Abstract: To improve the classification accuracy of rolling bearing fault diagnosis, an Logistic-Cauchy-Levy-subtraction average-based optimization (LCLSABO) algorithm was proposed. The algorithm combines Logistic mapping, Cauchy mutation strategy and Levy flight strategy to optimize the performance of the kernel extreme learning machine (KELM). Firstly, the Logistic mapping is used to optimize the population initialization in the subtraction-average-based optimization algorithm to enhance the population diversity. Secondly, Cauchy variation strategy and Levy flight strategy are incorporated to refine the displacement mechanism of the algorithm, improving global search capabilities and effectively avoiding local optima. Finally, the key parameters of KELM were optimized by the LCLSABO algorithm, and the LCLSABO-KELM model was established to classify and diagnose bearing faults. Experimental results show that compared with the SABO-KELM model, SSA-KELM model, PSO-KELM model and traditional KELM model, the fault diagnosis classification accuracy of the LCLSABO-KELM model is 98.63%, which is improved by 0.97%, 2.70%, 3.90%, and 11.30%, respectively. This demonstrates that the proposed method effectively extracts fault features and significantly enhances the classification accuracy of fault diagnosis, demonstrating its superior performance in rolling bearing fault diagnosis and classification.