Efficient Simulation of a Leak-Detection-and-Repair Program

Monte Carlo (MC) simulations are a common way to estimate the methane emissions and cost-effectiveness of leak detection and repair (LDAR) programs in the upstream oil and gas industry. In this paper we consider a simplified version of a LDAR program and demonstrate how to simulate the underlying system in an efficient manner, resulting in estimators of the quantities of interest that have a smaller variance than is possible using contemporary techniques. The proposed method relies on two ideas: the first is to leverage the underlying stochastic models to perform an event-driven simulation rather than a daily one; and the second is to use low-discrepancy sampling rather than plain random sampling to generate samples of potential scenarios for the underlying system, which results in a more systematic and balanced exploration of the scenario space. We show that in the context of a sensitivity analysis example, the proposed approach reduces the error of the estimates by factors of about 3 to 4 for the same computation time. This increased precision can provide conclusive statistical evidence that an LDAR program significantly reduces emissions compared to another, while the naive method's error is often too large to draw any conclusion.