Signal quality based power output prediction of a real distribution transformer station using M5P model tree

Abstract

Forecasting the output power of transformer stations is a prominent analysis of power systems, which is key aspect of load balancing. Moreover, power forecasting may lead to maximized energy efficiency in the electrical grid. In this study, a fully operational real distribution transformer station which is resident in Istanbul is investigated. Transformer’s electrical characteristics which are strictly related with signal quality, are recorded every fifteen minute for four months, are employed to predict power output of the transformer. As a signal quality concern, a local electrical distribution company must supply stable and distortion free signal waveforms to the consumers where frequency-voltage stability, reduced THD values are satisfied. In order to estimate power output of the transformer, these parameters are subjected to well-known machine learning (ML) methods and consequently most effective methods are selected. In this paper M5P tree algorithm is promoted for total active power estimation of the analyzed transformer station. The M5P results have been compared with the recorded results using Root Mean Square Error (RMSE) and Correlation Coefficient (R) as 78.55 and 0.986, respectively. According to the results, the M5P model is performed quite efficiently in power estimation for the output power of distribution transformer station.