Comparison of kernel functions of support vector machines: A case study for the solar cell output power prediction

Abstract

Four kernel functions of support vector machines (SVM), namely, radial basis function, sigmoid function, linear function and polynomial function, were applied for the prediction of solar cell output power. Two types of SVM model such as epsilon-SRV and nu-SVR were chosen for each kernel function. Measured values of temperature T (°C) and irradiance E (〖kWh.m〗^(-2)) were used as inputs and solar cell output power P (kW) was used as output. The accuracy of each kernel function was evaluated using well known statistical parameters. Radial basis function using nu-SVR and polynomial function using epsilon-SVR provided similar and better results than other kernels. However, polynomial function has taken more analysis run time while radial basis function used more number of support vectors than other kernels. They may be more computationally expensive.