Detection of False Data Injection of PV Production

Abstract

Due to cyber attack threats to the cyber physical systems which compose modern smart grids additional layers of security could be valuable. The potential of data tampering in the smart grid spurs the research of data integrity attacks and additional security means to detect such tampering. This paper conducts a study of photovoltaic based production data tampering as a detection problem and shows a set of machine learning models and highlights the best performing of the set at the detection task. The signal is observed daily and data tampering by increasing to 110%-150% of original signal is detected with over 80% accuracy and under 10% false alarm. This paper finds that the artificial neural network (ANN) slightly out performs the support vector machine (SVM) at the detection task, however the SVM is a much faster algorithm to fit the data with.