On the basis of studying the shortcomings of traditional AIC in its criterion order determination method , we propose a new method, which uses the modified particle swarm optimisation to determine the order of ARMA (r, m) model.The basic idea is that since the particle swarm optimisation is prone to early mature and is easy to fall into local optimum , which results in premature convergence and fail to achieve the optimal solution , so we propose the issue that to use the search method of dividing the parent group into sub-groups to avoid the algorithm falling into local optimum , to let sub-groups to obtain the optimal solutions through self-searching respectively , and to employ the optimal solutions as the new generation of particle population , then to continue the search to automatically generate the optimal solution , and to accurately determine the order of ARMA model , in this way it overcomes the shortcomings of AIC ’ s order determination criterion in cumbersome calculation and inaccurate order determination .MATLAB is used in instance simulation experiment , it proves that the method is simple and feasible .%在研究传统AIC准则定阶方法的缺点的基础上，提出用改进的粒子群算法对ARMA（ r，m）模型定阶的新方法。基本思想是，由于粒子群算法容易早熟，易陷入局部最优，从而导致过早收敛，得不到最优解，因此提出利用母群划分子群的搜索方法避免算法陷入局部最优的问题。子群各自进行搜索得出最优解，由最优解作为新一代粒子种群，继续搜索自动生成最优解，对ARMA模型进行准确定阶，克服了AIC定阶准则的计算法繁琐、定阶不精确的缺点。通过MATLAB进行实例仿真验证，证明了该方法简单可行。
展开▼
