Hydrological model time-varying parameter prediction method based on ensemble Kalman filtering and deep residual network

Abstract

The invention discloses a hydrological model time-varying parameter prediction method based on ensemble Kalman filtering and a deep residual network. The hydrological model time-varying parameter prediction method comprises the following steps: acquiring watershed multi-source data; acquiring time-varying parameter historical time sequence data of the hydrological model by using an ensemble Kalman filtering method; constructing a deep residual network prediction model based on the convolutional neural network; and training the model, inputting the obtained hydrological and meteorological data in the future climate mode into the trained model, and outputting hydrological model time-varying parameters in the future scene. According to the method, the historical data of the time-varying parameters of the hydrological model are acquired by using a data assimilation method, the future change of the parameters of the hydrological model is predicted through the deep residual network and future climate mode data, and a brand new data driving method is provided for dynamic prediction of the parameters of the hydrological model; and powerful technical support is provided for water resource management and climate change challenges in the future.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydrological prediction, and particularly relates to a method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network. Background Technique

[0002] In previous hydrological simulation studies, most key parameters (such as basin water storage capacity, infiltration rate, etc.) were regarded as fixed values. However, these parameters are often affected by multiple factors such as climate change, land use change, and human activities, and thus exhibit significant time-varying characteristics, resulting in a decline in the prediction accuracy of hydrological models when facing complex and non-stationary hydrological processes. Therefore, obtaining historical stage data of time-varying parameters of hydrological models and predicting future trends not only helps improve the accuracy of hydrological simulation, but also provides a more scientific and timely decision-making basis for flood warning, water resource management, etc.

[0003] In order to obtain historical stage data of time-varying parameters of hydrological models, many studies have started to use data assimilation methods, including the Kalman filter method. However, traditional Kalman filter usually assumes that the system is linear or weakly non-linear, and it is difficult to cope with complex hydrological processes.

[0004] With the development of artificial intelligence technology, many studies have used machine learning and deep learning to solve various problems in the field of hydrology. In the field of future prediction of hydrological data, deep learning algorithms such as convolutional neural network (CNN) and long short-term memory network (LSTM) have more advantages in solving non-linear problems compared with traditional statistical methods. However, deep learning algorithms represented by CNN still have certain defects. For example, they are prone to problems such as gradient disappearance when constructing multi-level non-linear mappings, which restricts the application of the deep network structure of the model and cannot fully exert the advantages of deep learning in complex pattern recognition. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network, so as to solve the technical problems in the prior art that the parameters of hydrological models cannot be effectively obtained using observed data and the future data trends cannot be reasonably predicted under the background of climate change.

[0006] To solve the above technical problems, the present invention is implemented as follows: A method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network includes the following steps: Step S1: Obtain multi-source data: Integrate multi-dimensional and multi-source data as input information for subsequent model construction and parameter calibration.

[0007] Step S2: Use the Ensemble Kalman Filter method to obtain the historical time series data of the time-varying parameters of the hydrological model. The Ensemble Kalman Filter can better handle the nonlinear problems in the system by sampling the uncertainty of the state, ensuring high estimation accuracy even in nonlinear situations. At the same time, the dynamic data assimilation process of the Ensemble Kalman Filter method can continuously adjust the hydrological model parameters to capture the time-varying characteristics of the system, thereby improving the accuracy of model prediction, while traditional methods often lack such a real-time update mechanism.

[0008] Step S3: Construct a prediction model based on the combination of a convolutional neural network and a deep residual network. Through this prediction model, extract the local features in the input data, capture the global features through multi-level processing, and then capture the complex nonlinear relationship between multi-source data and the time-varying parameters of the hydrological model. The deep residual network ResNet can automatically extract the local features in the input data and capture complex nonlinear relationships through multi-layer convolution and pooling operations. In particular, it can solve the degradation problem by making the gradient flow into the external network more easily.

