KR102254817B1 - Method and apparatus for real-time ensemble streamflow forecasting - Google Patents

Abstract

A flow forecast method and apparatus are disclosed. A flow rate forecasting method according to an embodiment includes determining one of a pre-parameter distribution and a post-parameter distribution of a rainfall-runoff model as a first parameter distribution, an alternative model of a rainfall-runoff model constructed based on the first parameter distribution Determining any one of surrogate models as a first replacement model, estimating a parameter of the first replacement model based on the first parameter distribution; And performing data assimilation processing on the parameters of the first replacement model, and inputting the parameter on which the data assimilation processing has been performed into the first replacement model, and predicting a streamflow in real time.

Description

Real-time ensemble flow forecast method and device {METHOD AND APPARATUS FOR REAL-TIME ENSEMBLE STREAMFLOW FORECASTING}

The following embodiments relate to a flow rate forecasting method and apparatus, and to an integrated modeling framework that simultaneously satisfies accuracy, predictability, and computational efficiency in order to perform real-time ensemble probability flood forecasting.

Flood damage is increasing worldwide due to urbanization and climate change. Therefore, a lot of budget and effort are required to reduce flood damage caused by heavy rains and typhoons. Methods to reduce flood damage include structural countermeasures for constructing structures such as dams, levees, and reservoirs, and non-structural countermeasures such as flood forecasting and warning systems. Unlike structural countermeasures that require a lot of construction cost and time, non-structural countermeasures can have a quick effect in preventing human damage through forecasting and warning of floods. The flood forecasting and warning system aims to secure time for the residents of the predicted areas to evacuate. As a result, it is possible to reduce personal and property damage in the area expected to be damaged, and it plays a role of supplementing structural countermeasures.

In the field of flood prediction systems worldwide, research on the Ensemble Prediction System (EPS) is increasing. In addition, studies on Ensemble Streamflow Prediction (ESP), which use it to predict flow, have been actively conducted in the field of hydrology. For such ensemble flow prediction, it is necessary to accurately analyze the total uncertainty of the probability prediction in order to express the range of the probability of occurrence, and it is necessary to improve the efficiency so as to have sufficient advance forecasting time.

The embodiments attempt to integrate the advantages of three modeling techniques: surrogate modeling, parameter inference, and data assimilation.

In the embodiments, 18 scenarios are devised according to the construction method of the alternative model, the parameter calculation method used as the model input, and whether the data assimilation parameters are updated, and the optimal combination is found by analyzing the advantages and disadvantages of the scenarios, and the application of the integrated framework. I would like to investigate the possibility and effect.

The embodiments attempt to generate an accurate ensemble for parameters and state variables based on a technique for evaluating the uncertainty of a rainfall-runoff model.

The examples seek to establish a high-efficiency alternative model to quantify the uncertainty and offset the cost of performing data assimilation.

A flow rate forecasting method according to an embodiment includes determining one of a pre-parameter distribution and a post-parameter distribution of a rainfall-runoff model as a first parameter distribution; Determining any one of surrogate models of the rainfall-runoff model constructed based on the first parameter distribution as a first replacement model; Inferring a parameter of the first replacement model based on the first parameter distribution; Performing data assimilation processing on the parameters of the first replacement model; And inputting the parameter on which the data assimilation process has been performed into the first replacement model, and predicting a streamflow in real time.

The performing of the data assimilation processing may include performing the data assimilation processing based on at least one of an ensemble Kalman filter (EnKF) and a dual ensemble Kalman filter (Dual EnKF).

Estimating the parameters of the first replacement model may include: initializing the parameters of the first replacement model based on the first parameter distribution; And initializing a state variable of the first replacement model based on the first parameter distribution.

The determining of the first parameter may include determining the pre-parameter based on past observation data and assumptions; Determining the posterior parameter distribution based on a Generalized Likelihood Uncertainty Estimation (GLUE) technique; And determining one of the pre-parameter distribution and the post-parameter distribution as the first parameter distribution.

