Chiller Fault Detection and Diagnosis Based on Genetic Algorithm and Support Vector Machine with Parameter Tuning

Abstract

In view of too many features and low accuracy characteristics in chiller fault detection and diagnosis(FDD),a hybrid FDD model based on genetic algorithm(GA) and SVM with parameter tuning is presented.The GA-SVM wrapper can carry out fault feature extraction and model training simultaneously.Studying on seven typical chiller faults shows that eight indicative features,centered around the core refrigeration cycle,can be selected from the original 64 features,with a rise in FDD accuracy from 96.95% to 99.53% and a sharp drop in FDD time(over 70%).Using hit rate(HR) and false alarm rate(FAR) to evaluate the model performance on individual fault,the presented hybrid model behaves much better than the SVM model without feature extraction and the PCA-SVM model.It has a promising future in chiller FDD applications.