An Adaptive Metamodel And Its Application On Integrated Electromechanical Optimization Of Reflector Antennas

Microwave reflector antenna is a kind of typical electromechanical system. The mechanical structure is not only the guarantee and bearing-structure of the antenna, but also affects the electrical performance of the antenna.Large microwave reflector antenna typically requires extensive use of simulation-based analysis tools. Re-analysis on electromagnetic field is used to get the ideal electrical performance, especially for the higher frequency which demands the mesh size be within 1/5 to 1/8 wavelength. The high computational cost often limits the application of design optimization. To improve the design quality and computational efficiency, this thesis focuses on the problem of metamodel techniques and its applicability. The main work is as follows:1. The process of building adaptive metamodel is expounded. Latin Hypercube Design sampling method and Kriging model are described in detail. Four infill criteria, as well as their advantages and disadvantages are discussed and analyzed. By using tests functions to make comparative research on the four infill criteria, results show that the capability of minimizing the response surface and minimizing the lower confidence bound in both aspects of optimization efficiency and prediction accuracy is better than maximizing expected improvement and maximizing the probability of improvement.2. In order to improve the predicting ability in the entire design space and ensure the accuracy of metamodel in the important domain, multi-point adaptive metamodel integrated important design domain is proposed. Two new samples are selected to update metamodel. One is chose in the region where uncertainty may be large for global exploration, and the other one is for local exploitation in the space around current best optimum. To improve the efficiency of local exploitation, important design domain is used to manage the metamodel. A number of multimodal numerical benchmark problems and I-beam design problem are used to test the validity and efficiency of the method. The comparison results demonstrate that the method proposed in this thesis outperforms several existing algorithms, in terms of predicting accuracy, computational efficiency and robustness.3. An optimization design model of microwave reflector antenna is proposed by using the adaptive metamodel above. Then the model is applied to the traditional optimization and electromechanical synthetic optimization of 8m reflector antenna. Compared with other methods, the proposed method obtains a better result in the aspects of predicting accuracy and the computational efficiency.