基于随机森林的变压器油中溶解气体浓度预测

对油中溶解气体浓度进行分析及发展趋势预测,可以为变压器的状态评估提供重要的依据。传统的离线DGA方法因易导致延迟判断变压器的运行状态,造成一定的经济损失,现已不适用于油中溶解气体浓度分析及预测。因此,提出一种基于随机森林的变压器油中溶解气体浓度预测模型,以更准确地分析与预测油中溶解气体浓度。该模型以7种气体浓度构成特征向量空间,作为可视输入,并以目标气体浓度作为输出。试验结果表明,相较于传统的机器学习方法(BPNN、RBF和SVM),随机森林模型能更准确地预测油中溶解气体浓度,且需要调整参数少、训练效率高。通过算例分析,验证了该方法的有效性。

Concentration prediction of dissolved gases in transformer oil based on random forest

The analysis and prediction of the trend the dissolved gas will have in oil can provide crucial basis for estimating the running status of the transformer.The traditional off-line DGA excruciatingly leads to deferred estimation of the running status of the transformer and hence the financial loss will not be prevented.At present,the traditional off-line DGA does not apply in the interest.Therefore,a random forest(RF)model is first suggested for more accurately analyzing and predicting the trend of the dissolved gas in oil.Seven types of gas build the feature vector space as the input of the model,and hence the output is the objective dissolved gas in oil.Experiments show that the RF model is a more accurate approach to predict the dissolved gas by comparison with the traditional machine learning methods(BPNN,RBF and SVM),the RF model is highly efficient in training and need to adjust less parameters.As a result,case analysis verifies effectiveness of the proposed model.