论文部分内容阅读
尝试将“复制”、“杂交”、“变异”算子和“Metropolis”采样策略引入到微粒群算法(PSO)搜索进程,并将其用于前馈型多层神经网络(FMANN)连接权值优化当中,形成了1种新的非线性统计建模方法:混合微粒群神经网络系统(hybrid particle swarmoptimizer-artificial neural network,HPSO-ANN)。通过仿真对比及对152个HLA-A~*0201限制性T细胞表位活性预测表明:HPSO-ANN仅在少量增加CPU耗时的同时大大提高了算法前期全局搜索能力及后期局部收敛性,特别是对于非线性、高维数等复杂问题该法往往能够取得优于传统QSAR建模方法的实际效果。
Attempts to introduce the sampling strategy of “Copy ”, “Hybridization ”, “Variation ” and “Metropolis ” into the Particle Swarm Optimization (PSO) search process and use it for feed- One of the new nonlinear statistical modeling methods is the hybrid particle swarm optimization-artificial neural network (HPSO-ANN). Through the comparison of simulation and the prediction of 152 HLA-A * 0201-restricted T cell epitopes, HPSO-ANN only improves the global search capability and local convergence in the early stage, This method is often able to achieve better results than traditional QSAR modeling methods for complex problems such as nonlinearity and high dimensional.