Research On LSTM Regional People Flow Prediction Based On Attention Mechanism

In recent years,under the trend of high-speed and high-quality economic growth,the subway has been involved in the construction of various cities,and urban residents has been used to take subway as a major method of commuting,and the growing subway passenger flow may lead to traffic congestion,which can easily produce hidden safety hazards and traffic problems,bring inconvenience to subway operation and travel of those residents.As a kind of regional passenger flow,subway passenger flow has high research value;Accurate predictions can improve the coordinated management of traffic flow and ease the flow of people to a certain extent;in addition,it also helps the deployment of the advertising market,realizing accurate advertising and saving operating costs.Besides,it is of great significance for promoting green,lowcarbon,environmental protection and energy saving.Through generalizing the methods of short-term passenger flow prediction at home and abroad,a CNN-LSTM model with strong feature extraction ability is proposed according to the advantages and disadvantages of those methods,and on this basis,an improved method is proposed to establish a CNN-LSTM-Attention combination model,further improving the accuracy of short-term passenger flow prediction.The main work of this paper is as follows:(1)Clean and analyze characteristic of passenger flow data.The subway stations is adapted as pedestrian flow areas to collect card swiping data in the entry and exit.Besides,the weather data set is added to form multi-feature passenger flow data.The next step is to clean and process the raw data,and analyze the characteristics of subway passenger flow on the basis of data integration,then carry out subway passenger flow prediction research from the perspectives of weekdays and rest days,Subsequently,carry out passenger flow prediction research in the two directions of inbound and outbound on this basis.(2)Establish a CNN-LSTM short-term passenger flow prediction model.According to the multi-feature input,CNN-LSTM is used to model the short-term passenger flow data prediction,and the prediction results are visualized.(3)The attention mechanism improvement is introduced based on the CNN-LSTM model.The self-attention mechanism is connected to the LSTM layer,which can adaptively assign extracted weights for features at different time steps.Use the adaptive weight allocation which is unique to the self-attention mechanism,and display the predictions of the CNN-LSTM-Attention model in 2D and 3D stereoscopic visualization results from different angles through data training parameters.(4)Carry out comparative experiments and ablation experiments.BP,RNN,LSTM,CNN-LSTM models are used as the benchmark experimental comparison in these experiments.under the evaluation system standard of establishing the prediction results of the model,through analyzing the predictions of ablation experiments and comparative experiments,it is concluded that the CNN-LSTM-Attention combination prediction model proposed in this paper has better performance than other control models and combination models.