Probabilistic safety assessment and optimal control of hazardous technological systems: A marked point process approach: Dissertation

Probabilistic safety assessment (PSA) and decision analysis are methods used for supporting risk management of hazards arising from technological systems. These methods are applied more often also in operational risk management, for instance, in the nuclear safety field. Operational risk management sets new requirements for modelling of systems and problems, since the context is dynamic compared with the static decision-making situation assumed in conventional risk and decision analysis approaches. This thesis applies a marked point process approach to represent dynamically the hazards of a technological process. The approach is applied here to risk follow-up and the problem of optimal control. Risk follow-up by PSA provides a systematic method for analysing incidents. In a retrospective risk assessment, operational events can turn out to be important in several respects. In order to highlight such differences, several alternative approaches should be used in parallel, as presented in this thesis. A period of actual operating history from a Finnish nuclear power plant is analysed. The thesis models risk management as an optimal control problem for a stochastic process. The approach classes the decisions made by management into three categories according to the control methods of a point process: (1) planned process lifetime, (2) modification of the design, and (3) operational decisions. The approach is used for optimization of plant shutdown criteria and surveillance test strategies of a hypothetical nuclear power plant. The thesis promotes use of the utility function as the objective function in optimization of risk management strategies. Compared with present approaches based on e.g. probabilistic safety criteria and ALARP principle (As Low As Reasonably Practicable), the utility theory would increase coherence in the analysis of different problems. The choice of utility function is here related to the problem of risk acceptance, i.e. probabilistic safety criteria are analysed using a utility function model. Conditions for a utility function satisfying the risk acceptance criterion are derived.