Non-linear Multivariate Dynamics Modelled by PLSR
Information-rich high-speed, low-cost multi-channel measurements will increasingly become available for characterizing and quantifying chemical constituents and physical phenomena in e.g. biology. The question is how to parameterize, quantify and display and interpret this information. The paper discusses how bio-scientists can make use of PLS-based regression methods to explore multivariate nonlinear dynamics of complex systems. A novel application of nominal-level PLSR modelling in the time domain is presented. It allows pragmatic modelling of empirical time series data in order to search for plausible mechanistic explanations of dynamics in terms of systems of local linear differential equation. A recently developed approach to develop non-linear mechanistic models is then employed in order to merge the PLS regression coefficients from the nominal-level local linear modelling into a global non-linear differential equation system. The approach is demonstrated on a simulated data set
PLS regression time series non-linear dynamics Jacobian estimation
Harald Martens
The Norwegian Food Research Institute,The Norwegian U. of Life Sciences/IMT, Centre for integrative genetics (CIGENE), 1430 Aas, Norway, U. of Copenhagen (LIFE)
国际会议
The 6th International Conference on Partial Least Squares and Related Methods(第六届偏最小二乘及相关方法国际会议)
北京
英文
139-144
2009-09-04（万方平台首次上网日期，不代表论文的发表时间）