Flood forecasting method, device and system coupling physical mechanism and deep learning

Abstract

The application discloses a flood forecasting method and device coupled with a physical mechanism and deep learning, and a system thereof. The flood forecasting method comprises the following steps: acquiring DEM data, hydro-meteorological data and static attribute data of a target basin; dividing the target basin into a plurality of sub-basins based on the DEM data, extracting a river network, and calculating static attribute data of the river network, wherein the river network comprises a plurality of river channels; inputting the hydro-meteorological data of each sub-basin, the static attribute data of the basin and the static attribute data of the river network into a pre-trained distributed mixed hydrological model to obtain simulated flow of a hydrological observation station, and completing flood forecasting; and the distributed mixed hydrological model is coupled with the physical mechanism and the deep learning. The distributed mixed hydrological model has good interpretability, can realize high-precision daily runoff and flood simulation, and has high calculation efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of flood forecasting, and particularly relates to a flood forecasting method, device and system coupling physical mechanisms and deep learning. BACKGROUND

[0002] Flood, as one of the most destructive natural disasters, not only poses a serious threat to human safety, but also has a significant impact on infrastructure and socio-economic development. Flood forecasting models, as an important non-structural tool for flood management, play an irreplaceable role in protecting life and property safety, promoting sustainable use of water resources, supporting water infrastructure construction and operation, etc. Therefore, accurately predicting flood events, especially in ungauged basins, and implementing effective disaster prevention and mitigation measures in advance, is of great significance to mitigate the negative impact of floods.

[0003] Currently, flood forecasting models are mainly divided into two categories: process-driven models and data-driven models. Process-driven models are based on physical mechanisms and have clear physical interpretability, and can output additional state variables such as soil moisture and evapotranspiration. However, due to the complexity of the hydrological system, especially under the background of changes in hydrological conditions caused by climate change and human activities, it is challenging to accurately describe the rainfall-runoff process through precise physical equations. Data-driven models are trained through big data and usually perform well in simulation accuracy, and can predict the flow of unmeasured basins through regional training. However, such models lack physical interpretability and cannot output additional state variables through flow calibration alone, and are highly dependent on a large amount of training data. Currently, there is a lack of large sample flood data sets, which severely limits the application of data-driven models in operational flood forecasting.

[0004] In recent years, hybrid hydrological models that combine the advantages of process-driven models and data-driven models have gradually become a hot topic. By encoding physical mechanisms and neural networks in a unified deep learning framework and using backpropagation algorithm for synchronous parameter calibration, hybrid models not only achieve simulation performance comparable to data-driven models, but also maintain good physical interpretability. In addition, hybrid models can simultaneously simulate key hydrological variables such as evapotranspiration, soil moisture, and base flow, providing more comprehensive information support for flood forecasting. However, existing hybrid hydrological models still have deficiencies in considering spatial heterogeneity of precipitation and underlying surface, especially in the development of distributed hybrid hydrological models, which face two major challenges: first, distributed hybrid hydrological models need to be trained multiple times through recursive iteration to calibrate parameters, and river (reach) confluence methods that consider upstream and downstream flood evolution relationships significantly increase the computational burden of the model; second, how to reasonably determine sub-basin runoff parameters based on static properties of the underlying surface is still a key problem to be solved. SUMMARY

[0005] In response to the above problems, the present invention proposes a flood forecasting method, device and system that couples physical mechanisms and deep learning, aiming to effectively solve the difficulties of hybrid models in computational efficiency and accuracy.

[0006] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0007] In a first aspect, the present invention provides a flood forecasting method that couples physical mechanisms and deep learning, comprising:

[0008] Obtain DEM data, hydrological and meteorological data, and static attribute data of the target watershed;

[0009] Based on the DEM data, the target watershed is divided into a plurality of sub-watersheds, and a river network is extracted, and static attribute data of the river network is calculated, wherein the river network includes a plurality of river channels;

[0010] The hydrological and meteorological data of each sub-basin, the static attribute data of the basin, and the static attribute data of the river network are input into a pre-trained distributed hybrid hydrological model to obtain the simulated flow of the hydrological observation station and complete the flood forecast; the distributed hybrid hydrological model couples physical mechanisms and deep learning.

[0011] In combination with the first aspect, optionally, the distributed hybrid hydrological model includes a sub-basin runoff calculation model and a confluence calculation model that are sequentially arranged;

[0012] The mathematical expression of the sub-basin runoff calculation model is:

[0013] ,

[0014] The mathematical expression of the confluence calculation model is:

[0015] ,

[0016] Where, represents the outflow of sub-basin i at time t; HM represents the lumped hydrological model; represents the hydrological and meteorological data of sub-basin i at time t; represents the parameters of the lumped hydrological model of sub-basin i; represents the static attribute data of sub-basin i; FNN1 represents the feedforward neural network used to learn the mapping relationship between the static attribute data of the sub-basin and the parameters of the lumped hydrological model; W FNN1 Represents the parameters of FNN1; It represents the discharge of sub-basin i at time t after the outflow through the river channel or river section converges at the hydrological observation station; RM is the river channel confluence model or river section confluence model; parameters of the river confluence model or the river reach confluence model; represents static attribute data of the river or the river reach; FNN2 represents a feedforward neural network used for learning a mapping relationship between the static attribute data of the river or the river reach and the parameters of the river confluence model or the river reach confluence model; W FNN2 represents parameters of FNN2; represents a simulated flow of the hydrological observation station at time t.

