基于非负矩阵分解与改进极端学习机的变压器油中溶解气体浓度预测模型

Concentration Prediction Model of Dissolved Gases in Transformer Oil Based on NMF and Improved ELM

变压器油中溶解气体浓度是评估变压器绝缘状态的重要依据,对气体浓度进行有效预测,可以及时识别变压器潜伏性故障。文中提出一种基于非负矩阵分解(nonnegative matrix factorization,NMF)与改进极端学习机(extreme learning machine,ELM)组合的变压器中溶解气体浓度预测模型。该方法通过NMF算法对输入样本进行分解,同时引入Adaboost算法对极端学习机进行改进;将低维矩阵作为模型的训练样本输入,剔除冗余数据,提高预测精度。实例分析结果表明,文中提出的方法能有效地降低输入样本维数,提高预测精度,能较好地解决变压器油中溶解气体浓度预测问题。

The concentration of dissolved gases in transformer oil is an important parameter to evaluate insulation state of oil-immersed transformer. Effective concentration prediction of the dissolved gases can help to identify latent faults of transformer in time. In this paper, a concentration prediction model of dissolved gases in transformer oil is proposed based on the nonnegative matrix factorization （NMF） and the improved extreme learning machine （ELM）. The input data is decomposed by using the NMF algorithm to reduce the dimension of input variables. Then Adaboost algorithm is introduced to improve the extreme learning machine, and the derived nonnegative lower-dimension mapping matrix is taken as the inputs of the model for training. Simulation illustrates the effectiveness and availability of the proposed method in reducing the dimension of the input variables, performing the concentration prediction of dissolved gases in transformer oil, and improving the prediction accuracy.