A flood forecasting model and method based on coupling mechanism model and deep learning

Abstract

The present invention relates to the technical field of flood forecasting, and discloses a flood forecasting model and method of a coupling mechanism model and deep learning, wherein the flood forecasting model comprises: a common neural network layer takes basin characteristic data as input data and outputs first data; a P-RNN layer takes meteorological driving factor time series data as input data, collects the first data, and outputs current runoff characteristic data, and the P-RNN layer comprises preset recurrent neurons, which are obtained by encoding based on a hydrological mechanism model; a first recurrent neural network layer takes previous flood flow data as input data and outputs previous flood characteristic data; a second recurrent neural network layer takes previous flood characteristic data, current runoff characteristic data, and meteorological driving factor data as input data and outputs flood characteristic data; an output layer takes flood characteristic data as input data and outputs flood flow data. The present invention realizes the consistency between the hydrological model dominated by deep learning and physics, and improves the accuracy of flood forecasting.

Description

Technical Field

[0001] The present invention relates to the technical field of flood forecasting, and in particular to a flood forecasting model and method based on a coupling mechanism model and deep learning. Background Art

[0002] Flood disasters are one of the most frequent and destructive natural disasters in the world today. Accurate and timely flood forecasting can greatly help humans reduce the damage caused by such events and reduce losses.

[0003] Among them, hydrological physics models can provide in-depth understanding of physical processes, but are subject to many simplified assumptions and it is difficult to discover usable information from large amounts of data. Data-driven models can well simulate highly nonlinear and complex hydrological systems from data.

[0004] Since most deep learning methods do not understand causal relationships and can only capture the correlation between input and output, the lack of physical mechanisms has become the main limitation of hydrological prediction models based on deep learning. Therefore, how to quantitatively understand the physical process and achieve physical consistency between deep learning-led hydrological models has become a crucial issue. Summary of the invention

[0005] In view of this, the present invention provides a flood forecasting model and method of a coupling mechanism model and deep learning, which achieves the consistency between the hydrological model dominated by deep learning and physics, reduces the uncertainty of the data-driven model, and improves the accuracy of flood forecasting.

[0006] In the first aspect, the present invention provides a flood forecasting model of a coupling mechanism model and deep learning, characterized in that the flood forecasting model includes: an ordinary neural network layer, a P-RNN layer, a first recurrent neural network layer, a second recurrent neural network layer and an output layer, the P-RNN layer includes preset recurrent neurons, and the preset recurrent neurons are encoded based on the hydrological mechanism model; the ordinary neural network layer takes the basin characteristic data as input data and outputs the first data; the P-RNN layer takes the meteorological driving factor time series data as input data, collects the first data, and outputs the current runoff characteristic data; the first recurrent neural network layer takes the previous flood flow data as input data and outputs the previous flood characteristic data; the second recurrent neural network layer takes the previous flood characteristic data, the current runoff characteristic data and the meteorological driving factor data as input data, and outputs the flood characteristic data; the output layer takes the flood characteristic data as input data and outputs the flood flow data.

[0007] In this embodiment, the flood forecasting model proposed in this application includes a P-RNN layer, which has neurons encoded based on the hydrological mechanism model, so that the mechanism model simulation method is used as the computing kernel of the neural network, realizing the compact coupling of the two paradigms of mechanism model and deep learning model, and realizing the consistency between the deep learning-dominated hydrological model and physics, thereby reducing the uncertainty of the flood forecasting model based on the data-driven model and improving the accuracy of flood forecasting.

[0008] In an optional embodiment, the first data includes the average tension water storage capacity, the basin tension water storage capacity curve distribution index, the average free water capacity and the basin free water capacity distribution curve index, the P-RNN layer uses the meteorological driving factor time series data as input parameters, collects the first data, and outputs the current runoff characteristic data, including: the P-RNN layer uses the meteorological driving factor time series data as input parameters, predicts the total runoff currently generated in the basin, and outputs the current runoff characteristic data, and the model parameters of the P-RNN layer include the first data.

