【摘要】 为提高细菌群体趋药性(bacterial colony chemotaxis,BCC)算法优化后RBF神经网络(RBFNN)的泛化能力,提出了一种新的细菌编码方式。将隐层节点位置和相应的控制参数组成细菌,使RBFNN的两个参数同时得到优化;同时,在整体算法中融入最速下降法,利用其能快速收敛到极小点的特性,使算法速度得到提升。将此算法优化后的RBFNN用于IRIS和双螺旋分类问题,结果表明:算法速度提升了大约60%,同时泛化效果也得到提高。更多还原

【Abstract】 To improve the generation ability of radius basis function neural networks(RBFNN) optimized by bacterial colony chemotaxis(BCC) algorithm,a new coding style that becterial are composed by the locations of hidden units and controlling parameters is proposed,then both parameters of RBFNN can be sought automatically according to the new code.At the same time,the steepest descent algorithm is added into the whole algorithm,so that the calculation speed is improved for its speciality of converging into local minimus points rapidly.We apply the RBFNN optimized by this algorithm in the classification of IRIS and twin-screw,an approximate 60% improvement of speed is achieved and the generation ability is advanced.更多还原