CN111856621A - Integrated evolution SVD conversion method based on mode and observation data fusion - Google Patents
Integrated evolution SVD conversion method based on mode and observation data fusion Download PDFInfo
- Publication number
- CN111856621A CN111856621A CN202010691849.3A CN202010691849A CN111856621A CN 111856621 A CN111856621 A CN 111856621A CN 202010691849 A CN202010691849 A CN 202010691849A CN 111856621 A CN111856621 A CN 111856621A
- Authority
- CN
- China
- Prior art keywords
- forecasting
- forecast
- prediction
- mode
- factor
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
Abstract
An integrated evolution SVD conversion method based on mode and observation data fusion can be used for regional weather/extended period prediction of climate process, and the method comprises the steps of filtering small-scale fast-varying noise of observation forecast factors and forecast object variables, and extracting useful forecastable information; extracting evolution coupling modes between each forecasting factor and a future forecasting object in the early stage and the same stage through SVD; converting and reconstructing a forecasting factor expansion coefficient and an SVD (singular value decomposition) mode which are fused with observation and mode forecasting data, and integrating multi-factor results to obtain a real-time quantitative prediction result of future 30 weather temperature and rainfall; finally, by combining a process identification technology, the extended period prediction of the weather/climate process of 30 days in the future is realized; according to the prediction method, the spatial-temporal dynamic coupling relation between the prediction factors and the prediction objects is utilized, the theoretical advantages of statistics and dynamic mode prediction are combined, the regional process with predictability is subjected to extended-period prediction, and the stability and reliability of the prediction effect are guaranteed.
Description
Technical Field
The invention relates to the technical field of climate extension period prediction, in particular to an integrated evolution SVD conversion method based on mode and observation data fusion, which can be particularly used for regional summer heavy precipitation, high-temperature hot waves, winter low-temperature cold damage and other important disastrous weather or climate process extension period prediction.
Background
China is located in the east Asia monsoon region, is influenced by multiple factors of audiences, has complex and changeable climate and is a high-incidence region of climate disasters; annual climate disasters can cause billions of economic losses, which seriously affects the life and property safety of people; therefore, accurate short-term climate prediction information is urgently needed by various departments such as national agriculture, water conservancy, disaster prevention and reduction and the like and the masses of people so as to scientifically serve the healthy development of national economic construction and reduce casualties and economic losses caused by climate disasters.
The climate prediction means that the possible trend of climate development in a certain period in the future is deduced according to the evolution law of the climate in the past; at present, the conventional weather forecast mainly focuses on 10 days, the short-term weather forecast is mainly forecast on a monthly time scale, and the 10-30 days of extension period forecast between the two forecast becomes a forecast gap; however, the extended period prediction of strong processes such as strong precipitation, high temperature hot waves, low temperature cold damage and the like can reserve preparation time for disaster prevention deployment and emergency plans of government departments, and is one of the key meteorological service points urgently needed by the social public and the government departments; for the prediction of the extension period, information contained in an initial place is gradually dissipated along with the increase of the prediction time, but the action of external force is not completely dominant at the moment, so that the mutation mechanism of an extension period prediction object is very complex, and the method is an extremely difficult scientific problem; the current short-term climate prediction technology can be mainly divided into two types, one type is a statistical prediction method, and the other type is a dynamic numerical prediction method; practice shows that both the statistical method and the dynamic method have certain accuracy in short-term climate prediction, and both can reflect partial rules of atmospheric motion, but have respective inherent defects; the statistical prediction method is mainly based on the physical statistical rule mastered in historical observation data, and variables with predictability are selected to construct a statistical prediction model; the method has the advantages that the statistical model is good at grasping the statistical causal relationship between the climate variability signal and the forecast variable, and has certain forecasting capacity for events meeting stationarity conditions; however, due to the complexity of climate variability, under different ages or changes of the ages, the time-lag statistical relationship of the prediction model may also change, so that the prediction effect is unstable; the dynamics numerical prediction is that a closed equation set for describing atmospheric motion is constructed on the basis of the theories of hydrodynamics, atmospheric dynamics, thermodynamics and the like, and a numerical mode is established by adopting computational mathematics and a high-performance electronic computing mechanism to predict the climate; at present, the prediction capability of most power modes at home and abroad on an annular flow field is greatly improved, but larger errors exist in direct prediction on temperature and precipitation; therefore, the prediction of the extension period by means of the statistical model or the dynamic mode alone cannot meet the actual requirements of disaster prevention and reduction, and the prediction result of the dynamic mode needs to be explained and applied by some statistical methods, or the prediction information of the dynamic mode is added into the statistical model, and the idea of combining the dynamic and statistical methods is utilized to further improve the prediction capability.
In the east asian region, the signal is influenced by extension phase signals from tropical and medium and high latitude regions; for example, the remote response signals of tropical season internal oscillation, tropical sea temperature, medium and high latitude low frequency oscillation and polar sea ice can have significant influence on the climate in east asian region; therefore, it is necessary to traverse the factors for different forecast object variables to find the combination of the factor variables with the most significant extension phase signal and the highest forecast skill; meanwhile, a coupling mode which is closest to the relationship between the atmospheric extension period signal and the regional temperature/precipitation and evolves along with the time needs to be identified, and a good causal relationship between a forecasting factor and a forecasting variable is established, so that a forecasting result has a physical significance; moreover, how to utilize the forecasting advantage of the power mode on the circulation factor to form a power-statistics combined forecasting model, and the forecasting capability is improved to the maximum extent; these are all problems to be solved in the prediction of the extension period.
