Based on the data of excellent driver's operation records，a framework of energy-saving driving optimization was proposed using machine learning technology. Hierarchical decomposition was applied to integrated machine learning method to excavate the hidden driving patterns from the driving log data of excellent drivers. The learning and forecasting of speed information and gear information were separately carried out to realize the automatic driving decision for energy-saving optimization of a train，and the actual railway lines and locomotive data were used for experimental verification. The test results show that under the constraints of ensuring the safety, punctuality and stability of the train, the energy-saving driving program of the train can save about 7% energy when compared with the average level drivers.