Deep　learning　methods　are　widely　employed　for　transient　stability　analysis　in　power　systems.　However,　the　prediction　performance　can　be　adversely　affected　by　undesirable　factors　in　the　measured　data.　To　alleviate　the　effect　of　the　undesirable　factors,　a　hybrid　physics-based　and　data-driven　prediction　network　of　the　rotor　angle　trajectory　is　proposed　in　this　paper.　The　prediction　model　embeds　the　rotor　equation　to　form　a　dynamic　learning　model　that　conforms　to　the　actual　physical　law.　The　physical　consistency　of　the　prediction　results　is　guaranteed.　Meanwhile,　a　dynamic　error　derivative　integral　network　incorporating　the　Runge–Kutta　method　is　proposed　to　correct　the　final　results.　The　accuracy　of　the　prediction　can　be　improved.　Finally,　it　is　tested　in　the　IEEE　39-bus　system　and　the　East　China　Power　Grid　system.　The　test　results　show　that　the　model　significantly　outperforms　other　comparative　models.　And　the　dependence　on　the　quality　of　measured　data　can　be　alleviated　effectively.　©　2025　Elsevier　Ltd