CN117556197B - Typhoon vortex initialization method based on artificial intelligence - Google Patents
Typhoon vortex initialization method based on artificial intelligence Download PDFInfo
- Publication number
- CN117556197B CN117556197B CN202410038922.5A CN202410038922A CN117556197B CN 117556197 B CN117556197 B CN 117556197B CN 202410038922 A CN202410038922 A CN 202410038922A CN 117556197 B CN117556197 B CN 117556197B
- Authority
- CN
- China
- Prior art keywords
- data
- typhoon
- typhoons
- meteorological
- model
- 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
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 22
- 238000011423 initialization method Methods 0.000 title claims abstract description 17
- 238000000034 method Methods 0.000 claims abstract description 186
- 238000013136 deep learning model Methods 0.000 claims abstract description 79
- 238000012549 training Methods 0.000 claims abstract description 56
- 230000006399 behavior Effects 0.000 claims abstract description 23
- 230000009467 reduction Effects 0.000 claims abstract description 20
- 238000013480 data collection Methods 0.000 claims abstract description 11
- 230000008602 contraction Effects 0.000 claims abstract description 8
- 238000010801 machine learning Methods 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 7
- 238000003909 pattern recognition Methods 0.000 claims abstract description 7
- 238000013499 data model Methods 0.000 claims abstract description 5
- 230000008569 process Effects 0.000 claims description 84
- 238000004458 analytical method Methods 0.000 claims description 59
- 230000006870 function Effects 0.000 claims description 56
- 230000008859 change Effects 0.000 claims description 52
- 238000004364 calculation method Methods 0.000 claims description 30
- 238000012545 processing Methods 0.000 claims description 30
- 238000006243 chemical reaction Methods 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 21
- 238000010606 normalization Methods 0.000 claims description 21
- 238000011161 development Methods 0.000 claims description 18
- 230000018109 developmental process Effects 0.000 claims description 18
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 18
- 238000005070 sampling Methods 0.000 claims description 16
- 239000013535 sea water Substances 0.000 claims description 15
- 238000005516 engineering process Methods 0.000 claims description 14
- 230000008520 organization Effects 0.000 claims description 13
- 230000011218 segmentation Effects 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 11
- 238000001556 precipitation Methods 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 9
- 238000003908 quality control method Methods 0.000 claims description 9
- 230000002159 abnormal effect Effects 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 230000002829 reductive effect Effects 0.000 claims description 7
- 230000001932 seasonal effect Effects 0.000 claims description 7
- 101100366707 Arabidopsis thaliana SSL11 gene Proteins 0.000 claims description 6
- 101100366710 Arabidopsis thaliana SSL12 gene Proteins 0.000 claims description 6
- 230000003542 behavioural effect Effects 0.000 claims description 6
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 230000007774 longterm Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000005728 strengthening Methods 0.000 claims description 6
- 101100366711 Arabidopsis thaliana SSL13 gene Proteins 0.000 claims description 5
- 101100366561 Panax ginseng SS11 gene Proteins 0.000 claims description 5
- 230000009286 beneficial effect Effects 0.000 claims description 5
- 230000010354 integration Effects 0.000 claims description 5
- 230000007613 environmental effect Effects 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 101100366562 Panax ginseng SS12 gene Proteins 0.000 claims description 3
- 101100366563 Panax ginseng SS13 gene Proteins 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 238000013178 mathematical model Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 230000036961 partial effect Effects 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 3
- BULVZWIRKLYCBC-UHFFFAOYSA-N phorate Chemical compound CCOP(=S)(OCC)SCSCC BULVZWIRKLYCBC-UHFFFAOYSA-N 0.000 claims description 3
- 230000001737 promoting effect Effects 0.000 claims description 3
- 238000000547 structure data Methods 0.000 claims description 3
- 239000000758 substrate Substances 0.000 claims description 3
- 238000012731 temporal analysis Methods 0.000 claims description 3
- 238000000700 time series analysis Methods 0.000 claims description 3
- 238000004800 variational method Methods 0.000 claims description 3
- JTJMJGYZQZDUJJ-UHFFFAOYSA-N phencyclidine Chemical class C1CCCCN1C1(C=2C=CC=CC=2)CCCCC1 JTJMJGYZQZDUJJ-UHFFFAOYSA-N 0.000 claims 3
- 238000012216 screening Methods 0.000 claims 2
- 230000005856 abnormality Effects 0.000 claims 1
- 230000002708 enhancing effect Effects 0.000 claims 1
- 229920006395 saturated elastomer Polymers 0.000 claims 1
- 230000002265 prevention Effects 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 239000013589 supplement Substances 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000002401 inhibitory effect Effects 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001066 destructive effect Effects 0.000 description 1
- 238000012407 engineering method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
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
Abstract
The invention relates to a typhoon vortex initialization method based on artificial intelligence, which provides more comprehensive typhoon behavior information by combining various meteorological data sources, preprocesses data and converts the data into a data set suitable for a machine learning model, performs feature learning and pattern recognition on the data by applying a deep learning model comprising a contraction path and an expansion path and jump connection, extracts key features and patterns of typhoons, constructs typhoons vortex structures and predicts typhoons paths and strength by applying a trained deep learning model to real-time meteorological data, and updates and improves the prediction capability of the deep learning model by continuous data collection and model training. The method can realize more accurate typhoon vortex initialization, remarkably improve the accuracy and efficiency of typhoon path and strength prediction, and provide effective technical support for typhoon early warning and disaster prevention and reduction.
Description
Technical Field
The invention belongs to the technical field of weather forecast, relates to a typhoon vortex initializing method and technical improvement, in particular to an artificial intelligence-based typhoon vortex initializing method, aims at improving accuracy and efficiency of typhoon path and strength prediction, and provides scientific basis for disaster prevention and alleviation.
Background
Typhoons are powerful rotational storms formed under specific meteorological conditions mainly on tropical or subtropical oceans, have extremely high wind speeds and precipitation, are an extreme weather phenomenon, and have profound effects on global society, economy and people's life. The unpredictability and great destructive power of typhoons make accurate typhoons prediction an urgent and important task in meteorology. In order to realize typhoon prediction, the interaction of a plurality of systems such as ocean, atmosphere, land and the like needs to be deeply analyzed and simulated, which involves a large number of nonlinear, dynamic, random and uncertain factors, and high requirements are placed on the processing of data and the establishment of a model. Traditional typhoon prediction methods rely on classical meteorological models and limited observed data, but exhibit significant limitations in processing large amounts of complex data, particularly in typhoon vortex initialization.
Typhoon vortex initialization is a key process in meteorology and is the basis for realizing accurate weather forecast, and particularly in the prediction of complex weather phenomena such as typhoons. The process involves developing typhoons in a meteorological model by using observation data and a mode forecasting field under given boundary conditions through algorithms such as data assimilation and the like so as to acquire an initial state of relatively reasonable typhoons and promote the representativeness of typhoons initial vortex. This process is critical for accurately predicting typhoons paths and intensities, as the initial state of typhoons can affect the subsequent simulation results. However, due to limitations of observed data and defects of a model, the conventional typhoon vortex initialization method often cannot accurately reflect the actual strength and path of typhoons during initialization. In simulation, due to the limitation of model resolution and processing capacity, the initial vortex of typhoons is still inaccurate, so that a large strength forecasting error is caused, and the overall reliability of weather forecasting is affected.
At present, the common typhoon and typhoon vortex initialization method mainly comprises the following steps: one is the BVM method (Bogus Vortex Method) which artificially inserts a hypothetical typhoon vortex in a pattern forecasting farm to simulate the presence of typhoons, and then coordinates them with the surrounding environment through pattern balance adjustment. The method has the advantages of simplicity and easiness, and has the defects that parameters such as the position, the shape, the strength and the like of the vortex are required to be set manually, and the consistency with actual observation data is poor. And the BDAM method (Bogus Data Assimilation Method) is used for correcting vortex by using observation data on the basis of BVM so as to enable the vortex to be more approximate to a real typhoon state. The method has the advantages that the observed data can be utilized, and the defects that the quality and the quantity of the observed data are insufficient, and error propagation and nonlinearity problems exist in the data assimilation process. There is also the EnKF method (Ensemble Kalman Filter Method), which is a data assimilation method based on Monte Carlo (Monte Carlo) technology, that reflects the uncertainty of typhoons by generating a set of initial states, and then updating the set with observed data to obtain a set of posterior states. The method has the advantages of taking errors of observation and modes into consideration, and has the disadvantages of large calculation amount and high requirements on the distribution and quality of observation data.
However, the above-described conventional typhoon vortex initialization method has many problems in processing observed data. Insufficient observation data, data quality problems and simplified assumptions of the model often lead to large deviations of the predicted results from the actual situation. More importantly, these methods often fail to fully utilize the large amount of weather data available, particularly pattern forecast and analytical data, thereby limiting the performance and accuracy of vortex initialization. This limitation is mainly due to the deficiencies of the existing methods in terms of data processing and integration. For example, conventional typhoon vortex initialization techniques often have difficulty effectively integrating weather data from different sources and different types, such as satellite remote sensing data, weather radar data, ground observation station data, and the like. The variability in spatial and temporal resolution, type of parameters, etc. of these data makes their comprehensive application a complex challenge. In addition, due to technical limitations, conventional methods also exhibit limitations in processing large-scale, high-dimensional data, and it is difficult to implement deep mining and efficient application of the data.
In summary, typhoon prediction is an important and complex task in meteorology, and the accuracy of typhoon prediction directly influences disaster prevention and treatment. Traditional typhoon vortex initialization methods rely on limited observation data and empirical models, which greatly limit the accuracy of predictions. With the development of artificial intelligence technology, the possibility of improving typhoon predictions through more complex data analysis and pattern recognition has emerged. In order to overcome the limitation of the traditional method in terms of data utilization and feature recognition, the method improves typhoon vortex initialization by means of integrated innovation technology so as to improve accuracy and efficiency of typhoon prediction, and the method becomes a technical problem to be solved in the current meteorological field.
Disclosure of Invention
Object of the invention
In view of the above background and existing problems, especially the limitations of observed data in the typhoon vortex initializing process, the shortages caused by model simplifying assumptions, and the limitations of the traditional typhoon vortex initializing method in terms of data utilization and feature recognition, the invention aims to provide an artificial intelligence-based typhoon vortex initializing method. The method aims to overcome the limitation of the traditional method in the aspects of processing complex meteorological data and realizing accurate typhoon vortex initialization by combining a deep learning model with a large number of mode re-analysis and forecast data. According to the invention, through integrating various meteorological data sources such as analysis data, typhoon analysis data, mode forecast data, typhoon optimal path data and the like, a more comprehensive typhoon behavior view angle is provided, and the quality and accuracy of input data of a prediction model are effectively improved. Meanwhile, the invention can accurately identify and analyze different modes and characteristics of typhoons by applying the deep learning model, and provides more accurate information for typhoons vortex initialization. By combining multi-source data integration and advanced deep learning technology, the accuracy and efficiency of typhoon path and strength prediction are obviously improved, and the method has important practical application value for disaster prevention and alleviation.
