Accurate　estimations　of　the　state　of　charge　(SOC)　and　state　of　health　(SOH)　are　crucial　for　improving　battery　management　techniques.　However,　batteries　are　affected　by　temperature　and　aging,　leading　to　nonlinear　relationships　that　are　more　difficult　to　be　characterized.　This　article　proposes　an　SOC-SOH　joint　estimation　method　of　lithium-ion　battery　based　on　temperature-dependent　extended　Kalman　filter　(EKF)　and　deep　learning.　First,　the　battery　model　state,　control,　and　observation　matrices　with　temperature　and　capacity　variables　are　created　for　real-time　SOC　estimation　by　using　EKF　at　the　local　end.　Second,　battery　aging　features　are　extracted　and　weighted　using　convolutional　neural　networks　(CNNs)　and　attention　mechanisms　and　are　combined　with　a　gated　unit　to　solve　long　time　series　memory　problem　for　SOH　estimation　at　remote　computing　platform.　Finally,　the　dual　time-scale　joint　model　is　realized　by　real-time　SOC　estimation　on　the　local　controller,　and　the　SOH　can　be　calculated　on　the　remote　computing　platform　to　correct　the　available　capacity　to　further　update　SOC　at　the　end　of　the　discharge.　Through　1C　discharge　rate　cycle　experimental　validation,　the　root　mean　square　errors　of　SOC　and　SOH　estimation　were　within　1%.　Therefore,　the　proposed　joint　SOC-SOH　estimation　method　can　be　achieved　with　local　and　remote　computation.　©　1982-2012　IEEE.