KR20140004915A - Multi-model probabilistic prediction system and method thereof - Google Patents

Abstract

The present invention relates to a system and a method for predicting the probability of multi-models. The method can revise the systematic error of individual models participating in multi-models by predicting global climate using a multi-variable step-wise pattern projection method (MSPM) and can increase the accuracy of probability using a multi-model ensemble (MME) formed of only excellent models of which a predictive fluctuation range is revised. The system comprises: a systematic error revising part for revising the systematic error of the individual models using the MSPM; a model selecting part for selecting models based on the predictability of predictive parameters of the individual models revised by the MSPM; a predictive fluctuation range revising part for revising the predictive fluctuation range of the individual models selected by the model selecting part based on the fluctuation range of observation using a statistical method; and a probability predicting part for predicting the probability by forming the MME of the multi-models of which predictive fluctuation range is revised after the revision of the systematic error and the selection of the models. [Reference numerals] (10) Systematic error revising part; (20) Excellent model selecting part; (30) Predictive fluctuation range revising part; (40) Probability predicting part

Description

TECHNICAL FIELD The present invention relates to a multi-model probabilistic prediction system,

The present invention relates to global climate prediction. More particularly, the present invention relates to a global climate prediction method, and more particularly, to a system and method for correcting systematic errors of individual models participating in multiple models using a multivariate sequential pattern projection technique (MSPM) (MME), thereby improving the accuracy of the probability prediction.

As the damage to the global environment has increased rapidly, a system of predicting the earth change has been developed to predict the results and to obtain the results.

The global environment refers to one or more natural environmental elements constituting the global environment such as the atmosphere, reservoir, lithosphere, biosphere, and cryosphere.

These global environmental factors are either limited or impossible to test, and numerical prediction models of global environmental changes are obtained by substituting observations or experimental values into equations representing natural laws that dominate global environmental factors based on computers. have.

Because global environmental change occurs mainly on a large scale, it requires a lot of computation in order to obtain a high-resolution result desired by a user using a numerical prediction model.

Numerical models including regional climate models have very different forecasting levels for weather and climate, depending on the simulated environment of the model, such as usage model, spatial resolution, combination of physical processes, season, geographical location, and boundary conditions.

Especially, the forecasting level of temperature and precipitation closely related to our daily life is more affected by the model simulation environment. Recently, ensemble prediction techniques have been used as a solution to solve these problems.

The ensemble prediction is a method of predicting a plurality of simulation results calculated by using the same model or different initial conditions or using one or more models.

At this time, the method of averaging the simulation data uses a weighted average giving a high weight to a model with high accuracy through various analytical methods such as simple averaging which gives equal weight to ensemble members and empirical orthogonal analysis.

Simple averaging has the advantage of a simple calculation method, but there is a problem that the prediction level is not much improved than that of an individual model.

On the other hand, the application of complex statistical methods such as empirical orthogonal analysis has the advantage that the prediction level is better than that of the individual models. However, accurate statistical analysis has a problem that the amount of data is large and the calculation is complicated.

In order to solve the problem of the conventional global climate prediction method, the present invention uses a multivariate sequential pattern projection technique (MSPM) to correct systematic errors of individual models participating in multiple models, And an object of the present invention is to provide a multi-model probability prediction system and method capable of improving the accuracy of probability prediction by performing a multiple model ensemble (MME) using only the corrected best model.

The present invention provides a method and system for estimating the probability of a multi-model of a global circulation model by processing statistical methods on the predicted data of the global area (360 占 180 占) have.

The objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

According to another aspect of the present invention, there is provided a system for predicting a multi-model probability, comprising: a systematic error corrector for correcting a systematic error of an individual model using a multivariate sequential pattern projection method; A model selection unit for selecting a model based on the prediction variable prediction power of the individual model corrected through the multivariate sequential pattern projection technique (MSPM); A predicted fluctuation width correcting unit for correcting the predicted fluctuation width of the individual model selected by the model selecting unit based on the variation of the observation by a statistical method; And a probability predicting unit for performing a probability prediction by using a multiple model ensemble (MME) for multiple models having the systematic error correction and the post-model-selected predictive fluctuation correction.

