基于Elman神经网络的装备状态组合预测方法

Combination Forecasting Methods for Equipment Technical Condition Based on Elman Neural Network

针对单一预测模型在利用多维状态特征信息进行状态预测时效果常常不够理想的情况,提出以灰色理论等模型作为单项预测模型,运用Elman神经网络进行变权组合预测的建模方法;考虑神经网络容易因过拟合导致预测时泛化能力变差的问题,运用遗传算法对神经网络隐层节点数和训练误差阈值进行优化求解,建立了完整的基于Elman神经网络的组合预测建模方法;最后,通过案例分析验证了该预测方法的有效性,结果表明组合预测能够将三步以内的预测相对误差控制在10%以内,大大优于定权组合预测模型。

According to the problem that unique forecasting model usually can＇ t perform well as expected when the forecasting outcome of equipment condition relies on multi--dimensional characteristic parameters, a combination of forecasting output mode is taken as the combination forecasting method of condition. Considering the fact that static weighted combination forecasting methods often runs short in ability of generalization when dealing with complicate time series data, a variable weighted combination forecasting method based on Elman neural network is proposed. Genetic algorithm is utilized to optimize structural parameters of the network, the modebng process of forecasting mod- el is finally determined. Finally, the model is verified by study of a case.