A full-order observer estimates all states whether it is measured by a sensor or not.
It leads the increase of calculation cost and makes a observer design difficult.
A minimal-order observer use sensor outputs as known parameters and reduces a states to be estimated.
Let
z be measurable state by a sensor and
ξ be a state to be estimated, and let us consider a system
dtd[ξz]y=[A11A21A12A22][ξz]+[B1B2]u+[H1H2]v=z+w,
where
y,
u,
v, and
w are an output, and input, process noise, and observation noise.
This representation provides a following equation
{ξ˙=A11ξ+(A12z+B1u)+H1vz˙−A22z−B2u=A21ξ+H2v.
This equation includes two estimable parameters using sensor outputs.
A12z+B1u≡Uz˙−A22z−B2u≡Y.
These parameters simplify the equation as
{ξ˙=A11ξ+U+H1vY=A21ξ+H2v.
The simplified representation shows that
U and
Y works as an input and output.
An estimated input and measured output of a reduced-order system is expressed as
U^Yˉ=An12y+Bn1u=y˙−An22y−Bn2u.
Using these parameter, a Luenberger observer can be designed as
{ξ˙^=An11ξ^+U^+L(Yˉ−Y^)Y^=An21ξ^.
Expanding the equation provides an estimation formula.
ξ˙^=An11ξ^+U^+L(Yˉ−Y^)=An11ξ^+(An12y+Bn1u)+L((y˙−An22y−Bn2u)−An21ξ^)=(An11−LAn21)ξ^+(An12y+Bn1u)+L(y˙−An22y−Bn2u)=(An11−LAn21)ξ^+(An12−LAn22)y+(Bn1−LBn2)u+Ly˙
This formula includes a derivative of
y and is difficlult to calculate.
This problem is solved by introducing a variable transformation.
Pξ˙^−Ly˙⇔P˙∴P≡ξ^−Ly=(An11−LAn21)ξ^+(An12−LAn22)y+(Bn1−LBn2)u=(An11−LAn21)ξ^−(An11−LAn21)Ly+(An11−LAn21)Ly+(An12−LAn22)y+(Bn1−LBn2)u=(An11−LAn21)(ξ^−Ly)+((An11−LAn21)L+An12−LAn22)y+(Bn1−LBn2)u=(An11−LAn21)P+((An11−LAn21)L+An12−LAn22)y+(Bn1−LBn2)u=(sI−An11+LAn21)−1{(Bn1−LBn2)u+((An11−LAn21)L+An12−LAn22)y}
As a result, an estimated value can be obtained by following formula:
ξ^=(sI−An11+LAn21)−1{(Bn1−LBn2)u+((An11−LAn21)L+An12−LAn22)y}+Ly=(sI−An11+LAn21)−1{(Bn1−LBn2)u+(sL+An12−LAn22)y}.