Disclosure of Invention
The invention aims to provide an integrated evolution SVD conversion method based on mode and observation data fusion, which can effectively solve the problems in the background technology.
In order to solve the problems, the technical scheme adopted by the invention is as follows: an integrated evolution SVD conversion method based on mode and observation data fusion comprises the following steps:
Step 1: selecting forecast factor variables and forecast object variables for statistical investigation;
step 2: respectively preprocessing a forecast factor variable and a forecast object variable in observation, and filtering out fast-changing noise interference;
and step 3: selecting the space range of the forecast factor variable according to the lead-lag correlation relationship of the forecast factor and the forecast factor; reconstructing the forecast factor variable and the forecast object variable, and extracting the forecastable information;
and 4, step 4: decomposing the forecasting factor and the forecasting object after preprocessing and denoising reconstruction by using SVD (singular value decomposition), obtaining the evolution coupling mode of the forecasting factor and the forecasting object, and establishing an integrated evolution SVD conversion method model based on mode and observation data fusion;
and 5: performing space-time conversion on a forecasting factor expansion coefficient and an SVD (singular value decomposition) mode fused with observation and mode forecasting data, reconstructing forecasting variables, performing magnitude modulation, and integrating multi-factor results to obtain a final real-time forecasting result of a forecasting object;
step 6: and obtaining the extension period prediction of the important disastrous weather or climate process by combining a process identification technology according to the real-time prediction result of the forecast object.
As a further preferable scheme of the present invention, in step 1, the reanalyzed data includes atmospheric circulation dynamics such as temperatures, air pressures, wind speeds and humidity of high, middle and low layers of the convection layer and thermodynamic meteorological elements as factor variables for the process prediction in the extension period, and the station air temperature and precipitation in the modeling area as forecast object variables.
As a further preferable scheme of the invention, step 2, firstly, removing the climate state of the forecasting factor and the forecasting variable, and extracting an abnormal component; and secondly, performing 5-day moving average processing to filter small-scale interference and fast-changing noise.
As a further preferable scheme of the invention, step 3 is that the average air temperature and precipitation in the modeling area are respectively subjected to linear correlation with the contemporaneous and prophase forecast factor variables, and the spatial range of the forecast factor variables is determined according to the significance of the correlation coefficient.
As a further preferable scheme of the invention, step 3 utilizes principal component EOF analysis to perform space-time filtering processing on the forecast factor variable and the forecast object variable, and selects the modal forecast factor variable and the forecast object variable with the cumulative variance contribution up to 85% for reconstruction.
As a further preferable scheme of the invention, the forecasting factor of the model in the step 4 is a circulation factor sequence 30 days before and after the forecasting start day, and the forecasting object is a temperature and precipitation sequence 30 days in the future of the forecasting start day; selecting forecasting factors of m consecutive days before the forecasting initial day and N consecutive days after the forecasting initial day (m + N = 30) to construct an SVD left field oblique square difference matrix, constructing an SVD right field oblique square difference matrix by forecasting objects of 30 consecutive days after the forecasting initial day, carrying out SVD decomposition on the left field oblique square difference matrix and the right field oblique square difference matrix, and obtaining N groups of forecasting factors and the evolution coupling mode of the forecasting objects according to the decomposition:
As a further preferable scheme of the present invention, in the real-time prediction, when the real-time prediction factor is constructed, the reanalysis data of m consecutive days before the initial prediction day and the mode prediction data of n consecutive days after the initial prediction day are constructed into a prediction factor sequence of 30 consecutive days, the prediction factor sequence is projected to the SVD left singular feature vector by using the least square method, the obtained projection coefficient is used as the estimation value of the SVD right modal time coefficient, and the point multiplication is performed to the SVD right singular feature vector to obtain the prediction variable sequence of 30 days in the future:
taking the forecast variable sequence of the future 30 days obtained by accumulating the N groups of modes as a model forecast output result of an integrated evolution SVD conversion method based on mode and observation data fusion; and finally, performing magnitude modulation on the final prediction result by utilizing the ratio of the observation standard deviation and the standard deviation output by the integrated evolution SVD conversion method based on mode and observation data fusion:
finally, the final real-time forecasting result of the forecasting object is obtained by utilizing the multi-factor forecasting result integration
As a further preferred scheme of the invention, in the actual modeling process, a cross-checking method is used, a stable and reliable mode is reserved, and an optimal forecasting factor is determined; in actual operation, when reconstructing the M almanac recalculation data, 1 year is selected as a target year, and the rest M-1 years are selected as modeling objects to reconstruct the recalculation result of the target year; repeating the steps to obtain complete M-year back calculation data finally; and determining an optimal forecasting factor, an observation-mode combination number and an SVD mode number N according to the back calculation data to form a final integrated evolution SVD conversion method prediction model based on mode and observation data fusion.
