Abstract: Aiming at the problems of standard Invasive Weed Optimization(IWO) algorithm such as population diversity declining in the late evolution and easily trapping into local extremum, an improved algorithm, Multi Sub-population Invasive Weed Optimization algorithm based on K-means clustering(K-MSIWO), is proposed. In K-MSIWO, the weed population is divided into three sub-populations by using K-means clustering. The co-evolutionary relationship of individual and individual, sub-population and sub-population is established through intraspecific and interspecific competition to increase the diversity of the weed population. When the convergence velocity of algorithm begins falling, the random perturbation mutation strategy is adopted for the premature individual to help them out of local minimum. Experimental results on several benchmark functions show that the modified algorithm is applied to PID control parameter tuning of second-order and high-order system, the overshoots of the systems are reduced by 33.2% and 50% respectively compared with GA approaches, and K-MSIWO has better accuracy and consistency.
Invasive Weed Optimization(IWO),