基于LSTM的箱式变压器高压套管温度预测

针对箱式变压器环境封闭、散热性能差而导致变压器各部件温度较高，且变压器套管事故率高的现状，提出一种基于长短期记忆(Long Short-Term Memory, LSTM)神经网络的箱式变压器高压套管温度预测方法，对箱式变压器高压套管热流进行分析，建立基于LSTM的变压器高压套管温度预测模型，LSTM算法可以解决有效解决变压器高压套管温度预测所存在的非线性和时滞性的问题，通过红外传感技术对某小区箱式变压器高压套管相关数据进行在线监测，对现场数据进行预处理，通过算例分析验证了文中所提方法预测精度更高、误差更小、泛化能力更强。对比结果表明，所提方法优于普通循环神经网络(Recurrent Neural Network, RNN)和支持向量机(Support Vector Machine, SVM)预测方法，平均误差分别降低了27.4%和36.3%,预测精度更高，与变压器套管温度实测值更趋一致。

Temperature prediction of box-type transformer high-voltage bushing based on LSTM

Aiming at the current situation of high temperature of transformer components due to the closed environment and poor heat dissipation performance of box transformers, and the high accident rate of bushings, a box transformer based on Long Short-term Memory (LSTM) neural network is proposed. Prediction method of high voltage bushing temperature. First, analyze the heat flow of the high-voltage bushing of the box-type transformer. Then, a LSTM-based transformer high-voltage bushing temperature prediction model is established. The LSTM algorithm can effectively solve the problems of nonlinearity and time delay in the transformer high-voltage bushing temperature prediction. Finally, using infrared sensing technology to monitor the relevant data of the box-type transformer high-voltage bushing in a residential area, the field data is preprocessed, and the calculation example analysis is performed to verify that the proposed method has high prediction accuracy, small error and strong generalization ability. The comparison results show that the proposed method is better than ordinary recurrent neural network (RNN) and support vector machine (SVM) prediction methods. The average error is reduced by 27.4% and 36.3%, respectively, and the prediction accuracy is higher. It is more consistent with the measured value.