CN106405682B - A kind of precipitation predicting method and device - Google Patents
A kind of precipitation predicting method and device Download PDFInfo
- Publication number
- CN106405682B CN106405682B CN201610752369.7A CN201610752369A CN106405682B CN 106405682 B CN106405682 B CN 106405682B CN 201610752369 A CN201610752369 A CN 201610752369A CN 106405682 B CN106405682 B CN 106405682B
- Authority
- CN
- China
- Prior art keywords
- prediction
- rainfall
- time
- model
- precipitation
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
Landscapes
- Environmental & Geological Engineering (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to rainfall forecast technical field, in particular to a kind of precipitation predicting method and device.The precipitation predicting method includes: step a: setting time scale interval and prediction lag time, and establishes Multiple Time Scales SVM rainfall dislocation prediction model based on the corresponding meteorological data in time scale interval；Step b: computation model parameter gamma value and parameter C value, input SVM rainfall dislocation prediction model obtain prediction lag time corresponding prediction of precipitation result；Step c: rainfall probability is calculated according to the prediction of precipitation result.Compared with the existing technology, the embodiment of the present invention can preferably reflect the following rainfall trend more refined, have certain practical value, have preferable accuracy rate and the goodness of fit on rain time point.The present invention utilizes most basic meteorological element data, has better scalability, can adapt to the training and prediction of more meteorological element data.
Description
Technical field
The present invention relates to rainfall forecast technical field, in particular to a kind of precipitation predicting method and device.
Background technique
Heavy rainfall is closed in short-term and refers to that short period precipitation intensity is bigger in subrange, and rainfall reaches or surpasses
Cross the common weather phenomenon of a certain standard.Under the background of global climate abnormal increase, in addition urbanization process acceleration is common
Effect, city " urban dry island effect " and " tropical island effect " are more and more obvious, and the frequency and intensity for closing on catchment generation in short-term are deposited
The case where possibility being increasing, short-time strong rainfall in all parts of the country causes Urban Stagnant Floods to cause disaster in recent years, is appeared in the newspapers repeatly to city
Waterlogging problem is increasingly severe, and other than the deficiency of sewerage pipeline network in the planning and designing of city itself, another is critically important
The reason of be exactly city heavy rainfall in short-term occur probability and intensity it is increasing, extreme catchment takes place frequently, raininess record quilt
Constantly refresh, urban waterlogging is all caused and seriously affected to socio-economic development and people's production and living.
It is no matter external or domestic, especially for being located in for the Shenzhen in south China, in short-term rainfall nowcasting there is
Very big uncertainty, there are many factor that cause that the forecast is inaccurate true, and Precipitation Distribution in Time and Space is closed in subtropical zone coastal region in short-term
Local otherness be to cause the one of the major reasons of precipitation forecast error, Shenzhen often occur entire urban area occur it is more
The different weather conditions of kind, if short-time strong rainfall occurs in Luohu District, but but sky is sunny for Nanshan District.With Shenzhen's meteorology number of units
For, meteorological observatory, Shenzhen from starting cloth meteorological disaster Division warning in 2007, from 2012 by when dynamic to update publication each
Subregion weather forecast.According to existing research, the accuracy rate of Shenzhen's precipitation predicting in short-term is generally all relatively low at present, wherein
1 hour accuracy rate is slightly higher, has certain use value, and the 2nd hour backward, the 3rd hour accuracy rate decline it is obvious that making
It is to be improved with being worth.Therefore, what the probability and precipitation peak value that the following a certain moment rainfall in Accurate Prediction ground occurs occurred
Time point seems extremely important.
Traditional nowcasting method is mainly based upon satellite cloud picture and radar return is extrapolated, this is also that comparison is conventional
Nowcasting technology and current weather forecast in Weather radar system and strong weather warning business element.
With the development of Numerical Forecast Technology, various characterization heating power, power physical quantity be introduced into and be applied to and close on rainfall in short-term
In analysis prediction, various precipitation predicting models in short-term are established.But the deficiency of Numerical Forecast Technology is that calculation amount is huge, precision
Depending on initial input condition.Also there is the method using machine learning, but a large amount of sample is needed to be trained.And it is right
For rainfall forecast, training sample is more, and the irrelevant information of introducing is more, and the effect of forecast is simultaneously bad.In numerical model
In the also jejune situation of nowcasting technology, expert system represents the mainstream hair of current nowcasting business in the world substantially
Exhibition, is implemented using expert system for the diagnosis early warning of strong convective weather and nowcasting during Beijing Olympic Games.
Summary of the invention
The present invention provides a kind of precipitation predicting method and devices, it is intended to solve at least to a certain extent in the prior art
One of above-mentioned technical problem.
To solve the above-mentioned problems, the present invention provides the following technical scheme that
A kind of precipitation predicting method, comprising:
Step a: setting time scale interval and prediction lag time, and based on the corresponding meteorological number in time scale interval
According to establish Multiple Time Scales SVM rainfall dislocation prediction model；
Step b: computation model parameter gamma value and parameter C value, input SVM rainfall dislocation prediction model obtain prediction and prolong
Corresponding prediction of precipitation result of slow time；
Step c: rainfall probability is calculated according to the prediction of precipitation result.
The technical solution that the embodiment of the present invention is taken further include: described to establish SVM rainfall dislocation in advance in the step a
Survey the modeling formula of model are as follows:
Yt-1=R* (Xt-1)
Xt-1=[xt-h-d,xt-h-d+1,…,xt-d-1]T
Yt-1=[yt-h,yt-h+1,…,yt-1]T
Yt=R* (xt-d)
In above-mentioned formula, Xt-1, Yt-1For training sample, R* is the model established according to training sample, and t is to be predicted
Moment, h are the time scale interval of training sample, and d is prediction lag time, yt、xt-dRespectively precipitation predicting value, to be predicted
The input of rainfall.
