Method to provide sensitivity information for lost-sale (R, s,S) inventory systems

• US 6,006,202 A
• Filed: 09/03/1997
• Issued: 12/21/1999
• Est. Priority Date: 09/03/1997
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Cheng, Dinah W., Leung, Ying Tat, Ramaswamy, Sanjay Elathur
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Campa, John
• Application Number:
  US08/922,286
• Time in Patent Office: 839 Days
• Field of Search: 705/28, 705/29, 705/7, 705/8, 705/10, 364/918.4, 364/468.06, 364/468.07, 364/468.13, 364/468.14
• US Class Current: 705/28
• CPC Class Codes: G06Q 10/08 Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
• CPC Class Codes: G06Q 10/087 Inventory or stock management, e.g. order filling, procurement or balancing against orders
• CPC Class Codes: G06Q 10/0875 Itemisation or classification of parts, supplies or services, e.g. bill of materials

Abstract

Sensitivity information about average lost sales and inventory levels of a lost-sale (s,S) inventory system with respect to the two decision variables s and S is used for "what-if" analysis and optimization of the decision variables. Computer software implements an algorithm that provides sensitivity of average inventory levels as well as average lost sales with respect to changes in s and q(q=S-s) for a lost-sale inventory simulation for any specified demand distribution. The sensitivity information provided is exact when compared to a finite-difference simulation for the case of both discrete and continuous, stationary and non-stationary demands. A finite difference estimate of the sensitivity is obtained by running two simulations where the parameter in question (s or S) differ by a chosen small number, and then calculating the resulting difference in the performance measure of interest, in this case average inventory and lost sales.

Claims

1. A computer implemented method for generating sensitivity information about average lost sales and inventory levels of a lost-sale (R,s,S) inventory system where R is a review period for inventory, s is a specified reorder point, and S is a specified order up to a level value such that when an inventory position is below s at a start of a time period an order is placed to bring an inventory position up to S, said method comprising the steps of:

reading input data from databases on inventory and simulation parameters and initializing variables, said simulation parameters including a perturbation value Δ

;

for a time horizon under consideration, performing an inventory simulation with a perturbation analysis, the perturbation analysis including a tracking of a plurality of paths of events representing an inventory state of a nominal system based on s and S, and an inventory state of two perturbed systems based, respectively on s and q perturbed by the value Δ

, where a q value=(S-s), using a combination of an adjusted Infinitesimal Perturbation Analysis algorithm and a set of finite difference calculations;

based on the perturbation analysis for the time horizon under consideration, calculating an output average sensitivity information on performance measures reflecting average inventory and average lost sales per unit change in s and q, and updating average statistics;

writing said output average sensitivity information to an output database; and

managing an inventory level of said (R,s,S) inventory system based on said average sensitivity information.

2. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 1, wherein said step ofreading input data from databases on inventory and simulation parameters and initializing variables, said parameters including a perturbation value Δ

comprises the steps of;

initializing a reorder point value s, and a reorder-up-to level value S;

initialzing, for a stating time index t=0 with nominal system values corresponding to s and S values, a time-indexed nominal system physical inventory value, a time-indexed nominal system stock delivery value, a time-indexed nominal system stock on order value, and a time-indexed sum of the time-indexed nominal system physical inventory and the nominal system stock-on-order value; and

setting the Δ

value representing a delta change in the reorder point s and in a q point.

3. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 2 wherein said step of reading input data from databases on inventory and simulation parameters and initializing variables, said parameters including a perturbation value Δ

further comprises the steps of;

initializing a first plurality of flags representing a sequence of events associated with the physical inventory of the nominal system, a first perturbed system, and a second perturbed system, said first perturbed system being said system based on said s value perturbed by the Δ

value, said second perturbed system being said system based on said q value perturbed by the Δ

value;

initializing a second plurality of flags representing a system state of said nominal system, with respect to said time increment t, said system state including whether or not an order has been placed prior to t, whether or not an inventory is less than or equal to a zero value, and whether or not an order is being received.

4. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 3 wherein said step of performing an inventory simulation with perturbation analysis for a time horizon under consideration comprises the steps of:

determining if the nominal path and the first perturbed path and the nominal path and the second perturbed path are at a converged value in accordance with a flag from said first and second plurality of flags;

modifying the nominal system physical inventory value and the nominal system stock delivery value based on the nominal system stock on order value;

checking if an order is to be placed, based on the s value and the nominal system physical inventory value, and

if the order is to be placed, performing steps (a) and (b) of

(a) constructing a perturbed path with respect to the change in q corresponding to said Δ

value in response to first and second plurality of flags, and

(b) calculating a value for the order based on the nominal system physical inventory value and the S value;

setting said second plurality of flags based on the system state of the nominal system;

generating a demand value;

constructing a perturbed system path with respect to the change in s corresponding to said Δ

value based on a state of said first plurality of flags;

updating a plurality of period statistics representing said nominal system physical inventory value and a lost sale value;

tracing a path for the first perturbed system based on the plurality of period statistics from said updating step and a condition of said first and second plurality of flags;

updating a second plurality of period statistics representing said first perturbed system physical inventory value and a first perturbed system lost sale value;

cumulating at least one of said second plurality of difference statistics for the first perturbed system with respect to the change in s corresponding to the Δ

value;

tracing a path for the second perturbed system based on a state of said first and second plurality flags;

updating a plurality of period statistics for the second perturbed system with respect to the change in q corresponding to the Δ

value;

cumulating a plurality of difference statistics for the second perturbed system with respect to changes in q corresponding to the Δ

value;

updating a plurality of cumulative statistics for the nominal system based on said step of updating a plurality of period statistics;

performing an adjusted infinitetessimal perturbation analysis calculations based on a state of the second plurality of flags;

incrementing the value t of the time increment counter;

if the value t is less than T repeating the step of determining if the nominal path and the first perturbed path and the nominal path and the second perturbed path are at a converged value and all subsequent steps in order.

5. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 4, further comprising the step of replicating, for a time horizon under consideration, the step of performing an inventory simulation with perturbation analysis, wherein the step of generating a demand value in each replication uses a random demand value based on a predetermined statistical model, and

wherein said step of calculating an output average sensitivity information on performance measures reflecting average inventory and average lost sales per unit change in s and q, and updating average statistics where q=(S-s) outputs said average statistics reflecting said replications.

6. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 1, further comprising the step of replicating, for a time horizon under consideration, the step of performing an inventory simulation with perturbation analysis, wherein each replication uses a series of random demand values based on a predetermined statistical model, and

wherein said step of calculating an output average sensitivity information on performance measures reflecting average inventory and average lost sales per unit change in s and q, and updating average statistics where q=(S-s) outputs said average statistics reflecting said replications.

7. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 1, wherein said step of using a combination of an adjusted Infinitesimal Perturbation Analysis algorithm and a set of finite difference calculations includes:

a) using a predetermined number of switches to recognize events that trigger path changes,

b) switching between the adjusted IPA algorithm and the set of finite difference equations based on a configuration of said switches, and

c) computing gradient estimates as a sum ofthe adjusted IPA algorithm and the set of finite difference calculations over said time horizon.

8. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 1, wherein said managing step includes placing purchase orders to replenish depleted inventory based on said output sensitivity information calculated in said calculating step.

9. A computer implemented method for generating sensitivity information about average lost sales and inventory levels according to claim 1, wherein when said inventory position is below s said inventory position is zero.

10. A computer system for generating sensitivity information about average lost sales and inventory levels of a lost-sale (R,s,S) inventory system where R is a review period for inventory, s is a specified reorder point and S is a specified order up to level value such that when an inventory position is zero and falls below s at a start of a time period an order is placed to being an inventory position up to S comprising:

a first input database containing inventory parameters;

a second input database containing simulation parameters;

a processor connected to said first and second input databases, said processor reading said inventory parameters and said simulation parameter settings and executing methods for calculating average unit change in s and q over all replications and updating average statistics where a q value=(S-s), said methods being executed for a predetermined number of replications and for a time horizon under consideration, said processor having means for tracking of a plurality of paths of events representing an inventory state of a nominal system based on s and S, and an inventory state of two perturbed systems based, respectively, on s and q perturbed by the value Δ

, said processing using a combination of an adjusted Infinitesimal Perturbation Analysis algorithm and a set of finite difference calculation based on a convergence of said plurality of paths, and said processor calculating an average change over all replications, wherein application of said combination includes;

using a predetermined number of switches to recognize events that trigger path changes, and switching between the adjusted IPA algorithm and the set of finite difference equations based on a configuration of said switches; and

an output database connected to the processor, said processor outputting sensitivity information on performance measures to said output database, said (R,s,S) inventory system managing an inventory level of said system based on said average sensitivity information.

