基于核极限学习机和Bootstrap方法的变压器顶层油温区间预测
发布时间：2018-10-24 14:27
【摘要】：为准确估计变压器热状态,提出了一种基于核极限学习机和Bootstrap方法的变压器顶层油温区间预测模型。首先通过Bootstrap采样得到L组训练样本,分别训练L个核极限学习机模型对顶层油温进行拟合回归点预测;然后训练一个核极限学习机模型对顶层油温观测噪声方差进行回归估计;最后根据这L+1个核极限学习机模型的结果估计在某置信水平上的顶层油温的预测区间。算例仿真结果表明,该方法可以较好地考虑变压器顶层油温预测模型的不确定性,得到较为准确可靠的顶层油温预测区间;采用核极限学习机算法的顶层油温区间预测结果的不确定性小于BP神经网络和极限学习机,与采用支持向量机算法区间预测模型相当,但计算速度明显优于支持向量机。相比于传统的顶层油温点预测方法,所提区间预测方法可以为变压器的热状态估计、安全运行等提供更为合理和充分的辅助依据。
[Abstract]:In order to accurately estimate the thermal state of the transformer, a prediction model of the top oil temperature interval of the transformer based on the kernel limit learning machine and the Bootstrap method is proposed. First, L group training samples are obtained by Bootstrap sampling, then L kernel extreme learning machine models are trained to predict the top oil temperature, then a kernel extreme learning machine model is trained to estimate the noise variance of the top layer oil temperature observation. Finally, the prediction interval of the top oil temperature at a certain confidence level is estimated according to the results of the L 1 nuclear extreme learning machine model. The simulation results show that this method can take into account the uncertainty of the oil temperature prediction model of the top layer of transformer, and obtain a more accurate and reliable prediction interval of the top oil temperature. The uncertainty of the prediction results of the top oil temperature interval using the kernel extreme learning machine algorithm is less than that of the BP neural network and the ultimate learning machine, which is similar to the interval prediction model using the support vector machine algorithm, but the calculation speed is obviously better than that of the support vector machine. Compared with the traditional top-layer oil temperature point prediction method, the proposed interval prediction method can provide more reasonable and sufficient auxiliary basis for transformer thermal state estimation and safe operation.
【作者单位】： 山东大学电气工程学院;
【基金】：山东省科技发展计划项目(2014GGH204002) 国家电网公司科技项目(SGTYHT/15-JS-191)~~
【分类号】：TM41
[Abstract]:In order to accurately estimate the thermal state of the transformer, a prediction model of the top oil temperature interval of the transformer based on the kernel limit learning machine and the Bootstrap method is proposed. First, L group training samples are obtained by Bootstrap sampling, then L kernel extreme learning machine models are trained to predict the top oil temperature, then a kernel extreme learning machine model is trained to estimate the noise variance of the top layer oil temperature observation. Finally, the prediction interval of the top oil temperature at a certain confidence level is estimated according to the results of the L 1 nuclear extreme learning machine model. The simulation results show that this method can take into account the uncertainty of the oil temperature prediction model of the top layer of transformer, and obtain a more accurate and reliable prediction interval of the top oil temperature. The uncertainty of the prediction results of the top oil temperature interval using the kernel extreme learning machine algorithm is less than that of the BP neural network and the ultimate learning machine, which is similar to the interval prediction model using the support vector machine algorithm, but the calculation speed is obviously better than that of the support vector machine. Compared with the traditional top-layer oil temperature point prediction method, the proposed interval prediction method can provide more reasonable and sufficient auxiliary basis for transformer thermal state estimation and safe operation.
【作者单位】： 山东大学电气工程学院;
【基金】：山东省科技发展计划项目(2014GGH204002) 国家电网公司科技项目(SGTYHT/15-JS-191)~~
【分类号】：TM41
【相似文献】
相关期刊论文 前2条
1 陈霞;董朝阳;王殿辉;;Bootstrap方法在基于ARMA-GARCH模型的电力市场价格区间预测中的应用(英文)[J];电力科学与技术学报;2008年03期
2 王焱;汪震;黄民翔;蔡祯祺;杨o踥，
本文编号：2291666
本文链接：https://www.wllwen.com/kejilunwen/dianlidianqilunwen/2291666.html
