Quantitative risk analysis is essential for power system planning and operation. Monte Carlo methods are frequently employed for this purpose, but their inherent sampling uncertainty means that accurate estimation of this uncertainty is essential. Basic Monte Carlo procedures are unbiased and, in the limit of large sample counts, have a well-characterised error distribution. However, for small time budgets and ill-behaved distributions (such as those for rare event risks), we may not always operate in this limit. Moreover, multilevel Monte Carlo was recently proposed as a computationally efficient alternative to regular Monte Carlo. In this approach, great asymptotic speedups are achieved by reducing the number of full model evaluations. This further challenges the assumption that normally distributed errors can be used. This paper investigates the sampling error distributions for a practical resource adequacy case study, in combination with the Multilevel Monte Carlo method. It further proposes a practical test for validating error estimates, based on a bootstrap approach.

@en

For planning of power systems and for the calibration of operational tools, it is essential to analyse system performance in a large range of representative scenarios. When the available historical data is limited, generative models are a promising solution, but modelling high-dimensional dependencies is challenging. In this paper, a multivariate load state generating model on the basis of a conditional variational autoencoder (CVAE) neural network is proposed. Going beyond common CVAE implementations, the model includes stochastic variation of output samples under given latent vectors and co-optimizes the parameters for this output variability. It is shown that this improves statistical properties of the generated data. The quality of generated multivariate loads is evaluated using univariate and multivariate performance metrics. A generation adequacy case study on the European network is used to illustrate model's ability to generate realistic tail distributions. The experiments demonstrate that the proposed generator outperforms other data generating mechanisms.@en

Monte Carlo simulation is often used for the reliability assessment of power systems, but it converges slowly when the system is complex. Multilevel Monte Carlo (MLMC) can be applied to speed up computation without compromises on model complexity and accuracy that are limiting real-world effectiveness. In MLMC, models with different complexity and speed are combined, and having access to fast approximate models is essential for achieving high speedups. This paper demonstrates how machine-learned surrogate models are able to fulfil this role without excessive manual tuning of models. Different strategies for constructing and training surrogate models are discussed. A resource adequacy case study based on the Great Britain system with storage units is used to demonstrate the effectiveness of the proposed approach, and the sensitivity to surrogate model accuracy. The high accuracy and inference speed of machine-learned surrogates result in very large speedups, compared to using MLMC with hand-built models.@en