As a further preferable scheme of the present invention, in the process identification, in step 6, starting from a single station, the temperature or precipitation at the current day and 5 days before and after the current day of year 2010 in 1981-2010 is selected for cumulative probability calculation. When the cumulative probability of the air temperature or the precipitation is higher than a certain percentage, for example, higher than 90%, defining a distance threshold value corresponding to the probability as a threshold value of a strong high temperature or strong precipitation process; when the air temperature range is larger than the threshold value at a certain day, a strong high-temperature process is considered to occur at the day; when the precipitation distance is larger than the threshold value, the strong precipitation process is considered to occur in the day; when the accumulated probability of the air temperature is lower than a certain percentage, for example, lower than 10%, defining a pitch critical value corresponding to the probability as a threshold value of the strong and low temperature process; when the air temperature range of a certain day is less than a threshold value, a strong low-temperature process is considered to occur in the day; if the time lasts for 1 day or more and is not interrupted, defining the time duration as a strong process of the station; therefore, important disastrous weather/climate processes such as strong precipitation in summer, high-temperature hot waves, low-temperature cold damage and the like of a single station in 30 days in the future can be identified and predicted by combining a process identification method according to the station air temperature/precipitation forecast result of the model of the integrated evolution SVD conversion method based on mode and observation data fusion.
Compared with the prior art, the method has the following advantages:
the method can well reflect the physical mechanism that the climate abnormities of different regions are influenced by different atmospheric anomaly forcing factors of the tropical zone and the other zone according to the difference of different climate regions;
continuous time evolution information between various different forecasting factors and target area climate forecasting variables can be considered, a coupling mode which is most closely related to regional difference is found, the forecasting factors and the forecasting objects have good causal relationship, and the cause of climate abnormity can be physically explained and understood;
modeling by adopting an integrated evolution SVD conversion method based on mode and observation data fusion and combining power-statistics, fully utilizing the accuracy of recent observation and mode for factor prediction in a certain period, providing the advantage of combining power mode and statistical prediction for prediction based on SVD conversion by combining the two data, and ensuring the optimal performance and stable result of a prediction model;
the SVD conversion prediction technology considering the time evolution grasps the physical influence process of continuous change between the prediction factor and the prediction object, and the integration of different factor results can more stably improve the prediction accuracy;
Compared with quantitative prediction of temperature/precipitation, the method provided by the invention predicts the extension period process with predictability, and ensures the stability and reliability of the prediction effect.
Drawings
FIG. 1 is a simplified flow diagram of a prediction method of the present invention;
FIG. 2 is a flow chart of an extended period process prediction method of the present invention;
FIG. 3 is an example of winter half-year evolution SVD modeling;
FIG. 4 is a diagram of a technique distribution of temperature prediction in winter in a Bohai and Bohai sea test point region based on different SVD reconstruction modal numbers of slp factors and different observation-mode data combination schemes; FIG. 4a is a prediction skill distribution diagram of different SVD reconstruction modes of the test point region in the Bohai sea in winter air temperature based on the fusion integration of an slp factor and a 20 observation +10 mode combination scheme; FIG. 4b is a prediction skill distribution diagram of a different data integration combination scheme of the air temperature in winter in the Bohai sea test point region based on the slp factor and 4 groups of SVD reconstruction modes;
FIG. 5 is a diagram of a prediction skill distribution of precipitation in winter in a test point region of Bohai and Bohai sea based on different SVD reconstruction modal numbers of slp factors and different observation-mode data combination schemes; FIG. 5a is a prediction skill distribution diagram of different SVD reconstruction modes of test point region winter precipitation in Bohai sea based on fusion integration of slp factor and 15 observation +15 mode combination scheme; FIG. 5b is a prediction skill distribution diagram of a different data integration combination scheme of precipitation in winter in the Bohai sea test point region based on slp factors and 8 groups of SVD reconstruction modes;
FIG. 6 is a prediction skill distribution diagram of winter air temperature in a Bohai and Bohai sea test point region based on an integration evolution SVD conversion method based on mode and observation data fusion of an slp factor; FIG. 6a is a prediction skill distribution diagram for forecasting winter air temperature in a Bohai sea test point region 5 days in advance by an integration evolution SVD conversion method based on mode and observation data fusion based on an slp factor; FIG. 6b is a prediction skill distribution diagram for forecasting the winter air temperature in a Bohai sea test point region 10 days in advance by an integration evolution SVD conversion method based on the fusion of a pattern and observation data and based on an slp factor; FIG. 6c is a prediction skill distribution diagram for forecasting the winter air temperature in a Bohai sea test point region 15 days in advance by an integration evolution SVD conversion method based on the fusion of the pattern and the observation data and based on the slp factor; FIG. 6d is a prediction skill distribution diagram for forecasting the winter air temperature in a Bohai sea test point region 20 days in advance by an integration evolution SVD conversion method based on the fusion of the pattern and the observation data and based on the slp factor; FIG. 6e is a prediction skill distribution diagram for forecasting the winter air temperature in a Bohai sea test point region 25 days in advance by an integration evolution SVD conversion method based on the fusion of the pattern and the observation data and based on the slp factor; FIG. 6f is a prediction skill distribution diagram for forecasting the winter air temperature in a Bohai sea test point region 30 days in advance by an integration evolution SVD conversion method based on the fusion of a pattern and observation data and based on an slp factor;
