Critical power system assets are becoming more and more digital, which makes them much more vulnerable to online dangers. Traditional tracking methods often can’t find strikes that are planned or done in secret, so more advanced methods are needed. This work proposes a scientifically robust methodology for effective tracking that integrates graph-theoretic modelling, state estimation, and probabilistic reasoning. Statistical theories are used to build intruder and anomaly detection algorithms, and assessment metrics reveal how accurate and reliable the detection is. At result we found that the deep neural network (DNN) are better than the machine learning model to discovering the pattern and effective for monitoring.