Typhoon prediction method based on support vector machine
By combining support vector machine models with multi-source meteorological data, the shortcomings of traditional typhoon forecasting methods in terms of initial condition sensitivity and nonlinear feature capture are addressed. This enables high-precision prediction of typhoon formation, path, and intensity, improving forecast accuracy and adaptability, and providing strong support for disaster prevention and mitigation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-04-03
AI Technical Summary
Traditional typhoon forecasting methods are sensitive to initial conditions, making it difficult to accurately predict typhoons in the early stages of their formation. Furthermore, they struggle to capture the nonlinear characteristics of typhoon changes, resulting in insufficient forecasting capabilities.
By employing a support vector machine model and combining historical typhoon data with multi-source meteorological data, a predictive model for typhoon formation, path, and intensity is constructed through data cleaning, normalization, feature extraction, and fusion. Incremental learning is then used to update the model.
It improves the accuracy and reliability of typhoon formation, path, and intensity prediction, has strong generalization ability and adaptability, can adapt to the typhoon prediction needs of different regions and types, and provides comprehensive disaster prevention and mitigation support.
Smart Images
Figure CN121786597A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field, specifically a typhoon prediction method based on support vector machines. Background Technology
[0002] Typhoons, as highly destructive natural disasters, pose a serious threat to the lives and property of people in coastal areas. Traditional typhoon forecasting methods have many limitations. For example, while numerical weather prediction models have high physical realism, they are extremely sensitive to initial conditions and perform poorly in simulating the early stages of typhoon formation. Statistical methods, although capable of rapidly processing large amounts of historical data, struggle to capture the nonlinear characteristics of typhoon changes. Therefore, there is an urgent need for an efficient and accurate typhoon forecasting method to improve the ability to predict typhoon formation, path, and intensity, providing strong support for disaster prevention and mitigation efforts. Summary of the Invention The purpose of this invention is to provide a typhoon forecasting method based on support vector machines to solve the problems mentioned in the background art.
[0003] To achieve the above objectives, the present invention provides the following technical solution, including: S1: Data collection and preprocessing, collecting historical typhoon data and meteorological parameters of the corresponding regions, as well as typhoon path data and multi-source typhoon data; and cleaning, normalizing and extracting features from the collected data, and fusing multi-source data. S2: Feature selection and construction. Preset meteorological parameters are selected as feature variables for typhoon formation prediction. Typhoon formation result labels are constructed to form a training sample set for typhoon formation prediction. Feature selection for typhoon path prediction and typhoon intensity prediction are also performed. S3: Model training. The processed typhoons generate a training sample set, and the radial basis function is selected as the kernel function to construct a typhoon generation and prediction support vector machine classification model. S4: Model application and update. After preprocessing the real-time collected meteorological data, the corresponding feature variables are extracted and input into the trained typhoon generation prediction support vector machine classification model to obtain the typhoon generation prediction results. New typhoon observation data and meteorological data are continuously collected to update the dataset.
[0004] As a further preferred embodiment of this technical solution: the historical typhoon data includes: the location, time, and intensity level of typhoon formation; the preset meteorological parameters include: absolute vorticity, relative humidity, relative vertical velocity, sea surface temperature, and vertical wind shear; the typhoon formation result label X, where X is 1 for formation and -1 for non-formation; the typhoon path data includes: the location coordinates of the typhoon center at different time points, its movement speed and direction, the guiding force of the surrounding environmental airflow, and typhoon intensity data; the multi-source typhoon data includes: satellite cloud images, radiosonde data, and ground meteorological observation data.
[0005] As a further preferred embodiment of this technical solution: the collected data is cleaned, normalized, and feature extracted; multi-source data is fused, including: cleaning the collected data to identify outliers and missing values; for missing values, imputation is performed using the mean, median, mode, or interpolation-based methods; for outliers, they are identified and corrected by setting thresholds, statistical methods, or cluster analysis; the normalization process uses the maximum minus minimum normalization method to map the data to the [0,1] interval, as shown in the formula: ,in For the normalized data, For raw data and These are the maximum and minimum values of the feature data, respectively; a convolutional neural network is used to extract features from the multi-source data, and the extracted deep features are then fused.
