Global　warming　has　led　to　an　increasing　number　of　extreme　weather　events,　sequential　attacks　of　such　events　cause　significant　damage　to　distribution　grids,　and　power　utilities　have　to　spend　huge　amounts　of　resources　to　strengthen　grid　structures.　To　improve　the　pre-event　resilience　of　distribution　systems　in　the　face　of　extreme　events　while　saving　costs,　this　paper　proposes　a　deep　reinforcement　learning-based　distribution　grid　hardening　strategy　that　combines　the　failure　rate　of　hardened　components　with　the　hardening　cost　to　form　a　bi-objective　optimization　model.　Simulations　are　performed　in　the　IEEE　37-bus　system　to　verify　that　the　hardening　strategy　found　by　the　proposed　approach　can　significantly　improve　system　resilience　under　different　hardening　conditions　and　disaster　situations.　©　2021　IEEE.