[0009] In an optional embodiment, the state variables of the preset recurrent neurons of the P-RNN layer include upper layer tension water storage, lower layer tension water storage, deep layer tension water storage, free water storage and runoff area.

[0010] In an optional implementation, the core calculation equation of the P-RNN layer is:

[0011] E t =f(WU t-1 , W.L. t-1 , P t , E.M. t |W kc , W c , W lm );

[0012] R t =g 1 (WU t , W.L. t , W.D. t , P t , E t |im,wm,b);

[0013]

[0014] RS t =ψ 0 (R t , P t , E t , S t , F R t |sm,ex);

[0015] RI t =ψ 1 (S t ,FR t |W ki );

[0016] RG t =ψ 2 (S t ,FR t |W kg );

[0017] WU t =ω 0 (WU t-1 ,WL t-1 ,WD t-1 ,R t ,P t ,E t |W um ,W lm ,W dm );

[0018] WL t =ω 1 (WU t-1 ,WL t-1 ,WD t-1 ,R t ,P t ,E t |W um ,W lm ,W dm );

[0019] WD t =ω 2 (WU t-1 ,WL t-1 ,WD t-1 ,R t ,P t ,E t |W um ,W lm ,W dm );

[0020] S t =n(S t-1 ,FR t-1 ,R t |W ki ,W kg );

[0021] FR t =φ(P t ,E t ,FR t-1 );

[0022] Among them, WU t , W.L. t , W.D. t , S t , F R t WU is the upper tension water storage, lower tension water storage, deep tension water storage, free water storage and runoff area at time t. t-1 , W.L. t-1 , W.D. t-1 , S t-1 , F R t-1 is the upper layer tension water storage, lower layer tension water storage, deep layer tension water storage, free water storage and runoff area at time t-1, W kc and W c are the evapotranspiration conversion coefficient and the deep evapotranspiration diffusion coefficient respectively; R t and are the flow rate generated by the permeable area and the flow rate generated by the impervious area of ​​the basin respectively; im is the proportion of impervious area; RS t , R.I. t , R.G. t are the surface runoff, subsoil flow and groundwater flow generated at time t; W ki and W kd are the outflow coefficients of subsoil flow and groundwater respectively; W um , W lm , W dm are the upper layer tension water capacity, the lower soil layer tension water capacity and the deep layer tension water capacity respectively; f, g 1 , g 2 , ψ 0 , ψ 1 , ψ 2 ,ω 0 ,ω 1 ,ω 2 , η is the calculation function.

[0023] In an optional embodiment, the flood forecasting model is built based on the Tensorflow framework; wherein the flood forecasting model includes multiple ordinary neural network layers, a P-RNN layer, multiple first recurrent neural networks, multiple second neural network layers and a recurrent neural network output layer.

[0024] In the second aspect, the present invention provides a training method for a coupled mechanism model and a deep learning flood forecasting model, and the training method for the coupled mechanism model and the deep learning flood forecasting model includes: constructing a training sample set, the training sample set includes pre-trained meteorological and hydrological data; based on a preset ratio, the training sample set is divided into a calibration sample set and a verification sample set; based on a preset loss function, the calibration sample set and the verification sample set are used to train the above-mentioned flood forecasting model, and the trained flood forecasting model is obtained by adjusting the epoch, batch-size parameters and the number of neurons of the flood forecasting model.

[0025] In a third aspect, the present invention provides a flood forecasting method, which includes: acquiring current hydrological data; inputting the current hydrological data into the above-mentioned flood forecasting model, predicting the flood characteristics of the current hydrological data, and obtaining current flood flow data.

[0026] In a fourth aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the coupling mechanism model of the second aspect and the training method of the deep learning flood forecasting model or the flood forecasting method of the third aspect by executing the computer instructions.

[0027] In a fifth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the coupling mechanism model of the second aspect and the training method of the deep learning flood forecasting model or the flood forecasting method of the third aspect.

