Real Time Determination of Rechargeable Batteries Type and the State of Charge via Cascade Correlation Neural Network

Transcription

1 ELEKTRONIKA IR ELEKTROTECHNIKA, ISSN , VOL. 24, NO. 1, 2018 Rel Time Determintion of Rechrgeble Btteries Type nd the Stte of Chrge vi Cscde Correltion Neurl Network Rif Byir, Emel Soylu Deprtment of Mechtronics Engineering, Technology Fculty, Krbuk University, Krbuk 78050, Turkey 1 Abstrct Btteries re used to store electricl energy s chemicl energy. They hve wide using re from portble equipment to electric vehicles. It is importnt to know the stte of chrge of bttery to use it efficiently. In this study, grphicl user interfce is developed using visul progrmming lnguge to monitor the electricl situtions of btteries. Cscde neurl network, which is one of the most chosen rtificil neurl networks, is used to determine the type nd stte of chrge of btteries. The softwre is ble to identify type nd stte of chrge of btteries online. Led cid, Lithium Ion, Lithium polymer, Nickel Cdmium, Nickel Metl Hydride rechrgeble btteries re used in experiments. The experimentl results indicte tht ccurte estimtion results cn be obtined by the proposed method. Index Terms Artificil neurl network; Bttery monitoring softwre; Rechrgeble btteries; Stte of chrge determintion. I. INTRODUCTION Although btteries seem to be simple, they re nonliner nd complex systems becuse of their physicl nd chemicl structure. Depending on the development of technology the usge re of btteries is incresing. It is importnt to estimte the stte of chrge (SOC) of the bttery ccurtely in bttery mngement systems to use the bttery efficiently. Mthemticl, electricl, electrochemicl methods re used to estimte to the SoC of the bttery; Mthemticl nd electrochemicl methods include complex equtions, nd these equtions must be redesigned for other types of btteries. The electricl method is esy to clculte, nd the user cn develop bttery model by looking t dtsheet of the bttery or by mesuring the bttery prmeters. Stisfctory bttery models cn be chieved using dtsets generted by electricl methods. Vrious SoC estimtion methods re proposed in the literture tht uses experimentl dtset. The dtset used in electricl bttery models cn be obtined by monitoring bttery voltge, current, electrochemicl impednce spectroscopy, etc. nd prmeters. Dt collecting is possible by mesuring prmeters while chrging the bttery, dischrging the bttery or in stedy stte. Most known methods re ANN, fuzzy logic, Klmn filter nd rdil bsis function neurl network (RBFNN). The rtificil Mnuscript received 11 August, 2017; ccepted 21 October, neurl network (ANN) method is esy to build becuse it doesn t hve complex mthemticl equtions nd works with high ccurcy [1] [7]. Developing fuzzy rules nd membership functions nd fuzzy outputs re difficult. It requires lot of dt nd expert knowledge to develop fuzzy system [8] [10]. The Klmn filter is computtionlly complex nd requires conditionl independence of the mesurement errors [11] [15]. RBFNN is esy to build, but it is slow when the dtset is lrge [16]. In [17] model is developed to estimte the usble cpcity of led cid btteries used in electric vehicles. High ccurcy is obtined using ANN. In [18] three lyer feed forwrd neurl network is used to estimte SoC of btteries. SoC is estimted under 5 % error rte with this method. In [2] bck propgtion neurl network is pplied successfully to estimte SoC of btteries used in electric vehicles. The SoC of the bttery cn be estimted while chrging, dischrging the bttery nd the stedy stte fter chrging. The open circuit voltge of the bttery is pplied s n input prmeter of neurl network. The simultion results suggest tht this method is suitble for hybrid electric vehicles. In [19] three-lyer bck propgtion neurl network is used to estimte the SoC of high powered bttery. Five input prmeters re pplied to the