Introduction Optimization Model Solution approaches Numerical Results Conclusion
Dynamic uncapacitated lot sizing with random demand under a fillrate constraint Horst Tempelmeier and Sascha Herpers Seminar f¨ ur SCM und Produktion Universit¨ at zu K¨ oln EURO Conference 2009
Bonn, July 2009
SCMP c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
1/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Agenda 1
Introduction The Problem Solution Approaches
2
Optimization Model Formulation
3
Solution approaches Exact solution Heuristic solution
4
Numerical Results Experiment 1 Experiment 2
5
Conclusion
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
2/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
1
Introduction The Problem Solution Approaches
2
Optimization Model Formulation
3
Solution approaches Exact solution Heuristic solution
4
Numerical Results Experiment 1 Experiment 2
5
Conclusion
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
The Problem Solution Approaches
3/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
The Problem Solution Approaches
Planning situation Dynamic and Random Demand
Demand (forecasted averages and variations) Period t µt σt
25/2009 ... ...
26/2009 ... ...
... ... ...
35/2009 ... ...
Holding costs Setup costs Service level
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
4/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
The Problem Solution Approaches
Alternatives
Common sense approach (MRP, APS) Compute safety stocks and add to forecasted demand (st , qt )-policy, (rt , St )-policy Use a stationary inventory policy with dynamic adjustment of parameters ”Static-dynamic uncertainty” strategy Fix replenishment periods in advance, adjust production quantity ”Static uncertainty” strategy Fix replenishment periods and quantities in advance
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
5/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
1
Introduction The Problem Solution Approaches
2
Optimization Model Formulation
3
Solution approaches Exact solution Heuristic solution
4
Numerical Results Experiment 1 Experiment 2
5
Conclusion
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Formulation
6/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Formulation
Model Formulation I Model SSIULSPqβc : Minimize Z =
T X t=1
s. t.
s · γt + h · E [It ]+
(1)
It−1 + qt − Dt = It
t = 1, 2, . . . , T
(2)
qt − M · γt ≤ 0
t = 1, 2, . . . , T
(3)
Itf ,prod = − [It−1 + qt ]−
t = 1, 2, . . . , T
(4)
Itf ,end = − [It ]−
t = 1, 2, . . . , T
(5)
Ft = Itf ,end − Itf ,prod
t = 1, 2, . . . , T
(6)
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
7/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Formulation
Model Formulation II lt = (lt−1 + 1) · (1 − γt )
t = 1, 2, . . . , T
l0 = −1
(8) t = 1, 2, . . . , T − 1
ωt = γt+1 ωT = 1 ( E
1−
E
(
(7)
(9) (10)
t X
j=t−lt t X
j=t−lt
Fj
Dj
)
) ≥ βc⋆
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
t ∈ {t | ωt = 1}
(11)
8/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Formulation
Symbols used I βc⋆ Dt Ft γt h It Itf ,end Itf ,prod lt M
target fillrate demand in period t (random variable) backorder in period t (random variable) binary setup indicator in period t inventory holding cost net inventory at the end of period t (random variable) backlog at the end of period t (random variable) backlog immediately after production in period t (random variable) number of periods since the last setup prior to period t large number
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
9/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Formulation
Symbols used II
ωt
qt s T [x]+ [x]−
indicator variable: ωt = 1, if production takes place in period t + 1; ωt = 0, otherwise production quantity in period t setup cost length of planning horizon = max{0, x} = min{0, x}
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
10/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Formulation
Expected Inventory
E {Itp }
=
Z
Q (t) 0 (t)
=Q
(Q (t) − y ) · fY (t) (y ) · dy
− E {Y (t) } + GY1 (t) (Q (t) )
t = 1, 2, . . . (14)
Q (t) – cumulated production quantity from period 0 to t Y (t) – cumulated demand from period 0 to t
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
11/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
1
Introduction The Problem Solution Approaches
2
Optimization Model Formulation
3
Solution approaches Exact solution Heuristic solution
4
Numerical Results Experiment 1 Experiment 2
5
Conclusion
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Exact solution Heuristic solution
12/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Shortest-Path Network E{C14 } E{C24 (P2 )}
1
E{C12 }
2
E{C23 (P2 )}
3
E{C34 (P3 )}
4
E{C13 }
E {Cτ t } = s + h ·
t−1 X ℓ=τ
" #+ ℓ X E Iτ −1 (Pτ ) + qτ⋆t − Di
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
(15)
i =τ
13/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Shortest-Path Network
1 1 1 1 1 1 1 .. .
