Multi-step forecasting method and device for river basin flood based on deep learning of transformer

Abstract

The present invention discloses a method and device for multi-step flood forecasting in a watershed based on deep learning of a converter. The method includes: considering the duration of runoff generation and convergence in the watershed, constructing a converter deep learning model based on an adaptive approach, intelligently matching a multi-step flood forecasting architecture, and using a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall; considering the regularization of the objective function, using the Adam optimization algorithm to calibrate the parameters of the converter deep learning model; applying an interpretability method to analyze the contribution of input factors to the accuracy of multi-step flood forecasting, and improving the physical interpretability of the deep learning model. The present invention constructs a converter deep learning model based on an adaptive approach to determine rainfall and flood effective input factors in different flood forecast periods; considering the dual self-attention mechanism and the calibration of model parameters based on the Adam optimization algorithm, thereby not only improving the accuracy of multi-step flood forecasting in the watershed, but also extending the flood forecast period.

Description

Technical Field

[0001] The present invention relates to the technical field of watershed hydrological models, and in particular to a watershed flood multi-step forecasting method and device based on converter deep learning. Background Art

[0002] The data-driven hydrological model treats the studied basin or region as a dynamic system, using input (generally upstream rainfall and upstream main and tributary water) and output (generally basin or regional control section flow) data, such as artificial neural network to establish a mathematical relationship between input and output, and then infer new output from new input. This type of model has been widely used in the fields of water conservancy project planning and design, flood forecasting, water resources development and utilization, and provides theoretical support for solving various engineering hydrological problems and improving people's understanding of hydrological laws.

[0003] With the development of machine learning, data-driven hydrological models have made great progress. However, this type of model only cares about the accuracy and forecast period of simulated forecast results, does not consider the physical causal relationship between input and output, and cannot intelligently match corresponding flood forecasting models and plans according to the objective laws of floods of different magnitudes and durations, which reduces the interpretability and generalization ability of such models. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a basin flood multi-step forecasting method and device based on converter deep learning, which can improve the basin flood forecasting accuracy and extend the flood forecast period, and enhance the generalization ability of the hydrological model.

[0005] According to one aspect of the present invention, a method for multi-step forecasting of river basin floods based on converter deep learning is provided, comprising:

[0006] Obtain source data to be forecasted;

[0007] Input the source data to be predicted into the trained converter deep learning model, and output the predicted flow; wherein the training of the converter deep learning model includes:

[0008] Considering the duration of runoff generation and convergence in the basin, a converter deep learning model based on the adaptive approach is constructed. The converter deep learning model intelligently matches the multi-step flood forecasting architecture and adopts a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall.

[0009] According to the converter deep learning model, the objective function regularization is considered, the Adam optimization algorithm is used to calibrate the model parameters, and the trained converter deep learning model is output.

[0010] As a further technical solution, the method also includes: applying interpretability methods to analyze the contribution of input factors to the accuracy of multi-step flood forecasts and improving the physical interpretability of deep learning models.

[0011] As a further technical solution, the adaptive approach of the converter deep learning model includes a multi-scale decomposer and a multi-scale aggregator, wherein the multi-scale decomposer is designed to extract the duration characteristics of three types of floods through Fourier transform functions and characterize the magnitude characteristics of three types of floods through kernel functions, and the multi-scale aggregator is designed to aggregate the output of the multi-scale decomposer into the final output of the adaptive approach through a weighting method, and intelligently adapt the converter deep learning model to the three types of floods.

[0012] As a further technical solution, the training of the converter deep learning model also includes: for local flow information, an internal self-attention mechanism is used to capture the dependency between the flow of the upstream nearby hydrological station and the flow to be forecasted; for global rainfall information, an external self-attention mechanism is used to capture the dependency between the upstream rainfall center rainfall and the flow to be forecasted.

[0013] According to one aspect of the present invention, a basin flood multi-step forecasting device based on converter deep learning is provided, comprising:

[0014] The first main module is used to obtain source data to be forecasted;

[0015] The second main module is used to input the source data to be predicted into the trained converter deep learning model and output the predicted flow rate; wherein the training of the converter deep learning model includes:

[0016] Considering the duration of runoff generation and convergence in the basin, a converter deep learning model based on the adaptive approach is constructed. The converter deep learning model intelligently matches the multi-step flood forecasting architecture and adopts a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall.

[0017] According to the converter deep learning model, the objective function regularization is considered and the Adam optimization algorithm is used to calibrate the model parameters.

