一种蚁群算法优化的BP神经网络技术研究  

Research on BP Neural Network Technology of Ant Colony Algorithm Optimization

作　　者：徐益民 杨余旺[1] 郭利强 XU Yimin;YANG Yuwang;GUO Liqiang(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094;Huaihai Industries Group,Changzhi 046000)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094 [2]淮海工业集团,长治046000

出　　处：《计算机与数字工程》2022年第11期2373-2376,2460,共5页Computer & Digital Engineering

摘　　要：BP神经网络被广泛用于分类与回归问题,对初值选取较敏感,易陷入局部最优;ACO-BP算法可用于改进神经网络的训练,提升全局寻优能力,但训练中网络预测性能易出现抖动。基于ACO-BP算法,引入核光滑方法,采用自适应信息素挥发率,优化神经网络的训练。使用四折交叉验证方法,将ACO-BP算法、BP算法与改进ACO-BP算法,应用于UCI数据库中三组数据集进行验证。改进算法的收敛速度与ACO-BP算法相近;对复杂问题预测性能显著优于ACO-BP算法。实验结果表明,与ACO-BP算法相比,改进ACO-BP算法在加速网络收敛的同时,具有较强的鲁棒性与全局寻优能力。BP neural network is widely used in classification and regression problems,which is sensitive to the initial value selection and easy to fall into local optimum.ACO-BP algorithm can be used to improve the training of neural network,enhance the global optimization-seeking ability and accelerate the network convergence,but the network prediction performance in training is prone to jitter.Based on the ACO-BP algorithm,a kernel smoothing method is introduced in the pheromone update strategy,the pheromone release formula is updated,the pheromone upper limit restriction is removed,and the adaptive pheromone volatility is used to optimize the training of the neural network.Using a four-fold cross-validation method,the ACO-BP algorithm,the BP algorithm and the improved ACO-BP algorithm are applied to three datasets of Cancer,Iris and MPG in the UCI database for validation.The convergence speed of the improved algorithm is similar to that of the ACO-BP algorithm,the prediction performance for complex problems is slightly higher than that of the BP algorithm and significantly better than that of the ACO-BP algorithm.The experimental results show that,compared with the ACO-BP algorithm,the improved ACO-BP algorithm accelerates the convergence of the network while reducing the jitter of the network performance during training,and has stronger robustness and global optimization seeking ability.

关 键 词：人工神经网络 蚁群算法 混合算法 神经网络训练 启发式信息 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]