Search

Data-driven approach has emerged as a popular research method for fault detection and diagnosis (FDD) of variable refrigerant flow system. However, one serious issue has been raised in recent studies, which shows that training classifiers with datasets which suffer of imbalanced class distributions can significantly influence the final classification accuracy. So, aiming at the imbalanced dataset environment problem of variable refrigerant flow (VRF) system based on data-driven fault detect and diagnosis, we conducted a study. First of all, through comparative research, it is concluded that the geometric mean accuracy of the fault detection in the VRF system has reduced an average of 42.49% due to the imbalanced data environment. Then we proposed a Principal Component Analysis-Synthetic Minority Oversampling Technique (PCA-SMOTE) method to save this problem, and verified the reliability and versatility of the proposed method with three types of modeling algorithms commonly used in VRF systems. The result shows that this method not only has obvious improvement effect, but also has strong versatility. In the end, we gradually reduce the number of minority from 50 to 10 to explore minimum fault data requirements of this method. The final result shows that this method can still ensure that common machine learning algorithms can effectively detect and diagnosis the VRF system fault when only 10 fault data are included. [ABSTRACT FROM AUTHOR]

• An improved deep learning model based on tree structure is proposed in VRF system. • Pearson's correlation coefficient combines with isolation forest for data preprocessing • The physical significance of outliers in VRF system is explained. • The IF-CF model does not require complex hyper-parameter optimization strategy. This paper presents a hybrid deep forest approach for outlier detection and fault diagnosis. Isolation forest algorithm is combined with Pearson's correlation coefficient for outlier detection. The physical significance of outliers detected by the proposed algorithm is explained by origin analysis, which is rarely mentioned in existing studies. In addition, a novel non-neural network deep learning model-cascade forest model is proposed to fault diagnosis of HVAC system for the first time to achieve high precision accuracy in low-dimensional features. The proposed approach is validated with the refrigerant charge fault of VRF system. The results show that the isolation forest algorithm can improve the performance of fault diagnosis model and the mainly outliers of VRF system are defrosting data. The IF-CF model has short operation time, and high accuracy in low-dimensional features. When the dimension drops to 6, the accuracy of the IF-CF model is 94.16%, which is 5.26%, 10.02%, 5.87% and 3.34% higher than the IF-MLP, IF-BPNN, IF-SVM and IF-LSTM models, respectively. Moreover, IF-CF model does not require complex hyper-parameter optimization strategy because its maximum accuracy difference in different hyper-parameters is 2.04%. This study is enlightening which may inspire the potential of outlier detection technology and deep learning in HVAC field. [ABSTRACT FROM AUTHOR]

• A novel fault diagnosis strategy is proposed using the GMM method. • The principal component analysis is selected for data dimensionality reduction. • Evaluations are made using four faults data of the VRF system experiments. • The established PCA-GMM has better fault diagnosis performance and reduced runtime. The timely fault diagnosis of HVAC systems is important for building energy saving, equipment maintenance and indoor comfort. The Gaussian mixture model method has rarely been studied in the fault diagnosis application of HVAC systems. Therefore, a novel fault diagnosis strategy is proposed based on the Gaussian mixture model (GMM) method for the variable refrigerant flow air-conditioning system. To reduce excessive input variables resulting in large model complexity and long running time, the principal component analysis approach (PCA) is used to perform data dimensionality reduction. Therefore, the fault diagnosis model combining the Gaussian mixture model and principal component analysis is established, which is evaluated using the four types of faults of the variable refrigerant flow system. These faults include refrigerant undercharge, refrigerant overcharge, outdoor unit fouling and four-way reversing valve faults. Experiments are carried out under three heating conditions. Results show that the PCA-GMM approach can effectively reduce the running time. Especially for the VVV type model, the running time is reduced from 176.78 s to 15.18 s. Meanwhile, established PCA-GMMs still have good fault diagnosis correct rates when the input data dimension is reduced. Especially, some PCA-GMMs have fault diagnosis correct rates of over 99% when the number of principal components exceeds 7. [ABSTRACT FROM AUTHOR]

Energy consumption of heat pump system accounts for a large part of the total building energy consumption, and the energy-saving operation of heat pump system has always been the focus of researchers. A promising solution to tackling energy wastes during system operations is anomaly detection. In this study, we propose an anomaly detection method for GSHP system in a public building based on mode decomposition based LSTM and statistical modeling method Grubbs' test. The system energy consumption is predicted using mode decomposition based LSTM algorithm, and the difference between predicted value and actual value is used to detect the abnormal system energy consumption by Grubbs' test. Results show that detected anomalies can be summarily divided into three categories (parabola anomaly, abrupt anomaly and time related anomaly) depending on their characteristics, and the rationality of detected anomalies are evaluated through field investigation and expert knowledge. This work is enlightening and indicates that the proposed method would efficiently detect the abnormal performance of GSHP system, and find out unreasonable operating patterns. [ABSTRACT FROM AUTHOR]

