With　electric　vehicles＇　popularity,　a　surge　has　been　created　in　demand　for　charging　infrastructure.　As　a　result,　the　maintenance　of　charging　piles　has　become　a　critical　issue　that　requires　attention.　To　effectively　utilize　the　fault　features　of　the　front　and　back　circuits　in　case　of　the　charging　pile　fails,　a　multifeature　fusion　model　is　proposed　in　this　article.　First,　use　the　front-　and　back-stage　feature　information　fusion　module　to　fuse　the　collected　front-stage　fault　feature　quantity　signals　and　the　back-stage　fault　feature　quantity　signals.　Then,　the　spatial　and　temporal　feature　extraction　modules　are　used　to　mine　the　spatial　and　temporal　high-dimensional　features　in　parallel.　Finally,　through　the　spatiotemporal　feature　fusion　classification　module,　the　spatial　and　temporal　features　are　fused　and　classified　to　achieve　the　purpose　of　fault　diagnosis.　The　proposed　method　employs　deep　learning　techniques,　which　avoids　the　cumbersome　steps　involved　in　graphical　input　and　the　errors　arising　from　manually　selecting　features　in　traditional　deep　learning　algorithms　and　gives　full　play　to　the　parallel　diagnostic　performance　of　deep　learning.　The　simulation　results　demonstrate　that　the　proposed　method　outperforms　other　comparative　algorithms　in　terms　of　diagnostic　accuracy,　convergence　speed,　and　overfitting　suppression,　and　has　excellent　noise　immunity,　which　can　cope　with　the　noisy　situation　of　charging　piles.　In　the　experimental　test,　the　fault　diagnosis　accuracy　of　this　method　reached　96.36%,　and　its　recognition　sensitivity　for　most　fault　categories　was　higher　than　that　of　the　comparison　model,　which　further　verified　the　superiority　and　robustness　of　this　method.　©　2015　IEEE.