[0009] Step S4: Combine the hydrological data and meteorological data obtained in Step S1, and the time-varying parameters of the hydrological model constructed in Step S2, into a historical time series dataset in chronological order. Divide this dataset into a training set and a test set according to a ratio of 7:3. Use the mean square error MSE as the loss function, use the training set to train the deep residual network model constructed in Step S3, use the validation set to verify the trained model, and use the mean square loss function for target optimization to determine the coefficient R 2 as the evaluation index to evaluate the trained prediction model.

[0010] Step S5: Input the multi-source hydrological and meteorological data obtained from future climate models into the trained prediction model, and output the time-varying parameters of the hydrological model under future scenarios.

[0011] For further optimization, the data obtained in Step S1 includes historical meteorological data, soil data, land use data, terrain data, runoff observation and runoff characteristic data obtained through meteorological data products.

[0012] For further optimization, Step S2 specifically includes: Step S2.1: Set the state variable as the time-varying parameter of the hydrological model, and the observation variable as a certain input data obtained in Step S1. Construct a state transition equation with process noise and an observation equation with observation error: State transition equation: (1); Observation equation: (2); In the above formula, is the state variable for predicting the time-varying parameters of the hydrological model at the current moment; C t-1 is the value of the time-varying parameter of the hydrological model at the previous moment; f () is the non-linear function describing the evolution law of the state variable; is the process noise vector, representing the random change of the state variable caused by various uncertain factors between time steps (t - 1) and t; y t is the observed variable obtained at the current moment, H () is the observation operator that maps the state variable to the observation space to describe the observed variable; v t is the observation error vector at the current moment, representing the random change of the observed variable caused by measurement error.

[0013] Step S2.2: Initialize the background field and predict the state value at the next moment, specifically including: Step S2.2.1: Set an initial state value based on prior knowledge or historical data as the basis for subsequent update prediction.

[0014] Step S2.2.2: Combine the predicted state and new observation data, and use the Kalman gain to update the state of the hydrological model parameters: (3); In the above formula, C t is the value of the time-varying parameter of the updated hydrological model at the current moment; K is the weight factor, used to determine the influence degree of the observation data on the state correction during the update process; y t is the observation data obtained at the current moment.

[0015] For further optimization, the specific content of step S3 is as follows: Step S3.1: Concatenate the hydrological data and meteorological data at time t obtained in step S1 and the value of the time-varying parameter of the hydrological model at time t obtained in step S2 according to the dimensional requirements to form the input vector of the prediction model: (4); In the above formula, represents the input vector at time t; H(t) , M(t) , C(t) respectively represent hydrological data, meteorological data, and the value of the time-varying parameter of the hydrological model.

[0016] Step S3.2: Input the input vector into the convolutional neural network to convert it into an initial input feature map suitable for convolution operations. Use the convolutional kernel to slide on the input feature map to perform convolution operations. Multiple convolutional kernels operate in parallel to simultaneously extract various local features of the input data. After the convolution operation, add the bias term to the convolution result and then use it as the input of the activation function. After being processed by the activation function, a new feature map is obtained, and the expression is: (5); In the above formula, is the feature map output after convolution, bias adjustment, and non-linear activation, and will be used as the input of the next layer (if any) l to continue participating in the feature extraction process of the network; is the activation function, which enables the model to better fit complex non-linear relationships; is the weight of the convolutional kernel of the th layer. The convolutional kernel slides on the input feature map l to perform convolution operations and extract local feature patterns in the input data, represents the convolution operation; different (l) W (l) values determine the types of features captured by the convolutional kernel, such as edges, textures, etc. represents the output feature map of the l− 1st layer, which contains the data features extracted from the previous layers; is the bias term of the l th layer, which offsets and adjusts the result after the convolution operation, increases the fitting ability of the model, avoids all outputs being concentrated near the origin, and enables the model to better learn data features.

