Optimizing the performance of MLP and SVR predictors based on logical oring and experimental ranking equation

ABSTRACT

Improving conventional prediction systems is widely used to optimize the learning process, achieve higher performance, and avoid overfitting. This paper’s purpose is to propose a new predictor for solar tracking systems applications based on oring operator and ranking equation with a conventional predictor including Multi-Layer Perceptron (MLP) and Support Vector Machine Regression (SVR). The point of using oring and ranking equation is to create a new variable that stores the information of combined attributes. This process aims to increase the accuracy of predictors and increase the efficiency of intelligent solar tracking systems. The experiments used 6 different datasets for solar tracking systems. The results revealed that the proposed predictors performed better than conventional predictors. Using the proposed predictors has improved both Root Mean Square Error (RMSE) and Coefficient of Determination (R²). The developed MLP models showed lower RMSE and higher R² compared to conventional MLP models. The improvement ranges for using MLP are from 1.0013 to 1.4614 degrees for RMSE, and from 1.0019 to 1.4984 times for R², while the improvement ranges using SVM are from 1.001 to 1.988 degrees for RMSE and from 1.000 to 2.385 times for R².

KEYWORDS:

• Optimization model
• multilayer perceptron
• support vector machine regression
• single axis tracker
• prediction model

Nomenclature

Isc	=	Short Circuit Current
MLP	=	Multi-Layer Perceptron
Pr	=	Power Radiation
R2	=	Coefficient of Determination
RMSE	=	Root Mean Square Error
SVR	=	Support Vector Machine Regression
Voc	=	Open Crcuit Voltage

Disclosure statement

No potential conflict of interest was reported by the authors.