[0018] As a further technical solution, it also includes: a third main module, which is used to apply interpretability methods to analyze the contribution of input factors to the accuracy of multi-step flood forecasts and improve the physical interpretability of deep learning models.

[0019] As a further technical solution, the adaptive approach of the converter deep learning model includes a multi-scale decomposer and a multi-scale aggregator, wherein the multi-scale decomposer is designed to extract the duration characteristics of three types of floods through Fourier transform functions and characterize the magnitude characteristics of three types of floods through kernel functions, and the multi-scale aggregator is designed to aggregate the output of the multi-scale decomposer into the final output of the adaptive approach through a weighting method, and intelligently adapt the converter deep learning model to the three types of floods.

[0020] As a further technical solution, the training of the converter deep learning model also includes: for local flow information, an internal self-attention mechanism is used to capture the dependency between the flow of the upstream nearby hydrological station and the flow to be forecasted; for global rainfall information, an external self-attention mechanism is used to capture the dependency between the upstream rainfall center rainfall and the flow to be forecasted.

[0021] According to one aspect of the present invention, there is provided an execution device, comprising a processor and a memory, wherein the processor is coupled to the memory; the memory is used to store a program; and the processor is used to execute the program in the memory, so that the execution device executes the described method.

[0022] According to one aspect of the present specification, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to execute the method described above.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The present invention constructs a basin flood multi-step forecasting model based on converter deep learning, which can not only intelligently adapt to different flood magnitudes and flood durations, but also make full use of the local flow information of the upstream hydrological station in the basin and the global information of upstream rainfall; it not only combines the objective function regularization and Adam optimization algorithm to calibrate the model parameters, but also adopts the adjoint model based on Shapley value to analyze the contribution of input factors, thereby improving the efficiency of model parameter optimization, and also improving the model interpretability and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction is given below to the drawings used in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 A schematic flow chart of a method for multi-step flood forecasting in a watershed based on deep learning of a converter provided by an exemplary embodiment is shown.

[0027] Figure 2 A schematic diagram of a training process of a converter deep learning model based on adaptive logic provided by an exemplary embodiment is shown.

[0028] Figure 3 A schematic diagram of a post-interpretation flow chart of a converter deep learning model based on adaptive logic provided by an exemplary embodiment is shown.

[0029] Figure 4 A logical schematic diagram of a basin flood multi-step forecasting method based on converter deep learning provided by an exemplary embodiment is shown.

[0030] Figure 5 A logical schematic diagram of an internal self-attention mechanism and an external self-attention mechanism in a multi-scale converter provided by an exemplary embodiment is shown.

[0031] Figure 6 An architectural diagram of a basin flood multi-step forecasting model based on converter deep learning provided by an exemplary embodiment is shown.

[0032] Figure 7 A block diagram of a basin flood multi-step forecasting device based on converter deep learning provided by an exemplary embodiment is shown.

[0033] Figure 8 A block diagram of a deep learning model of a converter based on adaptive logic provided by an exemplary embodiment is shown.

[0034] Fig. 9 A block diagram of an execution device provided by an exemplary embodiment is shown. DETAILED DESCRIPTION

[0035] The terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to the steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0036] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely exemplary and do not necessarily include all the contents and operations / steps, nor do they necessarily have to be executed in the order described. For example, some operations / steps may be further decomposed, while some operations / steps may be combined or partially combined, so the actual execution order may change according to the actual situation.

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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 in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention are arbitrarily combined with each other to form a new technical solution. This combination is not restricted by the sequence of steps and / or the structural composition mode, but must be based on the ability of ordinary technicians in this field to achieve. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that this combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0038] Please refer to Figure 1 ,in Figure 1 A flowchart of a basin flood multi-step forecasting method based on converter deep learning provided by an exemplary embodiment is shown, the method comprising:

[0039] Step 1: Obtain the source data to be forecasted, including numerically predicted rainfall, measured rainfall, and measured flow at hydrological stations.

[0040] Step 2: Input the source data to be forecasted into the trained converter deep learning model and output the forecasted flow.

[0041] Among them, the training of the converter deep learning model includes: considering the duration of runoff generation and convergence in the basin, constructing a converter deep learning model based on an adaptive approach, the converter deep learning model intelligently matches the multi-step flood forecasting architecture, and uses a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall; according to the converter deep learning model, considering the objective function regularization, using the Adam optimization algorithm to calibrate the model parameters, and outputting the trained converter deep learning model.

[0042] It should be noted that the runoff generation and confluence time takes into account the differences in runoff generation and confluence time of floods of different magnitudes.