neurl network; these re bttery dischrge current, totl mpere-hour, open circuit voltge of the bttery, timedependent verge open circuit voltge nd twice of time dependent verge open circuit voltge. The dtsets re obtined while dischrging the bttery from full chrge to full dischrge. The Levenberg Mrqurt lgorithm is used in trining. Simultion nd mesurement results re compred to test the performnce of the rtificil neurl network. After ten minutes the SoC of the bttery cn be estimted with fewer thn 5 % error rte. Knowing the percentge of energy left in bttery gives the user informtion on how much time bttery will continue to operte without rechrging. It is importnt to chrge nd dischrge the bttery in the correct form to prevent fires nd explosions; furthermore, proper use of the bttery provides more efficiency nd longer life for users. On the other side of the spectrum, improper use will reduce the lifetime of the bttery, nd the defective bttery cretes chemicl pollution in nture. In this study, n experimentl setup is developed to monitor the btteries' electricl 25

2 prmeters. Specil softwre is designed to sve the mesured dt systemticlly nd determine the type nd SoC of rechrgeble btteries online. The softwre is lso ble to stop n experiment while the bttery is out of voltge, current or temperture boundries. Cscde Correltion Neurl Network (CCNN) is used to determine the type nd SoC of btteries while dischrging the bttery t constnt lod. The terminl voltge of the bttery, current, power dt is used to generte the dtset. Led cid (), Lithium Ion (), Lithium polymer (LiPo), Nickel Cdmium (), Nickel Metl Hydride () rechrgeble btteries re used in the experiments. The Wtt-hour vlues of experimentl btteries re chosen which re very similr; this is to successfully determine the type of btteries tht hve nerly the sme properties. The difference between this study nd other cdemic studies is the ide of determining the type of bttery nd the chrging, dischrging form. The experimentl setup nd dtbse structure cn be n exmple for people who re working on monitoring the electricl behviours of btteries. The gol of this study is to determine type nd SoC of rechrgeble btteries vi CCNN with high ccurcy. There re mny pplictions to determine the SoC of bttery but estimting the type of btteries is new study. Estimting of the bttery vi CCNN is nother innovtion of the study.,,, nd rechrgeble btteries re used in the experiments. II. RECHARGEABLE BATTERIES Btteries re prt of our everydy lives t the moment, ll of the wireless equipment tht opertes using electricl energy tke this power from btteries. In the modern dy, portbility is importnt, which in turn hs incresed the importnce of btteries. The usge rte of btteries in country is directly proportionl to the usge of technology. There re mny different rechrgeble bttery types, in this study,,,,, rechrgeble btteries hve been used. btteries re suitble for pplictions where weight nd dimensions of the bttery re not necessry. Therefore they re chep. Mostly these btteries re used in vehicles, medicl devices, nd motorized chirs for disbled people nd emergency lighting spotlights nd uninterruptible power supplies. btteries re more stble nd lightweight; the orgnic electrolyte provides the prcticl cell voltge to be bove 4 V. They hve high energy densities, nd they provide esy pplictions without the need for connecting severl cells in series [20] [22]. They re used in lptops, mobile phones, music plyers nd much more digitl portble devices. LiPo btteries re rechrgeble btteries tht continue on from the bttery technology. LiPo btteries hve high energy densities ccording to their volume nd weights becuse of this; they hve lrge usge re. They re used in electricl vehicles, lptops, nd mny electronic pplictions. The most importnt property of btteries is tht they hold the cpcity inside it without losing it, in