摘要
本文提出了一种基于选择性集成径向基函数神经网络(SE-RBFNN)的来波方向(DOA)估计方法,解决单个神经网络建模进行DOA估计精度低的问题。首先利用Bagging方法训练生成一定数量的RBFNN弱分类器,其估计精度低但泛化能力强;然后提出并运用等宽分箱—投票选择性集成方法剔除估计误差大的奇异值个体,优选部分RBFNN输出结果进行平均处理,从而获得了高精度的DOA估计。仿真结果表明了算法的有效性,相对单个RBFNN建模,构建的选择性集成模型能适应方向特征的变化,算法的来波估计精度显著提高。
A novel DOA estimation algorithm based on the equal-width-voting selective RBF neural network ensemble is proposed,which overcomes the low estimation precision and weak generalization ability flaws in the traditional single neural network estimation algorithm.Bagging algorithm is used to train amount individual RBFNN,which has low estimation precision but good generalization performance.Then the equal-width-voting selective ensemble algorithm is proposed to extract the appropriate number of ensemble members from the available individual networks,meanwhile the DOA estimation strong learner that has higher estimation precision and better generalization ability is constructed.Experimental results show that,compared with the single RBFNN estimation algorithm,the estimation precision and generalization ability is largely improved.
出处
《电路与系统学报》
北大核心
2013年第1期65-69,共5页
Journal of Circuits and Systems
基金
国家自然科学基金(60972161)
