The district transformer meter is an electrical energy metering device installed on the low-voltage distribution transformer, serving as a key component for electricity consumption data collection. Timely fault detection and identification are crucial aspects of the daily management of district transformer meters. However, existing methods mainly relies on expert experience or simple indicator-based methods, which results in issues such as insufficient diagnostic timeliness, low levels of intelligence and automation, and inefficient manual troubleshooting. When various wiring kinds of faults occur on transformer meters, the characteristics of the measured data are highly related to the fault type. Aiming at this key characteristic, a fault identification method of the meters based on Boruta and TimesNet model is proposed in this paper. First, the Boruta algorithm is used to screen the original features and select all the features that have significant contributions to the target variables. The TimesNet model is then used to convert one-dimensional time series into two-dimensional tensors to better capture time changes within and across periods. The model realizes supervised learning classification of labeled fault samples through feature extraction across time scales. An example analysis based on the measured data of 10kV station area in Shanghai proves the effectiveness of the proposed method.