Based On Neural Network And Time Series Prediction Method And Its Application Research

This paper considers the forecasting methods and application based on neuralnetwork model and time series model. The main purpose is to derive time series model,neural network model, combined model based on time series model and neural nerworkmodel, hybrid model based on time series model and neural nerwork model and theapplication of these models in the Canadian lynx date series forecasting, whichincluding the following contents:1. Time series model and neural nerwork model are presented. First introduces thebasic conception and basic theoretics, then do the numerical analysis, which areprepared for considering combined model and hybrid model.2. Combined model based on time series model and neural nerwork model arepresented, then combined model based on combined objective function is provided. Theprovided model combined time series model and neural nerwork model, the weightcoefficient of this model is decided by goal programming, the objective function of thisgoal programming considered both the sum of squared errors and maximum absoluteerror. The result of numerical analysis suggests that the provided model reduced the rootmean squared error and maximum absolute error at the same time.3. Hybrid model based on time series model and neural nerwork model arepresented, then hybrid model based on binomial smoothing is provided. First recovernonlinear information through neural network model based on binomial smoothing, thenachieve the residual of actual data series using actual values minus the forecastingvalues of the neural network model based on binomial smoothing, finally recover linearinformation from the residuals through auto-regression model. Emprical results indicatethat the forecasting accuracy of the provided model is over then neural network modelbased on binomial smoothing and the hybrid model based on time series model andneural nerwork model.