The　reasonable　construction　of　organic　Rankine　cycle　(ORC)　data-driven　model　is　the　basis　of　analysis,　prediction　and　optimization.　The　validity　of　data,　the　rationality　of　input　variables　and　the　adaptability　of　structural　parameters　are　the　key　to　the　efficient　construction　of　the　model.　In　the　modeling　process,　the　data　selected　based　on　experience,　the　introduction　of　redundant　variables　and　fixed　structural　parameters　will　reduce　the　prediction　accuracy　and　enhance　the　instability　of　the　model.　In　this　paper,　we　fully　consider　the　abnormal　characteristics　of　data,　the　coupling　correlation　of　variables　and　the　instability　characteristics　of　the　actual　operation　of　the　system,　and　propose　an　integrated　online　dynamic　modeling　scheme,　which　aims　to　improve　the　prediction　accuracy,　anti-interference　ability　and　adaptability　of　the　ORC　data-driven　model　under　changeable　environment.　Parameter　information　extraction　can　further　improve　the　prediction　accuracy　of　the　model.　Mean　absolute　error　can　improve　by　42.2%.　The　introduction　of　network　structure　design　strategy　can　reduce　the　construction　time　of　the　model　by　79.56%　and　improve　the　R-square　by　62.18%.　The　combination　of　adaptive　gradient　approach　and　neuronal　mapping　potential　evaluation　method　can　improve　the　search　speed　and　ensure　the　fast　convergence　of　the　model.　The　integrated　online　dynamic　modeling　scheme　proposed　can　provide　necessary　theoretical　guidance　and　new　modeling　idea　for　the　whole　process　design　of　ORC　data-driven　model.