Data-driven Energy Assessment And Diagnosis Of Variable Refrigerant Flow Air-conditioning Systems

Energy assessment and diagnosis is to evaluate the use of energy consumption and identify the abnormal energy use,which has great significance in the energy optimization and energy management.Energy assessment and diagnosis of variable refrigerant flow airconditioning(VRF)systems is essential to optimize the performance of VRF systems and improve building energy efficiency.Data-driven energy assessment and diagnosis methods have been widely used because of the advantages of high speed and high accuracy.This paper collects the relevant data of the VRF systems.The preprocessed normal energy data set use different algorithms to predict the energy consumption of the VRF systems and the best prediction model is determined.And also,the difference between the predicted value and the actual value as the input variable data are employed to establish the energy assessment and diagnosis models of the VRF systems based on one-class classification algorithms.In addition,the models are verified by the abnormal energy data set of the refrigerant charge failure.Based on normal energy data set of the VRF systems,the prediction model is established.According to the correlation analysis method,four characteristic variables are chosen that have an important influence on the energy of the VRF systems: compressor operating frequency,compressor exhaust temperature,outdoor ambient temperature and PLR.The local anomaly factor algorithm is applied to remove the anomaly samples in the original data set.In addition,data standardization is further to optimize the data.Four datadriven methods are employed to establish the prediction models and the prediction results show that the support vector regression algorithm is selected as the best prediction model.The genetic algorithm is used to optimize the parameters C and g of support vector regression algorithm and the optimal parameters C=1.0843 and g=14.101 are determined to optimize the model.Considering the actual operating conditions of the VRF systems,a cluster-based prediction model is proposed.The prediction models are built for different categories of samples and the prediction accuracy is slightly improved.Taking the simplicity of modeling and other factors into account,the GA-SVR model is determined as the best prediction model.Energy assessment and diagnosis model of the VRF systems is established based on the normal energy data set.The difference and ratio of the predicted energy value based on the GA-SVR prediction model and the actual value as input variables are applied to construct the OCSVM-based energy assessment and diagnosis model and the SVDD-based energy assessment and diagnosis model.Energy use anomaly data set are applied in the models.The results show that the OCSVM-based and SVDD-based energy assessment and diagnosis model can accurately identify the abnormal energy cases.In particular,when the refrigerant charge faults are serious,the abnormal energy recognition is more than 85%.