This paper proposes BLTranAD-NR, a deep learning model, for time series data anomaly detection, which improves the BLTranAD. The residual link is introduced into the model, so that the model can directly pass the input information to the subsequent network layer, thereby avoiding the loss of information in the deep transmission process. In addition, a layer normalization operation is added after the bidirectional LSTM. Layer normalization standardizes the features of each sample, allowing the model to better balance the information flow in forward and backward propagation during training. In view of the problems of high data dimension and sample imbalance in network traffic, data enhancement technology based on Gaussian mixture model is used to generate more abnormal samples. GMM can capture the potential distribution of data, thereby generating representative and high-quality synthetic data and increasing data diversity.