Comparison Between Robust and Stochastic Optimisation for Long-term Reservoir Operations Under Uncertainty

TL;DR

This paper compares robust and stochastic optimization methods for long-term reservoir management under uncertainty, proposing a new methodology that provides formal guarantees and combines both paradigms for improved solution confidence.

Contribution

It introduces a methodology that derives reservoir bounds with formal guarantees, compares robust and stochastic approaches, and combines them to enhance solution confidence.

Findings

Solutions vary substantially between robust and stochastic methods.

Combining approaches allows assigning confidence levels to solutions.

Methodology is efficient, easy to implement, and guarantees global optimality.

Abstract

Long-term reservoir management often uses bounds on the reservoir level, between which the operator can work. However, these bounds are not always kept up-to-date with the latest knowledge about the reservoir drainage area, and thus become obsolete. The main difficulty with bounds computation is to correctly take into account the high uncertainty about the inflows to the reservoir. In this article, we propose a methodology to derive minimum bounds while providing formal guarantees about the quality of the obtained solutions. The uncertainty is embedded using either stochastic or robust programming in a model-predictive-control framework. We compare the two paradigms to the existing solution for a case study and find that the obtained solutions vary substantially. By combining the stochastic and the robust approaches, we also assign a confidence level to the solutions obtained by stochastic programming. The proposed methodology is found to be both efficient and easy to implement. It relies on sound mathematical principles, ensuring that a global optimum is reached in all cases.