Deep learning of network traffic is a research method that mimics the structural functions of the human nervous system to identify, classify, and predict data. We propose a new model based on Conv-LSTM to improve the accuracy and efficiency of network encrypted traffic recognition. Based on the public CIC-ISD2017 dataset, the new model is tested and measured, and evaluated based on the constructed confusion matrix and ROC graph. Comparing it with traditional Conv-LSTM, decision tree method, and RF&LSTM methods, it was found that the new model performs better and can perform well in multi classification tasks, with an accuracy rate of up to 99.60%. This model provides a reference solution for relevant applications in the field of network security.