Impact of Vehicle-to-Grid (V2G) on Battery Life

Transcription

1 Impact of Vehicle-to-Grid (V2G) on Battery Life The Importance of Accurate Models David Howey, Jorn Reniers, Grietus Mulder, Sina Ober-Blöbaum Department of Engineering Science, University of Oxford EnergyVille, VITO, Belgium FPC 2018

2 Contents Introduction Battery models Optimal control Conclusions V2G, and batteries in power systems How well do 3 often-used models predict battery degradation? How well do these model perform in an optimal control application? 1

3 Contents Introduction Battery models Optimal control Conclusions V2G, and batteries in power systems How well do 3 often-used models predict battery degradation? How well do these model perform in an optimal control application? 2

4 Introduction: Vehicle-to-grid (V2G) Already 3million+ EVs globally, 10million in UK by EV Charging creates new business opportunities (e.g. ToU, V2G etc.) V2G = grid scale battery. Many grid services possible, most support main aim of all power systems: balancing supply and demand. V2G already starting to happen e.g. Ovo Energy and Nissan 2 Key question: battery health? 1. Aurora Energy Research 2018, Photo by CEphoto, Uwe Aranas /, via Wikimedia Commons 3

5 Introduction: Batteries in power systems Batteries are flexible and fast -> can solve many problems Source: EC JRC scientific and policy reports, Assessing storage value in electricity markets,

6 Introduction: Economic optimisation Liberalisation -> markets -> profit-driven Techno-economic studies on how to use batteries (maximise profit) Use (very) simple battery models Battery degradation? But, better models might increase lifetime -> increase total profit? Ref: Patsios et at., 2016 J. Energy Storage Fig. 13. Battery degradation for 24 h of operation for different combinations of floating SoC and maximum SoC limits. 5

7 Introduction: Li-ion battery degradation Various degradation mechanisms in Li-ion batteries Ref. Birkl et al., 2017 J. Power Sources 6

8 Introduction: Li-ion battery degradation Cause Degradation Mechanism Degradation Mode Effect Time SEI growth High temperature SEI decomposition Electrolyte decomposition Loss of lithium inventory High V cell /SOC cell Binder decomposition Capacity fade Current load Graphite exfoliation Structural disordering Loss of active anode material Low temperature Lithium plating/dendrite formation Power fade Stoichiometry Mechanical stress Loss of electric contact Electrode particle cracking Transition metal dissolution Loss of active cathode material Low V cell /SOC cell Corrosion of current collectors Processes leading to battery degradation Ref. Birkl et al., 2017 J. Power Sources 7

9 Contents Introduction Battery models Optimal control Conclusions V2G, and batteries in power systems How well do 3 often-used models predict battery degradation? How well do these model perform in an optimal control application? 8

10 Battery models 3 battery models with 3 degradation models Bucket model Equivalent circuit model Single particle model Compare degradation predictions with degradation experiments Bucket model Equivalent circuit model Single particle model 9

11 Battery models: bucket model Bucket of energy with only state of charge (z) 10

12 Battery models: bucket model Bucket of energy with only state of charge (z) Lost capacity (! "#$%,'( ) linear function of energy throughput Linear model -> easy optimisation Accuracy? 11

13 Battery models: equivalent circuit model Equivalent circuit with State of charge (z) Parallel current (! " ) 12

14 Battery models: equivalent circuit model Equivalent circuit with State of charge (z) Parallel current (! " ) Lost capacity as fit through empirical data set Ref. Schmalstieg et al., 2014, J. Power Sources Nonlinear voltage and lost capacity More accurate and reasonable complexity 13

15 Battery models: single particle model Single particle model (SPM) with Arrhenius temperature dependency Butler-Volmer kinetics Thermal PDE Ohmic heat, reaction heat, entropy change, convective cooling Fick s law of diffusion SEI layer growth Refs: Subtract li-ions from ve Christensen et al., 2005, J. Electrochemical Society Ning et al., 2004, J. Electrochemical Society Guo et al., 2011, J. Electrochemical Society 14

16 Battery models: Degradation data (Mat4Bat project) Degradation experiments (markers) degradation simulation using the bucket model (lines) Calendar ageing at various State-of-Charge and temperatures Cycle ageing between various SoC windows and with various charging currents (T=45 C) 15

17 Battery models degradation summary Compared with 2 data sets (both using NMC cells) Mat4Bat (parametrisation SPM), shown on previous slide Schmalstieg (parametrisation ECM) Extrapolation of cycle ageing? Root mean square error [%] Calendar ageing Cycle ageing Mat4Bat [%] Schmalstieg [%] Mat4Bat [%] Schmalstieg [%] Bucket n/a n/a Equivalent circuit Single particle

