The　nonlinear　characteristics　of　the　output　of　the　photovoltaic(PV)　array　and　the　maximum　power　point　tracking　algorithm　will　affect　the　operation　of　PV　array　protection　equipment.　In　order　to　identify　correctly　the　operating　status　of　PV　array,　a　novel　fault　diagnosis　method　that　combines　Bayesian　optimization　algorithm(BOA),　stacked　autoencoder(SAE),　and　ensemble　extreme　learning　machine(EELM)　is　proposed　in　this　study.　First,　the　time　sequence　waveform　of　PV　array　is　standardized.　Then,　SAE　is　used　to　automatically　extract　the　features　of　the　standardized　time　sequence　waveforms,　and　an　EELM　model　is　trained　for　fault　classification.　Finally,　BOA　optimizes　the　overall　diagnosis　model.　The　experimental　results　show　that　the　proposed　method　has　98.40%　and　98.10%　fault　diagnosis　accuracy　for　simulation　and　experiment,　respectively,　which　is　better　than　that　of　backpropagation　neural　network,　support　vector　machine,　random　forest,　etc.　©　2022,　Solar　Energy　Periodical　Office　Co.,　Ltd.　All　right　reserved.