NEW VARIANTS OF MARINE PREDATORS ALGORITHM AND THEIR APPLICATIONS

Abstract

This Thesis introduces a significant advancement in optimization algorithms through the development of two novel variants of the Marine Predators Algorithm (MPA): one for continuous problems and the other for discrete problems. Each variant is designed to address specific practical challenges in their respective optimization domains. The classical MPA, inspired by the foraging behavior of marine predators, has shown promise in solving a variety of continuous optimization problems. However, the classical MPA suffers from critical drawbacks, such as premature convergence, slow convergence speed, stagnation in local optima, and limited population diversity. To overcome these challenges, the Opposition-Based Local Escaping Marine Predators Algorithm (OLMPA) is developed for continuous optimization problems. The newly introduced OLMPA incorporates opposition-based learning (OBL) and a local escaping strategy to enhance the algorithm’s performance by improving diversity and solution quality in the search space. The primary improvement, OBL, strategically generates candidate solutions that are opposite to the current population, enhancing the diversity within the search space. The second improvement, the local escaping mechanism, rejuvenates stagnating regions of the search space by replacing the worst solutions with new ones, thereby escaping local optima and fostering more robust global search capabilities. The effectiveness of OLMPA is validated through extensive tests on 23 popular benchmark functions, where it significantly outperformed classical MPA in terms of accuracy and reliability. To further enhance the theoretical understanding ofMPA, this Thesis examines structural bias, a recently identified feature of metaheuristic algorithms. Structural bias causes populations to repeatedly revisit specific regions of the search space, hampering exploi ration. This behavior is analyzed using advanced tools, such as the Signature Test, the BIAS toolbox, and the Generalized Signature Test. It is shown that OLMPA exhibits significantly less structural bias compared to classical MPA. This reduction further highlights the improved search and exploration efficiency of OLMPA. The practical value of OLMPA is showcased through its application to two real-world problems: earthquake hypocenter determination and wind farm layout optimization. In the context of earthquake hypocenter determination, OLMPA contributes to find a rapid and accurate earthquake location, which can help mitigate disaster impacts and address the severe effects of forthcoming earthquakes through effective post-earthquake relief efforts. Testing data from the Garhwal region in India show that OLMPA outperforms other well-known algorithms in accuracy and computational efficiency. In the domain of renewable energy, OLMPA optimizes the placement of wind turbines within a farm to maximize power generation while minimizing wake losses. Its ability to address these challenges underscores its potential to promote sustainable energy solutions. Across various experiments, OLMPA consistently achieves superior performance over classical MPA, demonstrating its efficacy in handling complex real-world scenarios. Being a continuous optimizer, the major limitation of MPA is its inapplicability to discrete optimization problems. Hence, the second variant, the Discrete Marine Predators Algorithm (D-MPA), extends the capabilities of MPA to discrete optimization problems, particularly combinatorial optimization. By integrating permutation operators and the 2-opt local search technique, D-MPA effectively addresses discrete optimization problems such as the Symmetric Traveling Salesman Problem (STSP). The algorithm is tested on 36 STSP instances and compared with popular state-of-the-art algorithms. For several instances, including krob100, pr107, pr144, ul59, and tsp225, the proposed algorithm outperformed the best-known results to date, setting new benchmarks and demonstrating remarkable performance. Moreover, for most STSP instances, D-MPA consistently outperformed other state-of-the-art algorithms, highlighting its efficiency and solution quality. This Thesis significantly improves the theoretical understanding and practical appliii cations of MPA. By addressing challenges such as premature convergence and structural bias and introducing robust variants for continuous and discrete optimization. As a result, OLMPA and D-MPA emerged as powerful tools for solving a wide spectrum of optimization problems.