Continual Refoircement Learning Using Real-World Data for Intelligent Prediction of SOC Consumption in Electric Vehicles

Juan P. Ortiz, German P. Ayabaca, Angel R. Cardenas, Diego Cabrera, Juan D. Valladolid

Research output: Contribution to journalArticlepeer-review

6 Scopus citations

Abstract

The accelerated migration towards electric vehicles (EV) presents several problems to solve. The main aspect is the management and prediction of the state of charge (SOC) in real long-range routes of different variations in altitude for a more efficient energy consumption and vehicle recharge plan. This paper presents the implementation of a new algorithm for SOC estimation based on continuous learning and meta-experience replay (MER) with reservoir sample. It combines the reptile meta-learning algorithm with the experience replay technique for stabilizing the reinforcement learning. The proposed algorithm considers several important factors for the prediction of the SOC in EV such as: speed, travel time, route altimetry, consumed battery capacity, regenerated battery capacity. A modified principal components analysis is used to reduce the dimensionality of the route altimetry data. The experimental results show an efficient estimation of the SOC values and a convergent increase in knowledge while the vehicle travels the routes.

Original languageEnglish
Pages (from-to)624-633
Number of pages10
JournalIeee Latin America Transactions
Volume20
Issue number4
DOIs
StatePublished - 1 Apr 2022

Bibliographical note

Publisher Copyright:
© 2003-2012 IEEE.

Keywords

  • Electric vehicles (EV)
  • Meta-experience replay (MER
  • Neural networks (NN)
  • Principal component analysis (PCA)
  • Reinforcementlearning (RL)
  • State of charge (SOC)

Fingerprint

Dive into the research topics of 'Continual Refoircement Learning Using Real-World Data for Intelligent Prediction of SOC Consumption in Electric Vehicles'. Together they form a unique fingerprint.

Cite this