Highlights • Three data-mining models are applied for cooling load demand forecasting. • Short and medium-term intervals are applied for future energy demand forecasting. • Environment variables effect on cooling load demand is also analyzed. • Model's performance and accuracy has been estimated at selection of different hidden neurons. • The best cooling load forecasting performance achieved at the short-term interval. Abstract In this study, data-mining techniques comprising three forecasting algorithms for accurate and precise cooling load requirement prediction in the building environment, with the primary aim and the objective of improving the load management are applied. Three state-of-the-art cooling load prediction algorithms are – multiple-linear regression (MLR) model, Gaussian process regression (GPR) model and Levenberg–Marquardt backpropagation neural network (LMB-NN) model. The Pearson correlation analysis is practiced calculating the correlation between actual cooling load demand and input feature variables of climate parameters. The impact of climate variability on the building load requirement is also analyzed. Forecasting intervals are divided into two basic parts: (i) 7-day ahead prediction; and (ii) 1-month ahead prediction. To assess the prediction performance, four performance evaluation indices are applied, which are: (i) coefficient of correlation (R); (ii) mean absolute error (MAE); (iii) mean absolute percentage error (MAPE); and (iv) coefficient of variation (CV). The model's performance is compared with the selection of different hidden neurons at different load conditions. The MAPE for 7-day ahead prediction interval by MLR, GPR and LMB-NN model is 13.053%, 0.405% and 2.592%, respectively. Furthermore, the data-mining algorithms are compared and validated with the previous study, and the MAPE of Bayesian regularization neural networks is calculated 2.515% for 7-day ahead prediction. It was witnessed that the algorithms could be applied to facilitate the building cooling load prediction, by applying a relatively limited number of parameters related to energy usage as well as environmental impact in the building environment. The forecasting results show that the three algorithms are effective in predicting the irregular behavior in the data as well as cooling load demand prediction. [ABSTRACT FROM AUTHOR]

• An online compressor liquid floodback fault diagnosis method is proposed for VRF system. • The proposed method fills the online liquid floodback fault diagnosis knowledge gap. • The online diagnosis results show the proposed method has extremely high accuracy. • The comparison results show that the BPNN method is better than CART method. Compressor liquid floodback not only can drop the performance of the variable refrigerant flow (VRF) system, but also may cause mechanical failure. Therefore, this is necessary to perform a timely online diagnosis of that fault. The present paper proposes an online compressor liquid floodback fault diagnosis method for VRF system based on back-propagation neural network (BPNN), which fills the online compressor liquid floodback fault diagnosis knowledge gap. The proposed method main article context as follow: Firstly, the sensors in VRF system record and save the data every three seconds to form the raw dataset. Secondly, after preprocessing the raw dataset, the correlation analysis is used to filter the data variables. Thirdly, the BPNN model is established by using back-propagation neural network algorithm. The testing dataset was used to verify the reliability of the model. The online dataset was used to test the model's online diagnostic capability, and its results were also compared and analyzed to the Classification and Regression Tree (CART) model. This result indicates that the BPNN method has a low degree of over-fitting and high reliability, and its online diagnostic accuracy is up to 99.48%. By comparing results of the method BPNN and CART, it shows that the former is superior to the latter, regardless of the diagnostic accuracy or the online diagnostic stability. [ABSTRACT FROM AUTHOR]

Malfunctions would occur in a variable refrigerant flow (VRF) system after years of operation or inappropriate maintenance, thus causing unnecessary energy waste and even occupant discomfort. This study presents a machine learning based malfunction diagnosis strategy that combines the recursive feature elimination algorithm (RFE) and the classification algorithms for the typical malfunctions of VRF system. RFE based on Random Forest (RF) model firstly serves as the feature selection process to evaluate variables importance, thus acquiring the key variables related to malfunction. Then five kinds of machine learning classification models are trained using the chosen key variables to diagnosis refrigerant leakage malfunction. By comparison, the AdaBoost.M1 (ABM) model shows the most desirable performance on the all nine malfunction severity levels. The results show that the RFR-RF based feature selection method can select the most six critical variables and the ABM model established based on the six variables achieves admirable diagnostic accuracy and AUC value for faults corresponding to nine severity levels. [ABSTRACT FROM AUTHOR]

Highlights • A new charge fault diagnosis method for VRF system. • The temperature difference variable is introduced to improve the diagnosis effect. • Especially increases the distinction between normal and excess charge fault. Abstract VRF (Variable Refrigerant Flow) system is an efficient and energy-saving air-conditioning system. The VRF system performance is significantly affected by the quantity of refrigerant (charge). Both insufficient and excessive refrigerant charge will lead a dramatic decline of air conditioning system performance. Accordingly, it is necessary to forecast refrigerant charge in existing systems. This study proposed a charge fault diagnosis method based on expert modification C5.0 decision tree. Initially, C5.0 decision tree is used to diagnosis VRF system charge faults in nine different conditions. Furthermore, in order to reduce error rate of the decision tree, expert experience is used to modify the decision tree. And temperature differential variables are used to improve the diagnosis result. The result s of the C5.0 decision tree using temperature differential variables shows fault diagnosis error rates under expert experience guidance are 10% lower than without expert experience. Besides, this method an increased accuracy in classifies between excessive and normal charge states. [ABSTRACT FROM AUTHOR]

