CN109783774B - Temperature set forecasting method and system - Google Patents
Temperature set forecasting method and system Download PDFInfo
- Publication number
- CN109783774B CN109783774B CN201811549198.3A CN201811549198A CN109783774B CN 109783774 B CN109783774 B CN 109783774B CN 201811549198 A CN201811549198 A CN 201811549198A CN 109783774 B CN109783774 B CN 109783774B
- Authority
- CN
- China
- Prior art keywords
- forecast
- field
- time
- forecasting
- mode
- 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
- 238000013277 forecasting method Methods 0.000 title claims abstract description 10
- 238000011478 gradient descent method Methods 0.000 claims abstract description 28
- 238000000034 method Methods 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 17
- 238000010276 construction Methods 0.000 claims description 10
- 230000001932 seasonal effect Effects 0.000 claims description 5
- 230000009191 jumping Effects 0.000 claims description 2
- 238000012935 Averaging Methods 0.000 claims 1
- 238000012937 correction Methods 0.000 abstract description 17
- 230000000694 effects Effects 0.000 abstract description 6
- 238000007689 inspection Methods 0.000 abstract description 4
- 230000035772 mutation Effects 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 13
- 238000001556 precipitation Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 241000219094 Vitaceae Species 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 235000021021 grapes Nutrition 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Air Conditioning Control Device (AREA)
Abstract
The temperature set forecasting method and the system provided by the invention start from a plurality of numerical mode forecasting fields through temperature set forecasting, carry out dynamic error correction and forecasting effect inspection on each forecasting field according to the forecasting field and the observation field in the past fifteen days, adjust the temperature mutation condition, calculate each mode weight by adopting a gradient descent method, and further obtain the optimal ground air temperature forecasting field by weighted average.
Description
Technical Field
The invention relates to the technical field of weather forecast, in particular to a temperature set forecast system and a method.
Background
Internationally, krishnamurti et al (1999) originally proposed the idea of multi-mode aggregate forecasting, and statistically optimizes a plurality of different mode forecasting results to finally obtain an aggregate forecasting result. A large number of experiments show that the prediction of the set effectively reduces the errors of the prediction of the season and the weather and the prediction of the weather, and the prediction effect is far better than that of single-mode and multi-mode set average. In terms of precipitation forecasting, krishnamurti et al conduct super aggregate forecasting study on precipitation, precipitation in the period of rain and typhoon landing strong precipitation in the southern sea monsoon burst in the chinese monsoon area based on five central global modes of the TIGGE data, uk weather bureau (United Kingdom Meteorological Office, UKMO), us environmental prediction center (National Centers for Environmental Prediction, NCEP), middle european weather prediction center (European Centre for Medium-Range Weather Forecasts, ECMWF), australian weather bureau (Bureau of Meteorology, australia, BOM), chinese weather bureau (China Meteorological Administration, CMA), and discuss extending forecasting timeliness from 1-3d to 10d, with minimal super aggregate root mean square error for 4-10d forecasting. The multi-mode integrated forecasting technology has the advantages of effectively improving season climate forecasting skills, improving medium-short term forecasting accuracy, being simple and practical and the like in the fields of conventional weather factor forecasting such as air temperature and the like and strong weather forecasting such as precipitation, typhoon and the like, and has been widely researched and applied internationally.
Compared with the international intensive research and application, the multi-mode integrated forecasting technology has been tried initially in the aspects of ground temperature and the like, wherein the starting of China is late. Zhao Shengrong (2006) based on the high-temperature forecast of the China national weather center T213 mode, the German weather office business mode and the Japanese weather hall business mode 2m, a BP neural network method is utilized to establish a temperature integrated forecast system of more than 600 stations in China, and the result shows that the temperature forecast of the set is obviously superior to the forecast results of the 3 modes alone, and the accuracy of the forecast results is improved to a certain extent. The intelligent cooperative flight and the like use the central set forecasting results of the TIGGE dataset European middle weather forecasting center (ECMWF), the Japanese weather hall (JMA), the national environmental forecasting center (NCEP) and the British weather office (UKMO) and the like to conduct set forecasting experiments on ground temperature and rainfall, and the results show that although the applicability of the multi-mode integration method for different forecasting elements is different, the forecasting effect of the multi-mode integration method is superior to that of a single center as a whole.
