Owing to the lack of early fault alert mechanisms for smart meters, a method that characterizes the operating condition of the smart meters using Sate Model is proposed. The technique is based on a variational feature state space matrix, which is made up of feature parameters that were taken from the power consumption change data. The state model is automatically generated from the variational feature state space matrix using the variational auto-encoder. By calculating the difference between a fixed normal state model and the real-time state model and comparing it to an automatically created number, smart meters can provide an early warning. The suggested approach's efficacy is demonstrated through experiments, and its properties are contrasted with the method based on high-dimensional feature space. The proposed approach can alarm earlier than the comparison method and has an accuracy rate for the frequent smart meter problems of up to 91.67%.