Imprecise probabilistic framework for power grids risk assessment and sensitivity analysis

Reliability of electric power supply and its robustness are major concerns for the future interconnected power grid. A comprehensive analysis of vulnerabilities and risks should take into account rare threatening scenarios such as cascading sequences, malicious attacks, sabotages, extreme weather conditions but also less unlikely events such as components tripping or malfunctioning. Generally speaking, frameworks for probabilistic risk assessment rely on a variety of model assumptions and availability of good quality data. Many probabilistic metrics for power networks risk assessments have been introduced and consider uncertainties. Nevertheless, several cases are the ones for which the available information is partially corrupted or of scarce quality, therefore classical probabilistic approaches are difficult to be implemented with-out subjective assumptions. Considering the reviewed works in literature, many lack to account for situations where the data is affected by imprecision or information have poor quality or is limited. The authors believe that further efforts should be devoted to the improvement of probabilistic risk assessment frameworks by including rational treatment of imprecision in the data and without introducing unwarranted assumptions. In this work, a power grids risk assessment framework based on imprecise probability theory is introduced. Adopting an imprecise probabilistic framework, relevant sources of uncertainty have been propagated through numerical model of the gird and their effects quantified in the risk metrics. In particular, the effect of imprecise data describing the stochastic load model and lack of information about the system components (e.g. line failure rate) has been analysed. Computational time is indeed a strong burden for imprecise probabilistic frameworks. Hence, linear approximation of sensitivity measure, i.e. line outage distribution factors, and parallelization strategy has been used as valuable tool to reduce the computational wall-clock time for the analysis. The framework has been tested on a realistic power network, two different propagation strategies adopted and sensitivity analysis for imprecise input performed. The results are compared and discussed, both aleatory and epistemic uncertainties propagated in the risk index and the most relevant sources of epistemic uncertainty identified using sensitivity analysis.