Multi-unit risk aggregation with consideration of uncertainty and bias in risk metrics

The risk significance of multi-unit events has received much interest, especially since the 2011 Fukushima–Daiichi accident. However, there have been limited experiences in performing a multi-unit probabilistic risk assessment (MUPRA), considering the interactions among multiple reactor units and fuel storage facilities on a site. While considerable research and development efforts have been devoted over the past few years to MUPRAs, there is still no consensus on a unified MUPRA methodology. A site-based risk model is of great importance for comprehensive risk-informed applications and safety goals evaluations. Further, a site-based MUPRA model needs to aggregate risks from various reactor units’ internal and external sequences of events and modes of operation. Risk values obtained from various initiating events and modes of operation, however, are usually biased very differently due to the various factors such as conservatism and modeling assumption used during their developments. As such, to appropriately aggregate various sequences of events, they should be least biased. This paper discusses the framework of a probabilistic aggregation approach for the multi-unit risk metrics to make them least biased using expert elicitation. More importantly, the paper, through a sensitivity analysis, shows that a biased risk metric, even when it is not aggregated, could mask important contributors to risk and thereby yield incorrect risk contributors. This is done by comparing the risk insights from importance measures for various modeling scope and assumptions that make a risk metric biased. Through an example, it is shown the masking of important risk contributors and human error events would be an important impediment in the MUPRAs. The example provides an illustration of the aggregation process of multi-unit risk metrics for both external events and internal events. The paper demonstrates the importance of biased correction before approaching to multi-unit risk aggregation, especially for the external events.