Research And Application Of Parameters Optimization And Residual Modification Approach In Forecasting Models Optimization

Forecast aims to provide the required future information for decision makers. It has been paid more and more attention because of the importance in decision-making. But how to find out the change and motion law in complex stochastic systems and predict the future trend? For this purpose, various forecast models have been proposed to predict and analyze different systems. However, all-purpose models are impossible and every model has its own drawbacks and shortcomings. At the same time, since traditional model was always established for the forecasting and analysis of a particular system at first, they surely have limitations while application. But in practice the simple, effective and accurate models are required. In order to developing reliability and accuracy future information, in-deep studies should been done on traditional forecasting models to improve the traditional approach and enhance forecasting abilities.Optimal methods are formed on the basis of higher requirements for models predicting abilities. The arising of various optimization algorithms and technologies bring new vitality to traditional models. By introducing optimal methods to traditional models, the forecasting precision could be improved effectively.Particle Swarm Optimization (PSO) is a kind of intelligent optimization technique. For the features of simple structure, easy to implement, PSO currently has been widely applied to various areas such as function optimization, neural network training, pattern recognition and signal processing. On the other hand, PSO provides valid solutions for the optimization of nonlinear continuous functions, combinatorial optimization, mixed integer nonlinear optimization and so on. Also, using PSO to search the optimal solutions of equations is another important application. Thus, PSO could be applied to optimize the parameters of forecasting model, achieving forecast abilities improvement objective.Errors are unavoidable in forecasting process. So, the forecast precision can be reduced by using appropriate model to simulate and predict the residual series and then adding predicted residual to corresponding original model forecast value. Since error directly related to whether a model is high-precision and is the most important model evaluation criteria, modified residual to develop more accurate model is another effective way for model improvement. On the basis of studying and review model optimization methods, parameters optimization method and residual modification approach are mainly applied in this paper. Firstly, PSO is introduced to the optimization of three grey derivation models:GIM(1)(grey linear power index model),GLPM(1)(grey logarithm power model)å'ŒGPPM(1)(grey parabola power model). Then, the proposed three hybrid models are applied to forecast grain yield in China. Secondly, by applying residual modification approach to the optimization of high-precision model, this paper put forwards residual modification seasonal ARIMA model, and the experiment results conclude that residual modification approach is effective and feasible in the optimization and improvement of seasonal ARIMA.