Self-powered　neutron　detector　(SPND)　is　an　important　nuclear　sensing　device　in　the　core,　whose　health　status　affects　the　safe　operation　of　the　reactor　directly.　Considering　the　measurement　correlation　between　SPNDs　at　different　positions　in　the　reactor,　a　twin　model-based　anomaly　detection　method　for　SPND　signals　in　the　core　was　proposed　in　this　paper.　The　characteristics　of　measurement　signals　of　neighboring　SPNDs　were　extracted　by　the　Random　Forest　Regression　(RFR)　algorithm　based　on　the　historical　operation　data　of　SPND,　and　a　twin　model　was　built　for　SPNDs　which　outputs　the　same　as　its　physical　entity.　Twin　model　and　entity　sensing　coexisted.　The　residual　error　between　the　actual　observation　value　of　SPND　and　the　twin　model　estimation　value　was　calculated　to　serve　as　the　anomaly　detection　criterion,　which　realized　the　identification　and　location　of　single-point　and　multi-point　SPND　anomalies.　The　experiments　show　that　the　prediction　error　of　the　twin　model　proposed　in　this　paper　attains　the　order　of　1×10−10,　which　has　a　very　high　output　consistency.　The　identification　accuracy　of　anomaly　detection　can　reach　over　99%　under　various　abnormal　states　of　SPND　signals,　and　the　single-point　and　multi-point　abnormal　SPND　can　be　accurately　identified,　which　has　a　high　reference　value　for　improving　the　reliability　and　safety　of　the　state　monitoring　of　neutron　flux　measurement　system　in　the　core.　©　2023　Yuan　Zi　Neng　Chuban　She.　All　rights　reserved.