Research for Classification Method of Battery Based on State of Health

Transcription

1 pp Research for Classfcaton Method of Battery Based on State of Health Yu Zhlong and L Ran School of Automaton, Harbn Unversty of Scence and Technology, Harbn, Chna E-mal: zlyu@hrbust.edu.cn Abstract In order to solve the problems that lthum-on power battery cannot reflect state of health(soh) n sortng process, the parameters whch can reflect the battery SOH, such as capacty, AC resstance and self-dscharge current, were used as the nput vector of battery sortng model, fuzzy c-means clusterng analyss and support vector machne based on cross valdaton algorthm to the battery for classfcaton and recognton were used, and lthum power battery sortng model was establshed and the same batch of power battery were separatng tested and accordng to the expermental results, batteres was dvded nto groups. The test results showed that: battery electrochemstry was havng a good consstency. The varaton of capactance was less than 5% whle there was 1500 cycle lfe. Keywords: power battery; state of health; fuzzy c-means clusterng analyss; SVM 1. Introducton There are many parameters reflectng the lthum power battery capacty performance [1] and health status, ncludng physcal propertes, electrochemcal propertes. After the end of the battery package producton and formaton, we need combnaton classfcaton of the battery characterstcs, to meet the needs of groupng of battery. Electrochemcal consstency detecton accuracy drectly affect battery performance [2] after groupng. The parameter of battery characterstcs manly ncludes: the qualty of the battery, the open crcut voltage, dscharge plateau voltage, current, depth of dscharge, the rated capacty, AC resstance, self-dscharge rate and SOC SOH. The SOC, SOH s drectly reflectng the power of battery and the degree of deteroraton, but n practcal applcaton process there s not a partcularly good way to measure, and parameters whch can be measured can be the bass for the classfcaton of ndustral feld. Due to the wde use of power battery, the core technology of the management system whch can mprove monomer battery electrochemcal consstency and mprove the effcency of the battery group demands an urgent mprovement. Thus, electrochemcal separaton technology consderng battery group measured volume has become a hot topc n the research of battery technology. Conventonal battery classfcaton methods are: Statc volumetrc method, voltage assembly method, charge and dscharge effcency method, nternal matchng method, nternal resstance and capacty sortng,cyclc data test method. Multcrtera sortng method [3-13]. Paolo Raspaand the others [3] n order to mprove the cell unformty, used self-organzng map neural network method, to develop a method for the homologous cell selecton and classfcaton. Zhhua Duo, Gechen L [4], put forward a strategy of rapd cell classfcaton, usng fuzzy decson makng comparatve analyss of battery charge and dscharge curves and dscharge capacty, and completed a sortng of battery consstency. Due to the dfference n the battery producton process n control of electrochemcal characterstcs and packagng technology, the electrochemcal propertes of the battery of ISSN: IJSIA Copyrght c 2016 SERSC

2 the same formula and the same batch wll dffer [10], and the electrochemcal dfference does not have a very accurate standard to measure the electrochemcal characterstcs of the monomer battery belongs to whch category, so n the process of groupng, cell sortng s only accordng to certan rules of sortng, as far as possble to reflect the electrochemcal cell consstency[14-15]. 2. Battery Classfcaton Model Due to the battery n the classfcaton process does not have a quanttatve classfcaton standard, each classfcaton s as far as possble to mprove the consstency of the electrochemcal cell n the same category. Therefore, the selecton of parameters reflectng the electrochemcal characterstc and categores of the correct dvson of the battery s the key to battery classfcaton technology. So the man dea of classfcaton method n the essay s: use data mnng technques to obtan cell classfcaton data, and use fuzzy c-means clusterng algorthm to determne the cluster centers of the cell samples, and use characterstcs of the sample to determne cell classfcaton numbers and use SVM calculaton method to the category formed by the sample for dscrmnatve tranng and classfcaton [16-17] Battery Parameter Selecton Consderng the work characterstcs of electrc core n the batteres wll reflect the battery deteroraton of the AC resstance, self-dscharge current and actual capacty, SOC and measurable quanttes n battery operaton process: use the open crcut voltage, dscharge plateau voltage, voltage change rate n the process of dschargng as battery electrochemcal separaton sample data of battery sortng, and obtan the sample set for x Vo Vds SOC V R m Isds P [,,,,,,, ] Among them, V o s open crcut voltage, V ds s dscharge plateau voltage, V s voltage change rat; R s AC resstance, m s the qualty of battery, I sds s selfdscharge current, P s actual dscharge capacty of battery. Usng the eght measurable quanttes adopted by formula (1) as the sample data of battery classfcaton can fully reflects the battery's electrochemcal and physcal propertes, and the AC resstance of battery, self-dscharge current reflects the battery's SOH, dscharge capacty and the open crcut voltage, and voltage of dscharge platform reflects the lfe of the battery, and SOC and voltage changng rate n the course of the dschargng s a reflecton of the dynamc characterstcs of the battery but cycle, and battery qualty reflects the physcal quantty of actve substances of the cell. So cell sample classfcaton set formed by the actual measurement parameters of the battery can fully reflect characterstcs of the battery Establshment of FCM-SVM classfcaton model The specfc steps to be establshed based on the FCM-SVM battery classfcaton model are as follows: Step 1, Use the fuzzy C mean clusterng algorthm to classfy the all learnng samples of battery, get the clusterng center m, 1,2,, c, Step 2, Calculate the Eucldean dstance between cluster centers,.e., d, j m mj,, j 1,2, c, j (1) 48 Copyrght c 2016 SERSC

