Research And Implementation Of High-dimensional State Estimation Algorithm For Time Series Data Based On Manifold Learning

The state estimation of high-dimensional time series data has broad applicationprospects in computer animation, financial information management, medical andbiological engineering etc, as the complexity of the state estimation algorithm willincrease exponentially as the number of dimensions increases, and the state system isnonlinear in most applications, so compared with the low-dimensional non-sequentialdata,high-dimensional time series data is more difficult to estimate the state, manifoldlearning as a non-linear method can extract the low-dimensional manifold embedded inhigh-dimensional space, so the state estimation of high-dimensional time series databased on manifold learning become a hot research of many scholars.Since the complexity and the unknown configuration of the high-dimensional timeseries data, make it is difficult to intuitive grasp and analysis in the application. in theparticle filter, when a large number of samples and the dimension is too high, the numberof particles increases with increasing number of dimensions which make particlecollection update becomes difficult, finally resulting in low efficiency of the algorithmand decline the accuracy of state estimation.This article studies local linear embedding,isometric mapping, laplacian Eigenmapand the gaussian process latent variable model algorithms, using them to reduct thedimension of high-dimensional time series data, by comparing the reconstruction errors,ultimately select the Gaussian process latent variable model as the dimensionalityreduction algorithm of high-dimensional time-series data. in this article, firstly, usingGaussian process latent variable model algorithm modeling train set generates motionempirical model, and then construct latent variable motion model, finally using latentvariable motion model to improve the particle filter algorithm. experiments show thatafter using the latent variable motion model to improve the particle filter algorithm thereal-time feature and accuracy of the state estimation has significantly improved.