The invention discloses an abnormal driving behavior online identification method based on an Encoder-Decoder attention network and an LSTM (Long Short Term Memory). The method is composed of three main modules, namely an encoder-decoder based on LSTM, an attention mechanism and a classifier based on SVM, and comprises the steps of input encoding, attention learning, feature decoding, sequence reconstruction, residual calculation and driving behavior classification. According to the method, on the basis of mobile phone multi-sensor fusion data, on the basis of driving behavior data characteristics and behavior pattern analysis, an Encoder-Decoder deep learning model, an Attention attention mechanism and an SVM classification model are fused to recognize abnormal driving behaviors. The method has the advantages of being easy in data acquisition, non-intrusive, low in cost and the like, not only considers the time correlation of the driving behaviors, but also considers the difference of different moments, can perform online identification on abnormal driving behaviors in an end-to-end mode, can provide a method basis for driving behavior evaluation and safety early warning, and has a good application prospect. The method has significant meaning for intelligent driving system design and traffic safety decision making.