[0043] Please refer to Figure 2 , Figure 2 A training flow chart of a deep learning model of a converter based on an adaptive approach provided by an exemplary embodiment is shown. The training process is implemented through steps 2.1 to 2.2.

[0044] Please refer to Figure 3 , Figure 3 A schematic diagram of a post-interpretation process of a deep learning model of a converter based on adaptive logic provided by an exemplary embodiment is shown. After the model training is completed, the contribution is analyzed through step 2.3 to improve the physical interpretability of the deep learning model.

[0045] Please refer to Figure 4 , Figure 4 The logical principle diagram of a multi-step flood forecasting method for a watershed based on deep learning of a converter provided by an exemplary embodiment is shown. The source data for forecasting is calculated by the forecasting logic to output the final forecast flow.

[0046] In step 2.1, considering the different flood magnitudes and durations of super floods, major floods, and small and medium floods, an adaptive approach is adopted to intelligently match the multi-step flood forecasting architecture according to different flood characteristics, where the adaptive approach consists of a multi-scale decomposer and a multi-scale aggregator;

[0047] The multi-scale decomposer aims to extract the duration characteristics of three types of floods through Fourier transform function and characterize the magnitude characteristics of three types of floods through kernel function. The flood duration characterization formula is as follows

[0048] X T =IDFT({p1, p2, p3}, B, φ)

[0049] Where: B and φ represent the phase set and amplitude set of the three types of floods, p1p2 and p3 are the occurrence frequencies of the three types of floods, IDFT is the inverse function of the Fourier transform function, and X T The flood duration characterization formulas for three types of floods are given below.

[0050] The flood magnitude characterization formula is as follows

[0051]

[0052] Where: X is the flood sample set, L(·) is the weight control function of the kernel function result, Softmax(·) is the conversion function of the multi-classification problem output, They are the conversion functions of three types of flood kernel functions.

[0053] The expressions to characterize the three types of floods by flood duration and flood magnitude are as follows:

[0054] X trans =Linear(X T , X M )

[0055] Where: X trans are representations of three types of floods, and Linear(·, ·) is a linearized function.

[0056] The adaptive approach is achieved by

[0057] R(X trans )=Softmax(X trans W r +γ·Softplus(X trans W noise ))

[0058] γ~N(0,1)

[0059] Where: R(X trans ) is the adaptive pathway function, W r and W noise is a learnable parameter for generating weights, γ is a random number that follows a standard normal distribution, and Softplus(·) is a power-exponential logarithmic conversion function.

[0060] The multi-scale aggregator aims to aggregate the multi-scale decomposer outputs into the final output of the adaptive pathway through a weighting method, intelligently adapting the converter deep learning models of the three types of floods.

[0061] If the flood samples can be decomposed into D batches, the multi-scale aggregation calculation formula is as follows:

[0062]

[0063]

[0064] Where: is the output of the dth batch of adaptive pathways that meet the characteristics of the three types of floods, Φ(·) is the indicator function, R(X trans ) d is the dth batch output of the adaptive pathway, T d (·) is the conversion function of the d-th batch output, is the d-th batch output of the multi-scale aggregator, and D is the number of batches.

[0065] Considering the flood formation mechanism, the floods at different hydrological control stations in the basin are not only affected by the flow of the upstream adjacent hydrological stations (runoff generation and confluence), but also by the upstream rainfall distribution (rainfall center). The information of the upstream adjacent hydrological stations is called local information, and the upstream rainfall center is called global information.

[0066] Please refer to Figure 5 For local flow information, the internal self-attention mechanism is used to capture the dependency between the flow of the upstream adjacent hydrological station and the flow to be predicted. The calculation formula is as follows

[0067]

[0068]

[0069] Where: is the dependency between the flow of the upstream adjacent hydrological station and the flow to be forecasted in the dth batch, is the intermediate variable of the internal self-attention mechanism, L j is the embedding dimension of the jth local information input factor (such as upstream flow), A intra is the dependency between the flow of the upstream adjacent hydrological station and the flow to be predicted, and Concat(·) is the concatenation function.

[0070] Please refer to Figure 5 , for the global rainfall information, an external self-attention mechanism is used to capture the dependency between the upstream rainfall center and the flow to be forecasted. The calculation formula is as follows

[0071]

[0072] L j =S·L j

[0073] Where: A inter is the dependence between the upstream rainfall center rainfall and the flow to be forecast, U inter , G inter 、V inter is the intermediate variable of the external self-attention mechanism, S is a subset of the input factors (such as upstream rainfall), and L′ j is the embedding dimension of the jth global information input factor (such as upstream rainfall).