The technical solution that the embodiment of the present invention is taken further include: the step a further include: in the step b, inputted
Meteorological data include each automatic Weather Station wind speed, temperature, air pressure and humidity.
The technical solution that the embodiment of the present invention is taken further include: in the step b, the model parameter gamma value and
The calculation of parameter C value specifically: parameter gamma value and parameter C value, the evaluation mark of cross validation are determined by cross validation
Standard is that TS scores, and selects TS and scores highest parameter gamma value and parameter C value as final suitable parameters.
The technical solution that the embodiment of the present invention is taken further include: in the step c, the rainfall probability calculation formula
Are as follows:
In above-mentioned formula, ytIt is model prediction of precipitation as a result, working as ytWhen > 0, P (yt) > 50%, conversely, P (yt)≦
50%；That is:
Another technical solution that the embodiment of the present invention is taken are as follows: a kind of precipitation predicting device, including model building module, mould
Type training module and precipitation predicting module；The model building module is for when being arranged time scale interval and prediction lag
Between, and Multiple Time Scales SVM rainfall dislocation prediction model is established based on the corresponding meteorological data in time scale interval；The model
Training module is used for computation model parameter gamma value and parameter C value, and input SVM rainfall dislocation prediction model obtains prediction lag
Time corresponding prediction of precipitation result；The precipitation predicting module is used to calculate rainfall according to the prediction of precipitation result general
Rate.
The technical solution that the embodiment of the present invention is taken further include: the model building module establishes SVM rainfall dislocation prediction
The modeling formula of model are as follows:
Yt-1=R* (Xt-1)
Xt-1=[xt-h-d,xt-h-d+1,…,xt-d-1]T
Yt-1=[yt-h,yt-h+1,…,yt-1]T
Yt=R* (xt-d)
In above-mentioned formula, Xt-1, Yt-1For training sample, R* is the model established according to training sample, and t is to be predicted
Moment, h are the time scale interval of training sample, and d is prediction lag time, yt、xt-dRespectively precipitation predicting value, to be predicted
The input of rainfall.
The technical solution that the embodiment of the present invention is taken further include: the meteorological data that the model training module is inputted includes
Wind speed, temperature, air pressure and the humidity of each automatic Weather Station.
The technical solution that the embodiment of the present invention is taken further include: the model training module computation model parameter gamma value
With the calculation of parameter C value specifically: determine parameter gamma value and parameter C value, the evaluation of cross validation by cross validation
Standard is TS scoring, selects TS and scores highest parameter gamma value and parameter C value as final suitable parameters.
The technical solution that the embodiment of the present invention is taken further include: the rainfall probability calculation formula of the precipitation predicting module
Are as follows:
In above-mentioned formula, ytIt is model prediction of precipitation as a result, working as ytWhen > 0, P (yt) > 50%, conversely, P (yt)≦
50%；That is:
Compared with the existing technology, the beneficial effect that the embodiment of the present invention generates is: the precipitation predicting of the embodiment of the present invention
Method and device determines model parameter by cross validation by establishing SVM rainfall dislocation prediction model using meteorological data
Gamma value and parameter C value, to obtain optimal prediction of precipitation as a result, and calculating finally according to prediction of precipitation result
Rainfall probability.Compared with the existing technology, the embodiment of the present invention can preferably reflect the following rainfall trend more refined, have
Certain practical value has preferable accuracy rate and the goodness of fit on rain time point.The present invention utilizes most basic meteorology
Factor data has better scalability, can adapt to the training and prediction of more meteorological element data.
Detailed description of the invention
Fig. 1 is the flow chart of the precipitation predicting method of the embodiment of the present invention；
Fig. 2 is the SVM rainfall dislocation prediction model schematic diagram of the embodiment of the present invention；
Fig. 3 is the structural schematic diagram of the precipitation predicting device of the embodiment of the present invention；
Fig. 4 is time scale 3 hours SVM prediction rainfall in following 1 hour and true rainfall comparison diagram (" western beautiful " automatic Weather Station
In June, -2014 rainfall in 2 months in 2014)；
Fig. 5 is time scale 3 hours SVM prediction rainfall in following 2 hours and true rainfall comparison diagram (" western beautiful " automatic Weather Station
2 months in June, -2014 of rainy season rainfall in 2014)；
Fig. 6 is time scale 3 hours SVM prediction rainfall in following 1 hour and true rainfall comparison diagram (" western beautiful " automatic Weather Station
From when 6 days 11 April in 2014 in three days)；
Fig. 7 is time scale 3 hours SVM prediction rainfall in following 2 hours and true rainfall comparison diagram (" western beautiful " automatic Weather Station
From when 6 days 11 April in 2014 in three days)；
