we have known the upstream water level and flow data for 1998. Inputting this data into a model trained with topographic data from 2021 and water level data from 2018 to 2023, the model can then output the predicted downstream outlet boundary water level for 1998.

[0049] For example, given the measured data of the Jiujiang River section on September 10, 2020, the goal is to reconstruct the downstream water level that should have been present in 2011 based on the topography. By inputting the upstream flow data from September 3 to September 9, 2020, the topographic feature data from 2011, and the hourly water level data from September 3 to September 9, 2020, into the model, the model output will be the downstream water level that should have been present on September 10, 2020, based on the 2011 topography. This reconstruction makes the hydrological data under the 2011 and 2020 topographic conditions comparable in terms of water level-discharge relationship, providing strong support for further analysis of the impact of topographic changes on the hydrological system.

[0050] In one embodiment, the acquisition of target river channel topographic feature data further includes the following methods: Historical topographic features, hydrodynamic data, and external influencing factors of the target river channel were collected. Among them, historical topographic features included cross-sectional measurement data and DEM data of the target river channel at multiple periods; hydrodynamic data included water level, flow rate, flow velocity, shear stress distribution, and sediment concentration at the corresponding periods; and external influencing factors included the sediment load of the watershed.

[0051] The topographic feature data is parameterized by cross-section to construct a topographic feature vector. The construction of the topographic feature vector includes parameterizing the cross-section, such as the width of the main channel, the average water depth, the change in riverbed elevation, and the curvature.

[0052] A prediction model based on spatiotemporal deep learning is constructed, taking hydrodynamic factors and historical topographic sequences as inputs and downstream riverbed elevation changes or cross-sectional morphology as outputs. The prediction model construction includes the use of a spatiotemporal deep learning model: the time series part uses LSTM / GRU to extract long-term evolution patterns; the spatial feature part uses convolutional neural networks (CNN) or graph neural networks (GNN) to extract cross-sectional spatial distribution patterns; the model input is the hydrodynamic-topographic feature data sequence of the past N years; the model output is the riverbed elevation distribution or cross-sectional morphology for the next M years.

[0053] The model was trained using historical data, and the accuracy was evaluated using a validation set. The model was trained using historical data, and the validation was performed using a test set for a portion of the time period. The accuracy was evaluated using error metrics (RMSE, riverbed elevation deviation, and cross-sectional area change rate).

[0054] Input the future hydrodynamic scenario and the current topography to obtain the prediction results of the future river channel topography.

[0055] The specific process includes: inputting future water level data and flow data along with current topographic features, predicting the riverbed erosion and deposition evolution in a specified future year, and outputting a predicted cross-sectional morphology change map and a riverbed elevation change curve.

[0056] For example, the topographic feature data of a future target river channel output above can be used as input data for training. If it is to predict the downstream outlet boundary water level of a future topography at a certain time in the present or past, the upstream water level and flow process data at that time in the present or past can be input into the model trained by the future topography and the corresponding water level data. The output of the model is the predicted value of the downstream outlet boundary water level at that time in the present or past under the future topography, which can further expand the prediction range.

[0057] A computer device includes a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described above.

[0058] A computer-readable storage medium storing computer instructions for causing a computer to perform the above-described method.

[0059] This invention fully considers the impact of topographic changes on the water level-discharge relationship. By constructing an LSTM model, it can accurately predict the adaptive water level process at the downstream outlet boundary under different topographic conditions and the same upstream discharge process. This effectively avoids the problem of simulation result distortion caused by using the same water level-discharge process curve, and improves the accuracy of hydrodynamic analysis.

[0060] It enables water level prediction based on past topography and water level reconstruction based on past topography, making hydrological data under different topographic conditions comparable. This provides reliable data support for research on the impact of topographic changes on river hydrodynamic performance, discharge capacity, and hydrological systems.

[0061] By employing an LSTM network, which has the powerful ability to process time-series data and capture long-term dependencies, it can better fit the complex nonlinear relationship between water level, flow rate and topography, thus improving prediction accuracy.

[0062] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.

[0063] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are explicitly stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby explicitly incorporated into the detailed description, with each claim representing a separate preferred embodiment of the invention.

