Individual Identification of Multiple EV–Charger Pairs via Modulated Charging Current Technology

Individual Identification of Multiple EV–Charger Pairs via Modulated Charging Current Technology Narutaka Nomura, Masaki Imanaka, Hiroyuki Baba, Daisuke Kodaira Institute of Systems and Information Engineering University of Tsukuba Tsukuba, Japan [email protected]

Introduction Background Problem Many existing AC chargers lack a bidirectional communication mechanism with connected EVs. e.g., IEC 61851-1 • Aggregators cannot flexibly control EV charging. • EV charging services like Plug and Charge cannot be provided. already implemented • Which EV is connected? • What is the EV’s SoC? • Which charger is connected? can be implemented Aggregator EV-cloud Charger-cloud already implemented AC charger EV Missing link AC Charger EV 1

Previous Research Ref Methods Features [1] A Biometric Second Factor Authentication Sheme IMU gesture biometrics authenticate charging user. smartwatch IoT and blockchain charging/billing IoT charging system enabling user-linked billing. IoT sensors Deep CNN license plate recognition License-plate recognition Cameras to identify vehicles. Sturgess et al., 2023 [2] Martins et al., 2019 [3] Jain et al., 2016 Hardware Limitation Most methods require additional hardware, which typically leads to high installation and maintenance costs. 2

Our Proposed Method Modulated Charging Current Technology (MCCT) 60 sec time interval Charger EV MCCT (Matching application) Charger-cloud Charger command pattern 25, 14, 30, 12, 21 (A) EV-cloud EV 24.4 25 13.4 14 29.0 30 11.6 12 20.3 21 Correlation: 0.99 • MCCT utilizes charging current as information transmission. • MCCT can be implemented using only existing AC charging infrastructure. 3

Limitation of Our Proposed Method • Correlation alone fails when current waveforms are similar. 23.4 A 24 A 14.5 A 15 A 29.0 A 30 A 20.5 A Ch_1 EV_1 21 A 14.5 A 15 A 5.5 A 6A 20.3 A 21 A 11.7 A Ch_2 EV_2 12 A Correlation Ch_1 Ch_2 EV_1 0.9999 0.9999 EV_2 0.9999 0.9999 it becomes impossible to identify which EV is connected to which charger.... Research Purpose Develop a robust method for identifying multiple EV–charger pairs 4

Individual Identification Process Correlation Distance-based metric Normalization Cost matrix Hungarian method Matching result Hadamard product Convert to cost matrix Matching Probability : High : Low 5

Individual Identification Process Correlation Distance-based metric Matching Probability : High : Low 6

Method • Key idea: Combining two metrics to accurately capture similarity. • Motivation: Two metrics complement each other’s weaknesses. Correlation Distance-based metric • It captures overall similarity between EV-side and charger-side current. • It captures local similarity. - 1.0 0 0 1.0 Negative correlation Positive correlation • We use “Euclidean distance” and “Dynamic Time Warping (DTW)”. Distance high similarity low similarity Correlation–Euclidean and Correlation–DTW 7

Individual Identification Process Correlation Distance-based metric -1 to 1 0 < distance Normalization 0 to 1 Matching Probability : High : Low 8

Individual Identification Process Hadamard product 𝑐𝑖𝑗 = 𝑎𝑖𝑗 ∙ 𝑏𝑖𝑗 𝑐13 = 𝑎13 ∙ 𝑏13 = 0.95 ∙ 0.32 = 0.30 Matching Probability : High : Low 9

Individual Identification Process Cost matrix 𝑑13 = 1 − 0.30 = 0.70 Combination matrix 𝑑𝑖𝑗 = 1 − 𝑐𝑖𝑗 Convert to cost matrix Matching Probability : High : Low 10

Individual Identification Process Cost matrix Hungarian method Matching result Matching Probability : High : Low 11

Hungarian Method • Key idea: Globally optimal one-to-one matching that avoids getting trapped in local optima Hungarian method : An algorithm that finds the optimal matching by repeatedly subtracting the minimum values from the rows and columns of the cost matrix. Cost matrix minimum value of each row stop when a one-to-one matching is possible with only zero element. Subtract the minimum value from each row 12

Simulation dataset • We generated simulation data and compared it with the measured data. 6, 18, 30, 12, 24 A Command value Charger current (measured) Charger current (generated) EV current (measured) EV current (generated) Steady state Experimental resolution time Transient state current Charger 5 sec 0.1 A EV 60 sec 1A ➢ Steady state: Generated data was very close to measured data. ➢ Transient state: Generated slopes are slightly smoother. 13

Simulation result • KeyScalability with the Number of EV–Charger Pairs Correlation–Euclidean (transient) Correlation–DTW (transient) Correlation (transient) Correlation–Euclidean (steady) Correlation–DTW (steady) Correlation (steady) State measured when current value is changing State measured when current value is stable ➢ Steady state: All metrics remain above 99% accuracy across all scales. ➢ Transient state: As the number of pairs increases, overall accuracy drops. 14

Conclusion Summary • We proposed two combination metrics. • We obtain the optimal matching using Hungarian method. • We evaluated scalability as the number of EV–charger pairs increased. Conclusion • When charging currents are measured in steady state, accuracy remains above 99% for all pairs. 15

Thank you for your kind attention.

Appendix

EV charging system modulated 5 time (6–30 A) at 60 sec interval (i) Charging Current Command 12 A BMS (Battery management system) OBC (On board charger) Console 12 A 11.6 A EV cloud Charger cloud measured every 5 sec at 0.1 A measured every 60 sec at 1 A (iv) Measurement Charger (ii) Control Pilot (iii) Control OBC BMS EV 18

Individual Identification Process • Let three EV-charger pairs start charging at the same time. Ch_1 Ch_2 Ch_3 12, 30, 21, 9, 15 (A) 18, 9, 27, 12, 15 (A) 6, 18, 24, 30, 15 (A) EV_1 EV_2 EV_3 19

Distance-based metric Euclidean distance : compares values at the same timestamps—simple and fast, but sensitive to time shifts. Dynamic Time Warping (DTW) : aligns in time by stretching and compressing—robust to delays and ramp-rate differences. Correlation–Euclidean and Correlation–DTW 20