The Application Of Ensemble Neural Network In Time-Series

The analysis and prediction of the time-series is a very important research aspect in predictive areas, which is a premise of scientific decision-making. Also time-series prediction is a cross area of diversity subjects. In this paper, based on the theory of the neural network, machine learning and time-series prediction, the ensemble neural network model is applied into the time-series prediction. In the same time, the theory's research, methods and model's construction are deemed as the emphases of our work.This paper simply introduces the definition of the time-series prediction and some diversified prediction models and mainly analyze how to apply the models of ensemble neural networks into time-series' prediction, including diversified ensemble models, constructing process, learning rules and training methods. Then these constructed ensemble models are used to the prediction of time-series and the generalization of the ensemble models is investigated. In addition, the constructed ensemble models are used to predict some real time-series and the results are very perfect, which show that our constructed models have better prediction and generalization. The constructed ensemble model is efficient and feasible.As a result, the complexity of the ensemble model is studied. An effective method to evaluate the ensemble models is presented. In the same time, the effective exploration is made to improve the ensemble model, and the way is put forward to construct the ensemble model with various individual models. Ultimately this paper uses the real time-series to testify the above.