Prediction Of The FAST Node Displacements Based On Neural Network

FAST(Five-hundred-meter Aperture Spherical Radio Telescope)is put forward by the Chinese astronomical community to build the world's largest single aperture antenna.The construction of this project has an important scientific significance in the astronomical community.Prediction of the FAST node displacements is the basic work and one of the core difficulties to implement the deformation of the whole network controll tactics of the main active reflecor and adaptive modeling research.Given this backdrop,prediction of the FAST node displacements have significant implication for theory and practice no matter in the field of industrial control or in the progress of designing and implementing the project.First,this paper summarizes the theory of neural network prediction,especially illustrates how to select the data set and train the neural network.Then,the FAST node displacements control principle has been understood further by analyzing the controll tactics of the main active reflecor and the structure of the cable-net.And the analysis of the factors influencing the accuracy of the node displacements helps to confirm the inputs and outputs of the prediction model.After that prediction of the FAST node displacements based on neural network is proposed definitely.Next,the paper explains the related theories of RBF neural network algorithm and clarifies the steps to implement the algorithm.The simulation results show the effectiveness of the RBF neural network algorithm.In addition,the paper introduces another algorithm—ELM neural network algorithm.After discussing the principle and characteristics of the ELM algorithm,the algorithm is applied to the FAST node displacements prediction model.By compared with the simulation results of the RBF algorithm,the ELM algorithm has much advantage in training time which can meet the real-time requirement of the system,but it also has some defects.Aiming at the hidden layer nodes number of ELM neural network need to be set manually,which can lead to network overfitting and affectts the algorithm's robustness and generalization ability,the improved ELM algorithm is proposed.Finally,the improved ELM algorithm is applied to the FAST node displacements prediction model.The simulation results show that the improved ELM algorithm has better prediction accuracy than the traditional algorithms and its training time can also meet the real-time requirement of the system.The main contribution of the paper is to establish the FAST node displacements prediction model based on the neural network algorithm.The results of simulation have verified the effectiveness and feasibility of the improved algorithms.It has important practical significance to the research of the entire network adjustment of FAST active reflector.