[0028] In a sixth aspect, the present invention provides a computer program product comprising computer instructions for causing a computer to execute the training method of the coupling mechanism model and deep learning flood forecasting model of the second aspect or the flood forecasting method of the third aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 It is a structural diagram of the coupling mechanism model and deep learning flood forecasting model of this application;

[0031] Figure 2is a flow chart of a method for constructing a flood forecasting model based on a coupling mechanism model and deep learning according to an embodiment of the present invention;

[0032] Figure 3 is a flow chart of a method for training a coupled mechanism model and a deep learning flood forecasting model according to an embodiment of the present invention;

[0033] Figure 4 is a flow chart of a flood forecasting method according to an embodiment of the present invention;

[0034] Figure 5 is a schematic diagram of the principle of a flood forecasting model according to an embodiment of the present invention;

[0035] Figure 6 is a structural block diagram of a training device for a flood forecasting model according to an embodiment of the present invention;

[0036] Figure 7 is a structural block diagram of a flood forecasting device according to an embodiment of the present invention;

[0037] Figure 8 It is a structural schematic diagram of a computer device provided by an optional embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0039] At present, researchers have developed a large number of flood forecasting models. Generally speaking, flood forecasting models are mainly divided into two categories, namely traditional hydrological physics models and data-driven deep learning models. Among them, traditional hydrological physics models can provide in-depth understanding of physical processes, but are subject to many simplified assumptions and it is difficult to find usable information from a large amount of data; while data-driven deep learning models can well simulate highly nonlinear and complex hydrological systems from data. Since the 1970s, data-driven models have been used for hydrological simulation and prediction, and have even been placed on an equal footing with hydrological physics models. In addition, due to the increasing computer performance and the emergence of available big data-driven deep learning models, the problems that were once difficult to solve in hydrological forecasting have been successfully overcome.

[0040] However, since most deep learning methods do not understand causal relationships and can only capture the correlation between input and output, the lack of physical mechanisms has become the main limitation of hydrological prediction models based on deep learning. Therefore, how to quantitatively understand the physical process and achieve consistency between the deep learning-led hydrological model and physics is a problem that needs to be solved at present.

[0041] The current research methods are divided into four categories. The first research method is to use empirical knowledge to divide the runoff process into multiple sub-processes and build a neural network model for each sub-process; the second research method is to use the output of the traditional hydrological physics model together with other related factors as the input of the deep learning model to predict runoff through loose coupling; the third research method is to use the physical constraint rules as the penalty term of the loss function in the artificial neural network; the fourth research method is to consider the deep learning runoff prediction model of the hydrological physics process through the attention mechanism inside the deep learning network.

[0042] However, existing coupling methods are mostly based on standard neural network models, and the integration of physical mechanisms is still in the simple combination stage, which makes it impossible to achieve consistency between deep learning-dominated hydrological models and physics.

[0043] Based on this, this application uses the hydrological mechanism model as the computational kernel of the neural network neurons, and compactly couples the two paradigms of the mechanism model and the deep learning model. This application proposes a P-RNN layer with a physical mechanism, and constructs the recurrent neurons of the P-RNN layer encoded by the principle method of the hydrological mechanism model. The hydrological model dominated by deep learning is consistent with physics, thereby reducing the uncertainty of the flood forecasting model based on the data-driven model and improving the accuracy of flood forecasting.

[0044] Specifically, the present application provides a flood forecasting model of a coupling mechanism model and deep learning. Among them, the flood forecasting model includes: a normal neural network layer, a P-RNN layer, a first recurrent neural network layer, a second recurrent neural network layer and an output layer. The P-RNN layer includes preset recurrent neurons, which are encoded based on the hydrological mechanism model; the normal neural network layer uses the basin characteristic data as input data and outputs the first data; the P-RNN layer uses the meteorological driving factor time series data as input data, collects the first data, and outputs the current runoff characteristic data; the first recurrent neural network layer uses the previous flood flow data as input data and outputs the previous flood characteristic data; the second recurrent neural network layer uses the previous flood characteristic data, the current runoff characteristic data and the meteorological driving factor data as input data, and outputs the flood characteristic data; the output layer uses the flood characteristic data as input data and outputs the flood flow data.

[0045] Specifically, see Figure 1 , Figure 1It is a structural diagram of the coupling mechanism model and deep learning flood forecasting model of this application.

