ZHANG Ruoyu 1SHI Zheng 1GUO Jian 1GONG Xin 1LEI Min1
作者信息
- 1. State Grid Shandong Electric Power Company Qingdao Power Supply Company,Qingdao 266000,China
- 折叠
摘要
Abstract
Transformer top oil temperature can be used to measure the thermal characteristics of the transformer,and is an important parameter for assessing transformer condition.Due to the on-site complex working environment,the on-line monitoring data of top oil temperature often have a large number of missing values,which seriously affects the prediction of top oil temperature.Meanwhile,affected by many factors such as load,environmental temperature,environmental wind speed,solar radiation,etc.,the information of different time scales are aliased within the monitoring data,making it difficult to achieve the desired accuracy by single prediction model.This paper proposes a method for predicting the top oil temperature of transformer based on data quality improvements.First,supplement the top oil temperature data of transformer containing missing values by the Gray-Markov model to obtain a complete and continuous time series.Then,ensemble empirical mode decomposition is used to decompose the transformer top oil temperature time series into multiple time series components to eliminate the interaction between different time scale information and reduce the prediction difficulty.Finally,extreme learning machine(ELM)sub-prediction models are established to predict the components,and the prediction results of all sub-models are summed to obtain the final prediction results for the transformer top oil temperature.Through field data verification,the proposed method was used to predict the transformer top oil temperature in the next two days.Mean absolute percentage error(MAPE)is 5.27%,root mean square error(RMSE)is 2.459 2,correlation coefficient(CC)is 0.832 6,and coefficient of determination(CD)is 0.682 9.Compared with back propagation(BP)neural network model,support vector machines(SVM)model and long short term memory network(LSTM)model,MAPE is respectively reduced by 69.56%,61.92%and 43.45%,and RMSE is respectively reduced by 67.02%,59.87%and 36.58%.Compared with LSTM model,CC and CD are respectively improved 289.61%and 559.1%.关键词
变压器/顶层油温/马尔可夫过程/集合经验模态分解/极限学习机Key words
transformer/top oil temperature/Markov process/ensemble empirical mode decomposition/extreme learning machine分类
信息技术与安全科学
