基于LSTM网络的变压器油中溶解气体浓度预测

电力变压器作为电力系统中传输和变换电能的主要设备,其安全稳定性运行在电网中起着重要的作用。对变压器油中溶解气体浓度变化的趋势进行预测,可为其运行状态评估提供重要依据,鉴于此提出了一种基于长短期记忆网络(LSTM)的变压器油中溶解气体浓度预测模型。该模型克服了传统神经网络在序列预测方面存在的"梯度消散"问题,利用油中溶解气体的序列数据对长短期记忆网络进行训练,得到最优的预测模型参数。以变压器油中溶解的7种特征气体浓度为输入,以待预测气体的浓度为输出。通过算例分析表明,相比于传统的机器学习预测方法支持向量机(support vector machine, SVM)与反向传播神经网络(back propagation neural network,BPNN),本文所提的LSTM预测模型更能准确地预测油中溶解气体的浓度。

Prediction of dissolved gas concentration in transformer oil based on LSTM network

Power transformers play an important role in the safe and stable operation of the power grid as the main equipment for transmitting and transforming electrical energy in power systems, Predicting the trend of dissolved gas concentration in transformer oil can provide an important basis for its operational status assessment. In view of that, a long-short-term memory network(LSTM) based prediction model for dissolved gas concentration in transformer oil is proposed. The model overcomes the problem of "gradient dissipation" in sequence prediction of the traditional neural networks. It uses the sequence data of dissolved in oil to train the long-short-term memory in order to obtain the optimal prediction model parameters. The model sets the concentration of the seven characteristic gases dissolved in the transformer oil as input, and the concentration of the gas to be predicted as output. The example analysis shows that Compared with traditional machine learning prediction methods, support vector machine(SVM) and back propagation neural network(BPNN), the LSTM prediction model can predict the concentration of dissolved gases in oil more accurately.