CN115049013A - Short-term rainfall early warning model fusion method combining linearity and SVM - Google Patents
Short-term rainfall early warning model fusion method combining linearity and SVM Download PDFInfo
- Publication number
- CN115049013A CN115049013A CN202210791058.7A CN202210791058A CN115049013A CN 115049013 A CN115049013 A CN 115049013A CN 202210791058 A CN202210791058 A CN 202210791058A CN 115049013 A CN115049013 A CN 115049013A
- Authority
- CN
- China
- Prior art keywords
- early warning
- rainfall
- svm
- short
- rainfall early
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Operations Research (AREA)
- Human Resources & Organizations (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Evolutionary Biology (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Algebra (AREA)
- Development Economics (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a short-term rainfall early warning model fusion method combining linearity and SVM, which comprises the following steps: acquiring the total zenith troposphere delay under the same time-space condition by using a ground-based Beidou station, acquiring multiple meteorological parameters, and performing inversion according to the total zenith troposphere delay and the multiple meteorological parameters to obtain the atmospheric degradable water; determining a forecasting factor according to the atmospheric degradable water yield; acquiring a forecasting factor threshold according to rainfall data and a forecasting factor; constructing a first early warning model according to a forecast factor threshold; constructing a second early warning model based on the SVM according to rainfall data, multiple meteorological parameters and atmospheric water-reducing capacity; utilizing the first early warning model and the second early warning model to predict rainfall, and respectively obtaining a first rainfall early warning time sequence and a second rainfall early warning time sequence; and fusing the first rainfall early warning time sequence and the second rainfall early warning time sequence to obtain a third rainfall early warning time sequence. The invention realizes the short rainfall early warning with higher precision by fusing two models, namely the linear model and the SVM model.
Description
Technical Field
The invention relates to a short-time rainfall early warning model fusion method combining linearity and SVM, and belongs to the field of GNSS meteorology and machine learning.
Background
The short-time heavy rainfall can cause disasters such as urban waterlogging, casualties, economic loss and the like, and the short-time heavy rainfall event becomes one of the most damaging weather phenomena in the world. The frequency of extreme rainfall events may show an increasing trend in the context of global warming. Therefore, the method has important significance and economic value for disaster prevention and reduction by accurately, timely and reliably predicting rainfall, especially short-time heavy rainfall.
The water vapor product obtained by the traditional meteorological detection technology cannot capture the high-time-space water vapor transmission and evolution process thereof, and the short-time weather phenomenon cannot be accurately predicted due to the loss of data information. With the development of GNSS (Global Navigation Satellite System) technology, the water vapor product obtained based on the technology has the advantages of high spatial and temporal resolution, all weather, low cost and high precision. With the completion of global deployment of the Beidou system in China, the water vapor product obtained based on the Beidou system can play a greater role and application value in GNSS meteorology. How to fully utilize the water vapor products inverted by the Beidou system and exert the important efficiency of the water vapor products becomes one of the research focuses of the invention.
In recent years, a short-term rainfall early warning model is built based on GNSS PWV (global navigation satellite system) to gradually attract attention of partial scholars and develop to a certain extent, and the model fits the PWV accumulation growth trend before rainfall occurs based on a least square algorithm so as to realize early warning of rainfall events. However, the model has the defects of less rainfall prediction factors and lower precision.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides a short-time rainfall early warning model fusion method combining linearity and an SVM (Support Vector Machine), a Beidou satellite system receiver is used for inverting the high-precision PWV content in the zenith direction by combining multiple meteorological parameters to respectively construct a linear short-time rainfall early warning model and an SVM short-time rainfall early warning model, and the high-precision short-time rainfall early warning is realized by fusing the two models.