Here, the systematic error correcting unit divides the global region of the global circular model into unit grids in order to correct the systematic error of the model used for prediction, and each unit grid divided by the unit grid divider. A correlation region estimator that estimates the correlation region with high correlation between the predictor and the predictor, and an association relationship by statistically analyzing the time series variability of the predictor and the pattern variability of the associated region of the predictor. And an extractor and a predictor corrector for correcting a predictor corresponding to a unit grid of a solution to be actually predicted through the correlation information extracted in the coupling pattern by the correlation extractor.

In order to select a model based on the predictive power of the predicted variable of the individual model corrected through the MSPM algorithm, the model selecting unit selects a corrected predicted variable of past data common to all models participating in the multiple model ensemble (MME) If the predictive force is equal to or greater than the reference value, it is determined as an excellent model. If the predictive force is equal to or less than the reference value, the predictive force is regarded as low and excluded at the time of performing the MME.

The predictive fluctuation width correcting unit multiplies the corrected predicted value per unit lattice of the individual model of the solution to be actually predicted by the ratio of the standard deviation of the observed data to the standard deviation of the predicted value corrected through the 23-year MSPM algorithm, And the correction is performed.

According to another aspect of the present invention, there is provided a method for predicting a probability of a multi-model sequential pattern projection method (MSPM) by statistically processing prediction data on a global region of a global circulation model, Correcting systematic errors of individual models; Selecting a model based on the predictive variable prediction power of the individual model corrected through the multivariate sequential pattern projection technique (MSPM); Correcting the predicted fluctuation width of the selected individual model on the basis of the fluctuation width of the observation by a statistical method; And performing a probability prediction by using a multi-model ensemble (MME) for multiple models having the systematic error correction and the post-model-selected prediction-amplitude correction.

Here, the probability prediction information by the probability prediction step is a probability value with respect to a normal level, a normal level, and a lower level than a normal level.

The step of correcting the systematic error of the individual model, the process of dividing the global region of the global circular model into a unit grid, and the high correlation between the predictors and predictors corresponding to each unit grid based on past data The process of estimating the region, statistically analyzing the time series variability of the predictor variable and the pattern variability of the associated region of the predictor, extracting the correlation, and the unit of the solution to be actually predicted through the information extracted from the combined pattern. Characterized in that the process of correcting the predicted variable corresponding to the grid.

The historical data of the global circulation model used to extract the associations use the period (1981-2003) common to all models participating in the MME, and the predictive factors are 500 hPa altitude, 850 hPa temperature, And the predicted parameters corrected from the respective predictive factors are finally simply averaged.

The step of selecting the model based on the predicted predictive power of the corrected individual model is performed when the predictive power of the corrected predictive variable of the past data common to all the models participating in the multiple model ensemble (MME) And the reference value for selecting the model is set to a correlation coefficient of 0.3 between the predicted variable and the time series volatility of the observed data.

The step of correcting the predicted fluctuation width of the individual model based on the fluctuation of the observations by a statistical method includes a step of correcting the corrected predicted value per unit lattice of the individual model of the solution to be actually predicted by using the 23- And multiplying the ratio of the standard deviation of the observed data to the standard deviation to correct the predicted variation.

)과 분산(

)으로부터 기후자료의 확률분포를 추정하는 과정과,상기 확률분포로부터 평년보다 높음, 평년 수준, 평년보다 낮음, 세 개의 카테고리를 정의하는 기준점 a, b를 추정하는 과정과, 예측력 높은 모델로 선정되어 보정된 예측자료의 확률분포를 이용하여 기후자료 확률분포로부터 결정된 2개의 기준값에 따른 각 카테고리에 해당하는 확률을 정규확률밀도 함수로 계산하는 과정으로 이루어지는 것을 특징으로 한다.And performing the probability prediction by using the multiple model ensemble (MME)

) And variance (

), The process of estimating the probability distribution of climate data, the process of estimating the reference points a and b that define three categories from the probability distribution, higher than normal years, lower than normal years, and average years, and models with high predictive power. Using the probability distribution of the corrected prediction data, it is characterized by consisting of calculating the probability corresponding to each category according to the two reference values determined from the probability distribution of climate data as a normal probability density function.

The multi-model probability prediction system and method according to the present invention have the following effects.

First, the systematic errors of individual models participating in multiple models are corrected using the multivariate sequential pattern projection method (MSPM), and the accuracy of probability prediction is improved by performing a multiple model ensemble (MME) .

