DEVELOPMENT OF ANN ENSEMBLE USING MULTIOBJECTIVE OPTIMIZATION FOR HYBRID FRACTAL ANTENNA DESIGN by Sahajpal Singh
Abstarct
Diversity and accuracy are the two critical requirements for an effective ANN ensemble. Since accuracy and diversity are conflicting with each other, therefore it is difficult to improve both objectives at same time; it means the creation of an ANN ensemble with accurate and diverse members leads to a multi-objective optimization problem. For addressing the conflicting nature of accuracy and diversity in ANN ensemble learning, multi-objective evolutionary algorithms have been successfully developed to solve optimization problems having multiple conflicting objectives. In proposed thesis work, Multi-Objective Particle Swarm Optimization (MOPSO) algorithm is used to address the tradeoff between diversity and accuracy. PSO is an efficient, simple, and population-based technique. Therefore, it can be naturally extended to deal with the multi-objective optimization problems. For a multi-objective optimizations problem, PSO can be modified in two ways. First, each objective function is treated separately, and second, all objective functions are evaluated for each particle. The mean square error (MSE) of individual ANNs is used as first objective function that describes the accuracy of ANN ensemble candidates; the correlation penalty function from Negative Correlation Learning is used as second objective for finding the diversity of ANN ensemble. Both objectives are minimized for optimizing the best values of accuracy and diversity for designing ANN ensemble. The optimized ANN ensemble is then used for finding the optimal dimensions of proposed hybrid fractal antenna. The proposed hybrid fractal antenna is designed by modifying the boundaries of the prefractal antenna by adding Koch curves at particular locations. The dimensions of the proposed antenna for resonant frequency of 2.45 GHz are optimized by using MOPSO based NNE, then for that optimized dimensions, the antenna is simulated and fabricated which shows its good matching between simulation and experimental results.
