Research On Optimization Of RBF Neural Network By Improved QPSO Algorith And Its Application In Deformation Prediction

Formation monitoring plays an important rolein and after construction link.The goal of formation monitoring is analyzing the deformation tendency by using the dataso that we can predict the deformation trend reasonably to ensure the safety of project implementation and avoid economic property damage and the threat of people's safety.Therefore,predicting the deformation trend is more and more important.Radial basis function neural network(RBFNN)is a kind of simulated operation model of the human brain neurons.Through the input samole data,RBFNN can adjust the weights of connections between neurons and the parameters continously to appropriate model to forecast analysis.With strong nonlinear approximation ability,RBFNN can have perfect effect on the nonlinear time series of prediction of deformation.Due to the hidden layer is the most crucial part of RBFNN and the the parameters are difficult to determine,we must try to find anther ways to optimize the parameters of RBFNN to exploit the advantages to the full.Intelligent optimization algorithm is a kind of technology forsolving the optimization problem of application.Among intelligent optimization algorithm,particleswarm optimization(PSO)algorithm is a classical swarm intelligence algorithm,with the advantages of simple structure,easy to describe and implementand with good global search capability,and so on,which is widely used in manyfields.But the standard PSO algorithm also has some shortcomings such as prematureconvergence and poor local search capability.So this thesis presents an improved PSO algorithmon the basis ofquantum particle swarm optimization(QPSO)propoesd by predecessorscalled WG-QPSOalgorithm which implied to the ANN.Finally the model is applied to predict the deformation trend.The main research content are as follows:(1)Analyze currently commondeformation trend predicting model.On the basis of the analysis,summarize the advantages and disadvantages of them.Then focus on studying the theory of RBF NN and explore the reason of defects,and accordingly improve parameters by using one of intelligentalgorithms——PSO algorithm.(2)By studying and analyzing PSOalgorithm,we have found that it will easily prematureconvergence which weaken its performance.In order to improve its performance,we bring QPSO algorithm into it.(3)AlthoughQPSO algorithm is better than PSO algorithm in performance,it'll still get into the local optimum and can not get out of it.Therefore it was presented an QPSO algorithm combined with mixed-probability and inertia weightmutationoperator which we named it WG-QPSO.Inertia weightmutationoperator is mainly used to make QPSOalgorithm have the ability of getting out of the local optimum,so that the algorithm can search of the optimum solution in other areas.(4)Analyze standard PSO algorithm and WG-QPSO algorithm combined with RBF NN.So that we can get a complete model by studying the feasibility and making up combination programs.And then with different parameters(5)According to the deformation monitoring examples,choose different kinds of PSO algorithm combined with RBFNN to predict deformation trend by Matlab in Windows 7 system.First of all,determine the topological structure through experiments.And then compare the results of PSO algorithm with different parameter setting in order to get the best parameter setting.With the best parameter setting,compare and analyze of the predicted results to get the applicability and advantages of the algorithm we purposed.