A new benchmark optimization problem of adaptable difficulty: theoretical considerations and practical testing

In this paper, we present a new benchmark problem for testing both local and global optimization techniques. This problem is based on ideas from groundwater hydraulics and simple Euclidian geometry and has the following attractive features: (a) known values of the infinite global optima, which can be classified in a restricted number of sets, with known location in the search space (b) simple form and (c) quick computation of objective function values. Moreover, the number of local optima sets, their location in the search space and thus the respective values of the objective function can be easily determined by the user, without affecting the global optimum value. In this way, the difficulty of finding the global optimum can be changed from quite small to almost insurmountable, as demonstrated by applying five widely used optimization methods, namely genetic algorithms, sequential quadratic programming, simulated annealing, Knitro and branch and bound. Moreover, some observations on the different behavior of optimization methods are discussed.