FIG. 7 is a diagram of a prediction skill distribution of precipitation in winter in a test point region of the Bohai and Bohai sea based on an olr + slp + hgt500 integrated evolution SVD conversion method based on mode and observation data fusion; FIG. 7a is a prediction skill distribution diagram for forecasting winter precipitation in a Bohai sea test point region 5 days in advance by an integration evolution SVD conversion method based on olr + slp + hgt500 integration and based on mode and observation data fusion; FIG. 7b is a prediction skill distribution diagram for forecasting winter precipitation in a Bohai sea test point region 10 days in advance by an integration evolution SVD conversion method based on olr + slp + hgt500 integration and based on mode and observation data fusion; FIG. 7c is a prediction skill distribution diagram for forecasting winter precipitation in a Bohai sea test point region 15 days in advance by an integration evolution SVD conversion method based on olr + slp + hgt500 integration and based on mode and observation data fusion; FIG. 7d is a prediction skill distribution diagram for forecasting winter precipitation in a Bohai sea test point region 20 days in advance by an integration evolution SVD conversion method based on olr + slp + hgt500 integration and based on mode and observation data fusion; FIG. 7e is a prediction skill distribution diagram for forecasting winter precipitation in a Bohai sea test point region 25 days in advance by an integration evolution SVD conversion method based on olr + slp + hgt500 integration and based on mode and observation data fusion; FIG. 7f is a prediction skill distribution diagram for forecasting winter precipitation in a test point region of the Bohai sea in 30 days in advance by an integration evolution SVD conversion method based on olr + slp + hgt500 integration and based on mode and observation data fusion.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
The invention provides an integrated evolution SVD conversion method based on mode and observation data fusion, which comprises the following steps:
step 1: selecting forecast factor variables and forecast object variables for statistical investigation;
step 2: respectively preprocessing a forecast factor variable and a forecast object variable in observation, and filtering out fast-changing noise interference;
and step 3: selecting the space range of the forecast factor variable according to the lead-lag correlation relationship of the forecast factor and the forecast factor; reconstructing the forecast factor variable and the forecast object variable, and extracting the forecastable information;
and 4, step 4: decomposing the forecasting factor and the forecasting object after preprocessing and denoising reconstruction by using SVD (singular value decomposition), obtaining the evolution coupling mode of the forecasting factor and the forecasting object, and establishing an integrated evolution SVD conversion method model based on mode and observation data fusion;
and 5: performing space-time conversion on a forecasting factor expansion coefficient and an SVD (singular value decomposition) mode fused with observation and mode forecasting data, reconstructing forecasting variables, performing magnitude modulation, and integrating multi-factor results to obtain a final real-time forecasting result of a forecasting object;
Step 6: and obtaining the extension period prediction of the important disastrous weather or climate process by combining a process identification technology according to the real-time prediction result of the forecast object.
As a further preferable scheme of the present invention, in step 1, the reanalyzed data includes atmospheric circulation dynamics such as temperatures, air pressures, wind speeds and humidity of high, middle and low layers of the convection layer and thermodynamic meteorological elements as factor variables for the process prediction in the extension period, and the station air temperature and precipitation in the modeling area as forecast object variables.
As a further preferable scheme of the invention, step 2, firstly, removing the climate state of the forecasting factor and the forecasting variable, and extracting an abnormal component; secondly, performing 5-day sliding average processing to filter small-scale interference and fast-changing noise; and obtaining a factor variable for filtering the fast-changing noise and a forecast variable for filtering the fast-changing noise.
As a further preferable scheme of the invention, step 3 is that the average air temperature and precipitation in the modeling area are respectively subjected to linear correlation with the contemporaneous and prophase forecast factor variables, and the spatial range of the forecast factor variables is determined according to the significance of the correlation coefficient.
As a further preferable scheme of the invention, step 3, performing space-time filtering processing on the forecast factor variable and the forecast object variable by using principal component EOF analysis, and selecting the modal forecast factor variable and the forecast object variable with the cumulative variance contribution up to 85% for reconstruction; and obtaining a forecasting factor with forecasting significance and a forecasting object with forecasting property.
As a further preferable scheme of the invention, the forecasting factor of the model in the step 4 is a circulation factor sequence 30 days before and after the forecasting start day, and the forecasting object is a temperature and precipitation sequence 30 days in the future of the forecasting start day; selecting forecasting factors of m consecutive days before the forecasting initial day and N consecutive days after the forecasting initial day (m + N = 30) to construct an SVD left field oblique square difference matrix, constructing an SVD right field oblique square difference matrix by forecasting objects of 30 consecutive days after the forecasting initial day, carrying out SVD decomposition on the left field oblique square difference matrix and the right field oblique square difference matrix, and obtaining N groups of forecasting factors and the evolution coupling mode of the forecasting objects according to the decomposition:
wherein the content of the first and second substances,tfor the forecast start day, X is the SVD left field and consists of forecast factor sequences 30 days (m + n) before and after the forecast start day; y is an SVD right field and consists of a forecast object sequence of 30 days in the future of a forecast initial day;
is a warpPredictor after SVD decompositioniThe left singular feature vector is a vector of the left singular feature vector,
is a corresponding firstiA time coefficient;
for the forecast object after SVD decompositioniThe right-hand singular feature vector is,
is a corresponding firstiTime coefficient, N is the total number of modes required for constructing the prediction model,ix、iyas longitude and latitude points of the spatial range of the forecast factor, istaThe site is a test point area site.
