Meteorological pattern-conditioned generative machine learning in model predictive control optimization for wind power dispatch

doi:10.1016/j.segan.2026.102473

Meteorological pattern-conditioned generative machine learning in model predictive control optimization for wind power dispatch

2026-08-04

0

PRE

AI

DOI：10.1016/j.segan.2026.102473

Original

Original request for help

Save

Abstract

En 中文

• Proposes a Conditional-Generation-Prediction-Optimization (C-G-P-O) framework that integrates meteorological wind pattern–conditioned generative models with model predictive control (MPC) for power dispatch. • Develops and compares three conditional generative models—WP-CGAN, WP-CVAE, and WPCDiffusion— to generate wind power data tailored to specific meteorological patterns (synoptic, mesoscale, and local scales). • Demonstrates that forecasters trained on conditionally generated data achieve lower prediction errors (MAPE) and yield economic performance closer to the true cost than those trained on original or unconditionally generated data. • Validates the framework on IEEE 30-bus and 118-bus systems, showing that WP-CDiffusion generally provides the most accurate predictions and best economic outcomes across all wind patterns.

Keywords:

Conditional generative machine learning

CDiffusion

CGAN

CVAE

Model predictive control

Wind power

Meteorological pattern-conditioned generative machine learning in model predictive control optimization for wind power dispatch

Abstract

Journal

Organization