CN114492569A - A Typhoon Track Classification Method Based on Width Learning System - Google Patents

Abstract

The invention discloses a typhoon path classification method based on a width learning system. The invention specifically comprises the following steps: s1, performing cluster analysis on the typhoon path by using the existing typhoon path data through an improved DBSCAN algorithm, and establishing a category label matrix of the typhoon path; s2, performing feature representation on the typhoon paths by using an improved hierarchical clustering algorithm, and unifying dimensions of all the typhoon paths; and S3, constructing a typhoon path classification model based on the width learning system, taking the type label matrix for establishing the typhoon path established in S1 and the typhoon path feature matrix in S2 as input samples, and calculating a weight matrix from system input to output so as to realize the classification of the typhoon paths. The method of the invention adopts the width learning system to classify the typhoon paths, thereby realizing the automatic classification of the typhoon paths.

Description

A Typhoon Track Classification Method Based on Width Learning System

technical field

The invention belongs to the technical field of typhoon path classification, and particularly relates to a typhoon path classification method based on a width learning system.

Background technique

Typhoon is one of the most serious natural disasters in the world. Therefore, it is of great scientific significance to study the variation law of typhoon activity and its causes for typhoon forecasting and disaster prevention and mitigation.

Breadth Learning System (BLS) is an incremental learning algorithm based on random vector function link neural network. Because its training process does not need to repeatedly iterate sample data and calculates the network output layer weight matrix by ridge regression to solve the pseudo-inverse, it is the same as Compared with ordinary neural networks, it is more suitable for processing tasks that require a large amount of computation, such as data classification.

The breadth learning system consists of a feature map layer, an enhanced node layer and an output layer, where the feature map layer and the enhanced vector layer jointly serve as the input of the system. The feature mapping layer realizes the feature extraction of the samples by randomly generating weights through the feature mapping function. The enhancement node layer performs enhancement calculation on the feature vector through orthogonal normalized random weights, and introduces the activation function function to enhance the nonlinear classification ability of the model, so as to achieve the purpose of fully extracting the feature information of the sample data. Finally, by performing pseudo-inverse operation on the composite matrix of the feature map layer and the enhancement node layer, the weight matrix from the input to the output of the system can be obtained. When using the breadth learning system to solve the classification problem, the data set to be classified and the label set of the data set to be classified are used as the input of the breadth learning system, and the breadth learning system is trained, and the breadth learning system outputs the probability that each data belongs to each category. The largest category is the category in which the data is judged.

The typhoon path is a time series composed of the positions of the typhoon center point during the movement of the typhoon. Because the typhoon movement is affected by the regional environment and climate, its movement path is complex, so it is difficult to classify and analyze the typhoon path. At present, the classification research of typhoon track mainly focuses on cluster analysis of typhoon, and uses similarity threshold to classify typhoon track. However, this method is only suitable for a small number of typhoon track classifications. When the typhoon track set is large, the definition of the similarity threshold is difficult and the generalization is poor.

SUMMARY OF THE INVENTION

In order to solve the above problems existing in the prior art, the present invention proposes a typhoon path classification method using a width learning system combined with typhoon path feature representation.

The method of the present invention comprises the following steps:

S1. Construct a set of typhoon path types. The improved DBSCAN algorithm is used to cluster the typhoon paths, and the dynamic time warping algorithm is used to replace the similarity calculation method in the DBSCAN algorithm to calculate the similarity between the paths. category label matrix;

S2. Construct the characteristic matrix of the typhoon path. The improved hierarchical clustering algorithm is used to unify the length of the typhoon path through the characteristic representation. Because the typhoon path points are a data sequence arranged in strict chronological order, the present invention improves the hierarchical clustering algorithm, and only the adjacent points of each path are Perform hierarchical clustering analysis, take the Euclidean distance (Euclidean distance) of adjacent points as the metric, merge the points with smaller Euclidean distances, and unify the length of all typhoon paths to realize the characteristic representation of each typhoon path , get the typhoon track characteristic matrix;

S3. Establish a multi-classification model of typhoon paths based on the width learning system, and use the model to classify typhoon paths. Width Learning System (BLS) is an incremental learning algorithm based on random vector function link neural network. The present invention uses the typhoon track type label matrix obtained by S1 and the typhoon track feature matrix obtained by S2 as the input of the model, and the feature mapping layer The weights are randomly generated to extract the features of the input samples. The enhancement node layer performs enhancement calculation on the feature vector through orthogonal normalized random weights, and uses the activation function function to enhance the nonlinear classification ability of the model. Finally, through the feature mapping layer and the enhancement node layer. The synthetic matrix performs pseudo-inverse operation to obtain the weight matrix from the input to the output of the system. The output value of each row of the output matrix is the probability that each path obtained by BLS belongs to each category, and the index of the maximum value of each row is each test. The category determined by the path, so as to realize the classification prediction of the typhoon path.

