基于主元分析的冷水機組傳感器故障檢測效率研究
發(fā)布時間：2018-11-25 07:53
【摘要】：傳感器故障不僅會影響制冷空調(diào)系統(tǒng)的運行狀況，也會導致運行能耗的增加。傳感器的故障檢測、診斷及重構(gòu)研究是制冷空調(diào)領(lǐng)域與自動控制領(lǐng)域的一個交叉研究方向，近年來逐漸受到關(guān)注和重視。冷水機組是制冷空調(diào)系統(tǒng)的主要供能設(shè)備，，也是制冷空調(diào)系統(tǒng)運行與耗能的核心設(shè)備，冷水機組傳感器的故障檢測、診斷及重構(gòu)研究，具有非常重要的理論研究意義和工程應(yīng)用價值。 主元分析是傳感器故障檢測、診斷及重構(gòu)研究中常用的數(shù)據(jù)分析方法。首先分析整理了以Q統(tǒng)計量為檢驗標準的基于主元分析的傳感器故障檢測、診斷及重構(gòu)策略。結(jié)合熱平衡原理以及冷水機組運行的控制邏輯，分析和篩選了冷水機組中的常用8個傳感器——冷凍水側(cè)供水溫度、回水溫度、流量，冷卻水側(cè)供水溫度、回水溫度、流量，機組功率以及制冷劑流量調(diào)節(jié)裝置反饋信號組成了主元分析的耦合模型，并分析了不同傳感器不同故障程度下的故障檢測效率特點。然后采用了實測數(shù)據(jù)和模擬數(shù)據(jù)數(shù)據(jù)進行分析和驗證工作。結(jié)果表明，不同傳感器在不同故障條件下，檢測效率差異很大�；谥髟治龅膫鞲衅鞴收蠙z測方法在小偏差故障條件下的故障檢測效率較低，而且部分傳感器的整體故障檢測效率偏低。 針對傳感器故障檢測效率有待進一步提高的問題，從訓練矩陣優(yōu)化、測量數(shù)據(jù)優(yōu)化、檢驗標準優(yōu)化三方面分析了一系列改善和提高基于主元分析的水冷冷水機組傳感器故障檢測效率的方法。 在訓練矩陣優(yōu)化方面，依據(jù)距離度量的概念，建立了兩種訓練矩陣優(yōu)化的方法。一種是以標準化原始數(shù)據(jù)歐氏距離作為異常數(shù)據(jù)的判斷標準，剔除歐氏距離z得分大于2的異常數(shù)據(jù)，建立了結(jié)合基于距離度量異常值剔除的主元分析故障檢測策略；另一種以Q統(tǒng)計量閾值Qα作為異常數(shù)據(jù)的判斷標準，采用嚴格的自適應(yīng)異常數(shù)據(jù)循環(huán)剔除方法，建立了自適應(yīng)主元分析故障檢測策略。兩種方法的主要目的均是通過剔除原始數(shù)據(jù)中偏離聚集中心的數(shù)據(jù)，減少異常數(shù)據(jù)對主元分析正交投影空間的影響。 在測量數(shù)據(jù)優(yōu)化方面，采用小波變換方法優(yōu)化原始訓練數(shù)據(jù)和后續(xù)被測數(shù)據(jù)，去除數(shù)據(jù)中的噪聲。由于小波變換具有可變的層次性，因此進一步對比分析了不同小波分解層次對檢測效率的影響。當分解層次越多時，檢測效率提高越明顯。 在檢驗標準優(yōu)化方面，通過多統(tǒng)計量的交叉檢驗提高故障檢測效率。對比分析了Q統(tǒng)計量、T~2統(tǒng)計量和HawkinsT~2_H統(tǒng)計量對于不同傳感器不同故障的檢測效率。通過主元空間統(tǒng)計量——T~2統(tǒng)計量和殘差空間統(tǒng)計量——Q統(tǒng)計量及Hawkins T~2_H統(tǒng)計量的交叉檢驗，能明顯提高在小偏差條件下的整體故障檢測效率。為了進一步提高對故障的及時檢測，以訓練矩陣Q統(tǒng)計量的均值作為預期均值，采用Q統(tǒng)計量的累積和控制圖進行在線檢測及其效率分析，利用誤差的時間累積性提高對微小偏差故障的檢測效率。 結(jié)果表明，上述方法均能改善和提高冷水機組傳感器的故障檢測效率，從而促進傳感器故障診斷及數(shù)據(jù)重構(gòu)研究的敏感性。
[Abstract]:The sensor failure will not only affect the operating conditions of the refrigeration and air conditioning system, but also result in an increase in operating energy consumption. The fault detection, diagnosis and reconstruction of the sensor is a cross-research direction in the field of refrigeration and air-conditioning and automatic control, and has been paid more attention and attention in recent years. Chiller is the main energy-supply equipment of the refrigeration and air-conditioning system. It is also the core equipment for the operation and energy consumption of the refrigeration and air-conditioning system. The fault detection, diagnosis and reconstruction of the water chilling unit sensor has very important theoretical research significance and engineering application value. The main element analysis is the data analysis commonly used in the research of sensor fault detection, diagnosis and reconstruction Methods: First, the fault detection, diagnosis and reconstruction of the sensor based on the primary element analysis of Q statistics is analyzed. the water supply temperature, the water return temperature, the flow rate, the water supply temperature of the cooling water side and the water return temperature are analyzed and screened in combination with the heat balance principle and the control logic of the operation of the water chilling unit, The coupling model of the main element analysis is composed of the flow rate, the unit power and the feedback signal of the refrigerant flow regulating device, and the failure detection efficiency of different sensors under different fault conditions is analyzed. Characteristics. The measured data and the analog data data are then used for analysis and verification. The results show that the detection efficiency of different sensors is different under different fault conditions. The sensor fault detection method based on the primary element analysis is low in fault detection efficiency under the condition of small deviation fault, and the whole fault detection efficiency of the partial sensor In view of the problem that the sensor fault detection efficiency is to be further improved, the fault detection of the water-cooled water chilling unit based on the main element analysis is analyzed from the aspects of the training matrix optimization, the measurement data optimization and the test standard optimization. The