The　accurate　prediction　for　the　remaining　useful　life　(RUL)　of　bearing　is　essential　for　ensuring　the　reliable　operation　and　establishing　an　effective　maintenance　strategy.　During　the　whole　life　of　bearing,　the　vibration　signal　contains　more　and　more　nonlinear　information　of　performance　degradations.　A　novel　model　named　dynamically　activated　convolutional　network　(DACN)　is　proposed　to　extract　the　nonlinear　information　of　vibration　signal.　DACN　can　execute　the　feature　extraction　on　the　input　and　adaptively　activate　the　features.　Based　on　DACN,　DACN-ConvLSTM　is　proposed　for　the　RUL　prediction　of　rolling　bearing.　In　model,　ConvLSTM　can　mine　the　timing　information　between　adjacent　signal　samples.　which　improves　the　prediction　accuracy　of　bearings.　Finally,　the　model　is　verified　by　two　cases.　The　features　are　visualized　by　using　t-SNE　method　which　provides　effective　prediction　pictures.　The　experiment　results　present　the　proposed　DACN-ConvLSTM　method　possesses　the　higher　RUL　prediction　accuracy　compared　with　other　methods,　and　can　maintain　the　prediction　performance　on　different　datasets.