Comparison of Linear Regression and ANN of Fish Oil Biodiesel Properties Prediction

Abstract

Abstract The straight fish oil based biodiesel is used as fuel in CI engines by adjusting physical properties such as density and viscosity and increasing the temperature. In the present study, the variation of density and viscosity of fish oil based biodiesel blend with temperature are studied. The volumetric fraction of biodiesel in the range of 10 to 100% is added with diesel and change in the properties of viscosity and densities are investigated. From the experimental results, density (ρ) and kinematic viscosity (ʋ) of fish oil are synchronized with diesel properties at B40 at 40-50ºC. Artificial Neural Network (ANN) and linear regression modeling are employed for predicting the density and viscosity of fish oil biodiesel blends. For training the ANN, 60% of data has been used a training set, 30% of data has been used a testing set and the remaining 10% of total data has been used for validation. It is observed that the MAPE (Mean Absolute Percentage Error) obtained from the ANN is only 5.27% whereas in the regression prediction the error is 23.9 %. Hence, it is established that ANN gives the best performance over regression analysis and found suitable for predicting the biodiesel properties with better accuracy. Thus, it eliminates instrumental usage and man-hours.