scatter search approach for routing home care nurses
DESCRIPTION
Scatter Search Approach for Routing Home Care Nurses. by B. Elbenani , University of Mohamed V, Rabat, Marocco J.A. Ferland , University of Montreal, Canada V. Gascon , University of Quebec at Three-Rivers, Canada. Outline Presentation. Problem Formulation Solution approach - PowerPoint PPT PresentationTRANSCRIPT
Scatter Search Approachfor
Routing Home Care Nurses
by
B. Elbenani, University of Mohamed V, Rabat, Marocco
J.A. Ferland, University of Montreal, Canada
V. Gascon, University of Quebec at Three-Rivers, Canada
Faculty of Information Technology
University of Science
Vietnam National University of Ho Chi Minh City
March 2010
Outline Presentation
Problem
Formulation
Solution approach
Numerical results
Problem
nurses
Sector:patients
Patient: Visit time windows Duration (treatment)
Blood sample or not
Continuity of care
Nurse: regular
on call
Formulation
First set of constraints: VRPTW
Nurse Vehicule
Patient Customer
Usual formulation of VRPTW to specify a route for
each nurse over the network of the area where the
objective is to reduce the total travel cost
Formulation
Second set of constraints: Blood sample const raints
Blood sample limited time before the nurse
returned to the clinic
10h00 blood sample < 11h00
return before 11h00
blood sample < 10h00
return before 10h00
Complex constraints to be included only if the nurse
has to execute blood sample.
Formulation
salary of each nurse (labor cost)
traveling cost (from the VRPTW element)
additional cost if a patient is visited by
Components of the objectiv
a nurse
from anot
e functio
her secto
ad
n
r
:
ditional cost if a patient is not visited by
his regular nurse for the continuity of care
Current solution approach
Solution generated manually.
Solution generated for each sector individually
because the nurses and the patients belong to specific sector.
A nurse may visit patients from other sector if it is
required to complete her daily workload.
Our approach tries to comply with the sector constraint
Our solution approach
adaptive
Scatter search
memory includi
relying on an
ng pools of solu
method
tions
Scatter Search approach
Generate a set of feasible solutions for the problem.
These solutions may be improved using som
Init
e pr
ialisation
ocedure.
pA omo olng t inchese sol luding tutions, generate he best soluti a ons.
pool including
the most different from those in the first pool
Among the rest of the solutions, select a second
to explore more
extensively the feasible dom
ain.
some solutions from the pools according to some criteria.
Generate offspring solutions by (according to some operator)
different number of the selected sol
Select
reco
ution
mbining
Improve
s.
ea
Step 1
ch offspring solution according to some improving procedure.
some solutions from the pools according to some criteria.
Generate offspring solutions by (according to some operator)
different number of the selected sol
Select
reco
ution
mbining
Improve
s.
ea
Step 1
ch offspring solution according to some improving procedure.
For each offspring solution
if it is better than the worst solution in the first pool,
then it replaces the later in the first pool;
if not, and if it is more different from
Step
th
2
e
os
in the first pool
than one of the solution in the second pool, then it replaces
the later in the second pool;
otherwise, it is discarded.
If the pools have not been modified for a fixed number of iterations,
and the time limit is not reached, then repeat the to
generate a new set of feasible solutio
Step 3
initial
ns.
Otherwise,
isation
repeat if the time limit is not reStep 1 ached.
For each offspring solution
if it is better than the worst solution in the first pool,
then it replaces the later in the first pool;
if not, and if it is more different from
Step
th
2
e
os
in the first pool
than one of the solution in the second pool, then it replaces
the later in the second pool;
otherwise, it is discarded.
Initialisation
Our approach tries to comply with the sector constraint
Procedure to generate a feasible solution for the global problem
, determine a good solution by applying adapted
version of Tabu Search procedure
For each sec
s used to so
Phase
lve V
tor
1
RPTW
Our approach tries to comply with the sector constraint
Procedure to generate a feasible solution for the global problem
, determine a good solution by applying adapted
version of Tabu Search procedure
For each sec
s used to so
Phase
lve V
tor
1
RPTW
Combine the solutions for the sectors to obtain a feasible
solution for the global p r
Phase 2
oblem.
this feasible solution for the global problem:
Eliminate some nurses having small workload by
reassingning their patients to other nurses
Apply
I
a
mp
T
rove
abu
P
S
hase 3
ear o
ch t
improve the solution of
the global problem
Initialisation
Use the former procedure to generate a set of feasible solutions
for the global problem.
Generate the pool of solutions including the best solutiofirst .
Generate the second pool of solutions inc
ns
luding the most different
from those in the firs
t pool
Step 1: recombining solutions
Procedure to generate a new offspring solution for the global problem
1. For each nurse, select a route in a solution of the pools:
Select randomly one of the two pools Select a solution in the selected pools The nurse's route in the selected solution is assigned to her roulette wheel, tournament, ...
2. Repair process to generate a new solution:
Combine the nurse routes
If a patient is visited by several nurses, eliminate him in
all these routes, except one
in the first route encountered
in the best solution
in the follow up nurse's route
Assign left out patient follow up nurse
nurse from his sector
nurse from other sector
new fictitious nurse
Procedure to generate a new offspring solution for the global problem1. For each nurse, select a route in a solution of the pools:
Select randomly one of the two pools Select a solution in the selected pools The nurse's route in the selected solution is assigned to her
2. Repair process to generate a new solution:
Combine the nurse routes If a patient is visited by several nurses, eliminate him in
all these routes, except one
Assign left out patient
3. Generate an offspring solution Redifine new sector including patients visited by
the nurses of the sector in the current solution
Solve the problem for each sector using an adapted
Tabu Search procedure
The initial solution for the global problem
is generated from those for the sectors.
