基于组合式神经网络的转子系统状态预测模型

针对反映转子系统工作状态的特征参数时间序列具有不确定性的、差异较大的分段函数变化规律的特点，提出了一种组合式神经网络转子系统状态预测模型。该模型将故障诊断和状态预测有机地结合起来，利用转子系统当前状态特征参数样本，通过故障诊断系统判断预测时的转子系统工作状态模式；从多种神经网络预测模型组全而成的预测模型库中调用同核工作状态模式相应的神经网络预测模型，对多种特征参数时间序列进行预测；依据预测出的未来革一时刻的多种特征参数，利用故障诊断系统判断转子系统的未来工作状态模式。仿真试验结果表明，该模型可以对转子系统状态进行可靠的预测。文中详细讨论了模型的建立和仿真实验结果。

A ROTOR STATE FORECASTINGMODEL BASED ON MODULAR NEURAL NETWORKS

According to the characteristics of the rotor state feature data time series, this paper presents a novel rotor state forecasting model based on modular neural networks.The forecasting and the fault diagnosis are integrated in the model.With the feature data sample,the fault diagnosis subsystem is first applied to diagnose the rotor state.In accordance with the rotor state,the corresponding forecasting modules based on neural networks are activated to forecast the different feature data time series.In the end,the fault diagnosis subsystem is further applied to diagnose the future rotor state with the forecasted feature data sample.The simulation results show that this model improves greatly the reliability to the rotor state forecasting.The structure of the model and simulation results are discussed in detail.