基于组合预测模型的变压器油中溶解气体质量浓度的预测

Forecast of Mass Concentration of Dissolved Gas in Transformer Oil Based on Combinative Forecasting Model

为了对变压器油中溶解气体质量浓度进行准确预测,提出将BP神经网络、灰色理论和线性回归预测算法进行综合,采用最优加权组合预测模型,对油中溶解气体质量浓度的发展趋势进行预测。该方法先对这3种单项预测方法根据各自的预测误差,按照预测误差平方和最小的原则计算各自的权重,然后加权综合建立最优组合预测模型,再计算出变压器油中溶解气体的质量浓度。实例分析证明该组合预测方法不仅可以有效地降低单项预测算法的预测误差,提高预测模型的预报能力,同时还增强了预测的稳健性。

For accurately forecast mass concentration of dissolved gas in transformer oil, it is proposed to integrate BP neural networks with gray theory and regression forecasting method and forecast the development trend of mass concentration of dissolved gas in transformer oil by use of optimal weighted combination forecast model. In accordance to the forecast errors, weight of each of the three single forecast methods is firstly calculated in this method on principle to maximize forecasting error square sum. Secondly, optimal combinative model is built with weighted synthesis, and finally mass concentration of dissolved gas in transformer oil is calculated. The example analysis proves that the forecast method can not only reduce forecast errors of single forecast method and improve forecast capacity of the model but also strengthen the forecast robustness.