Abstract: Abstract: In this paper, we propose a malicious traffic detection method based on a multi-model parallel fusion network to address the issue of incomplete reflection of original traffic information and reduced accuracy in identifying malicious traffic by a single sequential deep learning detection model. This method adopts a parallel approach, combining one-dimensional convolutional neural network (1D-CNN) and bidirectional long short-term memory network (Bi-LSTM) for feature extraction and traffic recognition. Each branch directly focuses on the original traffic, extracting both spatial and temporal features. A common fully connected layer is used for feature fusion, which can more accurately reflect the original traffic information and effectively improve the accuracy of malicious traffic identification. Experimental results on the open-source UNSW-NB 15 dataset demonstrate the superior performance of the proposed method in terms of feature extraction capability, robustness, online learning ability and and detection rate for malicious traffic detection.