Abstract: Forecasting of cooper price is an important area of international commodity research. A new multi-scale combined forecasting model is built in this paper by using empirical mode decomposition (EMD), artificial neural network (ANN), support vector machine (SVM) and time series methods based on the idea of decomposition-reconstruction-integration. During the model building process, a new idea to use run length judgment method to reconstruct the component sequences is proposed. Then this model is used to analyze the fluctuation characteristics and trend of LME copper price. Copper price series is decomposed and reconstructed into high frequency, low frequency and trend sequences which can be explained from the angle of irregular factors, major events and long-term trend. Empirical analysis shows that comparing with the gray model GM (1, 1), Elman and some other single models and ARIMA-SVM combined model, multi-scale combined model obtained the best forecasting result.

Key words: multi-scale model, EMD, SVM, ANN, run length judgment method