[0006] As a further preferred embodiment of this technical solution: the typhoon path prediction feature selection selects the typhoon's current location coordinates, its movement speed and direction over a past period, and the guiding force of the surrounding airflow as feature variables to construct an input feature vector for the support vector regression model; the typhoon intensity prediction feature selection extracts the typhoon's temperature change rate, humidity change rate, and wind speed change rate features from satellite cloud images, as well as radiosonde data and ground meteorological observation data, to construct a comprehensive feature vector.
[0007] As a further preferred embodiment of this technical solution: the function formula in S3: model training is: ,in and For feature vectors, The kernel function parameters control the width of the radial basis functions. The support vector machine (SVM) divides the sample data into two classes by finding the optimal classifying hyperplane; the classification decision function is: ,in The number of support vectors. For Lagrange multipliers, 'b' represents the class label of the support vector, and 'b' represents the bias term.
[0008] As a further preferred embodiment of this technical solution: In step S3, model training divides the training sample set into k subsets, and sequentially... One subset is used as the training set, and the remaining subset is used as the validation set. For each subset combination, a support vector machine model is trained and the classification accuracy on the validation set is calculated.
[0009] As a further preferred embodiment of this technical solution: S3: In model training, based on the time series characteristics of typhoon path data, a time series cross-validation method is adopted to divide the training data into multiple subsets in chronological order, so that the training set contains earlier data and the validation set contains slightly later data; for each validation set, the SVR model is trained and the path prediction error is calculated, and the error indicators include: root mean square error and / or mean absolute error.
[0010] As a further preferred embodiment of this technical solution: S3: In the model training, the prediction of typhoon intensity also adopts the support vector machine regression model, with the input being the fused multimodal deep feature vector and the output being the observed value of typhoon intensity.
[0011] As a further preferred embodiment of this technical solution: In S4: Model application and update, real-time typhoon observation data is input into the trained typhoon path prediction support vector regression model to predict the typhoon's path coordinates over a future period of time. The feature vectors extracted from real-time typhoon multi-source data are input into the trained typhoon intensity prediction support vector regression model to obtain the predicted value of the typhoon intensity.
[0012] As a further preferred embodiment of this technical solution: S4: In the application and updating of the model, new typhoon observation data and meteorological data are continuously collected over time to update the dataset. The support vector machine model is retrained and optimized periodically or when a preset amount of new data is accumulated to adapt to changes in typhoon characteristics and updates in data distribution. An incremental learning method is adopted to update and train the model using new data on the basis of the original model, thereby reducing computational costs. The support vector machine model includes a support vector machine classification model and a support vector regression model.
[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves high-precision prediction of typhoon formation, path and intensity by using an optimized support vector machine model and carefully selected feature variables, significantly improving the accuracy and reliability of typhoon prediction. It adopts a small sample learning method, which can adapt to the prediction needs of different regions and different types of typhoons, and has strong generalization ability and adaptability.
