Traditional air conditioning fault sensing methods require intrusive installation of sensors inside the system with high data acquisition frequency, causing that the conditions of Widespread Internet of Things cannot support these methods. Aiming at the above problems, this paper proposes two methods with only smart socket and temperature sensors. The frequency of data collection is low, and the amount of data storage is reduced by the mutation preservation mechanism. The first method proposes an approximate physical model based on first-order equivalent thermal parameter model, the residual value analysis of prediction and measured values is used to detect real-time faults. The second method is based on support vector data description (SVDD). The model derives a description of the normal data and enables online separation of normal and fault data. The results show that the real-time false positive rate of Method 1 is 1.46%, and that of Method 2 is zero.