Font Size: a A A

Multivariable Nonlinear Time Series Modeling And Forecasting Based On Particle Swarm Optimization

Posted on:2015-05-15Degree:MasterType:Thesis
Country:ChinaCandidate:X J SongFull Text:PDF
GTID:2298330422482423Subject:Probability theory and mathematical statistics
Abstract/Summary:PDF Full Text Request
In recent years, nonlinear time series is in rapid development time. Among them,univariate or multivariate nonlinear time series are obtained from complex dynamical systems.In practical problems, a complex system is often described using a number of state variables,therefore, multivariate nonlinear time series have a major significance.At present, domestic and overseas scholars mainly research on multivariate chaotic timeseries. The phase space reconstruction is the basis of its analysis and forecast, and the corepart of the phase space reconstruction is the determination of time-delays and embeddingdimensions. Although many scholars have proposed a variety of methods to determine theparameters, there are still some limitations. Firstly, the most methods analyze the dependencerelation of the variables using the statistical theory, remove redundant variables and use therest of chaotic time series as the reconstruction variables; secondly, they respectivelydetermine the time-delays and embedding dimensions, which ignores their correlation; finally,they reconstruct the phase-space using obtained parameters to predict the sequences. In thesame away, these parameters ignore the correlation with the forecasting models, which maylead to unsatisfactory predicting results. Although, many scholars subsequently propose thenew methods to determine the time-delays and embedding dimensions at the same time, theystill don’t combine the forecasting models to determine the parameters.To solve the above problems, this paper proposes a multivariable nonlinear time seriesanalysis method based on binary particle swarm optimization-radial basis function neuralnetwork. The core part of this method is using the binary particle swarm optimizationalgorithm combined with the radial basis function neural network prediction method todetermine the parameters of the phase space reconstruction, then we use the optimalparameters to reconstruct the phase space and build model. Then this method is used tosimulate several typical multivariate chaotic time series, which is compared with thetraditional parameter determination method under the same conditions. simulation resultsshow that the new method has better prediction accuracy, and response the feasibility andpracticability of this method intuitionally.
Keywords/Search Tags:multivariate nonlinear time series, binary particle swarm optimization, radialbasis function neural network, time-delay, embedding dimension
PDF Full Text Request
Related items