Fig. 8 is that the whole accuracy rate of the following 1 hour precipitation predicting of time scale SVM rainfall in 3 hours dislocation prediction model is shown
It is intended to；
Fig. 9 is that time scale is 3 hours SVM precipitation predicting TS appraisal result schematic diagrames；
Figure 10 is the SVM precipitation predicting TS appraisal result schematic diagram that the time scale of different websites is 6 hours；
Figure 11 is the SVM precipitation predicting TS appraisal result schematic diagram that time scale is 12 hours；
Figure 12 is that time scale is 24 hours SVM precipitation predicting TS appraisal result schematic diagrames；
Figure 13 is that time scale is 48 hours SVM precipitation predicting TS appraisal result schematic diagrames；
Figure 14 is that time scale is 72 hours SVM precipitation predicting TS appraisal result schematic diagrames；
Figure 15 is different time scales SVM precipitation predicting TS scoring mean value；
Figure 16 is that different time scales SVM precipitation predicting TS standards of grading are poor.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
Referring to Fig. 1, being the flow chart of the precipitation predicting method of the embodiment of the present invention.The precipitation predicting of the embodiment of the present invention
Method the following steps are included:
Step 100: setting time scale interval and prediction lag time, and it is based on the corresponding meteorology in time scale interval
Data establish Multiple Time Scales SVM rainfall dislocation prediction model；
In step 100, the present invention be using current time as boundary, with the meteorological data of certain time before current time come
Trained SVM rainfall dislocation prediction model is done, prediction model is misplaced by the SVM rainfall trained to predict the following certain time
Rainfall.Certain time before current time is " time scale interval ", and the following certain time of prediction is i.e. " when prediction lag
Between ", Multiple Time Scales, which refer to, can be set multiple time scale intervals.Time scale interval and prediction lag time can bases
Practical application is set, and the time scale interval in the embodiment of the present invention is respectively set are as follows: 3h, 6h, 8h, 12h, for 24 hours, 48h,
72h indicates to predict subsequent rainfall with 3h, 6h, 8h, 12h, the automatic Weather Station meteorological data for 24 hours, in 48h, 72h.The prediction lag time
It is respectively set are as follows: 0h, 1h, 2h, 3h, 4h, 5h indicate the rainfall of prediction future 0h, 1h, 2h, 3h, 4h, 5h.
The characteristics of handling small sample for SVM (support vector machines, Support Vector Machine) and advantage are established
SVM rainfall dislocation prediction model handles the meteorological data of multiple automatic Weather Stations, and the input parameter of model includes the wind of each automatic Weather Station
The meteorological datas such as speed, temperature, air pressure, humidity and parameter gamma value and parameter C value etc., in other embodiments of the present invention,
Mode input parameter can also include other meteorological datas.For the rainfall probability prediction at each moment and rainfall time to peak point
Prediction, it is contemplated that rainfall misplace prediction model in time series with the meteorological condition correlation for the previous time period closed on more
Greatly.Although from certain following rainfall data is also contained in the farther historical data of predicted time point, the information also body
Now in the meteorological data closer from predicted time point, therefore is handled and closed on using the Multiple Time Scales SVM of relatively closely spaced
Meteorological data can also reflect the variation of precipitation time series to a certain extent.
Specifically, the modeling pattern of SVM rainfall dislocation prediction model is as follows:
Yt-1=R* (Xt-1) (1)
Xt-1=[xt-h-d,xt-h-d+1,…,xt-d-1]T (2)
Yt-1=[yt-h,yt-h+1,…,yt-1]T (3)
Yt=R* (xt-d) (4)
In above-mentioned formula, Xt-1, Yt-1For training sample, R* is the model established according to training sample, and t is to be predicted
Moment, h are the capacity (time scale interval) of training sample, and d is prediction lag time, yt, xt-dRespectively precipitation predicting value,
The input of rainfall to be predicted.
Step 200: utilizing cross validation computation model parameter gamma value and parameter C value, and by model parameter gamma value
With parameter C value input SVM rainfall dislocation prediction model, prediction lag time corresponding prediction of precipitation result is obtained；
In step 200, in the use process of SVM method, parameter gamma value and parameter C are determined by cross validation
Value, the evaluation criterion of cross validation are that TS scores, and select the TS highest model parameter that scores and are input to for final suitable parameters
In model, to obtain optimal prediction of precipitation result.Specifically as shown in Fig. 2, being the SVM rainfall dislocation of the embodiment of the present invention
Prediction model schematic diagram.In embodiments of the present invention, optimal parameter gamma value are as follows: 0.1, optimal parameter C value are as follows:
1x103。
In modeling, the checkout procedure predicted, the embodiment of the present invention was used in the period of in September, -2015 in January, 2013
Observational data is as sample data, in addition to the sample size of the island MarginZhou website is smaller, single automatic Weather Station website per hour
Sample size >=20000 amount to >=20000 moment, and the sample size of most of websites is sufficiently large, therefore model has centainly
Stability.