The determining of the posterior parameter may include setting a condition corresponding to the performance of the rainfall-runoff model; And determining a response data set that satisfies the condition from among the output data of the rainfall-runoff model.

The setting of the condition may include determining a likelihood function; And determining a threshold value of the likelihood function.

The determining of the threshold may include setting an accuracy index; Setting an efficiency index; And determining a threshold value of the likelihood function based on the accuracy index and the efficiency index.

The determining of the first parameter may include setting a period sufficient to remove the uncertainty of the initial state of the rainfall-runoff model.

The determining as the first replacement model may include building a polynomial chaos expansion (PCE) model based on the first parameter distribution.

The step of constructing the polynomial chaos extension model may include determining the number of learning designs for constructing the polynomial chaos extension model; And determining the order of the polynomial for constructing the polynomial chaos expansion model.

The flow forecasting apparatus according to an embodiment determines any one of a pre-parameter distribution and a post-parameter distribution of a rainfall-runoff model as a first parameter distribution, and replaces the rainfall-runoff model constructed based on the first parameter distribution. Determine any one of the models as a first replacement model, estimate a parameter of the first replacement model based on the first parameter distribution, perform data assimilation processing on the parameter of the first replacement model, and the And a processor that predicts a streamflow in real time by inputting a parameter on which data assimilation processing has been performed into the first replacement model.

The processor may perform the data moving process based on at least one of an ensemble Kalman filter and a dual ensemble Kalman filter.

The processor may initialize a parameter of the first substitution model based on the first parameter distribution, and initialize a state variable of the first substitution model based on the first parameter distribution.

The processor determines the pre-parameter based on past observation data, determines the post parameter based on a Generalized Likelihood Uncertainty Estimation (GLUE) technique, and determines any one of the pre-parameter and the post-parameter as the first parameter. Can be determined by

The processor may set a condition corresponding to the performance of the rainfall-runoff model, and may determine a response data set that satisfies the condition from among output data of the rainfall-runoff model.

The processor may determine a likelihood function and determine a threshold value of the likelihood function.

The processor may set an accuracy index, set an efficiency index, and determine a threshold value of the likelihood function based on the accuracy index and the efficiency index.

The processor may set a period sufficient to remove the uncertainty of the initial state of the rainfall-runoff model.

The processor may build a polynomial chaos expansion model based on the first parameter distribution.

The processor may determine the number of learning designs for constructing the polynomial chaos extension model, and determine the order of the polynomial for constructing the polynomial chaos extension model.

Embodiments may incorporate advantages of three modeling techniques: surrogate modeling, parameter inference, and data assimilation.

In the embodiments, 18 scenarios can be devised according to a method of constructing an alternative model, a method of calculating a parameter used as a model input, and whether parameters are updated during a data assimilation procedure, and an optimal combination can be presented by analyzing the advantages and disadvantages of the scenarios. In this case, the applicability and effectiveness of the integrated framework can be investigated.

Embodiments may generate an accurate ensemble for parameters and state variables based on an uncertainty evaluation technique of a rainfall-runoff model.

Embodiments can build a high-efficiency replacement model to offset the cost of quantifying the uncertainty calculated through parameter inference and data assimilation.

1 is a flowchart illustrating a flow rate forecasting method according to an exemplary embodiment.
2 is a diagram for explaining a real-time flow rate forecasting method according to an exemplary embodiment.
3 is a diagram for describing a method of setting conditions for evaluating uncertainty according to an exemplary embodiment.
4 is a diagram for describing a method of determining a threshold value of a likelihood function based on an accuracy index and an efficiency index, according to an exemplary embodiment.
5 is a diagram for describing a total calculation time according to an exemplary embodiment.
6 is a diagram for explaining an effect of a data moving picture processing according to an exemplary embodiment.
7 is a diagram for describing a flow rate forecasting method according to an exemplary embodiment.
8 is a flowchart illustrating a flow rate forecasting device according to an exemplary embodiment.

