A novel Q-learning-inspired Mountain Gazelle Optimizer for solving global optimization problems

P Sarangi and S Mohapatra and P Mohapatra, INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS, 16, 6167-6213 (2025).

DOI: 10.1007/s13042-025-02620-1

Q-learning, an eminent reinforcement learning (RL) approach, has garnered substantial research attention in recent years owing to its effectiveness in solving intricate problems and attain noteworthy results in a range of applications. In this study, the Mountain Gazelle Optimizer (MGO) is explored as a promising metaheuristic algorithm, primarily due to its biologically inspired mechanisms that emulate the adaptive and dynamic behaviors of gazelles in nature. However, despite its strong performance, MGO has inherent limitations, such as a tendency to become trapped in suboptimal search regions during early iterations, making it challenging to escape local optima. Therefore, to circumvent these shortcomings, this paper introduces a novel Q-learning-inspired Mountain Gazelle Optimizer (QLMGO), integrating chaotic and random opposite-based learning (ROBL) strategies to enhance optimization performance. The key innovation of QLMGO lies in its dynamic switching mechanism, enabled by Q-learning, which adaptively selects between ROBL and chaotic strategies to optimize the search process. Initially, Q-learning is utilized to regulate the switching mechanism, ensuring efficient exploitation of the search space. During the update phase, QLMGO dynamically chooses the most effective strategy, either ROBL for intensified local search or chaotic exploration for escaping local optima, to accelerate convergence towards the global optimal solution. The performance of QLMGO was rigorously evaluated against well- established optimization algorithms using 23 CEC2005 functions, 10 advanced CEC2019 functions, 30 CEC2017 test functions, and six real- world engineering problems. To ensure a robust and precise assessment, statistical analyses including the Wilcoxon rank-sum test, Friedman test, and t test were conducted. The empirical results from benchmark functions and engineering applications demonstrate the superiority of QLMGO in solving both constrained and unconstrained optimization problems efficiently, thereby validating its effectiveness as an innovative optimization approach.

Return to Publications page