Step 300: final rainfall probability is calculated according to prediction of precipitation result；
In step 300, rainfall probability is calculated by following formula:
In formula (5), ytIt is model prediction of precipitation result；It is found that working as ytWhen > 0, P (yt) > 50%, conversely, P
(yt)≤50%.Then final rainfall probability is as follows:
Referring to Fig. 3, being the structural schematic diagram of the precipitation predicting device of the embodiment of the present invention.The rainfall of the embodiment of the present invention
Prediction meanss include model building module, model training module and precipitation predicting module；
Model building module is based on time scale interval pair for time scale interval and prediction lag time to be arranged
The meteorological data answered establishes Multiple Time Scales SVM rainfall dislocation prediction model；Wherein, the present invention is used using current time as boundary
The meteorological data of certain time does trained SVM rainfall dislocation prediction model before current time, passes through the SVM rainfall trained
Prediction model misplace to predict the rainfall of the following certain time.Certain time before current time is i.e. " between time scale
Every ", the following certain time of prediction is " prediction lag time ", Multiple Time Scales refer to multiple time scales can be set between
Every.Time scale interval and prediction lag time can be set according to practical application, the time ruler in the embodiment of the present invention
Degree interval be respectively set are as follows: 3h, 6h, 8h, 12h, for 24 hours, 48h, 72h, indicate with 3h, 6h, 8h, 12h, for 24 hours, in 48h, 72h
Automatic Weather Station meteorological data predicts subsequent rainfall.The prediction lag time is respectively set are as follows: 0h, 1h, 2h, 3h, 4h, 5h indicate prediction
The rainfall of following 0h, 1h, 2h, 3h, 4h, 5h.It is pre- to establish SVM rainfall dislocation for the characteristics of handling small sample for SVM and advantage
Survey the multiple automatic Weather Stations of model treatment meteorological data, the input parameter of model include the wind speed of each automatic Weather Station, temperature, air pressure,
The meteorological datas such as humidity and parameter gamma value and parameter C value etc., in other embodiments of the present invention, mode input parameter is also
It may include other meteorological datas.For the rainfall probability prediction at each moment and rainfall time to peak point prediction, it is contemplated that drop
Rain dislocation prediction model is bigger with the meteorological condition correlation for the previous time period closed in time series.Although when from prediction
Between put and also contain certain following rainfall data in farther historical data, but the information has been also embodied in from predicted time
In the closer meteorological data of point, therefore closing on meteorological data using the Multiple Time Scales SVM of relatively closely spaced processing also can be one
Determine the variation for reflecting precipitation time series in degree.
Specifically, the modeling formula of SVM rainfall dislocation prediction model is as follows:
Yt-1=R* (Xt-1) (1)
Xt-1=[xt-h-d,xt-h-d+1,…,xt-d-1]T (2)
Yt-1=[yt-h,yt-h+1,…,yt-1]T (3)
Yt=R* (xt-d) (4)
In above-mentioned formula, Xt-1, Yt-1For training sample, R* is the model established according to training sample, and t is to be predicted
Moment, h are the capacity (time scale interval) of training sample, and d is prediction lag time, yt, xt-dRespectively precipitation predicting value,
The input of rainfall to be predicted.
Model training module is used to utilize cross validation computation model parameter gamma value and parameter C value, and by model parameter
Gamma value and parameter C value input SVM rainfall dislocation prediction model, obtain prediction lag time corresponding prediction of precipitation result；
Wherein, in the use process of SVM method, parameter gamma value and parameter C value are determined by cross validation, cross validation is commented
Price card standard be TS scoring, select TS score highest model parameter be final suitable parameters be input in model, to obtain
Optimal prediction of precipitation result.In embodiments of the present invention, optimal parameter gamma value are as follows: 0.1, optimal parameter C value
Are as follows: 1x103。
In modeling, the checkout procedure predicted, the embodiment of the present invention was used in the period of in September, -2015 in January, 2013
Observational data is as sample data, in addition to the sample size of the island MarginZhou website is smaller, single automatic Weather Station website per hour
Sample size >=20000 amount to >=20000 moment, and the sample size of most of websites is sufficiently large, therefore model has centainly
Stability.
Precipitation predicting module is for calculating final rainfall probability according to prediction of precipitation result；Wherein, by following public
Formula calculates rainfall probability:
In formula (5), ytIt is model prediction of precipitation result；It is found that working as ytWhen > 0, P (yt) > 50%, conversely, P
(yt)≤50%.Then final rainfall probability is as follows:
It is of the invention practical in order to prove, pass through the history meteorological data of the whole 44 Meteorological Automatic Station points in Shenzhen
It tests to above embodiments.According to weather forecast Short-and-medium Term Weather Forecast quality inspection method, there are two types of precipitation inspection parties
Method, whether there is or not the inspection of precipitation and for the inspection of certain magnitude precipitation, using whether there is or not the inspection method of precipitation in the embodiment of the present invention,
Precipitation forecast inspection-classification is shown in Table 1.
1 precipitation forecast inspection-classification table of table
For the rainfall inspection of each automatic Weather Station and quality of forecast, TS methods of marking is commonly used, the calculation formula of TS scoring is such as
Under:
In above-mentioned formula, NAkTo forecast (secondary) number in correct station, NBkCall out the stops for sky (- secondary) number, NCkTo fail to report station (secondary)
Number, is shown in Table 1, TSkIt scores for the TS of numerical forecast, FOM is rate of failing to report, and FAR is empty report rate.
Precipitation probability accuracy rate compares:
It is 3 hours SVM precipitation predictings and true rainfall that " western beautiful " automatic Weather Station time scale is had chosen in the embodiment of the present invention
Comparison diagram, as shown in Figure 4, Figure 5, Figure 6 and Figure 7, Fig. 4 are time scale 3 hours SVM prediction rainfall in following 1 hour and true drop
Rain comparison diagram (in June, -2014 rainfall in 2 months in 2014 of " western beautiful " automatic Weather Station), Fig. 5 are that 3 hours SVM of time scale are 2 hours following
Predict rainfall and true rainfall comparison diagram (2 months in June, -2014 of the rainy season rainfall in 2014 of " western beautiful " automatic Weather Station), Fig. 6 is time ruler
Spend 3 hours SVM prediction rainfall in following 1 hour and true rainfall comparison diagram (" western beautiful " automatic Weather Station 11 Shi Qisan on April 6th, 2014
In it), Fig. 7 is time scale 3 hours SVM prediction rainfall in following 2 hours and true rainfall comparison diagram (" western beautiful " automatic Weather Station 2014
From when on April 6,11 in three days).From the point of view of the precipitation time series of " western beautiful " automatic Weather Station in January, 2013 in September, -2015, drop
Rain mainly concentrates annual spring, summer, and within this period, whether the density and intensity of rainfall are all bigger, for dropping
The prediction of rain and rain time point, the SVM rainfall dislocation prediction model that time scale is 3 hours all give preferable prediction knot
Fruit.Especially meteorological data of the prediction result based on automatic Weather Station, and the geographical distribution position of each automatic Weather Station is different, due to depth
The climate characteristic of ditch between fields, the meteorological condition of each automatic Weather Station present position and rain fall are and conventional there are apparent difference
Weather forecast is often directed to the Changes in weather (such as rainfall) within the scope of the whole city of Shenzhen, it is difficult to accomplish small scale, fining
Rainfall forecast, therefore the prediction result of the embodiment of the present invention can preferably reflect the following rainfall trend more refined, for certainly
Social and economic activities in dynamic station peripheral extent provide meteorological support, have certain practical value.And from Fig. 4, Fig. 5, Fig. 6
With Fig. 7 as can be seen that whether 1 hour future of prediction, or following 2 hours precipitation predictings, SVM rainfall dislocation prediction model
Preferable accuracy rate and the goodness of fit are shown on rain time point.
