Automated Energy Profile Forecasting

Universal forecasting startegy on energy data with AutoML.

Dataset

• Smart meters in London
• Elia grid load
• More...

AutoML libraries

• TPOT
• Autosklearn
• H2O.ai
• Facebook Prophet

Training pipeline

• Request local weather data from Darksky API(Optional)

• Feature construnction with lag/ahead

• Training process

• Evaluation and confidence interval construction

TPOT

1. Costomize your configuration of algorithm searching by modifyingtpot_multi.py.

2. Train your model:

from tpot import TPOTRegressor
import tpot_multi
tpot_reg = TPOTRegressor(config_dict = tpot_multi)
tpot_reg.fit(train_X, train_y)

3. Time and space complexity issue emerges for multioutput regression with large number of features. See Dask For more detail, see Customizing TPOT

Autosklearn

• Support multioutput now,

• A confidence interval method for regression estimator...

H2O

• Not supporting multioutput regression

Facebook Prophet

• Not so many experiments yet, support multioutput and confidence interval, but poor results.