[0046] (1) The general neural network layer includes NN neural units. The general neural network layer uses the watershed characteristic data as input data, processes the watershed characteristic data using the NN neural units, and calculates and outputs the first data. The first data is used as a parameter of the hydrological mechanism model in the P-RNN layer.

[0047] Among them, the basin characteristic data include soil type data and land use coverage data. The hydrological mechanism model is the Xin'anjiang model, and the first data are the four model parameters in the Xin'anjiang model. Specifically, they are the average tension water storage capacity (WM), the basin tension water storage capacity curve distribution index (B), the average free water capacity (SM) and the basin free water capacity distribution curve index (EX).

[0048] Furthermore, the core calculation equation of the NN neural unit in the common neural network layer is:

[0049]

[0050] Among them, y j is the first data of the jth output, w ij is the weight parameter, x i is the i-th input data of the ordinary neural network layer, v j is the jth bias term, and σ is the activation function.

[0051] (2) The P-RNN layer has a physical mechanism. The P-RNN layer takes the time series data of meteorological driving factors as input data, uses the preset recurrent neurons of the P-RNN layer to predict the total runoff currently generated in the basin, and calculates and outputs the current runoff characteristic data.

[0052] The P-RNN layer includes a custom preset recurrent neuron, wherein the preset recurrent neuron is constructed and encoded based on the principle method of the hydrological mechanism model. The first data output by the common neural network layer is used as part of the parameters of the hydrological mechanism model.

[0053] The time series data of meteorological cycle driving factors include rainfall time series data and pan evaporation time series data. Specifically, the rainfall time series data is (P 0 , P 1 ,…,P t-1 , P t ), the evaporation time series data of the evaporation dish is (EM 0 , E.M. 1 , …, EM t-1 , E.M. t ).

[0054] In one implementation, the hydrological mechanism model is the Xin'anjiang model.

[0055] Furthermore, the preset recurrent neuron has five state variables, which are upper tension water storage (WU), lower tension water storage (WL), deep tension water storage (WD), free water storage (S) and flow-generating area (FR).

[0056] Furthermore, the core calculation equation of the preset recurrent neuron is:

[0057] R t =g 1 (WU t , W.L. t , W.D. t , P t , E t |im,wm,b);

[0058]

[0059] RS t =ψ 0 (R t , P t , E t , S t , F R t |sm,ex);

[0060] RI t =ψ 1 (S t , F R t |W ki );

[0061] RG t =ψ 2 (S t , F R t |W kg );

[0062] WU t =ω 0 (WU t-1 , W.L. t-1 , W.D. t-1 , R t , P t , E t |W um , W lm , W dm );

[0063] WL t =ω 1 (WU t-1 , W.L. t-1 , W.D.t-1 , R t , P t , E t |W um , W lm , W dm );

[0064] WD t =ω 2 (WU t-1 , W.L. t-1 , W.D. t-1 , R t , P t , E t |W um , W lm , W dm );

[0065] S t =η(S t-1 , F R t-1 , R t |W ki , W kg );

[0066] FR t =φ(P t , E t , F R t-1 ).

[0067] Among them, WU t , W.L. t , W.D. t , S t , F R t WU is the upper tension water storage, lower tension water storage, deep tension water storage, free water storage and runoff area at time t. t-1 , W.L. t-1 , W.D. t-1 , S t-1 , F R t-1 is the upper layer tension water storage, lower layer tension water storage, deep layer tension water storage, free water storage and runoff area at time t-1, W kc and W c are the evapotranspiration conversion coefficient and the deep evapotranspiration diffusion coefficient respectively; R t and are the flow rate generated by the permeable area and the flow rate generated by the impervious area of ​​the basin respectively; im is the proportion of impervious area; RS t , R.I. t , R.G. t are the surface runoff, subsoil flow and groundwater flow generated at time t; W ki and W kdare the outflow coefficients of subsoil flow and groundwater respectively; W um , W lm , W dm are the upper layer tension water capacity, the lower soil layer tension water capacity and the deep layer tension water capacity respectively; f, g 1 , g 2 , ψ 0 , ψ 1 , ψ 2 ,ω 0 ,ω 1 ,ω 2 , η is the calculation function.

