基于数据挖掘算法的变压器热点温度时序预测方法

为了解决油浸式电力变压器热点温度预测方法缺乏对短期热点温度走势的预测，无法满足动态增容决策要求的问题，以SFPSZ-180000/220型变压器为研究对象，首先，研究对比发现变压器的热点温度与负载率相关性最大，在此基础上，建立了基于支持向量回归的局部地区负荷预测模型，为变压器热点温度预测提供数据；其次，提出了基于数据挖掘算法的变压器热点温度时序预测方法，并在此基础上分别建立了支持向量回归、BP神经网络、决策树3种数据挖掘预测模型；最后，对一般输入-输出的建模方法的预测结果与基于时间延迟方法的预测结果,以及不同时间延迟下3种数据挖掘模型的预测结果进行了对比分析。结果表明，有外在输入的支持向量回归预测模型结果比BP神经网络和决策树吻合度更高，并且时间延时更小，预测结果精确度更高。有外在输入的支持向量回归预测模型在很大程度上提高了变压器热点温度预测精度，其预测结果可为变压器的动态增容决策提供有效参考。

Prediction method on time series of transformer hot-spot temperature based on data mining algorithm

In order to solve the problem that prediction method of hot-spot temperature of oil-immersed power transformer lacks the prediction of short-term hot-spot temperature trends, and can not meet the requirements of dynamic capacity-increase decision-making, the SFPSZ-180000/220 transformer was taken as the research object. Firstly, by comparison, the hot-spot temperature and the load rate of the transformer showed the greatest correlation, on this basis, a local area load forecasting model based on support vector regression was established to provide related data for predicting the transformer hot spot temperature. Secondly, the prediction method on time series of transformer hot-spot temperature based on data mining algorithm was proposed, and three data mining prediction models of support vector regression, BP neural network, and decision tree were established. Finally, the prediction results of the general input-output modeling method were compared with the prediction results based on the time delay method and the prediction results of the three data mining models under different time delays. The results show that the prediction results of the support vector regression model with external input has the better prediction consistency, the smaller time delay and the higher accuracy comparing to that of BP neural network and decision tree. The support vector regression prediction model with external input greatly improves the prediction accuracy of transformer hot-spot temperature, and the results can provide an effective reference for the dynamic capacity increase decision-making .