Setup 2 0 1 1 1 1 0 .. .
in period 3 4 0 0 0 0 1 0 1 1 1 1 1 0 .. .. . .
5 0 0 0 0 1 0 .. .
On hand inventory E {I5p } 79.61 81.14 87.01 103.89 118.95 86.95 .. .
Table: Expected on-hand inventory at the end of period 5 as a function of the setup pattern
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
14/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Solution Procedure 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20:
M := U := {2, 3, . . . , T } E := {(τ, t)|τ = 1, 2, . . . , T ; t = τ + 1, τ + 2, . . . , T − 1} for all ((0, t) ∈ E with t ∈ U) do Predecessor(t):= 1; C (t) := E{C1t } end for while (M = 6 ∅) do Select τ ∈ M with minimum C (τ ) M := M \ τ ; U := U \ τ if (τ = T ) then end else for all ((τ, t) ∈ E with t ∈ U) do M := M ∪ t if (C (τ ) + E{Cτ t } < C (t)) then Predecessor(t):= τ C (t) := C (τ ) + E{Cτ t } end if end for end if end while
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
15/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Dynamic Lot Sizing Heuristic 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12:
τ := 1 while (τ < T ) do t := τ while (t < T ) do if (Cτ t ≤ Cτ,t+1 ) then t := t + 1 else Make current lotsize for period τ permanent. τ := t + 1 end if end while end while
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
16/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Silver-Meal Rule
s +h· E {Cτ t } =
t X ℓ=τ
" #+ ℓ X E Iτ −1 (Pτ −1 ) + qτ∗t − Di i =τ
t −τ +1
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
(16)
17/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Least-Unit-Cost rule
E {Cτ t } = E
" #+ t ℓ X X s +h· Iτ −1 (Pτ −1 ) + qτ∗t − Di i =τ
ℓ=τ
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
t P
i =τ
Di
(17)
18/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Exact solution Heuristic solution
Least-Total-Cost rule
E {Cτ t } = E
s +h·
t X ℓ=τ
"
Iτ −1 (Pτ −1 ) + qτ∗t −
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
ℓ X i =τ
Di
#+
(18)
19/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
1
Introduction The Problem Solution Approaches
2
Optimization Model Formulation
3
Solution approaches Exact solution Heuristic solution
4
Numerical Results Experiment 1 Experiment 2
5
Conclusion
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Experiment 1 Experiment 2
20/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Expected Demands
Series # 1 2 3 4
E {Dt } 92 80 50 10
92 100 80 10
92 125 180 15
92 100 80 20
92 92 92 92 92 50 50 100 125 125 0 0 180 150 10 70 180 250 270 230
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
92 92 93 100 50 100 100 180 95 40 0 10
21/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Parameters
Series # T 1 2 3 4
12 12 12 12
s
TBO
CVD
βc⋆
500 500 500 500
1–12 1–12 1–12 1–12
{0.1, 0.2, 0.3, 0.4} {0.1, 0.2, 0.3, 0.4} {0.1, 0.2, 0.3, 0.4} {0.1, 0.2, 0.3, 0.4}
{0.5, 0.525, 0.05, . . . , 0.975} {0.5, 0.525, 0.05, . . . , 0.975} {0.5, 0.525, 0.05, . . . , 0.975} {0.5, 0.525, 0.05, . . . , 0.975}
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
22/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Results
1
LUC SM LTC PPA AC Groff
Average cost increase (%) 4.9 5.0 5.3 5.8 6.1 24.5
2
SM LTC LUC PPA AC Groff
5.9 7.2 7.5 7.9 7.9 24.8
Series # Heuristic
