This paper combines a variational denoising approach with a dimensionality reduction of multivariate time series by dynamical component analysis (DyCA), a dimensionality reduction method for high-dimensional dynamical signals governed by a low-dimensional system of ordinary differential equations (ODEs). While DyCA has been successfully applied to these types of signals in the past, it has not been robust to noise or incomplete data. The proposed approach simultaneously denoises the given time series and reduces the dimensionality of the data using DyCA while retaining the most important dynamic structures.