Specific structural or functional descriptions disclosed in this specification are exemplified only for the purpose of describing embodiments according to a technical concept, and the embodiments may be implemented in various different forms and are limited to the embodiments described herein. It doesn't work.

Terms such as first or second may be used to describe various components, but these terms should be understood only for the purpose of distinguishing one component from other components. For example, a first component may be referred to as a second component, and similarly, a second component may be referred to as a first component.

When a component is referred to as being "connected" or "connected" to another component, it is understood that it may be directly connected or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there is no other component in the middle. Expressions describing the relationship between the elements, for example, "between" and "just between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present specification, terms such as "comprise" or "have" are intended to designate that the specified features, numbers, steps, actions, components, parts, or combinations thereof exist, but one or more other features or numbers, It is to be understood that the presence or addition of steps, actions, components, parts, or combinations thereof does not preclude the possibility of preliminary exclusion.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the relevant technical field. Terms as defined in a commonly used dictionary should be construed as having a meaning consistent with the meaning of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in the present specification. Does not.

The embodiments may be implemented in various types of products such as a personal computer, a laptop computer, a tablet computer, a smart phone, a television, a smart home appliance, an intelligent vehicle, a kiosk, and a wearable device. Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. The same reference numerals shown in each drawing indicate the same members.

1 is a flowchart illustrating a flow rate forecasting method according to an exemplary embodiment.

Referring to FIG. 1, steps 110 to 150 according to an embodiment may be performed by a flow forecasting device, and the flow forecasting device may be one or more hardware modules, one or more software modules, or It can be implemented in various combinations.

Although the operations of FIG. 1 may be performed in the illustrated order and manner, the order of some operations may be changed or some operations may be omitted without departing from the spirit and scope of the illustrated embodiment. Multiple operations shown in FIG. 1 may be performed in parallel or simultaneously.

In step 110, the flow rate forecasting device determines one of a prior parameter distribution and a posterior parameter distribution of the rainfall-runoff model as the first parameter distribution. The rainfall-runoff model is a deterministic model whose output is fixedly determined according to the input, and may be a model for the relationship between rainfall for a certain period of time in a river basin and the resulting runoff into the river. . The rainfall-runoff model according to an embodiment may include a Needbør-Afstrømnings Model (NAM).

The preparametric distribution of the rainfall-runoff model can be determined based on past observational data and assumptions of distribution. Pre-parameters of the rainfall-runoff model may include parameters having a range as shown in Table 1 below.

The flow forecasting device may determine a prior parameter distribution by randomly extracting a predetermined number (eg, 1,000) of data within a minimum value and a maximum value interval from the past observation data of the parameters of Table 1.

The distribution of the posterior parameters of the rainfall-runoff model can be determined based on the uncertainty evaluation technique. Regarding the uncertainty evaluation technique, the ensemble flow forecasting technique predicts the future floodgate by combining the hydrological conditions at the forecast point and past meteorological data through a rainfall-runoff model under the assumption that meteorological phenomena occurring in the past can be reproduced in the future. It may be a long-term hydrological forecasting technique. For example, a number of hydrological scenarios can be calculated from a rainfall-runoff model by utilizing the hydrological conditions at the forecast point and all possible weather scenarios that may occur in the future, and the representative values can be used as future hydrological forecast values.

Since the ensemble flow forecasting method may have uncertainties due to the structure and parameters of the rainfall-runoff model, meteorological input data, and initial conditions, it may not be possible to accurately model the hydrological behavior of complex natural watersheds. In such an environment, the flow forecasting device may use an uncertainty evaluation technique to extract data that satisfies a result close to an optimal flow rate forecast.

