Adaptive bio-inspired firefly and invasive weed algorithms for global optimisation with application to engineering problems

The focus of the research is to investigate and develop enhanced version of swarm intelligence firefly algorithm and ecology-based invasive weed algorithm to solve global optimisation problems and apply to practical engineering problems. The work presents two adaptive variants of firefly algorithm by introducing spread factor mechanism that exploits the fitness intensity during the search process. The spread factor mechanism is proposed to enhance the adaptive parameter terms of the firefly algorithm. The adaptive algorithms are formulated to avoid premature convergence and better optimum solution value. Two new adaptive variants of invasive weed algorithm are also developed seed spread factor mechanism introduced in the dispersal process of the algorithm. The working principles and structure of the adaptive firefly and invasive weed algorithms are described and discussed. Hybrid invasive weed-firefly algorithm and hybrid invasive weed-firefly algorithm with spread factor mechanism are also proposed. The new hybridization algorithms are developed by retaining their individual advantages to help overcome the shortcomings of the original algorithms. The performances of the proposed algorithms are investigated and assessed in single-objective, constrained and multi-objective optimisation problems. Well known benchmark functions as well as current CEC 2006 and CEC 2014 test functions are used in this research. A selection of performance measurement tools is also used to evaluate performances of the algorithms. The algorithms are further tested with practical engineering design problems and in modelling and control of dynamic systems. The systems considered comprise a twin rotor system, a single-link flexible manipulator system and assistive exoskeletons for upper and lower extremities. The performance results are evaluated in comparison to the original firefly and invasive weed algorithms. It is demonstrated that the proposed approaches are superior over the individual algorithms in terms of efficiency, convergence speed and quality of the optimal solution achieved.