摘要
针对传统混沌时间序列预测模型的复杂性、低精度性和低时效性的缺点,在倒差商连分式基础上提出全参数连分式模型,并利用量子粒子群优化算法优化模型参数,将参数优化问题转化为多维空间上的函数优化问题.以二阶强迫布鲁塞尔振子和三维二次自治广义Lorenz系统为模型,通过四阶Runge-Kutta法产生混沌时间序列,并利用基于量子粒子群优化算法的全参数连分式、BP神经网络和RBF神经网络分别对混沌时间序列进行单步和多步预测.仿真结果表明,基于量子粒子群优化算法的全参数连分式结构简单、精度高、效率高,该预测模型可被推广和应用.
In view of the complexity, low precision and low timeliness of traditional chaotic time series prediction models, a model about full-parameters continued fraction is proposed on the basis of the inverse difference quotient continued fraction. The quantum particle swarm optimization algorithm is used for parameters optimization of the model,thus the parameters optimization problem is transformed into the function optimization problem on the multidimensional space. Second order forced Brussels vibrator and three-dimensional quadratic autonomous generalized Lorenz system are taken as models respectively, then chaotic time series which will be used as the simulation object can be attained according to the fourth order Runge-Kutta method. Intercomparison experiments among the model about full-parameters continued fraction based on the quantum particle swarm optimization algorithm, the BP neural network and the RBF neural network are conducted on single-step and multi-step prediction for chaotic time series. The simulation results show that the fullparameters continued fraction based on the quantum particle swarm optimization algorithm has simpler structure, higher precision and higher efficiency, so this prediction model can be widely promoted and applied.
出处
《控制与决策》
EI
CSCD
北大核心
2016年第1期52-58,共7页
Control and Decision
基金
国家自然科学基金项目(61463047)
自治区研究生科研创新项目(XJGRI2015029)
关键词
全参数连分式
量子粒子群优化算法
混沌时间序列预测
full-parameters continued fraction
quantum particle swarm optimization algorithm
chaotic time series prediction
