Optimizing Thermal Load in Compact Buildings: A Comparative Analysis of Single and Hybrid Metaheuristics for Balanced HVAC Efficiency

Abstract

This work is a comprehensive comparison study of five metaheuristic algorithms — namely GA, PSO, SA, GA-PSO, and GA-PSO-SA — using 3,000 model evaluations across 30 independent runs to optimize compact building thermal loads. Statistical analysis demonstrates that the SA, GA-PSO, and GA-PSO-SA present similar accuracy (RMSE: 3.55±0.16 kWh/m²) without any statistical difference (p> 0.97), contradicting the hypothesis that complexity in hybrid promotes the actual performance. GA appears to be the best compromise, with the highest efficiency ratio (11.47), providing 97.5% of SA's accuracy at 42% lower computational cost. The instability shown by PSO is quite alarming (CV: 14.215%, performance spread: 53.6%) and clearly indicates premature convergence, which sharply contradicts its claim of outperforming in continuous optimization [1]. Sensitivity analysis results show that envelope thermal properties, particularly the wall U-value (NSC=1.43), have 7.5× more influence on prediction performance than building orientation, providing evidence supporting the argument that input data quality outweighs algorithm choice for HVAC design-type applications.

Keywords

• Thermal load optimization
• Genetic algorithm
• Particle swarm optimization
• Simulated annealing
• Single and hybrid algorithms