[0064] Those skilled in the art will also understand that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments herein can be implemented as electronic hardware, computer software, or a combination thereof. To clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art can implement the described functionality in alternative ways for each specific application; however, such implementation decisions should not be construed as departing from the scope of this disclosure.

[0065] The steps of the methods or algorithms described in conjunction with the embodiments herein can be directly embodied in hardware, software modules executed by a processor, or a combination thereof. The software modules can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in a user terminal. Alternatively, the processor and storage medium can exist as discrete components in the user terminal.

[0066] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. This software code can be stored in memory units and executed by a processor. The memory units can be implemented within the processor or outside the processor; in the latter case, they are communicatively coupled to the processor via various means, as is well known in the art.

[0067] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that the various embodiments can be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term "comprising" as used in the specification or claims is interpreted in a manner similar to the term "including," as interpreted when used as a conjunction in the claims. Additionally, the use of any term "or" in the specification of the claims is intended to mean "non-exclusive or."

Claims

1. A LSTM-based water level flow boundary prediction method, characterized in that, The application relates to a method for predicting a downstream water level of a river channel. The method comprises the following steps: acquiring historical terrain feature data, water level data and flow data of a target river channel, and performing data cleaning and standardization; taking upstream flow data, water level data and terrain features of the river channel as inputs, and taking downstream outlet boundary water level of the river channel as an output, to build an LSTM model comprising at least two LSTM layers; training and verifying the LSTM model by using a data set corresponding to different historical terrains of the river channel, and evaluating the prediction accuracy of the LSTM model; predicting the downstream water level based on future flow process data under a past terrain condition of the river channel; 2. The method of claim 1, wherein, restoring the downstream water level based on past flow process data under a present terrain condition of the river channel. The data cleaning process comprises the following steps: identifying missing values of the water level data and / or the flow data; if the missing amount is within a normal range, the missing values are supplemented by linear interpolation according to adjacent data before and after the missing values; identifying abnormal values of the water level data and / or the flow data; 3. The method of claim 2, wherein, if the abnormal values are determined to be caused by a monitoring device fault, the abnormal values are deleted and supplemented by interpolation. The data standardization process comprises the following steps: converting a paper map into a raster data format through a digitizer; extracting terrain feature information, including elevation and slope feature information; 4. The method of claim 2, wherein, standardizing the terrain feature information, the water level data and the flow data, so that the values of the data are within a range of 0-1.

5. The method of claim 1, wherein, The terrain feature data comprises terrain elevation data obtained through terrain measurement and river channel width data obtained by using a geographic information system; the water level and flow data are measured data from one or more hydrological stations on the river section, and the time span covers a time period corresponding to the collected terrain feature data.

6. The method of claim 1, wherein, The first LSTM layer of the LSTM model comprises 64 neurons, which are used to preliminarily extract time sequence features and nonlinear relationships in the data; the second LSTM layer comprises 32 neurons, which are used to further process and screen the features; and the output layer comprises one neuron, which is used to output a water level prediction value of a downstream outlet boundary.

7. The method of claim 1, wherein, The input gate, the forgetting gate and the output gate of the LSTM model all adopt a sigmoid activation function, and the cell state update part adopts a tanh activation function; the optimization algorithm is an Adam optimization algorithm, and the learning rate is set to 0.

001.

8. The method of claim 1, wherein, The process of predicting the downstream water level based on the past terrain condition of the river channel and the future flow process data comprises the following steps: inputting future upstream water level and flow process data of the period into the LSTM model trained by the past terrain and corresponding water level data, and the output result of the LSTM model is a prediction value of the downstream outlet boundary water level of the future period under the past terrain.

9. A computer device, comprising: The process of restoring the downstream water level based on the present terrain condition of the river channel and the past flow process data comprises the following steps: inputting past upstream water level and flow process data of the period into the LSTM model trained by the present terrain and corresponding water level data, and the output result of the LSTM model is a prediction value of the downstream outlet boundary water level of the past period under the present terrain. The application further relates to a system for predicting a downstream water level of a river channel. A memory and a processor, which are connected in communication with each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method of any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1-8.

Images