Least squares-support vector machine prediction method based on adaptive particle swarm

Abstract

The invention relates to a least squares-support vector machine (LS-SVM) prediction method based on an adaptive particle swarm. According to the method, the inertia weight is adjusted according to the degree of convergence of groups and the adaptive values of individuals. Training is speeded up. A matrix equation appearing in an LS-SVM is iteratively calculated using the algorithm, and matrix inversion is avoided. Memory is saved. The optimal solution is obtained. By using the method, the training sample can be simplified effectively, and the training speed is improved. Moreover, the method has the advantages of high classification accuracy, fast convergence and good generalization ability. The problems in prediction like high feature dimension, redundancy between samples and a limited number of samples are solved.

Description

technical field [0001] The invention relates to a least square support vector machine prediction method based on adaptive particle swarm. Background technique [0002] Let the training set S={(x i ,y i )|i=1,2,…,m}, where x i ∈R n and y i ∈R is the input data and output data respectively, and the least squares support vector machine LS-SVM uses the structural risk minimization-SRM criterion to construct the minimum objective function J(ω,e) and its constraints are as follows: [0003] [0004] s.t.y i =w T Φ(x i )+b+e i [0005] Among them, w is the weight vector, γ is a constant, b is a constant bias, e i for the deviation. [0006] In order to solve the optimization problem of the above formula, it is transformed into solving the following linear equations: [0007] [0008] where Q=y i the y j Φ(x i ) T Φ(x i )=y i the y j K(x i ,x j ), K(x i ,x j ) is a kernel function satisfying the Mercer condition, I is the identity matrix, L=[1,1,…,1]∈R ...

AI Technical Summary

Problems solved by technology

[0018] It can be seen from the model of the particle swarm optimization algorithm PSO that if the acceleration factor c 1 ,c 2 Or if the value of the inertial weight factor ω is large, the particle swarm may miss the optimal solution, causing the algorithm not to converge. Mature convergence, resulting in slower convergence speed in the later stage of the algorithm, and the accuracy may also decrease

Images