Local prediction of Complex Time Series based on Support Vector Machine and Differential Evolution algorithm

Prediction on complex time series has received much attention during the last decades. Global model is the main tool for time series predicting during the last decades, but it suffers low prediction efficiency, low prediction accuracy and high computation complexity for model training and updating. In recent years, local model for time series prediction draws widely attention for its more accuracy prediction ability, lower complexity of models and lower computation complexity of modeling. In this paper, a new scheme for time series prediction is proposed, in which nearest neighbor searching technique is used to searching the top k most similar data samples of the data point waiting for prediction, and then support vector regressing model is constructed with the top k most similar data point with differential evolution algorithm to do SVR training and parameter optimization. This proposed method is applied to three real world complex time series. The method provides relatively better prediction performance in comparison with the others.