Abstract: To address the deficiency in the local search ability of the artificial bee colony algorithm, a multi-strategy self-optimizing artificial bee colony algorithm based on reinforcement learning is proposed. This algorithm combines the Q-learning method in reinforcement learning with the artificial bee colony algorithm. The distance between the best value of the population and the individual fitness value, along with the diversity of the population are used as the basis for dividing the state. The algorithm creates an action set that contains multiple search strategies, adopts the ε-greedy strategy for selecting the best, produces high-quality offspring, and achieves intelligent selection of the ABC (Artificial Bee Colony) algorithm update strategy. Through 20 test functions and application in stock prediction, the results show that the proposed algorithm has better performance, a better balance between exploration and exploitation, faster convergence speed, and better self- optimizing ability.

Key words: artificial bee colony algorithm, reinforcement learning, multi-strategy, Q-learning, self-optimizing