基于时序NARX自适应神经网络的油浸式变压器绕组温度预测

Winding Temperature Prediction of Oil-immersed Transformer Based on TimeSeries NARX Adaptive Neural Network

变压器绕组热点温度过高会导致绝缘老化速度变快,剩余寿命变短。为此提出了一种基于时序性外因非线性自回归(NARX)的自适应神经网络模型以获得更精准的绕组热点温度预测数据。首先,确定影响变压器绕组温度的外部特征因子种类;然后,对变压器绕组热点数据和其他数据进行预处理;最后,将处理后的数据输入时序NARX自适应神经网络模型进行训练和调参,完成模型的构建。经实例验证,提出的外因NARX自适应神经网络绕组热点温度预测模型能对不同类型变压器数据进行特定的预处理,并且与支持向量机回归、回归树、高斯核回归方法相比,预测误差更小,在提高精度上具有更大优势。

High winding hot spot temperature of transformer will lead to faster insulation aging and shorter residual life.An adaptive neural network model based on time-series exogenous nonlinear autoregression(NARX)is proposed to obtain more accurate winding hot spot temperature prediction data.Firstly,the types of external characteristic factors that affect the winding temperature of the transformer are determined.Then,the transformer winding hot spot data and other data are preprocessed.Finally,the processed data is input into the time-series NARX adaptive neural network model for training and parameter tuning,and the model construction is completed.After case verification,the proposed external-cause NARX adaptive neural network winding hot spot temperature prediction model can perform specific preprocessing on different types of transformer data.At the same time,compared with the support vector machine regression,decision tree regression and Gaussian kernel regression methods,the prediction error of this method is smaller,and it has greater advantages in improving accuracy.