نظارت بر دستگاه سنجش و داده کاوی تکنیک های هوشمند برای پیش بینی عملکرد سیستم تبرید

A major challenge in many countries is providing sufficient energy for human beings and for supporting economic activities while minimizing social and environmental harm. This study predicted coefficient of performance (COP) for refrigeration equipment under varying amounts of refrigerant (R404A) with the aids of data mining (DM) techniques. The performance of artificial neural networks (ANNs), support vector machines (SVMs), classification and regression tree (CART), multiple regression (MR), generalized linear regression (GLR), and chi-squared automatic interaction detector (CHAID) were applied within DM process. After obtaining the COP value, abnormal equipment conditions can be evaluated for refrigerant leakage. Analytical results from cross-fold validation method are compared to determine the best models. The study shows that DM techniques can be used for accurately and efficiently predicting COP. In the liquid leakage phase, ANNs provide the best performance. In the vapor leakage phase, the best model is the GLR model. Experimental results confirm that systematic analyses of model construction processes are effective for evaluating and optimizing refrigeration equipment performance

Data mining (DM) techniques have formed a branch of applied Artificial Intelligence (AI) since the 1960s, several major kinds of data mining methods such as generalization, characterization, classification, clustering association, evolution, pattern matching, data visualization and meta-rule guided mining (Liao, Chu, & Hsiao, 2012). Prediction is one of major data mining functions used in the applications (Köksal, Batmaz, & Testik, 2011). Its broad applications include marketing, healthcare, civil engineering, and many others (Chen et al., 2013, Chou, 2009, Kao et al., 2011, Koyuncugil and Ozgulbas, 2012, Küçüksille et al., 2009 and Moreno Sáez et al., 2013). In addition, Köksal et al. (2011) comprehensively reviewed DM applications in manufacturing industries (Köksal et al., 2011). However, DM is rarely applied in the energy field, particularly to support energy efficiency. Taiwan, which imports 99.4% of its energy needs, has already begun replacing conventional meters with smart meters. Taiwan Power Company plans to install 1 million smart meters for its customers before 2015. To achieve this target, smart meter devices will be installed in 10 thousands households in 2012, and installations will gradually increase to 1 million devices during 2013 to 2015 (Lee, 2011). The aim of this policy is to improve energy efficiency and reduce carbon emissions in Taiwan. Based on the overseas experience in similar projects, policy makers have predicted that the devices will reduce carbon dioxide emissions in the future. Moreover, in many countries, current policies for reducing emissions coupled with growing public awareness of increased utilities price have increased the use of smart meters as monitoring tools (Bennett et al., 2013, Li et al., 2011 and Usman and Shami, 2013). Therefore improving refrigeration system performance via smart meter and data mining is an important research issue. As energy conservation and carbon reduction have recently become an important issue, investigation of large consumption energy systems has been prioritized (Bektas Ekici and Aksoy, 2011, Kalogirou, 2000, Oğuz et al., 2010, Rodger, 2014 and Soyguder and Alli, 2009). One of the facilities being studied is refrigeration systems for preserving food and for air conditioning, which constantly consume energy. Refrigeration systems are ubiquitous and can be found in many locations, including factories, households, offices, etc. Despite their wide spread use, the performance of these systems has not been fully investigated (Ahmed et al., 2011, Ozgoren et al., 2012 and Şahin, 2011). Thus, developing an appropriate methodology for predicting refrigeration system performance based on refrigerant conditions is imperative. Although many studies of this problem have been performed, available studies using data mining approach of such energy systems are still rare. Notably, the electrical properties of refrigerant amount used in vapor compression refrigeration systems can only be determined through experiments. This study designed laboratory experiments to achieve this goal. All experimental data were retrieved from smart meters, which were also used to monitor electricity usage and user behavior. The DM techniques were used as analytical tools to predict the coefficient of performance (COP) under different refrigerant amounts. The DM techniques compared in this study included artificial neural networks (ANNs), support vector machines (SVMs), classification and regression tree (CART), multiple linear regression (MLR), generalized linear regression (GLR), and chi-squared automatic interaction detector (CHAID) techniques. The indicators used to evaluate model performance were mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and correlation coefficient. Cross-fold validation was also performed to ensure a balanced view and to avoid bias from data. The rest of this paper is organized as follows. Research findings from previous studies are summarized in Section 2. Section 3 presents the data mining methodology used in this work. Section 4 describes the experimental design and monitoring system. Section 5 discusses the model implementation and analytical outcomes, i.e., model settings, cross-fold validation and the analysis results. The last section gives concluding remarks.

The aim of this paper is to evaluate the usefulness and compare the performance of data mining techniques in predicting COP of refrigeration equipment under varying settings for R404A refrigerants. Predicting coefficient of performance is useful in equipment monitoring in utility companies and facility owners. The six data mining techniques compared in this study were ANNs, SVMs, MR, CHAID, CART and GLR. The comparison is performed in several steps. Firstly, Pearson correlation coefficient is calculated for attribute selection. Secondly, cross-fold validation models are introduced to reduce prediction bias. The final step is to determine the SI values for each model under different Pearson correlation coefficients. The SI value is calculated by averaging 1 − R, MAPE, MAE, RMSE and the number of attributes. The research findings can be summarized as follows. (a) In liquid leakage phase, ANNs accurately predict COP values. When using SI5 value as the performance index, L13 provides the best prediction performance (|r| > 0.7). In vapor leakage phase, the best model is the GLR model (V24) with |r| > 0.6. (b) The amount of refrigerant substantially affects equipment performance. Therefore, equipment performance should be judged in terms of cooling capacity and cooling time. (c) The proposed refrigeration system performs best when the amount of R404A is 3kg. (d) In vapor phase and |r| > 0 to |r| > 0.5 cases, MAPE evaluation method shows more consistent results compared to both SI evaluations. (e) Model comparison results show that the SI5 method produces a more reliable and more comprehensive performance evaluation compared to MAPE method alone for |r| > 0.6 to |r| > 0.9 models. However, one significant limitation of this work is that it used default settings in the single model. Therefore, further studies are needed to investigate model parameters optimization for predicting COP as well as detect abnormal electricity usage and to reduce unnecessary energy consumption. The investigation is aligned with present global trends in terms of its use for reducing carbon emissions and for addressing customer concerns about high electricity costs. Another research direction suggested for future studies is to use cloud monitoring systems to diagnose equipment performance, including abnormal signal characteristics.