Fig. 8 is that the whole accuracy rate of the following 1 hour precipitation predicting of time scale SVM rainfall in 3 hours dislocation prediction model is shown
It is intended to.It, which is calculated, is based on entire sample space, indicates that each website is predicted accurate rainfall and predicted accurately total without the rainfall moment
Several ratios with sample size, it can be clearly seen that the whole predictablity rate of each website is all relatively good, in addition to " seamount " website,
Whether the prediction at current time, or following 1 hour to following 5 hours precipitation predicting, the whole prediction of most of websites
Accuracy rate is all 90% or more, and the accuracy rate of different prediction times is not much different, and the whole prediction rate of " stone dragon is young " website exists
94% or more.Wherein the whole prediction effect of " seamount " website is poor.According to analysis, the quality of data of part website is inadequate
Good, this has influenced the prediction result of model.Therefore, the good quality of data is to improve the good guarantee of model prediction accuracy rate.
Prediction of Precipitation result TS is examined
Table 2 gives the TS score value for the prediction result that time scale used by the embodiment of the present invention is 3 hours, from
In January, 2013 in September, -2015 period in, temporal resolution 1h, for meteorological condition hourly (wind speed, temperature,
Humidity, air pressure) modeling progress precipitation predicting.Require for different forecast moment (it is current, 1 hour, 2 hours, 3 hours, it is 4 small
When, 5 hours), calculate the TS score value of prediction result.
The rainfall probability that closes in short-term that 2 time scale of table is 3 hours predicts TS score value
In general, the TS scoring of each website is relatively high, prediction effect is good, and the prediction result of different websites is not yet
Equally, the rainfall of possible specific position website is influenced also different by other factors, and is not difficult to find out, is prolonged after predicted time
When, TS scoring decreases.
The TS score value of each prediction period of all websites is as shown in figure 9, it is that SVM rainfall in 3 hours is pre- that Fig. 9, which is time scale,
Survey TS appraisal result schematic diagram.In fig. 9 it can be seen that the time scale of different websites is the dislocation prediction mould of SVM rainfall in 3 hours
The TS appraisal result of type, hence it is evident that in the TS scoring of each website, according to preceding 3 hours meteorological datas to current time precipitation predicting
TS scoring highest, that is, rainfall accuracy rate highest.This also illustrates the time point closer with preceding 3 hours meteorological datas,
Rainfall and first 3 hours meteorological condition correlations are bigger.In 44 Meteorological Automatic Stations, the TS scoring of " agriculture garden " automatic Weather Station is most
Height, the precipitation predicting TS scoring at current time is close to 50%, and the TS scoring of " Shenzhen Airport south " automatic Weather Station is minimum, current time
Precipitation predicting also has 40% or more.With prolonging after predicted time point, TS scoring constantly reduces, that is, precipitation predicting in short-term
Accuracy rate is declined, as seen from the figure, in the SVM precipitation predicting that time scale is 3 hours, even if predicted time point is prolonged after being
5 hours, the prediction result that TS scoring also can achieve 35% or more, SVM rainfall dislocation prediction model did not occurred TS scoring
The case where (predictablity rate) reduces rapidly, i.e. model prolong the rainfall within 5 hours for after also has certain predictive value.
Meanwhile in Fig. 9 it can be seen that other than the precipitation predicting at current time, the precipitation predicting TS scoring at other moment connects very much
Closely, difference and little, therefore illustrate that the SVM rainfall dislocation prediction model of the embodiment of the present invention has certain stability.
Figure 10 is the SVM precipitation predicting TS appraisal result schematic diagram that the time scale of different websites is 6 hours, with Fig. 9 into
Row compares, it can clearly be seen that TS scoring is small compared to 3 when being divided into 6 hours between the time scale of meteorological data training sample is
When model prediction declined, even to the precipitation predicting at current time, TS scoring also only have 40% or so.And other are pre-
The TS scoring for surveying the moment is gradually decreased with prolonging after the time, and following 5 hours precipitation predicting TS score minimum, most of websites
The TS scoring of following 5 hours precipitation predictings all can only achieve 30% or so.