The flow forecasting apparatus according to an embodiment may quantify and evaluate uncertainty based on an uncertainty evaluation technique. Further, the flow rate forecasting apparatus may set conditions for evaluating uncertainty, and extract data satisfying a result close to an optimal flow rate forecast based on the corresponding condition. A set of data that satisfies a result close to an optimal flow rate forecast that satisfies the corresponding condition may be referred to as a behavior data set. As an example of an uncertainty evaluation technique, the flow forecasting device may determine a response data set based on a Generalized Likelihood Uncertainty Estimation (GLUE) technique.

The accuracy-related performance of an uncertainty evaluation technique can vary depending on how you set the conditions for evaluating uncertainty. Hereinafter, a method of setting conditions for evaluating uncertainty according to an exemplary embodiment will be described in detail with reference to FIGS. 3 to 4.

In step 120, the flow rate forecasting device determines any one of meta models of the rainfall-runoff model constructed based on the first parameter distribution as the first replacement model.

The replacement model may mean a model of a model, and in an embodiment, a deterministic model of a rainfall-runoff model, and may be referred to as a surrogate model. Flow forecasting devices can build high-efficiency alternative models to offset the cost of quantifying uncertainty. For example, the flow forecasting device can build an alternative model that has similar performance to the rainfall-runoff model and can significantly reduce the computational speed. For example, the flow forecasting device may build a polynomial chaos expansion (PCE) model based on a response data set.

The flow forecasting device may build a replacement model based on a prior parameter distribution, or build a replacement model based on a posterior parameter distribution. The polynomial chaos expansion model constructed based on the prior parameter distribution may be referred to as PCE-I, and the polynomial chaos expansion model constructed based on the posterior parameter distribution may be referred to as PCE-II.

The polynomial chaos replacement model can provide a functional approximation of the rainfall-runoff model through a spectral representation of a properly constructed polynomial function. Since the polynomial chaos replacement model approximates a complex rainfall-runoff model in a combined form of a simple polynomial, it can perform much faster calculations than the rainfall-runoff model.

In order to construct a polynomial chaos substitution model, it may be necessary to determine the _{PCE coefficient (y α ).} The PCE coefficient can be determined by determining the number of training sets (N) and the order of the polynomial (p). The number of training sets can be referred to as the number of experimental designs for training.

Increasing the order of the polynomial can increase the accuracy of the polynomial chaos substitution model. However, since the number of constant values to be estimated increases exponentially as the order increases, increasing the number of input data and increasing the order may cause calculation inefficiency.

According to an embodiment, the flow forecasting device may determine the number of training sets (N) and the order of the polynomial (p) that can balance accuracy and efficiency. For example, the flow rate forecasting device may determine the number of training sets (N) as 50 and the order of the polynomial as 3rd or 4th.

In step 130, the flow rate forecasting device estimates a parameter of the first replacement model based on the first parameter distribution. For very short-term forecasts, parameter specification and state initialization can play an important role.

Accordingly, the flow rate forecasting device may detail the parameters of the first replacement model based on the first parameter distribution. Also, the flow forecasting device may initialize the state variable of the first replacement model based on the first parameter distribution. In particular, parameter estimation based on posterior parameter distribution can greatly improve future forecasting performance and minimize uncertainty. Parameter estimation based on the prior parameter distribution may be referred to as a “random approach”, and parameter estimation based on the posterior parameter distribution may be referred to as a “selected approach”.

In step 140, the flow rate forecasting device performs data assimilation processing on the parameters of the first replacement model. Data assimilation can mean the process of generating optimal model initial data using all useful observation data to explain the current state. In order to improve forecast accuracy, data assimilation may be necessary to reduce the uncertainty of the initial condition of the model.