[0017] With reference to the first aspect, optionally, the sub-basin is divided into a plurality of river reaches, and the river reach from the upstream to the downstream is denoted as river reach . When the Muskingum confluence calculation method is adopted, each river reach is divided into equal-length river reaches, and the river reach index satisfies , the outflow of the sub-basin is sequentially confluenced from the upstream to the downstream, and for the upstream river reach , the inflow thereof is the outflow of the corresponding sub-basin i, and the outflow of the river reach is calculated according to the following formula:

[0018] ,

[0019] In the formula, Q and Q respectively represent the outflow of the river reach at time t and t , Q and Q respectively represent the outflow of the sub-basin i at time t and t , and Q represents the lateral inflow of the river reach at time t ; for the downstream river reach , the inflow thereof is equal to the outflow of the upstream adjacent river reach , and the outflow of the river reach is calculated according to the following formula: ;

[0020] ;

[0021] In the formula, Q and Q respectively represent the outflow of the river reach at time t and t . ​​

[0022] Danghe section When , calculate according to the following formula Moment sub-basin The flow at the hydrological observation station after the outflow passes through the river section :

[0023] ;

[0024] Where, for Moment sub-basin The outflow is the flow at the hydrological observation station after the confluence of the river section;

[0025] Parameters C1, C2, C3 and C4 are calculated using the following formulas:

[0026] ,

[0027] Where, Indicates river channel The river section The storage time constant, Indicates river channel The river section The weight factor of Δt represents the calculation time step; if the river section Located in a sub-basin Inside, here , is the total number of sub-basins, then The runoff will be distributed to the river section as lateral inflow , Indicates river channel The river section exist The specific calculation formula for the lateral inflow at the moment is as follows:

[0028] ,

[0029] Where, Represents a subbasin Middle River Section The proportionality coefficient, Represents a subbasin At the moment The outflow, Represents a subbasin Middle River Section Average catchment area controlled, Represents a subbasin area;

[0030] The parameters and Adaptive learning by feedforward neural network FNN2, whose value is determined by the river section The specific calculation formula is as follows:

[0031] ,

[0032] Where, Indicates river channel The river section Static attribute data.

[0033] In combination with the first aspect, optionally, the sub-basin The river course to the downstream hydrological observation station is recorded as the river course When the diffusion wave confluence method is used, the movement of flood in the river channel is described by the convection-diffusion equation, and combined with the initial conditions and boundary conditions, we get Moment sub-basin The flow calculation formula of the outflow at the hydrological observation station is as follows:

[0034] ,

[0035] Where, for Moment sub-basin The outflow is the flow at the hydrological observation station after the confluence of the river section. Indicates river channel exist The inflow at time i is also the outflow of subbasin i; Indicates that river channel i is The impulse response function corresponding to time is, For the river length; Indicates the maximum time step considering historical inflow, which is a hyperparameter;

[0036] parameter and Adaptive learning by feedforward neural network FNN2, whose value is determined by the river The specific calculation formula is as follows:

[0037] ,

[0038] Where, Indicates river channel Static attribute data.

[0039] In combination with the first aspect, optionally, the loss function Loss of the confluence calculation model is:

[0040] ,

[0041] ,

[0042] In the formula, is the number of hydrological observation stations, N is the total number of hydrological observation stations, NSE represents the Nash efficiency coefficient, represents the time, represents the length of the time series, represents the observed flow of the sub-basin , represents the simulated flow of the sub-basin , represents the average value of the observed flow of the sub-basin .

[0043] In combination with the first aspect, optionally, when the convolutional neural network confluence method is adopted, each river has a unique static attribute, which corresponds to a different convolution kernel, the flow of the sub-basin outflow after confluence in the hydrological observation station The calculation formula is:

[0044] ,

[0045] In the formula, represents the weight of the corresponding convolution kernel of the river , represents the position of each weight, , represents the total number of weights, represents the static attribute data of the river , represents the outflow of the sub-basin at time.

[0046] In combination with the first aspect, optionally, the loss function Loss of the confluence calculation model is:

[0047] ,

[0048] ,

[0049] In the formula, λ1 and respectively represent the coefficients of each component, and the sum of the two is 1; and respectively represent the weights of positions and in the corresponding convolution kernel of the river , represents the total number of sub-basins; is an activation function; an index representing a convolution kernel with the largest weight, ; representing a river channel a first-order difference of weights in a corresponding convolution kernel; is the number of hydrological observation stations, N is the total number of hydrological observation stations, and NSE represents the Nash efficiency coefficient.

[0050] In combination with the first aspect, optionally, the training process of the distributed hybrid hydrological model comprises:

[0051] In the pre-training stage, the distributed hybrid hydrological model is pre-trained using daily scale hydro-meteorological data, basin static attribute data and static attribute data of the river network, to obtain a pre-training model;

[0052] In the fine-tuning stage, the pre-training model is trained using hourly scale hydro-meteorological data, basin static attribute data and static attribute data of the river network, to obtain a final distributed hybrid hydrological model.

[0053] In the second aspect, the application provides a flood forecasting device coupling physical mechanisms and deep learning, comprising:

[0054] A data acquisition module is configured to acquire DEM data, hydro-meteorological data and basin static attribute data of a target basin;

[0055] A sub-basin division module is configured to divide the target basin into a plurality of sub-basins based on the DEM data, extract a river network, and calculate static attribute data of the river network, wherein the river network comprises a plurality of river channels.

[0056] A forecasting module is configured to input the hydro-meteorological data of each sub-basin, the basin static attribute data and the static attribute data of the river network into a pre-trained distributed hybrid hydrological model, to obtain simulated flow of a hydrological observation station and complete flood forecasting; the distributed hybrid hydrological model couples physical mechanisms and deep learning.

[0057] In the third aspect, the application provides a flood forecasting system coupling physical mechanisms and deep learning, comprising a storage medium and a processor;

[0058] The storage medium is configured to store instructions;

[0059] The processor is configured to operate according to the instructions to execute the method according to any one of the first aspect.

[0060] Compared with the prior art, the application has the following beneficial effects:

[0061] The application provides a distributed hybrid hydrological model capable of coupling a physical mechanism and deep learning, which can guarantee high-precision daily runoff and flood simulation under the premise of good interpretability and improve model calculation efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor under the premise of the drawings.

[0063] Figure 1 The flow of the flood forecasting method of coupling a physical mechanism and deep learning of an embodiment of the present application is intended.

[0064] Figure 2 The schematic diagram of a river basin of an embodiment of the present application is shown.

[0065] Figure 3 The schematic diagram of model performance comparison in time division strategy of an embodiment of the present application is shown.

[0066] Figure 4 The schematic diagram of model performance comparison in time-space division strategy of an embodiment of the present application is shown.

[0067] Figure 5 The curve of the prediction model in the daily flow and flood model of the GX station of an embodiment of the present application is shown.

[0068] Figure 6 The schematic diagram of the loss curve and time consumption of the prediction model of an embodiment of the present application is shown. DETAILED DESCRIPTION

[0069] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0070] In addition, if the description of "first", "second" and the like is involved in the embodiments of the present application, the description of "first", "second" and the like is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can be explicitly or implicitly included at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor in the protection scope required by the present application.