[0017] Then, input the feature map output by the convolutional layer into the pooling layer. The pooling layer uses the pooling function Pool() to downsample the feature map and extract the maximum value within the local area, which is expressed as: (6); In the above formula, Pool(⋅) is the pooling function used to downsample the feature map output by the convolutional layer.

[0018] Step S3.3 constructs a deep residual network using residual connections: To construct a deep network and avoid the problem of gradient disappearance, residual connections are adopted in each residual module. Specifically, in the first residual module, directly add the initial input feature map and the feature map obtained after being processed by Step S3.2 as the output of this residual module; multiple residual modules are stacked in sequence, and the output of the previous residual module is used as the input of the next residual module; through multi-layer residual calculations, the entire deep residual network is expressed as: (7); In the above formula, is the residual function of the entire deep residual network, reflecting the feature information extracted through all levels from the initial input feature map to the final output; is the set of all parameters to be learned in the model, including the convolutional kernels and biases of each layer in all residual modules. These parameters are continuously adjusted during the training process of the network so that the residual function can better fit the data. L is the total number of residual modules, that is, the number of layers of the network, l ∈ [1, L ; is the output feature map after being processed by L residual modules, and it is the final feature representation after the initial input feature map undergoes multiple convolutional, activation, and pooling operations.

[0019] Step S3.4: Combine the historical time series data of the original time-varying parameters of the hydrological model C(t) with the residual information learned by the deep residual network to predict the time-varying parameters of the hydrological model at future times, and its expression is: (8); In the above formula, is the predicted value of the time-varying parameters of the hydrological model at future times.

[0020] Through this step, based on the historical time series data C(t) of the time-varying parameters of the hydrological model, a deep residual network is superimposed, and the residual learned according to the current hydrological, meteorological and other data is used to correct and predict the future parameters of the hydrological model. For example, if the historical time series data of the time-varying parameters of the hydrological model represents the current basin water storage capacity, and the residual captures the non-linear impact of current precipitation, temperature and other meteorological data on the water storage capacity, then the sum of the two can obtain the predicted value of the water storage capacity at future times.

[0021] Compared with the prior art, the beneficial effects of the present invention are: 1. The present invention uses a combined data assimilation method of ensemble Kalman filter and hydrological model to construct a historical time series data of hydrological model parameters corrected based on observational data. This process effectively integrates multi-source data such as water level, meteorology, and underlying surface, obtains historical data closer to the true value, and makes up for the uncertainty of a single model.

[0022] 2. The prediction model designed by the present invention, which combines a convolutional neural network and a deep residual network, captures the complex non-linear relationship between various natural conditions such as hydrology and meteorology and the parameters of the hydrological model. The use of residual connections effectively alleviates the problem of gradient disappearance, enables the model to have a deeper number of layers, and thus improves the expression ability.

[0023] 3. The deep residual network model described in the present invention can, in the case of hydrological and meteorological data provided by future climate models, use the current state as a basis and correct the future change trend of hydrological model parameters by learning the residuals. It provides a brand-new data-driven method for the dynamic prediction of future hydrological model parameters, thereby providing stronger technical support for future water resource management and coping with climate change challenges. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flowchart of the method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network according to the present invention; Figure 2 is a schematic diagram of the deep residual neural network model according to the present invention; Figure 3 is the time-varying trend of the water storage capacity of different continents in the future scenario; Figure 4 is the time-varying trend of the water storage capacity of different climate zones in the future scenario. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The technical solution of the present invention will be described in detail below in conjunction with the embodiments, but the protection scope of the present invention is not limited to the described embodiments.

[0026] In this embodiment, calculations and analyses are performed on a total of 2,683 river basins worldwide.

[0027] As Figure 1 shown, the method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network specifically includes the following steps: Step S1: Obtain multi-source data of global river basins required for calculations. The data used in this embodiment includes historical meteorological data, soil data, land use data, terrain data, runoff observation and runoff characteristic data at the global scale obtained through meteorological products, and future hydrological and meteorological data obtained through climate models.