Frosting cleaning process for dust removal is a simple and direct way to keep heat exchangers clean, and the dust removal effectiveness is strongly influenced by surface property. An object of this paper is to investigate the frosting cleaning process on super-hydrophobic surface and to determine the interaction between the frost layer and dust particles. Frosting-defrosting process on four kinds of test samples were studied by experiments. The results showed that, compared with the clean surfaces, the frost layer structure was denser and more uniform on dusty surfaces. On the super-hydrophobic surface, as a kind of heterogeneous nucleation point, the dust particle promoted frost layer growth by 57%. During the defrosting process on dusty test samples, the melted water and dust particles were reverse osmosed the frost layer. The reverse osmosed rate increased with frosting time, and it was over 0.5 under all experimental conditions. Compared with the bare surface, super-hydrophobic surface displayed a different defrosting behavior and had a shorter defrosting time. On the super-hydrophobic surface, deposited dust particles were encased in the frost-water mixture block and flaked off the surface under gravity, and the dust removal rate was up to 94.5% in this work. Because of the reverse osmosed dust movement, the wrap effect, and the weak adhesion of super-hydrophobic surface, the dust-frost mixture is easily removed. Thus, utilizing the frosting-defrosting characteristics of super-hydrophobic surface for promoting the dust removal has a good application prospect. [ABSTRACT FROM AUTHOR]

This paper proposed a universal virtual variable-speed compressor power (VVCP) sensor for VRF system based on 20-coefficient model. Compressor power can be obtained by the VVCP sensor using three input parameters (frequency, condensing temperature and evaporation temperature) which are measured by system itself. The performance of the proposed VVCP sensor is evaluated using experiments data. Experimental conditions include cooling, heating and non-standard refrigerant charge levels. The result shows that the mean square percentage errors (MSPE) are 9.9% under cooling conditions and 7.96% under heating conditions. The MSPE are 9.88% at steady state and 9.75% at dynamic state when the VVCP sensor is applied under nine different refrigerant charge levels. It demonstrated that the proposed VVCP sensor can obtain compressor power under cooling, heating and non-standard refrigerant charge levels, which could be applied to do operational monitoring and fault detection and diagnosis for VRF system at low cost. [ABSTRACT FROM AUTHOR]

This paper presents an algebraic solution of erroneous sensor's undetectable boundary to evaluate the sensitivity of chiller sensor fault detection based on principal component analysis. Q -statistic of PCA is normally applied as a collective statistical index to detect sensor fault by comparing its value with the threshold. However, Q -statistic has no specific physical meaning and cannot evaluate the sensitivity of the provided method for sensor fault detection. We analyzed the definition of Q -statistic and derived the numerical value of the minimum range not to detect sensor fault. Bias sensor fault of a fielded screw chiller was studied for each sensor in PCA model by introducing different severity levels. Results showed that each sensor has different fault detection sensitivity using the same PCA model. The undetectable boundary can be a criterion used to evaluate the detection sensitivity of PCA-based method easily. [ABSTRACT FROM AUTHOR]

• Deep reinforcement learning (DRL) based models are proposed for HVAC system energy consumption prediction. • Autoencoder (AE) algorithm is used to extract the high-level features of state space of DDPG. • The DDPG based models are applied for energy consumption forecasting over short-term time horizon. • Improvement in prediction performance is observed when combine AE with DDPG. • The DDPG based prediction models, outperform common supervised models like BP neural network and support vector machine. Short-term energy consumption prediction has fundamental importance in many HVAC system management tasks, such as demand-side management, short-term maintenance, etc. Currently, the prevailing data-driven techniques, especially supervised machine learning methods, are widely applied for short-term energy consumption prediction. Deep reinforcement learning (DRL), as the state-of-the-art machine learning techniques, have been applied for HVAC system control, but rarely for energy consumption prediction. In this paper, a DRL algorithm, namely Deep Deterministic Policy Gradient (DDPG), is firstly introduced for short-term HVAC system energy consumption prediction. Moreover, Autoencoder (AE), which is powerful in processing data in their raw form, is incorporated into DDPG method to extract the high-level features of state space and optimize the prediction model. The operation data of the ground source heat pump (GSHP) system of an office building in Henan province, China is used to train and assess the proposed models. The results demonstrate that the proposed DDPG based models can achieve better prediction performance than common supervised models like BP Neural Network and Support Vector Machine. This study is an enlightening work which may inspire other researchers to tap the potential of DRL algorithms in this field. [ABSTRACT FROM AUTHOR]