[0071] Embodiment 1

[0072] The flood forecasting method coupling physical mechanism and deep learning is provided in the embodiments of the present application, comprising the following steps:

[0073] (1) obtaining DEM data, hydro-meteorological data and basin static attribute data of a target basin;

[0074] (2) dividing the target basin into a plurality of sub-basins based on the DEM data, extracting a river network, and calculating static attribute data of the river network, the river network comprising a plurality of river channels;

[0075] (3) inputting the hydro-meteorological data of each sub-basin, the basin static attribute data and the static attribute data of the river network into a pre-trained distributed hybrid hydrological model to obtain simulated flow of a hydrological observation station, and completing flood forecasting; the distributed hybrid hydrological model couples physical mechanism and deep learning.

[0076] In the above scheme, a distributed hybrid hydrological model capable of coupling physical mechanism and deep learning is proposed, which can realize high-precision daily runoff and flood simulation under the premise of ensuring good interpretability, and can improve the calculation efficiency of the model.

[0077] In one specific embodiment of the present application, the distributed hybrid hydrological model comprises a sub-basin runoff calculation model and a confluence calculation model arranged in sequence.

[0078] The mathematical expression of the sub-basin runoff calculation model is:

[0079] ,

[0080] The mathematical expression of the confluence calculation model is:

[0081] ,

[0082] In the formula, Qi(t) represents the outflow of sub-basin i at time t; HM represents the lumped hydrological model; Xi(t) represents the hydro-meteorological data of sub-basin i at time t; Pi represents the parameters of the lumped hydrological model of sub-basin i; Xi represents the basin static attribute data of sub-basin i; FNN1 represents a feedforward neural network for learning the mapping relationship between the basin static attribute data of sub-basin and the parameters of the lumped hydrological model; W FNN1 Pi represents the parameters of the lumped hydrological model of sub-basin i; Qi(t) represents the outflow of sub-basin i at time t; RM represents the river confluence model or the reach confluence model; Pi represents the parameters of the lumped hydrological model of sub-basin i; Xi represents the basin static attribute data of sub-basin i; FNN1 represents a feedforward neural network for learning the mapping relationship between the basin static attribute data of sub-basin and the parameters of the lumped hydrological model; W FNN2 Pi represents the parameters of the lumped hydrological model of sub-basin i; Qi(t) represents the outflow of sub-basin i at time t; RM represents the river confluence model or the reach confluence model;

[0083] In one specific embodiment of the present application, the sub-basins are divided into reaches , and the river confluence model or the reach confluence model is used to calculate the outflow of each reach . In one specific embodiment of the present application, the sub-basins are divided into reaches , and the river confluence model or the reach confluence model is used to calculate the outflow of each reach . In one specific embodiment of the present application, the sub-basins are divided into reaches .

[0084] ,

[0085] In one specific embodiment of the present application, the sub-basins are divided into reaches , and the river confluence model or the reach confluence model is used to calculate the outflow of each reach . In one specific embodiment of the present application, the sub-basins are divided into reaches , and the river confluence model or the reach confluence model is used to calculate the outflow of each reach . In one specific embodiment of the present application, the sub-basins are divided into reaches , and the river confluence model or the reach confluence model is used to calculate the outflow of each reach . In one specific embodiment of the present application, the sub-basins are divided into reaches , and the river confluence model or the reach confluence model is used to calculate the outflow of each reach Lateral inflow at the moment; for the downstream river section , , whose inflow is equal to the adjacent upstream reach The outflow of the river section The outflow is calculated as follows:

[0086] ;

[0087] Where, and Respectively represent river channels Middle River Section exist and The outflow of the moment;

[0088] Danghe section When , calculate according to the following formula Moment sub-basin The flow at the hydrological observation station after the outflow passes through the river section :

[0089]

[0090] Where, for Moment sub-basin The outflow is the flow at the hydrological observation station after the confluence of the river section;

[0091] Parameters C1, C2, C3 and C4 are calculated using the following formulas:

[0092] ,

[0093] Where, Indicates river channel The river section The storage time constant, Indicates river channel The river section The weight factor of Δt represents the calculation time step; if the river section Located in a sub-basin Inside, here , is the total number of sub-basins, then The runoff will be distributed to the river section as lateral inflow , Indicates river channel The river section exist The specific calculation formula for the lateral inflow at the moment is as follows:

[0094] ,

[0095] wherein, denotes a sub-basin of the middle reach of the river, denotes the outflow (i.e. the yield) of the sub-basin at the time , denotes the average catchment area controlled by the sub-basin of the middle reach of the river, denotes the area of the sub-basin ;

[0096] The parameters and are adaptively learned by a feedforward neural network FNN2, and the values thereof are determined by the static attribute data of the reach , and the specific calculation formula is as follows:

[0097] ,

[0098] wherein, denotes the static attribute data of the reach in the river channel .

[0099] In one specific implementation of the embodiment of the application, the river channel from the sub-basin to the downstream hydrological observation station is denoted as the river channel , when the diffuse wave (DW) confluence method is adopted, the movement of flood in the river channel is described by a convection diffusion equation, and combined with initial conditions and boundary conditions, a definite problem is formed, wherein the convection diffusion equation is:

[0100] ,

[0101] The mathematical expressions of the initial conditions and the boundary conditions are:

[0102] ,

[0103] wherein, denotes the flow at any position in the river channel; x is a spatial variable, i.e. any spatial position; t is a time variable, C denotes wave speed, and D denotes diffusion coefficient;

[0104] The convection diffusion equation is solved by using Laplace transform, and the following general solution is obtained:

[0105] ,

[0106] wherein, is the impulse response function of the convection diffusion equation under the initial conditions and the boundary conditions, represents the inflow of the river at time t - τ, , Indicates the maximum time length, which is a hyperparameter;

[0107] Finally got Moment sub-basin The flow calculation formula of the outflow at the hydrological observation station is as follows:

[0108] ,

[0109] Where, for Moment sub-basin The outflow is the flow at the hydrological observation station after the confluence of the river section. Indicates river channel exist The inflow at time i is also the outflow of subbasin i; Indicates that river channel i is The impulse response function corresponding to time is, For the river length; Indicates the maximum time step considering historical inflow, which is a hyperparameter;

[0110] parameter and Adaptive learning is performed by the feedforward neural network FNN2, and its value is determined by the static attribute data of the river channel i. The specific calculation formula is as follows:

[0111] ,

[0112] Where, Indicates river channel Static attribute data.

