Comparative Study of Load Demand Forecast Using Non-Linear Regression and Neural Network Techniques: A Case Study

Abstract

Demand load forecasting is the estimation of electrical load that
will be required by a certain geographical area with the use of
previous electrical load data in the said geographical area. This
has become necessary due to the increasing number of
prospective power users around the world to address for likely
shortage of electricity and to further plan for resources
budgeting and power real time availability. However, this
research work carried out a short term comparative study of
electric load demand forecast of the University of Benin,
Ugbowo campus between 1st to 30th September, 2019. The
forecasting approaches used in realizing this task are the nonlinear regression and artificial neural network (ANN)
approaches which was analyzed using MATLAB 2015 software.
The current load demand was presented and modeled using the
two approaches, the ANN gave an optimal result of 0.0021%
mean absolute percentage error (MAPE) for all the days and all
the constraint parameters used for the training of the model.
Following the validation of the ANN model with the non- linear
regression (NLR) model, it was observed that the artificial neural
network gave the best optimal result of 0.0021% MAPE for all
constraints applied while the non-linear regression model gave
an optimal result of 0.0448% MAPE. Therefore, the artificial
neural network (ANN) model is considered to produce a more
accurate result than the non-linear regression model as
confirmed by the result of validation clearly confirming model
suitability for the analysis