Figure 11 is the SVM precipitation predicting TS appraisal result schematic diagram that time scale is 12 hours.Occur compared to Fig. 9 and Figure 10
Decline by a relatively large margin, the TS fall that scores is about 8%, it can be seen that even to the precipitation predicting at current time, when
Between scale be 12 hours SVM TS scoring also there was only 33% or so.And the precipitation predicting at other following moment is then shown
Lower TS scoring, predictablity rate decline are obvious.Equally, it is SVM precipitation predicting TS scoring in 24 hours that Figure 12, which is time scale,
It is 48 hours SVM precipitation predicting TS appraisal result schematic diagrames, Figure 14 is that time scale is that result schematic diagram, Figure 13, which are time scale,
72 hours SVM precipitation predicting TS appraisal result schematic diagrames.It can be found that with the increase at time scale interval, precipitation predicting TS
Scoring constantly reduces, and the accuracy rate of prediction also gradually reduces, when time scale is 72 hours, the precipitation predicting TS at current time
Scoring only 20% or so, well below time scale be 3 hours SVM precipitation predictings TS appraisal result, and other when
The precipitation predicting TS scoring at quarter also only has 15% or so.The horizontal precipitation predicting accuracy rate of TS scoring is relatively low, short
When precipitation predicting in do not had practical application value.
The characteristics of can be seen that according to the TS appraisal result of Fig. 9-Figure 14 for SVM sample present treatment, between time scale
Every bigger, that is, the capacity of training sample is bigger, and the TS scoring of model prediction is lower, and in turn, time scale interval is smaller,
That is the capacity of training sample is smaller, and the TS scoring of model prediction is higher.In addition, also illustrating that the rainfall at a certain moment is adjacent
History meteorological condition correlation is bigger, when the time point of the meteorological condition of historical juncture and rainfall generation is apart from each other (such as 72
Hour), the rainfall data at the following moment farther out which is included is than relatively limited, it is difficult to effectively to it is following farther out when
The precipitation predicting at quarter provides valuable reference, it may also be said to which closing on for future time instance has contained meteorology in historical data
The changed information of condition, the information are more valuable to the prediction of the following rainfall.Therefore, as seen from the figure, time scale interval
When being 3 hours, the TS scoring highest of the SVM rainfall dislocation prediction model prediction result after training, either prediction following 1 are small
When, 2 hours or 5 hours, TS scoring near or above 40%, has biggish practical value and realistic meaning.
The stability of model
The stability of model determines that can model continuously and effectively be predicted.The embodiment of the present invention calculates multiple times
The TS average and standard deviation of the multiple websites of scale, the stability for descriptive model.The mean value of multiple websites is closer, standard
Difference is smaller, and the prediction result for showing that model provides is more stable.
Figure 15 is different time scales SVM precipitation predicting TS scoring mean value, and the statistics of Cong Tuzhong can be seen that time scale
The precipitation predicting TS scoring mean value of precipitation predicting TS scoring mean value highest when being 3 hours, current time is more than 40%, future
1 hour precipitation predicting TS scores mean value close to 40%, and following 2 hours to following 5 hours TS scoring mean values are also all close
40%, which reflects the overall central tendencies of the precipitation predicting TS of each website scoring.By figure it can also be seen that time scale is
Precipitation predicting TS scoring mean value at 6 hours decreases, and with the increase of prediction lag time, fall is also increasing
Greatly, when the prediction lag time is 0 small when (current time), all website TS scoring mean values drop to 38%, and when prediction
When delay time is 5 hours, all website TS scoring mean values drop to 30% or so.Secondly, time scale be 8 hours, it is 12 small
When, when 24 hours, 48 hours, 72 hours precipitation predicting TS scoring mean value all present with time scale and prediction lag
The training sample of the increase of time and the trend gradually reduced, this prediction model that shows to misplace with rainfall increases, and draws in model
Enter the more meteorologic factor unrelated with the following rainfall, increases the complexity and uncertainty of model.
Figure 16 is that different time scales SVM precipitation predicting TS standards of grading are poor, and TS standards of grading difference expresses all websites
Precipitation predicting TS scoring deviates the degree of mean value, time scale is 3 hours, 6 hours, 8 hours, 12 hours, 24 hours, it is 48 small
When, 72 hours when the difference prediction lag time TS standards of grading difference it is smaller, only 0.015 or so.Pass through the system of the figure
Meter analysis can it is further seen that, TS standards of grading difference reflects the dispersion degree of all website TS appraisal results, is that the TS is commented
Divide result relative to a kind of probabilistic measurement of TS scoring mean value, it is pre- to represent most of website rainfalls for lesser standard deviation
The TS appraisal result of survey all relatively ensemble average values, and as shown in Figure 15, whole TS grade average is preferably up to 40%
More than, and TS appraisal result standard deviation is smaller preferably, represents so more stable, also illustrates that the TS appraisal result of different websites
Numerical fluctuations are smaller, conversely, then fluctuating larger.
It follows that the model stability that the precipitation predicting method and device of the embodiment of the present invention is established is preferable, do not stood
The influence of point position and environment, the equal table of rainfall of all automatic websites different for diverse geographic location, geographical social environment
Preferable prediction effect is revealed, the rainfall simultaneously for all websites following different prediction lag times is also shown preferably
Prediction effect, especially time scale is 3 hours SVM precipitation predictings, and effect is best.
