electric load forecasting

Data exploration

Univariate analysis using the average temperature as the only regressor

• Model complexity vs RMSE

• Response curve and performance

Modelling with polynomial regression using average temperature and timestamp

• Model complexity vs RMSE

• Response surface: 5th degree vs 5th degree + Fourier terms

Modelling with Stepwise regression method

• Response surface: 5th degree + Fourier terms vs stepwise + Fourier terms

Final comparison using average temperature: Stepwise + Fourier vs 8-neuron MLP

• Neuron selection using k-fold cv and 1-SD rule

• Response surface and performance

Final model using all 25 temperatures

• Neuron selection using k-fold cv and 1-SD rule

• Goodness of Fit

• Residue histogram

• Prediction

Performance of all models on test set

MODELLO	RMSE	MAPE	R^2
Polinomio 5° grado + Fourier	17.93	10.14%	0.8507
Stepwise + Fourier	15.96	8.75%	0.8818
MLP con 8 neuroni	14.93	8.06%	0.8968
MLP con 8 neuroni (25 temp)	10.79	6.03%	0.9460
MLP con 19 neuroni (25 temp)	10.35	5.79%	0.9503