Research On Optimization Design Of Electric Apparatus Using Artificial Neural Networks

The increase voltage level of high voltage transmission system, miniaturization and complexity of the electric apparatus have brought forward higher requirement. The structure optimization plays a very important role for investigating and developing the electric apparatus accordingly. There are many variables for the optimization of the electric apparatus. However, for the traditional optimizing methods, it is very difficult to solve the intricate computation for the optimization mathematical model with higher efficiency and precision. Aiming at some problems, presently, the intelligent optimization approaches have widely applied for designing the electric apparatus.In this thesis, the Artificial Neural Network (ANN) has been improved for overcoming the disadvantages of conventional ANN. And novel dynamic ANN with optimization mechanism of "Step-by-Step", that is, training precision and the solution of the optimum is performed in dynamic and parallel processes, has been proposed. Moreover, the combination of the dynamic ANN with the numerical computation of the electric field has been applied for designing the high voltage electric apparatus for realizing the optimization with higher approaching capability and efficiency. During the whole optimization process, the optimizing direction is determined by the programming. And with the purposeful sampling and gradual reducing the sample space, large numbers of tedious sampling and training can be avoided and the efficiency of the optimization of the electric apparatus can be improved.The automation generating technology of the meshing plays a very important and necessary role for providing enough useful samples. In this thesis, the Advancing Front Method is used to generate the meshing, and the sampling sets selecting with higher quality has been realized.Finally, the dynamic ANN is employed for solving the complexity nonlinear oscillating mathematical function and searching the optimum, studying the relationship of the approaching capability and the ANN network. Moreover, the convergency speed and the range of thenumber of the neurons in hidden layer and the likes have been analyzed for testing the reliability and efficiency of the optimization technique in this thesis. Furthermore, based on the above works, the optimization of the contact of a 220kV SF6 circuit breaker and disc insulator of 500kV GIS have been optimized. From the optimization results, the feasibility and validity of the proposed dynamic ANN have been verified. And the basis of numerical computation and optimization for electric apparatus has been provided.