Leopard Tortoise Optimization Algorithm

To address the issues of oscillation, premature convergence, and highly redundant search paths inherent in traditional swarm intelligence optimization algorithms for high-dimensional, multi-peak, and plateau-type optimization problems, this paper proposes a novel optimization algorithm—the Leopard Tortoise Optimization Algorithm. This algorithm does not simply simulate animal behavior; instead, it starts from the structural characteristics of real-world complex optimization problems, introducing the concept of non-uniform steady-state search. It treats the search space as a holistic system composed of multiple stable patches and sparse reachable channels. The algorithm achieves dynamic pruning and structured exploration of the search space through steady-state patch modeling, anisotropic direction encoding, a reachability-based cross-patch transition mechanism, and self-negating search memory. Unlike the traditional continuous search assumption, this method emphasizes the long-term survivability of the search region rather than instantaneous optimality, thus exhibiting stronger stability and interpretability in complex optimization scenarios.