CONES: This git repository aims to couple the CFD software OpenFOAM with any other kind of open-source codes. It is currently employed to carry out sequential Data Assimilation techniques and, more specifically, an Ensemble Kalman Filter (EnKF). The communications between the EnKF and OpenFOAM are performed by a coupler called CWIPI.
6
stars
1
forks
source link
About how to include the uncertainty of the measurements #8
Hello, I opened this issue to discuss how the uncertainty $\sigma_m$ of the measurements is carried out now and how it should be improved in the future. Now there is a variable that is called "typeInputs", which has three possible values:
0: all observations of the same nature ($u_x, u_y, u_z, p, C_f$) have a common $\sigma_m$. This is because the uncertainty matrix $\mathbf{R}$ is directly related by $\mathbf{R} = \sigma_m \mathbf{I}$. At the present time, this is the recommended input in case of heterogeneous observations.
1: inputs are expressed as a percentage, which means that $\mathbf{R}_k = \sigma_m \mathbf{y}_k \mathbf{I}$. It is very convenient in case of homogeneous observation, but it does not work with very small quantities since coefficients of $\mathbf{R}_k$ become very small, and the following inversion to calculate the Kalman gain $\mathbf{K}$ crashes.
2: uncertainties are expressed by means of a potential function $\sigma_m = a + b y^c$. (By the way, we need to define the direction $y$ of the potential function as an input, it is not currently done).
Do you have any ideas about how we should improve these inputs? In particular, I would like to employ percentages with heterogeneous data as default.
Hello, I opened this issue to discuss how the uncertainty $\sigma_m$ of the measurements is carried out now and how it should be improved in the future. Now there is a variable that is called "typeInputs", which has three possible values:
Do you have any ideas about how we should improve these inputs? In particular, I would like to employ percentages with heterogeneous data as default.