A King Mongkut's University of Technology Thonburi team compared single-objective linear programming with genetic-algorithm multi-objective optimisation using Thailand's 2024 generation data across 12 source/technology categories and three objectives: unit cost, greenhouse-gas emissions and employment. Three separate LP runs totalled about 0.6 seconds, versus about 4 seconds for MOO, with both using roughly 3,500–3,800 MB. LP produced each objective's extreme optimum, including THB 2,684.14/MWh and 373.00 kgCO2e/MWh, while MOO generated a simultaneous compromise. Potential savings of THB 69,538.44 million, 14.782 MtCO2e and about 638 additional jobs are model scenarios, not realised system outcomes.
Key findings
- Three LP runs totalled slightly over 0.6 seconds versus about 4 seconds for MOO, while memory was similar at approximately 3,500–3,800 MB on the studied machine and settings.
- Actual average cost of THB 3,012.12/MWh fell to THB 2,684.14 under LP and THB 2,964.41 under MOO; LP was lower because it optimised cost alone.
- Actual emissions of 442.72 kgCO2e/MWh fell to 373.00 under LP and 420.91 under MOO for the selected mixes.
- Employment calculated for the actual mix was 3,766.82 jobs and the LP maximum was 4,405.35—about 638 additional jobs—conditional on the employment factors used.
- Annualising across 212,020,370 MWh gave LP potentials of THB 69,538.44 million saved and 14.782 MtCO2e avoided; MOO delivered less extreme values while addressing conflicting objectives simultaneously.
Why this matters globally
The comparison clarifies that the fastest solver is not necessarily the most policy-complete. LP suits one dominant objective and boundary exploration; MOO exposes economic, environmental and social Pareto trade-offs. The logic transfers internationally, but numerical results must be rebuilt for each market, resource mix, network and constraint set.
Thai researcher contribution
The Joint Graduate School of Energy and Environment at KMUTT turned Thailand's 2024 power-system data into a transparent comparison spanning cost, carbon and jobs, supported by KMUTT, TSRI and NSRF. It directly links optimisation engineering with Thailand's just-transition debate.
Limitations to consider
This is a deterministic one-year model aggregating the system into 12 categories with linear relationships. It does not fully represent hourly unit commitment, ramping, reserves, reliability, transmission congestion, storage, curtailment, fuel-price uncertainty, all contractual/legal constraints or renewable variability. Jobs derive from employment factors, not labour-market observation, and savings assume the mix can change without transition costs. Runtime and memory depend on hardware, software, GA settings and repeated random runs; one setup is not a universal benchmark. Each LP optimum addresses only one objective.
Verify the original sources
EnergiesEnergies (MDPI)↗DOI: 10.3390/en19194708