Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantification of Residential Flexibility Potential using Global Forecasting Models

Lorenzo Nespoli, Vasco Medici

TL;DR

The paper tackles quantifying residential flexibility potential under deferrable loads without historical data by training a non-parametric global forecaster (LightGBM) on simulated device responses. It uses the metamodel as a surrogate in optimization to minimize day-ahead costs and peak tariffs, and to analyze rebound and comfort constraints, demonstrating that accuracy is sufficient to bypass full simulations in closed-loop emulations. Key contributions include energy-imbalance aware training, rebound characterization, dynamic grouping, and open-/closed-loop emulations that enable rapid policy testing. The results show practical potential for scalable, data-light evaluation and optimization of demand-side flexibility for DSOs.

Abstract

This paper proposes a general and practical approach to estimate the economic benefits of optimally controlling deferrable loads in a Distribution System Operator's (DSO) grid, without relying on historical observations. We achieve this by learning the simulated response of flexible loads to random control signals, using a non-parametric global forecasting model. An optimal control policy is found by including the latter in an optimization problem. We apply this method to electric water heaters and heat pumps operated through ripple control and show how flexibility, including rebound effects, can be characterized and controlled. Finally, we show that the forecaster's accuracy is sufficient to completely bypass the simulations and directly use the forecaster to estimate the economic benefit of flexibility control.

Quantification of Residential Flexibility Potential using Global Forecasting Models

TL;DR

The paper tackles quantifying residential flexibility potential under deferrable loads without historical data by training a non-parametric global forecaster (LightGBM) on simulated device responses. It uses the metamodel as a surrogate in optimization to minimize day-ahead costs and peak tariffs, and to analyze rebound and comfort constraints, demonstrating that accuracy is sufficient to bypass full simulations in closed-loop emulations. Key contributions include energy-imbalance aware training, rebound characterization, dynamic grouping, and open-/closed-loop emulations that enable rapid policy testing. The results show practical potential for scalable, data-light evaluation and optimization of demand-side flexibility for DSOs.

Abstract

This paper proposes a general and practical approach to estimate the economic benefits of optimally controlling deferrable loads in a Distribution System Operator's (DSO) grid, without relying on historical observations. We achieve this by learning the simulated response of flexible loads to random control signals, using a non-parametric global forecasting model. An optimal control policy is found by including the latter in an optimization problem. We apply this method to electric water heaters and heat pumps operated through ripple control and show how flexibility, including rebound effects, can be characterized and controlled. Finally, we show that the forecaster's accuracy is sufficient to completely bypass the simulations and directly use the forecaster to estimate the economic benefit of flexibility control.
Paper Structure (23 sections, 23 equations, 10 figures, 7 tables)

This paper contains 23 sections, 23 equations, 10 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Contributions
  4. Problem statement and system description
  5. Global forecasting modes for flexibility simulation and control
  6. Dataset generation
  7. Model description
  8. Ablation studies
  9. Sampling schemes
  10. Energy unbalance awareness
  11. Characterization of the rebound effect
  12. Using metamodels for optimal flexibility control
  13. Controlling multiple groups
  14. Ensuring comfort for the end users
  15. Using metamodels for closed loop emulations
  16. ...and 8 more sections

Figures (10)

  • Figure 1: Left: a random sample of daily scenarios for the force-off signal. Center: ratio of active signals for a given timestep of the day. Right: distribution of the number of active timesteps among all possible scenarios.
  • Figure 2: Sampling strategies for building the final training set. Left: the total number of controllable devices is increased linearly, picking randomly between households with an HP or an EH. Left: the number of controllable devices is increased by independently co-varying the number of HPs and EHs.
  • Figure 3: Random example of day-ahead metamodel's forecasts, for different numbers of HPs and EHs, where the force off was activated at least once, for the energy-aware metamodel trained using the grid sampling strategy
  • Figure 4: Performances for the four tested metamodels, in terms of nMAE as a function of the step ahead.
  • Figure 5: Left: cumulative distributions of the relative energy imbalance for different models. Right: empirical cumulative density functions of absolute relative energy imbalance for different models.
  • ...and 5 more figures