Wednesday, November 26, 2014

Example of use of Unit Loss Function

In a previous post I showed how to write and solve a unit loss function G(k). Here I show how to use it.

Here I solve some (s,Q) inventory models. These models have a variable Q: order quantity. It is sometimes argued to use the EOQ order quantity for this. EOQ stands for Economic Order Quantity. In the model I try to see how much difference it makes when we use this EOQ or just solve for Q in the (s,Q) models directly.

The results are:

----    155 PARAMETER results 

                            EOQ      Qendog        Qeoq

EOQ .Q                 5200.000
CSEO.Q                             5876.746    5200.000
CSEO.Total Cost                 6337360.618 6337934.428
CSEO.Inv+Short Cost              137360.618  137934.428
CSEO.s                             5658.711    5749.015
CSEO.k                                2.094       2.152
CIS .Q                             5846.467    5200.000
CIS .Total Cost                 6331413.493 6331942.157
CIS .Inv+Short Cost              131413.493  131942.157
CIS .s                             5292.515    5373.301
CIS .k                                1.860       1.912

We can see for two inventory models (Cost per Stockout Event – CSOE and Cost per Item Short – CIS) the differences in Q and s are significant. However the effect on total cost and relevant cost is more limited: things are pretty flat out there.

In practice formulas based on the first order conditions are used to solve these inventory models. However I think there is a case to be made to look at the original cost functions and optimize these directly. This relates more to the original problem and also we may be a little bit more flexible if we want to add a few more wrinkles. In some cases there are closed solutions, e.g. the well known EOQ formula is:

image

In the model below again we use the original cost function and minimize that for Q. Note that for other cases it may not be that easy to find closed solutions.

$ontext

  
Some (s,Q) inventory models

  
We evaluate two inventory models:
     
(1) Cost per Stockout Event (CSOE) Model
     
(2) Cost per Item Short (CIS) Model

  
It is sometimes suggested to input Q*=EOQ into these models
  
opposed to optimizing directly for Q. Here we try to
  
see how much a difference this makes.

$offtext

scalars
   D        
'mean demand ($/year)'           /62000/
   sigma_D  
'standard deviation of demand'    /8000/
   c        
'cost per item ($/item)'           /100/
   cK       
'order cost ($/order)'       /3270.9678/
   ci       
'annual inventory cost'
   h        
'holding charge (% of unit cost)' /0.15/
   mu_dl    
'mean over lead time'
   sigma_dl 
'sigma over lead time'
   L        
'lead time (days)'                  /14/
   B1       
'CSOE penalty'                   /50000/
   cs       
'Item short cost'                   /45/
   Qeoq     
'EOQ'
;

ci = h*c;
mu_dl = D/(365/L);
sigma_dl = sigma_D/sqrt(365/L);

parameter results(*,*,*);

*------------------------------------------------------
* Deterministic EOQ model
*------------------------------------------------------

variables
    tc 
'total cost'
    Q  
'order quantity'
;

* prevent division by zero
Q.lo = 0.1;

equations
    costdef1   
'total cost calculation for simple deterministic case'
;

costdef1..
   tc =e= c*D + cK*(D/Q) + ci*(Q/2);

model eoq /costdef1/;
solve eoq minimizing tc using nlp;

Qeoq = Q.l;
results(
'EOQ','Q','EOQ') = Qeoq;


*------------------------------------------------------
*  Cost per Stockout Event Model
*------------------------------------------------------

positive variables
   k
   PStockout 
'P[x>=k]'
   s         
'order point'
;
equations
    costdef2 
'CSOE total cost function'
    cdf      
'this implements P[x>=k]'
    sdef     
'calculation of order point s'
;

costdef2..
   tc =e=  c*D + cK*(D/Q) + ci*(Q/2+k*sigma_dl) + B1*(D/Q)*PStockOut;

cdf..
   Pstockout =e= 1-errorf(k);

sdef..
   s =e= mu_dl + k*sigma_dl;

model csoe /costdef2,cdf,sdef/;



*------------------------------------------------------
*  Cost per Item Short (CIS)  Model
*------------------------------------------------------

variables
    G  
'unit loss function'
;
equations
    costdef3
'CIS total cost function'
    Gdef    
'this implements G(k)'
;

costdef3..
   tc =e=  c*D + cK*(D/Q) + ci*(Q/2+k*sigma_dl) + cs*sigma_dl*G*(D/Q);

Gdef..
   G =e= 1/sqrt(2*pi)*exp(-0.5*sqr(k)) - k * (1-errorf(k));

model cis /costdef3,sdef,Gdef/;



*------------------------------------------------------
*  Results with Q endogenous
*------------------------------------------------------

solve csoe minimizing tc using nlp;

results(
'CSEO','Total Cost','Qendog') = TC.L;
results(
'CSEO','Inv+Short Cost','Qendog') = TC.L-c*D;
results(
'CSEO','Q','Qendog') = Q.L;
results(
'CSEO','s','Qendog') = s.L;
results(
'CSEO','k','Qendog') = k.L;

solve cis minimizing tc using nlp;

results(
'CIS','Total Cost','Qendog') = TC.L;
results(
'CIS','Inv+Short Cost','Qendog') = TC.L-c*D;
results(
'CIS','Q','Qendog') = Q.L;
results(
'CIS','k','Qendog') = k.L;
results(
'CIS','s','Qendog') = s.L;


*------------------------------------------------------
*  Results with Q fixed to EOQ
*------------------------------------------------------

Q.fx = Qeoq;

solve csoe minimizing tc using nlp;

results(
'CSEO','Total Cost','Qeoq') = TC.L;
results(
'CSEO','Inv+Short Cost','Qeoq') = TC.L-c*D;
results(
'CSEO','Q','Qeoq') = Q.L;
results(
'CSEO','k','Qeoq') = k.L;
results(
'CSEO','s','Qeoq') = s.L;

solve cis minimizing tc using nlp;

results(
'CIS','Total Cost','Qeoq') = TC.L;
results(
'CIS','Inv+Short Cost','Qeoq') = TC.L-c*D;
results(
'CIS','Q','Qeoq') = Q.L;
results(
'CIS','k','Qeoq') = k.L;
results(
'CIS','s','Qeoq') = s.L;

display results;

No comments:

Post a Comment