摘要
为了提高增量映射学习(IPL)算法的效率,调整了径向基神经网络基函数的中心及方差,以达到调整采样算子的目的,同时,通过神经元函数相关性的计算,确定添加新神经元时,相关函数的阈值,为系统结构调整提供相应依据.新方法步骤相对简单,所以算法速度较快;仿真结果表明,由于系统参数得到调整,对于同一问题,改进IPL算法得到的径向基神经网络结构较一般算法得到的网络结构简单,输出结果也较为精确.
Firstly,we adjusted the parameters of the radial basis function neural network, including the center and the covariance of the base function,and readoped the sampling operator of the algorithm;Then we raised a threshold to decide whether the radial basis function neural network needs a new neuron to reduce the output error of the system or not,and we could get it through calculating the correlation of the system's hidden neurons' base functions and the correlation between the (neuron's) base functions and stock functions.Through iteration of training,adjustment and selection,this method could adjust the system structure logically. Because of the simplicity of the new method,the improved incremental projection learning (IPL) algorithm's calculation speed was faster than before. The simulation results showed that the new algorithm can induce a simpler network structure than the former algorithm,and the output of the new IPL inducing network is more accurate than before.
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2004年第4期655-658,共4页
Control Theory & Applications
关键词
增量映射学习(IPL)算法
径向基(RBF)神经网络
三相训练法
incremental projection learning (IPL) algorithm
radial basis function (RBF) network
three_phase_training algorithm