[0068] (3) The first recurrent neural network layer includes a first recurrent neural unit. The first recurrent neural network layer uses the previous flood flow data as input data, processes the previous flood flow data using the first recurrent neural unit, and calculates and outputs the previous flood characteristic data.

[0069] In one implementation, the first recurrent neural network layer uses a gated recurrent neural network Gated Recurrent Unit (GRU). The core calculation equation of the first recurrent neural network layer is:

[0070] R t =σ(X t W xr +H t-1 W hr );

[0071] Z t =σ(X t W xz +H t-1 W hz );

[0072]

[0073]

[0074] O t =H t W hq +b q .

[0075] Among them, X t is the flood flow data at time t, R t To reset the gate, Z t To update the gate, is the hidden state variable of the neuron in the gated recurrent neural network, H t , H t-1 are the neuron state variables of the gated recurrent neural network at time t and time t-1, O tis the neuron output variable, W xr , W hr , W xz , W hz , W xh , W hh , W hq and b q are the hyperparameters of the neurons in the gated recurrent neural network, and σ is the activation function.

[0076] (4) The second recurrent neural network layer includes a second recurrent neural unit. The second recurrent neural network layer uses the current runoff characteristic data output by the P-RNN layer, the previous flood characteristic data output by the first recurrent neural network layer, and the time series data of meteorological driving factors as input data, uses the second recurrent neural unit to predict the flood, and calculates and outputs the flood characteristic data.

[0077] In one implementation, the second recurrent neural network layer uses a gated recurrent neural network (GRU).

[0078] Among them, the core calculation equation of the second recurrent neural network layer is:

[0079]

[0080] Among them, Y t is the flood characteristic data output by the second recurrent neural network layer, is the current runoff feature data output by the P-RNN layer, is the previous flood characteristic data output by the first recurrent neural network layer, is the time series data of meteorological driving factors, and G is the core calculation equation of heating of gated recurrent neural network.

[0081] (5) The output layer includes fully connected neurons. The output layer uses the flood characteristic data output by the second recurrent neural network layer as input data, processes the flood characteristic data using fully connected neurons, and calculates and outputs the flood flow data of the current river section.

[0082] In this embodiment, the flood forecasting model proposed in the present application includes a P-RNN layer, which has neurons encoded based on the hydrological mechanism model, so that the mechanism model simulation method is used as the computing kernel of the neural network, realizing the compact coupling of the two paradigms of mechanism model and deep learning model, achieving the consistency between the deep learning-dominated hydrological model and physics, making up for the lack of physical mechanism in the data-driven hydrological model, thereby reducing the uncertainty of the flood forecasting model based on the data-driven model and improving the accuracy of flood forecasting.

[0083] In this embodiment, a method for constructing a flood forecasting model based on a coupling mechanism model and deep learning is also provided. Figure 2 is a flow chart of a method for constructing a coupling mechanism model and a deep learning flood forecasting model according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps:

[0084] Step S201: Data collection and preprocessing.

[0085] Collect meteorological and hydrological data. Among them, meteorological and hydrological data include watershed characteristic data, meteorological driving factor time series data and previous flood flow data. Specifically, meteorological and hydrological data include watershed DEM elevation data, soil type raster data, vegetation cover raster data, hourly rainfall station rainfall data, evaporation pan daily evaporation data and hourly previous flood flow data.

[0086] The meteorological and hydrological data are preprocessed. Preprocessing includes data segmentation, resolution adjustment, and missing value supplementation. Specifically, preprocessing includes automatic sub-basin division based on DEM data, resampling soil type raster data and vegetation cover raster data to a uniform resolution, interpolating missing values ​​in hourly rainfall data, and segmenting daily evaporation data from evaporation pans into hourly data.