[0014] 2. This invention considers multiple aspects such as typhoon formation, path, and intensity, providing comprehensive forecast information for disaster prevention and mitigation. It helps to formulate more scientific and reasonable prevention measures, effectively processes various meteorological data and satellite image data, makes full use of the advantages of multi-source data, improves the accuracy and stability of forecast results, supports real-time data updates and dynamic model optimization, and can reflect the latest changes in typhoons in a timely manner, ensuring the timeliness and accuracy of forecast results. Attached Figure Description
[0015] Figure 1 This invention relates to a typhoon prediction method based on support vector machines. Figure 2 This invention relates to a typhoon prediction method based on support vector machines. Figure 3 This invention relates to a typhoon prediction method based on support vector machines. Figure 4 This invention relates to a typhoon prediction method based on support vector machines. Figure 5 This invention relates to a typhoon prediction method based on support vector machines. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0017] Example Please see Figures 1-5 As shown, the present invention provides a technical solution including: S1: data collection and preprocessing, collecting historical typhoon data, including the location, time, intensity level of typhoon formation, and meteorological parameters of the corresponding area, such as absolute vorticity, relative humidity, relative vertical velocity, sea surface temperature, and vertical wind shear, as well as collecting typhoon path data and multi-source typhoon data; S1: data collection and preprocessing also includes data cleaning, normalization processing, and feature extraction, and fusing multi-source data; S2: Feature selection and construction. Preset meteorological parameters such as absolute vorticity, relative humidity, relative vertical velocity, sea surface temperature and vertical wind shear are selected as feature variables for typhoon formation prediction. Typhoon formation result label X is constructed. X is 1 if generated and -1 if not generated. Training sample set for typhoon formation prediction is formed. S2: Feature selection and construction also includes feature selection for typhoon path prediction and feature selection for typhoon intensity prediction. S3: Model training. The processed typhoons generate a training sample set, and the radial basis function is selected as the kernel function to construct a typhoon generation and prediction support vector machine classification model. S4: Model application and update. After preprocessing the real-time collected meteorological data, the corresponding feature variables are extracted and input into the trained typhoon generation prediction support vector machine classification model to obtain the typhoon generation prediction result Y. As time goes by, new typhoon observation data and meteorological data are continuously collected to update the dataset. In this embodiment, specifically: S1: Data collection and preprocessing also includes collecting typhoon path data, namely the location coordinates of the typhoon center at different time points, and typhoon intensity data, including typhoon center air pressure, wind speed, etc. In addition, multi-source data of the typhoon are collected, including satellite cloud images, radiosonde data, ground meteorological observation data, etc., to ensure the comprehensiveness and diversity of the data, and to provide a rich information foundation for subsequent feature extraction and model training. In this embodiment, specifically: data cleaning removes outliers and missing values from the data. For missing values, mean imputation, median imputation, mode imputation, or interpolation-based methods can be used to fill in the missing values. For outliers, they can be identified and corrected by setting thresholds, using statistical methods, or cluster analysis. Normalization uses the maximum minus minimum normalization method to map the data to the [0,1] interval, as shown in the formula: ,in For the normalized data, For raw data and The maximum and minimum values of the feature data are respectively identified; a convolutional neural network is used to extract features from the multi-source data, and the extracted deep features are then fused. Feature extraction and fusion can also be performed on satellite cloud images. Deep learning methods such as convolutional neural networks can be used to extract the deep features of the images to capture the morphology and change features of typhoons in satellite images, providing visual information support for subsequent path and intensity prediction. In this embodiment, for multi-source data, such as satellite cloud images, radiosonde data, and ground meteorological observation data, for feature extraction from satellite cloud images, a pre-trained convolutional neural network (such as ResNet, VGG, or Inception) can be selected. The preceding convolutional and pooling layers are retained, while the final fully connected layer is removed. The satellite cloud image is then input to extract feature vectors. For feature extraction from radiosonde and ground meteorological observation data, each feature of the radiosonde and ground meteorological observation data can be embedded into a fixed-dimensional vector space. One-dimensional convolutional layers (1D CNN) are used to extract features from the radiosonde data, and max pooling layers are used to compress the feature maps into fixed-length feature vectors. Subsequent feature fusion can be implemented according to actual needs, through methods such as concatenation (merging all feature vectors into a longer feature vector), weighted summation (assigning a weight to each feature vector and obtaining the fused feature vector through weighted summation), or multimodal fusion networks (constructing a small neural network that takes all feature vectors as input and obtains the fused feature vector through nonlinear transformation; for example, fully connected layers and activation functions (such as ReLU) can be used for feature fusion). I will not go into details here.
