A novel marine predators algorithm with adaptive update strategy

The marine predators algorithm (MPA) is a metaheuristic algorithm for solving optimization problems. MPA divides the whole optimization process into three phases evenly, and each phase corresponds to a different search agent update strategy. Such a setup makes MPA inflexible when facing different optimization problems, which affects the optimization performance. In this paper, we propose a novel modified MPA hybridizing by Q-learning (QMPA), which applies reinforcement learning to the selection of update strategy, and selects the most appropriate position update strategy for search agents in different iteration stages and states. It can effectively compensate for the deficiency of MPA's adaptive ability when facing different optimization problems. The performance of QMPA is tested on classical benchmark functions, the CEC2014 test suite, and engineering problems. In the classical benchmark functions test, QMPA is compared with MPA in 10, 30, and 50 dimensions. QMPA performs better than MPA for seven of the ten functions when the dimension is 10 and 30. The results of dimension 50 show that QMPA outperforms MPA in 5 functions and is close to it in 4 functions. Then, comparing QMPA with algorithms such as grey wolf optimizer, particle swarm optimization, slime mould algorithm, sine cosine algorithm, reptile search algorithm, and aquila optimizer, the results show that QMPA has the best performance on 22 of the total 30 functions in the CEC2014 test suite. Finally, QMPA is tested on two commonly used real-world engineering problems and gives the most optimal results. In general, the adaptive update strategy proposed in this paper improves the optimization performance of the MPA algorithm in terms of convergence and stability.