[0087] In a specific embodiment, the meteorological and hydrological data include STRM elevation data with a resolution of 90m×90m, soil type data from the China Soil Database, China's land use data for 2020 with a resolution of 30 meters, hourly rainfall station rainfall data, evaporation pan daily evaporation data, and hourly previous flood flow data. Preprocessing includes using ARCGIS software and DEM data to automatically divide the study basin into multiple sub-basins, using ARCGIS software to resample soil type raster data and vegetation cover raster data to a resolution of 30 meters, using linear interpolation to fill in the missing values ​​of the original hourly rainfall data, and dividing the evaporation pan daily evaporation data into hourly data by dividing by 24.

[0088] Step S202: constructing a flood forecasting model.

[0089] Build a flood forecasting model based on the Tensorflow framework.

[0090] In one implementation, the flood forecasting model includes multiple common neural network layers, a P-RNN layer, multiple first recurrent neural networks, multiple second neural network layers and a recurrent neural network output layer.

[0091] Furthermore, after the flood forecasting model is constructed, the hyperparameters of the flood forecasting model are calibrated.

[0092] In one implementation, the present application performs hyperparameter calibration on the flood forecasting model through an error back propagation algorithm. Specifically, when updating the GRU neural network weight parameter (W xr , W hr , W xz , W hz , W xh , W hq , b q ), the embedding layer weight parameters (W kc , W c , W ki , W kd , W um , W lm , W dm ). According to the chain rule, we can get the partial derivatives of each parameter Finally, each parameter in the flood forecasting model is updated using the gradient descent method according to the partial derivative of each parameter.

[0093] Step S203: training a flood forecasting model.

[0094] See also Figure 3 , Figure 3 is a flow chart of a method for training a flood forecasting model based on a coupling mechanism model and deep learning according to an embodiment of the present invention, such as Figure 3 As shown, the process includes the following steps:

[0095] Step S301: construct a training sample set.

[0096] The training sample set includes pre-trained meteorological and hydrological data, which include basin characteristic data, meteorological driving factor time series data and previous flood flow data.

[0097] Specifically, the pre-trained meteorological and hydrological data include pre-processed soil type raster data, vegetation cover raster data, hourly rainfall station rainfall data, evaporation pan daily evaporation data and hourly previous flood flow data. Based on the pre-processed soil type raster data, vegetation cover raster data, hourly rainfall station rainfall data, evaporation pan daily evaporation data and hourly previous flood flow data, three formats of model training sample sets are generated. Among them, the sample sets generated by soil type raster data and vegetation cover raster data are used as input data of the ordinary neural network layer, the sample sets generated by hourly rainfall station rainfall data and evaporation pan daily evaporation data are used as input data of the P-RNN layer with physical mechanism, and the sample sets generated by hourly previous flood flow data are used as input data of the first recurrent neural network layer.

[0098] Step S302: Based on a preset ratio, the training sample set is divided into a calibration sample set and a validation sample set.

[0099] In one implementation, the preset ratio is 8:2 for the calibration sample set and the validation sample set. According to the ratio of 8:2, the training sample set is divided into the calibration sample set and the validation sample set.

[0100] Step S303: Based on a preset loss function, the calibration sample set and the validation sample set are used for training to obtain a trained flood forecasting model.

[0101] Specifically, based on the preset loss function, the calibration sample set and the verification sample set are used to train the flood forecasting model constructed above, and the trained flood forecasting model is obtained by adjusting the epoch, batch-size parameters and the number of neurons of the flood forecasting model.

[0102] Step S204: Flood forecast model accuracy assessment.

[0103] The flood forecast model is comprehensively evaluated using multiple evaluation criteria. If the accuracy of the flood forecast model meets the requirements, the training step of step S203 is stopped; if it does not meet the requirements, the training step of step S203 is repeated.

[0104] In one implementation, the evaluation criteria include four evaluation criteria: NSE, R2, RMSE, and MAE.

[0105] Specifically, the flood forecasting model is comprehensively evaluated using the four evaluation criteria of NSE, R2, RMSE, and MAE. If the accuracy of the flood forecasting model meets the requirements, the training step of step S203 is stopped; if it does not meet the requirements, the training step of step S203 is repeated.

