Short-term Prediction Method of Resources Occupancy for LBS Guidance System

Abstract

Guidance system can improve the resources utilization of LBS (Location Based Service) system effectively, and the validity of guidance system depends on accurate predicting of the near-future system resources occupancy and its trend. Considering the strong randomicity of short-term characteristic of the trend, a multi-model fusion method integrating adaptive filter and ARMA (Auto Regressive Moving Average) model is proposed to predict the resources occupancy and trend accurately. With the method, observation series of recent resources occupancy is decomposed to different scales and reconstructed by wavelet transform firstly. According to the different features of series at different scale, for approximate signal, the adaptive filter algorithm is used to predict the trend of resources occupancy at coarse scale, and for detail signals in multiple fine scales, ARMA algorithms are adopted. Finally, integrating the prediction results of multi-model from different scales, the resources occupancy and trend with high prediction accuracy can be obtained. Experiment results show that the resources occupancy prediction accuracy of proposed method is higher than that of typical algorithms such as exponential smoothing and weighted Markov algorithms.