Prediction Method Of Complex Dynamic Systems

The complex dynamic system has kinds of uncertainties,so it is very difficult to research the functions and operational mechanism and to predict its development trend.This article studies the predictability of the dynamic complex system from two aspects. The main research works are as follows:1. This article firstly considers the basic aspect of the forecast science and makes an preliminary theoretical analysis according to the martingale and related concepts,then defines the dynamic complex system's predictability.2. This article explores three important areas of time series data,contrasts and analyses euclidean distance and dynamic time warping distance,and points out their adwantages and disadvantages respectively.It also compares various methods of time series classifications and points out disadvantages of the neural network classification such as long network training cycle.3. Empirical mode decomposition method is a decomposition method based on the local signal's characteristics,but the endpoints effect has not been fundamentally resolved. This paper presents and compares three effective methods to deal with the problem above. The method of combining the empirical mode decomposition with principal component analysis is presented.The principal components which are extracted by the principal component analysis are used as input nodes of the RBF neural networks. The experiments show that the method of combining the empirical mode decomposition with principal component analysis optimizes the input variables and reduces the network training cycle. Further the method has higher prediction accuracy than the method to apply single RBF neural network.Some simulation examples show their effectiveness.