基于混合模态分解和LSTM-CNN的变压器油中溶解气体浓度预测

对油中溶解气体浓度进行预测，可提前掌握变压器运行趋势。提出一种基于混合模态分解和LSTMCNN(long short-term memory network-convolution network)网络的预测方法，实现精准的气体浓度预测。首先，为消除分解中模态混叠和残余白噪声的影响，对气体序列进行ICEEMDAN分解，以削弱序列的非平稳性；然后，使用VMD对聚合重构后的高频分量进行二次分解，降低高频分量的复杂度；最后，为了增强模型对序列时间特征和空间特征的拟合，采用结合时间注意力机制的LSTM-CNN网络对分解分量分别进行预测并重构气体浓度数据。算例验证表明，所提出的模型相比其他模型具有更强的预测性能，为后续故障预测提供有力支撑。

Prediction of Dissolved Gas Concentration in Transformer Oil Based on Hybrid Mode Decomposition and LSTM-CNN

Predicting the concentration of dissolved gas in oil can help to know in advance the operation trend of transformers. A prediction method is thus proposed based on hybrid mode decomposition and LSTM-CNN network to achieve accurate gas concentration prediction. Firstly, in order to eliminate the influence of mode aliasing and residual white noise in the decomposition, the gas sequence is decomposed with ICEEMDAN to weaken the non-stationarity of the sequence. Then, the VMD is used to decompose the high frequency components after aggregation reconstruction to reduce the complexity of the high frequency components. Finally, in order to enhance the fitting of the model to the temporal and spatial features of the sequence, the TA-LSTM-CNN is used to predict the decomposition components and reconstruct the gas concentration data. Case study shows that the proposed model has stronger prediction performance than other models, which can provide strong support for subsequent fault prediction.