Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion-Based Forecasting for Uncertainty-Aware Model Predictive Control

Stelios Zarifis, Ioannis Kordonis, Petros Maragos

TL;DR

This work tackles uncertainty in partially observable systems by integrating diffusion-based probabilistic forecasts with Model Predictive Control (MPC). It introduces Diffusion-Informed MPC (D-I MPC), leveraging TimeGrad/DDPM forecasts to guide both deterministic and stochastic MPC in a partially observed setting, demonstrated on Battery Energy Storage System (BESS) energy arbitrage in the NYISO market. The results show that diffusion-informed planning substantially outperforms classical forecasters and model-free RL baselines, with near-oracle performance in stochastic MPC and robust improvements across metrics. The findings highlight diffusion models as a powerful forecasting backbone for uncertainty-aware control and point to promising extensions such as scenario-tree MPC and hybrid model-based/model-free strategies.

Abstract

We propose Diffusion-Informed Model Predictive Control (D-I MPC), a generic framework for uncertainty-aware prediction and decision-making in partially observable stochastic systems by integrating diffusion-based time series forecasting models in Model Predictive Control algorithms. In our approach, a diffusion-based time series forecasting model is used to probabilistically estimate the evolution of the system's stochastic components. These forecasts are then incorporated into MPC algorithms to estimate future trajectories and optimize action selection under the uncertainty of the future. We evaluate the framework on the task of energy arbitrage, where a Battery Energy Storage System participates in the day-ahead electricity market of the New York state. Experimental results indicate that our model-based approach with a diffusion-based forecaster significantly outperforms both implementations with classical forecasting methods and model-free reinforcement learning baselines.

Diffusion-Based Forecasting for Uncertainty-Aware Model Predictive Control

TL;DR

This work tackles uncertainty in partially observable systems by integrating diffusion-based probabilistic forecasts with Model Predictive Control (MPC). It introduces Diffusion-Informed MPC (D-I MPC), leveraging TimeGrad/DDPM forecasts to guide both deterministic and stochastic MPC in a partially observed setting, demonstrated on Battery Energy Storage System (BESS) energy arbitrage in the NYISO market. The results show that diffusion-informed planning substantially outperforms classical forecasters and model-free RL baselines, with near-oracle performance in stochastic MPC and robust improvements across metrics. The findings highlight diffusion models as a powerful forecasting backbone for uncertainty-aware control and point to promising extensions such as scenario-tree MPC and hybrid model-based/model-free strategies.

Abstract

We propose Diffusion-Informed Model Predictive Control (D-I MPC), a generic framework for uncertainty-aware prediction and decision-making in partially observable stochastic systems by integrating diffusion-based time series forecasting models in Model Predictive Control algorithms. In our approach, a diffusion-based time series forecasting model is used to probabilistically estimate the evolution of the system's stochastic components. These forecasts are then incorporated into MPC algorithms to estimate future trajectories and optimize action selection under the uncertainty of the future. We evaluate the framework on the task of energy arbitrage, where a Battery Energy Storage System participates in the day-ahead electricity market of the New York state. Experimental results indicate that our model-based approach with a diffusion-based forecaster significantly outperforms both implementations with classical forecasting methods and model-free reinforcement learning baselines.

Paper Structure

This paper contains 19 sections, 15 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Diffusion Models
  4. Denoising Diffusion Probabilistic Models
  5. TimeGrad Model
  6. Reinforcement Learning
  7. Methodology
  8. Problem Definition
  9. Probabilistic Forecasting with Diffusion Models
  10. Model Predictive Control with Diffusion Models
  11. Stochastic Model Predictive Control with Diffusion Models
  12. Results
  13. Experimental Setup
  14. Model Comparison and Observations
  15. Visualization of the Method
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Strategy planned by the MPC Optimizer. The dotted line represents the forecasted prices, the black line represents the actual electricity prices, the blue line represents the battery's SoC, and the cyan shaded areas represent the 90% and 50% quantiles of forecast distributions.
  • Figure 2: Strategy planned by the SMPC Optimizer. The cyan lines represent the generated trajectories, the black line represents the actual prices (published in the future) and the blue line represents the battery's SoC.
  • Figure 3: Anticipated and actual rewards (3-day windows) for various models, sorted by how closely the anticipated matches the actual reward.