【摘要】 针对深度信念网络在参数训练过程中易陷入局部搜索、收敛速度慢等问题,提出利用改进的混沌免疫算法优化深度信念网络参数的方法。采用混沌初始化和自适应变异,提高抗体种群的多样性;引入可变选择算子,加快算法的寻优速度。函数拟合实验和滚动轴承故障诊断实验结果表明,与粒子群优化算法和基本的克隆选择算法相比,该算法能够得到更优的网络参数,提高了深度信念网络的特征提取能力,加快了网络训练的收敛速度。更多还原

【Abstract】 Aiming at the problem that deep belief network is easy to fall into local search and slow convergence in parameter training,a method of optimizing deep belief network parameters based on the improved chaos immune algorithm was proposed.The chaos initialization and self-adaptive mutation were used to raise the diversity of antibody populations.The variable selection operator was introduced to accelerate the search speed of the algorithm.Results of simulation experiments on the function fitting and rolling bearing fault diagnosis show that the proposed method can obtain better network parameters compared to the particle swarm optimization algorithm and the basic clone selection algorithm.It can not only improve the feature extraction ability of deep belief network,but accelerate the convergence speed of network training.更多还原