电磁发射系统监测量预测方法

The Prediction Method of Monitoring Quantities of Electromagnetic Emission System

对设备监测量的数值预测是进行故障预测与健康管理(PHM)研究的重要环节之一。以电磁发射系统中分段供电直线电机的定子温度为例,分别基于自回归积分滑动平均(ARIMA)模型、卡尔曼滤波模型、反向传播(BP)神经网络模型和一种新的以工况信息为外部输入的非线性自回归神经网络(NARX)模型,实现了对定子温度多时间尺度的预测。ARIMA模型为其他三种模型提供了时序数据分析时确定阶数的依据。在不同于训练数据集的试验数据上应用四种预测模型,比较和分析了四种方法得到的多时间尺度预测结果:对于不超过1min的短时温度预测,四种方法都具有较好的效果;对于1~4min的中长时间预测,引入工况信息的NARX神经网络方法具有优势。四种方法对分段供电直线电机定子温度预测都不具有超过4min的预测能力。

The numerical prediction of the main state of the equipment is one of the most significant parts of prognostic and health management(PHM)study.Taken the stator temperature of a segment-powered linear motor in an electromagnetic launch system as an example,the multiple time scales prediction of stator temperature is achieved respectively based on the autoregressive integrated moving average(ARIMA)model,Kalman filtering model,back propagation(BP)neural network model and a new nonlinear autoregressive neural network(NARX)model with external input of working conditions.The ARIMA model provides the basis for determining the number of orders in time series data analysis for the other three models.With the test data different from training data,the four methods are compared and the multiple time scales prediction results are obtained.For short-term temperature predictions up to 1 minute,all four methods have better effects;for moderate to long-term predictions from 1 minute to 4 minutes,the NARX neural network method with the working condition information has advantages;all four methods do not have the predictive ability of more than 4 minutes for the stator temperature prediction of the segment-powered linear motor.