基于追逐粒子与自适应变异的PSO-SVM算法与养殖水质评价研究  

PSO-SVM algorithm based on chasing particle and adaptive mutation and evaluation of aquaculture water quality

作　　者：魏超 张刚[1] 韩祥兰[2] 蔡永胜 冯志敏[1] WEI Chao;ZHANG Gang;HAN Xiang-lan;CAI Yong-sheng;FENG Zhi-min(Faculty of Maritime and Transportation, Ningbo University, Ningbo 315211, China;Branch of Information and Computer Science, Ningbo Institute of Technology, Zhejiang University, Ningbo 315110, China)

机构地区：[1]宁波大学海运学院,浙江宁波315211 [2]浙江大学宁波理工学院信息与计算机科学分院,浙江宁波315110

出　　处：《宁波大学学报（理工版）》2018年第3期20-26,共7页Journal of Ningbo University：Natural Science and Engineering Edition

基　　金：国家自然科学基金(51675286);宁波市自然科学基金(2017A610085);浙江省公益技术项目(2017C32014)

摘　　要：为提高标准PSO算法对SVM参数优化选取的精度和速度,提出以粒子群适应度均方差为判别依据,用追逐粒子位置替代陷入局部优化时的群体最优位置,对粒子的速度进行自适应调整;利用两个距离最远的粒子位置构成的圆形区域作为最优位置的吸引域,引入位置变异因子,使粒子随着迭代次数有规律地跳出局部最优位置吸引域,从而对粒子位置进行自适应变异.通过对4种不同适应度函数迭代100次的收敛测试,并与另外2类不同粒子群算法比较,结果表明改进算法收敛精度提高了20倍以上.将改进的PSO-SVM算法与另外3种分类算法对采集的5 439组养殖水质数据按照从优到差5个级别进行水质评价分类试验,发现改进的PSO-SVM算法收敛速度和收敛精度最高,对水质评价的错误率为1.54%,能有效地满足实际水质分类需求.Aim to improve the accuracy and speed of SVM classification in water quality evaluation, the PSO optimization method is applied to optimize the key parameters of SVM algorithm in this paper. A new adaptive adjustment of particle velocity method is proposed, which is based on the consideration of taking particle swarm fitness variance as evaluation criterion and using chasing particle position to replace local optimization particle position. First, an attracting domain is defined by the distance between farthest two particles. Then, by position variation factor making the particle jump out of the local optimal attracting domain regularly with the constantly iterations. Thus, the particle location can be adaptively mutated. Convergence tests on four different functions shows that the proposed improved algorithm is increased by more than 20 times in convergence accuracy comparing to another two kinds of particle swarm algorithm. Last, the improved PSO-SVM is applied in five-level water quality classification test based on the collected 5439 group water quality data. Contrast with the other three classification methods, the improved PSO-SVM algorithm has the highest convergence rate and convergence accuracy. The error rate of water quality evaluation is 1.54%, which can effectively meet the actual water quality classification requirements.

关 键 词：水质评价 支持向量机 粒子群算法 自适应变异 

分 类 号：X824[环境科学与工程—环境工程]