Skip to main content

Hybrid multi-criteria and machine learning framework for site selection of PRO–based salinity gradient power plants

Abstract

Pressure Retarded Osmosis (PRO) is considered a promising approach to salinity-gradient energy harvesting at freshwater–seawater interfaces and enhances integrated water–energy systems. But the efficiency and sustainability of PRO power generation systems are very sensitive to the site choice, requiring integrated evaluation of environmental, infrastructural, and economic factors. In this study, we propose a PRO-specific hybrid decision-support framework combining the AHP and GMDH to merge a structured multi-criteria prioritisation framework with a nonlinear interaction-based feedback-refinement methodology for a PRO-specific process. A key contribution of this work is the establishment of a standalone PRO Location Suitability Index (PLSI) that explicitly incorporates key PRO operational sensitivities into a unified hybrid AHP–GMDH decision structure, thereby allowing process-aware macro-scale screening framework. The study employs comparative case studies on Oman, Norway, and Australia, contrasting hydro-climatic and infrastructural conditions, to illustrate the methodology. According to the results Oman is the highest (PLSI = 0.718), followed by Norway (0.526) and Australia (0.256). This results in factor prioritization whereby coastal length (30.62%) and maintenance infrastructure availability (18.67%) are identified as the identified as the most influential factors within the evaluated framework. The stability of rankings was further tested in terms of factor configurations (top-5, top-10, and full 20 factor sets) and sensitivity tests of weight-perturbation, indicating stable prioritization within the tested uncertainty range under plausible uncertainty. The new proposed AHP–GMDH–PLSI framework can be summarized as a a potentially scalable and transferable toolto aid early-stage PRO siting and sustainable salinity-gradient power planning in the wider water–energy nexus.