The existing set forecasting technology has obvious development, but the set forecasting model building method is single, a set average or linear weighted average method is adopted, the correction of systematic errors caused by the errors of numerical mode forecasting is still to be improved, particularly, the mode errors under the condition of abrupt temperature change (abrupt temperature rise or temperature drop) are larger, extreme weather such as high temperature and the like often occurs in Shenzhen or even Guangdong Australian regions, the deviation of numerical mode forecasting products under the condition is larger, and the construction of a stable temperature set forecasting system is needed to provide reliable reference for weather forecasting.
Disclosure of Invention
Accordingly, it is desirable to provide a stable temperature set forecasting system, a temperature set forecasting method and system for providing reliable reference for weather forecasting, and a computer program product for implementing the method.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
in one aspect, the invention provides a temperature set forecasting method, which comprises the following steps:
acquiring a current temperature forecasting field of each mode forecasting field;
judging whether the difference value between the current temperature forecast value of each mode forecast field and the forecast value at the same time of the previous day is larger than or smaller than a certain threshold value; if not, executing the next step; if yes, jumping to the next step, and executing the next step;
correcting result deviation of the current temperature mode prediction field;
calculating the weight of each mode prediction field by adopting a gradient descent method;
obtaining an optimal ground air temperature forecasting field by adopting weighted average according to the weights of all the mode forecasting fields;
and outputting the optimal ground air temperature forecasting field.
In some preferred embodiments, in the step of determining whether the difference between the current temperature forecast value and the forecast value at the same time of the previous day for each mode forecast field is greater than or less than a certain threshold, the threshold may be calculated based on different regional and seasonal statistics.
In some preferred embodiments, the step of correcting the result deviation of the current temperature prediction field specifically includes the following steps:
defining a first formula, wherein the first formula is b k,r,f ＝F k,r,f -O r+f Wherein F k,r,f Is the forecast of the k-th set member in advance of f time at r time, O r+f Is the observed value of the moment r+f;
defining a second formula, wherein the second formula is
Q 1 ，Q 2 And Q 3 Forecast deviation b at the same time of fifteen days k,r,f The 1,2,3 25% percentile values of the sequence are used for calculating the forecast average deviation value of the reporting time which is the time f advanced by the r moment in the past 15 days;
defining a third formula for revising result deviation of the current temperature prediction field, wherein the third formula is that
Wherein (1)>
The kth set member r curr Forecast of the moment in advance of the time f->
The kth set member calculated for the above step has last fifteen days r (and r curr At the same time) forecast average deviation value of advance f time,/>
Then it is the kth set member r curr The time is advanced by the deviation of the f time to correct the forecast field.
In some preferred embodiments, in the step of calculating the prediction field weight of each mode by using the gradient descent method, the method specifically includes the following steps:
defining a root mean square error cost function, wherein the root mean square error cost function is
Wherein m is the number of sliding training period data sets, F (x r,f ) Is the forecast of f time in advance of r time in training period, O r+f Is the actual observation value of the training period r+f moment, < >>
Wherein (1)>
To unify the weights, the weighted average yields the predicted value F (x r,f )；
Solving the root mean square error cost function optimal solution by using a gradient descent method so as to minimize the error;
and obtaining the optimal weight of each mode forecasting field according to the optimal solution.