As a further preferable scheme of the present invention, in the real-time prediction, when the real-time prediction factor is constructed, the reanalysis data of m consecutive days before the initial prediction day and the mode prediction data of n consecutive days after the initial prediction day are constructed into a prediction factor sequence of 30 consecutive days, the prediction factor sequence is projected to the SVD left singular feature vector by using the least square method, the obtained projection coefficient is used as the estimation value of the SVD right modal time coefficient, and the point multiplication is performed to the SVD right singular feature vector to obtain the prediction variable sequence of 30 days in the future:
wherein the content of the first and second substances,
in order to forecast the target day of the day,
is a forecast amount. Taking the forecast variable sequence of the future 30 days obtained by accumulating the N groups of modes as a model forecast output result of an integrated evolution SVD conversion method based on mode and observation data fusion; and finally, performing magnitude modulation on the final prediction result by utilizing the ratio of the observation standard deviation and the standard deviation output by the integrated evolution SVD conversion method based on mode and observation data fusion:
wherein the content of the first and second substances,
the prediction result is obtained after the magnitude modulation;
in order to observe the standard deviation of the light,
outputting a standard deviation for an integrated evolution SVD conversion method model based on mode and observation data fusion; the remaining variables are as above.
And finally, integrating the multi-factor forecasting results to obtain the final real-time forecasting result of the forecasting object.
As a further preferred scheme of the invention, in the actual modeling process, a cross-checking method is used, a stable and reliable mode is reserved, and an optimal forecasting factor is determined; in actual operation, when reconstructing the M almanac recalculation data, 1 year is selected as a target year, and the rest M-1 years are selected as modeling objects to reconstruct the recalculation result of the target year; repeating the steps to obtain complete M-year back calculation data finally; and determining an optimal forecasting factor, an observation-mode combination number and an SVD mode number N according to the back calculation data to form a final integrated evolution SVD conversion method prediction model based on mode and observation data fusion.
As a further preferable scheme of the invention, in the process identification of step 6, starting from a single station, selecting the temperature or precipitation at the current day and 5 days before and after the current day of year 2010 in 1981-2010 for cumulative probability calculation; when the cumulative probability of the air temperature or the precipitation is higher than a certain percentage, for example, higher than 90%, defining a distance threshold value corresponding to the probability as a threshold value of a strong high temperature or strong precipitation process; when the air temperature range is larger than the threshold value at a certain day, a strong high-temperature process is considered to occur at the day; when the precipitation distance is larger than the threshold value, the strong precipitation process is considered to occur in the day; when the accumulated probability of the air temperature is lower than a certain percentage, for example, lower than 10%, defining a pitch critical value corresponding to the probability as a threshold value of the strong and low temperature process; when the air temperature range of a certain day is less than a threshold value, a strong low-temperature process is considered to occur in the day; if the time lasts for 1 day or more and is not interrupted, defining the time duration as a strong process of the station; therefore, important disastrous weather/climate processes such as strong precipitation in summer, high-temperature hot waves, low-temperature cold damage and the like of a single station in 30 days in the future can be identified and predicted by combining a process identification method according to the station air temperature/precipitation forecast result of the model of the integrated evolution SVD conversion method based on mode and observation data fusion.
As a specific embodiment of the present invention:
the method is specifically applied to the prediction of the air temperature and the precipitation process of the Bohai region of the Bohai sea as an example; forecasting factor signals are derived from an obvious circulation factor field in a Bohai and hoop region, forecasting objects are temperature and precipitation in winter and half a year (10 months and 16 days to 5 months and 14 days in the next year) in the Bohai and hoop region, and a process identification technology is combined to forecast the low-temperature cold damage and strong precipitation process; the method mainly comprises the following steps:
(1) selection of forecasting factor and data preprocessing
Selection of predictor variables and predictor variables:
selecting reanalysis data of sea level air pressure slp and external long wave radiation olr and 500 hPa potential altitude field hgt500 as prediction factor variables, and using air temperature and precipitation observation data of a 316 station in a Bohai and Bohai region as prediction object variables; and respectively removing climate states of the forecasting factors and the forecasting object, reserving abnormal components, performing 5-day sliding average processing, and filtering out the fast-changing noise.
(2) De-noising reconstruction of predictor and predictor objects
Selecting a spatial range of a forecast factor:
and respectively solving linear correlation between the average air temperature and precipitation sequence of the ring Bohai 316 station area and slp, olr and hgt500 at the same period and the early period, and determining the modeling space range of the slp, olr and hgt500 factors in air temperature and precipitation prediction according to the significance of correlation coefficients.
Denoising and reconstructing:
and (3) respectively carrying out EOF decomposition on the slp, olr and hgt500 processed in the step (1) and the air temperature and precipitation data of the 316 station in the Bohai and Bohai region, and selecting a mode with the cumulative variance contribution reaching 85% for reconstruction to extract useful forecastable information.
(3) Establishment of model for integrated evolution SVD conversion method prediction based on mode and observation data fusion
Firstly, establishing an evolution SVD:
coupling the preprocessed slp, olr and hgt500 and reconstructed data of temperature and precipitation of 316 stations in the Bohai and Bohai regions by using an SVD (singular value decomposition) method, and establishing an integrated evolution SVD conversion method model based on mode and observation data fusion; and selecting slp, olr and hgt500 variables which are m days (namely-m +1 d-0 d) before the forecast initial day and N days (namely 1 d-nd) after the forecast initial day in the SVD left field to form a forecast factor sequence (wherein m + N = 30), and selecting an air temperature and precipitation sequence which is 30 days in the future of the forecast initial day in the SVD right field to obtain N groups of forecast factors and the coupling modes of forecast objects.
