广义回归神经网络在变压器绕组热点温度预测中的应用

电力变压器的绕组热点温度是影响其绝缘性能的主要因素之一,因此有必要进行电力变压器绕组热点温度预测以提高电力变压器的运行可靠性。变压器内部温度受诸多因素的影响,且计算涉及到传热学、流体力学和电磁学等边缘学科,以致其计算复杂,不宜使用。广义回归神经网络(GRNN)具有较强的非线性映射能力和柔性网络结构以及高度的容错性和鲁棒性等特点,将其应用于变压器绕组热点温度的预测,克服了基于误差反向传播算法的人工神经网络(BPNN)预测时训练过程中存在局部最小点、收敛速度慢等缺点。将预测结果与实测值进行对比,结果表明GRNN神经网络的预测结果与实测值具有较好的一致性。

Application of Generalized Regression Neural Network to Transformer Winding Hot Spot Temperature Forecasting

The winding hot-spot temperature of transformer is the key factor which affects the insulating performance,so it is necessary to forecast the transformer winding hot spot temperature so as to improve the operational reliability of transformer.However,the inner temperature of transformer is affected by many factors,and the heat transfer,fluid mechanics,electromagnetism and other interdisciplinary increase the complexity of calculation.Generalized regression neural network(GRNN) has strong nonlinear mapping capability,flexible network structure,high fault tolerance and robustness,etc,and adopting GRNN to predict the hot spot of transformer winding will overcome the shortcomings of back propagation neural network(BPNN),such as the local minimum and low convergence speed and so on.Comparison of the results of GRNN and the measured data show that the results of GRNN are in accordance with the measured values.