The precipitation predicting method and device of the embodiment of the present invention is by establishing SVM rainfall dislocation prediction mould using meteorological data
Type determines model parameter gamma value and parameter C value by cross validation, to obtain optimal prediction of precipitation as a result, simultaneously root
Final rainfall probability is calculated according to prediction of precipitation result.Compared with the existing technology, the embodiment of the present invention can preferably reflect
The following rainfall trend more refined, has certain practical value, has preferable accuracy rate and kiss on rain time point
It is right.The present invention utilizes most basic meteorological element data, has better scalability, can adapt to more meteorological element data
Training and prediction.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (8)
1. a kind of precipitation predicting method characterized by comprising
Step a: setting time scale interval and prediction lag time, and built based on the corresponding meteorological data in time scale interval
Vertical Multiple Time Scales SVM rainfall dislocation prediction model；
Step b: computation model parameter gamma value and parameter C value, input SVM rainfall dislocation prediction model, when obtaining prediction lag
Between corresponding prediction of precipitation result；
Step c: rainfall probability is calculated according to the prediction of precipitation result；
Wherein, step a includes: that trained SVM is of the meteorological data of certain time before current time using current time as boundary
Rainfall dislocation prediction model, the rainfall of the following certain time is predicted by the SVM rainfall dislocation prediction model trained, when
Certain time before the preceding moment is " time scale interval ", and the following certain time of prediction is " prediction lag time ", when more
Between scale be refer to be arranged multiple time scale intervals, time scale interval and prediction lag time according to practical application into
Row setting；
In the step a, the modeling formula for establishing Multiple Time Scales SVM rainfall dislocation prediction model are as follows:
Yt-1=R* (Xt-1
Xt-1=[xt-h-d,xt-h-d+1,…,xt-d-1]T
Yt-1=[yt-h,yt-h+1,…,yt-1]T
Yt=R* (xt-d)
In above-mentioned formula, Xt-1, Yt-1For training sample, R* is the model established according to training sample, when t is to be predicted
It carves, h is the time scale interval of training sample, and d is prediction lag time, yt、xt-dRespectively precipitation predicting value, drop to be predicted
The input of rain.
2. precipitation predicting method according to claim 1, which is characterized in that in the step b, the meteorological number that is inputted
According to wind speed, temperature, air pressure and humidity including each automatic Weather Station.
3. precipitation predicting method according to claim 2, which is characterized in that in the step b, the model parameter
The calculation of gamma value and parameter C value specifically: determine that parameter gamma value and parameter C value, intersection are tested by cross validation
The evaluation criterion of card is TS scoring, selects TS and scores highest parameter gamma value and parameter C value as final suitable parameters.
4. precipitation predicting method according to claim 1, which is characterized in that in the step c, the rainfall probability meter
Calculate formula are as follows:
In above-mentioned formula, ytIt is model prediction of precipitation as a result, working as ytWhen > 0, P (yt) > 50%, conversely, P (yt)≤50%；
That is:
5. a kind of precipitation predicting device, which is characterized in that including model building module, model training module and precipitation predicting mould
Block；The model building module is based on time scale interval pair for time scale interval and prediction lag time to be arranged
The meteorological data answered establishes Multiple Time Scales SVM rainfall dislocation prediction model；The model training module is joined for computation model
Number gamma value and parameter C value, input SVM rainfall dislocation prediction model, obtain prediction lag time corresponding prediction of precipitation knot
Fruit；The precipitation predicting module is used to calculate rainfall probability according to the prediction of precipitation result；
Wherein, the model building module is specifically used for using current time as boundary, with the meteorology of certain time before current time
Data do trained SVM rainfall dislocation prediction model, and to misplace prediction model by the SVM rainfall trained following certain to predict
The rainfall of time, the certain time before current time are " time scale interval ", and the following certain time of prediction is " prediction
Delay time ", Multiple Time Scales are to refer to that multiple time scale intervals, time scale interval and prediction lag time is arranged
It is set according to practical application；
The model building module establishes the modeling formula of SVM rainfall dislocation prediction model are as follows:
Yt-1=R* (Xt-1)
Xt-1=[xt-h-d,xt-h-d+1,…,xt-d-1]T
Yt-1=[yt-h,yt-h+1,…,yt-1]T
Yt=R* (xt-d)
In above-mentioned formula, Xt-1, Yt-1For training sample, R* is the model established according to training sample, when t is to be predicted
It carves, h is the time scale interval of training sample, and d is prediction lag time, yt、xt-dRespectively precipitation predicting value, drop to be predicted
The input of rain.
6. precipitation predicting device according to claim 5, which is characterized in that the meteorology that the model training module is inputted
Data include wind speed, temperature, air pressure and the humidity of each automatic Weather Station.
7. precipitation predicting device according to claim 6, which is characterized in that the model training module computation model parameter
The calculation of gamma value and parameter C value specifically: determine that parameter gamma value and parameter C value, intersection are tested by cross validation
The evaluation criterion of card is TS scoring, selects TS and scores highest parameter gamma value and parameter C value as final suitable parameters.