(II) technical scheme
In order to achieve the aim of the invention and solve the technical problems, the invention adopts the following technical scheme:
an artificial intelligence-based typhoon vortex initialization method is used for initializing typhoon vortex and is characterized by at least comprising the following steps:
SS1 Multi-Source Meteorological data Collection
Collecting and analyzing various different types of meteorological data of various meteorological data sources to acquire historical and real-time behavioral information of typhoons, wherein the various different types of meteorological data at least comprise analysis data, typhoon mode forecast data and typhoon optimal path data, the analysis data is a past meteorological condition data set reconstructed by combining historical meteorological observation data with a meteorological model, the typhoon analysis data is meteorological data formed after power initialization and assimilation observation are carried out on typhoon events, the typhoon mode forecast data is typhoon future behavioral meteorological data generated based on the meteorological model forecast, and the typhoon optimal path data is typhoon actual moving path data obtained after various information is integrated;
SS2 Multi-Source Meteorological data pretreatment
Preprocessing the multisource meteorological data collected in step SS1 to convert into a data set suitable for a machine learning model, wherein the data interpolation process estimates the value of the point to be interpolated by considering the distance between the point to be interpolated and the known data point to fill the data blank or convert the data to uniform resolution, the data normalization process at least comprises unit conversion, scaling to uniform interval and normalization process to convert the data to uniform format and scale, and the feature engineering at least comprises basic meteorological feature extraction, time sequence feature extraction, advanced feature construction, feature selection and data dimension reduction to extract and select the features most useful for typhoon prediction tasks from the original data;
SS3 deep learning model training
Performing feature learning and pattern recognition on the data set preprocessed in the step SS2 to extract key features and patterns of typhoons, and optimizing performance and accuracy of the deep learning model through a training and verifying process, wherein the deep learning model is used for capturing context information and accurately positioning typhoons and at least comprises a contracted path and a symmetrical expanded path, the contracted path is used for extracting advanced features in data and comprises a plurality of convolution layers and a pooling layer, the expanded path is used for recovering detailed information in the data and comprises a plurality of convolution layers and an upsampling layer, jump connection is further arranged between the convolution layers and the upsampling layer, and is used for connecting data between low-level features in the contracted path and the advanced features in the expanded path, and the training and verifying process is used for training and verifying the model by using the segmented data set so as to evaluate accuracy and generalization capability of the model and ensure effectiveness of the model in practical application;
SS4 typhoon vortex initialization
Applying the deep learning model trained in the step SS3 to real-time meteorological data to construct a typhoon vortex structure and predicting typhoon paths and strength, wherein the typhoon vortex structure is a three-dimensional structure at least comprising the shape, the size, the strength and wind field distribution information of typhoons, and the predicted typhoon paths and strength are calculated according to the typhoon vortex structure and the meteorological model, and the moving direction, the speed and the change trend of typhoons in a certain time in the future;
SS5 continuous learning and optimization model
The prediction capability of the deep learning model is continuously updated and improved through continuous data collection and model training, wherein the data collection comprises the steps of acquiring latest meteorological data and typhoon data in real time and feeding back the difference between a prediction result and an actual observation result of the deep learning model, and the model training comprises the steps of adjusting parameters and structures of the deep learning model according to the change of the data and the fed back error so as to adapt to the continuously changed meteorological conditions and continuously improve the prediction accuracy.
In a preferred embodiment of the present invention, in the step SS1, the analysis data is a past weather condition data set reconstructed by combining historical weather observation data with a weather model, the past weather condition data set being used to reflect a historical weather environment of typhoons occurring and developing, provide a comprehensive view angle for historical behavior of typhoons, and provide background information for typhoon vortex initialization, and the generation of the analysis data includes at least the following sub-steps:
SS11. Collecting and organizing historical weather observation data comprising at least weather data covering a wide time and space range obtained by ground, ship, sounding and/or satellite weather observation means and comprising a plurality of weather elements including at least wind speed, air pressure, temperature and/or humidity and typhoon information comprising at least path, intensity and/or wind field of typhoons, the organizing of the historical weather observation data comprising quality control and format conversion of the data for subsequent processing and analysis;
SS12, selecting a proper weather model, interpolating and normalizing historical weather observation data in the step SS11 according to the input requirement of the weather model to ensure the consistency and the integrity of the data, and simultaneously taking the space-time resolution and the coverage range of the data into consideration to meet the precision and the stability of the model and then inputting the data into the weather model to generate model prediction data, wherein the weather model is a mathematical model which is established based on a physical equation and a numerical method and is used for simulating and predicting weather phenomena, and comprises a global weather model, a regional weather model or a numerical weather forecast model;
SS13, fusion and optimal integration are performed on the historical meteorological observation data input in step SS12 and the model prediction data output by the meteorological model by using a data assimilation technology, so as to reduce errors between the observation data and the model prediction data and improve the credibility and representativeness of the data, and meanwhile, the dynamic and physical significance of the data are maintained so as to reflect the characteristics and behaviors of typhoons, wherein the data assimilation technology is a variational method, a kalman filtering method or an integrated kalman filtering method;
SS14. Outputting the results of step SS13 to obtain the analysis data, i.e., the reconstructed past meteorological conditions data set, which provides a continuous and consistent reconstruction of past meteorological conditions, which in turn serves as an important input for typhoon vortex initialization, provides the necessary historical meteorological information for deep learning model training, and enhances the accuracy of model to typhoon behavior prediction.
Further, in the above substep SS11, the arrangement of the historical meteorological observation data includes performing quality control and format conversion on the data, where the quality control of the data includes checking whether there is a missing, abnormal, erroneous and/or inconsistent problem on the data, and correcting or rejecting the problem data; the data is subjected to format conversion, including converting the data into a unified data format and scale suitable for a machine learning model, and performing unit conversion, scaling to a unified interval and/or standardization processing on the data so as to ensure the consistency and the integrity of the data.
In a preferred embodiment of the present invention, in step SS1, the typhoon analysis data is weather data formed after power initialization and assimilation observation for typhoon events, and the typhoon analysis data at least includes:
the method comprises the steps that typhoons path data, namely longitude and latitude coordinates of a typhoons center at different time points, are used for describing movement tracks and speeds of typhoons and the relation between the typhoons and geographic positions, and are obtained by interpolation and smoothing processing of typhoons optimal path data provided by each national grade weather bureau and/or international weather organization or by satellite images and ground observation results;
the method comprises the steps that the strength data of typhoons, namely the lowest air pressure and the highest air speed of a typhoon center at different time points, are used for describing the strength grade and the change trend of typhoons and the relation with meteorological conditions, and are obtained by estimating and correcting typhoons best path data provided by various national grade meteorological authorities and/or international meteorological organizations or by using satellite images and ground observation results;
the three-dimensional structure data of typhoons, namely wind speed, wind direction, temperature, humidity and/or air pressure information at different radiuses and heights in the typhoons influence range, is obtained by taking analysis data as a background field, assimilating satellite and ground observation data and carrying out vortex initialization on the basis of the analysis data to obtain three-dimensional data which can more represent the vortex structure of typhoons, and the vortex initialization adopts a meteorological model to integrate from the previous moment of analysis moment to the analysis moment, then transplants the vortex of typhoons to the previous moment, and repeats the steps until the vortex intensity is close to the optimal path intensity.
In a preferred embodiment of the present invention, in step SS1, the typhoon pattern prediction data is typhoon future behavior weather data generated based on weather model prediction, and the typhoon pattern prediction data includes:
typhoon path forecast data, namely data for predicting the moving direction and speed of a typhoon center according to a meteorological model, are used for describing the future moving track and position of typhoons and the relation between the typhoons and geographic positions; the typhoon path forecast data is generated from typhoon mode forecast data provided by each national grade weather bureau and/or international weather organization or by utilizing an independently developed numerical weather forecast model;
typhoon intensity forecast data, namely, data for forecasting the lowest air pressure and the maximum air speed of a typhoon center according to a meteorological model, are used for describing the future intensity level and change trend of typhoons and the relation with meteorological conditions; the typhoon intensity forecast data is generated from typhoon mode forecast data provided by each national grade weather bureau and/or international weather organization or by utilizing an independently developed numerical weather forecast model;
typhoon wind field forecast data, namely, data for forecasting wind speeds and wind directions at different radiuses and heights near the typhoon center according to a meteorological model, are used for describing future structure and power characteristics of typhoons and the relation between the typhoons and an environmental flow field; the typhoon wind field forecast data is generated from typhoon mode forecast data provided by each national grade weather bureau and/or international weather organization or by utilizing an independently developed numerical weather forecast model;
Typhoon rainfall forecast data, namely data for predicting the rainfall and the rainfall types in different areas and at different heights near the center of typhoons according to a meteorological model, are used for describing the future water vapor transmission and rainfall distribution characteristics of typhoons and the relationship with thermodynamic conditions;
the typhoon path forecast data, typhoon intensity forecast data, typhoon wind field forecast data and typhoon precipitation forecast data are generated from typhoon mode forecast data provided by each national weather office and/or international weather organization or by utilizing an independently developed numerical weather forecast model.
In a preferred embodiment of the present invention, in the step SS1, the typhoon optimal path data is typhoon actual moving path data obtained by integrating various information, and the typhoon actual path data is obtained by summarizing and analyzing information recorded by weather observation stations, satellite observations and other related technical means, and provides accurate information about the typhoon actual moving track, including detailed analysis of path change, moving speed and deviation from a prediction model.
In a preferred embodiment of the present invention, in the step SS2, the data interpolation process estimates the value of the point to be interpolated by considering the distance between the point to be interpolated and the known data point to fill the data blank or convert the data into a uniform resolution, wherein the data interpolation process uses the Cressman spatial interpolation method, and the method performs interpolation calculation according to the following formula:
In the method, in the process of the invention,V(P) Representing points to be interpolatedPIs a function of the number of (c),V(P i ) Representing known data pointsP i Is a function of the number of (c),W(d i ) The representation is based on distanced i And is the distanced i Is a decreasing function of (a),d i representing points to be interpolatedPWith known data pointsP i The distance between the two plates is set to be equal,nrepresenting the number of data points involved in interpolation.
In a preferred embodiment of the present invention, in the step SS2, the data interpolation process further includes performing a weighted average of space and time on the data to consider the space-time correlation and reliability of the data and to improve the continuity and representativeness of the data, and performing a data interpolation and a weighted average calculation of space and time on the data according to the following formula:
in the method, in the process of the invention,V(P) Representing points to be interpolatedPIs a function of the number of (c),V(P i ) Representing known data pointsP i Is a function of the number of (c),W s (d i ) The representation is based on distanced i Is a function of the spatial weight of (c) the (c),W t (d i ) The representation is based on timet i Is a function of the time weight of (c),d i representing points to be interpolatedPWith known data pointsP i The spatial distance between the two substrates is set to be equal,t i representing points to be interpolatedPWith known data pointsP i The time difference between the two times of the two,nrepresenting the number of data points involved in interpolation.
Further, the spatial weight functionW s (d i ) And a time weighting functionW t (d i ) Expressed in the form of the following gaussian function:
in the method, in the process of the invention,σ s andσ t standard deviation of space and time are respectively represented for controlling the weight decay rate of space and time respectively.
