State　of　charge　(SOC)　is　a　vital　parameter　which　helps　make　full　use　of　battery　capacity　and　improve　battery　safety　control.　In　this　paper,　an　improved　adaptive　dual　unscented　Kalman　filter　(ADUKF)　algorithm　is　adopted　to　realize　co-estimation　of　the　battery　model　parameters　and　SOC.　Notably,　the　covariance　matching　method　that　can　adapt　the　system　noise　covariance　and　the　measurement　noise　covariance　is　used　to　improve　the　estimation　accuracy.　Besides,　singular　value　decomposition　(SVD)　is　utilized　to　deal　with　the　non-positive　error　covariance　matrix　in　both　unscented　Kalman　filters,　further　enhancing　the　stability　of　estimation　algorithm.　Verification　results　under　Dynamic　Stress　test　and　Federal　Urban　Driving　Schedule　test　indicate　that　improved　ADUKF　can　achieve　more　accurate　SOC　estimates　with　error　band　controlled　within　2.8%,　while　that　of　traditional　dual　unscented　Kalman　filter　(DUKF)　can　only　be　controlled　within　5%.　Moreover,　robustness　analysis　is　also　conducted　and　the　validation　results　present　that　the　proposed　algorithm　can　still　provide　precise　SOC　prediction　results　under　some　disturbances,　such　as　erroneous　initial　SOC,　inaccurate　battery　capacity,　and　various　ambient　temperatures.　©　2020　John　Wiley　&　Sons　Ltd