11. A computer implemented method for generating sensitivity information as recited in claim 10, wherein said processor includes:

a) means for using a predetermined number of switches to recognize events that trigger path changes,

b) means for switching between the adjusted IPA algorithm and the set of finite difference equations based on a configuration of said switches, and

c) means for computing gradient estimates as a sum of the adjusted IPA algorithm and the set of finite difference calculations over said time horizon.

12. A computer-implemented method for generating sensitivity information with respect to s and S values in an (R,s,S) inventory system, comprising the steps of:

initializing a reorder point s value, and a reorder-up-to level S value;

initializing, for a starting time index t=0, a time-indexed nominal system physical inventory value, a time-indexed nominal system stock delivery value, a time-indexed nominal system stock on order value, and a time-indexed sum of the nominal system physical inventory and the nominal system stock on order value, said nominal system values corresponding to said s and S values;

setting a Δ

value representing a delta change in the reorder point s and in a q point, where q=(S-s);

initializing a first plurality of flags representing a sequence of events associated with the physical inventory of the nominal system, a first perturbed system, and a second perturbed system, said first perturbed system based on said s value perturbed by the Δ

value, said second perturbed system based on said q value perturbed by the Δ

value;

initializing a second plurality of flags representing a system state of said nominal system, with respect to said time increment t, said system state including whether or not an order has been placed prior to t, whether or not an inventory is less than or equal to a zero value, and whether or not an order is being received;

determining if the nominal path and the first perturbed path and the nominal path and the second perturbed path are at a converged value in accordance with flags from said first and second plurality of flags;

modifying the nominal system physical inventory value and the nominal system stock delivery value based on the nominal system stock on order value;

checking if an order is to be placed, based on the s value and the nominal system physical inventory value, and

if the order is to be placed, performing steps (a) and (b) of

(a) constructing a perturbed path with respect to the change in q corresponding to said Δ

value in response to first and second plurality of flags, and

(b) calculating a value for the order based on the nominal system physical inventory value and the S value;

setting said second plurality of flags based on the system state of the nominal system;

generating a demand value;

constructing a perturbed system path with respect to the change in s corresponding to said Δ

value based on a state of said first plurality of flags;

updating a plurality ofperiod statistics representing said nominal system physical inventory value and a lost sale value;

tracing a path for the first perturbed system based on the plurality of period statistics from said updating step;

updating a second plurality of period statistics representing said first perturbed system physical inventory value and a first perturbed system lost sale value;

cumulating at least one of said second plurality of difference statistics for the first perturbed system with respect to the change in s corresponding to the Δ

value;

tracing a path for the second perturbed system based on a state of said second plurality flags;

updating a plurality of period statistics for the second perturbed system with respect to the change in q corresponding to the Δ

value;

cumulating a plurality of difference statistics for the second perturbed system with respect to changes in q corresponding to the Δ

value;

updating a plurality of cumulative statistics for the nominal system based on said step of updating a plurality of period statistics;

performing adjusted infinitesimal perturbation analysis calculations based on a state of the second plurality of flags;

incrementing the value t of the time increment counter;

if the value t is less than T repeating the step of determining if the nominal path and the first perturbed path and the nominal path and the second perturbed path are at a converged value and all subsequent steps in order,

else

updating a plurality of average statistics representing a sensitivity change in an average inventory and average lost sale performance of the inventory system with respect to the Δ

change in s and q, where said plurality of average statistics are used to identify an optimum s and q value for an inventory system, and

managing an inventory level of said (R,s,S) inventory system based on said optimum s and q value.

13. A method for generating a sensitivity information according to claim 12 further comprising the steps of:

(c) initializing a replication counter value n to a starting value;

(d) performing the step of initializing a first plurality of flags and all subsequent steps in order;

(e) incrementing the replication counter value; and

(f) repeating said steps (d) and (e), wherein the step of generating a demand generates a stream of random values for each replication, based on a predetermined demand function, until the replication counter value equals a predetermined number N,

whereby said step of updating a plurality of average statistics representing a sensitivity change in an average inventory and average lost sale performance of the inventory system with respect to the Δ

change in s and q reflects the average of said average statistics over the N replications.