In a preferred embodiment of the present invention, in the step SS2, the data normalization process at least includes unit conversion, scaling to a unified range, and normalization process to convert the data into a consistent format and scale, and at least includes the following sub-steps:
SS21. Converting the data into units, converting the meteorological data of different units into unified standard units, so as to eliminate the influence of unit difference on model training;
SS22. Scaling the data to a unified interval, scaling all meteorological data types to a unified interval range of 0 to 1 or-1 to 1 according to a certain scale to eliminate the influence of different measurement range differences on model training, and scaling the data using a min-max normalization method shown in the following formula:
in the method, in the process of the invention,Xas the raw data is to be processed,X norm in order to normalize the data, the data is,X min andX max respectively minimum and maximum values of the same type of data in the data set;
SS23, carrying out standardization processing on the data, converting the meteorological data into a standard format with a mean value of 0 and a standard deviation of 1 so as to eliminate the influence of data distribution difference on model training, wherein the standardization formula is as follows:
in the method, in the process of the invention,Xas the raw weather data, the weather data,μas the mean value of the data,δas the standard deviation of the data, ZIs the standardized meteorological data.
In a preferred embodiment of the present invention, in the step SS2, the feature engineering includes basic weather feature extraction, time-series feature extraction, advanced feature construction, feature selection and data dimension reduction to extract and select features most useful for typhoon prediction tasks from raw data, and at least includes the following sub-steps:
SS2A. Extracting basic meteorological features of data, wherein the basic meteorological features are indexes reflecting typhoon development information and at least comprise Vertical Wind Shear (VWS), conditional Ocean Heat Content (COHC), 200 hundred Pa divergence (D200), relative Humidity (RHMD), sea level air pressure change (dMSLP) and/or typhoon moving Speed (SPD);
SS2B. extracting time series features of the data, extracting trend analysis and seasonal patterns from the time series of meteorological data, including analyzing long-term temperature and/or barometric pressure trend and identifying seasonal precipitation patterns and/or wind direction changes in the data, by which time series analysis long-term and periodic changes related to typhoon development are captured;
SS2C, performing advanced feature construction on the data, comparing and constructing advanced features based on the historical contemporaneous data to highlight the comparison and abnormal features of the current meteorological conditions and the historical modes, and extracting the abnormal or remarkable features of the current data through comparison analysis;
SS2D, selecting the data, at least comprising correlation analysis and feature importance assessment, wherein the most relevant features are screened out by analyzing the correlation between each feature and typhoon intensity and path prediction, and the most helpful features are screened out by assessing the contribution degree of each feature to the prediction model;
SS2E, judging whether to perform data dimension reduction on the data according to the number of the features, if the number of the features is larger than a preset threshold, performing data dimension reduction on the data, removing redundant and irrelevant features in the data through data dimension reduction processing, and reserving the features which are most important for prediction, so that the number and complexity of the features are reduced, a model is simplified, and the calculation load is reduced.
Further, the above sub-step SS2A is during the process of extracting the basic meteorological features of the data:
the Vertical Wind Shear (VWS) refers to the direction and speed change of wind speed at different heights, and its calculation formula is:
in the method, in the process of the invention,u 850 andv 850 respectively representing east-west and north-south wind speeds at 850 hundred Pa altitude,u 200 andv 200 representing east-west and north-south wind speeds at 200 hundred pascals altitude, respectively, and wherein lower vertical wind shear favors typhoon strength and development;
The Conditional Ocean Heat Content (COHC) refers to the heat stored in the ocean, and the high ocean heat content is beneficial to providing more energy for typhoons and promoting the strengthening of typhoons, and the calculation formula is as follows:
in the method, in the process of the invention,c p is the specific heat capacity of the seawater,ρis the density of the seawater, and is the density of the seawater,Tis the temperature of the seawater, and is the temperature of the seawater,T 26 the isotherm depth is 26C,D 26 is the minimum of 26 ℃ isotherm depth and seabed depth;
the 200 hundred Pa divergence (D200) refers to the horizontal wind field divergence condition with the height of about 12 km in the atmosphere and corresponding to the air pressure of 200 hundred Pa, and the calculation formula is as follows:
in the method, in the process of the invention,uandvrespectively denote east-west wind speed and south-north wind speed,xandyrespectively longitude and latitude, positive divergence, i.e. the air flow is in a divergent state and is related to the intensification of the tropical cyclone, as it contributes to the outflow of air from the upper layer of the tropical cyclone and brings more upward air flow and energy to the cyclone center, negative divergence, i.e. the air flow is in an aggregated state, possibly inhibiting the development of the tropical cyclone;
the Relative Humidity (RHMD) refers to the actual temperatureTSaturated water vapor pressure and dew point temperatureT d The following saturated water vapor pressure ratio has the following calculation formula:
in the method, in the process of the invention,e sat (Td) Is the saturated water vapor pressure at the dew point temperature, e sat (T) The distribution and the change of the water vapor content in the atmosphere are provided for the saturated water vapor pressure at the actual temperature, and the change of the Relative Humidity (RHMD) has important influence on accurately predicting the precipitation potential and typhoon intensity change;
the sea level air pressure change (dMSLP) refers to the change rate of the sea level air pressure, the change rate is expressed by the difference of the sea level air pressure values (MSLP) of the front time and the back time, so as to reflect the change of the sea level air pressure at different time points, and the calculation formula is as follows:
in the method, in the process of the invention,MSLP(t1)、MSLP(t2) Respectively the first timet1. Second time of dayt2, sea level air pressure value, sea level air pressure change (dMSLP) is one of key indexes for monitoring typhoons, and the reduction of the sea level air pressure indicates the strengthening of typhoons;
the typhoon moving Speed (SPD) refers to the average moving speed of a typhoon center in a certain time, and relates to a path prediction and a possibly affected area thereof, the typhoon moving speed is calculated by dividing the changing distance of the typhoon center position between two time points by a time interval to obtain accurate moving speed estimation, and the calculation formula is as follows:
in the method, in the process of the invention,x 1 ,y 1 andx 2 ,y 2 representing the geographical coordinates, delta, of the typhoon centre at two different points in time, respectively tIs the time difference between these two time points.
In a preferred embodiment of the present invention, in the step SS3, the deep learning model is a deep learning model of the uiet based on a Convolutional Neural Network (CNN), the deep learning model of the uiet has a U-shaped structure for extracting and fusing key features and patterns in meteorological data and includes a contracted path and a symmetrical extended path, wherein the contracted path is composed of a plurality of convolutional layers and pooled layers, and is used for capturing context information and performing downsampling; the expansion path consists of a plurality of convolution layers and an up-sampling layer and is used for recovering detail information and up-sampling; and jump connection is further arranged between the contracted path and the expanded path and used for splicing the low-level features and the high-level features so as to enhance the expression capacity of the features.
Further, the Unet deep learning model learns and extracts key features and patterns of typhoons, including at least shape, size, intensity and movement pattern of typhoons, based on the input meteorological data and through a multi-layered structure thereof, wherein,
for the shape of typhoons, the Unet deep learning model captures and reconstructs the spatial structure of typhoons through convolution operation of a contracted path and an expanded path and combination of jump connection and an up-sampling layer, and at least comprises the steps of identifying and extracting the eye wall, spiral rain belt characteristics and changes thereof at different time points of typhoons, so that the outline and the boundary of typhoons are identified and the shape information of typhoons is obtained;
For the typhoon size, the Unet deep learning model adjusts the typhoon resolution through the feature extraction operation of the convolution layers of the contracted path and the expanded path and the downsampling and upsampling operation of the pooling layer and the upsampling layer, and at least comprises the steps of analyzing typhoon images on a plurality of scales to measure the spatial coverage range and the boundary of the typhoon images, calculating the geometric properties of the typhoon and the changes of the typhoon at different time points, so as to obtain the typhoon size and area information;
for the strength of typhoons, the Unet deep learning model fuses the characteristics of different layers through jump connection operation of a contracted path and an expanded path, and obtains the strength information of typhoons at least comprising central air pressure, maximum air speed and maximum wind radius of the typhoons center and surrounding areas by comprehensively analyzing various meteorological parameters at least comprising air speed, air pressure and temperature and dynamic factors at least comprising ocean heat content and vertical wind shear;
and for the movement mode of typhoons, the Unet deep learning model analyzes the movement trend and direction of typhoons through the characteristic extraction operation of the convolution layers of the contracted path and the expanded path, and at least comprises the steps of predicting the potential movement direction and speed of typhoons by analyzing the space-time characteristics and the atmospheric circulation mode of meteorological data and combining pressure field data so as to obtain the movement mode information of typhoons.
Further, the process of training the Unet deep learning model involves adjusting weights of the network to minimize the difference between the predicted output and the actual result, and the model training process at least comprises the following sub-steps:
SS31, initializing weights: when training is started, randomly initializing weights in a Unet deep learning model network to break the symmetry of the network, wherein the weights represent the connection strength between layers of the network and determine the initial state and convergence speed of the network;
SS32. Forward propagation: giving a group of input meteorological data and corresponding typhoon segmentation labels, performing feature extraction and segmentation on the input data by a network through a multi-layer structure comprising a contraction path and an expansion path to obtain a predicted output of the network, performing forward propagation on the input meteorological data through the network in each step of training, multiplying the data in each layer by weight and adding bias, and performing nonlinear conversion through an activation function, wherein the nonlinear conversion comprises transmission through a convolution layer, a pooling layer and an up-sampling layer to generate the predicted output of typhoon features;
SS33, calculate loss: calculating a loss function value of the network based on the predicted output and the actual result of the network to reflect the segmentation error of the network, the loss function being defined using a mean square error, a cross entropy loss, or a Dice coefficient;
SS34. Counter propagation: calculating the gradient of each layer in the network according to the loss function value, namely the partial derivative of the loss function on each layer weight to reflect the influence of the weight change on the loss function, and calculating the contribution degree of each weight on the final loss based on the gradient of the loss function and a back propagation algorithm, wherein the process starts from an output layer and propagates back to an input layer by layer;
SS35, weight update: according to the calculated gradient of back propagation and combining with the learning rate, updating and adjusting the weight of each layer in the network by using a random gradient descent or Adam optimization algorithm, so that the weight changes along the opposite direction of the gradient, thereby reducing the loss function value and further reducing the difference between the prediction output and the actual result;
SS36, iterative process: the process of forward propagation, loss calculation, back propagation and weight update is repeated until the loss function value converges such that the performance of the model on the training data reaches an acceptable level or reaches a preset number of iterations.
(III) technical effects
Compared with the prior art, the typhoon vortex initialization method based on artificial intelligence has the following beneficial and remarkable technical effects:
(1) The invention provides a more comprehensive typhoon behavior view angle by combining multisource meteorological data, including analysis data, typhoon mode forecast data, typhoon optimal path data and the like, and the comprehensive data utilization mode remarkably improves the accuracy and reliability of typhoon prediction. The invention not only can utilize historical and real-time observation data, but also can combine the prediction data of the mode and the best path data, thereby better reflecting the actual condition and future trend of typhoons and providing a more reliable data basis for typhoons vortex initialization.
(2) The data interpolation, normalization and characteristic engineering method adopted in the data preprocessing stage provides high-quality data input for training of the deep learning model. In particular, the Cressman interpolation method and the data normalization step ensure that data of different sources and resolutions can be effectively integrated and normalized. Compared with the prior art, the method can solve the problems of data blank, inconsistent data, data redundancy and the like, converts the data into uniform resolution, format and scale through data interpolation and data standardization, and extracts and selects the most useful features for typhoon prediction tasks from the original data through feature engineering, so that the dimension and complexity of the data are reduced, and the effectiveness and usability of the data are improved.
