Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Probabilistic Forecasting of Real-Time Electricity Market Signals via Interpretable Generative AI

Xinyi Wang, Qing Zhao, Lang Tong

TL;DR

The paper tackles probabilistic forecasting of real-time electricity-market signals with a nonparametric, interpretable approach. It introduces WIAE-GPF, a Weak Innovation AutoEncoder-based Generative Probabilistic Forecasting framework that yields samples from the conditional distribution of future market signals by leveraging a Wiener-Kallianpur innovation representation. The authors prove Bayesian sufficiency and structural convergence under ideal training, and validate the method against multiple baselines on real IS0 data, showing superior probabilistic performance and competitive point forecasts. The work advances practical market operations by providing an interpretable, theoretically grounded alternative to black-box generative models, with potential extensions to nonstationary regimes and regime-switching analyses.

Abstract

This paper introduces a generative AI approach to probabilistic forecasting of real-time electricity market signals, including locational marginal prices, interregional price spreads, and demand-supply imbalances. We present WIAE-GPF, a Weak Innovation AutoEncoder-based Generative Probabilistic Forecasting architecture that generates future samples of multivariate time series. Unlike traditional black-box models, WIAE-GPF offers interpretability through the Wiener-Kallianpur innovation representation for nonparametric time series, making it a nonparametric generalization of the Wiener/Kalman filter-based forecasting. A novel learning algorithm with structural convergence guarantees is proposed, ensuring that, under ideal training conditions, the generated forecast samples match the ground truth conditional probability distribution. Extensive tests using publicly available data from U.S. independent system operators under various point and probabilistic forecasting metrics demonstrate that WIAE-GPF consistently outperforms classical methods and cutting-edge machine learning techniques.

Probabilistic Forecasting of Real-Time Electricity Market Signals via Interpretable Generative AI

TL;DR

The paper tackles probabilistic forecasting of real-time electricity-market signals with a nonparametric, interpretable approach. It introduces WIAE-GPF, a Weak Innovation AutoEncoder-based Generative Probabilistic Forecasting framework that yields samples from the conditional distribution of future market signals by leveraging a Wiener-Kallianpur innovation representation. The authors prove Bayesian sufficiency and structural convergence under ideal training, and validate the method against multiple baselines on real IS0 data, showing superior probabilistic performance and competitive point forecasts. The work advances practical market operations by providing an interpretable, theoretically grounded alternative to black-box generative models, with potential extensions to nonstationary regimes and regime-switching analyses.

Abstract

This paper introduces a generative AI approach to probabilistic forecasting of real-time electricity market signals, including locational marginal prices, interregional price spreads, and demand-supply imbalances. We present WIAE-GPF, a Weak Innovation AutoEncoder-based Generative Probabilistic Forecasting architecture that generates future samples of multivariate time series. Unlike traditional black-box models, WIAE-GPF offers interpretability through the Wiener-Kallianpur innovation representation for nonparametric time series, making it a nonparametric generalization of the Wiener/Kalman filter-based forecasting. A novel learning algorithm with structural convergence guarantees is proposed, ensuring that, under ideal training conditions, the generated forecast samples match the ground truth conditional probability distribution. Extensive tests using publicly available data from U.S. independent system operators under various point and probabilistic forecasting metrics demonstrate that WIAE-GPF consistently outperforms classical methods and cutting-edge machine learning techniques.
Paper Structure (27 sections, 3 theorems, 26 equations, 8 figures, 6 tables)

This paper contains 27 sections, 3 theorems, 26 equations, 8 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Summary of Contributions
  4. Organization and Notations
  5. Innovation Representation Learning
  6. Strong and Weak Innovation Representations
  7. Innovation Representation Learning
  8. WIAE Learning
  9. WIAE-GPF and its Properties
  10. Structure of WIAE-GPF
  11. Structural Convergence of WIAE-GPF
  12. From GPF to Point and Quantile Forecasting
  13. Evaluation Metrics
  14. WIAE-GPF for Market Operations
  15. Baseline Methods in Comparison
  16. ...and 12 more sections

Key Result

Lemma 1

Let $({\bm X}_t)$ be a stationary time series for which the weak innovation representation exists. Let $({\bf V}_t)$ be the weak innovation representation of $({\bm X}_t)$. Then, for all ${\bm x}$ and ${\bm X}_{0:t}={\bm x}_{0:t}$,

Figures (8)

  • Figure 1: Forecasting pipeline for WIAE-GPF.
  • Figure 2: An autoencoder interpretation of innovation representations.
  • Figure 3: Training schematics of WIAE. Dash lines indicate the flow of training information.
  • Figure 4: Real-time, day-ahead LMPs, and load at Long Island, and real-time LMP at NYC July 2023.
  • Figure 5: Trajectories of the real-time price at LONGIL in July (top row) and October (bottom row) 2023, and its prediction generated by selective methods.
  • ...and 3 more figures

Theorems & Definitions (3)

  • Lemma 1: Bayesian Sufficiency of Multivariate Weak Innovations
  • Theorem 1: Validity of WIAE-GPF
  • Theorem 2