8. precipitation predicting device according to claim 5, which is characterized in that the rainfall probability meter of the precipitation predicting module
Calculate formula are as follows:
In above-mentioned formula, ytIt is model prediction of precipitation as a result, working as ytWhen > 0, P (yt) > 50%, conversely, P (yt)≤50%；
That is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610752369.7A CN106405682B (en) | 2016-08-29 | 2016-08-29 | A kind of precipitation predicting method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610752369.7A CN106405682B (en) | 2016-08-29 | 2016-08-29 | A kind of precipitation predicting method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106405682A CN106405682A (en) | 2017-02-15 |
| CN106405682B true CN106405682B (en) | 2019-04-26 |
Family
ID=58003706
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610752369.7A Active CN106405682B (en) | 2016-08-29 | 2016-08-29 | A kind of precipitation predicting method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106405682B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110603465B (en) * | 2017-03-30 | 2021-12-28 | 精准天气预报股份有限公司 | System and method for forecasting snowfall probability distribution |
| CN107703564B (en) * | 2017-10-13 | 2020-04-14 | 中国科学院深圳先进技术研究院 | Rainfall prediction method and system and electronic equipment |
| CN109239807A (en) * | 2018-06-29 | 2019-01-18 | 平安科技（深圳）有限公司 | Rainfall appraisal procedure and system and terminal |
| CN108983323A (en) * | 2018-08-08 | 2018-12-11 | 湖北河海科技发展有限公司 | Precipitation forecast method and early warning platform based on optical flow method |
| CN109326087B (en) * | 2018-10-29 | 2020-09-08 | 广东奥博信息产业股份有限公司 | Urban waterlogging early warning method and device based on drainage pipe network monitoring |
| CN109447378B (en) * | 2018-12-26 | 2020-06-02 | 广东海洋大学 | Tibet plateau rainfall prediction method |
| CN110765878A (en) * | 2019-09-20 | 2020-02-07 | 苏州大圜科技有限公司 | Short-term rainfall prediction method |
| CN115438841A (en) * | 2019-11-13 | 2022-12-06 | 吉林农业大学 | Training method and prediction method based on artificial intelligence accurate prediction rainfall model |
| CN111427101B (en) * | 2020-04-07 | 2022-04-26 | 南京气象科技创新研究院 | Thunderstorm strong wind grading early warning method, system and storage medium |
| CN112528563B (en) * | 2020-12-08 | 2024-04-19 | 丹华水利环境技术(上海)有限公司 | Urban waterlogging early warning method based on SVM algorithm |
| CN113128778B (en) * | 2021-04-27 | 2022-08-26 | 最美天气(上海)科技有限公司 | Model training method based on graded TS meteorological scoring |
| CN114036204A (en) * | 2021-10-21 | 2022-02-11 | 河海大学 | Quantitative rainfall intermediate-term forecasting method based on open weather forecast product |
| CN114841476B (en) * | 2022-07-04 | 2022-09-02 | 水利部交通运输部国家能源局南京水利科学研究院 | Urban rainwater resource utilization space-time dynamic allocation and transaction method and system |
| CN115049013A (en) * | 2022-07-05 | 2022-09-13 | 中国电建集团西北勘测设计研究院有限公司 | Short-term rainfall early warning model fusion method combining linearity and SVM |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0921883A (en) * | 1995-07-04 | 1997-01-21 | Nippon Telegr & Teleph Corp <Ntt> | Hybrid meteorological forecast device |
| JP5090955B2 (en) * | 2008-02-13 | 2012-12-05 | 日本電信電話株式会社 | Lightning prediction device, method and program thereof |
| CN102221714B (en) * | 2011-03-11 | 2013-10-23 | 钱维宏 | Medium-range forecast system and method for low temperature, rain and snow and freezing weather based on atmospheric variable physical decomposition |
| JP6612523B2 (en) * | 2015-05-13 | 2019-11-27 | 株式会社東芝 | Flood warning system and notification method |
-
2016
- 2016-08-29 CN CN201610752369.7A patent/CN106405682B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106405682A (en) | 2017-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106405682B (en) | A kind of precipitation predicting method and device | |
| Mwangi et al. | Forecasting droughts in East Africa | |
| CN107292098A (en) | Medium-and Long-Term Runoff Forecasting method based on early stage meteorological factor and data mining technology | |
| Kyselý et al. | Comparison of regional and at‐site approaches to modelling probabilities of heavy precipitation | |
| KR101383617B1 (en) | Method and apparatus for predicting daily solar radiation level | |
| CN110070144A (en) | A kind of lake water quality prediction technique and system | |
| CN108227041A (en) | Horizontal visibility forecasting procedure based on website measured data and model results | |
| CN107169258A (en) | A kind of multi-source weather information data assimilation method and its application in rainfall forecast | |
| CN110598352B (en) | Drainage basin water supply forecasting method | |
| CN109325633A (en) | A kind of weather forecast set member choosing method and system | |
| CN102520464A (en) | Regional waterlogging forecasting system and forecasting method thereof | |
| KR101646587B1 (en) | Hybrid seasonal prediction of the Western North Pacific Tropical Cyclone Activity | |
| CN109143408A (en) | Combine short-term precipitation forecasting procedure in dynamic area based on MLP | |
| CN105447593A (en) | Rapid updating mixing assimilation method based on time lag set | |
| CN117787081A (en) | Hydrological model parameter uncertainty analysis method based on Morris and Sobol methods | |
| CN113592132B (en) | Rainfall objective forecasting method based on numerical weather forecast and artificial intelligence | |
| CN104834966A (en) | Fishery forecasting method based on ant colony classifying algorithm | |
| LU503105B1 (en) | Rice Productive Potential Simulation Method Based on Coupling of Land System and Climate Change | |
| Zhang et al. | Study on snowmelt runoff simulation in the Kaidu River basin | |
| CN103308956A (en) | Method for pre-judging future monthly average cloud cover in target area by utilizing cloud climatology data | |
| CN110619433B (en) | Rapid selection method and system for power grid heavy rain numerical mode parameterization scheme | |
| CN113592186B (en) | Hydrological prediction state variable real-time correction method based on real-time measured diameter flow data | |
| Herath et al. | Evaluation of HEC-HMS model for water resources management in Maha Oya Basin in Sri Lanka | |
| CN117172037B (en) | Distributed hydrologic forecasting method, device, computer equipment and medium | |
| CN101464979A (en) | System and method for evaluating crop yield representation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liu Jun Inventor after: Cao Chunyan Inventor after: Chen Jinsong Inventor after: He Jiajia Inventor after: Chen Kai Inventor before: Liu Jun Inventor before: Cao Chunlan Inventor before: Chen Jinsong Inventor before: He Jiajia Inventor before: Chen Kai |
|
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |