From one-step ahead to multi-step ahead forecasting

[jacktang]

Hello there! Now I created one SNARIMAX model on hourly dataset with one step forecasting:

    model = time_series.SNARIMAX(
        p=2,      
        d=1,      
        q=1,     
        m=24,     
        sp=28,    
        sq=28,     
        sd=0,     
        regressor=(
            preprocessing.StandardScaler() |
            tree.HoeffdingTreeRegressor()
        )
    )

and the performance was good:

Metrics on original scale:
MAE: 4.14
RMSE: 9.97
MAPE: 1.33%

and i extended to day ahead forecasting (24 hours ahead) and it never got good performance. Even I simplify the model with LinearRegression or PARegressor. Any suggestions on multi-step forecasting? or can you share how to build multi-step forecasting model?