Research On Fault Diagnosis Of Water Chiller Based On GA-LightGBM Model

According to research conducted by relevant institutions,HVAC energy consumption accounts for 60% of the total energy consumption in commercial buildings.As the core component of the HVAC system,chiller units come in various types and have complex structures,making them prone to failures and "running with faults." If a chiller unit experiences a failure and continues to operate with faults for an extended period of time,it can result in wasted energy,equipment damage,and potentially even safety accidents.Therefore,fault detection and diagnosis of chiller units is of great research value and significance.In this paper,we focus on chiller units and use a data-driven approach to study fault diagnosis from three aspects: feature selection,imbalanced data between normal and fault samples,and model optimization.The main research content of this paper includes:(1)Address the challenge of reduced operational efficiency and model performance due to the multitude of feature parameters in the operational data of chiller units for fault detection and diagnosis.The Relief F algorithm is proposed to rank the importance of chiller unit features,and four machine learning algorithms,including Random Forest,Support Vector Machine,BP Neural Network,and Light GBM,are used to establish chiller unit fault detection and diagnosis models under different feature dimensions,followed by the analysis of experimental results.Finally,the paper determines the appropriate feature dimension for chiller unit fault diagnosis and concludes that the Light GBM algorithm model performs the best.(2)Address the problem of imbalanced data between normal and fault samples during the actual operation process of chiller units,we propose the SMOTEENN-CGAN fault data generation method based on two data generation methods: SMOTE and CGAN.This method generates minority class fault data to solve the imbalanced data problem.Finally,we propose the SC-Light GBM chiller unit fault diagnosis model in combination with the Light GBM algorithm.The diagnostic results show that the SC-Light GBM model performs the best under multiple imbalanced data sets and fault levels,and can effectively solve the data imbalance problem of chiller units.(3)To further improve the performance of the Light GBM fault diagnosis model,we use genetic algorithm and particle swarm optimization algorithm to optimize the four important hyperparameters of the Light GBM fault diagnosis model: tree＿num,learning＿rate,max＿depth,and min＿child＿weight.We construct chiller unit fault diagnosis models based on GALight GBM and PSO-Light GBM.The results show that the GA-Light GBM model has the best accuracy under three data sets with imbalanced ratios of 10:1,20:1,and 50:1.Compared with the Light GBM model without parameter optimization,the accuracy of the GA-Light GBM model is increased by 1.16%,1.64%,and 3.36% in the three data sets,respectively.