[0074] In step 2.2, the converter deep learning model parameters are calibrated using the following formula:

[0075]

[0076] Where: L is the regularization objective function; MSE is the mean square error; Q s (t) is the measured discharge at the sth hydrological station in the basin; is the forecast flow of the sth hydrological station in the basin; μ is the regularization coefficient; ω i is the i-th weight parameter of the hidden layer; M is the number of neurons in the hidden layer.

[0077] Considering that the use of regularized objective function to calibrate model parameters is prone to cause the curse of dimensionality problem, the Adam optimization algorithm is used to learn in batches (the number of batches is equal to that of step one) to calibrate the parameters of the converter deep learning model to improve the efficiency of parameter optimization.

[0078] In step 2.3, the contribution analysis method of multi-step flood forecast accuracy is implemented through the following model, using Shapley value to describe the adjoint model of the interpretation:

[0079]

[0080] Where: x′ j ∈{0, 1} indicates whether the jth input factor can be observed, 0 means it cannot be observed, and 1 means it can be observed; K is the number of input factors; ψ j is the Shapley value of the jth input factor; ψ0 is the constant of the explanatory model, whose value is equal to the predicted mean of all training samples; x′ is the set describing whether the input factor can be observed; g(x′) is the adjoint model function of the input factor.

[0081] For a trained model, we define f x (S) = E[f(x)|x S ], then the Shapley value calculation of the jth input factor is performed by the following formula

[0082]

[0083] Where: x is the input factor set; f x (S) is a model trained using a subset S of the input factor set x.

[0084] Shapley value ψ j It can be a positive value, a negative value, or 0, representing that the j-th input factor is positively correlated, negatively correlated, or uncorrelated with the output variable, respectively. The larger its absolute value, the greater the contribution of the corresponding input factor to the accuracy of multi-step flood forecasting.

[0085] Figure 6 The diagram shows an architecture diagram of a basin flood multi-step forecasting model based on converter deep learning provided by an exemplary embodiment. Using the trained model for flood multi-step forecasting can improve the basin flood forecasting accuracy and extend the flood forecast period, thereby enhancing the generalization ability of the hydrological model.

[0086] The implementation basis of each embodiment of the present invention is to implement programmed processing through a device with a processor function. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the present invention are encapsulated into various modules. Based on this reality, on the basis of the above embodiments, an embodiment of the present invention provides a basin flood multi-step forecasting device based on converter deep learning, which is used to execute the basin flood multi-step forecasting method based on converter deep learning in the above method embodiment.

[0087] Please refer to Figure 7 , which shows a block diagram of the basin flood multi-step forecasting device based on converter deep learning provided by the present application, the device includes: a first main module 102, used to obtain the source data to be forecasted; a second main module 104, used to input the source data to be forecasted into the trained converter deep learning model, and output the forecast flow.

[0088] Please refer to Figure 8 , which shows a block diagram of the deep learning model of the converter based on the adaptive approach provided by the present application, the model includes:

[0089] The first training module 202 considers the duration of runoff generation and confluence in the basin, builds a deep learning model of the converter based on the adaptive approach, intelligently matches the multi-step flood forecasting architecture, and uses a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall;

[0090] The second training module 204 considers the regularization of the objective function and uses the Adam optimization algorithm to calibrate the parameters of the converter deep learning model.

[0091] The multi-step flood forecasting device for a river basin based on deep learning of a converter provided in an embodiment of the present invention adopts the current situation that the prior art does not consider the physical causal relationship between input and output and cannot intelligently match the corresponding flood forecasting models and schemes according to the objective laws of floods of different magnitudes and durations. Figure 5 Several modules in it can not only intelligently adapt to different flood magnitudes and durations, but also make full use of the local flow information of the upstream hydrological stations in the basin and the global information of upstream rainfall. It can improve the accuracy of basin flood forecasting, extend the flood forecast period, and enhance the generalization ability of the hydrological model.