[0113] In a specific implementation of the embodiment of the present invention, the loss function Loss of the confluence calculation model is:

[0114] ,

[0115] ,

[0116] Where, is the number of the hydrological observation station, N is the total number of hydrological observation stations, NSE is the Nash efficiency coefficient, Indicates the moment, represents the length of the time series, Represents a subbasin The observed flow rate, Represents a subbasin The simulated flow rate, sub-basin the average value of the observed flow.

[0117] In an embodiment of the present application, when a convolutional neural network (CNN) confluence method is used, each river has a unique static attribute, which corresponds to a different convolution kernel, sub-basin the flow at the hydrological observation station after the outflow of the river confluence The calculation formula is:

[0118] ,

[0119] In the formula, the weight of the corresponding convolution kernel of the river, the position of each weight, , the total number of weights, the static attribute data of the river, the yield of the sub-basin at the time t. In the calculation formula of the loss function Loss of the confluence calculation model is:

[0120]

[0121] ,

[0122] ,

[0123] In the formula, λ1 and respectively represent the coefficients of each component, and the sum of the two is 1; and respectively represent the weights of the positions and in the corresponding convolution kernel of the river, the total number of sub-basins; is an activation function; the index of the convolution kernel with the largest weight, ; the first-order difference of the weight in the corresponding convolution kernel of the river. In an embodiment of the present application, the training process of the distributed mixed hydrological model comprises:

[0124] In an embodiment of the present application, the training process of the distributed mixed hydrological model comprises:

[0125] ​​​In the pre-training stage, the distributed hybrid hydrological model is pre-trained using daily scale hydro-meteorological data, basin static attribute data and static attribute data of river network to obtain a pre-training model;

[0126] In the fine-tuning stage, the pre-training model is trained using hourly scale hydro-meteorological data, basin static attribute data and static attribute data of river network to obtain a final distributed hybrid hydrological model.

[0127] The flood forecasting method in the embodiments of the present application will be described in detail below in combination with Figure 1 and a specific embodiment.

[0128] Step 1, input data preparation. For the target basin, daily scale or hourly scale precipitation and evapotranspiration data of hydrological observation stations in the target basin, daily flow data of hydrological observation stations (i.e. hydrological sites) and hourly scale field flood records (flood records are hourly flow data) are collected, and high-precision digital elevation model (DEM) data (resolution needs to be within 100 meters) and basin static attribute data are obtained. The basin static attribute data covers key attributes such as soil type, vegetation coverage and terrain features, providing basic data support for subsequent model construction.

[0129] Step 2, input data preparation. Based on the location of the outlet hydrological observation station of the basin and the DEM data, the basin boundary is extracted and the river network is divided, and the basin is further divided into several sub-basins (this process is realized by using existing technology). Combined with the sub-basin boundary and the data collected in step 1, the average precipitation and evapotranspiration time series of each sub-basin are extracted, and the corresponding basin static attribute data are obtained. In addition, based on the DEM data, the static attribute data of the river network from the outlet of each sub-basin to the hydrological observation station are calculated, which include the average length of the river, the slope, the elevation, the sinuosity and the upstream catchment area, providing spatial heterogeneity information for model construction.

[0130] Step 3: Construction of a distributed hybrid hydrological model. The construction of the distributed hybrid hydrological model needs to ensure differentiability and be fully programmed in a deep learning framework to ensure gradient calculation, thereby realizing the calibration of distributed hybrid hydrological model parameters. The model architecture mainly consists of two core modules: one is the sub-basin runoff calculation module, and the other is the river (river section) confluence calculation module. The sub-basin runoff calculation module adopts a traditional lumped hydrological model (HM), such as the Xinanjiang model (China), the HBV model (Sweden), the Tank model (Japan), the Sactomento model (USA), etc. In order to adapt to the characteristics of different sub-basins, a feedforward neural network (FNN1) is introduced to establish a mapping relationship between the static properties of each sub-basin and the corresponding HM physical parameters, so as to achieve refined and differentiated parameter configuration. Subsequently, the runoff results of each sub-basin will be input into the river (river section) confluence calculation module, and the final simulated flow will be obtained by simulating the propagation and convergence process of water flow in the river network. Assume that the basin is divided into several sub-basins, and the sub-basin number i satisfies , the sub-basin The river course to the downstream hydrological observation station is recorded as the river course The above specific process can be expressed as follows:

[0131] (1);

[0132] Where, (m 3 / s) indicates Moment sub-basin The yield; HM represents the lumped hydrological model; express Moment sub-basin Hydrological and meteorological data (including precipitation, evapotranspiration, etc.); Represents a subbasin Parameters of the lumped hydrological model; represents the static attribute data of sub-basin i, which needs to be normalized; FNN1 represents the feedforward neural network used to learn the mapping relationship between the static attribute data of the sub-basin and the parameters of the lumped hydrological model; W FNN1 Represents the parameters of FNN1; (m 3 / s) represents the discharge of sub-basin i at time t after the outflow passes through the river channel or river section and reaches the hydrological observation station; RM is the river channel confluence model or river section confluence model; are the parameters of the river channel confluence model or river reach confluence model; represents the static attribute data of a river channel or river section, which needs to be normalized; FNN2 represents a feedforward neural network used to learn the parameter mapping relationship between the static attribute data of a river channel or river section and the river channel confluence model or river section confluence model; W FNN2 Represents the parameters of FNN2; (m 3 Qsim(t) represents the simulated flow at the hydrological observation site at time t, which is the sum of the outflows of all sub-basins.

[0133] Three methods were developed in the river (river reach) confluence calculation process of this embodiment. Specifically, they include confluence methods based on Muskingum (MK), diffusion wave (DW) and convolutional neural network (CNN). Feedforward neural network (FNN2) is used to map the relationship between the static properties of each sub-basin to a specific cross-section of the river and the physical parameters of the corresponding confluence method.