[0028] All data uses a spatial resolution of 0.5°. Given the different spatio-temporal spans of different data sources, the time span of all historical data is selected as 1980 - 2015. The specific data and sources can be seen in Table 1: Table 1 Types and Sources of Obtained Data ; In addition, 20 climate models were selected in this embodiment to predict future climate information. All data are from the sixth phase of the Coupled Model Intercomparison Project (CMIP6). The data cover the historical period (1970 - 2014), the near - future period (2015 - 2059), and the far - future period (2060 - 2100). The future climate models are shown in Table 2.

[0029] Table 2 Climate Models ;

[0030] Step S2: Use the Ensemble Kalman Filter method to construct the historical time - series data of the time - varying parameters of the hydrological model.

[0031] In this embodiment, the catchment water storage capacity (CWSC) is used as the target parameter of the hydrological model. At the beginning of the calculation, a random water storage capacity value is given to each catchment. Then, the Ensemble Kalman Filter algorithm is used to identify the parameters, and the method of updating the hydrological model parameters and state variables simultaneously is adopted. Specifically, it is expressed as: Step S2.1: Based on the hydrological model, the prediction process of the Ensemble Kalman Filter algorithm is represented by the following equation: (9); In the above formula, is the CWSC value predicted based on the time - varying parameters of the hydrological model at the current time; C t-1 is the CWSC value of the hydrological model at the previous time; f () is the non - linear function describing the evolution law of the state variables; is the process noise, representing the state change caused by internal or external uncertainties of the system.

[0032] Step S2.2: Initialize the background field and predict the state value at the next time, which specifically includes:

[0033] Step S2.2.1: Set an initial state value based on prior knowledge or historical data as the basis for subsequent prediction.

[0034] Step S2.2.2: Combine the predicted state and the new observation data, and use the Kalman gain to update the state of the hydrological model parameters: (10); In the above formula, C t is the updated CWSC value of the hydrological model at the current time; K is the weight factor, which is used to determine the influence degree of the observation data on the state correction during the update process; y t is the actual observation data obtained at the current time, which is the historical observation data of hydrology, meteorology, and underlying surface in this example; His a mapping operator used to map CWSC to the observation space.

[0035] In this embodiment, through the above calculations, the global basin water storage capacity CWSC values from 1980 to 2015 are obtained. Among them, the basin water storage capacity values on the grid at 23.75°N and 31.25°E are shown in Table 3.

[0036] Table 3 Basin water storage capacity CWSC values at 23.75°N and 31.25°E from 1980 to 2015 ; Step S3: Construct a prediction model based on the combination of a convolutional neural network CNN and a deep residual network (Deep Residual Network, ResNet). The entire network structure can not only extract local features in the input data, but also capture global features through multi-level processing, and then establish a fine mapping relationship. The schematic diagram of the deep residual network model structure is as Figure 2 shown. The specific steps are as follows: Step 3.1: Concatenate the hydrological and meteorological data obtained in Step 1 and the time series data of the basin water storage capacity parameters in Step 2 to form the input vector of the model, which is expressed as: (11); In the above formula, represents the input vector at time t; H(t) , M(t) , C(t) respectively represent the historical time series data of hydrology, meteorology, and water storage capacity.

[0037] Step 3.2: Input the input vector into the convolutional neural network, which is transformed into an initial input feature map suitable for convolution operations. Use the convolution kernel to slide on the input feature map to perform convolution operations, and multiple convolution kernels operate in parallel to extract multiple local features of the input data at the same time; after the convolution operation is completed, add the bias term to the convolution result and then use it as the input of the activation function. After being processed by the activation function, a new feature map is obtained, which is: (12); In the above formula, is the feature map output by the l th layer after convolution, bias adjustment, and non-linear activation, and will be used as the input of the next layer (if any) to continue participating in the feature extraction process of the network; is the activation function, which enables the model to better fit complex non-linear relationships; is the convolution kernel weight of the l th layer. The convolution kernel slides on the input feature map Slide up to perform a convolution operation to extract local feature patterns from the input data. The symbol * represents the convolution operation. Different W (l) values determine the types of features captured by the convolution kernel, such as edges, textures, etc. To represent the output feature map of the l− first layer, which contains various data features extracted from the previous layers; is the bias term of the l layer, which offsets and adjusts the result after the convolution operation, increases the fitting ability of the model, avoids all outputs concentrating near the origin, and enables the model to better learn data features.