The flow rate forecasting apparatus according to an embodiment may perform the data assimilation processing based on at least one of an ensemble Kalman filter (EnKF) and a dual ensemble Kalman filter (Dual EnKF). The ensemble Kalman filter is a Monte Carlo approach to Bayesian update, given the probability density function (pfd) of the model variable and the probability of the data, and can calculate the probability density function by considering the data uncertainty through Bayesian theory. . The ensemble Kalman filter expresses the distribution of model variables as a set of variables and an ensemble, and instead of using a covariance matrix, it can use the covariance of samples calculated from the ensemble.

In step 150, the flow rate forecasting device inputs the parameter on which the data assimilation process has been performed into the first replacement model, and predicts a streamflow in real time. Forecasting the flow rate may be to predict a hydrograph, and the flow forecasting device may also predict the occurrence of a flood through the flow rate forecast. The flow forecasting device may perform a probability flood forecast in real time.

2 is a diagram for explaining a real-time flow rate forecasting method according to an exemplary embodiment.

Referring to FIG. 2, the real-time flow rate forecasting method according to an embodiment may include a warm-up period, a calibration period, and a forecasting period.

In the step of determining the distribution of the first parameter, the flow forecasting device may remove uncertainty in the initial state of the rainfall-runoff model through the warm-up period. In the case of performing the warm-up period, compared to the case in which the warm-up period is not performed, the GLUE technique can generate a response data set much faster. In FIG. 2, n may mean the number of samples of the posterior parameter distribution, and n _w may mean the number of samples of the prior parameter distribution.

An alternative model can be built during the calibration period. The PCE-I model can be built based on the prior parameter distribution, and the PCE-II model can be built based on the posterior parameter distribution.

The flow rate forecasting apparatus may estimate a parameter of the first substitute model based on the first parameter distribution, and may perform data assimilation processing on the parameter of the first substitute model. Thereafter, in the forecasting period, the flow rate forecasting device may perform a probability flood forecast in real time.

3 is a diagram for describing a method of setting conditions for evaluating uncertainty according to an exemplary embodiment.

Referring to FIG. 3, steps 310 to 340 may be performed by the flow rate forecasting apparatus described with reference to FIG. 1.

In step 310, the flow forecasting device may sample data based on past observation data.

The flow forecasting device may randomly extract a predetermined number (eg, 1,000) of data within a minimum value and a maximum value interval from past observation data. For example, the flow forecasting device may sample data based on a Monte Carlo method. The Monte Carlo method may be a simulation method for calculating a value of a desired function using randomly extracted random numbers. However, the Monte Carlo method is only an exemplary method, and is not limited thereto, and various methods of sampling data may be applied.

In step 320, the flow rate forecasting device may input the sampled data into the rainfall-runoff model.

In step 330, the flow rate forecasting device may set a condition corresponding to the performance of the rainfall-runoff model. The flow forecasting device may determine a likelihood function and a threshold value of the likelihood function for evaluating the performance of the rainfall-runoff model. The likelihood function according to an embodiment may include NSE (Nash-Sutcliffe Efficiency), peak error (PE), and volume error (VE). NSE may correspond to a shape of a hydrograph, a peak error may correspond to a peak value of a flow rate, and a volume error may be a likelihood function corresponding to a volume of a flow rate.

Conventionally, in the uncertainty evaluation technique, there has not been a clear method for determining the threshold value of the likelihood function. The flow rate forecasting apparatus according to an embodiment may define an accuracy index (AI) and an efficiency index (EI), and determine a threshold value of a likelihood function based on the accuracy index and the efficiency index.

More specifically, the accuracy index may be defined as in Equation 1.

및

는 시간 t에서 불확실성 분포의 2.5 및 97.5 퍼센타일(percentile)에 대응하는 유량(discharge) 값일 수 있다.In Equation 1, U may mean the time average of the uncertainty identified between the 2.5th to 97.5th percentile of 1,000 behavioral ensemble results for the entire calculation time in the flow rate. k may be an index corresponding to various threshold values, T may be the number of total computational time steps for the total time, and t may be an index for the time from 1 to T.

And

May be discharge values corresponding to 2.5 and 97.5 percentiles of the uncertainty distribution at time t.