[0134] For the MK confluence method, each river is divided into equal-length river reaches, each with a fixed length of 2 km, and the river reach index satisfies . The outflow of a sub-basin is sequentially confluenced from upstream ( ) to downstream ( ). For the upstream river reach , its inflow is the outflow of the corresponding sub-basin i, and the outflow of the river reach is calculated according to the following formula:

[0135] (2);

[0136] In the formula, and represent the outflows of river reach in river i at times and , respectively, and represent the outflows of sub-basin i at times and , respectively, represents the lateral inflow of river reach in river i at time . For the downstream river reach ( ), its inflow is equal to the outflow of the upstream adjacent river reach , and the outflow of the river reach is calculated according to the following formula:

[0137] (3);

[0138] When river reach , it is calculated according to the following formula:

[0139] (4);

[0140] ​The parameters C1, C2, C3 and C4 in the formulas (2), (3) and (4) are calculated by the following formula:

[0141] (5);

[0142] In the formula, represents the storage time constant (parameter) of the river section in the river channel, represents the weight factor (parameter) of the river section in the river channel, and Δt represents the calculation time step. If the river section is located in the interior of a sub-basin , here , the runoff of the sub-basin will be distributed as a lateral inflow to the river section , and the specific calculation formula is as follows: (6);

[0143] In the formula,

[0144] represents the proportion coefficient of the river section in the sub-basin , represents the runoff (i.e. outflow) of the sub-basin at time t, represents the average catchment area controlled by the river section in the sub-basin , and represents the area of the sub-basin . The parameters and in the MK confluence method are adaptively learned by the feedforward neural network FNN2, and their values are determined by the static attributes of the river section , and the specific calculation formula is as follows: (7);

[0145] In the formula, represents the static attribute data of the river section j in the river channel i.

[0146] For the DW confluence method, the runoff of each sub-basin to a specific cross section is calculated in parallel based on the concept of unit line. When the downstream section of the river channel is a free downstream boundary not affected by factors such as backwater effect, the movement of flood in the river channel can be described by the convection-diffusion equation (8), and combined with the initial condition and boundary condition given by equation (9), a definite problem is formed:

[0147]

[0148] ​​​​​(8);

[0149] (9);

[0150] where, is the flow rate at any location in the river, x is the spatial variable, i.e., any spatial location, t is the time variable, C represents the wave speed, and D represents the diffusion coefficient. The above equation can be solved using the Laplace transform to obtain the general solution as follows:

[0151] (10);

[0152] where, is the impulse response function of equation (8) under the initial and boundary conditions of equation (9), represents the inflow of the river at time t - τ, , represents the maximum time length, which is a hyperparameter. Specifically, for the outflow of any sub-basin , its flow rate at the hydrological observation station can be calculated using the following equation:

[0153] (11);

[0154] where, represents the inflow of river i at time t - τ (which is also the outflow of sub-basin i), represents the impulse response function corresponding to river i, where the specific meaning of x is the length of river i. The parameters and in the DW confluence method are adaptively learned by the feedforward neural network FNN2, and their values are determined by the static attributes of river i. The specific calculation formula is as follows:

[0155] (12);

[0156] In the above equation, represents the static attribute data of river i.

[0157] For the CNN confluence method, the inspiration comes from the similarity between CNN and the unit line structure concept. The core of one-dimensional CNN is to use a fixed-size convolution kernel to perform convolution operation along the specified dimension with a defined stride, thereby effectively extracting features from the input data. Although the size of the convolution kernel remains unchanged, its weights can be updated through training iterations. Given the inflow time series of a river, setting the stride to 1, the CNN confluence method assumes that the convolution kernel function is similar to a unit line, allowing the outflow of the river to be calculated by convolution. In this invention, we extend its application to river confluence, where each river has unique static attributes, corresponding to a different convolution kernel. The weights of the convolution kernel are determined by mapping these attributes through the feedforward neural network FNN2, as follows:

[0158] (13);

[0159] wherein, denotes the river corresponding to the weight of the convolution kernel, denotes the position of each weight, , denotes the total number of weights, denotes the static attribute data of the river , denotes the runoff of the sub-basin at time , and denotes the outflow of river i at time t. To establish a connection between the convolution kernel and the unit line and enhance the interpretability of the model, an additional physical constraint is imposed on . Specifically, the activation function of the output layer of FNN2 is set to Softmax (a general function in mathematics), ensuring that falls within the range of 0 to 1 and the sum is 1. In addition, a regularization constraint term is added to the loss function of the neural network to ensure that the shape of the convolution kernel mimics the unimodal form of the unit line.

[0160] In the distributed hybrid hydrological model of the invention, the loss function Loss of the MK and DW confluence methods is as follows:

[0161] (14);

[0162] (15);

[0163] wherein, is the number of hydrological observation stations, N is the total number of hydrological observation stations, NSE represents the Nash efficiency coefficient, denotes the time, denotes the length of the time series, denotes the observed flow, denotes the simulated flow, denotes the average value of the observed flow. For the loss of the CNN confluence method, a regularization term is added, and the specific formula is as follows:

[0164] (16);

[0165] (17);

[0166] wherein λ1and λ2denote the coefficients of each component, and the sum of the two is 1; denote the coefficients of each component, and the sum of the two is 1; and denote the weights at positions and and denote the weights at positions denotes the total number of sub-basins; is an activation function; denotes the index of the convolution kernel with the largest weight, ; denotes the first-order difference of the weights in the corresponding convolution kernel. Step 4: After the runoff of the above sub-basins and the confluence process of the river or river section are sequentially calculated in the forward direction to obtain the simulated flow of the hydrological observation station, the gradient is calculated by back propagation according to the loss function Loss defined by formula (14) or (16), and the neural network weight W is updated with a specific step size η. Finally, the parameter update is realized in an "end-to-end" manner under the constraint of physical information, and the specific formula is as follows:

[0167]

[0168] (18)

[0169] (19)

[0170] wherein η is the number of iterations of model training; and θ is a physical parameter.

