The invention provides a method for optimizing energy conservation of locomotives and automatically driving the same on the basis of high-order correlation learning. The method includes carrying out experience grooming and preprocessing on actual driving data of drivers; constructing a feature set to describe driving states of the locomotives; describing the actual driving data of the drivers on the basis of the constructed feature set and constructing hypergraphs by the aid of training data which are the actual driving data of the drivers; acquiring hypergraph models by the aid of inductive learning procedures; allowing the hypergraph models obtained by means of learning to be accurate by means of constant strengthening and updating, in other words, in iteration training modes. The constructed hypergraph models can be used for generating driving operation in real time. The method has the advantages that training of the hypergraph models is free of constraints of time factors, and accordingly learning of the hypergraph models can be provided with excellent optimization spaces; excellent energy conservation effects can be realized when the hypergraph models are applied to acquire operation control tap positions of the locomotives in locomotive operation procedures; online control operation results of the locomotives can be used as data input, and accordingly the hypergraph models can be constantly adjusted and optimized.