Weather-Informed Probabilistic Forecasting and Scenario Generation in Power Systems

TL;DR

This work tackles high-dimensional probabilistic forecasting for load and renewable generation under weather-driven uncertainty. It unions weather-informed multivariate forecasting with a Gaussian copula to restore spatio-temporal dependencies and generate realistic scenarios, evaluated on a large MISO dataset with a 48-hour horizon. The Weather-Informed Temporal Fusion Transformer (WI-TFT) model, combined with copula-based scenario generation, achieves superior deterministic accuracy and markedly improved joint-scenario realism, outperforming ARIMA, DLinear, NLinear, and DeepAR, especially as lead time grows. The approach offers a scalable, practically impactful framework for risk-aware grid operations and decision-making under uncertainty, with future work on robustness to covariate noise and uncertainty quantification via conformal methods.

Abstract

The integration of renewable energy sources (RES) into power grids presents significant challenges due to their intrinsic stochasticity and uncertainty, necessitating the development of new techniques for reliable and efficient forecasting. This paper proposes a method combining probabilistic forecasting and Gaussian copula for day-ahead prediction and scenario generation of load, wind, and solar power in high-dimensional contexts. By incorporating weather covariates and restoring spatio-temporal correlations, the proposed method enhances the reliability of probabilistic forecasts in RES. Extensive numerical experiments compare the effectiveness of different time series models, with performance evaluated using comprehensive metrics on a real-world and high-dimensional dataset from Midcontinent Independent System Operator (MISO). The results highlight the importance of weather information and demonstrate the efficacy of the Gaussian copula in generating realistic scenarios, with the proposed weather-informed Temporal Fusion Transformer (WI-TFT) model showing superior performance.

Figures (7)

• Figure 1: Schematic of the proposed method for scenario generation in RES. (a) Schematic of the weather informed multivariate probabilistic forecasting for wind farms, given the historical wind power generation and the past and future weather covariates. (b) Schematic of estimating the Gaussian copula for a 2D example. (c) Schematic of the scenario generation steps given the estimated copula and estimated marginals for a 2D problem. Examples of generated scenarios using marginals are also presented in (c) with red triangles.
• Figure 2: Total-level forecasting error (i.e., MAE) in MW versus lead time for the time series models for (a) load, (b) wind power generation, and (c) solar power generation.
• Figure 3: Comparison of Marginal and Copula Scenarios. Generated scenarios for a load (LRZ1) using (a) marginals (i.e., no inclusion of copula) and (e) with the proposed copula method. (b) The ramps corresponding to the actuals and generated scenarios from marginals in (a). (f) The ramps for the actuals and generated scenarios using copula from (e). (c, d) The generated scenarios and the ramps for at total aggregated level using marginals and (g, h) using the proposed copula method.
• Figure 4: WI-TFT + Copula for Scenario Generation. Examples of the forecast and 5 generated scenarios for (a) load, (b) wind, and (c) solar power generation. For simplicity, only 5 scenarios are plotted for each case.
• Figure 5: DeepAR

Theorems & Definitions (1)

• Theorem 1: Sklar's Theorem