Unvaried-Time Series Analysis And Forecasting Based On Intelligent Computing

Unvaried-Time Series analysis and forecasting is an important portion of current signal process and economics. Its core problem is to establish the dynamic model to the inside contact of the series, and then makes use of the model proceed the forecasting. It primarily includes two contents: constructing predictor and the research of predictor algorithm. So in this paper, our works mainly concentrate on: the further development of the current predictor, and the improvement of predictor algorithm. Immunity Mix Algorithm based on the continuous differential function is proposed in this paper, and its astringency is proved. From here we get the accurate estimator method of the Autoregressive Moving Average model coefficient. The algorithm is simulated with the test function of optimization and actual time series, obtaining the good performance. For the general season time series, according to the model of Season Autoregressive Integrated Moving Average, The concept of horizontal and lengthways trend are gave, and a new season time series model is brought forward. And then the process of modeling is simplified consumedly. To the estimate problem of a kind of time series, the model performance is good. For general nonlinear time series (not-season time series), on the foundation of Pipelined Recurrent Neural Network, BFGS(Broyden-Fletcher-Goldfard-Shannon) is introduced in, so a study algorithm based on BFGS is put forward. The algorithm speeds up network study train, compared with the Real-time Recurrent Learning algorithm consumedly, and increases the accuracy of prediction. The simulation and contrast experiments result are provided in detail in this paper. Some experience regulations are summed up.