14. A computer-implemented method for generating sensitivity information with respect to s and S values in an(s,S) inventory system, comprising the steps of:

initializing a reorder point s value, a reorder-up-to level S value, a nominal system physical inventory value, a nominal system stock delivery value, a nominal system stock on order value, and a sum of the nominal system physical inventory and the nominal system stock-on-order value, where said nominal system values correspond to a state of a nominal inventory system based on said s and S values;

setting a Δ

value representing a delta change in the reorder point s and in a q point, where q=(S-s);

initializing a first plurality of flags representing a sequence of events associated with the physical inventory ofthe nominal system and a first perturbed system, respectively, said first perturbed system based on said s value perturbed by the Δ

value;

initializing a second plurality of flags representing a sequence of events associated with the physical inventory of the nominal system and a second perturbed system, respectively, said second perturbed system based on said q value perturbed by the Δ

value;

initializing a third plurality of flags representing a system state of said nominal system, with respect to a time increment t, said system state including whether or not an order has been placed prior to t, whether or not an inventory is less than or equal to a zero value, and whether or not an order is being received;

setting a counter value t for the time increment t to a starting number;

determining if the nominal path and the first perturbed path and the nominal path and the second perturbed path are at a converged value in accordance with a flag from said first, a flag from said second and a flag from said third plurality of flags;

modifying the nominal system physical inventory value and the nominal system stock delivery value based on the nominal system stock on order value;

checking if an order is to be placed, based on the s value and the nominal system physical inventory value, and

if the order is to be placed, performing steps (a) and (b) of

(a) checking at least one of said second plurality of flags and said third plurality of flags for starting a trace of the perturbed path with respect to changes in q, and constructing a perturbed path based on said checking, and

(b) calculating a value for the order based on the nominal system physical inventory value and the S value;

setting at least one of said third plurality of flags representing a system state of the nominal system based on at least one other of said third plurality;

generating a demand value;

checking said at least one of said first plurality of flags for starting a finite difference simulation with respect to changes in s and, based on said checking, constructing a perturbed system path with respect to changes in s;

updating a plurality ofperiod statistics representing said nominal system physical inventory value and a lost sale value;

setting at least one of said third plurality flags to a logical ON based on the plurality of period statistics from said updating step;

checking at least one of said first plurality flags and tracing a path for the first perturbed system based on said checking;

updating a second plurality of period statistics representing said first perturbed system physical inventory value and a first perturbed system lost sale value;

cumulating at least one of said second plurality of difference statistics for the first perturbed system with respect to changes in s;

checking at least one of said second plurality flags and tracing a path for the second perturbed system based on said checking;

updating a plurality of period statistics for the second perturbed system with respect to changes in q;

cumulating a plurality of difference statistics for the second perturbed system with respect to changes in q;

updating a plurality of cumulative statistics for the nominal system based on said step of updating a plurality of period statistics;

checking at least one of said third plurality of flags;

performing adjusted infinitesimal perturbation analysis calculations based on said checking;

incrementing the value t of the time increment counter;

if the value t is less than T repeating the step of determining if the nominal path and the first perturbed path and the nominal path and the second perturbed path are at a converged value and all subsequent steps in order,

else

updating a plurality of average statistics representing a sensitivity change in an average inventory and average lost sale performance of the inventory system with respect to the Δ

change in s and q and

managing an inventory level of said (R,s,S) inventory system based on said average statistics.

15. A method for generating a sensitivity information according to claim 14 further comprising the steps of:

(c) initializing a replication counter value n to a starting value;

(d) performing the step of initializing a first plurality of flags and all subsequent steps in order;

(e) incrementing the replication counter value; and

(f) repeating said steps (d) and (e), wherein the step of generating a demand generates a stream of random values for each replication, based on a predetermined demand function, until the replication counter value equals a predetermined number N,

whereby said step of updating a plurality of average statistics representing a sensitivity change in an average inventory and average lost sale performance of the inventory system with respect to the Δ

change in s and q reflects the average of said average statistics over the N replications.