# Solar Power driven EV Charging Optimization with Deep Reinforcement Learning

**Authors:** Stavros Sykiotis, Christoforos Menos-Aikateriniadis, Anastasios Doulamis, Nikolaos Doulamis, Pavlos S. Georgilakis  
**arXiv:** [2211.09479](https://arxiv.org/abs/2211.09479)  
**Date:** 2022-11-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.09479v1/

## Key points

- Decarbonizing the power sector is crucial for a sustainable future
- Integration of decentralized energy resources like EVs and solar PV is necessary for this transition
- Potential risk of bottlenecks in power distribution networks due to integration
- Real Time-of-Use tariffs can incentivize end-users to optimally shift EV charging load during high solar PV generation hours
- Deep Reinforcement Learning (DRL) used with historical measurements from Pecan Street dataset to shape flexibility potential reward that describes end-user charging preferences
- Proposed DQN EV optimal charging policy can reduce electricity bills by an average of 11.5% while achieving an average utilization of solar power at 88.4%
- Study cites relevant sources such as Dataport - Pecan Street Inc., City of Austin Fiscal Year 2018 Electric Tariff, and Idaho National Laboratory's Steady State Vehicle Charging Fact Sheet for further context and analysis
- Project has received funding from various organizations
- Optimizing EV charging is important for efficient use of renewable energy sources in the power sector

## Summary

The transition towards a sustainable future heavily relies on decarbonizing the power sector, which involves integrating decentralized energy resources like Electric Vehicles (EVs) and solar photovoltaic systems (PV) into residential power systems. To address the potential risk of bottlenecks in power distribution networks posed by this integration, a recent paper proposes optimizing domestic EV charging while prioritizing clean, solar energy consumption using Real Time-of-Use tariffs as a price-based Demand Response mechanism that incentivizes end-users to optimally shift EV charging load during hours of high solar PV generation. The proposed approach uses Deep Reinforcement Learning (DRL) with historical measurements from the Pecan Street dataset to shape a flexibility potential reward that describes end-user charging preferences. Experimental results demonstrate that the proposed DQN EV optimal charging policy can reduce electricity bills by an average of 11.5% while achieving an average utilization of solar power at 88.4%. The study also cites relevant sources such as Dataport - Pecan Street Inc., City of Austin Fiscal Year 2018 Electric Tariff, and Idaho National Laboratory's Steady State Vehicle Charging Fact Sheet for further context and analysis. This project has received funding from various organizations and highlights the importance of optimizing EV charging to ensure efficient use of renewable energy sources in the power sector.