New Hybrid Optimization Methodology to Identify Pollution Sources Considering the Source Locations and Source Flux as Unknown

Abstract

A hybrid methodology was developed for identifying unknown groundwater pollution sources. The source identification problem can be solved by using the inverse optimization approach. Being a mixed-integer problem, genetic algorithms (GAs) can be applied to solve the inverse optimization problem. Even though GA can determine the discrete variable efficiently, it is not efficient in obtaining the continuous variables. For such a situation, the gradient-based local search optimization is considered to be an efficient approach. Therefore, because of the advantages of GA and gradient-based approach, they are combined to form an improved approach. GA has been modified to handle the source variables of locations and fluxes differently. It has been observed that GA often gives near-optimal solution. As such, three local location search algorithms, i.e., longitudinal-transverse search (LTS), mutation search (MS), and ripple-migration search (RMS), have been proposed. The efficiency and the applicability of the proposed model were evaluated by applying it in a hypothetical study area. The results show that the proposed model is capable of obtaining optimal source locations and fluxes. The results obtained are superior to those obtained by using embedded optimization method. However, when computational efficiency was compared in terms of the number of function evaluations, it was found that the number of function evaluation for the LTS algorithm was minimum.