In some preferred embodiments, in the step of obtaining an optimal ground air temperature prediction field by using weighted average according to the weights of the prediction fields in each mode, the following formula is specifically adopted to obtain the optimal ground air temperature prediction field:
In another aspect, the present invention further provides a temperature set forecasting system, including:
the current temperature forecast field acquisition module is used for acquiring the current temperature forecast field of each mode forecast field;
the forecast value difference judging module is used for judging whether the difference value between the current temperature forecast value of each mode forecast field and the forecast value at the same time of the previous day is larger than or smaller than a certain threshold value;
the forecast result deviation correcting module is used for correcting the result deviation of the forecast field of the current temperature mode;
the gradient descent method calculation module is used for calculating the weight of each mode prediction field by adopting a gradient descent method;
the weighted average calculation module is used for obtaining an optimal ground air temperature forecasting field by adopting weighted average according to the weights of all the mode forecasting fields;
and the ground air temperature forecast value output module is used for outputting the optimal ground air temperature forecast field.
In some preferred embodiments, in the forecast value difference determination module, the threshold may be calculated based on different regional and seasonal statistics.
In some preferred embodiments, the forecast result deviation correction module includes:
a first construction unit for defining a first formula, wherein the first formula is b k,r,f ＝F k,r,f -O r+f Wherein F k,r,f Is the forecast of the k-th set member in advance of f time at r time, O r+f Is the observed value of the moment r+f;
a second construction unit for a second formula, the second formula being
Q 1 ，Q 2 And Q 3 Forecast deviation b at the same time of fifteen days k,r,f The 1,2,3 25% percentile values of the sequence are used for calculating the forecast average deviation value of the reporting time which is the time f advanced by the r moment in the past 15 days;
a third construction unit for defining a third formula for performing result deviation on the current temperature prediction fieldRevising the third formula to
Wherein (1)>
The kth set member r curr The moment is advanced by a forecast of f times,
the kth set member calculated for the above step has last fifteen days r (and r curr At the same time) forecast average deviation value of advance f time,/>
Then it is the kth set member r curr The time is advanced by the deviation of the f time to correct the forecast field.
In some preferred embodiments, the gradient descent method calculation module includes:
a first calculation unit for defining a root mean square error cost function, the root mean square error cost function being
Wherein m is the number of sliding training period data sets, F (x r,f ) Is the forecast of f time in advance of r time in training period, O r+f Is the actual observation value of the training period r+f moment, < >>
Wherein (1)>
To unify the weights, the weighted average yields the predicted value F (x r,f )；
The gradient descent method solving unit is used for solving the root mean square error cost function optimal solution by using a gradient descent method so as to minimize the error;
a second calculation unit for obtaining optimal forecast field weights of each mode according to the optimal solutionHeavy weight
In some preferred embodiments, the weighted average calculation module obtains the optimal ground air temperature prediction field by using the following formula:
The invention adopts the technical proposal has the advantages that:
the temperature set forecasting method and the system provided by the invention start from a plurality of numerical mode forecasting fields through temperature set forecasting, carry out dynamic error correction and forecasting effect inspection on each forecasting field according to the forecasting field and the observation field in the past fifteen days, adjust the temperature mutation condition, calculate each mode weight by adopting a gradient descent method, and further obtain the optimal ground air temperature forecasting value by weighted average.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart illustrating steps of a temperature set forecasting system according to embodiment 1 of the present invention.
Fig. 2 is a schematic structural diagram of a temperature set forecasting system provided in embodiment 2 of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1, the present invention provides a step flow chart of a temperature set forecasting method, which includes the following steps:
step S110: and acquiring the current temperature forecast field of each mode forecast field.
It can be understood that the temperature set forecast obtained by the invention is sent out from a plurality of numerical mode forecast fields, and the collection of the plurality of mode forecast fields can enable the forecast result to be more accurate and reliable.
Step S120: judging whether the difference value between the current temperature forecast value of each mode forecast field and the forecast value at the same time of the previous day is larger than or smaller than a certain threshold value; if not, go to step S130; if yes, go to the next step, and execute step S140. The threshold may be calculated based on different regional and seasonal statistics.
