OPTIMIZING BIODIESEL YIELD VIA AN ADAPTIVE LOCAL LINEAR REGRESSION MODEL WITH APPLICATION TO RESPONSE SURFACE METHODOLOGY

Abstract

In the estimation of maximum biodiesel yield, a proportion of oil palm and cotton-seed oil of ratio 60% to 40%
(P60C40) as given in the literature was collected for the production of biodiesel through ultrasound assisted trans-
esterification process. The challenge in the estimation of biodiesel yield, is to find an appropriate regression model
that would adequately capture maximum biodiesel yield production that would reflect the experimental yield of
biodiesel via three factors namely; reaction time, methanol-to-oil ratio and concentration of catalyst. The existing
techniques; second-order linear regression and the Artificial Neural Network (ANN) were utilized but could not
capture local variability in the data because the coded factors lack the axial (star) points which allows for the estimation
of curvature and maintain rotatability in the data. In other to address the challenge, we introduce an axial points to the
coded factors as employed in the experimental design known as the Circumscribed Central Composite Design (CCCD)
and the proposed adaptive local linear regression model to improve the goodness-of-fit statistics for Response Surface
Methodology (RSM) data. The results obtained show that the proposed adaptive local linear regression model gave
the maximum biodiesel yield of 96.31% that is approximately equal to the experimental biodiesel yield of 96.32%
over the Ordinary Least Squares (OLS) of 94.95%, Second-order linear regression model of 96.41% and ANN of
96.67% respectively. In addition, the proposed adaptive local linear regression model has the least residual error over
the other models utilized in this paper.