[0038] Then, the feature map output by the convolutional layer is input into the pooling layer. The pooling layer uses the pooling function Pool() to downsample the feature map and extract the maximum value within the local area.

[0039] Step S3.3 constructs a deep residual network using residual connections: To build a deep network and avoid the vanishing gradient problem, residual connections are adopted in each residual module. Specifically, in a residual module, the initial input feature map and the feature map obtained after being processed in Step S3.2 are directly added together as the output of this residual module; multiple residual modules are stacked in sequence, and the output of the previous residual module serves as the input of the next residual module; through multiple layers of residual calculations, the entire deep residual network is expressed as: (13); In the above formula, is the residual function learned by the entire deep residual network, reflecting the feature information extracted by all the layers passed from the input to the final output; is the set of all parameters to be learned in the model, including the convolution kernels and biases of each layer in all residual modules; L is the total number of residual modules, that is, the number of layers of the network, l ∈[1, L , is the output feature map after being processed by L residual modules, which is the final feature representation of the initial input feature map after multiple layers of convolution, activation, and pooling operations.

[0040] Step S3.4: Combine the historical time series data of the original time-varying parameters of the hydrological model C(t) with the residual information learned by the deep residual network to predict the time-varying parameters of the hydrological model at future times, and its expression is: (14); In the above formula, is the predicted value of the time-varying parameters of the hydrological model at future times.

[0041] Step S4: Combine the hydrological data and meteorological data obtained in Step S1, and the time-varying parameters of the hydrological model constructed in Step S2, into a historical time-series dataset in chronological order. Divide this dataset into a training set and a test set in a ratio of 7:3. Use the mean squared error (MSE) as the loss function, and use the training set to train the deep residual network model constructed in Step S3, and use the validation set to validate the trained model; use the mean squared loss function for objective optimization to determine the coefficient of determination R 2 as the evaluation metric.

[0042] In this embodiment, the coefficient of determination R of the final trained model in the training set and the test set 2 is 0.9291 and 0.8273 respectively, and the model performance is good, meeting the requirements of simulation and prediction.

[0043] Step S5: Input the hydrological and meteorological data in the future climate pattern obtained in Step S1 into the trained network model, and output the predicted value of the time-varying parameter basin water storage capacity of the hydrological model under future scenarios.

[0044] In this embodiment, the global basin water storage capacity is calculated and predicted, and finally the distribution of the global basin water storage capacity for many years in the future (2015 - 2100) under the SSP2-4.5 and SSP5-8.5 emission scenarios is obtained; among them, SSP2-4.5 represents a medium greenhouse gas emission scenario, and SSP5-8.5 represents a very high greenhouse gas emission scenario.

[0045] Generally speaking, there are obvious differences in the basin water storage capacity among different continents and regions, reflecting the comprehensive influence of various factors such as climate, terrain, and vegetation on water resource storage. Among them, in regions such as the Americas (especially the Amazon Basin in South America), central Africa, and southeastern and southern Asia, the basin water storage capacity is relatively high; while in arid and semi-arid regions, such as the Sahara Desert in northern Africa and the inland regions of central Asia, the basin water storage capacity is low. This is because humid regions have abundant precipitation and good vegetation cover, which is conducive to water resource storage; arid regions have scarce precipitation and strong evaporation, making it difficult to retain water resources.