18 Contents Introduction Battery models Optimal control Conclusions V2G, and batteries in power systems How well do 3 often-used models predict battery degradation? How well do these model perform in an optimal control application? 17

19 Optimal control framework Wholesale arbitrage à revenue, R Battery degradation à cost, C Battery state-space model, f Constraints, g Maximise either profit or revenue Day-ahead wholesale price in 2014 in Belgium 18

20 Results: Battery state of charge in 1 st week 3 battery models Maximise: R (revenue, solid) P (profit, dashed) 19

21 Results: Battery state of charge in 1 st week 3 battery models Maximise R (revenue, solid) P (profit, dashed) Access all capacity? BM 65% ECM 71% SPM 94% 20

22 Results: Battery state of charge in 1 st week 3 battery models Maximise R (revenue, solid) P (profit, dashed) Access all capacity? Decreased utilisation? 21

23 Results: Revenue & cost for whole year 3 battery models Maximise R (revenue, solid) P (profit, dashed) 22

24 Results: Revenue & cost for whole year 3 battery models Maximise R (revenue, solid) P (profit, dashed) Access all capacity? BM 65% ->R=76 ECM 71% ->R=82 SPM 94% ->R=108 23

25 Results: Revenue & cost for whole year 3 battery models Maximise R (revenue, solid) P (profit, dashed) Access all capacity? Decreased utilisation? BM: R 16 C 29 P

26 Introduction Battery models Optimal control Conclusions Results: Revenue & cost for whole year 3 battery models Maximise R (revenue, solid) P (profit, dashed) Access all capacity? Decreased utilisation? 25

27 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? How well do these model perform in an optimal control application? 26

28 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? Calendar ageing: quite good (but not bucket model) Calendar ageing Cycle ageing Mat4Bat [%] Schmalstieg [%] Mat4Bat [%] Schmalstieg [%] Bucket n/a n/a Equivalent circuit Single particle Root mean square error [%] of simulations vs 2 experimental data sets How well do these model perform in an optimal control application? 27

29 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? Calendar ageing: quite good (but not bucket model) Cycle ageing can be problematic more difficult to generalise. Calendar ageing Cycle ageing Mat4Bat [%] Schmalstieg [%] Mat4Bat [%] Schmalstieg [%] Bucket n/a n/a Equivalent circuit Single particle Root mean square error [%] of simulations vs 2 experimental data sets How well do these model perform in an optimal control application? 28

30 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? How well do these model perform in an optimal control application? Top-bottom: Access more capacity à increase revenue Max Revenue Max Profit Revenue Cost profit Lost capacity [%] Revenue Cost profit Lost capacity [%] Bucket Equivalent circuit Single particle Economic parameters and relative lost capacity at the end of the year 29

31 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? How well do these model perform in an optimal control application? Top-bottom: Access more capacity à increase revenue Left right: Reduce utilisation à decrease cost Max Revenue Max Profit Revenue Cost profit Lost capacity [%] Revenue Cost profit Lost capacity [%] Bucket Equivalent circuit Single particle Economic parameters and relative lost capacity at the end of the year 30

32 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? How well do these model perform in an optimal control application? Top-bottom: Access more capacity à increase revenue Left right: Reduce utilisation à decrease cost Both trends increase total profit Max Revenue Max Profit Revenue Cost profit Lost capacity [%] Revenue Cost profit Lost capacity [%] Bucket Equivalent circuit Single particle Economic parameters and relative lost capacity at the end of the year 31

33 Conclusions Introduction Battery models Optimal control Conclusions How well do 3 typical models predict battery degradation? How well do these model perform in an optimal control application? Top-bottom: Access more capacity à increase revenue Left right: Reduce utilisation à decrease cost Both trends increase total profit Total simulated profit over the year increased by 175% Techno-economic assessments underestimate potential NB: Vehicle batteries are not grid batteries! Cost per cycle much higher. Max Revenue Max Profit Revenue Cost profit Lost capacity [%] Revenue Cost profit Lost capacity [%] Bucket Equivalent circuit Single particle Economic parameters and relative lost capacity at the end of the year 32

34 Thanks! Introduction Battery models Optimal control Conclusions Paper: Improving optimal control of grid-connected lithium-ion batteries through more accurate battery and degradation modelling, J Power Sources 2018 Matlab code of the Single Particle Model on GitHub: Degradation data from MAT4BAT project ( funded by the EU Seventh Framework Programme (FP7/ ) under grant agreement n david.howey@eng.ox.ac.uk 33