Observation and mode data fusion transformation reconstruction and integrated prediction:
when a real-time forecasting factor is constructed, re-analysis data of slp, olr and hgt500 for m consecutive days before the forecasting initial day and forecasting data of slp, olr and hgt500 for n consecutive days after the forecasting initial day in a BCC _ CSM1.2 mode are constructed into a forecasting factor sequence for 30 consecutive days, the forecasting factor sequence is projected to a SVD left singular characteristic vector by using a least square method, an obtained projection coefficient is used as an estimation of a SVD right modal time coefficient, and the estimation coefficient is point-multiplied to the SVD right singular characteristic vector to obtain a forecasting variable sequence for 30 days in the future; taking the forecast variable sequence of the future 30 days obtained by accumulating the N groups of modes as a model forecast output result of an integrated evolution SVD conversion method based on mode and observation data fusion; and finally, performing magnitude modulation on the final prediction result by utilizing the ratio of the observation standard deviation and the standard deviation output by the integrated evolution SVD conversion method based on mode and observation data fusion.
And thirdly, determining the optimal mode number and the optimal observation and mode combination mode by cross inspection:
in the modeling process, a cross-checking method is used, stable and reliable mode number N is reserved, and an optimal forecasting factor and an m + N combination of the factors are selected; in actual operation, when reconstructing the M almanac history back calculation data, selecting 1 year as a target year, and the rest M-1 years as modeling objects to reconstruct the return result of the target year; repeating the steps to obtain complete M-year back calculation data finally; the back calculation data is subjected to a skill analysis to determine the final optimal parameters N, m and n.
In the inspection, the consistency of model prediction and actual observation can be characterized on significance by using a method of a Time Correlation Coefficient (TCC), and the formula is shown as the following formula:
wherein the content of the first and second substances,ithe date of the start of the newspaper is shown,jthe date of the target is indicated,
is represented iniSun pairjThe prediction of the day or days is made,
representsiThe average of the predictions for a day is,
to representjThe observation of the day is abnormal,
to represent
M represents a time series number, TCC ranges from-1 to 1, with the closer TCC is to 1 indicating the higher the model prediction skill.
(4) Identification prediction of heavy precipitation process
For Beijing station, selecting temperature and precipitation distance data of the current day and 5 days before and after the current day of year 1981-2010 for cumulative probability calculation; when the cumulative probability is higher than 90%, defining a distance critical value corresponding to the probability as a threshold value of a strong process; when the air temperature range is larger than the threshold value at a certain day, the strong high-temperature process is considered to occur in Beijing city at the certain day; when the precipitation distance is larger than the threshold value, the strong precipitation process is considered to occur in the day; when the cumulative probability of the air temperature is lower than 10%, defining a distance critical value corresponding to the probability as a threshold value of the strong and low temperature process; when the air temperature range of a certain day is less than a threshold value, a strong low-temperature process is considered to occur in the day; if the time lasts for 1 day or more and is not interrupted, defining the time duration as a strong process of the station; on the basis, according to the model prediction result of the integrated evolution SVD conversion method based on mode and observation data fusion, the extension period prediction is carried out on important disastrous weather and climate processes such as heavy rainfall in summer, high-temperature hot waves, low-temperature cold damage in winter and the like of a single station.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (9)
1. An integrated evolution SVD conversion method based on mode and observation data fusion comprises the following steps:
step 1: selecting forecast factor variables and forecast object variables for statistical investigation;
step 2: respectively preprocessing a forecast factor variable and a forecast object variable in observation, and filtering out fast-changing noise interference;
and step 3: selecting the space range of the forecast factor variable according to the lead-lag correlation relationship of the forecast factor and the forecast factor; reconstructing the forecast factor variable and the forecast object variable, and extracting the forecastable information;
and 4, step 4: decomposing the forecasting factor and the forecasting object after preprocessing and denoising reconstruction by using SVD (singular value decomposition), obtaining the evolution coupling mode of the forecasting factor and the forecasting object, and establishing an integrated evolution SVD conversion method model based on mode and observation data fusion;
And 5: performing space-time conversion on a forecasting factor expansion coefficient and an SVD (singular value decomposition) mode fused with observation and mode forecasting data, reconstructing forecasting variables, performing magnitude modulation, and integrating multi-factor results to obtain a final real-time forecasting result of a forecasting object;
step 6: and obtaining the extension period prediction of the important disastrous weather or climate process by combining a process identification technology according to the real-time prediction result of the forecast object.
2. The integrated evolution SVD conversion method based on the mode and observation data fusion of claim 1, wherein step 1 uses the atmospheric circulation dynamics such as the temperature, the air pressure, the wind speed, the humidity of the upper, middle and lower layers of the convection layer and the thermal meteorological elements in the reanalysis data as the factor variables of the extended period process prediction, and uses the station air temperature and the precipitation in the modeling area as the forecast object variables.
3. The SVD conversion method based on the integration evolution of the mode and observation data fusion according to claim 1, wherein the method comprises the steps of 2, removing the climate state from the forecast factor and the forecast variable, and extracting the abnormal component; and secondly, performing 5-day moving average processing to filter small-scale interference and fast-changing noise.
