| Electromagnetic inverse problems are very interesting and useful for industries, but hard to solve and analyze so far since they are highly nonlinear and there are no conventional methods to follow. Based on this reason, new techniques are needed to explore. In this dissertation, some methods in artificial intelligence are introduced to solve this kind of problems.;First, the fundamental theories of the artificial neural networks, genetic algorithms and evolutionary programming are presented, and their properties are analyzed. Then two examples of the inverse problems are given to solve by applying these methods and compare the corresponding results with the existing analytical solutions.;The results shows that the neural networks method can be used for the electromagnetic inverse problems, but with some limitations because of its divergence when the initial random numbers are not appropriate and the required accuracy is high. But the novel evolutionary programming technique is fully successful for both of the examples. It is never divergent under any conditions and always gives better accuracy, which means the evolutionary programming approach is stable.;Finally, this research work is summarized: This is the first time to attempt to solve electromagnetic inverse problems by introducing evolutionary technique, and this method has many advantages over the neural networks one. This work provides us new ideas and approaches when the classical methods for inverse problems are not exist or applicable. The evolutionary programming method is recommended based on the examples because of its stability.;The computer code for solving the examples with both artificial neural networks and evolutionary programming methods are attached in the appendix. This is also an indispensable content of the entire research. |
