CONCLUSION
Two nonparametric models Root Transform Local Linear Regression and
Kernel Density Estimation are proposed for estimating electric load PDF
over the Gaussian distribution used in literature to improve the
accuracy of electric load modeling. The performance of the nonparametric
techniques was compared alongside the Gaussian and Gamma distribution
and assessed using electric load data from over 2400 enterprise and
residential locations in the United Kingdom using RMSE, \(R^2\), Kolmogorov-Smirnov test and data splitting. Root Transform Local Linear
Regression had the best results across the board with the lowest Test
RMSE values and with the most locations producing p-values greater than
0.01 when conducting the KS test followed by Kernel Density Estimation.
The parametric distributions had overall the highest RMSE values per
location and the KS test null hypothesis was rejected for all locations
using those models. Further research areas would investigate the performance of RTLLR in power systems planning and optimization studies for predicting
stochastic load.