Compared with the prior art, the present invention has the following advantages:

First, the improved DBSCAN clustering algorithm is used in the method of the present invention to establish the type set of typhoon paths, and the type set can be automatically generated according to the similarity of each typhoon path in history.

Second, the method of the present invention adopts the feature representation method of improved hierarchical clustering, and unifies the scale of different typhoon path data to provide a basis for classification.

Third, the method of the present invention adopts the width learning system to classify the typhoon path, and realizes the automatic classification of the typhoon path.

Description of drawings

In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

FIG. 1 is a flowchart of a typhoon track classification method based on a width learning system in a specific embodiment of the present invention.

Detailed ways

As shown in Figure 1, the present invention comprises the following steps:

S1. Establish a typhoon track type set through a clustering algorithm.

Because the typhoon path is the trajectory sequence of the typhoon center in chronological order during the typhoon movement, and the DBSCAN algorithm is a density-based clustering algorithm, it can find clusters of any shape, automatically determine the number of clusters, and have a good understanding of noise. Robustness, so it is very suitable for cluster analysis of trajectory data such as typhoon paths, and automatically cluster the paths with large similarity into one category. The similarity is generally measured by distance, and the smaller the distance, the greater the similarity. The original DBSCAN algorithm uses the Euclidean distance measurement method when doing cluster analysis. This method is only suitable for typhoon paths with the same number of points and does not consider the fluctuation of the path. Therefore, the present invention uses dynamic time. The warping algorithm is used as a distance metric for paths. The dynamic time warping algorithm is to lengthen or shorten the different paths that need to be compared until the lengths are the same, and then calculate the distance between the paths. Therefore, the establishment process of the typhoon track type set is as follows:

(1) Set the scanning neighborhood radius eps, the minimum number of included paths minPts, select an unvisited path in the typhoon path sample to start, use the dynamic time warping algorithm to calculate the distance between the path and other paths, and find out the distance between the path and other paths. All nearby paths within and including eps.

(2) If the number of nearby paths is greater than or equal to minPts, the current path and its nearby paths form a cluster, and the departure path is marked as visited.

(3) Repeat steps (1)-(2) to process all paths in the cluster that are not marked as visited. If the number of nearby paths is less than minPts, the path is marked as noise data.

(4) If all paths in the cluster are marked as visited, repeat steps (2)-(3) until all objects are classified as a cluster or marked as noise data, and output the category label matrix of each path. The feature matrix of each path is denoted as the label matrix of the path features, which is denoted as Y.

其中x_i和y_i分别为点T_i的经度和纬度，x_i+1和y_i+1分别为点T_i+1的经度和纬度，具体计算步骤如下：S2. The data lengths of each typhoon path may be different, so the original data cannot be directly used to construct a feature matrix for classification. Considering that the typhoon path cannot be accurately fitted with a functional relationship, an improved hierarchical clustering algorithm is used to extract the same number of points from each typhoon path to represent the typhoon path. This process is named typhoon path feature representation in the present invention. The extracted points are called feature points, and each feature point constitutes a feature path. The calculation principle of the hierarchical clustering algorithm is to first calculate the distance between samples, and merge the points with the closest distance into the same class each time. Then, calculate the distance between the classes, merge the classes with the closest distance into one large class, and keep merging until the specified conditions are met. The invention uses the improved hierarchical clustering method to characterize the typhoon path, and only uses the Euclidean distance (Euclidean distance) of adjacent points as the metric standard, merges the points with small Euclidean distance, and merges the points with the smaller Euclidean distance. as the feature points of the path. If l _i is the distance from point T _i to point T _i+1 ,

where x _i and y _i are the longitude and latitude of the point T _i respectively, and x _i+1 and y _i+1 are the longitude and latitude of the point T _i+ 1 respectively. The specific calculation steps are as follows:

A1. Calculate the distance li between adjacent points on the original _typhoon path.

A2. Save all distances l _i in the set S, S={l ₁ , l ₂ ,..., l _n-1 }, where l ₁ is the distance between point T ₁ and point T ₂ , l ₂ is the distance between point T2 and point T3, l _{n-1 is} the distance between point Tn _{-1 and} point Tn, _i represents the serial number of the point, and n represents the total number of points.

