Support vector regression for voltage reference elements monitoring

We use support vector regression (SVR) for building predictive models for monitoring the behavior of voltage reference elements (VREs). The predictive abilities of SVR are compared to some standard neural network-based predictors. We also use v-SVM (support vector machine) technique, which allows automatic control of the model accuracy. We test SVR on different data sets by comparing the performance of different approximation techniques such as polynomial, radial basis function, anova and neural network. Experimental results confirm the potential applicability of SVR for voltage predictive modeling. The predictive models are further used to estimate the next voltage value without performing measurements with a high precision digital voltmeter. The models for short-term prediction are previously trained by using long-term measurements of voltage, preformed by a high-precision digital voltmeter. Predictive models are added to the voltage reference elements (VRE) and by simple voltage digitalization the predictor estimates the next voltage value. Due to the robustness of the predictors, the voltage estimation is allowed.