The invention provides a short-term rainfall early warning model fusion method combining linearity and SVM, which comprises the following steps:
acquiring total zenith troposphere delay under the same time-space condition by using a ground-based Beidou station, acquiring multiple meteorological parameters, and performing inversion according to the total zenith troposphere delay and the multiple meteorological parameters to obtain the atmospheric water reducible quantity;
step two, determining a forecasting factor according to the atmospheric degradable water obtained in the step one;
step three, acquiring a forecasting factor threshold according to rainfall data and the forecasting factor determined in the step two;
step four, constructing a first early warning model according to the forecasting factor threshold obtained in the step three;
step five, constructing a second early warning model based on the SVM according to rainfall data, multiple meteorological parameters and atmospheric water-reducing capacity;
step six, performing rainfall prediction by using the first early warning model obtained in the step four and the second early warning model obtained in the step five to respectively obtain a first rainfall early warning time sequence and a second rainfall early warning time sequence;
and step seven, fusing the first rainfall early warning time sequence and the second rainfall early warning time sequence obtained in the step six to obtain a third rainfall early warning time sequence.
Further, in the first step, the method for acquiring the multiple meteorological parameters interpolates the multiple meteorological parameters provided by the reanalysis data set to the Beidou website.
Further, the method for interpolating the multiple meteorological parameters provided by the reanalysis data set to the Beidou station is bilinear interpolation.
Further, in the second step, the forecast factor includes an atmospheric precipitation magnitude value.
Further, the forecast factor may further include an atmospheric degradable water amount change amount and an atmospheric degradable water amount change rate, which are calculated according to the atmospheric degradable water amount.
Further, in the third step, before the forecast factor threshold is obtained, fitting is performed on the atmospheric degradable water content time sequence.
And further, fitting the atmospheric water reducible quantity time sequence based on a least square algorithm.
Further, the method for obtaining the threshold value of the forecasting factor is a percentile method.
Further, the optimal percentile threshold value of the percentile method is determined according to the highest rainfall early warning accuracy rate and the lowest false alarm rate.
Further, in the fourth step, the constructed first early warning model is: and when the forecasting factor exceeds the forecasting factor threshold value, forecasting the rainfall early warning moment.
Further, in the seventh step, the method for fusing the first rainfall early warning time sequence and the second rainfall early warning time sequence includes: and keeping the same rainfall early warning time, and deleting different rainfall early warning times.
The invention has the beneficial effects that: establishing a short-term rainfall early warning model fusing the traditional linearity and the SVM by combining the advantages of the single models of the linearity and the SVM; the Beidou system is fully utilized to obtain high-precision water vapor products, and the universality and the reliability of the fusion rainfall early warning model are ensured; and multiple meteorological parameters are provided by combining the ERA5 data set, and interpolation is performed to the Beidou site, so that important meteorological data support is provided for the inversion of PWV based on the Beidou technology and the construction of a short-term rainfall early warning model based on an SVM algorithm.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings.
Fig. 1 is a schematic flow chart of the present invention, and specifically, the present invention provides a short-term rainfall early warning model fusion method combining linearity and SVM, which includes the following steps:
acquiring total zenith troposphere delay under the same time-space condition by using a ground-based Beidou station, acquiring multiple meteorological parameters, and performing inversion according to the total zenith troposphere delay and the multiple meteorological parameters to obtain the atmospheric water reducible quantity;
the method comprises the following steps that PWV (water volume) is an important parameter for building a rainfall early warning model, and the high-precision PWV content in the zenith direction of the receiver is inverted by the Beidou satellite system receiver in a research area. The method comprises the following steps of resolving a Beidou receiver real-time observation value by utilizing GAMITT or BERNASE software to obtain zenith troposphere total delay ZDD (Zenith troposphere delay) under the same space-time condition, wherein the ZTD mainly comprises zenith dry delay ZDD (Zenith hydro delay) and zenith Wet delay ZWD (Zenith Wet delay), wherein ZHD is calculated by a Saastamonen model:
in the formula, P sr The surface pressure (hPa), omega and H eg Respectively, the latitude (°) and elevation (m) of the survey station.
