Machine Learning Techniques to Predict Voltage Unbalance in a Power Transmission System

Abstract

Voltage unbalance is a growing issue that, among other things, can impact three-phase motor and drive loads, result in nuisance tripping of generation units and capacitor banks, and prevent optimization of conservative voltage regulation strategies. This difference between the three phases of voltage delivered to customers can damage the equipment of these customers as well as negatively impact the power system itself. This work presents an approach for predicting voltage unbalance using machine learning. Historical megawatt and megavar data–obtained through a Supervisory Control And Data Acquisition (SCADA) system–are used to train an artificial neural network model as a binary classifier with a portion of the data serving to validate the trained model. Voltage unbalance is predicted at an accuracy above 95% for eight substations within the power utility's extra-high voltage transmission network and over 91% for all 42 substations. The trained model is tested in a manner that would be employed using simulated data generated by state estimation software. This simulated data validates the model's capacity to predict the substation buses that would experience voltage unbalance.

Keywords

• Artificial neural network (ANN)
• automation
• classification
• prediction
• prediction model
• supervisory control and data acquisition (SCADA)