一种改进的随机向量函数链接网络集成模型

An improved random vector function link network ensemble model

针对传统随机向量函数链接网络集成模型时多样性不足和泛化性能差的问题,提出一种改进的随机向量函数链接集成模型.首先,通过6种简单回归模型替代传统随机向量函数链接网络中的直接链接;其次,采用高斯过程回归(Gaussian process regression,GPR)方法初始化隐含层参数,增强各基分类器的多样性;最后,使用不同的结合策略,集成具有差异性的基分类器得到预测模型.结果表明,改进的随机向量函数链接集成模型的预测精度明显高于其他传统集成模型,较传统随机向量函数链接网络具有更好的泛化性能.

Aiming at the problems of lack of diversity and poor generalization performance in traditional random vector functional link network ensemble model,an improved random vector functional link ensemble model is proposed,which replaces the direct link in the traditional random vector function link network through six simple regression models,and uses the GPR(Gaussian process regression)method to initialize the parameters of the hidden layer to enhance each base classification diversity between devices.Finally,the prediction model is obtained by integrating different base classifiers with different combination strategies.The results show that the prediction accuracy of the improved stochastic vector function link integration model is obviously higher than that of other traditional integration models,and the generalization performance is better than the traditional stochastic vector function link network.