Urban Traffic Flow Prediction Algorithm Based on Attention Mechanism and LSTM

Traffic flow prediction is one of the important techniques for urban intelligent transport construction. Due to the instability and complexity of the actual traffic flow changes in urban road networks, existing prediction models are difficult to extract the flow characteristics of traffic sections that are geographically on different grids at different times with limited training. Regarding this kind of problem, this paper proposes a prediction method based on transformer attention mechanism and multilayer LSTM model, firstly, the temporal characteristics of data are processed by four-layer LSTM network, then the attention mechanism of multiple heads is used to make predictions on spatial location characteristics, finally, the prediction results analysed by both of them are subjected to full-connectivity processing and constant adjustment of weights, and the extracted traffic flow characteristics with temporal and spatial characteristics are finally processed and outputted. spatial characteristics of the city is finally processed and the prediction results are output. After the related prediction experiments on the grid flow data and the related evaluation indexes of the comparative model experiments, the prediction effect of this model is better than the traditional prediction method, and it can make effective long- and short-term flow prediction of the traffic flow of the urban roads.