(3) The invention applies the deep learning model to perform feature learning and pattern recognition on the data set so as to extract key features and patterns of typhoons, and optimizes the performance and accuracy of the deep learning model through training and verification processes. Compared with the prior art, the method can automatically learn and extract the advanced features and modes of typhoons from complex meteorological data by utilizing the nonlinear fitting and self-adaptive learning capabilities of the deep learning model without manually setting parameters and thresholds, thereby overcoming the limitation of the traditional method in the aspects of processing the complex meteorological data and realizing accurate typhoon vortex initialization.
(4) The invention further adopts the Unet model as a deep learning model, and the model has a symmetrical U-shaped structure, comprises a contracted path, an expanded path and jump connection, and can simultaneously utilize the position information of a lower layer and the semantic information of a higher layer, thereby improving the precision and the robustness of segmentation. The Unet model, in its U-shaped configuration, is excellent in image and data processing of complex structures, in particular in feature extraction and pattern recognition. The model is capable of efficiently learning and extracting key typhoon features, such as shape, size, intensity and movement patterns, from meteorological data. The Unet provides higher prediction accuracy and better generalization capability than existing prediction models.
(5) The invention constructs the typhoon vortex structure and predicts the typhoon path and strength by applying the trained deep learning model to real-time meteorological data. Compared with the prior art, the method can realize the construction of the three-dimensional structure of the typhoon vortex, including the shape, the size, the intensity and the wind field distribution information of typhoons, so that the dynamic characteristics and the change rules of typhoons are reflected better, and more accurate information is provided for the prediction of typhoons paths and intensities. According to the typhoon vortex structure and the meteorological model, the movement direction, speed and change trend of typhoons in a certain time in the future can be calculated, so that the prediction of typhoons paths and intensities is realized, and effective technical support is provided for typhoon early warning and disaster prevention and reduction.
(6) The invention continuously updates and improves the prediction capability of the deep learning model through continuous data collection and model training. Compared with the prior art, the method can realize continuous learning and optimization of the deep learning model, and the latest meteorological data and typhoon data are obtained in real time, and the difference between the predicted result and the actual observed result of the deep learning model is fed back, so that the parameters and the structure of the deep learning model are adjusted according to the data change and the feedback error, the continuous weather condition is adapted, and the prediction accuracy is continuously improved.
Drawings
FIG. 1 is a schematic diagram of an implementation flow of an artificial intelligence based typhoon vortex initialization method of the present invention;
FIG. 2 is a schematic diagram showing the process of generating analysis data according to the present invention;
FIG. 3 is a schematic diagram of a data normalization process according to the present invention;
FIG. 4 is a schematic diagram of a feature engineering process according to the present invention;
fig. 5 is a schematic diagram of a training process of a deep learning model of the invention.
Detailed Description
For a better understanding of the present invention, the following examples are set forth to illustrate the present invention. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of the invention. The embodiments described below by referring to the drawings are illustrative and intended to explain the present invention and should not be construed as limiting the invention. 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. The following describes the structure and technical scheme of the present invention in detail with reference to the accompanying drawings, and an embodiment of the present invention is given.
Example 1
As shown in fig. 1, the typhoon vortex initialization method based on artificial intelligence of the invention mainly comprises the following steps when in implementation:
SS1 Multi-Source Meteorological data Collection
Collecting and analyzing various different types of meteorological data of various meteorological data sources to acquire historical and real-time behavioral information of typhoons, wherein the various different types of meteorological data at least comprise analysis data, typhoon mode forecast data and typhoon optimal path data, the analysis data is a past meteorological condition data set reconstructed by combining historical meteorological observation data with a meteorological model, the typhoon analysis data is the meteorological data formed after power initialization and assimilation observation are carried out on typhoon events, the typhoon mode forecast data is typhoon future behavioral meteorological data generated based on meteorological model forecast, and the typhoon optimal path data is typhoon actual moving path data obtained after various information is integrated;
SS2 Multi-Source Meteorological data pretreatment
Preprocessing the multisource meteorological data collected in the step SS1, wherein the preprocessing at least comprises data interpolation, data normalization and feature engineering to convert the multisource meteorological data into a data set suitable for a machine learning model, the data interpolation processing is used for estimating the numerical value of a point to be interpolated by considering the distance between the point to be interpolated and a known data point to fill a data blank or converting the data into uniform resolution, the data normalization processing at least comprises unit conversion, scaling to uniform intervals and normalization processing to convert the data into uniform formats and scales, and the feature engineering at least comprises basic meteorological feature extraction, time sequence feature extraction, advanced feature construction, feature selection and data dimension reduction to select features which are most useful for typhoon prediction tasks from original data;
SS3 deep learning model training
Performing feature learning and pattern recognition on the data set preprocessed in the step SS2 to extract key features and patterns of typhoons, and optimizing performance and accuracy of the deep learning model through a training and verifying process, wherein the deep learning model is used for capturing context information and accurately positioning typhoons and at least comprises a contracted path and a symmetrical expanded path, the contracted path is used for extracting high-level features in the data and comprises a plurality of convolution layers and a pooling layer, the expanded path is used for recovering detailed information in the data and comprises a plurality of convolution layers and an up-sampling layer, jump connection is further arranged between the convolution layers and the up-sampling layer, the data connection between low-level features in the contracted path and high-level features in the expanded path is used for performing model training and verification by using the segmented data set, so that accuracy and generalization capability of the model are evaluated, and effectiveness of the model in practical application is ensured;
SS4 typhoon vortex initialization
Applying the deep learning model trained in the step SS3 to real-time meteorological data to construct a typhoon vortex structure and predicting typhoon paths and strength, wherein the typhoon vortex structure is a three-dimensional structure at least comprising the shape, the size, the strength and wind field distribution information of typhoons, and the step of predicting the typhoon paths and the strength is to calculate the moving direction, the speed and the change trend of typhoons in a certain future time according to the typhoon vortex structure and the meteorological model;
SS5 continuous learning and optimization model
The prediction capability of the deep learning model is continuously updated and improved through continuous data collection and model training, wherein the data collection comprises the steps of acquiring latest meteorological data and typhoon data in real time and feeding back the difference between the prediction result and the actual observation result of the deep learning model, and the model training comprises the steps of adjusting parameters and structures of the deep learning model according to the change of the data and the fed back error so as to adapt to the continuously changed meteorological conditions and continuously improve the prediction accuracy.
In the embodiment of the invention, by combining the multi-source meteorological data and the deep learning technology, more accurate typhoon vortex initialization can be realized. The method fully utilizes the modern data analysis technology, remarkably improves the accuracy and efficiency of typhoon path and strength prediction, and provides effective technical support for typhoon early warning and disaster prevention and reduction.
Example 2
On the basis of the above-described embodiment 1, this embodiment serves as a further supplement to step SS1 therein.
In the typhoon vortex initialization method based on artificial intelligence, in step SS1, analysis data is a past meteorological condition data set reconstructed by combining historical meteorological observation data with a meteorological model, wherein the past meteorological condition data set is used for reflecting the historical meteorological environment of typhoon generation and development, providing a comprehensive view angle for the historical behavior of typhoon and background information for typhoon vortex initialization, and the generation of the analysis data is shown in fig. 2:
And SS11. Collecting and arranging historical meteorological observation data, wherein the historical meteorological observation data at least comprise meteorological data which cover a wide time and space range and comprise various meteorological elements and typhoon information, the meteorological elements comprise wind speed, air pressure, temperature and/or humidity, the typhoon information at least comprise the path, the intensity and/or the wind field of typhoons, and the arrangement of the historical meteorological observation data comprises quality control and format conversion of the data for subsequent processing and analysis.
In this sub-step, the sorting of the historical meteorological observation data includes quality control and format conversion of the data, wherein the quality control of the data includes checking whether there are missing, abnormal, erroneous and/or inconsistent problems with the data, and correcting or rejecting the problem data; the data is subjected to format conversion, including converting the data into a unified data format and scale suitable for a machine learning model, and performing unit conversion, scaling to a unified interval and/or standardization processing on the data so as to ensure the consistency and the integrity of the data.
SS12, selecting a proper weather model, interpolating and normalizing historical weather observation data in the step SS11 according to the input requirement of the weather model to ensure the consistency and the integrity of the data, and simultaneously taking the space-time resolution and the coverage range of the data into consideration to meet the precision and the stability of the model and then inputting the data into the weather model to generate model prediction data, wherein the weather model refers to a mathematical model which is established based on a physical equation and a numerical method and is used for simulating and predicting weather phenomena, and comprises a global weather model, a regional weather model or a numerical weather forecast model;
SS13, fusion and optimal integration are performed on the historical meteorological observation data input in step SS12 and the model prediction data output by the meteorological model by using a data assimilation technology, so as to reduce errors between the observation data and the model prediction data and improve the credibility and representativeness of the data, and meanwhile, the dynamic and physical significance of the data are maintained so as to reflect the characteristics and behaviors of typhoons, wherein the data assimilation technology is a variational method, a kalman filtering method or an integrated kalman filtering method;
and SS14. Outputting the result in the step SS13 to obtain analysis data, namely a reconstructed past meteorological condition data set, wherein the past meteorological condition data set provides continuous and consistent reconstruction of past meteorological conditions and then is used as an important input for typhoon vortex initialization, necessary historical meteorological information is provided for deep learning model training, and the accuracy of model on typhoon behavior prediction is enhanced.
Through the steps, the method can fully utilize the historical meteorological data and combine the modern meteorological model technology to provide comprehensive and deep historical background and environmental information for typhoon vortex initialization. This not only helps to understand the historical behavior patterns of typhoons in depth, but also provides a solid data base for future typhoons predictions and studies.
In a preferred embodiment of the present invention, in the step SS1, the typhoon analysis data is weather data formed after power initialization and assimilation observation for typhoon events, and the typhoon analysis data at least includes:
the method comprises the steps that typhoons path data, namely longitude and latitude coordinates of a typhoons center at different time points, are used for describing movement tracks and speeds of typhoons and the relation between the typhoons and geographic positions, and are obtained by interpolation and smoothing processing of typhoons optimal path data provided by each national weather office and/or international weather organization or by satellite images and ground observation results;
the strength data of typhoons, namely the lowest air pressure and the maximum air speed of a typhoon center at different time points, are used for describing the strength grade and the change trend of typhoons and the relation with meteorological conditions, and are obtained by estimating and correcting typhoons best path data provided by each national grade meteorological office and/or international meteorological organization or by using satellite images and ground observation results;
the three-dimensional structure data of typhoons, namely wind speed, wind direction, temperature, humidity and/or air pressure information at different radiuses and heights in the typhoons influence range, is obtained by taking analysis data as a background field, assimilating satellite and ground observation data and carrying out vortex initialization on the basis of the analysis data to obtain three-dimensional data which can more represent the vortex structure of typhoons, and the vortex initialization adopts a meteorological model to integrate from the previous moment of analysis moment to the analysis moment, then transplants the vortex of typhoons to the previous moment, and repeats the steps until the vortex intensity is close to the optimal path intensity. The key role of typhoon analysis data is to provide detailed and real-time information about specific typhoon events, which is critical for a deep understanding of typhoons dynamics and behavior patterns. By analyzing these data, it is possible to obtain deep insight concerning various aspects such as typhoon path change, intensity fluctuation, wind speed distribution, and precipitation pattern. Not only does this information help to reveal the mechanisms of typhoons formation and development, but it also provides a direct and specific input for the training of typhoons predictive models. In the present invention, typhoon analysis data is used as one of key inputs for typhoon vortex initialization. By integrating these real-time and detailed typhoon observation data into the deep learning model, the accuracy and reliability of the model to typhoon behavior prediction can be significantly improved. In addition, the application of typhoon analysis data is also beneficial to the model to better adapt to different typhoon events and enhance the prediction capability of the typhoon analysis data under diversified meteorological conditions.
