A method and system for intelligent analysis of stroke data distribution patterns

Abstract

The present invention relates to a method and system for intelligently analyzing the distribution patterns of stroke data. The method comprises: obtaining stroke case data; performing ZCA whitening on the stroke case data to obtain whitened data; performing dimensionality reduction on the whitened data using a stacked sparse autoencoder to obtain reduced-dimensionality data; clustering the reduced-dimensionality data using a learning vector quantization clustering method optimized by deep reinforcement learning to obtain clustering results; and generating stroke data distribution patterns based on the clustering results. Based on this method, the present invention can improve data processing efficiency, thereby improving the efficiency and accuracy of generating stroke data distribution patterns.

Claims

1. A method for intelligent analysis of stroke data distribution patterns, characterized in that: include: Obtain stroke case data; Performing ZCA whitening processing on the stroke case data to obtain whitened data; Performing dimensionality reduction processing on the whitening processing result by using a stacked sparse autoencoder to obtain dimensionality reduced data; Clustering the dimensionality reduction data using a learning vector quantization clustering method optimized by deep reinforcement learning to obtain a clustering result; generating a distribution pattern of stroke data based on the clustering results; The learning vector quantization clustering method optimized for deep reinforcement learning is a learning vector quantization clustering method that incorporates a state set and an action set of deep reinforcement learning; The reinforcement learning used is deep Q network; The process of constructing the action set includes: The action set is obtained through the exploration and utilization mechanism, and a reward function suitable for the clustering method is designed to determine the action corresponding to the maximum reward value. The action corresponding to the maximum reward is selected in each iteration to obtain the clustering result after the iteration. Among them, the reward function design: Where a i represents the i-th iteration from the action set A={a1,a2,...,a i ,...,a L }, Represents the execution of action a i The average distance between the centroid and each type of cluster samples, d i Represents the average distance between the original LVQ cluster samples and their centroids in the i-th iteration; The selection process of the action set is as follows: Adopting the "exploration" mechanism and the "exploitation" mechanism to selectively select data points and perform "closer or farther" operations on the prototype vector; Among them, the implementation process of the "utilization" mechanism is as follows: introduce a parameter m < z, and randomly select m samples from the input dataset to form a data subset X m , perform a "pulling closer" or "moving away" operation on the m data in the dataset and the prototype vector to obtain an action; z is the number of samples in the stroke dataset; The implementation process of the "exploration" mechanism is as follows: introduce the parameter v < z, the exploration coefficient ε is taken as 0.1, and randomly select v samples from the stroke dataset to form a data subset X v , and perform an operation of "getting closer" or "getting farther away" on all the data in the data subset X v with the prototype vector to obtain an action; If the sample x i With the prototype vector p j The labels of are the same, then a "closer" operation is performed. The "closer" operation is: use the formula p' j =x j +η(x i -x j ) updates the prototype vector; where η is the learning rate, p′ j is the updated prototype vector, x j is the prototype vector p j The corresponding attribute value; Sample x i and the updated prototype vector p′ j The distance between them is: ||p′ j -x i ||2=||x j +η(x i -x j )-x i ||2=(1-η)*||p j -x i || If the sample x i With the prototype vector p j If the labels are different, the "away" operation is performed. The "away" operation is: use the formula p' j =x j -η(x i -x j ) for the prototype vector p j Make updates; At this time, x i and the updated prototype vector p′ j The distance between them is: ||p′ j -x i ||2=||x j -η(x i -x j )-x i ||2=(1+η)*||p j -x i ||2。 2. The intelligent analysis method for stroke data distribution pattern according to claim 1, characterized in that: The performing ZCA whitening processing on the stroke case data to obtain whitened data specifically includes: Converting the stroke case data into a numerical matrix; Performing normalization processing on the numerical matrix to obtain a first matrix; Determining a sample covariance matrix of the first matrix, and determining eigenvalues ​​of the sample covariance matrix; Arrange the eigenvalues ​​in descending order, and extract the eigenvector of each eigenvalue after the descending order to obtain a second matrix; determining a rotation matrix of the second matrix; The whitening data is determined based on the rotation matrix.

3. The intelligent analysis method for stroke data distribution pattern according to claim 1, characterized in that: The selection process of the state set is: The prototype vector obtained in each iteration of the learning vector quantization clustering method is used as a state, and the number of iterations is used as the number of states to form the state set.

4. An intelligent analysis system for stroke data distribution patterns, characterized in that: The system is used to implement the intelligent analysis method for stroke data distribution rules according to any one of claims 1 to 3; the system comprises: Data acquisition module, used to acquire stroke case data; A whitening processing module, configured to perform ZCA whitening processing on the stroke case data to obtain whitened data; A dimensionality reduction processing module, configured to perform dimensionality reduction processing on the whitening processing result by using a stacked sparse autoencoder to obtain dimensionality-reduced data; A clustering processing module is used to cluster the dimensionality reduction data using a learning vector quantization clustering method optimized by deep reinforcement learning to obtain a clustering result; A rule generation module is used to generate a stroke data distribution rule based on the clustering result.

5. The intelligent analysis system for stroke data distribution pattern according to claim 4, characterized in that: The whitening processing module includes: A matrix conversion unit, used for converting the stroke case data into a numerical matrix; a normalization processing unit, configured to perform normalization processing on the numerical matrix to obtain a first matrix; an eigenvalue determining unit, configured to determine a sample covariance matrix of the first matrix and determine an eigenvalue of the sample covariance matrix; a feature extraction unit, configured to arrange the eigenvalues ​​in descending order and extract an eigenvector of each eigenvalue after the descending order arrangement to obtain a second matrix; a matrix determining unit, configured to determine a rotation matrix of the second matrix; A whitening processing unit is configured to determine the whitening processing data based on the rotation matrix.

6. The intelligent analysis system for stroke data distribution pattern according to claim 4, characterized in that: The cluster processing module includes: A clustering processing unit is used to cluster the dimensionality reduction data using a deep reinforcement learning optimized learning vector quantization DQN-LVQ clustering method to obtain a clustering result; the deep reinforcement learning optimized learning vector quantization clustering method is a learning vector quantization clustering method that incorporates a state set of deep reinforcement learning and an action set of deep reinforcement learning.

Images