Novel application of multi-model ensemble learning for fault diagnosis in refrigeration systems

Abstract

Despite the importance of fault diagnosis in refrigeration systems, the performance and improvement of most existing diagnostic models are limited by their reliance on a single method. This study proposes a novel application of ensemble learning that incorporates several intelligent ensemble members into an integrated model by means of majority voting. The ensemble members include k-nearest neighbour (KNN), support vector machine (SVM), decision tree (DT), random forest (RF) and logistic regression (LR). ASHRAE fault data were employed to establish the model. In addition, this study explores the integration of different subsets of ensemble members and revealed that the optimum subset combination comprised KNN, DT, and RF. Although the accuracy of this model was slightly lower than the model with all five ensemble members, it realised a substantial reduction in runtime. Compared to an SVM optimization model, the integrated model realised higher accuracy and reduced training time, without requiring parameter optimization. This achievement merits note as ensemble learning is traditionally associated high time-costs. Further investigation revealed that both the diversity and high accuracy of ensemble members are required to obtain an effective integrated model. These observations demonstrate the proposed model offers a promising alternative solution for fault diagnosis in refrigeration systems.