Abstract: Time-series models for forecasting face low performance due to non stationarity as well as distribution drift. To remedy this, we propose a novel adaptive normalization approach AdaNorm-TS.