essence, it hs the sme cpcity two weeks fter the lst chrging time. btteries re used in single or grouped form in drills, mesuring instruments, etc. The fst chrge of these types of btteries decreses their using life. With stndrd chrging, btteries hve n verge life of 5 yers. btteries hve more energy density thn btteries, but their rechrgeble number is lower. They re used in lptops, mobile phones, cmers, toys, etc. There is some memory effect in btteries. A. Experimentl Setup III. MATERIAL AND METHOD In electricl bttery test setups the experimentl setup is shped ccording to the mesuring prmeters. If the bttery s internl resistnce will be mesured, the internl resistnce meter is used, if the current will be mesured, the current sensor will be used, if voltge is mesured, the voltge sensor must be used. A chrger must be used to chrge the bttery, nd lod must be used to dischrge the bttery. If the temperture prmeter is necessry temperture sensors must be used in the system. The collected dt cn be processed by computer or embedded systems. Successful bttery models cn be obtined with collected dt from electricl mesurements. In this study, open circuit voltge, current, power, lod, mbient temperture nd bttery temperture re ll mesured during chrging nd dischrging of the btteries. The mesurement setup of this study is given in Fig. 1. To chrge the bttery, Imx B8+ chrge equipment nd to dischrge bttery Arry 3711A progrmmble DC lod equipment is used. A circuit is designed to choose the chrger or lod from softwre. A LTS25-NP current sensor, LV25P voltge sensor nd LM35 temperture sensor re lso locted on this circuit. Three btteries cn be connected to this circuit nd the experiment bttery cn be chosen from the softwre. There re lso contcts to control buttons of the chrger on this circuit. The contcts on this circuit re controlled by digitl I/O on Advntech USB-4716 dt cquisition (DAQ) crd. The output of K-type thermocouple is connected to digitl I/O of DAQ crd through the circuit. The progrmmble DC lod is connected to the PC vi Arry 3312 Seri-USB port converter. Squre codes re glued to ll btteries tht define their identity. Perkon Spider SP400 squre code reder is used to red codes. This equipment is connected to the computer vi USB port. A web cmer is used to wtch the experimentl setup. Seri to USB port converter Progrmmble DC lod Bttery chrger DC Power supply K type termocupl Bttery Fig. 1. Mesurement setup. Web-Cm Temperture trnsmitter D0 D1 B3 B2 B1 (D2-D3) Up (D4) Current Sensor Strt (D5) Down (d6) Stop (D7) Kullnıcı Aryüzü Computer Voltge sensor Tempertu re sensor Selection nd sensor circuit Squre code reder While dischrging the bttery, the current, voltge, lod nd power prmeters re tken from the lod equipment, while chrging the bttery, current nd voltge prmeters AGND A0 A1 A2 A3 DGND D0 D1 D2 D3 D4 D5 D6 D7 Dt cquisition crd 26

3 re mesured with sensors nd trnsferred to computer vi DAQ crd. While chrging, the bttery lod nd power prmeters re clculted using voltge nd current dt. Ttec 6 V 1.3 Ah bttery, Pnsonic CGR18650CG 3.7 V 2.2 Ah bttery, Power Xtr PX V 2 Ah bttery, AA Portble Portble Corp. CD- SC2200P 3.6 V 2.2 Ah bttery nd Gold Pek Group GP211AFH 3.6 V 2.1 Ah bttery re used in the experiments. The technicl informtion of these btteries is given in Tble I. TABLE I. THE TECHNICAL PROPERTIES OF BATTERIES. Li- Li- Property Ion Po Nominl voltge (V) Nominl cpcity (mah) Nominl cpcity mx (mah) Mx operting temperture ( o C) Min operting temperture ( o C) Stndrd chrge current (ma) Stndrd chrge time (h) Fst chrge current (ma) Fst chrge time (h) Deep chrge voltge (V) Weight (gr) Cycle life C rte Overchrge voltge (V) Wh (Vx Ah) Wh % difference from verge Wh The percentge of mximum difference with verge Wh vlue is 4,15 %. The btteries cpcities re very similr, nd this property mkes it difficult to determine the bttery type. Although