A New Machine Double-Layer Learning Method and Its Application in Non-Linear Time Series Forecasting

Machine learning is an effective method, whose aim is recognize unknown samples through learning from known samples. At present, artificial neural network (ANN), support vector machine (SVM) and genetic algorithm (GA) are the most popular machine learning methods, but they all have some defects as well as some merits. In this paper, a new machine double-layer learning strategy is put forward. It integrates the merits of ANN/SVM and GA. ANN/ SVM is used to carry out inner layer learning in order to obtain model´s inner parameters, and GA is used to implement outer layer learning so as to acquire model´s outer parameters. Therefore the new learning method need carry out double layers learning, by comparison to common machine learning, the new method possesses stronger self-adaptive ability, and it can make up the shortcomings of single learning method and fully assure model´s generalization ability. In the end, the machine double-layer learning method is applied for nonlinear time series forecasting, and examples show the correctness and validity of the new method.