The efficiency index may calculate the number of executions (P) of the rainfall-runoff model required until a predetermined number (for example, 1000) of response data sets are obtained. Furthermore, the number of executions (P) of the rainfall-runoff model may be calculated for each k. Based on this, the efficiency index may be defined as in Equation 2.

Here, P _max may be the maximum number of times the model is executed. The accuracy index and efficiency index may be determined between 0 and 1. After determining the accuracy index and the efficiency index, the flow forecasting device may determine a threshold value of the likelihood function based on the two indexes.

Referring to FIG. 4, the flow rate forecasting apparatus according to an embodiment may determine a value of a likelihood function corresponding to an intersection of an accuracy index and an efficiency index as a threshold value.

In the case of NSE, the closer to 1 may mean better model performance, and closer to 0 (%) for PE and VE may mean better model performance.

The accuracy index may have a trend proportional to the value of the likelihood function, and the efficiency index may have a trend that is inversely proportional to the value of the likelihood function. The flow rate forecasting apparatus according to an embodiment may set a threshold value in which accuracy and efficiency are balanced by determining a value of a likelihood function corresponding to an intersection of the accuracy index and the efficiency index as a threshold value. The threshold value determined according to the embodiment of FIG. 3 may be as shown in Table 2 below.

Referring back to FIG. 3, in step 340, the flow rate forecasting device may determine a response data set that satisfies the condition from among the output data of the rainfall-runoff model. The posterior parameter distribution may be determined based on the response data set.

5 is a diagram for describing a total calculation time according to an exemplary embodiment.

Referring to FIG. 5, the total calculation time according to an embodiment may include a replacement model construction time, a parameter inference time, and a data assimilation time. Although it may take a lot of time for parameter inference and data assimilation, which is necessary to improve accuracy, time can be saved by using an alternative model for parameter inference and data assimilation. For example, using an alternative model can be up to 80 times faster than otherwise.

6 is a diagram for explaining an effect of a data moving picture processing according to an exemplary embodiment.

Referring to FIG. 6, graphs are graphs showing performance according to three data assimilation methods (no data assimilation processing, ensemble Kalman filter, and dual ensemble Kalman filter) for 500 ensemble flood peaks during a forecast period.

Although there is no significant difference in actual results between the ensemble Kalman filter and the dual ensemble Kalman filter, there may be a remarkable improvement in reproducing the peak of the water level. If data assimilation processing is not used, the Selected approach gives good results, and if the selection approach is not available, data assimilation processing may need to be applied.

7 is a diagram for describing a flow rate forecasting method according to an exemplary embodiment.

Referring to FIG. 7, the flow forecast method according to an embodiment may incorporate advantages of three modeling techniques: surrogate modeling, parameter inference, and data assimilation.

Specifically, as the model 710 for flow rate prediction, there may be three models of alternative models PCE-I, PCE-I, and rainfall-runoff model (NAM), and the parameter inference method 720 is a selection approach and a random approach. There may be two methods, and the data assimilation processing method 730 may include no data assimilation processing, use of an ensemble Kalman filter, and three methods of using a dual ensemble Kalman filter.

Accordingly, in the flow rate forecasting method according to an embodiment, 18 scenarios may exist as shown in Table 3 below according to a method of constructing an alternative model, a method of calculating a parameter used as a model input, and whether parameters are updated during a data assimilation procedure.

According to the flow rate forecasting method according to an embodiment, the A15 method may be the most excellent method. The A15 method may be a method of determining a replacement model based on a posterior parameter distribution, estimating a parameter of the replacement model based on a selection approach, and performing data assimilation processing based on a dual ensemble Kalman filter on the parameter of the replacement model. .

8 is a flowchart illustrating a flow rate forecasting device according to an exemplary embodiment.

