神经网络应用于多元变量时间序列的建模研究

自然界复杂系统的动态特性通常包含在多个变量时间序列的演化轨迹中。采用主成分分析与神经网络相结合的方法，进行多元变量时间序列的建模和预测研究。主要思路是首先通过主成分分析法找到一组相互独立的输入变量；再利用神经网络实现多个变量之间以及变量当前状态和未来状态之间的函数映射。多组实例仿真结果证明了该方法进行多元变量预测的可能性和有效性。

Application of Neural Networks on Modeling of Multivariate Time Series

For a practical complex system,the internal dynamics is often contained in the time series of multiple model variables.A new methodology is studied to model and predict multivariate nonlinear time series based on combination of principal component analysis(PCA) and neural network.PCA finds the uncorrelated directions of maximum variance in the data space of different variables.Neural network makes predictions on the basis of approximating both the functional relationship among different variables and the map between current state and future state.Simulation results from different examples show the probability and the validity of the proposed method.