A modified flower pollination algorithm and carnivorous plant algorithm for solving engineering optimization problem

• Main Author: Ong, Kok Meng
• Format: Thesis
• Language: English; English; English
• Published: 2021
• Subjects: TA401-492 Materials of engineering and construction. Mechanics of materials
• Online Access: http://eprints.uthm.edu.my/1836/2/ONG%20KOK%20MENG%20-%20declaration.pdf; http://eprints.uthm.edu.my/1836/1/ONG%20KOK%20MENG%20-%2024p.pdf; http://eprints.uthm.edu.my/1836/3/ONG%20KOK%20MENG%20-%20full%20text.pdf; http://eprints.uthm.edu.my/1836/

Optimization in an essential element in mechanical engineering and has never been an easy task. Hence, using an effective optimiser to solve these problems with high complexity is important. In this study, two metaheuristic algorithms, namely, modified flower pollination algorithm (MFPA) and carnivorous plant algorithm (CPA), were proposed. Flower pollination algorithm (FPA) is a biomimicry optimisation algorithm inspired by natural pollination. Although FPA has shown better convergence than particle swarm optimisation and genetic algorithm in the pioneering study, improving the convergence characteristic of FPA still needs more work. To speed up the convergence, modifications of: (i) employing chaos theory in the initialisation of initial population to enhance the diversity of the initial population in the search space, (ii) replacing FPA’s local search strategy with frog leaping algorithm to improve intensification, and (iii) integrating inertia weight into FPA’s global search strategy to adjust the searching ability of the global strategy, were presented. CPA, on the other hand, was developed based on the inspiration from how carnivorous plants adapt to survive in harsh environments. Both MFPA and CPA were first evaluated using twenty-five well-known benchmark functions with different characteristics and seven Congress on Evolutionary Computation (CEC) 2017 test functions. Their convergence characteristic and computational efficiency were analysed and compared with eight widely used metaheuristic algorithms, with the superiority validated using the Wilcoxon signed-rank test. The applicability of MFPA and CPA were further examined on eighteen mechanical engineering design problems and two challenging real-world applications of controlling the orientation of a five-degrees-of-freedom robotic arm and moving-object tracking in a complicated environment. For the optimisation of classical benchmark functions, CPA was ranked first. It also obtained the first rank in CEC04 and CEC07 modern test functions. Both CPA and MFPA showed promising results on the mechanical engineering design problems. CPA improved over the particle swarm optimisation algorithm in terms of the best fitness value by 69.40-95.99% in the optimisation of the robotic arm. Meanwhile, MFPA demonstrated a better tracking performance in the considered case studies by at least 52.99% better fitness function evaluation and fewer number of function evaluations as compared with the competitors.