The　feature　extraction　is　key　step　to　fault　diagnosis　in　nonlinear　circuit.　An　improved　particle　swarm　optimization　(PSO)algorithm　are　presented　to　competent　for　the　identification　and　feature　extraction　problem　in　nonlinear　circuit　here.　Firstly,　the　output　response　from　the　circuit　under　diagnosis　will　be　collected　and　processing,　some　order　frequency-domain　kernel　function　under　can　be　extracted　as　circuit　feature　for　fault　diagnosis.　Secondly,　the　single　feature　or　whole　feature　group　of　circuit　can　be　seek　by　particle　swarm　optimization　solution,　the　improved　article　swarm　optimization　algorithm　apply　to　the　searching　for　optimization　solution　of　system　identification　based　on　frequency　domain　kernel　of　the　nonlinear　circuit.　At　last,　the　circuit　diagnosis　result　will　be　give.　The　results　shown:　the　algorithm　is　competent　for　the　system　identification　optimization　and　pattern　classification　problem　well,　the　diagnostic　efficiency　is　high.　©　2010　IEEE.