In order to accurately predict the traffic flow and fully extract the complex linear and nonlinear features and their dependence in the traffic flow, a CLABEK model integrating multi-dimensional spatial-temporal features is proposed. Among them, the temporal and spatial features, periodic features and additional features (holidays, weather conditions and temperature) are extracted by Conv-LSTM, BiLSTM and Dense neural networks respectively, and the nonlinear features of traffic flow are comprehensively obtained by combining the above models; then the linear features of traffic flow are extracted by Kalman Filtering. The comparative experiments on public data sets show that the CLABEK model performs the best in short-term traffic flow prediction task.