In a preferred embodiment of the present invention, in step SS1, the typhoon pattern prediction data is typhoon future behavior weather data generated based on weather model prediction, and the typhoon pattern prediction data includes:
typhoon path forecast data, namely data for predicting the moving direction and speed of a typhoon center according to a meteorological model, are used for describing the future moving track and position of typhoons and the relation between the typhoons and geographic positions; typhoon path forecast data are generated from typhoon pattern forecast data provided by various national weather authorities and/or international weather organizations or by using an independently developed numerical weather forecast model;
typhoon intensity forecast data, namely, data for forecasting the lowest air pressure and the maximum air speed of a typhoon center according to a meteorological model, are used for describing the future intensity level and change trend of typhoons and the relation with meteorological conditions; typhoon intensity forecast data are generated from typhoon pattern forecast data provided by various national weather authorities and/or international weather organizations or by using an independently developed numerical weather forecast model;
typhoon wind field forecast data, namely, data for forecasting wind speeds and wind directions at different radiuses and heights near the typhoon center according to a meteorological model, are used for describing future structure and power characteristics of typhoons and the relation between the typhoons and an environmental flow field; typhoon wind field forecast data are generated from typhoon mode forecast data provided by various national grade weather authorities and/or international weather organizations or by using an independently developed numerical weather forecast model;
Typhoon rainfall forecast data, namely data for predicting the rainfall and the rainfall types in different areas and at different heights near the center of typhoons according to a meteorological model, are used for describing the future water vapor transmission and rainfall distribution characteristics of typhoons and the relationship with thermodynamic conditions;
the typhoon path forecast data, typhoon intensity forecast data, typhoon wind field forecast data and typhoon precipitation forecast data are generated from typhoon mode forecast data provided by each national weather office and/or international weather organization or by using an independently developed numerical weather forecast model.
The key role of typhoon mode forecast data is to provide prediction of future typhoon behaviors, and provide important basis for understanding potential development trend of typhoons. By analysing these forecast data, the predictive performance of the different meteorological models, in particular the accuracy in predicting typhoon path and intensity, can be assessed and compared. This is critical to optimizing and calibrating artificial intelligence models, as it can reveal the behavior of the model under specific meteorological conditions, guiding the improvement and tuning of the model. In the invention, typhoon mode forecast data is not only used as an important input for training a deep learning model, but also used for verifying and optimizing a model forecast result. By integrating the prediction data based on the advanced numerical model into the deep learning model, the accuracy of the model for predicting the future behavior of typhoons can be remarkably improved, and the model is helped to adapt and respond to continuously-changing meteorological conditions better, so that the overall performance and efficiency of typhoons prediction and vortex initialization are enhanced.
In a preferred embodiment of the present invention, in the step SS1, the typhoon optimal path data is typhoon actual moving path data obtained by integrating various information, and the typhoon actual path data is obtained by summarizing and analyzing information recorded by weather observation stations, satellite observations and other related technical means, and provides accurate information about the typhoon actual moving track, including detailed analysis of path change, moving speed and deviation from a prediction model.
Typhoon best path data is critical to evaluate accuracy of weather prediction and effect of deep learning model. It can be used not only to verify the accuracy of the predictive model, but also to help identify possible defects or shortfalls of the model, guiding further optimization and tuning of the model. In the context of typhoon prediction, best path data is a key resource for understanding typhoon behavior and improving predictive models. In the present invention, typhoon best path data is used as an important reference for deep learning model training and verification. By analyzing the data, the model can learn the behavior mode of typhoons more accurately, and the accuracy of future typhoons path and strength prediction is improved. In addition, the data can also be used for evaluating the performance of the model in practical application, so that the model can effectively operate under the practical meteorological conditions.
Example 3
On the basis of the above-described embodiment 1, this embodiment serves as a further supplement to step SS2 therein.
In the typhoon vortex initialization method based on artificial intelligence, in step SS2, the data interpolation process estimates the numerical value of a point to be interpolated by considering the distance between the point to be interpolated and a known data point to fill in a data blank or convert the data into uniform resolution, wherein the data interpolation process adopts a Cressman spatial interpolation method, and the method carries out interpolation calculation according to the following formula:
in the method, in the process of the invention,V(P) Representing points to be interpolatedPIs a function of the number of (c),V(P i ) Representing known data pointsP i Is a function of the number of (c),W(d i ) The representation is based on distanced i And is the distanced i Is a decreasing function of (a),d i representing points to be interpolatedPWith known data pointsP i The distance between the two plates is set to be equal,nrepresenting the number of data points involved in interpolation.
Since the analytical meteorological data may come from different modes, they may differ in spatial and temporal resolution. Data interpolation is used to fill in missing values in the dataset or to convert the data to a uniform resolution. Interpolation of data is a key step in processing meteorological data, particularly when the data is from different models and has different spatial and temporal resolutions. According to the method, meteorological data with different sources and resolutions can be effectively unified into the same space frame by the Cressman interpolation method, and a continuous and complete data set is provided for further analysis and model training.
In a preferred embodiment of the present invention, in the step SS2, the data interpolation process further includes performing a weighted average of space and time on the data to consider the space-time correlation and reliability of the data and to improve the continuity and representativeness of the data, and performing a data interpolation and a weighted average calculation of space and time on the data according to the following formula:
in the method, in the process of the invention,V(P) Representing points to be interpolatedPIs a function of the number of (c),V(P i ) Representing known data pointsP i Is a function of the number of (c),W s (d i ) The representation is based on distanced i Is a function of the spatial weight of (c) the (c),W t (d i ) The representation is based on timet i Is a function of the time weight of (c),d i representing points to be interpolatedPWith known data pointsP i The spatial distance between the two substrates is set to be equal,t i representing points to be interpolatedPWith known data pointsP i The time difference between the two times of the two,nrepresenting the number of data points involved in interpolation.
Further, a spatial weighting functionW s (d i ) And a time weighting functionW t (d i ) Expressed in the form of the following gaussian function:
in the method, in the process of the invention,σ s andσ t standard deviation of space and time are respectively represented for controlling the weight decay rate of space and time respectively.
In the invention, the data points with shorter distance or time have larger weight and the data points with longer distance or time have smaller weight through the Gaussian function, thereby reflecting the space-time correlation and the credibility of the data and processing the data interpolation more smoothly, especially in the region with larger change in space and time. The space Gaussian weight function considers the geographical position relation among the data points, the time Gaussian weight function considers the timeliness and the relativity of the data, and by the method, the complementary information of the multi-source meteorological data can be effectively utilized, a reasonable weight is distributed to each point to be interpolated, so that a smoother and accurate interpolation result is obtained, and high-quality training data is provided for a deep learning model.
In a preferred embodiment of the present invention, in the step SS2, the data normalization process at least includes unit conversion, scaling to a uniform interval, and normalization process to convert the data into a uniform format and scale, and at least includes the following sub-steps, as shown in fig. 3:
SS21. Converting the data into units, converting the meteorological data of different units into unified standard units, so as to eliminate the influence of unit difference on model training;
SS22. Scaling the data to a unified interval, scaling all meteorological data types to a unified interval range of 0 to 1 or-1 to 1 according to a certain scale to eliminate the influence of different measurement range differences on model training, and scaling the data using a min-max normalization method shown in the following formula:
in the method, in the process of the invention,Xas the raw data is to be processed,X norm in order to normalize the data, the data is,X min andX max respectively minimum and maximum values of the same type of data in the data set;
SS23, carrying out standardization processing on the data, converting the meteorological data into a standard format with a mean value of 0 and a standard deviation of 1 so as to eliminate the influence of data distribution difference on model training, wherein the standardization formula is as follows:
in the method, in the process of the invention,Xas the raw weather data, the weather data,μas the mean value of the data, δAs the standard deviation of the data,Zis the standardized meteorological data.
Through the normalization step, the invention can ensure that the input data received by the model has comparability among different characteristics, thereby improving the performance and accuracy of the machine learning model in the training and prediction process. Normalization is particularly important for processing large-scale and complex meteorological data sets.
In a preferred embodiment of the present invention, in the above step SS2, the feature engineering includes basic weather feature extraction, time-series feature extraction, advanced feature construction, feature selection, and data dimension reduction to extract and select features most useful for typhoon prediction tasks from raw data, as shown in fig. 4:
SS2A. Extracting basic meteorological features of the data, wherein the basic meteorological features are indexes reflecting typhoon development information and at least comprise Vertical Wind Shear (VWS), conditional Ocean Heat Content (COHC), 200 hundred Pa divergence (D200), relative Humidity (RHMD), sea level air pressure change (dMSLP) and/or typhoon moving Speed (SPD);
SS2B. extracting time series features of the data, extracting trend analysis and seasonal patterns from the time series of meteorological data, including analyzing long-term temperature and/or barometric pressure trend and identifying seasonal precipitation patterns and/or wind direction changes in the data, by which time series analysis long-term and periodic changes related to typhoon development are captured;
SS2C, performing advanced feature construction on the data, comparing and constructing advanced features based on the historical contemporaneous data to highlight the comparison and abnormal features of the current meteorological conditions and the historical modes, and extracting the abnormal or remarkable features of the current data through comparison analysis;
SS2D, selecting the data, at least comprising correlation analysis and feature importance assessment, wherein the most relevant features are screened out by analyzing the correlation between each feature and typhoon intensity and path prediction, and the most helpful features are screened out by assessing the contribution degree of each feature to the prediction model;
SS2E, judging whether to perform data dimension reduction on the data according to the number of the features, if the number of the features is larger than a preset threshold, performing data dimension reduction on the data, removing redundant and irrelevant features in the data through data dimension reduction processing, and reserving the features which are most important for prediction, so that the number and complexity of the features are reduced, a model is simplified, and the calculation load is reduced.
Feature engineering is the process of extracting and selecting features from raw data that are most useful for predictive tasks. This may include creating new features (e.g., extracting trends and seasonal patterns from time series data) or selecting the most relevant feature subset. In the context of typhoon prediction, valid features may include wind speed, barometric pressure changes, temperature gradients, etc., which are important indicators of typhoon strength and path. Through the preprocessing steps, the quality and the applicability of a data set can be ensured, and a solid foundation is laid for subsequent deep learning model training and analysis. Through implementation of the sub-steps of the feature engineering, the feature set which is most critical to typhoon prediction can be effectively extracted and constructed from the original meteorological data, so that the performance and accuracy of the deep learning model in typhoon vortex initialization are greatly improved.