method of efficiency. In the optimization of training matrix, two kinds of training are set up according to the concept of distance measure. The invention relates to a method for optimizing the training matrix. The method comprises the following steps of: taking the Euclidean distance of the standardized raw data as the judgment standard of the abnormal data, and removing the abnormal data with the Euclidean distance z score of more than 2, and establishing a main element combined with the elimination of the abnormal value based on the distance metric; The fault detection strategy is analyzed, and the self-adaptive main element is established by using the strict self-adaptive abnormal data cyclic elimination method based on the Q statistic threshold Q value as the judgment standard of the abnormal data. The main purpose of the two methods is to reduce the abnormal data to the main element by eliminating the data from the collection center in the original data. The influence of the cross-projection space. In the aspect of data optimization, the method of wavelet transform is used to optimize the original training data and the follow-up. The data is used to remove the noise in the data. The wavelet transform has the variable hierarchy, so it is further compared and analyzed the different wavelength division. The effect of the solution level on the detection efficiency. The more the decomposition level The more the detection efficiency is, the more statistical it is to test the standard optimization. The efficiency of fault detection is improved by cross-checking of quantity. The statistics of Q statistics, T ~ 2 statistic and HawksT ~ 2 _ H statistic are compared and analyzed. The detection efficiency of different sensors of different sensors can be obviously improved by cross-checking the statistics of the statistics of the main element space _ T-2 and the statistic quantity of the residual space _ Q and the statistic quantity of Hawkins T-2 _ H. in order to further improve the timely detection of the fault, the average value of the statistical quantity of the training matrix Q is used as the expected mean value, the accumulation and control charts of the Q statistics are adopted to carry out on-line detection and the efficiency analysis, and the time accumulative property of the error is utilized. The results show that the method can improve and improve the fault detection efficiency of the water chilling unit sensor, so as to promote the transmission
【學位授予單位】：華中科技大學
【學位級別】：博士
【學位授予年份】：2013
【分類號】：TU831.4
本文編號：2355363
[Abstract]:The sensor failure will not only affect the operating conditions of the refrigeration and air conditioning system, but also result in an increase in operating energy consumption. The fault detection, diagnosis and reconstruction of the sensor is a cross-research direction in the field of refrigeration and air-conditioning and automatic control, and has been paid more attention and attention in recent years. Chiller is the main energy-supply equipment of the refrigeration and air-conditioning system. It is also the core equipment for the operation and energy consumption of the refrigeration and air-conditioning system. The fault detection, diagnosis and reconstruction of the water chilling unit sensor has very important theoretical research significance and engineering application value. The main element analysis is the data analysis commonly used in the research of sensor fault detection, diagnosis and reconstruction Methods: First, the fault detection, diagnosis and reconstruction of the sensor based on the primary element analysis of Q statistics is analyzed. the water supply temperature, the water return temperature, the flow rate, the water supply temperature of the cooling water side and the water return temperature are analyzed and screened in combination with the heat balance principle and the control logic of the operation of the water chilling unit, The coupling model of the main element analysis is composed of the flow rate, the unit power and the feedback signal of the refrigerant flow regulating