SIRO: A Deep Learning-Based Next-Generation Optimizer for Solving Global Optimization Problems

Abstract

This paper introduces the SIR Optimizer (SIRO), a novel next-generation learned metaheuristic algorithm inspired by biological systems and deep learning techniques. The optimizer uses the susceptible-infected-removed (SIR) epidemiological model to predict the population’s susceptibility, active infections, and recoveries. To enhance the search process, SIRO incorporates deep learning into its initialization and parameter tuning components, enabling intelligent and autonomous behaviour. By generating initial solutions based on neural models, the algorithm achieves efficient, effective, and robust search outcomes. To validate the effectiveness of SIRO, a set of numerical hybrid test functions from the CEC 2017 benchmark, each characterized by 30 dimensions were utilized. The experimental results were compared against various state-of-the-art algorithms, demonstrating that SIRO outperforms its competitors. Moreso, it delivers high-quality solutions while utilizing fewer control parameters. The incorporation of a learning process in SIRO leads to superior precision and computational efficiency compared to other optimization approaches in the existing literature.