Stochastic Modeling for Wind–Solar Power Forecasting: Uncertainty‑Driven Risk Decisions in Modern Grids

Abstract

The rapid rise of wind and solar capacity has transformed power generation but introduced severe forecasting uncertainty. Variable renewable production can swing from minutes to days, creating operational and economic risks for grids built around predictable dispatch. This narrative review surveys stochastic modeling techniques for wind and solar power forecasting published between 2015 and 2025, covering time‑series, spatial–temporal, hybrid, and deep learning approaches. Comparative evidence shows that autoregressive and Kalman‑filter models provide interpretable benchmarks yet struggle with non‑linearities; copula and vine‑copula schemes better capture spatial dependence; hybrid schemes that fuse numerical weather prediction with machine learning significantly reduce forecast errors; and emerging non‑stationary Gaussian processes and generative models further improve probabilistic accuracy. Persistent gaps include limited cross‑regional validation, short training periods, and inconsistent evaluation metrics. The review suggests that risk‑aware scheduling can leverage these probabilistic forecasts for chance‑constrained reserves and conditional‑value‑at‑risk unit commitment, enabling more reliable and economical integration of wind and solar power.