绝缘油是电抗器内部重要的绝缘介质,击穿电压是评估其绝缘特性的关键指标,与绝缘油的品质状态密切相关.本文共选取155组电抗器绝缘油进行实验,分别进行击穿电压的测定和多频超声信号在油样中传播衰减后信号的采集,分析多频超声声学参数和击穿电压之间的幅频响应、相频响应之间的关系,并基于多频超声检测技术提出结合灰狼优化算法(grey wolf optimizer,GWO)优化随机森林算法(random forest algorithm,RF)的击穿电压预测方法.结果表明:GWO-RF绝缘油击穿电压预测模型的预测值与实际值的平均相对误差为4.04%,预测准确率达到95.96%,相较于优化前的RF绝缘油击穿电压预测模型准确率提升了20.25%.结合多频超声检测技术和GWO-RF建立的并联电抗器绝缘油击穿电压预测模型,对击穿电压的预测具有可行性.
Research on prediction of breakdown voltage of insulating oil based on multi-frequency ultrasound and GWO-RF algorithm
Insulating oil plays a critical role as a dielectric medium in reactors,and the breakdown voltage is a key indicator evaluating its insulating properties,which is closely related to the quality of insulating oil.In this paper,155 reactor insulating oil samples were selected for experiments,which included the measurement of breakdown voltage and collection of multi-frequency ultrasound signals after propagation in the oil samples.The relationship between the breakdown voltage and the amplitude-frequency and phase-frequency responses of ultrasonic acoustic parameters was analyzed.A breakdown voltage prediction method was then proposed by combining multi-frequency ultrasound technology with a grey wolf optimizer(GWO)optimized random forest(RF)algorithm.The results show that the GWO-RF model achieves 4.04%of mean relative error and 95.96%of accuracy on the test set,and there is 20.25%of improvement in prediction accuracy compared to the unoptimized RF model.The proposed prediction model,which integrates multi-frequency ultrasound detection and GWO-RF optimization,demonstrates significant feasibility for predicting the breakdown voltage of insulating oil in reactor.
万方数据
维普
