Page 257 - Intelligent Digital Oil And Gas Fields

Integrated Asset Management and Optimization Workflows 207

Generation 1 Generation 4

Misfit Parameter 2 Misfit Parameter 2

Misfit Parameter 1 Misfit Parameter 1

Generation 7 Generation 11

Misfit Parameter 2 Misfit Parameter 2 2D Pareto front

Misfit Parameter 1 Misfit Parameter 1
Fig. 6.4 Behavior of the two-component objective space in a history matching-
workflow using the multiobjective genetic algorithm (MOGA) as a function of a number
of minimization iterations.

The main attributes of local optimization methods can be summarized as
follows:
• they generally seek a local solution and depend on derivatives of the cost
function and constraints;
• the solution is not guaranteed to have the lowest objective (when min-
imization) or the highest objective (when maximization) among all the
feasible points;
• they can be solved relatively easily, because they require the differentia-
bility of the objective and constraints;
• they require initial guess (estimate); and
• they are frequently supported by a solid convergence theory.