[0092] It should be noted that the device embodiments provided by the present invention are used to implement the methods in the above method embodiments as well as the methods in other method embodiments provided by the present invention. The only difference is that the corresponding functional modules are set, and the principle is basically the same as the principle of the above device embodiments provided by the present invention. As long as the technical personnel in this field refer to the specific technical solutions in other method embodiments on the basis of the above device embodiments, obtain the corresponding technical means and the technical solutions composed of these technical means by combining the technical features, and on the premise of ensuring the practicality of the technical solutions, they can improve the equipment in the above device embodiments to obtain the corresponding device class embodiments, which are used to implement the methods in other method class embodiments. For example:

[0093] Based on the content of the above device embodiment, as a preferred embodiment, the basin flood multi-step forecasting device based on converter deep learning provided in the embodiment of the present invention further includes:

[0094] The third main module is used to apply interpretability methods to analyze the contribution of input factors to the accuracy of multi-step flood forecasts and improve the physical interpretability of deep learning models.

[0095] Based on the contents of the above-mentioned device embodiments, as a preferred embodiment, in the basin flood multi-step forecasting device based on converter deep learning provided in the embodiments of the present invention, the adaptive approach of the converter deep learning model includes: a multi-scale decomposer and a multi-scale aggregator, wherein the multi-scale decomposer is intended to extract the duration characteristics of three types of floods through Fourier transform functions and characterize the magnitude characteristics of three types of floods through kernel functions, and the multi-scale aggregator is intended to aggregate the output of the multi-scale decomposer into the final output of the adaptive approach through a weighting method, and intelligently adapt the converter deep learning model of the three types of floods.

[0096] Based on the content of the above-mentioned device embodiment, as a preferred embodiment, the multi-step forecasting device for river basin flood based on converter deep learning provided in the embodiment of the present invention, the training of the converter deep learning model also includes: for local flow information, an internal self-attention mechanism is used to capture the dependency between the flow of the upstream adjacent hydrological station and the flow to be forecasted; for global rainfall information, an external self-attention mechanism is used to capture the dependency between the upstream rainfall center rainfall and the flow to be forecasted.

[0097] Since the above methods have been discussed in detail, the specific implementation of the above modules will not be repeated here.

[0098] Next, an execution device provided by an embodiment of the present invention is introduced. Fig. 9 , Fig. 9The schematic diagram of the structure of the execution device provided in the embodiment of the present invention is as follows. The execution device 300 can be specifically manifested as an autonomous driving vehicle, a mobile phone, a tablet, a laptop computer, a desktop computer, a monitoring data processing device, etc., which are not limited here. Figure 1 The functions of the execution device in the corresponding embodiment. Specifically, the execution device 300 includes: a receiver 301, a transmitter 302, a processor 303 and a memory 304 (wherein the number of the processor 303 in the execution device 300 can be one or more, Fig. 9 In the example of FIG. 3 , the processor 303 may include an application processor 3031 and a communication processor 3032. In some embodiments of the present application, the receiver 301, the transmitter 302, the processor 303 and the memory 304 may be connected via a bus or other means.

[0099] The memory 304 may include a read-only memory and a random access memory, and provides instructions and data to the processor 303. A portion of the memory 304 may also include a non-volatile random access memory (NVRAM). The memory 304 stores processor and operation instructions, executable modules or data structures, or subsets thereof, or extended sets thereof, wherein the operation instructions may include various operation instructions for implementing various operations.

[0100] The processor 303 controls the operation of the execution device. In a specific application, the various components of the execution device are coupled together through a bus system, wherein the bus system includes not only a data bus but also a power bus, a control bus, and a status signal bus, etc. However, for the sake of clarity, various buses are referred to as bus systems in the figure.

[0101] The method disclosed in the above embodiment of the present invention can be applied to the processor 303, or implemented by the processor 303. The processor 303 can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 303 or the instructions in the form of software. The above processor 303 can be a general processor, a digital signal processor (digital signal processing, DSP), a microprocessor or a microcontroller, and can further include an application specific integrated circuit (application specific integrated circuit, ASIC), a field programmable gate array (field-programmable gate array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The processor 303 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or a combination of hardware and software modules in the decoding processor to execute. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 304, and the processor 303 reads the information in the memory 304 and completes the steps of the above method in combination with its hardware.

[0102] The receiver 301 can be used to receive input digital or character information and generate signal input related to the relevant settings and function control of the execution device. The transmitter 302 can be used to output digital or character information through the first interface; the transmitter 302 can also be used to send instructions to the disk group through the first interface to modify the data in the disk group; the transmitter 302 can also include a display device such as a display screen.

[0103] In the embodiment of the present invention, the processor 303 is used to execute Figure 1 The specific manner in which the application processor 3031 in the processor 303 performs the above steps is the same as that in the present application. Figure 1 The corresponding method embodiments are based on the same concept, and the technical effects they bring are the same as those in this application. Figure 1 The corresponding method embodiments are the same. For specific contents, please refer to the description in the method embodiments shown above in this application, which will not be repeated here.

