Mining time series data based upon cloud model

In recent years many attempts have been made to index, cluster, classify and mine prediction rules from increasing massive sources of spatial time-series data. In this paper, a novel approach of mining time-series data is proposed based on cloud model, which described by numerical characteristics. Firstly, the cloud model theory is introduced into the time series data mining. Time-series data can be described by the three numerical characteristics as their features: expectation, entropy and hyper-entropy. Secondly, the features of time-series data can be generated through the backward cloud generator and regarded as time-series numerical characteristics based on cloud model. In accordance with such numerical characteristics as sample sets, the prediction rules are obtained by curve fitting. Thirdly, the model of mining time-series data is presented, mainly including the numerical characteristics and prediction rule mining. Lastly, a case study is carried out for the prediction of satellite image. The results show that the model is feasible and can be easily applied to other forecasting.