论文部分内容阅读
针对卫星姿控系统时间序列故障预测问题,给出了BP神经网络和小波神经网络、小波分解-LSTM网络相结合的故障预测方法.利用卫星正常运行时的数据训练BP神经网络,将其作为系统的标准模型,对卫星实时输出和标准模型输出之间的残差建立小波神经网络和小波分解-LSTM故障预测模型,并进行仿真对比分析.结果 表明,2种方法均能准确的对故障进行预测,由于基于小波分解-LSTM的方法对残差序列进行了多级小波分解,还利用了LSTM网络能选择性保留输入数据的特点,因此预测更准确,性能更优.“,”A new method based on BP neural network(BPNN),wavelet neural network(WNN) and wavelet decomposition-LSTM(wLSTM) network is proposed for predicting faults in the satellite attitude control system.Normal satellite attitude data is used to train BPNN which is used as the standard model of satellite attitude control system.The real-time attitude residuals is obtained by subtracting the BPNN output attitude angle from the real-time data of satellite attitude.The time series of the residuals are used to build WNN and wLSTM models to predict the faults of satellite attitude control system.A conclusion is given according to comparing the WNN and wLSTM that both the fault prediction methods can precisely predict the fault and the wLSTM model predicts more accurately because LSTM network can selectively retain the characteristics of input data.At the same time,it also provides a novel method for the prediction of complex system.