Accelerating system adequacy assessment using the multilevel Monte Carlo approach

Simon Tindemans,Goran Strbac

doi:10.1016/j.epsr.2020.106740

Abstract

Accurately and efficiently estimating system performance under uncertainty is paramount in power system planning and operation. Monte Carlo simulation is often used for this purpose, but convergence may be slow, especially when detailed models are used. Previously published methods to speed up computations may severely constrain model complexity, limiting their real-world effectiveness. This paper uses the recently proposed Multilevel Monte Carlo (MLMC) framework, which combines outputs from a hierarchy of simulators to boost computational efficiency without sacrificing accuracy. It explains which requirements the MLMC framework imposes on the model hierarchy, and how these naturally occur in power system adequacy assessment problems. Two adequacy assessment examples are studied in detail: a composite system and a system with heterogeneous storage units. An intuitive speed metric is introduced for easy comparison of simulation setups. Depending on the problem and metric of interest, large speedups can be obtained.

Highlights

Operational and planning problems in the power system domain often involve the assessment ofsystem performance across a range of probabilistically modelled scenarios
This paper considers how the Multilevel Monte Carlo (MLMC) framework [5] can be used to accelerate risk calculations, in particular in applications relating to system adequacy assessment of complex systems
This paper has set out how the MLMC approach can be applied to power system risk analysis, and to system adequacy assessment problems

Summary

Introduction

Operational and planning problems in the power system domain often involve the assessment of (sub-)system performance across a range of probabilistically modelled scenarios. For all but the simplest power system models, this cannot be done analytically, and Monte Carlo (MC) simulations are used instead. MC simulations are a powerful general purpose computation method with a long tradition in power system applications [1], but convergence to the correct answer may be slow. A number of different variance reduction methods exist to speed up convergence of Monte Carlo estimates, e.g. Importance sampling, has recently grown in popularity for power system applications, especially in combination with automatic tuning of model bias parameters using the cross-entropy approach [3], [4]. Implementing importance sampling typically requires deep insight into the model, and limits the design freedom, e.g. for simulations involving complex decision making or sequential actions

Objectives

Methods

Findings

Conclusion