


摘 要: 為了解決非線性系統中故障參數估計問題,提出多重漸消因子強跟蹤平方根無跡卡爾曼濾波(MST?SRUKF)算法。MST?SRUKF將多重漸消因子引入到協方差矩陣平方根中,推導適用于平方根無跡卡爾曼濾波(SRUKF)的漸消因子計算公式,從而實時調整SRUKF中的增益矩陣,保證其對模型存在較大誤差或者突變情況下的濾波精度。實驗結果表明,相比于SRUKF和強跟蹤無跡卡爾曼濾波(STUKF),MST?SRUKF對故障參數具有更高的估計精度。
關鍵詞: 強跟蹤濾波; 故障參數; 平方根無跡卡爾曼濾波; 非線性系統
中圖分類號: TN713?34; TP273 文獻標識碼: A 文章編號: 1004?373X(2016)20?0015?05
Abstract: To solve the problem of nonlinear system fault parameter estimation, the multiple fading factors strong tracking square root unscented Kalman filter (MST?SRUKF) algorithm is proposed. The multiple fading factors are introduced into covariance matrix square root by means of MST?SRUKF. Then the fading factor computational formula suitable for square root unscented Kalman filter (SRUKF) is deduced to adjust the gain matrix in SRUKF in real time to ensure filter accuracy when the model has big error or changes abruptly. The experiment result shows that, compared with SRUKF and strong tracking unscented Kalman filter (STUKF), the MST?SRUKF has higher estimation accuracy of fault parameter.
Keywords: strong tracking filter; fault parameter; square root unscented kalman filter; nonlinear system
0 引 言
一般情況下,系統的參數不能被直接測量,卡爾曼濾波算法能夠根據系統的輸出,間接估計出現故障的參數,被廣泛應用于故障參數辨識領域[1?3]。實際中大部分系統具有一定的非線性[4],因此需要采用非線性濾波算法實現故障參數估計[5]。擴展卡爾曼濾波(EKF)作為一種常用的非線性濾波算法[6?7],方法簡單且易于實現,但是當系統的非線性較強時,EKF的估計精度較差,甚至可能濾波發散。為了解決這個問題,有學者提出了無跡卡爾曼濾波(UKF)算法[8?9],文獻[10]根據不同UKF的殘差實現故障診斷,但是UKF在數值計算過程中存在著舍入誤差,隨著濾波的進行,累積的舍入誤差可能會導致濾波協方差矩陣不再保持正定性,造成濾波過程的不穩定。平方根UKF(SRUKF)能夠有效地解決舍入誤差引起的不穩定問題[11?12],文獻[13]利用SRUKF對渦扇發動機進行故障診斷,實驗結果表明SRUKF的估計誤差小于EKF算法和UKF算法。但是,當系統模型的不確定性較大或者系統出現突變故障時,普通的SRUKF對故障參數的估計精度不高,甚至可能出現濾波失效。文獻[14]提出了帶多重漸消因子的強跟蹤濾波算法,其通過在預測誤差方差陣中引入漸消因子,增強濾波算法對模型存在較大誤差情況下的魯棒性。文獻[15]將強跟蹤濾波應用于電路參數跟蹤問題,實時診斷電路的元件故障。文獻[16]提出強跟蹤UKF(STUKF)算法,但是與普通UKF相似,STUKF存在著濾波不穩定問題。
文獻[14]提出的強跟蹤濾波算法在求得漸消因子時,需要計算狀態方程和量測方程的一階偏導矩陣,但是SRUKF在本質上屬于非偏導矩陣計算的濾波算法,為此,本文推導出適用于SRUKF的多重漸消因子計算公式,使其適應SRUKF的本質特性,提出帶多重漸消因子的強跟蹤SRUKF(MST?SRUKF),并通過實驗證明了MST?SRUKF方法的有效性。
1 SRUKF多重漸消因子計算
式中:[λk]表示多重漸消因子矩陣;[λik≥1]為第i個狀態變量對應的漸消因子; [αi≥1]為第i個漸消因子的比例系數,如果某個狀態變量對應的狀態方程的誤差較大,則選擇一個較大的值,以增強濾波算法對該狀態變量的強跟蹤程度。
上面計算多重漸消因子公式需要計算狀態方程和量測方程的偏導矩陣。而SRUKF算法本質上為不需要計算偏導矩陣的濾波算法,為了保證多重漸消因子強跟蹤SRUKF的這一特點,下面推導SRUKF中基于非偏導矩陣計算的漸消因子計算公式。
由圖4可知,在k=200和k=400時刻MST?SRUKF的漸消因子出現峰值,其利用較大的漸消因子來增強對突變情況的跟蹤能力,提示了MST?SRUKF估計突變參數的機理。
通過在初始值已知的緩變故障、狀態初始值未知的緩變故障、模型參數存在容差的緩變故障和突變故障情況下對參數V的估計結果表明,本文提出的MST?SRUKF可以較好地對變化趨勢未知情況下的故障參數進行估計,其估計精度略高于STUKF,而對于普通SRUKF,由于其缺乏對故障參數變化函數未知情況下的強跟蹤能力,估計誤差較大。
4 結 論
本文研究了一種基于MST?SRUKF的不可直接測量故障參數估計算法。該算法通過在SRUKF的協方差矩陣平方根中引入多重漸消因子,增強其對模型誤差較大和狀態變量突變情況下的跟蹤能力。采用普通SRUKF和強跟蹤UKF(STUKF)作為對比方法,實驗結果表明,在緩變故障和突變故障情況下,本文提出的MST?SRUKF方法對故障參數的估計精度均高于其他兩種方法。
