I have been tasked to produce a 24 hour electricity demand forecast, further classify it to predict next day peak hours.
For that I made a dataset with several years of hourly electricity demand data for my city together with weather features (all that OpenWeather provides), plus street light on-off schedule, transportation, sunrise-sunset times and also 24h hourly forecast of generation and demand for the state we are part of, which are correlated to my target.
Problem is for all these years there is no data for weekends and holidays. So time series has explicit gaps in it. Sometimes there are 2 days a week, sometime 7-10 days in a row. The good thing though, I do not have to forecast these days also.
So, when I make tsibble data frame (with regular=False) and pass for STL decomposition to extract components it fails with:
.data contains implicit gaps in time. You should check your data and convert implicit gaps into explicit missing values using `tsibble::fill_gaps()` if required.
Can you please advise what is the right way to deal with this time series with explicit gaps in order to employ methods from Forecasting: Principles and Practice? What models will work (if any)?
Tried XGBoost Regressor already with extended feature set, mean, std, slopes and rolling windows etc. Got peak prediction result of 75%, but client is not happy with that...
Thank you in advance