Reliable Prediction System Based on Support Vector Regression with Genetic Algorithms

This study applies a novel neural-network technique, support vector regression (SVR), to predict reliably in dynamical system. The aim of this study is to examine the feasibility of SVR in state prediction by comparing it with the existing neural-network approaches. To build an effective SVR model, SVR´s parameters must be set carefully. This study proposes a novel approach, known as GA-SVR, which searches for SVR´s optimal parameters using genetic algorithms, and then adopts the optimal parameters to construct the SVR models. The application results of practical vibration data state forecasting measured from a Co2 compressor demonstrate that the GA-SVR model outperforms the existing neural network based on the criteria of mean absolute error (MAE) and root mean square error (RMSE).