Secondly, we can provide multi-model probability prediction information which can increase the prediction level by using MME with only the superior model that corrects the systematic error of the model and corrects the prediction fluctuation.

Third, by using multi - model probability prediction information with higher prediction level, it is possible to increase the reliability of climate prediction by reducing uncertainty in predicting climate.

1 is a block diagram of a multi-model probability prediction system according to the present invention;
2 is a flowchart showing a method of predicting a multi-model probability using a multi-model probability prediction system according to the present invention.
FIG. 3 is a diagram illustrating a systematic error correction of an individual model through an MSPM algorithm in a multi-model probability prediction system according to the present invention.
FIG. 4 is a block diagram illustrating an excellent model selection and prediction fluctuation correction in the multi-model probability prediction system according to the present invention.
5 is a graph showing probability prediction information production using multiple models in a multi-model probability prediction system according to the present invention.
6 is a graph showing prediction characteristics of a multi-model probability prediction system according to the present invention

Hereinafter, a preferred embodiment of a multi-model probability prediction system and method according to the present invention will be described in detail.

The features and advantages of the multi-model probability prediction system and method according to the present invention will be apparent from the following detailed description of each embodiment.

FIG. 1 is a configuration diagram of a multi-model probability prediction system according to the present invention, and FIG. 2 is a flowchart illustrating a multi-model probability prediction method using the multi-model probability prediction system according to the present invention.

The present invention relates to a method and apparatus for estimating multi-model probability prediction information (MME) using a multi-variable Step-wise Pattern Projection Method (MSPM) algorithm and correcting a systematic error of a model, .

As shown in FIG. 1, the multi-model probability prediction system according to the present invention includes a systematic error correction unit 10 for correcting a systematic error of an individual model, an excellent model selection unit 20 for selecting an excellent model having a high prediction power, A prediction fluctuation width correcting unit 30 for correcting a predicted fluctuation width between individual models, and a probability predicting unit 40 for performing a probability prediction using multiple models.

Here, the systematic error correction unit 10 includes a unit grid dividing unit (not shown) for dividing the global area of the global circulation model into a unit grid to correct systematic errors of the model using a MSPM (Multi-variable Step-wise Pattern Projection Method) 11), an associated region estimator (12) for estimating an associated region having a high degree of correlation between a prediction parameter and a prediction parameter corresponding to each unit grid divided by the unit grid division unit (11) (13) for statistically analyzing the pattern variability of the region where the predictive factors are related to each other and extracting an association, and an association extractor (13) And a predictive variable corrector 14 for correcting a predictive variable corresponding to a unit lattice of a solution to be solved.

In order to select an excellent model based on the predictive power of the individual model corrected through the MSPM algorithm, the best model selection unit 20 calculates the predicted power of the corrected predicted variables of the past data common to all the models participating in the MME If it is greater than or equal to the reference value, it is determined as an excellent model. If it is below the reference value, the model is regarded as having a low prediction power and is excluded when performing the MME.

In order to accurately estimate the uncertainty of the prediction, the predictive fluctuation width correcting unit 30 corrects the predicted fluctuation of the individual model by using a statistical method similar to the fluctuation of the observation, Is calculated by multiplying the corrected predicted value by the ratio of the standard deviation of the observed data to the standard deviation of the corrected predicted value through the 23-year MSPM algorithm.

The probability estimator 40 corrects the systematic errors of the individual models participating in the multiple models by using a multi-variable step-wise pattern projection method (MSPM) The accuracy of probability prediction is improved by performing a multi-model ensemble (MME) in a state in which the selection and prediction fluctuation are corrected.

The multi-model probability prediction method using the multi-model probability prediction system according to the present invention having the above-described configuration is as follows.

The multi-model probability prediction method according to the present invention improves the probability prediction accuracy by processing a statistical method on the predicted data for the global region (360 占 180 占) of the global circulation model.

Here, the probability prediction information means probability values for three categories: higher than normal / normal / lower than normal.

The method of predicting a multi-model probability according to the present invention includes a step S201 of correcting a systematic error of an individual model, a step S202 of selecting an excellent model having a high prediction power, a step S203 of correcting a prediction fluctuation width between individual models, And estimating a probability using the multiple models (S204).

