Swarm Intelligence Optimization is a series of relatively new optimization algorithm, and is one of the most focused fields in optimization nowadays.It simulates the swarm behavior of the social animals,using the information interchange and cooperation between individuals to achieve the optimization.Compared to the other optimization methods,it is easy to be implemented and is efficient.However,though some accomplishments have been obtained,generally speaking,the new field is still an open area,and there are still numerous problems to be solved,such as how to further improve the optimization efficiency and how to apply the other social animal system methodology to the optimization.The present thesis makes studies in algorithms and their applications.Chapter 2 to Chapter 5 study the parameter selection strategies,parameter update strategies,hybrid optimization scheme,and new algorithm design.Chapter 6 and Chapter 7 study their applications,using the 2-dof parallel manipulator as the experiment object,and calibrate all the kinematic parameters of the 2-dof parallel manipulator.In Chapter 2,orthogonal experiment is introduced into the parameter selection of the ant colony optimization,with the typical TSP problem as the test problem.In the study,three-level orthogonal experiment is done twice over the main four parameters, and the results show that the proposed method can select parameters in relatively few experiments and in higher precision.An ant colony optimization with negative feedback scheme is proposed,in which the positive pheromone is set on the good path and the negative pheromone is set on the bad path,and the TSP experimental results demonstrate that the negative feedback scheme outperforms the conventional one over the diversity maintenance.In Chapter 3,according to the principle of mechanics,a new adaptive inertia weight strategy on the dimension-level of each particle is proposed,in which the search is dependent on the searching state of each particle(including the current location and velocity). Based on this strategy,an inertia weight function is designed with the reference of Butter-worth filter function;Because the search step of particle swarm optimization is relatively large,and that of evolutionary strategies is relatively small,a hybrid optimization is proposed, wherein the good individual is updated in evolutionary strategies way and the bad individual is updated in particle swarm optimization way.The experimental results demonstrate that the hybrid optimization can improve the performance.In order to improve the multi-modal optimization ability of differential evolution, Chapter 4 introduces the concept of unstable global minima to the conventional crowding clustering,and elitism mechanism is added to crowding clustering,thus an improved crowding clustering is proposed,and theoretically proved that in this improved crowding clustering,the unstable global minima would not be cleared by any parent individual and would enter the next generation,and that local stable minima would have larger probability to enter the next generation.Differential evolution is incorporated with this improved crowding clustering and improved crowding clustering differential evolution(ICCDE) is proposed accordingly.In Chapter 5,the overall frameworks of the algorithms that is studied in the previous Chapters are firstly summarized.Then,the fish shoal behavior research result is intro- duced into the swarm intelligence continuous optimization algorithm in this paper,and a novel fish shoal algorithm is proposed.The algorithm simulates the movement of fish shoal in the space.According to the Euclidean distance,the neighborhood of the fish individual is divided into attraction zone,repulsion zone and neutral zone.The others can attract,repulse and do random attraction or repulsion accordingly.Furthermore,the movement trend to the food source is considered.The experiments are carried out on the benchmark functions for comparing novel fish shoal algorithm with artificial fish-swarm algorithm and particle swarm optimization.Finally,a linear change weight factor strategy is proposed,and experiments are done for comparison between the fixed weight factor strategy and the linear change weight factor strategy.Based on the close loop constraint of the planar 2-dof parallel manipulator,error functions are constructed.Further,through the kinematics study,the passive joint positions are eliminated from the error functions and more compact presentations are obtained, with the number of the parameters to be calibrated being decreased.Then,by using variable substitution,the product item is decoupled,and the error functions become more easier.In the system with three parameters of which are set fixed,the self-calibration of the planar 2-dof parallel manipulator are studied first by simulation.Through simulation, the most suitable error function and the most appropriate optimization algorithm are determined,and then the actua.system calibration are done thereafter,with the 12 parameters out of the 15 being calibrated.For the purpose of calibrating all the parameters,based on plane geometry analysis, three pose transformation are illustrated,and the subsequent multiple solution problem due to the pose transformation is analyzed,and the necessity of fixing three base coordinates are illustrated.Then,by means of external measurement,the strategy of eliminating multiple solutions and an integrated calibration based on closed-loop and open-loop are proposed.By applying differential evolution as the optimization algorithm,an integrated calibration without fixing any parameter is studied by simulation.Because the error function is typically a multi-modal function with numerous solutions,the improved crowding clustering differential evolution is applied.The results show that all the kinematic parameters of the manipulator can be calibrated with high precision. |