1) As a first hint I would ask myself whether the series looks mean-reverting. With knowledge to the process (low in mornings and evenings, high during the middays) the series does not seem to be mean-reverting. Therefore I would suggest it is not stationary. You can perform statistical test (Dickey-Fuller test).  

2) You can not use the model if series is not stationary. (Moreover it also makes no sense to use t as a variable since the series goes up in the morning but down in the evening. You would do better with some dummy time variables)  

3) You can try differencing of 1 or 2 orders in ARIMA model. Or you can perform the differencing first and then try some of the models you proposed in step 2 on the differenced time series (if it would be stationary).  

4) You would build more robust model if you connect all days into one time series, you would obtain a lot more observation points. Then you would also be able to model seasonal behavior of mornings/evenings (which is more complex topic although).