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Maximum cost increase (%) 30.4 30.4 29.1 43.5 30.1 52.7 32.7 28.7 31.3 49.9 32.9 60.2
% Optimal % Worst 56.6 56.0 48.6 57.8 48.6 3.9
3.76 3.97 7.00 19.23 7.52 75.24
42.8 41.5 39.9 51.4 38.3 0.4
4.79 5.31 5.83 17.50 5.31 69.90 23/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Results
3
AC PPA SM LTC Groff LUC
Average cost increase (%) 9.7 10.2 10.7 11.5 23.2 29.0
4
SM AC PPA LTC Groff LUC
6.0 13.2 15.3 15.8 17.1 39.1
Series # Heuristic
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Maximum cost increase (%) 37.0 55.8 36.6 41.1 58.9 67.9 30.9 50.3 59.0 52.3 47.6 61.2
% Optimal % Worst 35.4 46.0 29.5 33.2 10.3 1.8
3.02 3.96 1.04 5.83 33.13 54.69
49.9 27.8 34.2 22.5 14.3 3.4
1.88 0.73 14.58 2.19 7.81 74.48 24/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Results for Demand Series 1 40
Averagde cost increase (%)
35
Groff
30 25 20 15 10
AC
PPA
LTC
5
Silver-Meal
0
LUC
0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Fillrate
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
25/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Parameters
E {Dt } ≃ Uniform(0, 100) Series # T s TBO CVD 5 5 500 1–5 {0.1, . . . , 0.4} 6 10 500 1–5 {0.1, . . . , 0.4} 7 15 500 1–15 {0.1, . . . , 0.4} 8 20 500 1–15 {0.1, . . . , 0.4}
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
βc⋆ {0.5, 0.525, 0.05, . . . , 0.975} {0.5, 0.525, 0.05, . . . , 0.975} {0.5, 0.525, 0.05, . . . , 0.975} {0.5, 0.525, 0.05, . . . , 0.975}
26/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Results
5
SM AC LTC LUC Groff PPA
Average cost increase (%) 4.7 5.6 5.9 10.5 13.8 15.1
6
SM AC LTC Groff LUC PPA
4.2 10.2 11.5 13.8 15.4 19.9
Series # Heuristic
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Maximum cost increase (%) 35.4 39.7 63.4 64.4 60.5 56.7 32.7 41.2 44.7 70.5 55.3 60.4
% Optimal % Worst 65.4 61.2 62.6 50.6 34.7 38.6
14.47 13.07 11.16 25.25 36.77 42.72
43.7 21.6 16.5 13.9 12.6 12.2
3.75 10.05 13.35 27.95 21.25 43.78 27/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Experiment 1 Experiment 2
Results
7
LTC AC PPA LUC SM Groff
Average cost increase (%) 8.5 8.6 9.1 11.0 12.1 30.3
8
SM AC LTC PPA LUC Groff
8.8 10.7 11.6 12.5 13.9 29.9
Series # Heuristic
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
Maximum cost increase (%) 42.8 44.8 64.9 53.9 40.3 64.4 38.8 46.5 46.2 66.2 49.4 64.6
% Optimal % Worst 40.4 39.5 53.1 35.3 25.6 2.3
3.14 2.26 15.04 8.18 5.96 68.47
21.1 17.9 17.0 32.5 14.7 1.0
1.54 1.48 3.64 16.42 9.33 69.41 28/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
1
Introduction The Problem Solution Approaches
2
Optimization Model Formulation
3
Solution approaches Exact solution Heuristic solution
4
Numerical Results Experiment 1 Experiment 2
5
Conclusion
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
29/30
Introduction Optimization Model Solution approaches Numerical Results Conclusion
Conclusion
Exact solution for the stochastic Wagner-Whitin problem Adjusted cost criteria used in standard dynamic lot sizing heuristics Silver-Meal rule superior to Groff rule Directly applicable in ERP/AP systems Static uncertainty strategy: no nervousness, no bullwhip effect Target service level (instead of backorder costs) Possible extension: Capacities (done)
c Prof. Dr. Horst Tempelmeier — www.scmp.uni-koeln.de
30/30