Diagnosis for the refrigerant undercharge fault of chiller using deep belief network enhanced extreme learning machine

Abstract

HVAC systems accumulate large amounts of historical data in their long-term operation process, and the use of big data has become one of the effective tools for fault diagnosis. As a common fault of chiller in HVAC system, refrigerant undercharge fault is chosen for analysis in this paper. Firstly, the experimental data of normal operation and refrigerant undercharge fault of chillers in an industrial park are collected through the IoT intelligent terminal, and the fault sensitivity characteristics of the experimental data are evaluated. Combined with expert knowledge, 23 features were selected as the input of the model. Then, combining the advantages of deep learning and machine learning, a hybrid model of refrigerant undercharge fault diagnosis for chillers using Deep Belief Network (DBN) enhanced Extreme Learning Machine (ELM) is proposed, where the parameters of the hybrid model are optimized by Particle Swarm Optimization (PSO). Finally, the experimental results of the real chiller are shown that the hybrid model can maintain a high diagnosis accuracy in the case of the number of training data set is small and has the highest accuracy and robustness compared with models such as ELM, Support Vector Machines (SVM) and K-Nearest Neighbor (KNN). The comprehensive accuracy rate of the model reaches 99.86%, and it has better data analysis ability, which can be well applied in practical applications.