A novel interpretable fault diagnosis approach for chillers based on self-adaptive oversampling technique and tree-based multi-meta stacking ensemble method

Chillers are of great significance for temperature control and energy consumption optimization. Once the chiller fails, it will lead to insufficient cold supply and reduced production efficiency. The data-driven deep learning model is widely used in fault diagnosis because it can automatically extract features from a large amount of data. However, in practical engineering, it is often faced with the problems of insufficient samples and unbalanced data, and the black box characteristics of the model make fault diagnosis urgently need interpretability. In order to cope with the above challenges, this paper proposes an interpretable ensemble learning model combined with oversampling technology. Firstly, the SHapley Additive exPlanations and pearson correlation coefficient matrix are used to visually analyze the prediction process of the model, and the potential effective features are mined according to the contribution, importance and correlation. Then, according to the sample density of each category and the sample neighborhood distribution, the classification difficulty of the sample is evaluated, the attention to the difficult-to-classify samples is improved, and the number of samples of each category is balanced. Finally, a diverse meta-learner stacking ensemble learning structure is proposed. By combining multiple tree-based ensemble learning models for progressive learning and using multiple meta-learners to capture feature information from multiple scales at the same time, the output of the first stage is comprehensively corrected to obtain the final diagnosis result. The proposed diagnostic framework is verified in the RP-1043 dataset, and the diagnostic accuracy of the proposed model can reach 99.08%.