3 Step 3, Calculate dstance between each clusterng center and other clusterng centers, and order and number them accordng to the dstance, get dstance between d wth other clusterng centers., j p p p 1 2 c 1 Step 4, Compare the mnmum dstance of dfferent clusterng centers, then have the cluster centers ranked accordng to the dstance. The clusterng centers are ranked accordng to the dstance from large to small sort, and we get a new clusterng center label for M1, M2,, Mc. Step 5, Frst set the sample sequence set correspondng to M1 to be the postve class, and set sample sequence sets correspondng to other clusterng centers to be the negatve class, then structure support vector machne SVM 1, and set the sample sequence set correspondng to M1 to be the postve class, M3, M4,, Mc to be the negatve class, and structure support vector machne SVM 2, etc., fnally get c 1support vector machne classfer. In step 5, we need to buld c 1 support vector machne (SVM) classfer and a sngle bnary support vector machne usng C-SVC model to tran and specfc forms are as follows: Step 1, Set a known tranng set: Among them, 1,2,, l T ( x, y ) ( X Y) l x s cell sample collecton, and 8 x X R, y Y {1, 1}, After the tranng samples and test samples were normalzed, have the samples normalze nto the range of [0,1]. Step 2, The SVM kernel functon s selected as the radal bass functon. K( x, x ) exp( x x ), 0 2 Step 3, Accordng to the cross valdaton method to choose the approprate penalty factor, radal bass functon parameter. Tran support vector machne classfer, get the model and test t. Use fuzzy C means clusterng analyss and support vector machne algorthm based on cross valdaton are used to classfy the battery. The specfc model s shown n Fgure 1. Copyrght c 2016 SERSC 49

4 Fgure 1. SVM Algorthm based on FCM 3. Verfcaton of Classfcaton Experment 3.1. Data collecton Experments are usng Arbn battery chargng and dschargng equpment and the expermental object s 120 batches of IFR18650P lthum ron phosphate monomer battery, and ts rated capacty s of 1100mAh. Expermental way: under the room temperature of 20, usng standard current 1C of battery for chargng and dschargng experments, n small current of 1c we thnk that battery charge and dscharge effcency s 100%. The charge and dscharge current should be kept constant, and the ampere hour method s used to calculate SOC value; after electrc core chargng s completed and SOC s 100%, let the battery stew for 30 mnutes, and record the battery termnal voltage n statc state and measure the AC resstance, qualty of electrc core; n the dscharge, record voltage of electrc core dschargng platform, voltage changng rate and dschargng capacty of the battery. In dfferent SOC of battery, measure self-dscharge flow of electrc core randomly, and fnally get the 220groups of measurement data, as shown n Table 1 Table 1. Sample Data of Battery Parameters Samples OCV/V DPV/V SOC AC Resstance /mω Qualty/mg Selfdscharge current/μa Power /mah % % % % % Model Tranng and Smulaton Use 220 groups of data as the tranng and testng data for battery classfcaton 50 Copyrght c 2016 SERSC

5 model, ncludng 120 groups as the tranng data, another 100 groups for the test data. Have data clusterng analyss wth FCM algorthm, and the clusterng result curve as shown n Fgure 2. Fgure 2. Results of Cluster Analyss Accordng to the result of the clusterng analyss, usng the 120 sets of data for teraton, we got the most reasonable classfcaton 2classe, as shown n Fgure 2. Through clusterng analyss for measured data we obtaned 2 class, and elmnated the dfferences, and wth usage of SVM classfer we could classfy the sample cell data. In the tranng usng LIBSVM Matlab toolbox to tran two classfers, the tranng samples adopted the 120 groups of known data n the measurng samples, and used the remanng 100 sets of data for cross valdaton. Valdaton results are shown n Fgure 3. Fgure 3. Contour Map of Parameter Selecton Copyrght c 2016 SERSC 51

