State　estimation　is　a　key　function　of　the　energy　management　system　in　the　power　grid　dispatching　control　center,　which　can　be　used　to　estimate　the　operating　status　of　the　entire　power　network　and　enable　real-time　monitoring　of　the　power　system.　Currently,　the　availability　of　phasor　measurement　units　(PMUs)　alone　is　insufficient　to　meet　the　observability　requirements　for　state　estimation.　To　address　this,　integrating　conventional　measurement　systems,　like　supervisory　control　and　data　acquisition　(SCADA),　into　a　hybrid　measurement　setup　can　effectively　enhance　measurement　redundancy.　Meanwhile,　the　Kalman　filter　algorithm　used　for　dynamic　state　estimation　in　power　systems　demands　high　model　accuracy,　yet　its　tracking　capability　and　estimation　precision　are　compromised　under　the　influence　of　unknown　time-varying　noise.　To　mitigate　the　impact　of　unknown　measurement　noise　and　accurately　obtain　system　states,　this　paper　proposes　a　hybrid　measurement　state　estimation　method　based　on　the　improved　adaptive　robust　extended　Kalman　filter　(IAREKF)　algorithm.　Firstly,　the　power　measurements　of　SCADA　are　converted　into　equivalent　current　phasor　measurements　by　the　measurement　transformation　strategy,　which　linearizes　the　measurement　equations　and　enhances　computational　efficiency.　Simultaneously,　a　spatio-temporal　pseudo-measurement　generation　method　is　devised　that　incorporates　power　transfer　distribution　factors　and　temporal　correlations,　thereby　supplementing　missing　state　information　and　enabling　the　fusion　of　hybrid　measurements.　Secondly,　the　measurement　noise　covariance　is　estimated　by　analyzing　the　statistical　characteristics　of　the　measurement　devices.　An　equivalent　weight　function　is　constructed　based　on　the　standardized　innovation　to　modify　the　measurement　noise　covariance,　thereby　enhancing　the　robustness　of　the　algorithm.　Finally,　an　adaptive　forgetting　factor　method　is　employed　to　refine　the　Sage-Husa　noise　estimator.　Quantify　the　overall　filtering　performance　of　the　current　state　estimation　utilizing　the　ratio　of　the　trace　of　the　innovation　variance　matrix　to　the　trace　of　the　innovation　covariance　matrix　as　a　metric.　An　adaptive　forgetting　factor　is　determined　using　this　metric,　adjusting　the　level　of　discounting　applied　to　outdated　data　during　the　estimation　of　the　process　noise　covariance　matrix.　This　method　enhances　the　precision　of　the　adaptive　estimation　of　process　noise,　thereby　improving　the　algorithm′s　dynamic　tracking　performance　of　the　system　state.　Simulation　results　show　that　in　the　IEEE　39-bus　power　system,　under　conditions　of　unknown　model　and　noise　parameters,　the　proposed　method　achieves　a　mean　absolute　error　(MAE)　of　3.477×10-4(pu)　for　voltage　phase　angle　estimation,　and　MAE　of　0.03057°　for　voltage　phase　angle　estimation.　Compared　to　the　generalized　maximum　likelihood　iterated　extended　Kalman　filter　(GMIEKF)　and　adaptive　H∞　extended　Kalman　filter　(AHEKF)　algorithms,　the　MAE　reductions　are　37.51%　and　42.21%　for　voltage　magnitude　estimation,　and　14.97%　and　21.46%　for　voltage　phase　angle　estimation,　respectively.　Under　the　influence　of　bad　data,　the　MAE　reductions　are　37.26%　and　24.84%　for　voltage　magnitude　estimation,　and　12.95%　and　6.55%　for　voltage　phase　angle　estimation,　demonstrating　excellent　robustness.　During　system　transients,　the　proposed　method　maintains　high　estimation　accuracy,　showcasing　its　superior　dynamic　tracking　capability.　The　following　conclusions　can　be　obtained　through　simulation　analysis:　(1)　Facing　the　differences　in　characteristics　among　various　measurement　systems,　the　measurement　transformation　technology　is　adopted　to　achieve　the　fusion　of　hybrid　measurements.　Pseudo-measurements　are　constructed　based　on　the　spatio-temporal　characteristics　of　the　system　to　increase　the　frequency　of　state　estimation　to　synchronize　with　PMU　sampling,　enabling　a　more　rapid　acquisition　of　the　system＇s　dynamic　changes.　(2)　The　measurement　noise　covariance　is　estimated　by　analyzing　the　characteristics　of　measurement　devices.　Furthermore,　a　robust　factor　is　constructed　based　on　the　standardized　innovation　to　adjust　the　measurement　noise　covariance　matrix,　ensuring　robustness　against　bad　data.　(3)　An　adaptive　forgetting　factor-based　process　noise　estimation　algorithm　is　proposed.　Compared　to　the　traditional　Sage-Husa　noise　estimation　method,　this　algorithm　can　more　quickly　forget　previously　invalid　statistical　information,　leading　to　a　more　accurate　estimation　of　process　noise.　Consequently,　it　achieves　higher　estimation　accuracy　under　system　disturbances　or　abrupt　changes.　©　2025　China　Machine　Press.　All　rights　reserved.