Scribd
Upload
179 views

MATLAB Reconciliation

This document discusses using data reconciliation in MATLAB to estimate true values from redundant measurements. It covers linear and bilinear systems, how to formulate the problems as constrained optimization, and the analytical solutions. An example is provided for a linear system with mass balances and component flows. The objectives are to take advantage of redundant data to obtain more accurate estimates and their uncertainties.

Uploaded by

Allan Paolo
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
179 views

MATLAB Reconciliation

This document discusses using data reconciliation in MATLAB to estimate true values from redundant measurements. It covers linear and bilinear systems, how to formulate the problems as constrained optimization, and the analytical solutions. An example is provided for a linear system with mass balances and component flows. The objectives are to take advantage of redundant data to obtain more accurate estimates and their uncertainties.

Uploaded by

Allan Paolo
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

Data Reconciliation with MATLAB

... from actual experience!

Felipe Aristizbal

Department of Chemical Engineering


McGill University

Montral, Qubec
July 31, 2013
Objectives

We will cover:
The basic fundamentals of data reconciliation.
Linear and Bilinear systems.
MATLAB implementation.
Problem Description

Using mass balances, calculate the value of F6:

FI FI FI FI
101 103 105 106

1 3 5 6

Stream Name kt/h


FI FI FI-101 F1 110.5
102 104
FI-102 F2 60.8
2 4 FI-103 F3 35.0
FI-104 F4 68.9
FI-105 F5 38.6
FI-106 F6 ????
Many solutions possible. Redundant measurements.
Data Reconciliation

All measurements have errors.


Take advantage of redundant information.
Formulate as constrained optimization problem.
Analytical solution for linear and bilinear constraints.
Linear Constraints

Minimize: y ) =(y y )T V 1 (y y )
J(
Subject to: A
y =0

Where:
y : m by 1 vector of raw measurements.
y : m by 1 vector of estimates.
V : m by m covariance matrix.
A: Incidence matrix - mass balances.
Analytical Solution: Linear Data Reconciliation
Using Lagrange multipliers ....
... the estimates, y , can be calculated from:

W = I VAT (AVAT )1 A
y = Wy
y ) = WVW T
Cov(

Procedure (MATLAB)
1 Create y from measurements
2 Obtain V from measurement standard deviations (V = diag(i2 ))
3 Use mass balances to obtain A
4 Calculate W , y
5 Extra: what are estimates standard deviations,
i ?
p
i = diag(Cov(
y ))
Example: Linear DR

1 3 5 6

2 4

Stream Raw Standar Reconciled Standar


No. Measurement Deviation, Flow Deviation

(kt/h) (kt/h) (kt/h) (kt/h)
1 110.5 0.82 103.2 0.42
2 60.8 0.53 65.4 0.37
3 35.0 0.46 37.8 0.30
4 68.9 0.71 65.4 0.37
5 38.6 0.45 37.8 0.30
6 101.4 1.20 103.2 0.42
Example: Bilinear System

F2 F2
x2,1 f2,1
x2,2 f2,2
Transform to
F1 component flow F1
x1,1 F3 fij = Fi xij f1,1 F3
x1,2 x3,1 f1,2 f3,1
x3,2 f3,2

Non-Linear Constraints Linear Constraints

F1 x1,1 F2 x2,1 F3 x3,1 = 0 f1,1 f2,1 f3,1 = 0


x1,1 + x1,2 = 1 F1 f1,1 f1,2 = 0
... ...
Example: Bilinear System (cont.)
Approximate the variance of the new variables,

V (fij ) xij2 V (Fi ) + Fi2 V (xij )

Variable Raw Standar Reconciled


Name Measurement Deviation, Measurement
F1 1095.5 54.8 1010.2
x1,1 0.4822 0.0048 0.4808
x1,2 0.5170 0.0052 0.5192
F2 478.4 23.9 500.1
x2,1 0.9410 0.0094 0.9518
x2,2 0.0501 0.0005 0.0480
F3 488.2 24.4 510.1
x3,1 0.0197 0.0002 0.0190
x3,2 0.9748 0.0097 0.9810

You might also like