4. The SVD conversion method based on the integration evolution of the mode and observation data fusion according to claim 1, wherein the step 3 is to make linear correlation between the average air temperature and precipitation in the modeling region and the forecast factor variables at the same period and the previous period respectively, and determine the spatial range of the forecast factor variables according to the significance of the correlation coefficient.
5. The SVD conversion method based on the mode and observation data fusion of claim 1, wherein the method comprises step 3, performing spatio-temporal filtering on the forecast factor variable and the forecast object variable by using principal component EOF analysis, and selecting the modal forecast factor variable and the forecast object variable with the cumulative variance contribution up to 85% for reconstruction.
6. The SVD conversion method based on the integration evolution of the mode and the observation data fusion according to claim 1, wherein the forecasting factors of the model in step 4 are circulation factor sequences 30 days before and after the forecast start day, and the forecast objects are temperature and precipitation sequences 30 days in the future of the forecast start day; selecting forecasting factors of m consecutive days before the forecasting initial day and N consecutive days after the forecasting initial day (m + N = 30) to construct an SVD left field oblique square difference matrix, constructing an SVD right field oblique square difference matrix by forecasting objects of 30 consecutive days after the forecasting initial day, carrying out SVD decomposition on the left field oblique square difference matrix and the right field oblique square difference matrix, and obtaining N groups of forecasting factors and the evolution coupling mode of the forecasting objects according to the decomposition:
7. the integrated evolution SVD conversion method based on mode and observation data fusion according to claim 1, characterized in that, in real-time prediction, when constructing a real-time prediction factor, the step 5 constructs reanalysis data for m consecutive days before the prediction initial day and mode prediction data for n consecutive days after the prediction initial day into a prediction factor sequence for 30 consecutive days, projects the prediction factor sequence to SVD left singular feature vector by using least square method, and dot-multiplies the projection coefficient to SVD right singular feature vector as the estimated value of SVD right modal time coefficient to obtain the prediction variable sequence for 30 future days:
Taking the forecast variable sequence of the future 30 days obtained by accumulating the N groups of modes as a single-factor forecast output result of the model of the integrated evolution SVD conversion method based on mode and observation data fusion; and (3) performing magnitude modulation on the prediction result by utilizing the ratio of the observation standard deviation and the standard deviation output by the integrated evolution SVD conversion method based on mode and observation data fusion:
and finally, integrating the multi-factor forecasting results to obtain the final real-time forecasting result of the forecasting object.
8. The SVD conversion method based on the integration evolution of the mode and the observation data fusion according to claim 7, wherein the method comprises the steps of using a cross-check method in the actual modeling process, retaining stable and reliable modes, and determining the optimal forecasting factor; in actual operation, when reconstructing the M almanac recalculation data, 1 year is selected as a target year, and the rest M-1 years are selected as modeling objects to reconstruct the recalculation result of the target year; repeating the steps to obtain complete M-year back calculation data finally; and determining an optimal forecasting factor, an observation-mode combination number and an SVD mode number N according to the back calculation data to form a final integrated evolution SVD conversion method prediction model based on mode and observation data fusion.
9. The SVD conversion method based on the integration evolution of the mode and the observation data fusion as claimed in claim 1, wherein in the process identification, in step 6, the temperature or precipitation of the current day and 5 days before and after the current day of year 2010 in 1981-2010 is selected for the cumulative probability calculation from a single station; when the cumulative probability of the air temperature or the precipitation is higher than a certain percentage, for example, higher than 90%, defining a distance threshold value corresponding to the probability as a threshold value of a strong high temperature or strong precipitation process; when the air temperature range is larger than the threshold value at a certain day, a strong high-temperature process is considered to occur at the day; when the precipitation distance is larger than the threshold value, the strong precipitation process is considered to occur in the day; when the accumulated probability of the air temperature is lower than a certain percentage, for example, lower than 10%, defining a pitch critical value corresponding to the probability as a threshold value of the strong and low temperature process; when the air temperature range of a certain day is less than a threshold value, a strong low-temperature process is considered to occur in the day; if the time lasts for 1 day or more and is not interrupted, defining the time duration as a strong process of the station; therefore, important disastrous weather/climate processes such as strong precipitation in summer, high-temperature hot waves, low-temperature cold damage and the like of a single station in 30 days in the future can be identified and predicted by combining a process identification method according to the station air temperature/precipitation forecast result of the model of the integrated evolution SVD conversion method based on mode and observation data fusion.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010691849.3A CN111856621B (en) | 2020-07-17 | 2020-07-17 | Integrated evolution SVD conversion method based on mode and observation data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010691849.3A CN111856621B (en) | 2020-07-17 | 2020-07-17 | Integrated evolution SVD conversion method based on mode and observation data fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111856621A true CN111856621A (en) | 2020-10-30 |
