Design of Grey Prediction Controller Based on Adaptive Grey Wolf Optimizer

Abstract

Although there are theoretical advantages of grey prediction control when dealing with systems having poor information; but they have problems such as experience-dependent parameter tuning, fixed model structure and vulnerable to local optima during optimization. We propose integration of the adaptive grey wolf optimizer into the parameter collaborative tuning process of the grey prediction controller resulting in an integrated design approach. Algorithm level: analyse the discrepancy between the linear decay of the convergence factor and the search state in the standard grey wolf algorithm by suggesting an adaptive adjustment based on the variance of population distribution. Prove convergence via Markov chain theory without increasing additional computational overhead up to a constant factor magnitude. Modeling level: Construct a dynamic grey prediction module in the forward channel of the controller (using a rolling window and online parameter adjustment) tracking time-varying characteristics and designing an online correction strategy separating deviation temporal features suppressing prediction inaccuracies. Parameter optimization level: Adjust all adjustable parameters of the predictor and the controller into one optimization vector building a weighted objective function. Collaboration optimization lead us to use the Lyapunov method deriving sufficient conditions ensuring the stability of the closed-loop system. Coordination match of predictive modeling and control tuning under the same framework provide an effective way for the design of grey prediction controllers.