Chiller Fault Diagnosis Based on Automatic Machine Learning

Intelligent diagnosis is an important means of ensuring the safe and stable operation chillers driven by big data. To address problems input feature redundancy in intelligent reliance on human intervention selection model parameters, a chiller fault method was developed this study based automatic machine learning. Firstly, improved max-relevance min-redundancy algorithm used to extract information effectively automatically from training Then, long short-term memory (LSTM) mine temporal correlation between data, genetic employed train optimize obtain optimal neural network architecture hyperparameter configuration. Finally, transient co-simulation platform for building MATLAB as well Engineering Equation Solver built, effectiveness proposed verified using dynamic simulation dataset. The experimental results showed that, compared with traditional learning methods such recurrent network, back propagation support vector methods, LSTM provides significant performance improvement cases low severity complex faults, verifying superiority method.