Referring to FIG. 8, the flow rate forecasting apparatus 800 according to an embodiment includes a processor 810. The flow forecasting device 800 may further include a memory 830, a communication interface 850, and sensors 880. The processor 810, the memory 830, the communication interface 850, and the sensors 880 may communicate with each other through the communication bus 805.

The processor 810 determines any one of a pre-parameter distribution and a post-parameter distribution of the rainfall-runoff model as a first parameter distribution, and any of the replacement models of the rainfall-runoff model constructed based on the first parameter distribution One is determined as a first replacement model, and based on the first parameter distribution, a parameter of the first replacement model is estimated, a data assimilation process is performed on the parameter of the first replacement model, and the data assimilation process is performed. By inputting the performed parameters into the first replacement model, the flow rate is predicted in real time.

The memory 830 may store past observation data, a response data set, and the like. The memory 830 may be a volatile memory or a non-volatile memory.

According to an embodiment, the processor 810 may perform data moving process based on at least one of an ensemble Kalman filter and a dual ensemble Kalman filter. The processor 810 may initialize the parameters of the first replacement model based on the first parameter distribution, and initialize the state variables of the first replacement model based on the first parameter distribution. The processor 810 may determine a pre-parameter based on past observation data, determine a post parameter based on a Generalized Likelihood Uncertainty Estimation (GLUE) technique, and determine any one of the pre-parameter and the post-parameter as the first parameter. have. The processor 810 may set a condition corresponding to the performance of the rainfall-runoff model, and may determine a response data set that satisfies the condition from among the output data of the rainfall-runoff model. The processor 810 may determine the likelihood function and determine a threshold value of the likelihood function. The processor 810 may set an accuracy index, set an efficiency index, and determine a threshold value of the likelihood function based on the accuracy index and the efficiency index. The processor 810 may set a period sufficient to remove the uncertainty of the initial state of the rainfall-runoff model. The processor 810 may build a polynomial chaos expansion model based on the first parameter distribution. The processor 810 may determine the number of designs for constructing the polynomial chaos extension model, and determine the order of the polynomial for constructing the polynomial chaos extension model.

In addition, the processor 810 may perform at least one method described above with reference to FIGS. 1 to 7 or an algorithm corresponding to at least one method. The processor 810 may execute a program and control the flow rate forecasting device 800. The program code executed by the processor 810 may be stored in the memory 830. The flow rate forecasting device 800 is connected to an external device (eg, a personal computer or a network) through an input/output device (not shown), and may exchange data. The flow forecasting device 800 may be mounted on various computing devices and/or systems such as a smart phone, a tablet computer, a laptop computer, a desktop computer, a television, and a wearable device.

The embodiments described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices, methods, and components described in the embodiments are, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate (FPGA). array), programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. Further, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to operate as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or, to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

As described above, although the embodiments have been described by the limited drawings, a person of ordinary skill in the art can apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order from the described method, and/or components such as systems, structures, devices, circuits, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and those equivalent to the claims also fall within the scope of the claims to be described later.

Claims (21)

