Harmonic　state　estimation　is　the　main　method　of　harmonic　source　location　and　harmonic　responsibility　division,　also　is　a　necessary　means　of　reduction　harmonic　hazard.　However,　when　some　regions　in　the　power　system　are　unobservable,　the　traditional　harmonic　state　estimation　methods　cannot　estimate　accurately　due　to　the　underdetermined　state　equations.　In　order　to　solve　the　above　problems,　a　harmonic　state　estimation　method　based　on　the　network　equivalence　and　deep　learning　is　proposed　in　this　paper.　Firstly,　network　equivalence　and　coefficient　matching　are　proposed　for　locating　harmonic　source.　Then,　the　harmonic　state　estimation　model　is　constructed　based　on　generative　adversarial　network　(GAN),　and　the　mechanism　equation　is　fused　in　the　objective　function　to　make　the　model　include　the　power　grid　information.　Finally,　the　accuracy　of　the　proposed　method　is　verified　on　IEEE33　system.　The　results　show　that　the　proposed　method　can　accurately　locate　harmonic　source　and　has　high　precision　of　state　estimation.　©　2023　IEEE.