Electricity　theft　detection　is　a　major　concern　in　modern　power　systems,　and　has　drawn　the　attention　of　both　academia　and　industry.　However,　this　problem　has　not　been　fully　solved　because　of　its　complex　intrinsic　patterns,　which　are　beyond　the　capacity　of　human　decisions.　Machine　learning　methods　have　been　proven　to　be　effective　in　handling　such　complex　problems.　Based　on　an　open　electricity　theft　detection　dataset,　a　deep　learning-based　solution,　the　Temporal　Convolutional　Network　(TCN),　was　proposed　in　this　study　and　compared　with　existing　machine　learning　schemes　to　demonstrate　its　effectiveness.　The　influence　of　the　model　parameters　was　also　investigated　in　numerical　experiments.　The　proposed　TCN　model　is　effective　in　detecting　electricity　theft　behaviour　in　real-world　electricity　markets.　©　2024　SPIE.