the cpcities of these chosen btteries re similr, their chrging types nd chrging currents re different. B. Cscde Correltion Neurl Network The CCNN is developed by Fhlmn in 10. CCNN is supervised lerning lgorithm. CCNN begins with miniml network, then utomticlly trins nd dds new hidden units one by one, creting multi-lyer structure. The CCNN rchitecture hs severl dvntges over existing lgorithms: it lerns very quickly, the network determines its own size nd topology, it retins the structures it hs built even if the trining set chnges, nd it requires no bckpropgtion of error signls through the connections of the network [23]. An untrined cscde correltion network is blnk slte; it hs no hidden units. A cscde correltion network s output weights re trined until either the solution is found, or progress stgntes. If single lyered network will suffice, trining is complete. The weights of hidden neurons re sttic; once they re initilly trined, they re not touched gin. The fetures they identify re permnently cst into the memory of the network. Preserving the orienttion of hidden neurons llows cscde correltion to ccumulte experience fter its initil trining session. Few neurl network rchitectures llow this. If bckpropgtion network is retrined, it forgets it's initil trining [24]. The CCNN rchitecture is shown in Fig. 2. Inputs + 1 Inputs + 1 Inputs Initil Stte No Hidden Units Add Hidden Unit Fig. 2. The Cscde rchitecture. Add Hidden Unit 2 Outputs Outputs Outputs Initil stte nd fter dding two hidden units. The verticl lines sum ll incoming ctivtion. Box connections re frozen, X connections re repetedly trined. CCNN combines two ides: The first is the cscde rchitecture, in which hidden units re dded only one t time nd do not chnge fter they hve been dded. The second is the lerning lgorithm, which cretes nd instlls the new hidden units. For ech new hidden unit, the lgorithm tries to mximize the mgnitude of the correltion between the new unit's output nd the residul error signl of the net. IV. SPECIAL SOFTWARE TO DETERMINE TYPE AND SOC OF BATTERIES A grphicl user interfce is developed in Visul Studio 2010 softwre in C# progrmming lnguge to monitor conditions of btteries, sving mesurement dt to dtbse to determine the type nd SoC of btteries. Users cn dd new bttery to the dtbse. Users select the test bttery, durtion of the experiment, smple time, nd choose to chrge or to dischrge the bttery. When ll the djustments re mde n experiment code is generted utomticlly. A tble is creted clled this code in the dtbse, nd the mesurement dt is sved to this tble. The mesurement dt curves cn be seen online. The mesurements sved to the dtbse before cn be listed. The bttery cn be inserted into chrge-dischrge loop sfely becuse during the experiments the bttery is controlled if it chieved to criticl limit vlues of voltge, current, nd temperture. If one of these vlue is chieved the softwre close the system utomticlly nd generte lrms. The rest periods between chrging nd dischrging re djustble. The user cn generte the dtset nd normlize the dt to recognize the bttery for CCNN nd sve it in Excel formt. The input vribles of CCNN to recognize the bttery re voltge, current, power, voltge decresing ngle nd current decresing ngle. To determine the voltge nd current decresing ngles the bttery must be dischrged for determined time. 400 second is selected 27