[0104] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may 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-ROM, optical storage, etc.) containing computer-usable program code.

[0105] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, 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 device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0106] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0108] In summary, the present invention considers the duration of runoff generation and convergence in the basin, constructs a deep learning model of the converter based on the adaptive approach, intelligently matches the multi-step flood forecasting architecture, and uses a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall; considers the regularization of the objective function, and uses the Adam optimization algorithm to calibrate the parameters of the converter deep learning model; applies the interpretability method to analyze the contribution of input factors to the accuracy of multi-step flood forecasts, and improves the physical interpretability of the deep learning model. In this method, a deep learning model of the converter based on the adaptive approach is constructed to determine the effective input factors of rainfall and floods in different flood forecast periods; considers the dual self-attention mechanism and the calibration of model parameters based on the Adam optimization algorithm, thereby not only improving the accuracy of the multi-step flood forecast in the basin, but also extending the flood forecast period.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A multi-step flood forecasting method for a river basin based on deep learning of a converter, characterized in that: include: Obtain source data to be forecasted; Input the source data to be predicted into the trained converter deep learning model, and output the predicted flow; wherein the training of the converter deep learning model includes: Considering the duration of runoff generation and convergence in the basin, a converter deep learning model based on an adaptive approach is constructed. The converter deep learning model intelligently matches the multi-step flood forecasting architecture and uses a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall. The adaptive approach of the converter deep learning model includes a multi-scale decomposer and a multi-scale aggregator. The multi-scale decomposer aims to extract the duration characteristics of three types of floods through the Fourier transform function and characterize the magnitude characteristics of three types of floods through the kernel function. The multi-scale aggregator aims to aggregate the output of the multi-scale decomposer into the final output of the adaptive approach through the weighting method, and intelligently adapt the converter deep learning model of the three types of floods. For local flow information, an internal self-attention mechanism is used to capture the dependency between the flow of the upstream adjacent hydrological station and the flow to be forecasted. For global rainfall information, an external self-attention mechanism is used to capture the dependency between the rainfall in the upstream rainfall center and the flow to be forecasted. According to the converter deep learning model, the objective function regularization is considered, the Adam optimization algorithm is used to calibrate the model parameters, and the trained converter deep learning model is output.

2. The method for multi-step flood forecasting in a river basin based on converter deep learning according to claim 1 is characterized in that: The method also includes: applying an interpretability method to analyze the contribution of input factors to the accuracy of multi-step flood forecasting and improving the physical interpretability of the deep learning model.

3. A multi-step flood forecasting device based on converter deep learning, characterized in that: include: The first main module is used to obtain source data to be forecasted; The second main module is used to input the source data to be predicted into the trained converter deep learning model and output the predicted flow rate; wherein the training of the converter deep learning model includes: Considering the duration of runoff generation and convergence in the basin, a converter deep learning model based on an adaptive approach is constructed. The converter deep learning model intelligently matches the multi-step flood forecasting architecture and uses a dual self-attention mechanism to capture the local and global dependencies between floods and rainfall. The adaptive approach of the converter deep learning model includes a multi-scale decomposer and a multi-scale aggregator. The multi-scale decomposer aims to extract the duration characteristics of three types of floods through the Fourier transform function and characterize the magnitude characteristics of three types of floods through the kernel function. The multi-scale aggregator aims to aggregate the output of the multi-scale decomposer into the final output of the adaptive approach through the weighting method, and intelligently adapt the converter deep learning model of the three types of floods. For local flow information, an internal self-attention mechanism is used to capture the dependency between the flow of the upstream adjacent hydrological station and the flow to be forecasted. For global rainfall information, an external self-attention mechanism is used to capture the dependency between the rainfall in the upstream rainfall center and the flow to be forecasted. According to the converter deep learning model, the objective function regularization is considered and the Adam optimization algorithm is used to calibrate the model parameters.

4. The basin flood multi-step forecasting device based on converter deep learning according to claim 3 is characterized in that: Also includes: The third main module is used to apply interpretability methods to analyze the contribution of input factors to the accuracy of multi-step flood forecasts and improve the physical interpretability of deep learning models.

5. An execution device, characterized in that: It comprises a processor and a memory, wherein the processor is coupled to the memory; the memory is used to store programs; the processor is used to execute the programs in the memory, so that the execution device executes the method as claimed in any one of claims 1 to 2.

6. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method of any one of claims 1 to 2.

Images