Kazem Samimi, Esmaeil Zareie, Mohsen Omidavar, Javad Ghyasi, Parham Azimi, Mostafa Pouyakian,
Volume 15, Issue 3 (10-2025)
Abstract
Introduction: Fire risk assessment in oil storage tanks faces challenges due to incomplete, conflicting, and uncertain data, particularly when empirical evidence is limited. Traditional point-based likelihood estimates often fail to capture expert doubt and epistemic uncertainty. This study aims to develop and evaluate a novel hybrid framework combining Dempster-Shafer Theory (DST) and Bayesian Networks (BN) to improve the trustworthiness of fire risk prediction in such industrial settings.
Material and Methods: The proposed approach integrates DST to model expert uncertainty through interval probabilities (Bel–Pl) and BN to dynamically update causal relationships as new information appears. The study implements computational coding to enable DST calculations for five expert opinions across 243 scenarios, overcoming prior limitations in multi-expert modeling due to computational complexity.
Results: The hybrid DST-BN framework demonstrated superior ability to incorporate incomplete and conflicting expert data, reducing overconfidence linked to point estimates. Interval probabilities offered more trustworthy representations of epistemic uncertainty, while BN integration allowed traceable and adaptable causal modeling. The computational solution facilitated practical application of DST with multiple experts, enhancing the strength of the risk assessment.
Conclusion: This research provides an effective DST-BN hybrid methodology for assessing fire risk in fixed-roof oil tanks, improving accuracy and trustworthiness in complex industrial environments. By addressing the shortcomings of point-based methods and enabling multi-expert participation, the framework supports clearer and more defensible probabilistic inferences. Future work may focus on integrating real-time sensor data and AI-based decision systems to further strengthen dynamic risk assessment capabilities.