ZWD is obtained by subtracting ZHD from ZTD, and PWV is calculated as follows:
in the formula I w Refers to the liquid water density (g/m) 3 ),R t 、c 2 And c 3 Is a constant term, T am Is the atmospheric weighted mean temperature, T am From the surface temperature T s And calculating to obtain: t is am =70.2+0.72·T s 。
The multi-meteorological parameters required in the above formula are provided by reanalysis data sets, and ERA5 is the ECMWF fifth generation reanalysis data set, and can provide multi-meteorological parameters (such as PWV, T, P and rainfall data, etc.) with different spatial resolutions. Considering the limitation of the time length of the measured data, the GNSS station and the ERA5 grid point position are not overlapped in general, so that multiple meteorological parameters provided by ERA5 need to be interpolated to the beidou station, specifically, the interpolation method is bilinear interpolation. In one embodiment of the present invention, the accuracy of the meteorological data interpolated to the site by ERA5 is comparable to that provided by conventional observation methods.
Step two, determining a forecasting factor according to the atmospheric water-reducing capacity obtained in the step one;
in one embodiment of the present invention, the predictor includes a PWV value, and may further include a PWV change amount and a PWV change rate, which are calculated according to the PWV value. Firstly, fitting the PWV time sequence based on a least square algorithm:
where f (PWV) represents a fitting function of PWV values, x 1 ~x n Representing the time of day, n the size of the fitting window, η the number of fits, ω 1 ~ω n Represents the fitting coefficient, ω 0 Is a constant. In order to realize seamless fitting epoch-by-epoch PWV, a sliding step length is introduced to move a fitting window to slide backwards in sequence.
The calculation formula of the PWV variation amount and the PWV variation rate is as follows:
PWV Change amount f (PWV) max -f(PWV) min #(4)
In the formula, f (PWV) max And f (PWV) min Respectively referring to the maximum and minimum PWV values, T, within the fitting window max And T min Respectively, the time corresponding to the maximum and minimum values of PWV.
Step three, acquiring a forecasting factor threshold according to rainfall data and the forecasting factor determined in the step two;
according to the method, rainfall data and PWV data obtained by inversion of Beidou sites are divided according to four seasons of spring, summer, autumn and winter, and considering that a percentile method can automatically determine variable values corresponding to different percentiles in an unknown data set. Different percentiles (10,15,20,25 and 30) are tried, and the optimal percentile threshold is determined according to the highest rainfall early warning accuracy rate and the lowest false alarm rate.
The calculation formula of the rainfall early warning accuracy TDR is as follows:
in the formula, N tr Number of correct rainfall predictions, N total Refers to the actual number of rains.
The rainfall early warning false alarm rate FFR calculation formula is as follows:
in the formula, N ff Number of erroneous rainfall predictions, M total Which refers to the number of rainfall forecasts.
Step four, constructing a first early warning model according to the forecasting factor threshold obtained in the step three;
when any one of the PWV value, the PWV variation amount and the PWV variation rate exceeds a corresponding threshold value, the rainfall early warning moment is predicted.
Step five, constructing a second early warning model based on the SVM according to rainfall data, multiple meteorological parameters and atmospheric water-reducing capacity;
the SVM-based second early warning model relates to an internal and external coincidence experiment, the internal coincidence experiment is mainly used for modeling and checking the precision of the model, the external coincidence experiment is used for evaluating unknown data to drive the model to obtain the precision of the predicted rainfall, the internal and external coincidence experiment input parameters are the air pressure, the temperature, the rainfall and PWV data in the previous hour, and the output parameter is the rainfall in the next hour.
The SVM model mainly comprises the following two parts:
(1) the method comprises the following steps of (1) optimizing a regression function, wherein the function of the SVM is to construct a hyperplane of the regression function, and fit a target vector variation trend with minimum error:
in the formula, w and t are parameters, and θ (n) refers to a nonlinear function.
Defining a loss function E from the insensitive loss function epsilon ε :
In the formula, n i Finger input variable, n j Refers to the sample set output variable.
SVM is translated into the following optimization problem:
in the formula, xi i Andis a relaxation factor, meaning subject to an error threshold epsilonTraining an upper bound and a lower bound of error; and C is a penalty factor and refers to the error correction degree under the condition of misclassification.
(2) Solving the optimization problem, namely solving the optimization problem by introducing a Lagrange multiplier, wherein a regression function is as follows:
non-zero Lagrange multiplierThe corresponding input variables are support vectors, and the regression function can be further written as:
in the formula, n k Finger support vector, K (n) k And n) refers to the kernel function, which the present invention selects as the kernel function the most commonly used gaussian radial basis function.
