A smart energy cyber-physical system is characterized by the interactions among the sensing devices and the physical energy system. Big data analytics techniques can be leveraged to improve the situational awareness of the energy system. However, how to decide whether the sensor data are trustworthy and how to assess the system state even if those data contain uncertainties are quite challenging. In this talk, I will describe a systematic framework, which is based on partially observable Markov decision process, orthogonal matching pursuit, and empirical mode decomposition, to detect anomaly energy usage behavior through analyzing the massive smart meter data in a community. I will also discuss how this framework can be used for detecting smart grid cyberattacks such as energy theft. I will conclude the talk with some of the ongoing research in this topic.