基于PLS和GAs的径基函数网络构造策略  

RBFN Structure Determination Strategy Based on PLS and GAs

作　　者：赵伟祥[1] 吴立德[1] 

机构地区：[1]复旦大学计算机科学与工程系,上海200433

出　　处：《软件学报》2002年第8期1450-1455,共6页Journal of Software

基　　金：~~国家自然科学基金资助项目(69935010)

摘　　要：鉴于传统径基函数网络(radial basis function network,简称RBFN)构造策略的不足,提出了基于偏最小二乘法(partial least squares,简称PLS)和遗传算法(genetic algorithms,简称GAs)的RBFN构造策略和一种更有效的径基宽度取值方法.在这个集成构造策略中,PLS克服了K-Means算法求取径基易陷入局部最优的弊病,并使合成径基比由正交算法获取的径基更具代表性;而所提出的径基宽度取值方法和GAs则为网络性能和结构的实质性改善与优化提供了保障.实验证实了基于PLS和GAs的RBFN构造策略及所提出的径基宽度取值方法的优越性、可靠性和有效性.In view of the drawbacks of the conventional structure determination strategies for RBFN (radial basis function network), a RBFN structure determination strategy based on PLS (partial least squares) and Gas (genetic algorithms), and a more effective determination method for radial basis width are proposed in this paper. In this structure determination strategy, PLS avoids the local optima occurring in the process of calculating radial basis with K-Means algorithm and makes the synthesized radial bases more representative than the radial basis got by orthogonal algorithm. Furthermore the proposed determination method for radial basis width and GAs guarantee the substantial improvement and optimization on the network's performance and its structure. Finally, the experiments demonstrate the superiority, reliability, and the effectiveness of the proposed RBFN structure determination strategy based on PLS and GAs, and proposed determination method for radial basis width.

关 键 词：PSL GAS 径基函数网络 构造策略 神经网络 聚类 正交算法 偏最小二乘回归 遗传算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]