It will be appreciated that for each mode forecast field, the difference between the current forecast value and the forecast value at the same time of the previous day is greater than or less than a certain threshold (abrupt rise or fall in temperature), the error correction is abandoned, i.e
Wherein,,
the kth set member r curr Forecast of the moment in advance of the time f->
For the kth set member r curr If the time is equal, the error correction is not performed, namely, the breakthrough of temperature occurs at the moment, and more frequent temperature drop occurs, if the adjustment is performed according to the historical deviation, larger system errors are easily caused, and if the deviation correction of the mode prediction result is not performed, the deviation correction of the mode prediction result is performed.
Step S130: and correcting the result deviation of the current temperature prediction field.
In some preferred embodiments, the method specifically comprises the following steps:
step S131: defining a first formula, wherein the first formula is b k,r,f ＝F k,r,f -O r+f Wherein F k,r,f Is the forecast of the k-th set member in advance of f time at r time, O r+f Is the observed value of the moment r+f;
step S132: defining a second formula, wherein the second formula is
Q 1 ，Q 2 And Q 3 Forecast deviation b at the same time of fifteen days k,r,f The 1,2,3 25% percentile values of the sequence are used to calculate the forecast average deviation value for the time of the report as the time of the advance f of the r moment in the past 15 days.
It will be appreciated that the deviation from the actual is not known from the current numerical pattern results, as estimated by a dynamic average forecast deviation over the last 15 days (this time is adjustable).
Step S133: defining a third formula for revising result deviation of the current temperature prediction field, wherein the third formula is that
Wherein (1)>
The kth set member r curr Forecast of the moment in advance of the time f->
The kth set member calculated for the above step has last fifteen days r (and r curr At the same time) forecast average deviation value of advance f time,/>
Then it is the kth set member r curr The time is advanced by the deviation of the f time to correct the forecast field.
It will be appreciated that for each numerical pattern, each forecast length, the average deviation over fifteen days is revised
Is applied to the current integrated forecasting result to generate a forecasting field with deviation correction.
The step S131 to the step S133 can correct the result deviation of the current temperature prediction field.
Step S140: and calculating the weight of each mode prediction field by adopting a gradient descent method.
In some preferred embodiments, the method specifically comprises the following steps:
step S141: defining a root mean square error cost function, wherein the root mean square error cost function is
Wherein m is the number of sliding training period data sets, F (x r,f ) Is the forecast of f time in advance of r time in training period, O r+f Is the actual observation value of the training period r+f moment, < >>
Wherein (1)>
To unify the weights, the weighted average yields the predicted value F (x r,f )；
Step S142: solving the root mean square error cost function optimal solution by using a gradient descent method so as to minimize the error;
it can be understood that the gradient descent method is used to solve the root mean square error cost function optimal solution, namely, to differentiate each variable (weight), specifically:
wherein ECMWF (European middle weather forecast center), JMA (Japan weather forecast), GRAPES (Global model set forecast) and the like are set forecast members (which can be adjusted according to requirements), and it can be understood that proper learning rate is set, and the optimal solution of the gradient descent solution weight is set so as to minimize the error.
Step S143: obtaining optimal weights of all mode prediction fields according to the optimal solution
It can be understood that, because each mode has different forecasting characteristics, the system gives different weights to different modes, so that the mode weight with strong forecasting capability is relatively larger, and gradient descent is used for calculating the weight of each mode
Step S150: and obtaining the optimal ground air temperature forecasting field by adopting weighted average according to the weight of each mode forecasting field.
The optimal ground air temperature forecasting field is obtained by adopting the following formula:
Step S160: and outputting the optimal ground air temperature forecasting field.
The temperature set forecasting method provided by the invention starts from a plurality of numerical mode forecasting fields through temperature set forecasting, performs dynamic error correction and forecasting effect inspection on each forecasting field according to the forecasting field and the observation field in the past fifteen days, adjusts the temperature mutation condition, calculates each mode weight by adopting a gradient descent method, and further obtains an optimal ground air temperature forecasting value by weighted average.