[0018] In this embodiment, specifically: the typhoon path prediction feature selection selects the typhoon's current location coordinates, its movement speed and direction over a past period, and the guiding force of the surrounding airflow (e.g., wind field data of the surrounding area, calculating the guiding force of the airflow) and other feature variables to construct the input feature vector for the support vector regression model; the typhoon intensity prediction feature selection extracts the temperature change rate, humidity change rate, and wind speed change rate features of the typhoon from satellite cloud images, as well as radiosonde data and ground meteorological observation data, to construct a comprehensive feature vector; In this embodiment, specifically: S3: Model training divides the training sample set into k subsets, and sequentially... One subset is used as the training set, and the remaining subset is used as the validation set. For each subset combination, a support vector machine model is trained and the classification accuracy on the validation set is calculated. In this embodiment, specifically: S3: During model training, since the typhoon path data has time series characteristics, the time series cross-validation method is adopted to divide the training data into multiple subsets in chronological order. For example, the training set contains early data, and the validation set contains slightly later data. For each validation set, the SVR model is trained and the path prediction error is calculated. Common error indicators include root mean square error and mean absolute error. In this way, the performance of the model can be evaluated and optimized to ensure that the model has a stable classification effect on different data subsets. In this embodiment, typhoon path prediction and typhoon intensity prediction can be performed separately, and the support vector regression model used can be trained separately for each.
[0019] For typhoon track prediction, feature vectors representing the predicted typhoon track characteristics can be constructed, which serve as the input feature vectors for the Support Vector Regression (SVR) model. For the SVR model, a suitable kernel function (e.g., Radial Basis Function (RBF)) can be selected. For each validation set, the corresponding training set is used to train the SVR model, and the corresponding track prediction error is determined to evaluate the model performance. In implementation, the typhoon's current location coordinates (longitude and latitude), its past movement speed and direction, and the guiding force of surrounding air currents can be used as input features, with observed typhoon intensity values (e.g., central pressure, maximum wind speed) as the output target.
[0020] Time-series cross-validation can be used to divide the training data into multiple subsets in chronological order. The training set contains earlier data, and the validation set contains slightly later data. Assume the collected data includes: January 1, 2020: current location coordinates: (120.0, 15.0), speed over the past 3 hours: 10.0 km / h, direction of movement over the past 3 hours: 270 degrees (west), guiding force of surrounding airflow: wind speed 15.0 km / h, wind direction 240 degrees, typhoon intensity observation: central pressure 980 hPa, maximum wind speed 120 km / h; and January 2, 2020: current location coordinates: (119.5, 15.0), speed over the past 3 hours: 10.5 km / h, direction of movement over the past 3 hours: 270 degrees (west), guiding force of surrounding airflow: wind speed 14.5 km / h, wind direction 245 degrees, typhoon intensity observation: central pressure 978 hPa, maximum wind speed 125 km / h. km / h. Assume the period from January 1, 2020 to June 30, 2020 is selected as the training set, and the period from July 1, 2020 to July 31, 2020 is selected as the validation set. During training, the current location coordinates of the typhoon, its movement speed and direction over the past 3 hours, and the guiding force of the surrounding airflow from the training set from January 1, 2020 to June 30, 2020 can be used as input features. The observed typhoon intensity values (central pressure, maximum wind speed) for the corresponding dates from January 1, 2020 to June 30, 2020 can be used as output targets to train the support vector regression model. During the validation process, a pre-trained support vector regression model is used. The current location coordinates of the typhoon, its movement speed and direction over the past 3 hours, and the guiding force of surrounding airflow from July 1st to July 31st, 2020, are used as input features to predict the observed typhoon intensity for the corresponding dates. Based on the actual typhoon intensity observations (central pressure, maximum wind speed) for the corresponding dates from July 1st to July 31st, 2020, error indices (e.g., RMSE and MAE) between the predicted and actual values are determined. This allows for effective training and evaluation of the typhoon track prediction model, ensuring its predictive performance across different time periods.
