Deep Learning and Explainable AI Based Blast Wave Pressure Prediction in IoV Applications

Abstract

Predicting blast wave pressure is crucial for enhancing safety and response strategies in Internet of Vehicles (IoV) applications, particularly in the context of urban environments and high-risk areas. This paper presents a novel deep learning model designed for predicting blast frequencies in BLEVE (Boiling Liquid Expanding Vapor Explosion) scenarios. The article evaluates several activation functions, including ReLU, Mish, ELU, Silu, and Leaky ReLU, to determine their effectiveness in improving model accuracy. The results indicate that the Mish activation function achieves slightly lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) scores on the test dataset, with Train RMSE at 0.075 and Test RMSE at 0.099. The Silu activation function also demonstrates strong performance, yielding a Train RMSE of 0.074 and Test RMSE of 0.095, alongside high R² scores, indicating a good fit to the data. The paper further explores the interpretability of the model’s predictions using the LIME framework, revealing insights into how sensor positions and vapor heights influence blast frequency predictions. This research explores the potential of deep learning techniques in enhancing safety measures and predictive capabilities in hazardous scenarios, contributing valuable insights to the field of risk assessment and management in IoV applications.