[0106] Among them, NSE, R 2 The calculation equations of , RMSE and MAE are:

[0107]

[0108]

[0109]

[0110]

[0111] in, is the runoff simulation prediction value, y i Runoff simulations are based on real observations.

[0112] In this embodiment, a flood forecasting method using the above-mentioned flood forecasting model is also provided. Figure 4is a flow chart of a flood forecasting method according to an embodiment of the present invention. Figure 4 As shown, the process includes the following steps:

[0113] Step S401: Acquire current hydrological data.

[0114] Step S402: input the current hydrological data into the flood forecasting model, predict the flood characteristics of the current hydrological data, and obtain the current flood flow data.

[0115] In a specific embodiment, the flood forecasting model includes two common neural network layers, one P-RNN layer, two first recurrent neural network layers, two second recurrent neural network layers and an output layer. Figure 5 , Figure 5 4 is a schematic diagram of the principle of a flood forecasting model according to an embodiment of the present invention.

[0116] Among them, the flood forecasting model includes auxiliary data input layer 1: ordinary neural network layer, auxiliary data input layer 2: the first recurrent neural network layer as auxiliary, main data input layer: P-RNN layer and the second recurrent neural network layer.

[0117] Based on Figure 5 The schematic diagram of the flood forecasting model shown is a principle diagram of the flood forecasting model. The flood forecasting model embeds the Xinanjiang model into the neurons of the custom recurrent neural network, predicts the flood characteristics of the current hydrological data, and obtains the current flood flow data.

[0118] This application achieves physical consistency between the deep learning-led hydrological model and the data-driven hydrological model by embedding the Xin'anjiang model into the neurons of a custom recurrent neural network, solves the problem of lack of physical mechanism in the data-driven hydrological model, reduces the uncertainty of the flood forecasting model based on the data-driven model, and improves the accuracy of flood forecasting, which can better support flood forecasting work.

[0119] In this embodiment, a training device for a coupled mechanism model and a deep learning flood forecasting model is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" may be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0120] This embodiment provides a training device for a flood forecasting model of a coupling mechanism model and deep learning. Figure 6 As shown, Figure 6 1 is a structural block diagram of a training device for a flood forecast model according to an embodiment of the present invention, wherein the training device for a flood forecast model comprises:

[0121] The construction module 601 is used to construct a training sample set, which includes pre-trained meteorological and hydrological data.

[0122] The division module 602 is used to divide the training sample set into a calibration sample set and a validation sample set based on a preset ratio.

[0123] The adjustment module 603 is used to train the flood forecasting model based on a preset loss function using a calibration sample set and a verification sample set, and obtain a trained flood forecasting model by adjusting the epoch, batch-size parameters and the number of neurons of the flood forecasting model.

[0124] The training device of the coupling mechanism model and the deep learning flood forecasting model in this embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0125] In this embodiment, a flood forecasting device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made are not repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0126] This embodiment provides a flood forecasting device, such as Figure 7 As shown, Figure 7 is a structural block diagram of a flood forecasting device according to an embodiment of the present invention, the flood forecasting device comprising:

[0127] The acquisition module 701 is used to acquire current hydrological data.

[0128] The prediction module 702 is used to input the current hydrological data into the flood forecasting model, predict the flood characteristics of the current hydrological data, and obtain the current flood flow data.

[0129] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0130] The flood forecasting device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0131] The embodiment of the present invention also provides a computer device having the above Figure 6 The coupled mechanism model and the training device of the deep learning flood forecasting model are shown in Figure 7 The flood forecasting device shown.

[0132] See also Figure 8 , Figure 8 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 8 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.

[0133] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0134] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0135] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0136] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0137] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 8 The example of connecting through bus is taken in the following.