In the step S201 of correcting the systematic errors of the individual models, the global area of the global circulation model is divided into a grid of 144 × 73 with a unit of 2.5 ° × 2.5 ° as shown in FIG. 3, Based on the assumption that there is a correlation between predictive variables (eg, latitude 25 ° N, hardness 100 ° E) and predictor variables, And then corrects the predictive variable corresponding to the unit lattice of the solution to be actually predicted through the information extracted from the combined pattern.

Here, the historical data of the global circulation model used to extract associations use the period (1981-2003) common to all models participating in the MME.

In addition, the prediction factors include 500 hPa altitude, 850 hPa temperature, and sea surface pressure, and the predicted variables calibrated from each predictor are finally averaged, and the error correction algorithm of this model is defined as MSPM.

In step S202 of selecting an excellent model having a high predictive power, an excellent model is selected based on the predictive variable predictive power of the individual model corrected through the MSPM algorithm as shown in FIG.

If the predictive power of the corrected predictive variables of the past data is more than the reference value, it is determined as the superior model. If it is below the reference value, it is regarded as the low predictive power and is excluded when performing the MME.

Here, it is desirable that the reference value for selecting an excellent model be a correlation coefficient of 0.3 between the predicted variable and the time series volatility of the observed data.

The step of correcting the predicted fluctuation width between the individual models (S203) is performed before the MME of the individual predicted values of the individual models selected as the excellent models, as shown in FIG. 4, in order to accurately estimate the uncertainty of the predictions, Is to apply a technique to correct the variation of the observations similar to the statistical method.

In this step, by multiplying the corrected predicted value per unit lattice of the individual model of the solution to be actually predicted by the ratio of the standard deviation of the observed data to the standard deviation of the corrected predicted value through the 23-year MSPM algorithm, The variation is corrected to be similar to the fluctuation of the observation.

The step of estimating the probability using the multiple models (S204) is to finally generate the probability information based on the corrected predicted data of the model selected as the predictive model through various steps, as shown in FIG.

In other words, not a deterministic MME technique that determines the corrected values of several models as a single value, but to provide more useful information to users, it considers each model prediction data as one sample, The probability distribution is estimated by using the data and the probabilistic MME method is applied to provide probability information about higher / same level / lower than normal by utilizing this probability distribution.

)과 분산(

)으로부터 기후자료의 확률분포를 추정하고, 그 확률분포로부터 세 개의 카테고리를 정의하는 기준점 a, b를 추정한다.First, the average of observed climate data (

) And variance (

), And estimates the reference points a and b that define the three categories from the probability distribution.

Using the probability distribution of the corrected forecast data selected as a predictive model, the probability corresponding to each category based on the two reference values determined from the probability distribution of climate data is calculated as a normal probability density function.

Here, f (x) is a normalized probability density function defined as follows.

That is, the present invention provides multi-model probability prediction information with high prediction power by performing MME using only an excellent model that corrects a systematic error of a model and corrects a prediction fluctuation with an MSPM algorithm.

As shown in FIG. 5, the multi-model probability prediction system and method according to the present invention show improved predictive power as compared with the prior art probability prediction system and method.

The multi-model probability prediction system and method according to the present invention is related to a multi-model probability estimation (MMPP) system for global middle / long-term forecasting (monthly, seasonal), and utilizes a multivariate sequential pattern projection method (MSPM) It improves the accuracy of probability prediction by correcting the systematic error of individual models participating in multiple models, selecting an excellent model among individual models, and correcting the prediction fluctuation to perform a multiple model ensemble (MME).

As described above, it will be understood that the present invention is implemented in a modified form without departing from the essential characteristics of the present invention.

It is therefore to be understood that the specified embodiments are to be considered in an illustrative rather than a restrictive sense and that the scope of the invention is indicated by the appended claims rather than by the foregoing description and that all such differences falling within the scope of equivalents thereof are intended to be embraced therein It should be interpreted.