[0171] ​Step 5, pre-training and fine-tuning two-stage strategy. In actual flood forecasting, many basins lack continuous hourly precipitation and flow observation data, and usually only daily data are available, and only hourly scale records during specific flood events. The traditional method usually first calibrates the model parameters on the daily scale to obtain the initial state variables required for predicting flood events, and then converts the scale of some runoff parameters through expert experience, and adjusts the specific parameters using hourly scale flood event data. However, this method is strongly dependent on expert experience. The distributed hybrid hydrological model proposed in the present invention adopts a pre-training and fine-tuning two-stage strategy, making full use of daily data and hourly scale flood event data. In the pre-training stage, the daily outflow of the sub-basin is calculated using daily hydro-meteorological data and static attribute data of the sub-basin, and the daily outflow is converted to hourly outflow assuming that the daily flow is constant, and then the hourly outflow of the basin is obtained through river (reach) confluence routing. Finally, the average value of the 24-hour prediction value is taken as the daily flow prediction result, and the pre-training model is obtained. In the fine-tuning stage, based on the pre-training model, some model parameters in the sub-basin runoff calculation model are scaled to adapt to the hourly scale runoff calculation, and the observed flood event data and the initial state variables derived from the pre-training model are used to further calibrate the model.

[0172] Example verification.

[0173] 1. Invention area and data.

[0174] This embodiment takes the Xiu River Basin in Hunan Province as an example, and the specific reference is Figure 2 , the proposed distributed hybrid hydrological model is verified. The Xiu River Basin is a typical humid basin with an area of about 10,305 square kilometers, belonging to subtropical monsoon climate, with an average annual temperature of 18.0°C and an average annual precipitation of 1,561.2 mm. Influenced by atmospheric circulation, precipitation shows significant seasonal and spatial variability. There are multiple hydrological observation stations in the basin, including Ganxi Station (GX) in the lower reaches, Longjiashan Station (LJS) in the middle reaches, Wulipai Station (WLP), Kengkou Station (KK) and Anren Station (AR) in the upper reaches. In addition, there are 35 automatic telemetering precipitation stations and 1 evaporation station in the basin. The hydrological data comes from the Hydrological Yearbook and the Hunan Provincial Hydrological Bureau, covering daily precipitation, evaporation and flow records from January 1, 1980 to December 31, 1985 and from January 1, 2000 to December 31, 2006, as well as hourly flood data during the flood season.

[0175] 2. Data preprocessing.

[0176] Based on the 90-meter resolution DEM data, the river network of the Lishui watershed was extracted and divided into 81 discrete sub-basins. In the MK confluence method, the river network was further discretized into 487 segments, each with a length of about 2 kilometers. The Kriging interpolation method was used to interpolate the precipitation data from 35 weather stations to obtain the average precipitation of each sub-basin. Due to the lack of evaporation stations, the initial evapotranspiration input for all sub-basins was set to a uniform value. However, in the distributed hybrid hydrological model of this example, a specific evapotranspiration coefficient suitable for the properties of each sub-basin was determined through a neural network, thereby implicitly capturing the spatial heterogeneity of evapotranspiration. In addition, the model requires the input of static properties of sub-basins and river segments. 14 areal average properties related to terrain, soil, and vegetation were calculated for the 81 sub-basins, and 5 static properties of the river channel (such as channel length, slope, etc.) were derived from the DEM to make up for the lack of measured river data.

[0177] 3. Model construction and comparison.

[0178] In this example, the runoff part of the distributed hybrid hydrological model uses the Xin'anjiang model structure, and focuses on evaluating the effectiveness of three river (segment) confluence methods (MK, DW, and CNN) on daily flow and flood event simulation. Two validation strategies are used: time division strategy and space-time division strategy to evaluate the performance of the model at measured stations and unmeasured stations. Regarding the definition of "unmeasured stations", it is necessary to clarify its actual meaning. In theory, the model should be evaluated at all unmeasured river segments, but since there are an infinite number of unmeasured river segments in the actual watershed, in this invention, "unmeasured stations" refer to the case where the historical data of some of the four internal hydrological observation stations (LJS, WLP, KK, and AR) in the hypothetical watershed are unknown, and only the test period data is used for model driving and validation.

[0179] In the time division strategy, five scenarios are designed, using the data of 1 to 5 hydrological observation stations to train the model, and evaluating its effectiveness at each hydrological observation station. In the space-time division strategy, four scenarios are designed, each using a different number of measured stations to train the model, and evaluating the performance of the model at unmeasured stations. In the space-time division strategy, the GX station located at the outlet of the watershed is always designated as a measured station with historical data to meet the actual situation.

[0180] In addition, a lumped Xin'anjiang model (L-XAJ) is constructed as a comparison, which is implemented in the PyTorch framework and the parameters are updated through backpropagation. The upstream watershed of each station is considered as an independent watershed, and the average precipitation and evapotranspiration are calculated. The L-XAJ model is calibrated for each of the five watersheds and its performance is tested.

[0181] All models were calibrated using data from January 1, 1980 to December 31, 1985 and tested using data from January 1, 2000 to December 31, 2006 with a 1-year warm-up period. To account for modeling uncertainty, each model was trained 10 times using 10 different random seeds while keeping other hyperparameters unchanged. In the present invention, the distributed hybrid hydrological model is named as “DHFM-Routing-X”, where “Routing” represents the MK, DW and CNN river (reach) routing method, and “X” represents the number of gauging stations used for training (1, 2, 3, 4 and 5).

[0182] 4. Results analysis.

[0183] Figure 3 and Figure 4 show the NSE index results of several models in the time division strategy and the space-time division strategy, respectively, Figure 5 show the actual simulation curves, Figure 5 where Obs represents the observed value, Obs.(Day) represents the daily flow simulation, Obs.(Flood) represents the flood simulation, Q represents the flow, Day represents the day, and Hour represents the hour. The results show that the distributed hybrid hydrological model proposed in the present invention can better simulate the daily flow and flood events, and is suitable for daily flow and flood simulation in ungauged basins. Figure 6 show the calculation efficiency of the model, where the DW and CNN river (reach) routing method significantly improves the calculation efficiency and realizes the calculation in seconds, further verifying the high efficiency and practicability of the model in practical application.

