基于多频超声和人工神经网络的变压器油中微水含量检测

Detection of Micro-water Content in Transformer Oil Based on Multi Frequency Ultrasonic and Artificial Neural Network

变压器油中的微水含量是衡量变压器能否长期稳定运行的重要因素。本研究基于多频超声检测结合人工神经网络算法,提出一种变压器油中微水含量预测方法。首先,利用卡尔费休滴定法测定210组油样中的微水含量。其次,对210组油样进行多频超声检测,分析油样中微水含量与多频超声数据中振幅和相位信号的关系。最后,利用PCA将原始242维多频超声数据降为23维,结合BPNN和GRNN两种人工神经网络以及GA和PSO两种优化算法,建立了基于PCA-GA-BPNN和PCA-PSO-GRNN的两种变压器油中微水含量预测模型,并将预测结果与实际结果进行对比。结果表明:两种预测模型的预测准确率均超过90%,表明本研究提出的方法能够有效地检测变压器油中的微水含量。

The micro-water content in transformer oil is an important factor to measure whether the transformer can operate stably for a long time.Based on multi-frequency ultrasonic detection combined with artificial neural network algorithm,a method for predicting micro-water content in transformer oil was proposed in this study.Firstly,the micro-water content in 210 groups of oils was determined by Carl Fischer titration.Secondly,210 groups of oil samples were detected by multi-frequency ultrasound to analyze the relationship between micro water content in oil samples and amplitude and phase signals in multi-frequency ultrasonic data.Finally,the original 242-dimensional multi-frequency ultrasonic data was reduced to 23-dimensional by PCA.Two prediction models for micro-water content in transformer oil based on PCA-GA-BPNN and PCA-PSO-GRNN were established by combining with BPNN and GRNN artificial neural networks as well as GA and PSO optimization algorithms.The prediction results were compared with the actual results.The results show that the forecast accuracy of both models is higher than 90%,which indicates that the method proposed in this study can effectively detect the moisture content in transformer oil.