Novel Bacterial Foraging Optimisation Algorithms with Application to Modelling and Control of Flexible Manipulator Systems

Biologically-inspired soft-computing algorithms, which were developed by mimicking evolution and foraging techniques of animals in nature, have attracted significant attention of researchers. The works are including the development of the algorithm itself, its modification and its application in broad areas. This thesis presents works on biologically-inspired algorithm based on bacterial foraging algorithm (BFA) and its performance evaluation in modelling and control of dynamic systems. The main aim of the research is to develop new modifications of BFA and its combination with other soft computing techniques and test their performances in modelling and control of dynamic systems. Modification of BFA focuses for improving its convergence in terms of speed and accuracy. The performances of modified BFAs are assessed in comparison to that of original BFA. In the original BFA, in this thesis referred as standard BFA (SBFA), bacteria use constant chemotactic step size to head to global optimum location. Very small chemotactic step size around global optimum location will assure bacteria find the global optimum point. However, a large number of steps is needed for the whole optimisation process. Moreover, there is potential for the algorithm to be trapped in one of the local optima. On the contrary, big chemotactic step size will assure bacteria have faster convergence speed but the literature shows that it results oscillation around global optimum point and the algorithm potentially missing the global optimum point and leading to oscillation around the point. Thus SBFA can be improved by applying adaptable chemotactic step size which could change: very large when bacteria are in locations far away from the global optimum location, to speed up the convergence, and very small when bacteria are in the locations near the global optimum so that bacteria able to find global optimum point without oscillation. Here, four novel adaptation schemes allowing the chemotactic step size to depending on the cost function value have been proposed. The adaptation schemes are developed based on linear, quadratic and exponential functions as well as fuzzy logic (FL). Then, the proposed BFAs with adaptable chemotactic step size, i.e. linearly adaptable BFA (LABFA), quadratic adaptable BFA (QABFA), exponentially adaptable BFA (EABFA) and fuzzy adaptable chemotactic step size (FABFA), are validated by using them to find global minimum point of seven well-known benchmark functions commonly used in development of optimisation techniques development. The results show that all ABFAs achieve better accuracy and speed compared to those of SBFA. The ABFAs are then used in modelling and control of a single-link flexible manipulator system. This includes modelling (based on linear model structures, neural network (NN), and fuzzy logic (FL)), optimising joint-based collocated (JBC) proportional-derivative (PD) control, and optimising both PD and proportional integral derivative (PID) control of end-point acceleration feedback for vibration reduction of a single-link flexible manipulator. The results show that ABFAs outperform SBFA in terms of convergence speed and accuracy. Since all SBFA and ABFAs use the same general parameters and bacteria are initially placed randomly across the nutrient media (cost function), the superiority better performance of ABFAs are attributed to the proposed adaptable chemotactic step size.