Study On The Model And Algorithm Of Time Series Prediction

In each field of natural science and social science, a great deal of decision problems can not get away from the prediction which is the foundation of the decision. The best method to settle down prediction problems is to detect and find out the law in the dynamic state process or phenomenon. In natural, the needed information is usually insufficient, also the relative theories. The people's understanding to the thing is limited by the observed data ,namely time series. So we can make use of the existing history data to establish a model to predict the future.Firstly ,in this paper, introduces the current ways and models that the time series prediction adopt and the character and summarizes the aspects that the current ways and model are needed to be improved in the nonlinear system prediction. Secondly, introduces the good character and advantages of the artificial neural networks especially feed forward neural network (BP Neural Network) in the application of the nonlinear prediction models. Then, points out the weaknesses which exists in the BP Neural Network prediction model and its training algorithm: The BP neural network is a kind of static network that have no processing ability of time so that it can't identify the time series model. The standard BP algorithm is slowly constringency and is easy falling into the local minimum points. Aiming at these weaknesses, the paper presents a good identify ability of the AR prediction model which is used to identify the time series and set up the representative training samples in the BP Neural Network. Meanwhile, based on the analyzing the main...