| CN111856621B CN111856621B (en) | 2021-09-28 |
Family
ID=73001151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010691849.3A Active CN111856621B (en) | 2020-07-17 | 2020-07-17 | Integrated evolution SVD conversion method based on mode and observation data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111856621B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113255962A (en) * | 2021-04-16 | 2021-08-13 | 北京玖天气象科技有限公司 | Multi-mode sea wave integrated forecasting system and method |
| CN113779760A (en) * | 2021-08-10 | 2021-12-10 | 南京大学 | Power-statistics combined seasonal climate prediction method based on predictable climate modality |
| CN113807583A (en) * | 2021-09-18 | 2021-12-17 | 南京信息工程大学 | Low-frequency increment space-time coupling-based power statistics combined sub-season prediction method |
| CN113837620A (en) * | 2021-09-26 | 2021-12-24 | 华北电力大学 | Method for evaluating temperature extreme value change |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104392097A (en) * | 2014-10-24 | 2015-03-04 | 封国林 | Seasonal precipitation analogue prediction method based on seasonal prediction mode |
| CN110058328A (en) * | 2019-01-30 | 2019-07-26 | 沈阳区域气候中心 | Summer Precipitation in Northeast China multi-mode combines NO emissions reduction prediction technique |
-
2020
- 2020-07-17 CN CN202010691849.3A patent/CN111856621B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104392097A (en) * | 2014-10-24 | 2015-03-04 | 封国林 | Seasonal precipitation analogue prediction method based on seasonal prediction mode |
| CN110058328A (en) * | 2019-01-30 | 2019-07-26 | 沈阳区域气候中心 | Summer Precipitation in Northeast China multi-mode combines NO emissions reduction prediction technique |
Non-Patent Citations (3)
| Title |
|---|
| ZHIWEI ZHU ET AL.: "A spatial–temporal projection model for extended-range forecast in the tropics", 《CLIM. DYN.》 * |
| ZHIWEI ZHU ET AL.: "Statistical extended-range forecast of winter surface air temperature and extremely cold days over China", 《Q.J.R.METEOROL.SOC.》 * |
| 任宏利等: "MJO预报研究进展", 《应用气象学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113255962A (en) * | 2021-04-16 | 2021-08-13 | 北京玖天气象科技有限公司 | Multi-mode sea wave integrated forecasting system and method |
| CN113779760A (en) * | 2021-08-10 | 2021-12-10 | 南京大学 | Power-statistics combined seasonal climate prediction method based on predictable climate modality |
| CN113779760B (en) * | 2021-08-10 | 2023-08-04 | 南京大学 | Power-statistics combined season climate prediction method based on predictable climate mode |
| CN113807583A (en) * | 2021-09-18 | 2021-12-17 | 南京信息工程大学 | Low-frequency increment space-time coupling-based power statistics combined sub-season prediction method |
| CN113807583B (en) * | 2021-09-18 | 2023-05-05 | 南京信息工程大学 | Subseasonal prediction method based on dynamic statistics combination of low-frequency increment space-time coupling |
| CN113837620A (en) * | 2021-09-26 | 2021-12-24 | 华北电力大学 | Method for evaluating temperature extreme value change |
| CN113837620B (en) * | 2021-09-26 | 2024-02-02 | 华北电力大学 | Method for evaluating temperature extremum change |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111856621B (en) | 2021-09-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111856621B (en) | Integrated evolution SVD conversion method based on mode and observation data fusion | |
| Espeholt et al. | Skillful twelve hour precipitation forecasts using large context neural networks | |
| CN108508505A (en) | Heavy showers and thunderstorm forecasting procedure based on multiple dimensioned convolutional neural networks and system | |
| Shi et al. | Short-term wind power generation forecasting: Direct versus indirect ARIMA-based approaches | |
| CN109583653B (en) | Northwest Pacific ocean typhoon extension period forecasting method based on statistical model | |
| Saxena et al. | A review study of weather forecasting using artificial neural network approach | |
| Younes et al. | Solid waste forecasting using modified ANFIS modeling | |
| Liu et al. | Estimation of precipitation induced by tropical cyclones based on machine‐learning‐enhanced analogue identification of numerical prediction | |
| Gairaa et al. | Contribution of ordinal variables to short-term global solar irradiation forecasting for sites with low variabilities | |
| CN114936691A (en) | Temperature forecasting method integrating relevance weighting and space-time attention | |
| Tang et al. | A new hybrid forecasting model based on dual series decomposition with long-term short-term memory | |
| CN113984198B (en) | Shortwave radiation prediction method and system based on convolutional neural network | |
| Gao et al. | A multifactorial framework for short-term load forecasting system as well as the jinan’s case study | |
| CN114385611A (en) | Precipitation prediction method and system based on artificial intelligence algorithm and knowledge graph | |
| William et al. | Framework for IOT Based Real-Time Monitoring System of Rainfall Water Level for Flood Prediction Using LSTM Network | |
| Atasever et al. | A new approach to short-term wind speed prediction: the prophet model | |
| CN116090596A (en) | Typhoon intensity prediction method, typhoon intensity prediction device, typhoon intensity prediction equipment and storage medium | |
| Wu et al. | Prediction of egional temperature change trend based on LSTM algorithm | |
| McNown et al. | Forecasting annual water demands dominated by seasonal variations: the case of water demands in Mecca | |
| Jiang et al. | A hybrid degradation tendency measurement method for mechanical equipment based on moving window and Grey–Markov model | |
| Zhou et al. | An explainable recurrent neural network for solar irradiance forecasting | |
| Ng et al. | Statistical downscaled local climate model for future rainfall changes analysis: A case study of hyogo prefecture, Japan | |
| Patanè et al. | Model transferability for wave height prediction | |
| Wu et al. | A Statistical ARIMA Model to Predict Arctic Environment for NSR Shipping | |
| CN116384284B (en) | Red tide gridding forecasting method and system |
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 |