Aiming at the problems of uneven fault categories of mining excavator engine and small and medium samples leading to insufficient diagnostic accuracy, an improved whale search algorithm optimizing LightGBM for intelligent fault diagnosis of mining excavator engine is proposed. First, the feature extraction of the collected excavator engine fault data is utilized by the recursive feature cross-validation elimination method (RFECV) to remove redundant and irrelevant features. Second, Focal-loss is adopted as the loss function of LightGBM, and an improved whale algorithm is proposed for hyperparameter optimization of LightGBM to construct a new diagnostic model. Finally, the engine failure data of an excavator in a mine is utilized for validation and compared and analyzed with common integration models, tuning frameworks and diagnostic algorithms. The experimental results show that the proposed method has better diagnostic performance and achieves 98.08% accuracy and 98.51% F1 score, which can provide reference for the intelligent diagnosis of mining machinery and equipment.