DNN-based fire evacuation risk assessment for large and complex buildings

Ensuring safe and rapid evacuation during fires in large, complex buildings is a major challenge for building managers and emergency responders. Accurate and timely evacuation risk assessment is crucial for effective disaster mitigation and planning. However, existing methods often suffer from limited factor coverage, low efficiency, and poor real-time performance. To address these issues, this study proposes a deep neural network (DNN)-based evacuation risk assessment method. Twelve key influencing factors were identified, and an orthogonal experimental design was used to generate simulation data with PyroSim and Pathfinder. A DNN model was then trained on the data to predict evacuation risks, with its hyperparameters optimized using Bayesian Optimization (BO). A large university library served as a case study. Results show the proposed method accurately predicts evacuation times and effectively improves risk assessment performance.