Further, the above sub-step SS2A is during the process of extracting the basic meteorological features of the data:
vertical Wind Shear (VWS) refers to the direction and speed change of wind speed at different heights, calculated as:
in the method, in the process of the invention,u 850 andv 850 respectively representing east-west and north-south wind speeds at 850 hundred Pa altitude,u 200 andv 200 representing east-west and north-south wind speeds at 200 hundred pascals altitude, respectively, and wherein lower vertical wind shear favors typhoon strength and development;
the Conditional Ocean Heat Content (COHC) refers to the heat stored in the ocean, and a high ocean heat content is beneficial to providing more energy for typhoons and promoting the strengthening of typhoons, and the calculation formula is as follows:
in the method, in the process of the invention,c p is the specific heat capacity of the seawater,ρis the density of the seawater, and is the density of the seawater,Tis the temperature of the seawater, and is the temperature of the seawater,T 26 the isotherm depth is 26C,D 26 is the minimum of 26 ℃ isotherm depth and seabed depth;
200 hundred pascals divergence (D200) refers to the horizontal wind field divergence at a height of about 12 kilometers in the atmosphere and corresponding to 200 hundred pascals air pressure, and the calculation formula is:
in the method, in the process of the invention,uandvrespectively denote east-west wind speed and south-north wind speed,xandyrespectively longitude and latitude, positive divergence, i.e. the air flow is in a divergent state and is related to the intensification of the tropical cyclone, as it contributes to the outflow of air from the upper layer of the tropical cyclone and brings more upward air flow and energy to the cyclone center, negative divergence, i.e. the air flow is in an aggregated state, possibly inhibiting the development of the tropical cyclone;
Relative Humidity (RHMD) refers to the actual temperatureTSaturated water vapor pressure and dew point temperatureT d The following saturated water vapor pressure ratio has the following calculation formula:
in the method, in the process of the invention,e sat (Td) Is the saturated water vapor pressure at the dew point temperature,e sat (T) The distribution and the change of the water vapor content in the atmosphere are provided for the saturated water vapor pressure and the change of the Relative Humidity (RHMD) at the actual temperature, and the method has important influence on accurately predicting the precipitation potential and typhoon intensity change;
the sea level air pressure change (dMSLP) refers to the change rate of the sea level air pressure, the change rate is expressed by the difference of the sea level air pressure values (MSLP) of the front time and the back time, so as to reflect the change of the sea level air pressure at different time points, and the calculation formula is as follows:
in the method, in the process of the invention,MSLP(t1)、MSLP(t2) Respectively the first timet1. Second time of dayt2, sea level air pressure value, sea level air pressure change (dMSLP) is one of key indexes for monitoring typhoons, and the reduction of the sea level air pressure indicates the strengthening of typhoons;
typhoon moving Speed (SPD) refers to the average moving speed of a typhoon center in a certain time, and relates to the path prediction and possibly affected area of the typhoon center, the typhoon moving speed is calculated by dividing the changing distance of the typhoon center position between two time points by the time interval to obtain accurate moving speed estimation, and the calculation formula is as follows:
In the method, in the process of the invention,x 1 ,y 1 andx 2 ,y 2 representing the geographical coordinates, delta, of the typhoon centre at two different points in time, respectivelytIs the time difference between these two time points.
Example 4
On the basis of the above-described embodiment 1, this embodiment serves as a further supplement to step SS3 therein.
In the step SS3, the deep learning model is a Unet deep learning model based on a Convolutional Neural Network (CNN), and the Unet deep learning model is provided with a U-shaped structure for extracting and fusing key features and modes in meteorological data and comprises a contraction path and a symmetrical expansion path, wherein the contraction path consists of a plurality of convolution layers and pooling layers and is used for capturing context information and performing downsampling; the expansion path consists of a plurality of convolution layers and an up-sampling layer and is used for recovering detail information and up-sampling; and a jump connection is further arranged between the contracted path and the expanded path and used for splicing the low-level features and the high-level features so as to enhance the expression capability of the features.
The Unet model used in the invention is a deep learning model based on a Convolutional Neural Network (CNN), and can extract the characteristics of typhoon such as shape, size, strength, wind field distribution and the like from meteorological data and output the three-dimensional structure of typhoon. The core idea of the Unet model is to symmetrically connect the contracted path and the expanded path by utilizing a U-shaped structure, so as to realize fusion and segmentation of the multi-scale features.
The shrink path is made up of multiple convolution layers and pooling layers for capturing context information and downsampling. After each downsampling, the number of feature maps increases, but the resolution decreases, which helps to extract higher-level semantic features, but some detail and location information is lost. The last layer of the shrink path is a bottleneck layer that contains the features of the highest layer and is also the starting point of the expansion path.
The extended path is composed of a plurality of convolution layers and an up-sampling layer, and is used for recovering detail information and up-sampling. After each upsampling the number of feature maps decreases but the resolution increases, which helps to recover the low-level detail features but lacks some semantic information. Each layer of the expansion path is in jump connection with the corresponding layer of the contraction path, and the position information of the lower layer and the semantic information of the upper layer are spliced, so that the expression capability of the features is enhanced. The last layer of the extended path is an output layer that maps the feature map to the number of target classes using a 1 x 1 convolution, resulting in the final segmentation result.
Further, the Unet deep learning model learns and extracts key features and patterns of typhoons based on the input meteorological data and through a multi-layer structure thereof, the key features and patterns of typhoons including at least a shape, a size, an intensity and a movement pattern of typhoons, wherein,
For the shape of typhoons, the Unet deep learning model captures and reconstructs the spatial structure of typhoons through convolution operation of a contracted path and an expanded path and combination of jump connection and an up-sampling layer, and at least comprises the steps of identifying and extracting the eye wall, spiral rain belt characteristics and changes of the eye wall, the spiral rain belt characteristics at different time points of typhoons, so that the outline and the boundary of typhoons are identified and the shape information of typhoons is obtained;
for the size of typhoons, the Unet deep learning model adjusts the resolution of typhoons through the characteristic extraction operation of a convolution layer of a contraction path and an expansion path and the downsampling and upsampling operation of a pooling layer and an upsampling layer, and at least comprises the steps of analyzing typhoons images on multiple scales to measure the spatial coverage range and the boundary of the typhoons, calculating the geometric properties of the typhoons and the changes of the typhoons at different time points, so as to obtain the size and area information of the typhoons;
for the strength of typhoons, the Unet deep learning model fuses characteristics of different layers through jump connection operation of a contracted path and an expanded path, and obtains strength information of typhoons at least comprising central air pressure, maximum air speed and maximum air radius of typhoons and surrounding areas by comprehensively analyzing various meteorological parameters at least comprising air speed, air pressure and temperature and dynamic factors at least comprising ocean heat content and vertical wind shear;
For the movement mode of typhoons, the Unet deep learning model analyzes the movement trend and direction of typhoons through the characteristic extraction operation of the convolution layers of the contracted path and the expanded path, and at least comprises the steps of predicting the potential movement direction and speed of typhoons by analyzing the space-time characteristics and the atmospheric circulation mode of meteorological data and combining pressure field data, so that movement mode information of typhoons is obtained.
Further, the process of the Unet deep learning model training involves adjusting the weights of the network to minimize the difference between the predicted output and the actual results, the model training process is as shown in FIG. 5:
SS31, initializing weights: when training is started, randomly initializing weights in a Unet deep learning model network to break the symmetry of the network, wherein the weights represent the connection strength among layers of the network and determine the initial state and convergence speed of the network;
SS32. Forward propagation: giving a group of input meteorological data and corresponding typhoon segmentation labels, performing feature extraction and segmentation on the input data by a network through a multi-layer structure comprising a contraction path and an expansion path to obtain a predicted output of the network, performing forward propagation on the input meteorological data through the network in each step of training, multiplying the data in each layer by weight and adding bias, and performing nonlinear conversion through an activation function, wherein the nonlinear conversion comprises transmission through a convolution layer, a pooling layer and an up-sampling layer to generate the predicted output of typhoon features;
SS33, calculate loss: calculating a loss function value of the network based on the predicted output and the actual result of the network to reflect the segmentation error of the network, wherein the loss function is defined by using a mean square error, cross entropy loss or a Dice coefficient;
SS34. Counter propagation: calculating the gradient of each layer in the network according to the loss function value, namely the partial derivative of the loss function on each layer weight to reflect the influence of the weight change on the loss function, and calculating the contribution degree of each weight on the final loss based on the gradient of the loss function and a back propagation algorithm, wherein the process starts from an output layer and propagates back to an input layer by layer;
SS35, weight update: according to the calculated gradient of back propagation and combining with the learning rate, updating and adjusting the weight of each layer in the network by using a random gradient descent or Adam optimization algorithm, so that the weight changes along the opposite direction of the gradient, thereby reducing the loss function value and further reducing the difference between the prediction output and the actual result;
SS36, iterative process: the process of forward propagation, loss calculation, back propagation and weight update is repeated until the loss function value converges such that the performance of the model on the training data reaches an acceptable level or reaches a preset number of iterations.
The object of the present invention is fully effectively achieved by the above-described embodiments. Those skilled in the art will appreciate that the present invention includes, but is not limited to, those illustrated in the drawings and described in the foregoing detailed description. While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
Claims (14)
1. An artificial intelligence-based typhoon vortex initialization method is used for initializing typhoon vortex and is characterized by comprising the following steps of:
SS1 Multi-Source Meteorological data Collection
Collecting and analyzing various different types of meteorological data of various meteorological data sources to acquire historical and real-time behavioral information of typhoons, wherein the various different types of meteorological data comprise analysis data, typhoon mode forecast data and typhoon optimal path data, the analysis data is a past meteorological condition data set reconstructed by combining historical meteorological observation data with a meteorological model, the typhoon analysis data is meteorological data formed after power initialization and assimilation observation are carried out on typhoon events, the typhoon mode forecast data is typhoon future behavioral meteorological data generated based on meteorological model forecast, and the typhoon optimal path data is typhoon actual moving path data obtained after various information is integrated;
SS2 Multi-Source Meteorological data pretreatment
Preprocessing the multisource meteorological data collected in step SS1 to convert into a data set suitable for machine learning models, wherein the data interpolation process estimates the value of the points to be interpolated by considering the distance between the points to be interpolated and the known data points to fill the data blank or convert the data to a uniform resolution, the data normalization process comprises unit conversion, scaling to a uniform interval and normalization process to convert the data to a uniform format and scale, and the feature engineering comprises basic meteorological feature extraction, time series feature extraction, advanced feature construction, feature selection and data dimension reduction to extract and select features most useful for typhoon prediction tasks from the original data, and the following sub-steps are included:
SS2A. Extracting basic meteorological features of the data, wherein the basic meteorological features are indexes reflecting typhoon development information and comprise vertical wind shear VWS, conditional ocean heat content COHC, 200 hundred Pa divergence D200, relative humidity RHMD, sea level air pressure change dMSLP and/or typhoon moving speed SPD,
SS2B. extracting time series characteristics of the data, extracting trend analysis and seasonal patterns from the time series of meteorological data, including analyzing long-term temperature and/or barometric pressure trend and identifying seasonal precipitation patterns and/or wind direction changes in the data, by these time series analysis to capture long-term and periodic changes related to typhoon development,
SS2C.A high-level feature construction is performed on the data, high-level features are constructed on the basis of historical contemporaneous data comparison to highlight comparison and abnormal features of current weather conditions and historical patterns, the abnormal or remarkable features of the current data are extracted through comparison analysis,
SS2D, performing feature selection on the data, including correlation analysis and feature importance assessment, screening the most relevant features to the predictions by analyzing the correlation of each feature with typhoon intensity and path predictions, screening the most helpful features to the predictions by assessing the contribution of each feature to the prediction model,
SS2E, judging whether to perform data dimension reduction on the data according to the number of the features, if the number of the features is larger than a preset threshold, performing data dimension reduction on the data, removing redundant and irrelevant features in the data through data dimension reduction processing, and reserving the features which are most important for prediction, so that the number and complexity of the features are reduced, a model is simplified, and the calculation load is reduced;
SS3 deep learning model training
Performing feature learning and pattern recognition on the data set preprocessed in the step SS2 to extract key features and patterns of typhoons, and optimizing performance and accuracy of the deep learning model through a training and verifying process, wherein the deep learning model is used for capturing context information and accurately positioning typhoons and comprises a contracted path and a symmetrical extended path, the contracted path is used for extracting advanced features in data and comprises a plurality of convolution layers and a pooling layer, the extended path is used for recovering detailed information in the data and comprises a plurality of convolution layers and an upsampling layer, jump connection is further arranged between the convolution layers and the upsampling layer and is used for connecting data between low-level features in the contracted path and advanced features in the extended path, and the training and verifying process utilizes the segmented data set to perform model training and verification so as to evaluate accuracy and generalization capability of the model and ensure effectiveness of the model in practical application;
SS4 typhoon vortex initialization
Applying the deep learning model trained in the step SS3 to real-time meteorological data to construct a typhoon vortex structure and predicting typhoon paths and strength, wherein the typhoon vortex structure is a three-dimensional structure comprising the shape, the size, the strength and wind field distribution information of typhoons, and the predicted typhoon paths and strength are calculated according to the typhoon vortex structure and the meteorological model, and the moving direction, the speed and the change trend of typhoons in a certain future time;
SS5 continuous learning and optimization model
Continuously updating and improving the predictive capability of the deep learning model through continuous data collection and model training, wherein the data collection comprises the steps of acquiring latest meteorological data and typhoon data in real time and feeding back the difference between the predicted result and the actual observed result of the deep learning model, and the model training comprises the step of adjusting parameters and structures of the deep learning model according to the change of the data and the fed-back error.