The parameter threshold of the SVM model is determined by using a cross validation and grid search method.
Step six, carrying out rainfall prediction by using the first early warning model obtained in the step four and the second early warning model obtained in the step five, and respectively obtaining a first rainfall early warning time sequence and a second rainfall early warning time sequence;
and step seven, fusing the first rainfall early warning time sequence and the second rainfall early warning time sequence obtained in the step six to obtain a third rainfall early warning time sequence.
The fusion method comprises the following steps: the same rainfall early warning time obtained by the two rainfall early warning models is reserved, and different rainfall early warning times are deleted to obtain a final rainfall early warning time sequence.
And comparing each rainfall early warning time sequence with the actual rainfall occurrence time, counting the correct rainfall early warning times and the wrong rainfall early warning times, and calculating the rainfall early warning correct rate and the wrong rainfall early warning rate of the fusion model. The rainfall early warning accuracy and the false reporting rate of the three rainfall early warning models are compared, and the short-time rainfall early warning model combining linearity and SVM provided by the invention is proved to have optimal precision.
Claims (11)
1. A short-time rainfall early warning model fusion method combining linearity and SVM is characterized by comprising the following steps:
acquiring total zenith troposphere delay under the same time-space condition by using a ground-based Beidou station, acquiring multiple meteorological parameters, and performing inversion according to the total zenith troposphere delay and the multiple meteorological parameters to obtain the atmospheric water reducible quantity;
step two, determining a forecasting factor according to the atmospheric degradable water obtained in the step one;
step three, acquiring a forecasting factor threshold according to rainfall data and the forecasting factor determined in the step two;
step four, constructing a first early warning model according to the forecasting factor threshold obtained in the step three;
step five, constructing a second early warning model based on the SVM according to rainfall data, multiple meteorological parameters and atmospheric water-reducing capacity;
step six, carrying out rainfall prediction by using the first early warning model obtained in the step four and the second early warning model obtained in the step five, and respectively obtaining a first rainfall early warning time sequence and a second rainfall early warning time sequence;
and step seven, fusing the first rainfall early warning time sequence and the second rainfall early warning time sequence obtained in the step six to obtain a third rainfall early warning time sequence.
2. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 1, wherein in the first step, the method for obtaining multi-meteorological parameters interpolates the multi-meteorological parameters provided by the reanalysis data set to a Beidou site.
3. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 2, wherein the method of interpolating the Beidou site with multiple meteorological parameters provided by the reanalysis dataset is bilinear interpolation.
4. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 1, wherein in the second step, the forecasting factor comprises an atmospheric precipitation magnitude.
5. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 4, wherein the forecasting factors further include an amount of atmospheric reducible water change and an amount of atmospheric reducible water change rate, and the amount of atmospheric reducible water change rate are calculated from the amount of atmospheric reducible water.
6. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 1, wherein in the third step, the atmospheric water reducible volume time sequence is fitted before the forecasting factor threshold is obtained.
7. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 6, wherein the atmospheric degradable water content time sequence is fitted based on a least squares algorithm.
8. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 1, wherein in the third step, the method for obtaining the forecasting factor threshold is a percentile method.
9. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 8, wherein the optimal percentile threshold of the percentile method is determined according to the highest rainfall early warning accuracy rate and the lowest false alarm rate.
10. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 1, wherein in the fourth step, the constructed first early warning model is: and when the forecasting factor exceeds the forecasting factor threshold, forecasting the rainfall early warning moment.