[0184] Embodiment 2

[0185] Based on the same inventive concept as in Embodiment 1, the present invention provides a flood forecasting device coupled with physical mechanisms and deep learning, comprising:

[0186] a data acquisition module configured to acquire DEM data, hydro-meteorological data and basin static attribute data of a target basin;

[0187] a sub-basin division module configured to divide the target basin into a plurality of sub-basins based on the DEM data, extract a river network, and calculate static attribute data of the river network, wherein the river network comprises a plurality of river channels;

[0188] a forecasting module configured to input the hydro-meteorological data of each sub-basin, the basin static attribute data and the static attribute data of the river network into a pre-trained distributed hybrid hydrological model to obtain simulated flow of a hydrological observation station and complete flood forecasting.

[0189] The remaining parts are the same as in Embodiment 1.

[0190] Embodiment 3

[0191] The flood forecasting system coupling physical mechanism and deep learning provided in the embodiments of the present application comprises a storage medium and a processor.

[0192] The storage medium is configured to store instructions.

[0193] The processor is configured to operate according to the instructions to perform the method according to any one of the embodiments 1.

[0194] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0195] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0196] These computer program instructions can also be stored in a computer-readable memory capable of causing the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product comprising instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0197] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1steps of the functions specified in the one or more blocks.

[0198] The embodiments of the present application are described above with reference to the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, not restrictive, and a person of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims.

[0199] The basic principles and main features of the present application and the advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited to the above-described embodiments, and the above-described embodiments and descriptions in the specification are only illustrative of the principles of the present application, and various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A flood forecasting method coupling physical mechanism and deep learning, characterized in that, The method comprises the following steps: acquiring DEM data, hydro-meteorological data and static attribute data of a target basin; dividing the target basin into a plurality of sub-basins based on the DEM data, extracting a river network, and calculating static attribute data of the river network, wherein the river network comprises a plurality of river channels; inputting the hydro-meteorological data of each sub-basin, the static attribute data of the basin, and the static attribute data of the river network into a pre-trained distributed hybrid hydrological model to obtain simulated flow of a hydrological observation station, and completing flood forecasting; the distributed hybrid hydrological model couples physical mechanisms and deep learning; the distributed hybrid hydrological model comprises a sub-basin runoff calculation model and a confluence calculation model arranged in sequence; the mathematical expression of the sub-basin runoff calculation model is: the mathematical expression of the confluence calculation model is: In the formula, Q represents the outflow of sub-basin i at time t; HM represents the lumped hydrological model; Q represents the hydro-meteorological data of sub-basin i at time t; θ basin,i A represents the parameters of the lumped hydrological model of sub-basin i; A basin,i F represents the static attribute data of sub-basin i; FNN1 represents a feedforward neural network for learning the mapping relationship between the static attribute data of sub-basin and the parameters of the lumped hydrological model; W FNN1 W represents the parameters of FNN1; Q represents the flow of sub-basin i at time t after confluence through a river channel or a river reach; RM represents a river confluence model or a river reach confluence model; θ channel A represents the parameters of the river confluence model or the river reach confluence model; A channel F represents the static attribute data of the river channel or the river reach; FNN2 represents a feedforward neural network for learning the mapping relationship between the static attribute data of the river channel or the river reach and the parameters of the river confluence model or the river reach confluence model; W FNN2 W represents the parameters of FNN2; Q represents the simulated flow of the hydrological observation station at time t; Let the river channel from sub-basin i to the downstream hydrological observation station be river channel i. When the Muskingum routing method is used, each river channel i needs to be divided into J equal-length river sections, and the river section index j satisfies j ∈ [1, J]. The outflow of the sub-basin is sequentially routed from upstream to downstream. For the upstream river section j = 1, the inflow is the outflow of the corresponding sub-basin i The outflow of the river section j = 1 is calculated according to the following formula: wherein, and Qijtand Qijt-1denote the outflow of reach j in river i at time t and t-1, respectively, and Qitand Qit-1denote the outflow of sub-basin i at time t and t-1, respectively, Qj-1,t-1denotes the lateral inflow of reach j in river i at time t-1; for downstream reaches j, j = 2, 3,... J-1, whose inflow is equal to the outflow of the upstream adjacent reach j-1, the outflow of reach j is calculated according to the following equation: wherein and Qj-1,tand Qj-1,t-1respectively represent the outflow of reach j-1 in river i at times t and t-1. when the river section j = J, the flow of the sub-basin i at the hydrological observation station at time t is calculated according to the following formula In the formula, Qi(t) is the flow of sub-basin i at time t; and Qi(t-1) is the flow of sub-basin i at time t-1. parameters C1, C2, C3 and C4 are calculated according to the following formulas respectively: where K i,j represents the storage time constant of river reach j in river channel i, X i,j represents the weight factor of river reach j in river channel i, and Δt represents the calculation time step; if river reach j is located within a certain sub-basin k, where k≠i, k∈[1, n], n is the total number of sub-basins, then the runoff of sub-basin k will be distributed to river reach j as lateral inflow, represents the lateral inflow of river reach j in river channel i at time t, and its specific calculation formula is as follows: wherein, represents the proportional coefficient of river section j in sub-basin k, represents the outflow of sub-basin k at time t, represents the average catchment area controlled by river section j in sub-basin k, S k represents the area of sub-basin k; The parameter K i,j and X i,j Adaptively learned by the feedforward neural network FNN2, the value of which is determined by the static attribute data of the river reach j, and the specific calculation formula is as follows: wherein represents static attribute data for reach j in river course i.

2. The method of claim 1, wherein: when the diffusion wave confluence method is used, the movement of flood in the river channel is described by the convection diffusion equation, and combined with the initial condition and the boundary condition, the calculation formula of the flow of the outflow of the sub-basin i at the hydrological observation station at time t is obtained, which is specifically: where, Qi(t) is the flow at the hydrological observation station after the outflow of sub-basin i converges through the river reach at time t, basin,i (t-τ) represents the inflow of river i at time t-τ, which is also the outflow of sub-basin i; h channel,i (x,t) represents the impulse response function of river i at time t, x is the length of river i; t max represents the maximum time step considering historical inflow, which is a hyperparameter; Parameter C i and D i Adaptively learned by the feedforward neural network FNN2, whose value is determined by the static attribute data of the river channel i, and the specific calculation formula is as follows: In the formula, represents the static attribute data of the river channel i.