[0021] In this embodiment, specifically: S3: In the model training, the typhoon intensity prediction also adopts the support vector machine regression model, with the input being the fused multimodal deep feature vector and the output being the observed value of the typhoon intensity; For typhoon intensity prediction, satellite cloud image data, radiosonde data (including temperature, humidity, wind speed, and air pressure), and surface meteorological observation data (including temperature, humidity, wind speed, and air pressure) during the typhoon period can be obtained. The pixel values of the satellite cloud images are normalized to the range of [0, 1] or [-1, 1], and the images are cropped or scaled to a uniform size, such as 224x224. Features such as temperature, humidity, wind speed, and air pressure from the radiosonde and surface meteorological observation data are standardized to a distribution with a mean of 0 and a standard deviation of 1. Features relevant to typhoon intensity prediction are selected, such as the rate of change of temperature, humidity, wind speed, and air pressure. During feature extraction, a pre-trained convolutional neural network (such as ResNet, VGG, or Inception) can be selected. The last fully connected layer is removed, and the preceding convolutional and pooling layers are retained. The satellite cloud image is then used as input to extract feature vectors. For feature extraction from radiosonde and surface meteorological observation data, the rate of change of features such as temperature, humidity, wind speed, and air pressure can be calculated. A time step can be selected (e.g., 3 hours, 6 hours) to calculate the rate of change of these features within each time step. During feature alignment, ensure that all feature vectors have the same dimension. If they are not, a fully connected layer can be used to map the feature vectors to the same dimension. Feature fusion can be performed using methods such as concatenation, weighted summation, or multimodal fusion networks.
[0022] When constructing a Support Vector Regression (SVR) model, a suitable kernel function (e.g., Radial Basis Function (RBF)) can be selected. The fused feature vector is used as input, and the observed typhoon intensity is used as output. The SVR model is trained using training data. The performance of the SVR model for typhoon intensity prediction is evaluated using a validation set. For example, assuming historical data includes typhoon data from January 1, 2020 to December 31, 2020, divided by month, the training set includes typhoon data from January to October 2020, and the validation set includes typhoon data from November and December 2020. Temperature change rate, humidity change rate, and wind speed change rate are extracted from the training set to construct a comprehensive feature vector. The SVR model is trained using the training set data, and the model parameters are adjusted to optimize performance. The model's performance is evaluated using the validation set data, i.e., the root mean square error (RMSE) and mean absolute error (MAE) are calculated, and the model's prediction error is recorded.
[0023] In this embodiment, specifically: S3: The function formula in model training is: ,in and For feature vectors, The kernel function parameter controls the width of the radial basis function. The support vector machine (SVM) divides the sample data into two classes by finding the optimal classifying hyperplane. The classification decision function is: ,in The number of support vectors. For Lagrange multipliers, b is the class label for the support vector, and b is the bias term. A comprehensive feature vector is constructed to capture the key influencing factors of typhoon intensity changes and provide a strong basis for intensity prediction. In this embodiment, specifically: S4: In model application and updating, real-time typhoon observation data is input into the trained typhoon path prediction support vector regression model to predict the typhoon's path coordinates in the future. The feature vectors extracted from real-time typhoon multi-source data are input into the trained typhoon intensity prediction support vector regression model to obtain the predicted value of the typhoon intensity. In this embodiment, specifically: S4: In model application and updating, new typhoon observation data and meteorological data are continuously collected over time to update the dataset. The support vector machine model is retrained and optimized periodically or when there is enough new data to adapt to changes in typhoon characteristics and updates in data distribution. An incremental learning method is adopted to update and train the model using new data on the basis of the original model, thereby reducing computational costs.
