基于深度卷积神经网络的人员疏散时间快速预测方法

Evacuation time prediction is of critical importance to the management and control of pedestrian crowds and optimizing building layout. Current researches focus on establishing theoretical evacuation models. However, due the high spatial-temporal complexity of the evacuation models, it is difficulty to realize rapid evacuation time prediction. The complex building space environment and the spatial distribution of pedestrians are the decisive factor that determines the length of the evacuation time. This paper proposed a fast calculation method to determine evacuation time based on the image of the building layout and the distribution of pedestrians. Evacuation scenarios considering different number of evacuees, different number of exits, different exit widths, and different space layout forms have been designed. Then, with a cellular automata evacuation model (CA), simulations have been performed to build an evacuation database. Next, a fast evacuation time prediction model, namely, FastEvaNet, has been proposed. This deep convolutional neural network model was then trained and tested with the simulation database. Then, the reliability and efficiency of evacuation time prediction model has been evaluated. The results show that the MAPE value is 8.21%. The overall prediction accuracy of the model is good, and the generalization ability is strong. What is more, the computational time of evacuation time is almost independent of the complexity of the evacuation scenario. The computational efficiency is three orders of magnitude higher than that of the CA model. When the data flow is transmitted to the network for batch prediction, the efficiency can be four orders of magnitude higher.