[0138] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0139] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0140] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0141] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A flood forecasting model based on a coupling mechanism model and deep learning, characterized in that: The flood forecasting model includes: a common neural network layer, a P-RNN layer, a first recurrent neural network layer, a second recurrent neural network layer and an output layer, wherein the P-RNN layer includes preset recurrent neurons, which are obtained based on the Xin'anjiang model encoding, and the state variables of the preset recurrent neurons in the P-RNN layer include upper layer tension water storage, lower layer tension water storage, deep layer tension water storage, free water storage and runoff area; The common neural network layer uses the basin characteristic data as input data and outputs first data, wherein the first data includes average tension water storage capacity, basin tension water storage capacity curve distribution index, average free water capacity and basin free water capacity distribution curve index; The P-RNN layer uses the meteorological driving factor time series data as input data, collects the first data, and outputs the current runoff characteristic data; The first recurrent neural network layer takes the previous flood flow data as input data and outputs the previous flood characteristic data; The second recurrent neural network layer uses the previous flood characteristic data, the current runoff characteristic data and the meteorological driving factor data as input data, and outputs flood characteristic data; The output layer takes the flood characteristic data as input data and outputs flood flow data.

2. The coupling mechanism model and deep learning flood forecasting model according to claim 1, characterized in that: The P-RNN layer uses the meteorological driving factor time series data as input data, collects the first data, and outputs the current runoff characteristic data including: The P-RNN layer uses the meteorological driving factor time series data as input data, predicts the total runoff currently generated in the basin, and outputs the current runoff characteristic data. The model parameters of the P-RNN layer include the first data, and the meteorological driving factor time series data include rainfall time series data and evaporation pan evaporation time series data.

3. The coupling mechanism model and deep learning flood forecasting model according to claim 1, characterized in that: The core calculation equation of the P-RNN layer is: ； ； ； ； ； ； ； ； ； ； ； in, are the upper layer tension water storage, the lower layer tension water storage, the deep layer tension water storage, the free water storage and the runoff area at time t, are the upper layer tension water storage, the lower layer tension water storage, the deep layer tension water storage, the free water storage and the runoff area at time t-1, and are the evapotranspiration conversion coefficient and the deep evapotranspiration diffusion coefficient respectively; and They are the flow yield from the permeable area and the flow yield from the impervious area of ​​the basin respectively; is the proportion of impervious area; are the surface runoff, subsoil flow and underground flow generated at time t respectively; and are the outflow coefficient of the subsoil flow and the outflow coefficient of groundwater respectively; They are upper layer tension water capacity, lower soil layer tension water capacity and deep layer tension water capacity; is the calculation function; EM t E is the time series data of evaporation from the evaporating dish; t is the evapotranspiration data at time t, P t is the rainfall time series data; wm is the average tension water storage capacity, b is the distribution index of the basin tension water storage capacity curve, sm is the average free water capacity, and ex is the distribution index of the basin free water capacity curve.

4. The coupling mechanism model and deep learning flood forecasting model according to claim 1, characterized in that: The flood forecasting model is constructed based on the Tensorflow framework; wherein, the flood forecasting model includes multiple ordinary neural network layers, one P-RNN layer, multiple first recurrent neural networks, multiple second neural network layers and a recurrent neural network output layer.

5. A method for training a flood forecasting model based on a coupling mechanism model and deep learning, characterized in that: The method comprises: Constructing a training sample set, wherein the training sample set includes pre-trained meteorological and hydrological data; Based on a preset ratio, the training sample set is divided into a calibration sample set and a validation sample set; Based on a preset loss function, the flood forecasting model described in any one of claims 1 to 4 is trained using the calibration sample set and the verification sample set, and the trained flood forecasting model is obtained by adjusting the epoch, batch-size parameters and the number of neurons of the flood forecasting model.

6. A flood forecasting method, characterized in that: The method comprises: Get current hydrological data; The current hydrological data is input into the flood forecasting model described in any one of claims 1 to 4, and the flood characteristics of the current hydrological data are predicted to obtain the current flood flow data.

7. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the coupling mechanism model and the deep learning flood forecasting model training method described in claim 5 and the flood forecasting method described in claim 6 by executing the computer instructions.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which are used to enable a computer to execute the training method of the coupling mechanism model and deep learning flood forecasting model described in claim 5 and the flood forecasting method described in claim 6.

9. A computer program product, characterized in that It includes computer instructions, which are used to enable a computer to execute the training method of the coupling mechanism model and deep learning flood forecasting model described in claim 5 and the flood forecasting method described in claim 6.

Images