3. The method of claim 1 or 2, wherein: the loss function Loss of the confluence calculation model is: where s is the number of hydrological observation station, N is the total number of hydrological observation stations, NSE represents the Nash-Sutcliffe coefficient, t represents the time, T represents the length of time series, represents the observed flow of the sub-basin i, represents the simulated flow of the sub-basin i, represents the average value of the observed flow of the sub-basin i. 4.The flood forecasting method of coupling physical mechanism and deep learning according to claim 1, wherein: When the convolutional neural network confluence method is adopted, each river has a unique static attribute, which corresponds to a different convolution kernel, and the flow of sub-basin i at time t after confluence in the hydrological observation station The calculation formula is: In the formula, represents the weight of the convolution kernel corresponding to the river channel i, m represents the position of each weight, m∈[1, l], and l represents the total number of weights, represents the static attribute data of the river channel i, Q basin,i (t-m) represents the outflow of the sub-basin i at t-m.

5. The flood forecasting method coupling physical mechanism and deep learning according to claim 4, characterized in that: the loss function Loss of the confluence calculation model is: In the formula, λ1 and λ2 respectively represent the coefficients of each component, and the sum of the two is 1; and respectively represent the weights of positions m and m+1 in the convolution kernel corresponding to river channel i, n represents the total number of sub-basins; ReLu is an activation function; p represents the index of the convolution kernel with the largest weight, p∈[1,l]; represents the first-order difference of the weight in the convolution kernel corresponding to river channel i; s is the number of the hydrological observation station, N is the total number of the hydrological observation stations, and NSE represents the Nash efficiency coefficient.

6. The method of claim 1, wherein: The training process of the distributed hybrid hydrological model comprises: in the pre-training stage, the distributed hybrid hydrological model is pre-trained by using daily scale hydro-meteorological data, static attribute data of the basin and static attribute data of the river network to obtain a pre-training model; in the fine-tuning stage, the pre-training model is trained by using hourly scale hydro-meteorological data, static attribute data of the basin and static attribute data of the river network to obtain a final distributed hybrid hydrological model. 7.A flood forecasting device coupling physical mechanism and deep learning, characterized in that, The method comprises the following steps: a data acquisition module is configured to acquire DEM data, hydro-meteorological data and static attribute data of a target basin; a sub-basin division module is configured to divide the target basin into a plurality of sub-basins based on the DEM data, extract a river network, and calculate static attribute data of the river network, wherein the river network comprises a plurality of river channels; a prediction module is configured to input the hydro-meteorological data of each sub-basin, the static attribute data of the basin, and the static attribute data of the river network into a pre-trained distributed hybrid hydrological model to obtain simulated flow of a hydrological observation station, and complete flood forecasting; the distributed hybrid hydrological model couples physical mechanisms and deep learning; the distributed hybrid hydrological model comprises a sub-basin runoff calculation model and a confluence calculation model arranged in sequence; the mathematical expression of the sub-basin runoff calculation model is: the mathematical expression of the confluence calculation model is: In the formula, Q represents the outflow of sub-basin i at time t; HM represents the lumped hydrological model; Q represents the hydro-meteorological data of sub-basin i at time t; θ basin,i A represents the parameters of the lumped hydrological model of sub-basin i; A basin,i F represents the static attribute data of sub-basin i; FNN1 represents a feedforward neural network for learning the mapping relationship between the static attribute data of sub-basin and the parameters of the lumped hydrological model; W FNN1 W represents the parameters of FNN1; Q represents the flow of sub-basin i at time t after confluence through a river channel or a river reach; RM is a river confluence model or a river reach confluence model; θ channel A represents the parameters of the river confluence model or the river reach confluence model; A channel F represents the static attribute data of the river channel or the river reach; FNN2 represents a feedforward neural network for learning the mapping relationship between the static attribute data of the river channel or the river reach and the parameters of the river confluence model or the river reach confluence model; W FNN2 W represents the parameters of FNN2; Q represents the simulated flow of the hydrological observation station at time t; Let the river channel from sub-basin i to the downstream hydrological observation station be river channel i. When the Muskingum routing method is used, each river channel i needs to be divided into J equal-length river sections, and the river section index j satisfies j ∈ [1, J]. The outflow of the sub-basin is sequentially routed from upstream to downstream. For the upstream river section j = 1, the inflow is the outflow of the corresponding sub-basin i The outflow of the river section j = 1 is calculated according to the following formula: wherein, and Qijtand Qijt-1denote the outflow from reach j in river i at time t and t-1, respectively, and Qitand Qit-1denote the outflow from sub-basin i at time t and t-1, respectively, Qj-1t-1denotes the lateral inflow to reach j in river i at time t-1; for downstream reaches j, j = 2, 3,... J-1, whose inflow is equal to the outflow from the upstream adjacent reach j-1, the outflow from reach j is calculated according to the following equation: wherein and Qj-1,tand Qj-1,t-1respectively represent the outflow of reach j-1 in river i at times t and t-1. When the river section j = J, the outflow of the sub-basin i at the hydrological observation station after the confluence of the river section is calculated according to the following formula In the formula, Qi(t) is the flow of sub-basin i at time t; and Qi(t-1) is the flow of sub-basin i at time t-1. parameters C1, C2, C3 and C4 are calculated according to the following formulas respectively: where K i,j represents the storage time constant of river reach j in river channel i, X i,j represents the weight factor of river reach j in river channel i, and Δt represents the calculation time step; if river reach j is located within a certain sub-basin k, where k≠i, k∈[1, n], n is the total number of sub-basins, then the runoff of sub-basin k will be distributed to river reach j as lateral inflow, represents the lateral inflow of river reach j in river channel i at time t, and its specific calculation formula is as follows: wherein represents the proportional coefficient of river section j in sub-basin k, represents the outflow of sub-basin k at time t, represents the average catchment area controlled by river section j in sub-basin k, S k represents the area of sub-basin k; The parameter K i,j and X i,j Adaptively learned by the feedforward neural network FNN2, the value of which is determined by the static attribute data of the river reach j, and the specific calculation formula is as follows: wherein represents the static attribute data of the reach j in the river course i. 8.A flood forecasting system coupling physical mechanism and deep learning, characterized in that, The storage medium and the processor are included. The storage medium is configured to store instructions; the processor is configured to operate according to the instructions to execute the method according to any one of claims 1-6.

Images