Example 2
Referring to fig. 2, a schematic structure diagram of a temperature set prediction system is provided in the present invention, including:
the current temperature prediction field acquisition module 110: and acquiring the current temperature forecast field of each mode forecast field.
It can be understood that the temperature set forecast obtained by the invention is sent out from a plurality of numerical mode forecast fields, and the collection of the plurality of mode forecast fields can enable the forecast result to be more accurate and reliable.
The forecast value difference value judging module 120: and the method is used for judging whether the difference value between the current temperature forecast value of each mode forecast field and the forecast value at the same time of the previous day is larger than or smaller than a certain threshold value. The threshold may be calculated based on different regional and seasonal statistics.
It will be appreciated that for each mode forecast field, the difference between the current forecast value and the forecast value at the same time of the previous day is greater than or less than a certain threshold (abrupt rise or fall in temperature), the error correction is abandoned, i.e
Wherein,,
the kth set member r curr Forecast of the moment in advance of the time f->
For the kth set member r curr If the time is equal, the error correction is not performed, namely, the breakthrough of temperature occurs at the moment, and more frequent temperature drop occurs, if the adjustment is performed according to the historical deviation, larger system errors are easily caused, and if the deviation correction of the mode prediction result is not performed, the deviation correction of the mode prediction result is performed.
Forecast result deviation correction module 130: and correcting the result deviation of the current temperature prediction field.
In some preferred embodiments, the method specifically comprises the following steps:
a first construction unit: for defining a first formula, the first formula being b k,r,f ＝F k,r,f -O r+f Wherein F k,r,f Is the forecast of the k-th set member in advance of f time at r time, O r+f Is the observed value of the moment r+f;
a second construction unit: for defining a second formula, the second formula being
Q 1 ，Q 2 And Q 3 Forecast deviation b at the same time of fifteen days k,r,f 25% percentile values of 1,2,3 of the sequenceThe method is used for calculating the forecast average deviation value of which the reporting time is the time f advanced by the time r in the past 15 days.
It will be appreciated that the deviation from the actual is not known from the current numerical pattern results, as estimated by a dynamic average forecast deviation over the last 15 days (this time is adjustable).
A third construction unit: for defining a third formula for revising the result deviation of the current temperature prediction field, wherein the third formula is that
Wherein (1)>
The kth set member r curr The moment is advanced by a forecast of f times,
the kth set member calculated for the above step has last fifteen days r (and r curr At the same time) forecast average deviation value of advance f time,/>
Then it is the kth set member r curr The time is advanced by the deviation of the f time to correct the forecast field.
It will be appreciated that for each numerical pattern, each forecast length, the average deviation over fifteen days is revised
Is applied to the current integrated forecasting result to generate a forecasting field with deviation correction.
The module can be used for correcting the result deviation of the current temperature prediction field.
Gradient descent method calculation module 140: and calculating the weight of each mode prediction field by adopting a gradient descent method.
In some preferred embodiments, the method specifically comprises the following steps:
a first calculation unit: defining root mean square error cost functionNumber, the root mean square error cost function is
Wherein m is the number of sliding training period data sets, F (x r,f ) Is the forecast of f time in advance of r time in training period, O r+f Is the actual observation value of the training period r+f moment, < >>
Wherein (1)>
To unify the weights, the weighted average yields the predicted value F (x r,f )；
Gradient descent method solving unit: solving the root mean square error cost function optimal solution by using a gradient descent method so as to minimize the error;
it can be understood that the gradient descent method is used to solve the root mean square error cost function optimal solution, namely, to differentiate each variable (weight), specifically:
wherein ECMWF (European middle weather forecast center), JMA (Japan weather forecast), GRAPES (Global model set forecast) and the like are set forecast members (which can be adjusted according to requirements), and it can be understood that proper learning rate is set, and the optimal solution of the gradient descent solution weight is set so as to minimize the error.
