基于人工神经网络和多频超声波检测技术的变压器油界面张力预测

Prediction of Interfacial Tension of Transformer Oil Based on Artificial Neural Network and Multi-frequency Ultrasonic Testing Technology

绝缘油是油浸式变压器内部重要的液体绝缘介质,油的界面张力值与油的劣化程度密切相关。基于多频超声检测技术,建立了遗传算法(GA)优化的反向传播神经网络(BPNN)变压器油界面张力预测模型。以160组变压器油样为数据样本,其中150组为训练集,10组为预测集。基于不同界面张力的分子特性及多频超声波检测原理,以242维多频超声波检测数据为输入,变压器油界面张力为输出,通过试验法确定BP神经网络的隐层神经元个数为14,由此建立非线性映射关系,并用遗传算法优化BP神经网络的各层连接权值及阈值。研究结果表明:基于GA-BPNN的油界面张力预测模型的预测平均绝对百分比误差为9.76%,相比于传统的BPNN油界面张力预测模型,其预测结果与真实值拟合程度更高、误差更小。研究结果为基于多频超声波检测技术的变压器油界面张力等参量的监测奠定了基础。

Transformer oil is an important liquid insulating medium in oil-immersed transformer. The interfacial tension value of oil is closely related to the degree of oil deterioration. Based on the multi-frequency ultrasonic detection tech- nology, a predictive model designed to detect variable pressure oil interfacial tension by using the back propagation neural network (BPNN) which is optimized by the genetic algorithm (GA) was established. Data samples were collected from 160 sets of transformer oil samples, 150 of which were training sets and 10 were forecast sets. Based on the theory of molecular properties of different interfacial tension and multi-frequency ultrasonic testing, the 242 dimensions mul- ti-frequency ultrasonic testing data were adopted as the input, and interfacial tension of transformer oil as the output. The BP neural network hidden layer neurons number was determined to be 14 in the experiment, thus a nonlinear mapping re- lationship was established, and the genetic algorithm (GA) was adopted to optimize the BP neural network connection weights and threshold of every layer. The results show that the mean absolute percent error of oil interfacial tension pre- diction model based on GA-BPNN is 9.76%. Compared with the results from traditional BPNN oil interfacial tension prediction models, the prediction results are higher and the error is smaller. The research results can lay the foundation for monitoring the interfacial parameters such as the interfacial tension of transformer oil based on multi-frequency ultrasonic detection technology.