Deep Learning Technology for Chiller Faults Diagnosis

In order to timely detect the faults of the chillers and simultaneously reduce energy consumption, manpower and maintenance costs, this paper takes advantage of deep learning technology to accurately diagnose various chiller faults for centralized heating ventilation air-conditioning (HVAC) systems. Using the benchmark chiller data collected by the ASHRAE project 1043-rp, and long short-term memory (LSTM) neural network, we build a two-layer LSTM fault detection and diagnosis model for various chiller faults in different severity levels. The deep learning model parameters are optimized and cross-validated. In this way, the optimal model parameters are determined, resulting in the outperformance of the proposed deep learning neural network over existing machine learning techniques. In the experimental section, the proposed method is compared with the traditional recurrent neural network (RNN) and gated recurrent unit (GRU) neural network, over five different air conditioning fault severity, the proposed method generally outperforms the two traditional RNN frameworks with higher detection and diagnosis accuracy.