In　order　to　solve　the　problems　of　model-based　diagnosis　method　for　distribution　network　faults,　such　as　slow　diagnosis　speed,　low　accuracy　and　low　fault　tolerance　in　diagnosis　of　multiple　multi-phase　faults,　a　model-based　hierarchical　diagnosis　method　for　distribution　network　faults　is　proposed.　From　the　aspect　of　diagnosis　algorithm,　the　new　fitness　function　and　the　search　strategy　of　feature　learning　are　used　to　improve　the　diagnosis　speed　and　accuracy.　In　the　diagnosis　model,　the　single-layered　single　high-dimensio-nal　operations　is　transformed　into　two-layered　multiple　low-dimensional　operations　by　a　hierarchical　approach,　which　further　improves　the　diagnosis　speed　and　accuracy.　The　fault　tolerance　ability　of　the　first　layer　diagnosis　is　improved　by　defining　the　constraint　relation　of　the　equivalent　components,　and　that　of　the　second　layer　diagnosis　is　improved　by　using　the　redundancy　relation　of　voltage　constraints　and　current　constraints.　Numerical　examples　show　that　compared　with　other　diagnosis　methods,　the　speed　of　model-based　hierarchical　diagnosis　method　is　greatly　improved,　the　accuracy　is　always　maintained　near　the　ideal　value,　and　the　fault　tolerance　ability　is　obviously　enhanced.　In　the　fault　diagnosis　of　large-scale　distribution　network,　the　model-based　hierarchical　diagnosis　method　has　obvious　advantages.　©　2020,　Electric　Power　Automation　Equipment　Press.　All　right　reserved.