A second calculation unit: obtaining optimal solutions of the weights of the prediction fields of all modes according to the optimal solutions
It will be appreciated that the system will be due to the different forecasting characteristics of each modeGiving different modes different weights, so that the mode weight with strong forecasting capability is relatively larger, and calculating the weight of each mode by using gradient descent
Weighted average calculation module 150: the method is used for obtaining the optimal ground air temperature forecast value by adopting weighted average according to the forecast field weight of each mode.
Specifically, the weighted average calculation module obtains an optimal ground air temperature prediction field by adopting the following formula:
Ground air temperature forecast value output module 160: and outputting the optimal ground air temperature forecasting field.
The temperature set forecasting system provided by the invention starts from a plurality of numerical mode forecasting fields through temperature set forecasting, performs dynamic error correction and forecasting effect inspection on each forecasting field according to the forecasting field and the observation field in the past fifteen days, adjusts the temperature mutation condition, calculates each mode weight by adopting a gradient descent method, and further obtains the optimal ground air temperature forecasting field by weighted average.
Of course, the temperature set forecasting method and system of the present invention can also have various changes and modifications, and are not limited to the specific structure of the above embodiments. In general, the scope of the present invention should include those variations or alternatives and modifications apparent to those skilled in the art.
Claims (8)
1. A method for forecasting a temperature set, comprising the steps of:
acquiring a current temperature forecasting field of each mode forecasting field;
judging whether the difference value between the current temperature forecast value of each mode forecast field and the forecast value at the same time of the previous day is larger than or smaller than a certain threshold value; if not, executing the next step; if yes, jumping to the next step, and executing the next step;
correcting result deviation of the current temperature mode prediction field;
calculating the weight of each mode prediction field by adopting a gradient descent method;
obtaining an optimal ground air temperature forecasting field by adopting weighted average according to the weights of all the mode forecasting fields;
outputting the optimal ground air temperature forecasting field;
the step of correcting the result deviation of the current temperature prediction field specifically comprises the following steps:
defining a first formula, wherein the first formula is b k,r,f ＝F k,r,f -O r+f Wherein F k,r,f Is the forecast of the k-th set member in advance of f time at r time, O r+f Is the observed value of the moment r+f;
defining a second formula, wherein the second formula is
Q 1 ，Q 2 And Q 3 Forecast deviation b at the same time of fifteen days k,r,f The 1,2,3 25% percentile values of the sequence are used for calculating the forecast average deviation value of the reporting time which is the time f advanced by the r moment in the past 15 days;
defining a third equation for the currentRevising result deviation of the temperature prediction field, wherein the third formula is that
Wherein (1)>
The kth set member r curr Forecast of the moment in advance of the time f->
Forecast average deviation value of f time earlier for the last fifteen days r of the kth set member calculated for the above steps,/">
Then it is the kth set member r curr Correcting the forecast field by advancing the time by the deviation of the time f, and correcting the r time and the r curr At the same time.
2. The temperature set forecasting method of claim 1, wherein in the step of judging whether the difference between the current temperature forecast value and the forecast value of the same time of the previous day of each mode forecasting field is greater than or less than a certain threshold value, the threshold value can be calculated according to different regions and seasons.
3. The method of claim 1, wherein in the step of calculating the respective mode prediction field weights by using a gradient descent method, the method comprises the steps of:
defining a root mean square error cost function, wherein the root mean square error cost function is
Wherein m is the number of sliding training period data sets, F (x r,f ) Is the forecast of f time in advance of r time in training period, O r+f Is the actual observation value of the training period r+f moment, < >>
Wherein (1)>
To unify the weights, the weighted average yields the predicted value F (x r,f )；
Solving the root mean square error cost function optimal solution by using a gradient descent method so as to minimize the error;
and obtaining the optimal weight of each mode forecasting field according to the optimal solution.
