摘要
针对目前绝缘油电气与理化参量检测操作繁琐、实时性差的难题,提出了一种基于超声脉冲回波法与麻雀搜索算法(Sparrow Search Algorithm,SSA)优化随机森林算法(Random Forest,RF)的SSA-RF预测模型,对绝缘油击穿电压进行检测。首先,以STM32F407VGT6微控器为核心,搭建了绝缘油超声脉冲回波检测平台,以25#克拉玛依油为例,开展了油样加速热老化实验,并利用该平台采集油样的超声脉冲回波信号。然后,对回波信号进行时域和频域分析,得到162维超声回波信号特征参量,利用最大信息系数(Maximum Information Coefficient,MIC)进行特征参量筛选,获得与油样击穿电压强相关的82维特征参量。最后,构建了基于SSA-RF的绝缘油击穿电压检测模型,采用K折交叉验证进行训练,10次交叉验证的平均预测准确率为95.99%。测试集的预测准确率达到94.43%,相较于优化前的预测模型,其准确率提高了14.80个百分点。
To overcome current challenges in detecting cumbersome electrical and physicochemical parameters in real-time,this study proposes a method to detect insulating oil breakdown voltage using ultrasonic pulse-echo and a sparrow search algorithm optimized random forest(SSA-RF)prediction model.Initially,an ultrasonic pulse-echo detection platform for insulating oil was developed using the STM32F407VGT6 microcontroller core.Using 25#Karamay oil as an example,accelerated thermal aging tests were conducted on oil samples,with ultrasonic pulse-echo signals collected using this platform.Then,through time-domain and frequency-domain analyses of these signals,a total of 162-dimensional ultrasonic echo signal feature parameters were obtained.Feature parameter screening using maximum information coefficient identified 82 feature parameters strongly correlated with the oil sample s breakdown voltage.Finally,an insulating oil breakdown voltage detection model based on SSA-RF was developed.The model,trained by K-fold cross-validation,achieved an average prediction accuracy of 95.99%across 10 folds.The prediction accuracy on the test set reached 94.43%,representing an improvement of 14.80 percentage points compared to the baseline model prior to optimization.
作者
刘宏
俞华
王璇
梁基重
李帅
LIU Hong;YU Hua;WANG Xuan;LIANG Jizhong;LI Shuai(State Grid Shanxi Electric Power Co.Ltd.,Electric Power Research Institute,Taiyuan 030001,China)
出处
《西南大学学报(自然科学版)》
北大核心
2026年第1期266-278,共13页
Journal of Southwest University(Natural Science Edition)
基金
国网山西省电力公司科技项目(520530220004)
山西省重点研发计划项目(202402020101003)。