[0046] In addition, under the SSP2-4.5 scenario, the changes in the global hydrological system are relatively moderate, and the changes in the distribution of basin water storage capacity can be predicted and adapted to a certain extent. While under the SSP5-8.5 scenario, more intense climate changes will be triggered. The high radiative forcing leads to the instability of the climate system, changing the precipitation pattern, and making the changes in the distribution of basin water storage capacity more complex. There may be sudden increases and decreases in the water storage capacity in some areas, and frequent alternations between droughts and floods, posing greater challenges to regional water resource management and ecosystem stability.

[0047] Figure 3 are the time-varying trend lines of the water storage capacity of different continents under future scenarios; among them,Figure 3 Among them, (a), (b), (c), (d), (e), and (f) successively show the time-varying trend lines of the water storage capacity of Asia, Africa, Oceania, Europe, North America, and South America. In the figure, AWI-SSP245 represents the predicted water storage capacity of river basins on different continents globally under the medium greenhouse gas emission scenario using the AWI-CM-1-1-MR climate model developed by the Alfred Wegener Institute for Polar and Marine Research in Germany; AWI-SSP585 represents the predicted water storage capacity of river basins on different continents globally under the very high greenhouse gas emission scenario using the AWI-CM-1-1-MR climate model developed by the Alfred Wegener Institute for Polar and Marine Research in Germany. MPI-SSP245 represents the predicted water storage capacity of river basins on different continents globally under the medium greenhouse gas emission scenario using the AWI-CM-1-1-MR climate model developed by the Max Planck Institute for Meteorology in Germany; MPI-SSP585 represents the predicted water storage capacity of river basins on different continents globally under the very high greenhouse gas emission scenario using the AWI-CM-1-1-MR climate model developed by the Max Planck Institute for Meteorology in Germany.

[0048] Figure 4 It is the time-varying trend line of the water storage capacity of different climate zones globally in the future scenario. Figure 4 Among them, (a), (b), (c), (d), (e), and (f) successively show the time-varying trends of the water storage capacity of the globe, the humid zone, the equatorial zone, the arid zone, the inland zone, and the polar zone. From three aspects of spatial distribution, regional differences, and time-varying trends, it reveals the impact of climate change on the global hydrological system, providing multi-scale support for scientific research, policy-making, and water resource management.

[0049] As mentioned above, it is only an implementation example of the present invention in a specific river basin, but the protection scope of the present invention is not limited thereto. If those skilled in the art are inspired by it and design methods and embodiments similar to this technical solution without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network, characterized in that It includes the following steps: Step S1: Obtain multi-source data of the basin, and integrate multi-dimensional and multi-source data as the input information for subsequent model construction and parameter calibration; Step S2: Use the Ensemble Kalman Filter method to obtain historical time-series data of the time-varying parameters of the hydrological model; Step S3: Construct a prediction model based on the combination of a convolutional neural network and a deep residual network. Through this prediction model, extract local features from the input data, capture global features through multi-level processing, and then obtain the complex non-linear relationship between multi-source data and the historical time-series data of the time-varying parameters of the hydrological model; Step S4: Combine the data obtained in Step S1 and the historical time-series data of the time-varying parameters of the hydrological model constructed in Step S2 into a historical time-series data set in chronological order, and divide this data set into a training set and a test set in a ratio of 7:3; use the training set to train the prediction model constructed in Step S3, and use the validation set to validate the trained prediction model, and perform objective optimization with the mean square error MSE as the loss function to determine the coefficient R 2 as the evaluation index to evaluate the trained prediction model; Step S5: Input the multi-source hydrological and meteorological data obtained from future climate models into the trained prediction model, and output the time-varying parameters of the hydrological model under future scenarios.

2. The time-varying parameter prediction method for a hydrological model based on ensemble Kalman filter and deep residual network according to claim 1, wherein The multi-source data obtained in step S1 includes historical meteorological data, soil data, land use data, terrain data, runoff observation and runoff characteristic data obtained through meteorological data products.

