基于C-Prophet的变压器油温预测方法

A New Prediction Method of Transformer Oil Temperature Based on C-Prophet

变压器绕组温度过高会导致绝缘老化,严重危害电力设备正常运行。变压器油温能够作为判断绕组温度的辅助依据,然而变压器油温受季节因素、天气变化影响较大,目前针对变压器油温预测的精度有待进一步提升。针对此问题,首次将Prophet算法用于变压器油温预测,并提出一种与自适性噪声完备总体经验模态分解(CEEMDAN)相结合的变压器油温预测方法(CEEMDAN-Prophet,C-Prophet)。同时为进一步提升油温预测精度,在预测过程中综合考虑季节变化对变压器油温的影响。C-Prophet预测方法首先利用CEEMDAN对采集的油温信号进行分解,得到最终残差和K个本征模态分量。然后利用Prophet算法对各个分量进行预测,并将得到的N个模态分量的预测值进行求和得到最终的预测结果。算例结果表明,与卷积神经网络(convolutional neural networks,CNN)、长短期记忆网络(long short-term memory,LSTM)和随机森林算法相比,C-Prophet算法预测精度分别提高了54.40%、44.08%、20.09%,且C-Prophet算法在考虑季节因素后预测精度进一步提升了61.24%。

Excessive temperature of transformer windings causes insulation aging and bring about serious harms to the normal operation of the power equipment.Transformer oil temperature can be used as an auxiliary basis for judging the winding temperature.However,the temperature is often affected by seasonal factors and weather changes,and the present prediction accuracy of transformer oil temperature needs to be further improved.To this end,the Prophet algorithm is first applied to the prediction of transformer oil temperature,and a new transformer oil temperature prediction method(CEEMDANProphet,C-Prophet)combined with adaptive noise complete ensemble empirical mode decomposition(CEEMDAN)is proposed.Furthermore,to further improve the prediction accuracy,the influence of transformer temperature seasonality and weather factors are considered in the prediction process.This method first uses CEEMDAN to decompose the collected oil temperature signal to obtain the final residual and K intrinsic mode components.Then the Prophet algorithm is used to predict each component,and the predicted values of N modal components are summed to obtain the final prediction results.The case study results show that,compared with Convolutional Neural Networks(CNN),Long Short-Term Memory(LSTM)and random forest algorithm,the prediction accuracy by the new method is increased by 54.40%,44.08%,20.09%respectively.Moreover,the prediction accuracy of C-Prophet algorithm is further improved by 61.24%with seasonal factors considered.