The central result is a modeled cost comparison. For a representative closed-loop pumped-storage hydropower project, the optimized layout was assigned a total capital cost of USD 727 million, compared with USD 774 million for an initial geomorphon-heuristic layout. The paper reports this as a 6.1% reduction, or about USD 47 million, in modeled total project CAPEX.
A map-based search narrows the field
HERA-S is built as a three-part workflow linking geographic information systems, or GIS, engineering and optimization. It applies spatial filters to maps, searches for viable closed-loop site pairs, and then builds preliminary engineering designs while optimizing reservoir geometry.
The Rio de Janeiro case used state boundaries, transport routes, transmission lines and substations, along with regional geology and federal and state conservation-area layers. Its elevation input was a 90-meter SRTM digital elevation model, a gridded map of terrain height. The first topographic screen kept closed-loop pairs with more than 400 meters of gross head, or height difference, and a length-to-head ratio no greater than 10 to 1. Later screens considered distance to substations or transport corridors, protected reserves, geology and local watershed area.
From candidate sites to reservoir geometry
After those filters, the intensive search evaluated 1,000 MW closed-loop configurations with storage durations from six to 120 hours. It identified candidate sites for 12-hour systems. The paper also presents duration scenarios at a single 1 GW candidate site, with reported unit costs below USD 20 per kWh for multi-day configurations longer than 72 hours.
For the detailed example, the model represented a 1,000 MW, 12-hour project with 12,000 MWh of energy and about 14.4 million cubic metres of storage at a gross head of 424 metres. It treated reservoir geometry as a mixed-integer program, a mathematical search in which some terrain cells can be selected or rejected, with the goal of minimizing civil expenditure while meeting storage-volume and spatial-connectivity requirements.
Keeping the proposed reservoirs connected
Connectivity was a central engineering constraint. The formulation used directional separating-plane constraints and a Traveling Salesman Problem, or route-planning, formulation to prevent selected reservoir interiors from breaking into disconnected pieces and to avoid perimeter subtours, or loops that do not belong to the intended boundary.
The scale of the search made computational shortcuts important. HERA-S used separating planes and a progressive zoom-in approach: it worked on a coarse grid, clipped the area around a candidate boundary, and then returned iteratively to the native elevation-map scale. The paper reports optimal or near-optimal solutions for grids of up to 266 by 266 cells.
The savings remain inside the model
In the reported FICO Xpress solver run, the initial optimality gap, the solver's measure of distance from its best bound, exceeded 60%. It reached near zero in 320 seconds. The run found 61 feasible integer solutions, and the optimized reservoir cost was USD 127 million.
The lower total figure included lower-reservoir savings associated with using material excavated locally to a depth of up to 4 metres for retaining embankments. On the same modeled basis, the unit storage cost was about USD 64 per kWh, based on the USD 727 million total CAPEX. The paper says that figure was lower than an equivalent utility-scale battery energy-storage system for 12-hour discharge.
What the comparison does not settle
These comparisons should be read as outputs of one regional pre-feasibility exercise. The supplied analysis reports no uncertainty quantification, sensitivity analysis or external validation, and it does not provide detailed assumptions for the battery-storage comparator. The case was set in Rio de Janeiro State, so the reported site choices and costs have not been shown to transfer to other regions or to detailed surveys, construction or operation.
The figures therefore describe what the HERA-S workflow estimates under its stated spatial, engineering and cost assumptions. They do not establish that a candidate site will be built, that a facility will operate as modeled, or that pumped-storage hydropower is generally cheaper than battery storage.
A preprint with industry support
The supplied document is an arXiv preprint, version 1, dated 25 August 2026. It states that HERA-S was developed by PSR with support from EDF, CTG, Brookfield and Rio Light. The authors present the workflow as decision support for regional closed-loop pumped-storage prospecting and pre-feasibility cost optimization, while the supplied evidence does not include a field or operating test.
Paper data and sources
Original title: Optimization of Closed-Loop Pumped-Storage Hydropower Siting
Authors: Luiz Rodolpho Albuquerque, Rafael Kelman, Tarcisio Castro et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-25
DOI: Not available
Original paper · Full text