Sandpiper Food Search Algorithm: A New Optimization Algorithm With Limited Knowledge

Abstract Optimization problems in mechanical engineering play a crucial role in refining complex system designs, improving performance, and maximizing efficiency across various applications. These problems have become increasingly complex, stretching the limits of conventional methods such as convex optimization or the Newton-Raphson method. To address this challenge, numerous metaheuristic algorithms, often inspired by nature, have emerged including Particle Swarm Optimization, Genetic Algorithm, Bat Algorithm, Firefly Algorithm, and Cuckoo Search. However, most of these existing algorithms have global knowledge, which is unrealistic for some real-world complex systems such as underground mining and spacecraft trajectory. To bridge this gap, this article develops and presents a new biologically inspired optimization algorithm, the Sandpiper Food Search Algorithm. This algorithm is inspired by the food search behavior of sandpipers at the beach where each agent (sandpiper) explores the problem space to find the optimal area by exploiting the local search for candidate solutions around them. Moreover, this algorithm includes the wave action that forces these birds to shift from their current solution to increase exploration of the solution space. The performance of the algorithm is evaluated using four standard benchmark functions in comparison with the Firefly Algorithm. Firefly Algorithm was selected for comparison because it shares similar parameterization characteristics, and its use of decreasing light brightness with distance between fireflies mirrors the limitation of knowledge imposed by the visibility radius in sandpipers. This research provides a conceptual design of the new Sandpiper Food Search Algorithm and its performance when compared to Firefly Algorithm over four benchmark functions. Our work shows that the Sandpiper Food Search Algorithm outperforms the Firefly Algorithm in three of the four benchmark functions with at least 3% improvement in mean best solution and is on average 38% more reliable at finding a solution at least 95% of the optimal.