4. The method of claim 1, wherein in the step of obtaining an optimal ground air temperature prediction field by weighted averaging according to the respective model prediction field weights, the optimal ground air temperature prediction field is obtained by specifically using the following formula:
5. A temperature aggregate forecast system, comprising:
the current temperature forecast field acquisition module is used for acquiring the current temperature forecast field of each mode forecast field;
the forecast value difference judging module is used for judging whether the difference value between the current temperature forecast value of each mode forecast field and the forecast value at the same time of the previous day is larger than or smaller than a certain threshold value;
the forecast result deviation correcting module is used for correcting the result deviation of the current temperature forecast field;
the gradient descent method calculation module is used for calculating the weight of each mode prediction field by adopting a gradient descent method;
the weighted average calculation module is used for obtaining an optimal ground air temperature forecasting field by adopting weighted average according to the weights of all the mode forecasting fields;
the ground air temperature forecast value output module is used for outputting the optimal ground air temperature forecast field;
the forecast result deviation correcting module comprises:
a first construction unit for defining a first formula, wherein the first formula is b k,r,f ＝F k,r,f -O r+f Wherein F k,r,f Is the forecast of the k-th set member in advance of f time at r time, O r+f Is the observed value of the moment r+f;
a second construction unit for a second formula, the second formula being
Q 1 ，Q 2 And Q 3 Forecast deviation b at the same time of fifteen days k,r,f The 1,2,3 25% percentile values of the sequence are used for calculating the forecast average deviation value of the reporting time which is the time f advanced by the r moment in the past 15 days;
a third construction unit, configured to define a third formula for revising the result deviation of the current temperature prediction field, where the third formula is that
Wherein (1)>
The kth set member r curr Forecast of the moment in advance of the time f->
The kth calculated for the above stepForecast average deviation value of f time earlier by r time in last fifteen days of each collection member, +.>
Then it is the kth set member r curr Correcting the forecast field by advancing the time by the deviation of the time f, and correcting the r time and the r curr At the same time.
6. The temperature aggregate forecast system of claim 5, wherein in the forecast value difference determination module, the threshold is calculated based on different regional and seasonal statistics.
7. The temperature aggregate forecast system of claim 5, wherein the gradient descent method calculation module comprises:
a first calculation unit for defining a root mean square error cost function, the root mean square error cost function being
Wherein m is the number of sliding training period data sets, F (x r,f ) Is the forecast of f time in advance of r time in training period, O r+f Is the actual observation value of the training period r+f moment, < >>
Wherein (1)>
To unify the weights, the weighted average yields the predicted value F (x r,f )；
The gradient descent method solving unit is used for solving the root mean square error cost function optimal solution by using a gradient descent method so as to minimize the error;
a second calculation unit for obtaining the optimal prediction field weight of each mode according to the optimal solution
8. The temperature aggregate forecast system of claim 5, wherein the weighted average calculation module obtains the optimal ground air temperature forecast field by using the formula:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811549198.3A CN109783774B (en) | 2018-12-18 | 2018-12-18 | Temperature set forecasting method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811549198.3A CN109783774B (en) | 2018-12-18 | 2018-12-18 | Temperature set forecasting method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109783774A CN109783774A (en) | 2019-05-21 |
| CN109783774B true CN109783774B (en) | 2023-05-23 |
Family
ID=66497247
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811549198.3A Active CN109783774B (en) | 2018-12-18 | 2018-12-18 | Temperature set forecasting method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109783774B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111538935B (en) * | 2019-12-26 | 2023-08-25 | 北京玖天气象科技有限公司 | Fine precipitation fusion method, system, electronic equipment and storage medium based on terrain features and multi-source mode products |
