ニューラルネットワークを用いた電界最適化技術の開発

抄録

For the electrical insulation of high voltage apparatus we developed a highly efficient electric field optimization technique using neural network (NN). Once the NN learns the relationship between inputs and outputs, the developed optimization algorithm easily allows to get a target solution. The procedure is as follows: (1)we provide learning patterns and calculate learning data, such as electric field strength, electrode curvature, gap length and coordinate. (2)The NN learns teaching data. (3) Inputting optimum value of electric field strength to NN allows to get an optimum electrode contour. During the optimization process, we proposed an algorithm to determine the optimum field value from the electric field distribution on learning contours. As a result, it took only 12 seconds to reach the optimum contour by personal computer. Consequently, the proposed method in this paper realized the highly efficient electric field optimization by NN.