The Research Of Key Techniques For Function Mining And Time Series Analysis By Gene Expression Programming

Data Mining, as a cross-discipline, has received plenty of attentions in the database research groups since the last two decades. Among various data mining technologies, Evolution Computing has been widely applied in complex problem handling because it requires little field knowledge and no prior model. Particularly, applying Gene Expression Programming to function mining and time series analysis has won big success.However, these practices have exposed some limitations of traditional GEP. For instance, it is difficult to mine segmented function and to decide the embedding dimension for time series. Based on existing research, this thesis combines GEP with statistics and wavelet analysis to form a few new methods for function mining and time series prediction. The main results and contributions are as follows,1) Analyses the disadvantages of original GEP in segmented function mining and proposes Wavelet-based Segment Point Finding Algorithm. This algorithm applies a discrete wavelet transformation on training data. Through an analysis on wavelet detail coefficients, it can discover the segment points of segmented functions and hence enable GEP to mine segmented functions effectively.2) Proposes self-correlation coefficients based method to classify time series into Severe Vibrated Time Series and Mild Vibrated Time Series, and...