2. The method of initializing typhoon vortex based on artificial intelligence according to claim 1, wherein in the step SS1, the analysis data is a past weather condition data set reconstructed by combining historical weather observation data with a weather model, the past weather condition data set being used to reflect a historical weather environment of typhoons occurrence and development, provide a comprehensive view for historical behavior of typhoons, and provide background information for typhoons power initialization, the generation of the analysis data comprising the sub-steps of:
SS11. Collecting and organizing historical weather observations, including weather data obtained by ground, marine, sounding, and/or satellite weather observations covering a wide time and space range and including various weather elements including wind speed, air pressure, temperature, and/or humidity, and typhoon information including path, intensity, and/or wind farm of typhoons, including quality control and format conversion of the data for subsequent processing and analysis;
SS12, selecting a meteorological model, interpolating and normalizing historical meteorological observation data in the step SS11 according to the input requirement of the meteorological model to ensure the consistency and the integrity of the data, and simultaneously taking the space-time resolution and the coverage range of the data into consideration to meet the precision and the stability of the model and then inputting the data into the meteorological model to generate model prediction data, wherein the meteorological model refers to a mathematical model which is established based on a physical equation and a numerical method and is used for simulating and predicting meteorological phenomena, and comprises a global climate model, a regional climate model or a numerical weather forecast model;
SS13, fusion and optimal integration are performed on the historical meteorological observation data input in step SS12 and the model prediction data output by the meteorological model by using a data assimilation technology, so as to reduce errors between the observation data and the model prediction data and improve the credibility and representativeness of the data, and meanwhile, the dynamic and physical significance of the data are maintained so as to reflect the characteristics and behaviors of typhoons, wherein the data assimilation technology is a variational method, a kalman filtering method or an integrated kalman filtering method;
SS14. Outputting the results of step SS13 to obtain analysis data, i.e., a reconstructed past meteorological condition dataset that provides a continuous and consistent reconstruction of past meteorological conditions, which in turn serves as an important input for typhoon dynamics initialization, providing historical meteorological information for deep learning model training, enhancing model accuracy for typhoon behavior prediction.
3. The method of initializing typhoon vortex based on artificial intelligence according to claim 2, wherein in the substep SS11, the arrangement of the historical meteorological observation data includes performing quality control and format conversion on the data, wherein the quality control on the data includes checking whether there is a problem of missing, abnormality, error and/or inconsistency of the data, and correcting or rejecting the problem data; the data is subjected to format conversion, including converting the data into a unified data format and scale suitable for a machine learning model, and performing unit conversion, scaling to a unified interval and/or standardization processing on the data so as to ensure the consistency and the integrity of the data.
4. The method according to claim 1, wherein in the step SS1, the typhoon analysis data is weather data formed after power initialization and assimilation observation for typhoon events, and the typhoon analysis data includes:
The method comprises the steps that typhoons path data, namely longitude and latitude coordinates of a typhoons center at different time points, are used for describing movement tracks and speeds of typhoons and the relation between the typhoons and geographic positions, and are obtained by interpolation and smoothing processing of typhoons optimal path data provided by each national grade weather bureau and/or international weather organization or by satellite images and ground observation results;
the method comprises the steps that the strength data of typhoons, namely the lowest air pressure and the highest air speed of a typhoon center at different time points, are used for describing the strength grade and the change trend of typhoons and the relation with meteorological conditions, and are obtained by estimating and correcting typhoons best path data provided by various national grade meteorological authorities and/or international meteorological organizations or by using satellite images and ground observation results;
the three-dimensional structure data of typhoons, namely wind speed, wind direction, temperature, humidity and/or air pressure information at different radiuses and heights in the typhoons influence range, is obtained by taking analysis data as a background field, assimilating satellite and ground observation data and carrying out vortex initialization on the basis of the analysis data to obtain three-dimensional data which can more represent the vortex structure of typhoons, and the vortex initialization adopts a meteorological model to integrate from the previous moment of analysis moment to the analysis moment, then transplants the vortex of typhoons to the previous moment, and repeats the steps until the vortex intensity is close to the optimal path intensity.
5. The method according to claim 1, wherein in step SS1, the typhoon pattern prediction data is typhoon future behavior weather data generated based on weather model prediction, and the method comprises:
typhoon path forecast data, namely data for predicting the moving direction and speed of a typhoon center according to a meteorological model, are used for describing the future moving track and position of typhoons and the relation between the typhoons and geographic positions; the typhoon path forecast data is generated from typhoon mode forecast data provided by each national grade weather bureau and/or international weather organization or by utilizing an independently developed numerical weather forecast model;
typhoon intensity forecast data, namely, data for forecasting the lowest air pressure and the maximum air speed of a typhoon center according to a meteorological model, are used for describing the future intensity level and change trend of typhoons and the relation with meteorological conditions; the typhoon intensity forecast data is generated from typhoon mode forecast data provided by each national grade weather bureau and/or international weather organization or by utilizing an independently developed numerical weather forecast model;
typhoon wind field forecast data, namely, data for forecasting wind speeds and wind directions at different radiuses and heights near the typhoon center according to a meteorological model, are used for describing future structure and power characteristics of typhoons and the relation between the typhoons and an environmental flow field; the typhoon wind field forecast data is generated from typhoon mode forecast data provided by each national grade weather bureau and/or international weather organization or by utilizing an independently developed numerical weather forecast model;
Typhoon rainfall forecast data, namely data for predicting the rainfall and the rainfall types in different areas and at different heights near the center of typhoons according to a meteorological model, are used for describing the future water vapor transmission and rainfall distribution characteristics of typhoons and the relationship with thermodynamic conditions;
the typhoon path forecast data, typhoon intensity forecast data, typhoon wind field forecast data and typhoon precipitation forecast data are generated from typhoon mode forecast data provided by each national weather office and/or international weather organization or by utilizing an independently developed numerical weather forecast model.
6. The method according to claim 1, wherein in the step SS1, the typhoon optimal path data is typhoon actual moving path data obtained by integrating various information, and the typhoon actual path data is obtained by summarizing and analyzing information recorded by weather observation stations, satellite observation and other related technical means, and provides accurate information about the typhoon actual moving track, including detailed analysis of path change, moving speed and deviation from a prediction model.
7. The method according to claim 1, wherein in step SS2, the data interpolation process estimates the value of the point to be interpolated by considering the distance between the point to be interpolated and the known data point to fill the data blank or convert the data to a uniform resolution, wherein the data interpolation process uses the Cressman spatial interpolation method, which performs interpolation calculation according to the following formula:
In the method, in the process of the invention,V(P) Representing points to be interpolatedPIs a function of the number of (c),V(P i ) Representing known data pointsP i Is a function of the number of (c),W(d i ) The representation is based on distanced i And is the distanced i Is a decreasing function of (a),d i representing points to be interpolatedPWith known data pointsP i The distance between the two plates is set to be equal,nrepresenting the number of data points involved in interpolation.
8. The typhoon vortex initialization method based on artificial intelligence according to claim 1, wherein in the step SS2, the data interpolation process further comprises performing a weighted average of space and time on the data to consider space-time correlation and reliability of the data and to improve continuity and representativeness of the data, and performing a data interpolation and weighted average calculation of space and time on the data according to the following formula:
in the method, in the process of the invention,V(P) Representing points to be interpolatedPIs a function of the number of (c),V(P i ) Representing known data pointsP i Is a function of the number of (c),W s (d i ) The representation is based on distanced i Is a function of the spatial weight of (c) the (c),W t (d i ) The representation is based on timet i Is a function of the time weight of (c),d i representing points to be interpolatedPWith known data pointsP i The spatial distance between the two substrates is set to be equal,t i representing points to be interpolatedPWith known data pointsP i The time difference between the two times of the two,nrepresenting the number of data points involved in interpolation.
9. The artificial intelligence based typhoon vortex initialization method of claim 8, wherein the spatial weight function W s (d i ) And a time weighting functionW t (d i ) Expressed in the form of the following gaussian function:
in the method, in the process of the invention,σ s andσ t standard deviation of space and time are respectively represented for controlling the weight decay rate of space and time respectively.
10. The method of initializing typhoon vortex based on artificial intelligence according to claim 1, wherein in the step SS2, the data normalization process includes unit conversion, scaling to a uniform interval, and normalization process to convert data into a uniform format and scale, and includes the sub-steps of:
SS21. Converting the data into units, converting the meteorological data of different units into unified standard units, so as to eliminate the influence of unit difference on model training;
SS22. Scaling the data to a unified interval, scaling all meteorological data types to a unified interval range of 0 to 1 or-1 to 1 according to a certain scale to eliminate the influence of different measurement range differences on model training, and scaling the data using a min-max normalization method shown in the following formula:
in the method, in the process of the invention,Xas the raw data is to be processed,X norm in order to normalize the data, the data is,X min andX max respectively minimum and maximum values of the same type of data in the data set;
SS23, carrying out standardization processing on the data, converting the meteorological data into a standard format with a mean value of 0 and a standard deviation of 1 so as to eliminate the influence of data distribution difference on model training, wherein the standardization formula is as follows:
In the method, in the process of the invention,Xas the raw weather data, the weather data,μas the mean value of the data,δas the standard deviation of the data,Zis the standardized meteorological data.
