Support Vector Regression for the Prediction of Net Power Output in Downdraft Gasifier

Different technologies, including downdraft gasification technology, have been explored in converting biomass feedstock to generate power. Effective biomass utilization is premised on its predictability. However, the variability of biomass characteristics, process complexity, and controllability pose a significant threat to large-scale power generation through gasification. Conventional analytical techniques used to support decision on net power output depend on a plethora of parameters, which are sometimes poorly defined. Intelligent modelling approach provides a less expensive route for evaluating benefits and associated risks with the gasification process since it does not rely on the experimental verification method and can predict the real-time condition of the gasification system. Here, we propose a support vector regression method that eschews the process representation uncertainties to predict the net power output in a downdraft gasification process. SVR can handle linear and non-linear data and is less sensitive to outliers; hence, it is widely deployed in various real-world applications. Three kernel functions of SVR (Gaussian, Polynomial, and Linear) were used to determine the optimality. The input parameters considered in the model development are elemental properties (C, H, O S, N) proximate compositions (moisture, ash, volatile material, and fixed carbon), and operating parameters, which are gasifier temperature and air-to-fuel ratio. Dataset comprising 980 simulated data points from various types of biomasses under different operating conditions was used to train and test the models. The data was divided into a ratio of 70:30 for training and testing, respectively. The Polynomial kernel function outperformed the other variants with RMSE, MAD, MAPE, and R-Square of 28.14kW, 18.46kW, 16.26% and 0.87 respectively.