A3. Combine the two points with the smallest distance in the S set into one point, and use the mean of the longitude and latitude of the two points to represent the point to form a new path.

A4. Recalculate the distances between adjacent points on the new path according to step A2, and update the set S.

A5. Go to step A3 until the points of all new paths reach the length of the shortest path set in the sample set, and output the feature path set F.

S3. Establish a typhoon path classification model based on the width learning system, and perform classification prediction based on the model. Breadth Learning System (BLS) is an incremental learning algorithm based on random vector function link neural network. It takes feature path set F and label matrix Y as input, and the feature mapping layer randomly generates weights to extract features from input samples and enhance nodes. The layer performs enhancement calculation on the feature vector through orthogonal normalized random weights, and uses the activation function function to enhance the nonlinear classification ability of the model. The output weight matrix, the output value of each row of the output matrix is the probability of each path belonging to each category determined by BLS, and the index of the maximum value of each row is the category judged by each test path, so as to realize the typhoon path. Classification prediction.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features thereof can be equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention. scope.

Claims (3)

1. A typhoon path classification method based on a width learning system is characterized by comprising the following steps:

s1, constructing a typhoon path category set;

the improved DBSCAN algorithm is used for carrying out cluster analysis on the typhoon path, and specifically comprises the following steps: calculating the similarity between paths by using a dynamic time warping algorithm instead of a similarity calculation method in a DBSCAN algorithm, and automatically grouping paths with large similarity into one category; establishing a category label matrix of the typhoon path;

s2, constructing a feature matrix of the typhoon path;

the method utilizes an improved hierarchical clustering algorithm to unify the length of the typhoon path through feature representation, and specifically comprises the following steps: improving a hierarchical clustering algorithm, only performing hierarchical clustering analysis on adjacent points of each path, combining the points with smaller Euclidean distance by taking the Euclidean distance of the adjacent points as a measurement standard, unifying the lengths of all typhoon paths, and realizing the characteristic representation of each typhoon path; obtaining a typhoon path feature matrix;

s3, typhoon path classification is carried out by using a typhoon path multi-classification model based on a width learning system;

taking the typhoon path type label matrix obtained in the step S1 and the typhoon path feature matrix obtained in the step S2 as the input of the model;

randomly generating weights by a feature mapping layer in the typhoon path multi-classification model to perform feature extraction on input samples, performing enhanced calculation on feature vectors by an enhanced node layer through orthogonal normalized random weights, and enhancing the nonlinear classification capability of the model by using an activation function;

and finally, performing pseudo-inverse operation on a synthetic matrix of the feature mapping layer and the enhanced node layer to obtain a weight matrix from system input to output, wherein the output value of each row of the output matrix is the probability that each path obtained by the width learning system belongs to each category, and the position index of the maximum value of each row is the category judged by each test path, so that the classified prediction of the typhoon paths is realized.

2. The method for classifying typhoon paths based on the width learning system as claimed in claim 1, wherein the constructing step of the typhoon path class label matrix in S1 is:

s11, setting a scanning neighborhood radius eps, wherein the scanning neighborhood radius eps comprises the number minPts of paths at minimum, starting with an unaccessed path in an optional typhoon path sample, calculating the distance between the path and other paths by using a dynamic time warping algorithm, and finding out all nearby paths which are within the eps of the path;

s12, if the number of the nearby paths is more than or equal to minPts, the current path and the nearby paths form a cluster, and the departure path is marked as visited;

s13, repeating S11-S12, processing all paths which are not marked as visited in the cluster, and if the number of nearby paths is less than minPts, marking the paths as noise data;

and S14, if all paths in the cluster are marked as accessed, repeating the steps S12-S13 until all objects are classified into a certain cluster or marked as noise data, and outputting the category label matrix of each path.

3. The typhoon path classification method based on the width learning system as claimed in claim 1, wherein the typhoon path feature matrix constructing step in S2 is:

s21, calculating the distance l between each adjacent point on each original typhoon path_i；

S22, dividing all distances l_iStored in the set S; s ═ l₁,l₂,…,l_n-1In which l₁Is a point T₁And point T₂Distance between points, l₂Is a point T₂And point T₃Distance between l_n-1Is a point T_n-1And point T_nI represents the serial number of the points, and n represents the total number of the points;

s23, combining two points with the minimum distance in the set S into one point, and representing the point by the mean value of the longitude and the latitude of the two points to form a new path;

s24, recalculating the distance between the points adjacent to the newly generated point of the new path according to the step S22, and updating the set S;

and S25, turning to the step S23, and outputting the feature path set until the point number of all the new paths reaches the length of the shortest path set in the sample set.

Images