基于自适应变异PSO的ARMA模型参数寻优及预测应用  

Parameters optimization and prediction application of ARMA model based on PSO with adaptive mutation

作　　者：李怀俊[1] 谢小鹏[2] 李军[1] 

机构地区：[1]广东交通职业技术学院车辆安全工程技术中心,广州510650 [2]华南理工大学汽车摩擦学与故障诊断研究所,广州510640

出　　处：《计算机应用研究》2015年第4期1004-1006,1015,共4页Application Research of Computers

基　　金：广东省自然科学基金资助项目(S2011010002118);广东省高校优秀青年教师培养项目(Yq2013178)

摘　　要：为提高ARMA模型在时间序列预测中的精度,提出一种基于改进粒子群算法(AMPSO)的模型参数智能寻优估计方法。AMPSO算法以粒子熵的判别为依据,在寻优过程中对算法的关键参数进行多次自适应变异,以提高其跳出局部、面向全局寻优的能力。模型参数寻优过程基于ARMA(2n,2n-1)模型架构依次跳阶,每阶的参数初估后运用AMPSO算法展开寻优,适应度判定标准为当前模型残差方差最小。针对齿轮箱轴承的输出扭矩进行预测,结果表明,AMPSO算法在收敛性和寻优速度方面效果良好;参数寻优方法可显著提高参数预测精度,具有良好的工程应用价值。To improve the prediction precision with auto-regressive moving average（ ARMA） model in time series,this paper presented an ARMA parameters estimation method based on improved particle swarm optimization（ AMPSO）. The algorithm determined the adaptive mutation scheme with the key parameters of AMPSO based on the discrimination of swarm entropy to improve the ability of global optimization in searching process. It optimized the model parameters using jumping steps scheme from smallest step based on ARMA（ 2n,2n- 1） model,and used the AMPSO algorithm that the fitness criterion was to get the minimum residual variance after the initial parameters estimation in every step. Using the method to estimate the torque of the gear box bearing and the result shows there is good engineering application value with notable performance in astringency and optimization speed to AMPSO algorithm and the significant precision to the prediction method.

关 键 词：粒子熵 PSO ARMA模型 参数寻优 残差方差 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] O329[自动化与计算机技术—控制科学与工程]