旋转机组烈度趋势预测技术研究

The research on vibration intensity trend predicting technology of rotary machine

组合预报在时序预测中应用广泛 ,而传统线性、非线性组合预报模型对各单项预测方法权系数的计算过于繁琐 ,并且多为定权系数 ,现有的神经网络组合预报模型又存在着对不同预测方法的特点强调不足的缺陷 ,针对以上两点 ,提出一种新的神经网络架构———变权重神经网络架构。这种新型神经网络架构利用了神经网络的非线性映射能力 ,避免了求解组合权系数的繁琐过程 ,又克服了对不同预测方法的适用性强调不足的缺陷。作为验证 ,给出模型在大型旋转机组烈度趋势预测中的应用实例。

Grouping predicting has wide application in the time series predicting. In the traditional linear and nonlinear grouping predicting models, the calculating process is too complicated to get the result.And the weighting modulus is usually fixed. Meanwhile the present NN based grouping predicting models have limitations in emphasizing the high precision component models.Concerning the deficiency mentioned above, the new neural network based groupig predicting model——variable weighted neural network is put forward.The new method need not calculate the modulus, and testifies predicting result in the industrial application.