Currently,　high-speed　train　(HST)　is　mainly　controlled　by　manual　driving　and　automatic　train　protection　system,　which　may　reduce　the　comfort　of　passengers　and　impair　the　intelligence　of　train　operation.　In　recent　years,　some　intelligent　driving　methods　have　been　proposed　for　subway　line.　However,　because　of　the　continuous　rise　of　HST＇s　operation　speed　and　mileage,　the　driving　data　collected　from　HST　is　more　than　that　of　subway　and　the　intelligent　driving　model　will　be　complex　if　the　source　driving　data　is　directly　trained　by　data　mining　algorithms.　So,　the　source　driving　data　sets　are　classified　into　several　classes　in　terms　of　the　features　of　the　driving　data.　In　addition,　iterative　sparse　L0-norm　minimization　is　applied　to　sparsify　the　classified　driving　data　and　thus　the　redundant　data　will　be　deleted,　which　can　speed　up　the　computation　speed　of　learning　process.　Furthermore,　ensemble　CART,　including　B-CART　and　A-CART　are　used　to　find　the　driving　rules　of　both　experienced　drivers　and　ATO　controller.　Finally,　the　field　data　of　Beijing-Shanghai　high-speed　railway　and　ATO　simulation　data　are　used　to　test　the　performance　of　the　proposed　intelligent　driving　methods.　Compared　with　A-CART,　the　energy　consumption,　and　the　redundancy　of　the　training　data　set　of　S-A-CART　algorithm　can　be　respectively　decreased　by　0.27%　and　40%　and　the　passengers’　riding　comfort　can　be　increased　by　17.71%.　©　2018　Elsevier　Ltd