CAJ | 학술논문

提出了一种基于人工神经网络（ANN）的模拟空气（I-Air）击穿电压的预测方法。ANN采用BP网络模型，由输入层、隐藏层和输出层3层组成。依据I-Air中针对板（N-P）、球对板（S-P）电极的工频交流击穿试验数据，分别进行了针对不均匀电场和稍不均匀电场2类的人工神经网络的数据和网络测试。对于不均匀电场网络，利用同一气压下若干气隙长度的击穿电压，预测同一气压下的其他气隙长度的击穿电压；对于稍不均匀电场网络，利用若干气压下的击穿电压预测另外气压下的击穿电压，并用Matlab中的ANN工具箱实现了人工神经网络模型，通过比较实测结果和预测结果发现预测平均误差小于5%，取得了较好的预测结果。预测方法可以用来在一定范围内预测I-Air的击穿电压，大大减少了试验的时间和试验投资成本。
제출료일충기우인공신경망락（ANN）적모의공기（I-Air）격천전압적예측방법。ANN채용BP망락모형，유수입층、은장층화수출층3층조성。의거I-Air중침대판（N-P）、구대판（S-P）전겁적공빈교류격천시험수거，분별진행료침대불균균전장화초불균균전장2류적인공신경망락적수거화망락측시。대우불균균전장망락，이용동일기압하약간기극장도적격천전압，예측동일기압하적기타기극장도적격천전압；대우초불균균전장망락，이용약간기압하적격천전압예측령외기압하적격천전압，병용Matlab중적ANN공구상실현료인공신경망락모형，통과비교실측결과화예측결과발현예측평균오차소우5%，취득료교호적예측결과。예측방법가이용래재일정범위내예측I-Air적격천전압，대대감소료시험적시간화시험투자성본。
ABSTRACT:A method based on ANN (artificial neural net-works) to predict the breakdown voltages of I-Air(Imitated Air)gas is proposed in this paper. The proposed ANN using BP network model consists of one input layer,one hidden layers and one output layer. Using the experimental data for I-Air about N-P and S-P to non-uniform and slightest non-uniform electric field,the data training and network testing are per-formed. The ANN model is implemented using ANN tool box in MATLAB. Compared with the experimental data,it is found that by using the proposed ANN,the average relative errors on predicted breakdown voltages are less than 5% for training as well as for testing in all cases and better forecasting results can be thus obtained. This prediction method can efficiently predict the breakdown voltage of I-Air in a certain range. The cost of experiment time and investment are greatly diminished.