CAJ | 학술논문

可听噪声属于输电线路电磁环境的影响因子之一，其常规预测模型均存在使用条件受约束或预测误差偏大的问题．根据间接预测法的思想，以可听噪声通用表达式中的4个因素为输入变量，可听噪声值为输出变量，建立了三层结构的BP神经网络交流输电线路可听噪声预测模型．以国外多条输电线路可听噪声数据为样本，通过BP神经网络对样本进行训练，运用得到的模型对预测集线路的可听噪声值进行了预测．结果表明，与常规GE公式相比，采用BP神经网络预测模型的预测平均绝对误差要小1．6414 dB（A）．
가은조성속우수전선로전자배경적영향인자지일，기상규예측모형균존재사용조건수약속혹예측오차편대적문제．근거간접예측법적사상，이가은조성통용표체식중적4개인소위수입변량，가은조성치위수출변량，건립료삼층결구적BP신경망락교류수전선로가은조성예측모형．이국외다조수전선로가은조성수거위양본，통과BP신경망락대양본진행훈련，운용득도적모형대예측집선로적가은조성치진행료예측．결과표명，여상규GE공식상비，채용BP신경망락예측모형적예측평균절대오차요소1．6414 dB（A）．
Audible noise is one of impact factors for power line electromagnetic environment .Nevertheless , there are the problems for application constraints and unacceptable errors in the traditional prediction models .According to the theoretical of the indirect method , an audible noise prediction model with a structure of three layers for AC trans-mission lines was established based on BP neural network , in which the input variables are four factors in general ex-pression of audible noise and the output variable is audible noise value .Taking audible noise dates of abroad multiple transmission lines as an example , a train was acquired from the training dataset through BP neural network .Using the trained model , the audible noise values were predicted for the transmission lines in prediction dataset .The results showed that the value of the mean absolute error , which was predicted by the proposed model based on BP neural net-work, was 1.641 4 dB (A) which is less than that of GE formula .