Transformer Top Oil Temperature Interval Prediction Based on Kernel Extreme Learning Machine and Bootstrap Method

In order to estimate transformer thermal condition accurately, a transformer top oil temperature (TOT) prediction model based on kernel extreme learning machine (KELM) and Bootstrap method was proposed in this paper. Firstly, the Bootstrap method was utilized to get L sets of training samples and the L sets of samples were utilized to train L KELM models to get the point prediction results of the TOT. Then, a KELM model was trained to estimate the noise variance of TOT observational data. Finally, the results of L+1 KELM models were used to calculate the TOT prediction interval with a confidence level. Case study results show that the proposed model can take the uncertainty of TOT prediction model into account properly and get accurate and reliable prediction intervals. The uncertainty of KELM TOT prediction model is lower than BP network and extreme learning machine (ELM) and is comparable to support vector machine (SVM), but the calculation is significantly faster than SVM. Compared with traditional point-prediction model, the interval-prediction model proposed in this paper can support the transformer thermal state estimation and operation safety with more reasonable and sufficient subsidiary basis. © 2017 Chin. Soc. for Elec. Eng.