4 for this ppliction. The dtset to trin CCNN to determine SoC of the bttery cn be generted nd normlized. The SoC vlue is determined ccording to the mesurement dt. The input vribles of this CCNN re voltge, current, power nd time: t TA I t, 0 (1) n gi 1 gi TA ( ti 1 ti ), i 1 2 (2) m g i 1 g IA i ( t 1 ), i 1 2 i t i (3) QC SOC, (4) QMx TA IA SOC. (5) TA The SoC of the bttery cn be determined from the current curve of the bttery while dischrging the bttery. From full chrge to full dischrge the re under the current curve represents % SOC. In (1) the eqution of totl re (TA) under from full chrge to full dischrge of the bttery curve is given. In this eqution, I represents current vlue nd t represents time. In the softwre, the integrl cn be determined by the trpezoid method; this method is pplied s given in (2) where g i is current vlue of ith time, g i+1 is the current vlue of i+1 th time; t i is i th time vlue, t i+1 is i+1 th time vlue; n is number of mesurements. For m th mesurement dt, the re (IA) under up to this time cn be determined ccording to (3), IA represents the used cpcity of the bttery. The SoC of the bttery cn be determined by dividing the remining cpcity of the bttery to full cpcity of the bttery s given in (4). In this eqution, QC is the remining cpcity of the bttery nd QMx is the mximum cpcity of the bttery. So the rte of remining cpcity of the bttery cn be derived from (5). In this eqution (TA-IA) gives the remining cpcity of the bttery nd TA gives the mximum cpcity of the bttery [25]. useful. From this ide, we initilly tried to determine the type of bttery in Mtlb. There is bttery block in Simulink nd it supports mny types of rechrgeble btteries. By using the full chrge to full dischrge vlues of voltge nd current nd using the CCNN method we succeeded in determining the type of bttery. Then we studied this method in rel ppliction. The rchitecture of the CCNN used to determine the type of bttery is given in Fig. 3. There re five inputs, one hidden lyer nd five outputs in this rchitecture. The input vlues re current, voltge, power, V θ nd i θ. V θ is the ngle of the voltge drop nd i θ is the ngle of the current drop while dischrging the bttery. These vlues re determined by clculting the difference of vlues over 400 seconds. This time vlue is determined by trying nd considered s time unit. V is the difference of voltge vlue, nd i is the difference of current vlue fter 400 seconds. iθ is rctn( i) nd V θ is rctn( V). The input vlues of CCNN is normlized between 0 nd 1 dividing input vlue by the bsolute vlue of the mximum vlue of the input vector. In this eqution x is the normlized vlue, x is the vlue to be normlized nd x is the mximum vlue of input vector. The NN hs five outputs nd gives the result between 0 nd 1 for ech neuron. The mximum of these vlues represents the type of the bttery. Fig. 3. The structure of CCNN used to determine the type of bttery. The rchitecture of CCNN to determine the SoC of the bttery is given in Fig. 4. V. EXPERIMENTAL STUDY To obtin the dtset for usge to determine the type nd SoC of btteries the btteries re full chrged firstly then they re fully dischrged under constnt lods. 3 Ω, 5 Ω, nd 10 Ω constnt lod vlues re used. All experiments re done in n mbient temperture nd with helthy btteries. This experimentl dt is used s trining dt for CCNN to determine the type nd SoC of btteries. VI. DETERMINING TYPE AND SOC OF BATTERIES VIA CCNN There re mny studies on estimting the SoC of bttery but estimting the type of bttery is new study. In future, the usge of electricl crs will increse, nd the importnce of btteries will increse ccordingly. The users will not wit t chrge sttions. Insted they will chnge the bttery pcks in these situtions. So softwre tht determines the type nd SoC of bttery nd gives the informtion of how to chrge nd the usge of this bttery pck will be very Fig. 4. The structure of CCNN used to determine the SoC of bttery. There re four inputs, one hidden lyer nd one output in this rchitecture. The input vlues re current, voltge, power nd time (t). The output of NN is between 0 nd 1. 0 represents the fully dischrged bttery nd 1 represents the fully chrged bttery. The t vlue is clculted from (6). It is clculted from the chnge of voltge vlue. For ech rchitecture, the neuron number of the hidden lyer is determined by trying. The number tht gives the best result is chosen old voltge vlue = new voltge vlue, t + 1, t old voltge vlue new voltge vlue, 1. Mtlb Neurl Network Toolbox is used to trin neurl networks. A Mtlb function block is written to pply input vribles to CCNN. The outputs re compred with trgets (6) 28