6 As s shown n Fgure 3, by cross valdaton, the optmum parameters g c was selected. Usng optmal parameters to tran the classfcaton model, we got the 100% classfcaton accuracy rate. 4. Expermental Verfcaton To verfy the correctness of battery sortng results, we had a battery sortng of the same batch n accordance wth the model, and took 10 n the same category for battery cycle lfe experment. Expermental scheme s: under the temperature of 20, have1c charge and dscharge experments on the battery respectvely, wth charge cutoff voltage of 3.2V, dscharge cut-off voltage of 2.0V, and record the dscharge capacty of the battery each nterval of 50 cycles. The result s shown n Fgure 4. Fgure 4. Dscharge Capacty of Battery n Cycle Testng As s shown n chart 5, battery classfcaton algorthm can better reflect the battery electrochemcal consstency In cycle lfe experment of 10 batteres, the battery capacty can mantan a good attenuaton trend after 1500 cell cycles, and wth the relatve dsperson of battery capacty n a sngle measurement, capacty dfference maxmum less than 5% can ndcate that cell electrochemstry s n good consstency. Battery sortng method based on FCM-SVM can acheve satsfactory effect. 5. Concluson Ths essay adopts capacty, nternal resstance and self-dscharge current whch can reflect the electrochemcal propertes of lthum on power battery as the nput parameters of the battery model to construct the battery classfcaton model. Expermental results shows that usng the battery sortng results whch adopted ths essay s classfcaton algorthm can better take nto account the battery health effects but the man problems s battery parameter measurement n the classfcaton algorthm wll take a long tme and s not sutable for large quanttes of ndustral feld applcaton. 52 Copyrght c 2016 SERSC

7 Acknowledgement Ths paper s partally supported by Technologcal Innovaton Foundaton for Leaders of Dscplnes n Scence of Harbn (2014RFXXJ032). References [1] RAMADASS P, HARAN Bala and WHITE Ralph, Mathematcal modelng of the capacty fade of Lon cells [J], Journal of Power Sources, vol. 123, (2003), pp [2] L. Ran, W. Junfeng and W. Hayng, Research on consstency classfcaton of battery based on mult pont spectrum method, Power technology, vol. 10, (2010), pp [3] P. Raspa, L. Frncon and A. Mancn, Selecton of Lthum Cells for EV Battery Pack Usng Self- Organzng Maps, J Automotve Safety and Energy, vol. 2, no. 2, (2011), pp [4] D. Zhhua and L. Gechen, An algorthm for the dentfcaton of the power of waveform recognton, Power technology, vol. 22, no. 4, (1998), pp [5] W. Yao, Battery consstency matchng technology for electrc vehcles, Electrcal ndustry, vol. 6, (2002), pp [6] C. Shaoan, Z. Janqng and L. Hong, Electrochemcal mpedance spectroscopy (MH-N) analyss of cell cycle declne, Power supply technology, vol. 23, (1999), (S1) pp [7] R. L, J. Wu and H. Wang, Research on battery consstency classfcaton based on mult pont spectrum method, Power supply technology, vol. 10, (2010), pp [8] Y. Lnru, Z. Jnge and W. Yan, MH-N battery classfcaton method based on data fuson theory, Power supply technology, vol. 3, (2000), pp [9] K. Dae-Keun and S. Heon-Cheol, Investgaton on cell mpedance for hgh-power lthum-on batteres [J], J Sold State Electrochem, vol. 11, (2007), pp [10] G. Wejun, S. Zechang and W. Xuezhe, Two Methods Studyng Internal Resstance for Lthum-on Battery Lfe Estmaton Used n EV [J], J Automotve Safety and Energy, vol. 2, no. 2, (2011), pp [11] H. Wang, L. He and J. Sun, Study on correlaton wth SOH and EIS model of L-on battery [C]//Strategc Technology (IFOST), th Internatonal Forum on. IEEE, vol. 1, (2011), pp [12] W. Hayng, L. Yang and F. Hang, Estmaton of state of charge of batteres for electrc vehcles [J], Internatonal Journal of Control and Automaton, vol. 6, no. 2, (2013), pp [13] Z. Chen, Desgn and SOC estmaton of lthum on battery montorng and management system for electrc vehcle [D], Nanjng Unversty of Scence and Technology, (2013), pp [14] W. Hayng, X. Chao and F. Q, Lthum Battery SOC Estmaton Method Study Based on Prncpal Component Analyss[J], Internatonal Journal of Control and Automaton, vol. 8, no. 7, (2015), pp [15] X. Hu, Study on SOH estmaton method of power L on battery [D], Jln Unversty, (2013). [16] Y. Yang, Study on the method of fuzzy C mean clusterng n fuzzy clusterng analyss [D], Shenyang: Shenyang Unversty of Technology, (2009). [17] Y. Hajun, Research on fuzzy clusterng analyss technology and ts applcaton [D], Hefe: He Fe Unversty of Technology, (2006). Copyrght c 2016 SERSC 53

8 54 Copyrght c 2016 SERSC