Multimodal Fusion‐Based Cooperative Adaptive Control Method for Blade De‐Icing Using Multi‐Agent Systems

ABSTRACT

This paper proposes a multimodal fusion‐based cooperative adaptive de‐icing control method. The method aims to achieve optimal de‐icing performance while maintaining wind turbine efficiency and operational safety. First, a control‐oriented reduced‐order thermodynamic model of blade electric‐heating de‐icing is established from the Messinger‐type energy‐balance principle. Then, a smooth switching function is introduced to enable multimodal fusion, providing a unified representation of sensible heating and latent melting modes. Next, to handle unknown system dynamics, a neural network (NN)‐based state identifier is designed to approximate the unknown nonlinear dynamics and provide online state reconstruction from available local measurements. Each independently heated zone on the blade surface is treated as an agent, forming a multi‐agent (MA) consensus cooperative control framework that incorporates both its own and neighboring control‐effort costs. This framework transforms the de‐icing problem into a MA cooperative optimal control task. Using adaptive dynamic programming (ADP), a critic NN approximates each agent's optimal value function online, from which a near‐optimal consensus cooperative control law is derived. The control law achieves coordinated de‐icing across all heating zones while minimizing the prescribed cooperative performance index. Appropriate Lyapunov functions are constructed to rigorously establish the stability of the NN identifier and the closed‐loop cooperative control system, guaranteeing the uniform ultimate boundedness of the estimation, learning, and consensus errors. Finally, simulation results validate the effectiveness of the proposed method.