In the flow rate forecasting method in which each step is performed by the flow rate forecasting device,
Determining any one of a pre-parameter distribution and a post-parameter distribution of the rainfall-runoff model as a first parameter distribution;
Determining one of meta models of the rainfall-runoff model constructed based on the first parameter distribution as a first replacement model;
Estimating a parameter of the first replacement model based on the first parameter distribution;
Performing data assimilation processing on the parameters of the first replacement model; And
Inputting the parameter on which the data assimilation processing has been performed into the first replacement model, and predicting a streamflow in real time
Flow forecast method comprising a.
The method of claim 1,
The step of performing the data assimilation processing
Performing the data assimilation processing based on at least one of an ensemble Kalman filter (EnKF) and a dual ensemble Kalman filter (Dual EnKF)
Containing, flow rate forecasting method.
The method of claim 1,
Estimating the parameters of the first replacement model
Initializing the parameters of the first replacement model based on the first parameter distribution; And
Initializing a state variable of the first replacement model based on the first parameter distribution
Containing, flow rate forecasting method.
The method of claim 1,
The step of determining with the first parameter
Determining the prior parameter distribution based on past observation data;
Determining the posterior parameter distribution based on a Generalized Likelihood Uncertainty Estimation (GLUE) technique; And
Determining any one of the prior parameter distribution and the posterior parameter distribution as the first parameter distribution
Containing, flow rate forecasting method.
The method of claim 4,
The step of determining the posterior parameter is
Setting a condition corresponding to the performance of the rainfall-runoff model; And
Determining a response data set that satisfies the condition from among the output data of the rainfall-runoff model
Containing, flow rate forecasting method.
The method of claim 5,
The step of setting the condition is
Determining a likelihood function; And
Determining a threshold value of the likelihood function
Containing, flow rate forecasting method.
The method of claim 6,
The step of determining the threshold value
Setting an accuracy index;
Setting an efficiency index; And
Determining a threshold value of the likelihood function based on the accuracy index and the efficiency index
Containing, flow rate forecasting method.
The method of claim 1,
The step of determining with the first parameter
Setting a period sufficient to remove the uncertainty of the initial state of the rainfall-runoff model
Containing, flow rate forecasting method.
The method of claim 1,
The step of determining as the first replacement model
Building a polynomial chaos expansion (PCE) model based on the first parameter distribution
Containing, flow rate forecasting method.
The method of claim 9,
Building the polynomial chaos expansion model
Determining the number of learning designs for constructing the polynomial chaos expansion model; And
Determining the order of a polynomial for constructing the polynomial chaos expansion model
Containing, flow rate forecasting method.
A computer program stored in a computer-readable recording medium for executing the method of any one of claims 1 to 10 in combination with hardware.
Determine any one of the pre-parameter distribution and the post-parameter distribution of the rainfall-runoff model as a first parameter distribution, and first replace any one of the replacement models of the rainfall-runoff model constructed based on the first parameter distribution A model is determined, based on the first parameter distribution, a parameter of the first replacement model is estimated, a data assimilation process is performed on the parameter of the first replacement model, and the parameter on which the data assimilation process is performed is determined as the parameter. Processor that inputs into the first alternative model and predicts streamflow in real time
Containing, flow forecasting device.
The method of claim 12,
The processor is
A flow forecasting device for performing the data assimilation processing based on at least one of an ensemble Kalman filter and a dual ensemble Kalman filter.
The method of claim 12,
The processor is
Based on the first parameter distribution, initializes a parameter of the first replacement model, and initializes a state variable of the first replacement model based on the first parameter distribution.
The method of claim 12,
The processor is
Determining the pre-parameter based on past observation data, determining the post parameter based on a Generalized Likelihood Uncertainty Estimation (GLUE) technique, and determining any one of the pre-parameter and the post-parameter as the first parameter , Flow forecast device.
The method of claim 15,
The processor is
A flow forecasting device for setting a condition corresponding to the performance of the rainfall-runoff model, and determining a response data set that satisfies the condition from among output data of the rainfall-runoff model.
The method of claim 16,
The processor is
A flow rate forecasting device for determining a likelihood function and determining a threshold value of the likelihood function.
The method of claim 17,
The processor is
A flow forecasting device for setting an accuracy index, setting an efficiency index, and determining a threshold value of the likelihood function based on the accuracy index and the efficiency index.
The method of claim 12,
The processor is
A flow forecasting device for setting a period sufficient to remove the uncertainty of the initial state of the rainfall-runoff model.
The method of claim 12,
The processor is
A flow forecasting device for constructing a polynomial chaos expansion model based on the first parameter distribution.
The method of claim 20,
The processor is
A flow forecasting device for determining the number of designs for building the polynomial chaos expansion model, and determining the order of the polynomial for building the polynomial chaos expansion model.

Images