| Timely and accurate fault detection and diagnosis(FDD)of the air conditioning system can reduce energy waste and labor maintenance costs,consequently improving the air-conditioning efficiency.There are many reasons to perform FDD on air conditioning equipment with complex internal structure.The traditional machine learning methods have limitations with the complex and continuous industrial environment.In addition,actual air conditioning fault data is far less than the normal data.The fault label acquisition is difficult.This study performs air conditioning FDD using deep learning technology.The works that have been done include:(1)According to the time series of air conditioning fault data,this study builds a model of air conditioning fault diagnosis based on deep learning LSTM.This model overcomes the gradient descent problem of traditional recurrent neural networks and optimizes the parameters.The optimal LSTM model is finally determined under cross-validation.This model can accurately classify air-conditioning fault labels under different severity of air-conditioning faults.Compared with traditional recurrent neural network and gated recurrent unit network,the model has higher accuracy and better adaptability for fault diagnosis.(2)The premise of fault diagnosis includes accurately detecting whether there is any fault in a large amount of unbalanced data.This study combines the Relief F algorithm for feature selection and builds a dual-module LSTM model.It adds a module for air conditioner fault detection,forming an integration of air conditioner fault detection and diagnosis.Based on the detection and diagnosis under different severity levels,each type of fault is diagnosed one by one.(3)In order to solve the problem of difficult to obtain fault data of air conditioners and not many data labels,this study proposes a generative adversarial network(WGAN)using Wasserstein distance to measure the similarity between two probability distributions.At the same time,this model also combines XGBoost algorithm,VAE algorithm and semi-supervised combined support vector machine algorithm as the evaluation module of this model.The evaluation module can pick out high-quality samples.Finally,under different fault severity of air conditioners,this model provides different algorithms to diagnose multiple fault categories,such as memory recurrent neural network algorithm,random forest algorithm,support vector machine algorithm,decision tree algorithm,K nearest neighbor algorithm,etc.(4)This study designed and implemented a multi-model comprehensive air conditioning fault diagnosis software based on deep learning algorithms.It includes five parts: a module for preprocessing data,a module for generating fault data,a module for extracting high-quality samples,a module for visualizing the diagnosis results of different models,and an auxiliary function module.It can accurately diagnose the operation of the air conditioner.At the same time,it can provide comparison results of different diagnostic models.The interface of the air-conditioning diagnostic software is simple and friendly.The air conditioning diagnostic software can enhance user experience. |
