Controlling the Charging of Electric Vehicles with Neural Networks

Martin PilátIJCNN 2018 · DOI · OpenReview

Abstract

We propose and evaluate controllers for the coordination of the charging of electric vehicles. The controllers are based on neural networks and are completely de-centralized, in the sense that the charging current is completely decided by the controller itself. One of the versions of the controllers does not require any outside communication at all. We test controllers based on two different architectures of neural networks - the feed-forward networks and the echo state networks. The networks are optimized by either an evolutionary algorithm (CMA-ES) or by a gradient-based method. The results of the different architectures and the different optimization algorithms are compared in a realistic scenario. We show that the controllers are able to charge the cars while keeping the peak consumptions almost the same as when no charging is performed. Moreover, the controllers fill the valleys of the consumption thus reducing the difference between the maximum and minimum consumption in the grid.

Introduction

As electric vehicle (EV) adoption grows, the "simultaneous charging" problem looms: if millions of drivers plug in their cars immediately upon returning home, residential power demand could double, creating massive peaks that threaten grid stability. Existing solutions are often centralized, requiring a utility company to monitor and control every charger. This presents major hurdles regarding infrastructure requirements and user privacy. This paper explores the creation of "smart chargers" that are computationally light enough for cheap embedded systems and can operate either entirely locally or with minimal one-way broadcasts from the grid.

Methodology

The authors evaluate two neural network architectures: Feed-forward Neural Networks (NN), which are simple reactive models, and Echo State Networks (ESN), a type of reservoir computing with internal memory to factor in temporal trends. These were tested under two communication levels: Type H (Household), which is completely decentralized, and Type A (Aware), which receives a one-way broadcast of total grid consumption.

Optimization was performed using either CMA-ES (a derivative-free evolutionary strategy) or a gradient-based method. To ensure realistic results, the models were trained and tested using data from Pecan Street’s Dataport (real electricity consumption from 74 households) and the National Household Transportation Survey (to synthesize realistic car arrival and departure times).

Results and Discussion

The study found that decentralized coordination is highly effective. Surprisingly, "Type H" controllers (zero communication) performed nearly as well as "Type A" controllers, suggesting that local household consumption is a strong proxy for grid-wide behavior. ESNs generally outperformed simple feed-forward NNs due to their internal state allowing for smoother responses to baseline power usage.

Comparison of grid load over time for different control strategies versus the baseline (no charge).

As shown in the time-series data, the controllers successfully shifted charging to nighttime and midday "valleys," effectively flattening the total demand curve. CMA-ES consistently found better parameters than the gradient-based approach, which tended to be too stable and failed to react quickly enough to fill valleys.

Performance metrics for various controller and optimizer combinations.

Quantitatively, the best NN controllers kept the peak load around 140 kW, only slightly higher than the baseline peak without any EVs (135 kW), whereas an unmanaged "max charge" scenario hit 194.1 kW. The standard deviation of the grid load was reduced from 20.1 (baseline) to 15.2 using the ESN-A+CMA model.

Conclusion

Decentralized neural network controllers offer a viable, privacy-preserving solution for EV charging coordination. These models are lean enough to run on low-cost microcontrollers found in standard charging equipment. Future work will explore supervised learning to mimic optimal offline schedules and investigate the impact of real-world user override behavior.

Summary updated

Abstract

We propose and evaluate controllers for the coordination of the charging of electric vehicles. The controllers are based on neural networks and are completely de-centralized, in the sense that the charging current is completely decided by the controller itself. One of the versions of the controllers does not require any outside communication at all. We test controllers based on two different architectures of neural networks-the feed-forward networks and the echo state networks. The networks are optimized by either an evolutionary algorithm (CMA-ES) or by a gradient-based method. The results of the different architectures and the different optimization algorithms are compared in a realistic scenario. We show that the controllers are able to charge the cars while keeping the peak consumptions almost the same as when no charging is performed. Moreover, the controllers fill the valleys of the consumption thus reducing the difference between the maximum and minimum consumption in the grid.
BibTeX
@inproceedings{pilat2018controlling,
  author = {Martin Pilát},
  title = {{Controlling the Charging of Electric Vehicles with Neural Networks}},
  booktitle = {IJCNN},
  year = {2018},
  doi = {10.1109/ijcnn.2018.8489027},
}
Figure