5 nd success rte is clculted ccording to (7). SR is success rte, REN is right estimtion number, nd TSN is totl smple number in this eqution. A similr model is used to clculte the success rte of type of bttery determintion REN SR. (7) TSN The determintion of the type of btteries results re shown in Tble II nd estimtion of SoC of the btteries results re given in Tble III. 20 % of dt is used s test dt. The estimtion tolernce is ±1 %. For ech tble, it cn be seen tht the best results re obtined from 3 Ω constnt lod experiments. TABLE II. DETERMINING TYPE OF THE BATTERIES. Constnt lod vlue (Ω) 3 Ω 5 Ω 10 Ω Bttery Type Success rte of trining dt (%) Success rte of test dt (%) Averge success rte of trining dt (%) Averge success rte of test dt (%) ,11 87,584 91,552 96,016 86,969 91,47 TABLE III. DETERMINING SOC OF THE BATTERIES. Constnt lod vlue 3 Ω 5 Ω 10 Ω (Ω) Bttery Type Success rte of trining dt (%) Success rte of test dt (%) Averge success rte of trining dt (%) Averge success rte of test dt (%) ,88 98,18 96,506,03 98,628 96,28 After trining networks in Mtlb environment, the weight nd bis vlues of these networks re used in softwre. A function is written to find the type of bttery nd function is written to find the SoC of the bttery. There re three CCNN to determine the type of bttery. When the softwre is strted ccording to constnt lod vlue the CCNN tht will be used to determine the type of bttery is found. 400 seconds lter softwre cn determine the type of bttery. Then it determines the SoC of the bttery. There re fifteen CCNN to determine the SoC of the bttery. This CCNN is determined ccording to constnt lod vlue nd type of bttery. The softwre gives results for online mesurements. The window of online results is presented in Fig. 5. Fig. 5. Anlysis of bttery. 29 VII. CONCLUSIONS Type nd stte of chrge of rechrgeble btteries re estimted in this study vi CCNN. The mximum verge success rte is 96,016 % for estimting the type of bttery nd btteries cn be determined with % success during constnt 3 Ω, 5 Ω nd 10 Ω dischrging conditions The mximum verge success rte is,03 % for estimting the SoC of the bttery. In this study, helthy btteries re used in the experiments. This study cn be extended by tking into ccount the stte of helth of the btteries. In the estimtion, only voltge, current, lod nd power prmeters re used while dischrging the bttery but the bttery temperture nd mbient temperture re lso mesured nd sved to dtbse while chrging nd dischrging the bttery. This dt cn be used in future studies. The experimentl setup nd softwre cn be used for nother type of btteries too. The softwre is flexible nd cn be developed. The dtset obtined from the experiments is suitble to use with other rtificil intelligence techniques to determine the type nd SoC of bttery. This softwre cn be used in bttery mintennce services, bttery tests for bttery mnufcturers nd for determining undefined btteries efficiently. REFERENCES [1] K. L. Mn, K. Wn, T. O. Ting, C. Chen, T. Krilvicius, J. Chng, S. H. Poon, Towrds hybrid pproch to SoC estimtion for smrt Bttery Mngement System (BMS) nd bttery supported Cyber- Physicl Systems (CPS), in Proc. of the 2nd Bltic Congress Future Internet Communictions (BCFIC 2012), 2012, pp DOI: /BCFIC [2] B. Sun, L. Wng, The SOC estimtion of NIMH bttery pck for HEV bsed on BP neurl network, Int. Work. Intell. Syst. Appl. (ISA 2009), 2009, pp DOI: / IWISA [3] G. Dong, X. Zhng, C. Zhng, Z. Chen, A method for stte of energy estimtion of lithium-ion btteries bsed on neurl network model, Energy, vol. 90, pp , DOI: /j.energy [4] A. A. Hussein, Cpcity