[0024] Working principle or structural principle: First, historical typhoon formation data and meteorological parameter data from climate reanalysis datasets are collected. After cleaning and normalizing these data, meteorological parameters such as absolute vorticity, relative humidity, relative vertical velocity, sea surface temperature, and vertical wind shear are selected as feature variables to construct a training sample set for typhoon formation prediction. The generated result is labeled 1, and the ungenerated result is labeled -1. Using this training sample set, a support vector machine classification model is constructed by selecting a radial basis function as the kernel function. By adjusting the kernel function parameters and the regularization parameter C, cross-validation and grid search are used to train and optimize the model. The trained model can accurately predict typhoons. The generation location and number of typhoons are determined. For typhoon path prediction, years of optimal typhoon path data are collected, including the typhoon center coordinates and wind speed, as well as corresponding meteorological background field data. Feature variables such as the typhoon's current position, past movement speed and direction, and the guiding force of surrounding airflow are selected to construct an input feature vector. Support vector regression is used to train the model on the training data, and the model parameters are adjusted to obtain the typhoon path prediction model. Real-time typhoon observation data is input into this model to predict the typhoon's path coordinates over a future period. The model is dynamically adjusted and optimized as new observation data is updated. Regarding typhoon intensity prediction, typhoon... Multi-source data, including satellite cloud images, radiosonde data, and ground meteorological observation data, are used. Convolutional neural networks are employed to extract features from satellite cloud images, obtaining depth features. Simultaneously, key features from radiosonde and ground meteorological observation data, such as variations in temperature, humidity, wind speed, and air pressure, are extracted and fused to construct a comprehensive feature vector. Using this fused feature vector as input and observed typhoon intensity as output, a support vector machine regression model is constructed. The model is trained and optimized using cross-validation and grid search to determine the optimal model parameters. New typhoon data is then processed using the same feature extraction and fusion methods and input into the training model. The trained model yields typhoon intensity predictions, which are then compared with actual observations to assess the model's accuracy and reliability. In practical applications, the typhoon formation prediction model, path prediction model, and intensity prediction model work together to provide comprehensive and accurate information support for typhoon early warning and disaster prevention and mitigation. Before the typhoon season, the typhoon formation prediction model is used to assess the typhoon formation risk in specific sea areas and deploy monitoring resources in advance. After a typhoon forms, the path prediction model tracks its movement trajectory in real time and issues timely path warnings. Simultaneously, the intensity prediction model closely monitors changes in typhoon intensity, providing ample preparation time for super typhoons. This invention presents a typhoon prediction method based on support vector machines (SVMs), capable of simultaneously determining the location and number of typhoons. In future climate change scenarios, it can also directly combine meteorological data under climate change conditions to predict typhoon formation. It provides an effective solution for accurate prediction of typhoon formation, path, and intensity, offering strong technical support for disaster prevention and mitigation efforts and possessing significant practical application value. Regarding typhoon formation prediction, through analysis of historical typhoon data and model training, the method can accurately predict the formation location and number of typhoons 24–72 hours in advance, improving prediction accuracy by 15%–20% compared to traditional methods. In terms of typhoon path prediction, the method updates prediction results in real time; as new observational data is added, the error in path prediction gradually decreases. Through prediction experiments on multiple typhoon cases, the method's path prediction error averages out within 72 hours. The distances are all less than 100 kilometers, reducing the prediction error by 30% to 40% compared to traditional methods. Regarding typhoon intensity prediction, the method of this invention can accurately capture the changing trend of typhoon intensity, predicting the strengthening or weakening of typhoon intensity 12 to 24 hours in advance. Through prediction experiments on multiple typhoon intensity change processes, the correlation coefficient of the method in intensity prediction reached above 0.85, and the root mean square error was less than 15 kt (nautical miles per hour). Furthermore, the method of this invention has good data adaptability and scalability. When facing typhoon data from different sea areas and under different climatic conditions, the method of this invention can achieve good prediction results through appropriate adjustment and optimization of model parameters. At the same time, as new typhoon data is continuously accumulated and updated, the method of this invention can continuously improve prediction accuracy and performance through continuous training and optimization to adapt to the ever-evolving needs of the typhoon prediction field.
[0025] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that it can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0026] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A typhoon forecasting method based on support vector machines, characterized in that, include: S1: Data collection and preprocessing, collecting historical typhoon data and meteorological parameters of the corresponding areas, as well as typhoon path data and multi-source typhoon data; The collected data is cleaned, normalized, and its features are extracted; multi-source data is then fused. S2: Feature selection and construction. Preset meteorological parameters are selected as feature variables for typhoon formation prediction. Typhoon formation result labels are constructed to form a training sample set for typhoon formation prediction. Feature selection for typhoon path prediction and typhoon intensity prediction are also performed. S3: Model training. The processed typhoons generate a training sample set, and the radial basis function is selected as the kernel function to construct a typhoon generation and prediction support vector machine classification model. S4: Model application and update. After preprocessing the real-time collected meteorological data, the corresponding feature variables are extracted and input into the trained typhoon generation prediction support vector machine classification model to obtain the typhoon generation prediction results. New typhoon observation data and meteorological data are continuously collected to update the dataset.