10. Systematic Error Correction Section 20. Best Model Selection
30. Prediction Variable Width Correction Unit 40. The probability prediction unit

Claims (11)

A systematic error correction unit for correcting systematic errors of individual models using a multivariate sequential pattern projection technique (MSPM);
A model selection unit for selecting a model based on the prediction variable prediction power of the individual model corrected through the multivariate sequential pattern projection technique (MSPM);
A predicted fluctuation width correcting unit for correcting the predicted fluctuation width of the individual model selected by the model selecting unit based on the variation of the observation by a statistical method;
And a probability predicting unit for performing a probability prediction by using a multiple model ensemble (MME) for multiple models having the systematic error correction and the post-model predicted amplitude correction. 2. The apparatus of claim 1, wherein the systematic error correction unit corrects systematic errors of a model used for prediction,
A unit lattice division unit for dividing the global region of the global circulation model into unit lattices,
An association region estimating unit for estimating a correlation region having a high correlation between the predictor and the predictor corresponding to each unit grid divided by the unit grid dividing unit;
A correlation extracting unit for statistically analyzing the time series volatility of the predictive parameter and the pattern variability of the predictive factor,
And a prediction parameter correcting unit for correcting a prediction parameter corresponding to a unit cell of a solution to be actually predicted through association information extracted by the association extracting unit in the combining pattern. 2. The method of claim 1, wherein the model selection unit selects the model based on the prediction variable prediction power of the individual model corrected through the MSPM algorithm,
If the predictive power of the corrected predictive variable of past data existing in all models participating in the multiple model ensemble (MME) is greater than the reference value,
And if it is below the reference value, the predictive power is considered to be low and excluded when performing the MME. The apparatus according to claim 1,
And multiplying the corrected predicted value per unit lattice of the individual model of the solution to be actually predicted by the ratio of the standard deviation of the observed data to the standard deviation of the corrected predicted value through the 23-year MSPM algorithm to correct the predicted fluctuation. Model Probability Prediction System. A method for estimating probability by processing statistical forecasting data on the global area of the global circulation model,
Correcting systematic errors of individual models with a multivariate sequential pattern projection technique (MSPM);
Selecting a model based on predictive predictive power of individual models corrected through the multivariate sequential pattern projection technique (MSPM);
Correcting the predicted fluctuation width of the selected individual model on the basis of the fluctuation width of the observation by a statistical method;
And performing probability estimation by using a multi-model ensemble (MME) for multiple models having the systematic error correction and the post-model prediction post-prediction correction. [6] The method of claim 5, wherein the probability prediction information by the probability prediction step is a probability value with respect to a normal level, a normal level, and a lower level than a normal level. 6. The method of claim 5, wherein correcting systematic errors of the individual models comprises:
A process of dividing the global area of the global circulation model into unit lattices,
The process of estimating the region of high correlation between the predictors and predictors corresponding to each unit grid based on the historical data;
A step of statistically analyzing the time series volatility of the predictive variables and the pattern variability of the predictive factors,
And correcting a prediction parameter corresponding to a unit lattice of a solution to be actually predicted through information extracted from the combined pattern. The method of claim 5, wherein the historical data of the global circular model used for extracting the correlation uses a period (1981-2003) that is common to all models participating in the MME.
Wherein the predictor includes a 500 hPa altitude, a temperature of 850 hPa, and a sea surface pressure, and the predicted parameters calibrated from each predictor are finally averaged. 6. The method of claim 5, wherein the step of selecting a model based on the predictive variable prediction power of the corrected individual model comprises:
If the predictive power of the corrected predictive variable of past data existing in all models participating in the multiple model ensemble (MME) is greater than the reference value,
If it is below the reference value, the predictive power is considered to be low and excluded at the time of performing the MME,
Wherein the reference value for model selection is set to a correlation coefficient of 0.3 between the predicted variable and the time series volatility of the observed data. 6. The method of claim 5, wherein the step of correcting the predicted fluctuation width of the individual model based on the variation of the observation by a statistical method comprises:
And multiplying the corrected predicted value per unit lattice of the individual model of the solution to be actually predicted by the ratio of the standard deviation of the observed data to the standard deviation of the corrected predicted value through the 23-year MSPM algorithm to correct the predicted fluctuation. Model Probability Prediction Method. 6. The method of claim 5, wherein performing the probability estimation using the multiple model ensemble (MME)
Average of observed climate data (

) And variance (

), Estimating the probability distribution of climatic data,
Estimating reference points a and b that define three categories from the probability distribution, which are higher than normal, higher than normal, and lower than normal,
A multi-model comprising the step of calculating the probability corresponding to each category according to the two reference values determined from the probability distribution of climate data using the probability distribution function of the predicted data, which is selected as a model with high predictive power, and as a normal probability density function. Probability prediction method.

Images