Evaluation of Multi-Objective Methods for Portfolio Tradeoff Analysis

Abstract

AbstractRecent advances in multi-objective optimization technology enable improved solution quality, scalability, and flexibility for portfolio analysis.A multi-objective evolutionary algorithm (MOEA) with innovative constraint handling was used for portfolio planning of offshore exploration & production projects. The MOEA method was used for tradeoff analysis and portfolio optimization. In order to facilitate direct comparison with alternative methods, a multi-objective evaluation framework was developed, and a simplified test problem was derived from historical planning data. The MOEA method was compared to three other portfolio planning methods: rank & cut, sample & filter, and mixed integer linear programming (MILP).The study highlighted that the evaluation of multi-objective methods is itself a multi-objective problem. Methods that performed well on some criteria performed poorly on others, and no method dominated on all criteria. The presence of planning constraints proved critical, with unconstrained problems being easy to solve with traditional rank & cut, and constrained problems requiring some type of optimization method.The MOEA method demonstrated several advantages. Even though the test problem was designed specifically to be a good match for the MILP method, the MOEA method was competitive on all criteria and excelled on solution spread, flexibility, and scalability. Published MOEA methods have the potential to enable new types of portfolio tradeoff analysis, although successful applications will likely require enhancements to published constraint handling techniques.Multi-objective methods are in the early phases of petroleum industry adoption, so there is little knowledge of appropriate solution method evaluation criteria. The study highlights important and commonly overlooked issues in evaluating multi-objective methods.