2. The typhoon forecasting method based on support vector machine according to claim 1, characterized in that, The historical typhoon data includes: the location, time, and intensity level of the typhoon formation; The preset meteorological parameters include: absolute vorticity, relative humidity, relative vertical velocity, sea surface temperature, and vertical wind shear. The typhoon generation result label X is 1 if generated and -1 if not generated. The typhoon path data includes: the location coordinates of the typhoon center at different times, its movement speed and direction, the guiding force of the surrounding airflow, and the typhoon intensity data; The multi-source data for the typhoon includes: satellite cloud images, radiosonde data, and ground meteorological observation data.
3. The typhoon forecasting method based on support vector machine according to claim 2, characterized in that, The collected data undergoes cleaning, normalization, and feature extraction; multi-source data is fused, including: Clean the collected data and identify outliers and missing values. For missing values, fill them using the mean, median, mode, or interpolation-based methods; Outliers are identified and corrected by setting thresholds, using statistical methods, or cluster analysis. The normalization process employs the maximum minus minimum normalization method, mapping the data to the [0,1] interval, as shown in the formula: ,in For the normalized data, For raw data and These are the maximum and minimum values of the feature data, respectively. A convolutional neural network is used to extract features from the multi-source data, and the extracted deep features are then fused.
4. The typhoon forecasting method based on support vector machine according to claim 1, characterized in that, In the typhoon path prediction feature selection, the current location coordinates of the typhoon, its movement speed and direction over a past period, and the guiding force of the surrounding airflow are selected as feature variables to construct the input feature vector for the support vector regression model. The typhoon intensity prediction feature selection extracts the temperature change rate, humidity change rate, and wind speed change rate features of typhoons from satellite cloud images, as well as radiosonde data and ground meteorological observation data, to construct a comprehensive feature vector.
5. The typhoon forecasting method based on support vector machine according to claim 1, characterized in that, The function formula in model training, S3, is as follows: ,in and For feature vectors, The kernel function parameter controls the width of the radial basis function. The support vector machine divides the sample data into two classes by finding the optimal classification hyperplane. The classification decision function is: ,in The number of support vectors. For Lagrange multipliers, 'b' represents the class label of the support vector, and 'b' represents the bias term.
6. The typhoon forecasting method based on support vector machine according to claim 1, characterized in that, S3: Model training divides the training sample set into k subsets, and sequentially... One subset is used as the training set, and the remaining subset is used as the validation set. For each subset combination, a support vector machine model is trained and the classification accuracy on the validation set is calculated.
7. The typhoon forecasting method based on support vector machine according to claim 4, characterized in that, S3: In model training, based on the time series characteristics of typhoon path data, a time series cross-validation method is adopted to divide the training data into multiple subsets in chronological order, so that the training set contains earlier data and the validation set contains later data. For each validation set, a support vector regression (SVR) model is trained and the path prediction error is calculated. The error metrics include root mean square error and / or mean absolute error.
8. The typhoon forecasting method based on support vector machine according to claim 4, characterized in that, S3: In the model training, the typhoon intensity prediction adopts a support vector machine regression model. The input is the fused multimodal deep feature vector, and the output is the observed value of the typhoon intensity.
9. The typhoon forecasting method based on support vector machine according to claim 4, characterized in that, In step S4: Model application and update, real-time typhoon observation data is input into the trained typhoon path prediction support vector regression model to predict the typhoon's path coordinates over a future period of time. The feature vectors extracted from real-time typhoon multi-source data are input into the trained typhoon intensity prediction support vector regression model to obtain the predicted typhoon intensity.
10. The typhoon forecasting method based on support vector machine according to claim 1, characterized in that, S4: In model application and updates, new typhoon observation data and meteorological data are continuously collected over time to update the dataset. Periodically or when a preset amount of new data is accumulated, the support vector machine model is retrained and optimized to adapt to changes in typhoon characteristics and updates in data distribution. An incremental learning method is adopted to update and train the model using new data on the basis of the original model, reducing computational costs. The support vector machine model includes a support vector machine classification model and a support vector regression model.