Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Algorithmic Control Improves Residential Building Energy and EV Management when PV Capacity is High but Battery Capacity is Low

Lennart Ullner, Alona Zharova, Felix Creutzig

TL;DR

The paper addresses how prosumer households can optimize energy and EV charging when PV capacity is high but battery capacity is limited, using real-world data from 90 DACH households and comparing DRL (DDPG) with RB and MPC. It demonstrates that DRL can improve energy management and reduce costs primarily when there is meaningful optimization potential, such as frequent EV connections and PV surplus, while simple RB rules often suffice for PV-surplus BESS charging under fixed pricing. A detailed analysis of nine households shows two EV charging behavior clusters (self-optimizers and overnight chargers) and reveals that DRL’s advantages depend on the data’s richness, particularly PV surplus windows and EV-transaction patterns; a synthetic dataset with higher potential confirms DRL can realize substantial gains. The study concludes that DRL offers practical benefits in complex, dynamic HEM contexts but may provide limited improvements in scenarios with low optimization potential, guiding when to deploy DRL versus simpler forecasting-based controls for residential energy systems and grid decarbonization.

Abstract

Efficient energy management in prosumer households is key to alleviating grid stress in an energy transition marked by electric vehicles (EV), renewable energies and battery storage. However, it is unclear how households optimize prosumer EV charging. Here we study real-world data from 90 households on fixed-rate electricity tariffs in German-speaking countries to investigate the potential of Deep Reinforcement Learning (DRL) and other control approaches (Rule-Based, Model Predictive Control) to manage the dynamic and uncertain environment of Home Energy Management (HEM) and optimize household charging patterns. The DRL agent efficiently aligns charging of EV and battery storage with photovoltaic (PV) surplus. We find that frequent EV charging transactions, early EV connections and PV surplus increase optimization potential. A detailed analysis of nine households (1 hour resolution, 1 year) demonstrates that high battery capacity facilitates self optimization; in this case further algorithmic control shows little value. In cases with relatively low battery capacity, algorithmic control with DRL improves energy management and cost savings by a relevant margin. This result is further corroborated by our simulation of a synthetic household. We conclude that prosumer households with optimization potential would profit from DRL, thus benefiting also the full electricity system and its decarbonization.

Algorithmic Control Improves Residential Building Energy and EV Management when PV Capacity is High but Battery Capacity is Low

TL;DR

The paper addresses how prosumer households can optimize energy and EV charging when PV capacity is high but battery capacity is limited, using real-world data from 90 DACH households and comparing DRL (DDPG) with RB and MPC. It demonstrates that DRL can improve energy management and reduce costs primarily when there is meaningful optimization potential, such as frequent EV connections and PV surplus, while simple RB rules often suffice for PV-surplus BESS charging under fixed pricing. A detailed analysis of nine households shows two EV charging behavior clusters (self-optimizers and overnight chargers) and reveals that DRL’s advantages depend on the data’s richness, particularly PV surplus windows and EV-transaction patterns; a synthetic dataset with higher potential confirms DRL can realize substantial gains. The study concludes that DRL offers practical benefits in complex, dynamic HEM contexts but may provide limited improvements in scenarios with low optimization potential, guiding when to deploy DRL versus simpler forecasting-based controls for residential energy systems and grid decarbonization.

Abstract

Efficient energy management in prosumer households is key to alleviating grid stress in an energy transition marked by electric vehicles (EV), renewable energies and battery storage. However, it is unclear how households optimize prosumer EV charging. Here we study real-world data from 90 households on fixed-rate electricity tariffs in German-speaking countries to investigate the potential of Deep Reinforcement Learning (DRL) and other control approaches (Rule-Based, Model Predictive Control) to manage the dynamic and uncertain environment of Home Energy Management (HEM) and optimize household charging patterns. The DRL agent efficiently aligns charging of EV and battery storage with photovoltaic (PV) surplus. We find that frequent EV charging transactions, early EV connections and PV surplus increase optimization potential. A detailed analysis of nine households (1 hour resolution, 1 year) demonstrates that high battery capacity facilitates self optimization; in this case further algorithmic control shows little value. In cases with relatively low battery capacity, algorithmic control with DRL improves energy management and cost savings by a relevant margin. This result is further corroborated by our simulation of a synthetic household. We conclude that prosumer households with optimization potential would profit from DRL, thus benefiting also the full electricity system and its decarbonization.

Paper Structure

This paper contains 45 sections, 11 equations, 25 figures, 15 tables.

Table of Contents

  1. Introduction
  2. State-of-the-Art in prosumer household energy management
  3. Optimization behavior of prosumer households
  4. Optimization Approaches
  5. Deep Reinforcement Learning in HEM
  6. System Architecture
  7. Optimisation Task
  8. Reinforcement Learning
  9. States
  10. Actions
  11. Transition
  12. Reward
  13. Value Function
  14. Experimental Setup
  15. Data
  16. ...and 30 more sections

Figures (25)

  • Figure 1: System architecture
  • Figure 2: Markov Decision Process Sutton.1997
  • Figure 3: Household 01 on March 08 with real-world data, RWD (panel 1), RBPM (panel 2), MPC (panel 3), and DRL (panel 4). The stacked curves with filled-in areas represent available electricity (PV generation and BESS discharging), while the stacked bars represent electrical demands (household demand, EV charging, and BESS charging).
  • Figure 4: Start and finish times of EV charging transactions for cluster of overnight chargers (panel 1) and cluster of self-optimizers (panel 2)
  • Figure 5: PV peak relative to yearly total household demand. The gray dashed lines indicate industry guidelines for PV system sizing, with factors of 1.5 and 2.5. The dots represent 90 households, with dark-colored dots highlighting the selected nine households.
  • ...and 20 more figures