关键词: 白鲸优化算法（BWO）, 蝴蝶算法, 柯西变异, 动态反向学习, 轻量梯度提升机（LightGBM）

Abstract: Aiming at the shortcomings of the beluga whale optimization algorithm (BWO), which exhibits a slow convergence speed and is unable to escape the local optimum, an improved beluga whale optimization algorithm (MYBWO) based on butterfly search and dynamic reverse learning Cauchy variant is proposed. A nonlinear equilibrium factor is introduced to better balance the algorithm ability of global exploration and local exploitation. A butterfly search mechanism is introduced in the global exploration stage to enrich the diversity of populations and improve the search probability of the optimal solution. The dynamic opposite-learning and Cauchy variation strategies are integrated in the local exploitation stage to enhance the ability of the algorithm to jump out of the local optimum while expanding the population search scope. Simulation experiments are carried out using CEC2005 and CEC2019 test functions with different characteristics. The results show that compared with several selected comparison algorithms, MYBWO algorithm has higher optimization accuracy and faster convergence, which effectively solves the shortcomings of the algorithm that is easy to stagnate in the local optimum. In order to verify the practicality of the improved algorithm, MYBWO algorithm is used to optimize the LightGBM model to establish a new air quality prediction model. The experimental results prove that the prediction accuracy and stability of the model have been steadily improved.

Key words: beluga whale optimization (BWO), butterfly algorithm, Cauchy variation, dynamic opposite-learning, light gradient boosting machine (LightGBM)