11. The method of initializing typhoon vortex based on artificial intelligence according to claim 1, wherein the substep SS2A is during the basic meteorological feature extraction of data:
the vertical wind shear VWS refers to the direction and speed change of wind speed at different heights, and the calculation formula is as follows:
in the method, in the process of the invention,u 850 andv 850 respectively representing east-west and north-south wind speeds at 850 hundred Pa altitude,u 200 andv 200 representing east-west and north-south wind speeds at 200 hundred pascals altitude, respectively, and wherein lower vertical wind shear favors typhoon strength and development;
the conditional ocean heat content COHC refers to heat stored in the ocean, and the high ocean heat content is beneficial to providing more energy for typhoons and promoting the strengthening of the typhoons, and the calculation formula is as follows:
in the method, in the process of the invention,c p is the specific heat capacity of the seawater,ρis the density of the seawater, and is the density of the seawater,Tis the temperature of the seawater, and is the temperature of the seawater,T 26 the isotherm depth is 26C,D 26 is the minimum of 26 ℃ isotherm depth and seabed depth;
the 200-hundred-Pa divergence D200 refers to a horizontal wind field divergence condition which is at a height of 12 km in the atmosphere and corresponds to 200-hundred-Pa air pressure, and the calculation formula is as follows:
In the method, in the process of the invention,uandvrespectively denote east-west wind speed and south-north wind speed,xandyrespectively, longitude and latitude, positive divergence, i.e., airflow in a divergent state and associated with the enhancement of tropical cyclones, as it aids in the outflow of air from the upper layer of tropical cyclones and brings more upward airflow and energy to the cyclone center, negative divergence, i.e., airflow in an aggregate state, inhibits the development of tropical cyclones;
the relative humidity RHMD refers to the actual temperatureTSaturated water vapor pressure and dew point temperatureT d The following saturated water vapor pressure ratio has the following calculation formula:
in the method, in the process of the invention,e sat (Td) Is the saturated water vapor pressure at the dew point temperature,e sat (T) For saturated water vapour pressure at actual temperature, the variation of the relative humidity RHMD providesThe distribution and the change of the water vapor content in the atmosphere have important influence on accurately predicting the precipitation potential and typhoon intensity change;
the sea level air pressure change dMSLP refers to the change rate of the sea level air pressure, the change rate is expressed by the difference of the sea level air pressure value MSLP of the front time and the back time so as to reflect the change of the sea level air pressure at different time points, and the calculation formula is as follows:
in the method, in the process of the invention,MSLP(t1)、MSLP(t2) Respectively the first timet1. Second time of dayt2, the sea level air pressure value, the sea level air pressure change dMSLP is one of key indexes for monitoring the development of typhoons, and the reduction of the sea level air pressure indicates the strengthening of typhoons;
The typhoon moving speed SPD refers to the average moving speed of a typhoon center in a certain time, relates to a path prediction and a possibly affected area thereof, and calculates the typhoon moving speed by dividing the changing distance of the typhoon center position between two time points by a time interval so as to obtain accurate moving speed estimation, wherein a calculation formula is as follows:
in the method, in the process of the invention,x 1 ,y 1 andx 2 ,y 2 representing the geographical coordinates, delta, of the typhoon centre at two different points in time, respectivelytIs the time difference between these two time points.
12. The method according to claim 1, wherein in step SS3, the deep learning model is a deep learning model of a convolutional neural network CNN, the deep learning model of a U-shape for extracting and merging key features and patterns in meteorological data and comprising a contracted path and a symmetrical expanded path, wherein the contracted path is composed of a plurality of convolutional layers and pooled layers, and is used for capturing context information and downsampling; the expansion path consists of a plurality of convolution layers and an up-sampling layer and is used for recovering detail information and up-sampling; and jump connection is further arranged between the contracted path and the expanded path and used for splicing the low-level features and the high-level features so as to enhance the expression capacity of the features.
13. The method of initializing typhoon vortex based on artificial intelligence according to claim 12, wherein the ultraviolet deep learning model learns and extracts key features and patterns of typhoons, including shape, size, intensity and movement pattern of typhoons, based on the input meteorological data and through a multi-layered structure thereof, wherein,
for the shape of typhoons, the Unet deep learning model captures and reconstructs the spatial structure of typhoons through convolution operation of a contracted path and an expanded path and combination of jump connection and an up-sampling layer, and the method comprises the steps of identifying and extracting the eye wall, spiral rain belt characteristics and changes of the eye wall, the spiral rain belt characteristics at different time points of typhoons, so that the outline and the boundary of typhoons are identified and the shape information of typhoons is obtained;
for the typhoon size, the Unet deep learning model adjusts the typhoon resolution through the feature extraction operation of the convolution layers of the contracted path and the expanded path and the downsampling and upsampling operation of the pooling layer and the upsampling layer, and the method comprises the steps of analyzing typhoon images on multiple scales to measure the spatial coverage and the boundary of the typhoon images, calculating the geometric properties of the typhoon and the changes of the typhoon at different time points, and thus obtaining the typhoon size and area information;
For the strength of typhoons, the Unet deep learning model fuses the characteristics of different layers through jump connection operation of a contracted path and an expanded path, and obtains the strength information comprising central air pressure, maximum air speed and maximum wind radius of typhoons center and surrounding areas by comprehensively analyzing various meteorological parameters comprising air speed, air pressure and temperature and dynamic factors comprising ocean heat content and vertical wind shear;
for the movement mode of typhoons, the Unet deep learning model analyzes the movement trend and direction of typhoons through the characteristic extraction operation of the convolution layers of the contracted path and the expanded path, and comprises the steps of predicting the potential movement direction and speed of typhoons by analyzing the space-time characteristics and the atmospheric circulation mode of meteorological data and combining pressure field data, so that movement mode information of typhoons is obtained.
14. The artificial intelligence based typhoon swirl initialization method of claim 12, wherein the process of the une deep learning model training involves adjusting weights of a network to minimize a difference between a predicted output and an actual result, the model training process comprising the sub-steps of:
SS31, initializing weights: when training is started, randomly initializing weights in a Unet deep learning model network to break the symmetry of the network, wherein the weights represent the connection strength between layers of the network and determine the initial state and convergence speed of the network;
SS32. Forward propagation: giving a group of input meteorological data and corresponding typhoon segmentation labels, performing feature extraction and segmentation on the input data by a network through a multi-layer structure comprising a contraction path and an expansion path to obtain a predicted output of the network, performing forward propagation on the input meteorological data through the network in each step of training, multiplying the data in each layer by weight and adding bias, and performing nonlinear conversion through an activation function, wherein the nonlinear conversion comprises transmission through a convolution layer, a pooling layer and an up-sampling layer to generate the predicted output of typhoon features;
SS33, calculate loss: calculating a loss function value of the network based on the predicted output and the actual result of the network to reflect the segmentation error of the network, the loss function being defined using a mean square error, a cross entropy loss, or a Dice coefficient;
SS34. Counter propagation: calculating the gradient of each layer in the network according to the loss function value, namely the partial derivative of the loss function on each layer weight to reflect the influence of the weight change on the loss function, and calculating the contribution degree of each weight on the final loss based on the gradient of the loss function and a back propagation algorithm, wherein the process starts from an output layer and propagates back to an input layer by layer;
SS35, weight update: according to the calculated gradient of back propagation and combining with the learning rate, updating and adjusting the weight of each layer in the network by using a random gradient descent or Adam optimization algorithm, so that the weight changes along the opposite direction of the gradient, thereby reducing the loss function value and further reducing the difference between the prediction output and the actual result;
SS36, iterative process: the process of forward propagation, loss calculation, back propagation and weight update is repeated until the loss function value converges such that the performance of the model on the training data reaches an acceptable level or reaches a preset number of iterations.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410038922.5A CN117556197B (en) | 2024-01-11 | 2024-01-11 | Typhoon vortex initialization method based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410038922.5A CN117556197B (en) | 2024-01-11 | 2024-01-11 | Typhoon vortex initialization method based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117556197A CN117556197A (en) | 2024-02-13 |
| CN117556197B true CN117556197B (en) | 2024-03-22 |
Family
ID=89813186
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410038922.5A Active CN117556197B (en) | 2024-01-11 | 2024-01-11 | Typhoon vortex initialization method based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117556197B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117809203A (en) * | 2024-02-28 | 2024-04-02 | 南京信息工程大学 | Multi-task continuous learning cross-sea area tropical cyclone strength estimation method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116860824A (en) * | 2023-05-30 | 2023-10-10 | 南京信息工程大学 | Meteorological data knowledge discovery method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114547017B (en) * | 2022-04-27 | 2022-08-05 | 南京信息工程大学 | Meteorological big data fusion method based on deep learning |
-
2024
- 2024-01-11 CN CN202410038922.5A patent/CN117556197B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116860824A (en) * | 2023-05-30 | 2023-10-10 | 南京信息工程大学 | Meteorological data knowledge discovery method and system |
Non-Patent Citations (1)
| Title |
|---|
| 综述预测台风路径的传统模型与神经网络;王旭洋;;科学咨询(科技・管理);20200101(第01期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117556197A (en) | 2024-02-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111815037B (en) | Interpretable short-critical extreme rainfall prediction method based on attention mechanism | |
| Feng et al. | Modeling urban growth with GIS based cellular automata and least squares SVM rules: a case study in Qingpu–Songjiang area of Shanghai, China | |
| Sailor et al. | Climate change implications for wind power resources in the Northwest United States | |
| Olsson et al. | Neural networks for rainfall forecasting by atmospheric downscaling | |
| CN109272721B (en) | Landslide hazard forecasting method based on KPCA-FOA-LSSVM | |
| CN109738970A (en) | The method, apparatus and storage medium for realizing Lightning Warning are excavated based on lightning data | |
| CN108761574A (en) | Rainfall evaluation method based on Multi-source Information Fusion | |
| Ayet et al. | Nowcasting solar irradiance using an analog method and geostationary satellite images | |
| CN117556197B (en) | Typhoon vortex initialization method based on artificial intelligence | |
| CN111665575B (en) | Medium-and-long-term rainfall grading coupling forecasting method and system based on statistical power | |
| Shin et al. | A Bayesian network-based probabilistic framework for drought forecasting and outlook | |
| CN112966722A (en) | Regional landslide susceptibility prediction method based on semi-supervised random forest model | |
| CN115933008A (en) | Strong convection weather forecast early warning method | |
| CN113496310A (en) | Atmospheric pollutant prediction method and system based on deep learning model | |
| CN114385611A (en) | Precipitation prediction method and system based on artificial intelligence algorithm and knowledge graph | |
| CN114882373A (en) | Multi-feature fusion sandstorm prediction method based on deep neural network | |
| KR102411074B1 (en) | Devices and Methods for Detection and Prediction of Cold Water Mass Using Artificial Intelligence and Satellite Imaging | |
| Zhang | Remote sensing data processing of urban land using based on artificial neural network | |
| CN112200355B (en) | Landslide prediction method, landslide prediction device, landslide prediction equipment and storage medium | |
| Mihai et al. | Analyzing meteorological data using unsupervised learning techniques | |
| CN117332909B (en) | Multi-scale urban waterlogging road traffic exposure prediction method based on intelligent agent | |
| Li | Improving Wind Ramp Predictions Using Gabor Filtering and Statistical Scenarios | |
| CN117434624B (en) | Strong convection weather identification and development prejudgment method based on semantic segmentation | |
| KR102589878B1 (en) | System and Method for improving accuracy of satellite based sea wind speed performance using artificial intelligence | |
| Shield | Leveraging “Big Data” to Better Understand and Predict Deep Convection Initiation |
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 |