device, and the failure detection efficiency of different sensors under different fault conditions is analyzed. Characteristics. The measured data and the analog data data are then used for analysis and verification. The results show that the detection efficiency of different sensors is different under different fault conditions. The sensor fault detection method based on the primary element analysis is low in fault detection efficiency under the condition of small deviation fault, and the whole fault detection efficiency of the partial sensor In view of the problem that the sensor fault detection efficiency is to be further improved, the fault detection of the water-cooled water chilling unit based on the main element analysis is analyzed from the aspects of the training matrix optimization, the measurement data optimization and the test standard optimization. The method of efficiency. In the optimization of training matrix, two kinds of training are set up according to the concept of distance measure. The invention relates to a method for optimizing the training matrix. The method comprises the following steps of: taking the Euclidean distance of the standardized raw data as the judgment standard of the abnormal data, and removing the abnormal data with the Euclidean distance z score of more than 2, and establishing a main element combined with the elimination of the abnormal value based on the distance metric; The fault detection strategy is analyzed, and the self-adaptive main element is established by using the strict self-adaptive abnormal data cyclic elimination method based on the Q statistic threshold Q value as the judgment standard of the abnormal data. The main purpose of the two methods is to reduce the abnormal data to the main element by eliminating the data from the collection center in the original data. The influence of the cross-projection space. In the aspect of data optimization, the method of wavelet transform is used to optimize the original training data and the follow-up. The data is used to remove the noise in the data. The wavelet transform has the variable hierarchy, so it is further compared and analyzed the different wavelength division. The effect of the solution level on the detection efficiency. The more the decomposition level The more the detection efficiency is, the more statistical it is to test the standard optimization. The efficiency of fault detection is improved by cross-checking of quantity. The statistics of Q statistics, T ~ 2 statistic and HawksT ~ 2 _ H statistic are compared and analyzed. The detection efficiency of different sensors of different sensors can be obviously improved by cross-checking the statistics of the statistics of the main element space _ T-2 and the statistic quantity of the residual space _ Q and the statistic quantity of Hawkins T-2 _ H. in order to further improve the timely detection of the fault, the average value of the statistical quantity of the training matrix Q is used as the expected mean value, the accumulation and control charts of the Q statistics are adopted to carry out on-line detection and the efficiency analysis, and the time accumulative property of the error is utilized. The results show that the method can improve and improve the fault detection efficiency of the water chilling unit sensor, so as to promote the transmission
【學位授予單位】：華中科技大學
【學位級別】：博士
【學位授予年份】：2013
【分類號】：TU831.4
【參考文獻】
相關(guān)期刊論文 前10條
1 邱天;丁艷軍;吳占松;;基于霍金斯指標的傳感器故障重構(gòu)研究[J];傳感器與微系統(tǒng);2006年10期
2 畢小龍;王洪躍;司風琪;徐治皋;;基于核主元分析的傳感器故障檢測[J];動力工程;2007年04期
3 鄧勇;王彥;王超;;空調(diào)系統(tǒng)傳感器故障診斷方法[J];電子科技;2011年06期
4 孫宇乾;童創(chuàng)明;李安平;馮有前;;基于小波分析的信噪分離方法研究[J];彈箭與制導學報;2005年S8期
5 王希武;董光波;謝桂海;;基于小波變換的核磁共振FID信號的去噪方法研究[J];核電子學與探測技術(shù);2008年02期
6 王海清,余世明;基于故障診斷性能優(yōu)化的主元個數(shù)選取方法[J];化工學報;2004年02期
7 何平;剔除測量數(shù)據(jù)中異常值的若干方法[J];航空計測技術(shù);1995年01期
8 陳友明 ,郝小禮 ,彭建國;空調(diào)系統(tǒng)中傳感器故障檢測與診斷方法的研究[J];測控技術(shù);2002年11期
9 汪云亮;卜樂平;;應(yīng)用小波變換進行壓縮中分解層次的一種確定方法[J];艦船電子工程;2006年02期
10 孫勇;景博;覃征;張波;;基于小波分析的信噪分離方法研究[J];計量學報;2006年02期
本文編號：2355363
本文鏈接：http://www.sikaile.net/kejilunwen/sgjslw/2355363.html
