摘要
提出一种太阳黑子月均数混沌时序的模糊神经网络预测方法。该方法根据时间序列的延迟因子和饱和嵌入维数重构相空间,利用Lyapunov指数法判别时序系统的混沌特性,采用混合pi-sigma模糊神经推理方法拟合混沌吸引子特性。其中混合pi-sig-ma模糊神经网络以高斯基函数作为模糊子集的隶属度函数,在线动态调整隶属度函数和结论参数,并采用量子粒子群算法(QPSO)优化网络初始参数,提高预测准确度。该模型具有物理意义清晰、预测精度高以及预测结果确定等优点,仿真实验结果证明了该方法的有效性。
A fuzzy-neural network prediction method of sunspot month average number chaotic time series is proposed. The method reconstructs the phase space according to the delay factor and saturation embedded dimension of the time series, and uses Lyapunov exponents to estimate the chaotic characteristics of the time series system, uses hybrid pi-sigma fuzzy-neural reasoning method to fit the chaotic attractor characteristics. Where, the hybrid pi-sigma fuzzy neural network uses Gaussian base function as the membership functions of fuzzy sets, dynamically adjusts the membership functions and conclusion parameters online, and uses quantum particle swarm optimisation （QPSO） to optimise the initial parameters of network in order to improve the forecast accuracy. The model has clear physical significance, high prediction accuracy and determined prediction results. Simulation experimental results demonstrate the effectiveness of the method.
出处
《计算机应用与软件》
CSCD
北大核心
2013年第8期91-94,98,共5页
Computer Applications and Software
基金
国家安全生产监督管理总局安全生产科技发展指导性计划项目(06-472)
河北省教育厅科学研究基金项目(Z2006439)