Apply a Tabu Search to improve this solution of
the global problem
If the pools have not been modified for a fixed number of iterations,
and the time limit is not reached, then repeat the to
generate a new set of feasible solutio
Step 3
initial
ns.
Otherwise,
isation
repeat if the time limit is not reStep 1 ached.
For each offspring solution
if it is better than the worst solution in the first pool,
then it replaces the later in the first pool;
if not, and if it is more different from
Step
th
2
e
os
in the first pool
than one of the solution in the second pool, then it replaces
the later in the second pool;
otherwise, it is discarded.
Numerical Results
Numerical results
Numerical results for parameters with different problems
Problem Mean cost Mean # nurses
Mean # patients without
follow-up
Mean time(sec.)
Day 1 Our sol. 9050.6 12.0 0.8 589.1
CLSC 9623.0 13.0 0.0 -
Day 2 Our sol. 9545.0 14.0 1.0 651.7
CLSC 10673.0 16.0 0.0 -
Day 3 Our sol. 9998.2 13.0 0.0 408.5
CLSC 11413.0 15.0 0.0 -
Problem # of patients
# of regular nurses
# of nurses from the recall list
% of visits with blood
sample
# patients with follow-
up
Day 1 84 9 4 34.5 34
Day 2 92 13 3 45.6 38
Day 3 82 9 6 23.1 34
Real Data
Reducing the number of nurses
Mean number of patient
without follow up
Contraintes de prises de sang
1 P i, xIk Vj
kij
I, k, NV\i, xxVj
kji
Vj
kij
100
Ik,xVj
kj
10
I, kxxVj
kj
Vj
kjN
01
IkA,(i,j),xMbtrb kij
kjiji
ki 1
IkVi, f be ikii ,
IV, ki,j, ,xkij 10
IkVi, bki ,entieret 0
s s
ks kij ij
s S k I (i, j) A
Min c x
Sujet à
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(15)
(16)
Contraintes de prises de sang
1 P i, xIk Vj
kij
I, k, NV\i, xxVj
kji
Vj
kij
100
Ik,xVj
kj
10
I, kxxVj
kj
Vj
kjN
01
IkA,(i,j),xMbtrb kij
kjiji
ki 1
IkVi, f be ikii ,
IV, ki,j, ,xkij 10
IkVi, bki ,entieret 0
s s
ks kij ij
s S k I (i, j) A
Min c x
Sujet à
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(15)
(16)
Contraintes de prises de sang
1 P i, xIk Vj
kij
I, k, NV\i, xxVj
kji
Vj
kij
100
Ik,xVj
kj
10
I, kxxVj
kj
Vj
kjN
01
IkA,(i,j),xMbtrb kij
kjiji
ki 1
IkVi, f be ikii ,
IV, ki,j, ,xkij 10
IkVi, bki ,entieret 0
s s s s
ks k kij ij j
s S k I (i, j) A s S k I j P
Min c x y
Sujet à
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(15)
(16)
I, kPi, xxMHbbpPj
kij
Pj
kjp
kki
ss
11
11110,min1 10 (9)&(10)
I, kPi, xxMbHpPj
kij
Pj
kjp
ki
ss
10
101101 (8)
IkPixysVj
kji
ki
,,
Ik,Pi,yMbHsPj
kj
ki
10
Ik, yMHMbPk
kj
k
1010
I, kPi yMbb ki
kki ,110
IkPi, y ki ,1,0
(11)
(12)
(13)
(14)
(17)
Contraintes de prises de sang
1 P i, xIk Vj
kij
I, k, NV\i, xxVj
kji
Vj
kij
100
Ik,xVj
kj
10
I, kxxVj
kj
Vj
kjN
01
IkA,(i,j),xMbtrb kij
kjiji
ki 1
IkVi, f be ikii ,
IV, ki,j, ,xkij 10
IkVi, bki ,entieret 0
s s s s
ks k kij ij j
s S k I (i, j) A s S k I j P
Min c x y
Sujet à
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(15)
(16)
I, kPi, xxMHbbpPj
kij
Pj
kjp
kki
ss
11
11110,min1 10 (9)&(10)
I, kPi, xxMbHpPj
kij
Pj
kjp
ki
ss
10
101101 (8)
IkPixysVj
kji
ki
,,
Ik,Pi,yMbHsPj
kj
ki
10
Ik, yMHMbPk
kj
k
1010
I, kPi yMbb ki
kki ,110
IkPi, y ki ,1,0
(11)
(12)
(13)
(14)
(17)
Contraintes de prises de sang
1 P i, xIk Vj
kij
I, k, NV\i, xxVj
kji
Vj
kij
100
Ik,xVj
kj
10
I, kxxVj
kj
Vj
kjN
01
IkA,(i,j),xMbtrb kij
kjiji
ki 1
IkVi, f be ikii ,
IV, ki,j, ,xkij 10
IkVi, bki ,entieret 0
s s s s
ks k kij ij j
s S k I (i, j) A s S k I j P
Min x y c
Sujet à
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(15)
(16)
I, kPi, xxMHbbpPj
kij
Pj
kjp
kki
ss
11
11110,min1 10 (9)&(10)
I, kPi, xxMbHpPj
kij
Pj
kjp
ki
ss
10
101101 (8)
IkPixysVj
kji
ki
,,
Ik,Pi,yMbHsPj
kj
ki
10
Ik, yMHMbPk
kj
k
1010
I, kPi yMbb ki
kki ,110
IkPi, y ki ,1,0
(11)
(12)
(13)
(14)
(17)