fde estimtion in electric vehicles Li-ion btteries using rtificil neurl networks, IEEE Energy Convers. Congr. Expo., vol. 51, no. 3, pp , DOI: /TIA [5] Y. Wng, D. Yng, X. Zhng, Z. Chen, Probbility bsed remining cpcity estimtion using dt-driven nd neurl network model, J. Power Sources, vol. 315, pp , DOI: /j.jpowsour [6] A. Uysl, R. Byir, Rel-time condition monitoring nd fult dignosis in switched reluctnce motors with Kohonen neurl network, J. Zhejing Univ. C-Computers Electron., vol. 14, no. 12, pp , [7] X. Dng, L. Yn, K. Xu, X. Wu, H. Jing, H. Sun, Open-circuit voltge-bsed stte of chrge estimtion of lithium-ion bttery using dul neurl network fusion bttery model, Electrochim. Act, vol. 188, pp , DOI: /j.electct [8] P. Singh, R. Vinjmuri, X. Wng, D. Reisner, Fuzzy logic modeling of EIS mesurements on lithium-ion btteries, Electrochim. Act, vol. 51, no. 8 9, pp , DOI: /j.electct [9] H. Yildiz, A. Uysl, R. Byir, Fuzzy logic control of In-Wheel permnent mgnet brushless DC motors, in 4th Int. Conf. Power Eng. Energy Electr. Drives, 2013, pp DOI: /PowerEng [10] A. J. Slkind, C. Fennie, P. Singh, T. Atwter, D. E. Reisner, Determintion of stte-of-chrge nd stte-of-helth of btteries by fuzzy logic methodology, J. Power Sources, vol. 80, no. 1 2, pp , 19. DOI: / S () [11] F. Sun, X. Hu, Y. Zou, S. Li, Adptive unscented Klmn filtering for stte of chrge estimtion of lithium-ion bttery for electric vehicles, Energy, vol. 36, no. 5, pp , DOI: /j.energy [12] T. Okoshi, K. Ymd, T. Hirsw, A. Emori, Bttery condition monitoring (BCM) technologies bout led cid btteries, J. Power

6 Sources, vol. 158, no. 2, pp , DOI: /j.jpowsour [13] S. Lee, J. Kim, J. Lee, B. H. Cho, Stte-of-chrge nd cpcity estimtion of lithium-ion bttery using new open-circuit voltge versus stte-of-chrge, J. Power Sources, vol. 185, no. 2, pp , DOI: /j.jpowsour [14] L. Xu, J. Wng, Q. Chen, Klmn filtering stte of chrge estimtion for bttery mngement system bsed on stochstic fuzzy neurl network bttery model, Energy Convers. Mng., vol. 53, no. 1, pp , DOI: / j.enconmn [15] S. Wng, C. Fernndez, L. Shng, Z. Li, J. Li, Online stte of chrge estimtion for the eril lithium-ion bttery pcks bsed on the improved extended Klmn filter method, J. Energy Storge, vol. 9, pp , DOI: / j.est [16] W. He, D. Hung, D. Feng, The prediction of SOC of lithium btteries nd vried pulse chrge, Int. Conf. Mechtronics Autom. (ICMA 2009), 2009, pp DOI: /ICMA [17] C. C. Chn, E. W. C. Lo, S. Weixing, The vilble cpcity computtion model bsed on rtificil neurl network for led cid btteries in electric vehicles, J. Power Sources, vol. 87, no. 1 2, pp , DOI: / S () [18] C. Bo, B. Zhifeng, C. Binggng, Stte of chrge estimtion bsed on evolutionry neurl network, Energy Convers. Mng., vol. 49, no. 10, pp , DOI: /j.enconmn [19] C. Ci, D. Du, Z. Liu, J. Ge, Stte-of-chrge (SOC) estimtion of high power Ni-MH rechrgeble bttery with rtificil neurl network, in Proc. 9th Int. Conf. Neurl Inf. Process., (ICONIP 2002), 2002, vol. 2, pp DOI: /ICONIP [20] D. Aurbch, Y. Tlyosef, B. Mrkovsky, E. Mrkevich, E. Zinigrd, L. Asrf, J. S. Gnnrj, H.-J. Kim, Design of electrolyte solutions for Li nd Li-ion btteries: review, Electrochim. Act, vol. 50, no. 2 3, pp , DOI: /j.electct [21] J. O. Besenhrd, J. Yng, M. Winter, Will dvnced lithium-lloy nodes hve chnce in lithium-ion btteries?, J. Power Sources, vol. 68, no. 1, pp , 17. DOI: /S (96) [22] S. S. Zhng, A review on electrolyte dditives for lithium-ion btteries, J. Power Sources, vol. 162, no. 2, pp , DOI: /j.jpowsour [23] S. E. Fhlmn, The cscde-correltion lerning rchitecture, Crnegie Mellon Univ., [24] The Cscde Correltion Algorithm. Cornell University. [Online]. Avilble: ppofneurlnetworkcrystllogrphy/neurlnetworkcscdecorreltio n.htm. [25] E. Soylu, R. Byir, Mesurement of electricl conditions of rechrgeble btteries, Mes. Control, vol. 49, no. 2, pp , DOI: /