參考文獻
[1] FAN Jiajie, YUNG Kam?Chuen, PECHT Michael. Predicting long?term lumen maintenance life of LED light sources using a particle filter?based prognostic approach [J]. Expert systems with applications, 2015, 42(5): 2411?2420.
[2] JESUSSEK M, ELLERMANN K. Fault detection and isolation for a railway vehicle by evaluating estimation residuals [J]. Procedia lutam, 2015, 13(1): 14?23.
[3] ALROWAIE F, GOPALUNI R B, KWOK K E. Fault detection and isolation in stochastic non?linear state?space models using particle filters [J]. Control engineering practice, 2012, 20: 1016?1032.
[4] WANG Shiyuan, FENG Jiuchao, TSE Chi K. Novel cubature Kalman filtering for systems involving nonlinear states and linear measurements [J]. AEU?international journal of electronics and communications, 2015, 69(1): 314?320.
[5] WANG Jinjiang, WANG Peng, GAO R X. Enhanced particle filter for tool wear prediction [J]. Journal of manufacturing systems, 2015, 36: 35?45.
[6] SINGLETON R K, STRANGAS E G, AVIYENTE S. Extended Kalman filtering for remaining?useful?life estimation of bearings [J]. IEEE transactioins on industrial electronics, 2015, 62(3): 1781?1790.
[7] KULIKOV Gennady Yu, KULIKOVA Maria V. High?order accurate continuous?discrete extended Kalman filter for chemical engineering [J]. European journal of control, 2015, 21: 14?26.
[8] PARTOVIBAKHSH M, LIU Guangjun. An adaptive unscented Kalman filtering approach for online estimation of model parameters and state?of?charge of lithiumion batteries for autonomous mobile robots [J]. IEEE transactions on control systems technology, 2015, 23(1): 357?363.
[9] SULLIVAN Edmund. An intuitive look at the unscented Kalman filter [J]. Journal of the Acoustical Society of America, 2014, 136(4): 2213?2216.
[10] MIRZAEE Amin, SALAHSHOOR Karim. Fault diagnosis and accommodation of nonlinear systems based on multiple?model adaptive unscented Kalman filter and switched MPC and H?infinity loop?shaping controller [J]. Journal of process control, 2012, 22: 626?634.
[11] 胡志坤,劉斌,林勇,等.電池SOC的自適應平方根無跡卡爾曼濾波估計算法[J].電機與控制學報,2014(4):111?116.
[12] JAFARZADEH S, LASCU C, FADALI M S. Square root unscented kalman filters for state estimation of induction motor drives [J]. IEEE transactions on industry applications, 2013, 49(1): 92?99.
[13] 胡宇,張世英,楊月誠,等.基于超球體平方根無跡Kalman濾波算法的渦扇發動機氣路部件故障診斷[J].航空動力學報,2014,29(3):689?695.
[14] 周東華,席裕庚,張鐘俊.一種帶多重次優漸消因子的擴展卡爾曼濾波器[J].自動化學報,1991,17(6):689?695.
[15] 李雄杰,周東華.基于強跟蹤濾波器的模擬電路故障在線診斷方法[J].電工技術學報,2007,22(5):13?17.
[16] 王小旭,趙琳,夏全喜,等.基于Unscented變換的強跟蹤濾波器[J].控制與決策,2010,25(7):1063?1068.