3. The time-varying parameter prediction method of the hydrological model based on the ensemble Kalman filter and the deep residual network according to claim 2, characterized in that Step S2 specifically includes: Step S2.1: Set the state variable as the time-varying parameter of the hydrological model, and the observation variable as a certain input data obtained in step S1. Construct a state transition equation with process noise and an observation equation with observation error: State transition equation: (1); Observation equation: (2); In the above formula, is the state variable predicted based on the time-varying parameters of the hydrological model at the current moment; C t-1 is the value of the time-varying parameter of the hydrological model at the previous moment; f () is the non-linear function describing the evolution law of the state variable; is the process noise vector, representing the random change of the state variable caused by various uncertain factors between time step (t - 1) and t; y t is the observed variable obtained at the current moment, H () is the observation operator that maps the state variable to the observation space to describe the observed variable; v t is the observation error vector at the current moment, representing the random change of the observed variable caused by measurement errors; Step S2.2: Initialize the background field and predict the state value at the next moment, specifically including: Step S2.2.1: Set an initial state value based on prior knowledge or historical data as the basis for subsequent prediction; Step S2.2.2: Combine the predicted state and new observation data, and update the state of the hydrological model parameters using the Kalman gain. The expression is: （3）； In the above formula, C t is the time-varying parameter value of the hydrological model at the current moment after update; K is the weight factor, which is used to determine the influence degree of the observed data on the state correction during the update process; y t is the observed data obtained at the current moment.

4. The method for predicting time-varying parameters of a hydrological model based on ensemble Kalman filter and deep residual network according to claim 3, characterized in that Step S3 specifically includes: Step S3.1: Concatenate the hydrological data and meteorological data at time t obtained in step S1, and the time-varying parameter value of the hydrological model at time t obtained in step S2 according to the dimensional requirements to form the input vector of the prediction model: （4）； In the above formula, represents the input vector at time t; H(t) , M(t) , C(t) respectively represent hydrological data, meteorological data, and time-varying parameter values of the hydrological model; Step S3.2: Input the input vector into the convolutional neural network to convert it into an initial input feature map suitable for convolution operations. Use the convolution kernel to slide on the input feature map to perform convolution operations. Multiple convolution kernels operate in parallel to extract multiple local features of the input data at the same time. After the convolution operation is completed, add the bias term to the convolution result, and then use it as the input of the activation function. After being processed by the activation function, a new feature map is obtained. The expression is: （5）； In the above formula, is the feature map output after convolution, bias adjustment, and non-linear activation for the l th layer; is the activation function; is the convolutional kernel weight for the l th layer; represents the output feature map of the l− 1st layer; represents the convolution operation; is the bias term for the l th layer; Then, input the feature map output by the convolutional layer into the pooling layer, and use the pooling function Pool() to downsample the feature map to extract the maximum value within the local area, which is expressed as: （6）； In the above formula, Pool(⋅) is the pooling function used to downsample the feature map output by the convolutional layer; Step S3.3 uses residual connections to construct a deep residual network: In the residual module, directly add the initial input feature map and the feature map obtained after being processed in step S3.2 as the output of the first residual module. Multiple residual modules are stacked in sequence, and the output of the previous residual module is used as the input of the next residual module. Through multi-layer residual calculations, the entire deep residual network is obtained, which is expressed as: （7）； In the above formula, is the residual function of the entire deep residual network, reflecting the feature information extracted through all levels from the initial input feature map to the final output; is the set of all parameters to be learned in the model, including the convolutional kernels and biases of each layer in all residual modules; L is the total number of residual modules, that is, the number of layers of the network, l ∈ [1, L ; is the output feature map after being processed by L residual modules, which is the final feature representation after the initial input feature map undergoes multiple convolutional, activation, and pooling operations. Step S3.4: Combine the historical time series data of the original time-varying parameters of the hydrological model C(t) with the residual information obtained by the deep residual network to predict the time-varying parameters of the hydrological model at future moments, and its expression is: （8）； In the above formula, is the predicted value of the time-varying parameter of the hydrological model at a future time.

Images