Use of Bayesian Networks as a Decision Support System for the rapid loss assessment of infrastructure systems - BRGM - Bureau de recherches géologiques et minières Access content directly
Conference Papers Year : 2018

Use of Bayesian Networks as a Decision Support System for the rapid loss assessment of infrastructure systems


The loss assessment of infrastructure systems has emerged as an essential aspect of the risk and resilience analysis of exposed communities. Predicting the performance loss of critical infrastructure before an event is useful to plan mitigation strategies, while a rapid loss assessment in the short-term (i.e. in the crisis period immediately following the disaster) is especially helpful for emergency responders as it contributes to situational awareness (e.g. knowledge of the areas in urgent need of basic utilities, accessibility of strategic locations, etc.). To this end, conventional approaches to model and simulate infrastructure systems include a probabilistic risk framework, where Monte Carlo simulations are performed from the generation of earthquake events to the computation of the system performance indicators. Alternatively, Bayesian Networks (BNs) have been recently used to structure the links and statistical dependencies between the uncertain variables involved in the analysis chain (e.g. earthquake magnitude and location, ground-motion field, damage state of infrastructure components, response of the system, etc.), thanks to the convenient use of conditional probabilities through Bayes' rule. BNs may be used in a predictive (forward analysis), where all sources of uncertainties are propagated in order to obtain a probabilistic distribution of the variables of interest. On the other hand, BNs also have the ability to perform a diagnostic (backward) analysis, where the prior distribution of given variables is updated from evidence collected on fixed variables (e.g. field observations or measures). The latter property is especially relevant in the context of crisis management, since ex-ante predictive loss models may be updated thanks to the resolution of a BN with incoming evidence, thus contributing to a progressive refinement of the estimated consequences of an earthquake event. One of the main issues preventing the application of BNs in an operational context resides in the computational complexity, which generates intractable datasets when large real-world systems are considered. Moreover, formulating an exact BN, which depicts all links between variables, requires the implementation of accurate rules between the components' states (i.e. damage states of individual infrastructure systems) and the performance of the whole system. This constraint limits most BN models to a connectivity analysis, while it has been shown that capacity or serviceability analyses provide a much more accurate picture of the situation. Therefore an approximate BN formulation is presented in the present study, in order to allow for a quick and efficient Bayesian updating of predictive models in near-real-time. The proposed approach is based on two distinct steps, as follows:  Generation of a learning dataset through Monte Carlo simulations: thousands of loss scenarios, accounting for all types of uncertainties, are sampled through infrastructure modelling and simulation tools (e.g. OOFIMS platform from the FP7 SYNER-G project).
Fichier principal
Vignette du fichier
2018-06_earthquake.pdf (1.19 Mo) Télécharger le fichier
abstract1_BN.pdf (279.35 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Origin : Files produced by the author(s)

Dates and versions

hal-01654871 , version 1 (04-12-2017)


  • HAL Id : hal-01654871 , version 1


Pierre Gehl, Francesco Cavalieri, Paolo Franchin, Caterina Negulescu, Kristel Carolina Meza Fajardo, et al.. Use of Bayesian Networks as a Decision Support System for the rapid loss assessment of infrastructure systems . 16th European Conference on Earthquake Engineering - 16ECEE, Jun 2018, Thessalonique, Greece. ⟨hal-01654871⟩


107 View
92 Download


Gmail Facebook X LinkedIn More