| CN115827710B (en) * | 2022-12-20 | 2023-06-27 | 中国民用航空局空中交通管理局航空气象中心 | Screening method and system for meteorological element multi-mode forecasting result |
| CN117849908B (en) * | 2024-03-08 | 2024-05-10 | 江苏省气候中心 | Plum-entering and plum-exiting date prediction method and device in plum rainy season based on mode circular flow field |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447593A (en) * | 2015-11-16 | 2016-03-30 | 南京信息工程大学 | Rapid updating mixing assimilation method based on time lag set |
| CN108399469A (en) * | 2018-01-31 | 2018-08-14 | 中国人民解放军空军研究院战场环境研究所 | A kind of weather phenomenon forecasting procedure based on deep learning and numerical weather forecast |
| CN108983324A (en) * | 2018-08-09 | 2018-12-11 | 深圳市雅码科技有限公司 | A kind of temperature forecast method and system based on Kalman filtering |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140336934A1 (en) * | 2013-05-07 | 2014-11-13 | Atomic Energy Council - Institute Of Nuclear Energy Research | Ensemble wind power forecasting platform system and operational method thereof |
-
2018
- 2018-12-18 CN CN201811549198.3A patent/CN109783774B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447593A (en) * | 2015-11-16 | 2016-03-30 | 南京信息工程大学 | Rapid updating mixing assimilation method based on time lag set |
| CN108399469A (en) * | 2018-01-31 | 2018-08-14 | 中国人民解放军空军研究院战场环境研究所 | A kind of weather phenomenon forecasting procedure based on deep learning and numerical weather forecast |
| CN108983324A (en) * | 2018-08-09 | 2018-12-11 | 深圳市雅码科技有限公司 | A kind of temperature forecast method and system based on Kalman filtering |
Non-Patent Citations (1)
| Title |
|---|
| 基于TIGGE资料的地面气温多模式超级集合预报;林春泽等;《应用气象学报》;20091215(第06期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109783774A (en) | 2019-05-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109783774B (en) | Temperature set forecasting method and system | |
| Nipen et al. | Adopting citizen observations in operational weather prediction | |
| CN110363327B (en) | ConvLSTM and 3D-CNN-based short rainfall prediction method | |
| Sun et al. | Projected changes in temperature and precipitation in ten river basins over China in 21st century. | |
| Stathopoulos et al. | Wind power prediction based on numerical and statistical models | |
| US20160203245A1 (en) | Method for simulating wind field of extreme arid region based on wrf | |
| CN111898660B (en) | Hydrological simulation method for average fusion of multi-source data based on Bayesian mode | |
| CN111257970B (en) | Precipitation prediction correction method and system based on aggregate prediction | |
| CN106022470B (en) | A kind of tropospheric delay correction method based on BP-EGNOS Fusion Model | |
| CN111007582B (en) | High-resolution ensemble forecasting system suitable for multiple elements | |
| Roulin et al. | Post‐processing of medium‐range probabilistic hydrological forecasting: impact of forcing, initial conditions and model errors | |
| CN113419246B (en) | Nudging approximation multi-moment 3DVar analysis field method for high-frequency assimilation of radar data | |
| CN113435630B (en) | Basin hydrological forecasting method and system with self-adaptive runoff yield mode | |
| CN105975791A (en) | Sparse region rainfall estimation fusion method based on dual-smoothing method | |
| CN112859202A (en) | Method and equipment for forecasting air temperature | |
| CN110896761A (en) | Irrigation decision-making method and system for greenhouse | |
| CN105447593A (en) | Rapid updating mixing assimilation method based on time lag set | |
| Gunavathi et al. | Assessment of various bias correction methods on precipitation of regional climate model and future projection | |
| CN106355264A (en) | Combined prediction method of reference crop evapotranspiration | |
| CN114330478B (en) | Wind speed classification correction method for wind speed forecast of power grid | |
| CN113361190B (en) | CMONOC-ZTD-based troposphere delay correction method | |
| AU2021105536A4 (en) | A High Spatial-Temporal Resolution Method for Near-Surface Air Temperature Reconstruction | |
| Fooladmand | Comparing reference evapotranspiration using actual and estimated sunshine hours in south of Iran | |
| Feng | A new method for ionospheric short-term forecast using similar-day modeling | |
| Stalenberg et al. | Using historical normals to improve modern monthly climate normal surfaces for Madagascar |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |