Journal
| Title | Performance Analysis of Support Vector Regression Machine Models in Day-Ahead Load Forecasting Posted by Lemuel Clark Velasco |
| Authors | Velasco, Lemuel Clark P; Polestico, Daisy Lou L; Abella, Dominique Michelle M; Alegata, Genesis T; Luna, Gabrielle C |
| Publication date | 2020/2/13 |
| Journal | Advances in Intelligent Systems and Computing |
| Volume | 1130 |
| Pages | 541-555 |
| Publisher | Springer Nature |
| Abstract | Support vector machines (SVM) is a machine learning framework that has exhibited optimum performance in the functions of classification and clustering. This study explored Support Vector Regression Machines (SVRM) as a specialized application of SVM in predictive functions by conducting a performance analysis of various SVRM models for day-ahead load forecasting. In order to find an appropriate SVRM model that can yield promising forecasting results, data preparation which involved data representation and feature selection was conducted for the electric load dataset and found out that only the time attribute has a relevant relationship with the consumed electric load. Through the selection of an appropriate kernel along with its SVRM parameters and SVRM architecture, it was found out that the Radial Basis Function kernel along with SVRM parameters: c = 110, g = 0.001, e = 0.01 and p = 0.005 implemented in an SVRM architecture that uses: day before, two days before, seven days before, and fourteen days before electric load data as input for the SVRM model yields the best forecasting results. The results generated and obtained by this study clearly suggests that with proper data representation, feature selection, kernel selection, parameter selection and architecture selection, SVRM can go beyond clustering and classification by being a viable forecasting technique for day-ahead electric load forecasting. |
| Index terms / Keywords | Machine Learning Performance, Support Vector Machines, Support Vector Regression Machines, Electric Load Forecasting |
| DOI | https://doi.org/10.1007/978-3-030-39442-4_40 |
| URL | https://link.springer.com/chapter/10.1007/978-3-030-39442-4_40 |