Font Size: a A A

Research On Chaotic Time Series Prediction And Weak Target Signal Detection Methods

Posted on:2008-09-05Degree:MasterType:Thesis
Country:ChinaCandidate:M LiFull Text:PDF
GTID:2178360242964908Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
The development history and its importance of chaos exploration are reviewed,also, the definitions and charactions of chaos are summarized. Moreover, some kinds of representative strategies of chaotic systems and the methods of chaotic time series prediction and weak targets signal detection are introduced in detail. The significance of research on the thesis is presented as well.The theory and methods of phase-space reconstruction are introduced in detail, and the methods of phase-space reconstruction are given. In those methods of estimating delay timeτhave autocorrelation function, multiple autocorrelation and mutual information. The method of estimating embedding dimension m is Cao, the method of estimatingτand m at one time is C-C. The C-C method is used to reconstruct phase space of familiar chaotic systems. The results have shown that this method can better reconstruct phase space of chaotic time series.The Largest lyapunov exponents of time series are calculated and estimated whether the series have the character of chaos by the theory of lyapunov exponents, applying small data sets to calculate the largest Lyapunov exponent from the time series, then, the chaotic time series is predicted by the disciplinarian of data series based on the theory of lyapunov exponents. The prediction method not only doesn't structure model but also has high precision and strong reliability. Results show that the method can calculate accurate lyapunov exponents and can better predict chaotic time series in short time.Volterra adaptive filter is used to predict chaotic time series based on the phase-space reconstruction of delay-coordinate embedding of dynamic system. Some kinds of low-dimensional chaotic time series are predicted by using second-order Volterra adaptive filter. An adaptive high-order nonlinear Fourier infrared filter is proposed to make prediction of high-dimensional chaotic time series. The experiments show that Volterra adaptive filter can accurately predict multidimensional chaotic time series when the length of Volterra filter is long enough.By incorporating modified genetic algorithm with the neural network, a novel hybrid genetic neural network method for predicting chaotic time series based on the theory of phase-space reconstruction is presented. The chaotic time series is reconstructed by using multiple correlation and Cao methods, and the modified genetic algorithm is used to optimize the structure, the initial weights and thresholds of neural network, then neural network is trained to search for the optimal solution. The availability of this algorithm is proved by predicting chaotic time series, and the precision of this algorithm compared with those of BP and RBF algorithms. The computer simulations have shown that the nonlinear fitting and precision of this algorithm are better than those of BP and RBF algorithms.A method of target detection with differential evolution RBF neural network model based on the theory of chaos and technology of phase-space reconstruction is proposed. The central vectors, radical widths and output connection weights of RBF neural network are optimized by differential evolution (DE) algorithm. This model is used to predict chaotic time series and combined with support vector machine (SVM) classifiers to detect target signal, then comparing with experimental simulation of simple RBF neural network and ordinary differential evolution algorithm. Finally, the effectiveness of the propose scheme is demonstrated by the computer simulation.Based on the theory of phase-space reconstruction and fuzzy, a method of chaos time series prediction and target detection with T-S fuzzy clustering model is proposed. Chaotic time series is predicted and weak target signal is detected by fuzzy clustering model, and the target signal is decided by adaptive threshold decision. In T-S fuzzy model the Premise and conclusion are identified separately, since simplify the steps of the identification, and then improved the generalization ability, also resolved the system problem of the complicated degree exaltation but the rule few aggrandizement. This scheme is proved by comparing with RBF network on detecting weak signal. Finally, the effectiveness of the propose scheme is demonstrated by the computer simulation.A novel neural network algorithm based on hybrid particle swarm optimization (HPSO) with adaptive mutation is presented. The RBF neural network (RBFNN) is optimized by PSO based on adaptive population mutation and individual annealing operation. The adaptive mutation and annealing operation are used to adjust and optimize the population, so the global convergence ability of PSO is improved; the parameters and structures of RBF are optimized. The algorithm is used to predict chaos time series and detect weak target signal in the background of chaos. Finally, the simulation shows that the algorithm has powerful nonlinear prediction ability and can better detect weak target signal.
Keywords/Search Tags:Chaotic time series prediction, Largest lyapunov exponent, Volterra adaptive filter, Genetic algorithm, Differential evolution algorithm, Fuzzy clustering, Particle swarm optimization, Weak target signal detection
PDF Full Text Request
Related items