11. The short-term rainfall early warning model fusion method combining linearity and SVM of claim 1, wherein in the seventh step, the method for fusing the first rainfall early warning time sequence and the second rainfall early warning time sequence comprises: and keeping the same rainfall early warning time, and deleting different rainfall early warning times.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210791058.7A CN115049013A (en) | 2022-07-05 | 2022-07-05 | Short-term rainfall early warning model fusion method combining linearity and SVM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210791058.7A CN115049013A (en) | 2022-07-05 | 2022-07-05 | Short-term rainfall early warning model fusion method combining linearity and SVM |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115049013A true CN115049013A (en) | 2022-09-13 |
Family
ID=83165899
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210791058.7A Pending CN115049013A (en) | 2022-07-05 | 2022-07-05 | Short-term rainfall early warning model fusion method combining linearity and SVM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115049013A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115356789A (en) * | 2022-10-08 | 2022-11-18 | 南京气象科技创新研究院 | Plum rain period short-time strong precipitation grading early warning method |
CN117669793A (en) * | 2023-10-20 | 2024-03-08 | 广东省气象台(南海海洋气象预报中心、珠江流域气象台) | Rainfall frequency estimation method and device for combined satellite and site data |
CN117669793B (en) * | 2023-10-20 | 2024-06-04 | 广东省气象台(南海海洋气象预报中心、珠江流域气象台) | Rainfall frequency estimation method and device for combined satellite and site data |
-
2022
- 2022-07-05 CN CN202210791058.7A patent/CN115049013A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115356789A (en) * | 2022-10-08 | 2022-11-18 | 南京气象科技创新研究院 | Plum rain period short-time strong precipitation grading early warning method |
CN117669793A (en) * | 2023-10-20 | 2024-03-08 | 广东省气象台(南海海洋气象预报中心、珠江流域气象台) | Rainfall frequency estimation method and device for combined satellite and site data |
CN117669793B (en) * | 2023-10-20 | 2024-06-04 | 广东省气象台(南海海洋气象预报中心、珠江流域气象台) | Rainfall frequency estimation method and device for combined satellite and site data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20220043182A1 (en) | Spatial autocorrelation machine learning-based downscaling method and system of satellite precipitation data | |
CN112905560B (en) | Air pollution prediction method based on multi-source time-space big data deep fusion | |
CN107703564B (en) | Rainfall prediction method and system and electronic equipment | |
CN107316095B (en) | Regional weather drought level prediction method coupled with multi-source data | |
CN111401602B (en) | Assimilation method for satellite and ground rainfall measurement values based on neural network | |
CN110232471B (en) | Rainfall sensor network node layout optimization method and device | |
CN112016052B (en) | Near-surface daily maximum air temperature estimation method, system and terminal based on multi-source data | |
Hong et al. | Basis for recommending an update of wind velocity pressures in Canadian design codes | |
CN113344291B (en) | Urban inland inundation range forecasting method, device, medium and equipment | |
CN112232554A (en) | Construction method of local short-term rainfall forecast model based on BP neural network | |
CN114611778B (en) | Reservoir water level early warning method and system based on warehousing flow | |
CN115563540A (en) | Urban waterlogging prediction method combining multi-source data and deep learning and application thereof | |
CN113704693A (en) | High-precision effective wave height data estimation method | |
CN113281754A (en) | WRF-Hydro key parameter calibration method for quantitatively estimating rainfall by integrating rainfall station with radar | |
CN115049013A (en) | Short-term rainfall early warning model fusion method combining linearity and SVM | |
CN117233869A (en) | Site short-term wind speed prediction method based on GRU-BiTCN | |
CN117277304A (en) | Photovoltaic power generation ultra-short-term power prediction method and system considering sunrise and sunset time | |
CN115236770B (en) | Nonlinear short-time adjacent precipitation prediction method based on space-time stacking and sample reconstruction | |
CN116609859A (en) | Weather disaster high-resolution regional mode forecasting system and method | |
CN116186493A (en) | Meteorological forecasting method, device, equipment and medium based on numerical mode correction | |
Donnelly et al. | Short-term forecasting of nitrogen dioxide (NO 2) levels using a hybrid statistical and air mass history modelling approach | |
CN114529035A (en) | CART-based wind speed forecasting method of multi-mode integrated model | |
Vishwakarma et al. | Assessment of CMIP5 and CORDEX-SA experiments in representing multiscale temperature climatology over central India | |
Ozkaya et al. | Analyzing radar rainfall estimate errors with three vector norms: application to weather radar rainfall data in Muğla, Turkey | |
Aly et al. | Evaluation of improvised spatial interpolation methods for infilling missing precipitation records |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |