Study On State Prediction Methods For Spacecraft Propulsion System Pipeline Based On Process Neural Network Ensemble

Some key technologies including pipeline fault simulation,pipeline state prediction for a certain type of satellite propulsion system were studied in the thesis.The methods of pipeline state prediction based on process neural network ensemble were designed and implemented.The DFH satellite propulsion system was taken as an example to analyze the main failure modes and effects of the pipeline of spacecraft propulsion system in the thesis.And based on the modular modeling idea,a simulation model of a certain type of satellite propulsion system was established.Based on the simulation model,six kinds of typical pipeline fault effects were simulated respectively.Then,three single-process neural networks were established to predict the pipeline state based on the fault data of the numerical simulation.At the same time,the model of process neural network ensemble was established and three different conclusion synthesis methods were adopted for the deficiency of single process neural network prediction.Finally,the factors influencing the accuracy of network integration prediction were analyzed from the perspective of improving network integration generalization ability and the process neural network ensemble based on the dynamic weight conclusion method was verified.The research results show that the process neural network ensemble based on the dynamic weight conclusion method is the better and the more accurate for pipeline state prediction of spacecraft propulsion system,and it can provide an effective way for pipeline fault detection of spacecraft propulsion system.