Black-box Models for Reference Voltage Monitoring

I. Nančovska Šerbec and D. Fefer (Slovenia)


Time Series, Black-Box Models, Prediction, SupportVector Regression, Neural Networks.


In this paper we use predictive models for monitoring the behavior of voltage reference elements (VRE-s). The predictive abilities of different paradigms, such as neural network-based predictors, support vector machine (SVM) for regression, regression trees and simple linear regression are compared. Predictors are evaluated on reference voltage time series and on two other artificial data sets, such as Lorenz attractor and Fractional Brownian Motions. Models are treated as black-boxes in the way they fit input-output data. It means that no physical knowledge about the underlying system is used but the chosen model structure belongs to families that are known to have good flexibility and have been "successful in the past". The predictive models are used to estimate the next voltage values without performing measurements. The models for short-term prediction are previously trained by using long-term measurements of voltage, performed by high-precision digital voltmeter. Due to the robustness of the predictors, the voltage estimation is allowed by using the predictor and the voltage estimation by AD converter. Thus, we obtain transparent and adaptive measurement system.

Important Links:

Go Back