Short-Term Traffic Flow Prediction: A Method of Combined Deep Learnings

Short-term traffic flow prediction can provide a basis for management and support travelers to make decisions. Accurate short-term also provides necessary conditions the sustainable development of environment. Although application deep learning methods has achieved good accuracy, problem combining multiple improve accuracy single method still margin in-depth research. In this article, combined (CDLP) model including two paralleled models, CNN-LSTM-attention CNN-GRU-attention model, is established. one-dimensional convolutional neural network (1DCNN) used extract local trend features RNN variants (LSTM GRU) with attention mechanism are long temporal dependencies features. Moreover, dynamic optimal weighted coefficient algorithm (DOWCA) proposed calculate weights goal minimizing sum squared errors CDLP model. Then, neuron number, loss function, optimization algorithm, other parameters discussed set through experiments. Finally, training test established processing data collected from field. The trained tested, results obtained analyzed. It indicates that fit change very well better performance. Furthermore, under same dataset, compared baseline models. found higher than