Optimization problem

Impact of changes in the size of the front area in a two-zone

warehouse on total costs
Meiliang Zhao

September 2024

1 Problem definition:
In common B2C (business to consumer) warehouses, warehouses are usually divided into two main areas: the
reserve area and the forward area. Reserve Area: The reserve area is the main storage area of the warehouse,
used to store large quantities of inventory. This part of the area is large, with high storage density and relatively
low cost. The reserve area is mainly used to replenish the forward area and meet long-term inventory needs.
Forward Area: The forward area is dedicated to fast picking, and storing small, high-frequency, and easy-to-
pick small items. The forward area is designed to improve picking efficiency by reducing the walking distance of
pickers. Due to the limited area of the forward area, it needs to be replenished from the reserve area regularly.
The area and configuration of the forward area will directly affect the replenishment frequency, picking efficiency
and overall operating costs.

1.1 Core problem:

The total area of the warehouse is fixed. Under this premise, it is necessary to decide how to adjust the area of
the forward area to minimize the total operating cost. The total cost consists of the following parts:

• Picking cost:It is directly related to the size and layout of the forward area. Within a certain range, the
larger the area, the higher the picking efficiency and the lower the picking cost.
• Replenishment cost: When the forward area is small, the storage capacity is limited and more frequent
replenishment from the reserve area is required, which increases the replenishment cost.

• Inventory holding cost: The holding costs of the forward area and the reserve area are different. The
unit area holding cost of the forward area is usually higher.
• Out-of-stock cost: When the forward area is too small, it may lead to high-frequency out-of-stock,
affecting order fulfilment and increasing out-of-stock costs.

1.2 Overall layout of warehouse

• 1.Total area of warehouse:The total area of the warehouse is A square meters, which is a fixed value.
The warehouse is divided into the forward area and the reserve area.
• 2.Relationship between the forward area and the reserve area: The forward area is used for fast
picking of high-frequency goods, and the reserve area is used for bulk storage. The area of the forward
area Af and the area of the reserve area As meet the total area restriction: Af + As = A
• 3.Front area layout design:The front area adopts a modular design and is adjusted with a basic unit
of 25 square meters (5 meters × 5 meters square) to ensure the convenience of adjusting warehouse
operations.

• 4.Integerity of adjustments:Each adjustment of the front zone area can only be an integer multiple of
25 square meters to maintain the practical operability of the adjustment.
Optimization problem

1.3 research assumptions

1. Demand Stability Assumption: It is assumed that customer demand is stable and known during
the study period, and the demand quantity Dt follows a normal distribution Dt ∼ N (µ, σ 2 ). Seasonal
fluctuations, promotional activities, or sudden spikes in demand are not considered.
2. Replenishment Strategy Assumption: Replenishment is completed before each picking cycle to ensure
that the front zone inventory is sufficient before each picking. The frequency of replenishment is inversely
proportional to the area of the front zone.
3. Picking and Replenishment Path Assumption: Pickers use an ’S’ path strategy, moving along the
aisles within the front zone to minimize repetitive routes and unnecessary movement. The length of the
picking path is primarily affected by the area of the front zone.
4. Storage Strategy Assumption: The storage density in the front zone is relatively low, and the holding
cost per unit area in the front zone is higher than that in the reserve zone. It is assumed that the holding
cost per unit area CAf in the front zone is twice that in the reserve zone CAs .

5. Time and Resource Limitation Assumption: Each picking and replenishment operation is completed
by a single operator, and there is a fixed time limit for each picking and replenishment cycle. Both picking
and replenishment must be completed within each cycle.
6. Cost Composition Assumption: Only direct operational costs are considered, including picking costs,
replenishment costs, holding costs, and stockout costs. Other indirect costs are not considered. The
specific forms of costs are: picking costs decrease with the increase in the front zone area; replenishment
costs increase with the decrease in the front zone area; holding costs are proportional to the areas of the
front and reserve zones; stockout costs increase with the decrease in the front zone area.
7. Adjustment Unit Assumption: Adjustments can increase or decrease the front zone area but must be
continuous and reasonable. Partial adjustments are not allowed, and adjustments must be in units of 25
square meters (i.e., each adjustment is a 5×5 meter square).
8. Stockout Handling Assumption: When a stockout occurs in the front zone, pickers will directly pick
from the reserve zone and trigger additional replenishment operations.
9. Front Zone Area Reasonableness Assumption: The front zone area Af must be no less than the
basic requirement of 25 square meters and must be less than the reserve zone area As , i.e., Af < As .

2 Parameter Definition

Symbol Description
A Total warehouse area (square meters), a fixed constant.
Af Forward area (square meters), the main decision variable to be optimized.
As Reserve area (square meters), calculated as As = A − Af .
∆A Adjustment unit area (square meters), set to 25 square meters, the basic ad-
justment unit.
CAf Holding cost per unit area of the forward area (cost per square meter per day),
assumed to be twice that of the reserve area.
CAs Holding cost per unit area of the reserve area (cost per square meter per day),
typically CAf = 2 × CAs .
Cpick (Af ) Picking cost (daily picking cost), a function that generally decreases as the
forward area increases.
Crepl (Af ) Replenishment cost (daily replenishment cost), increases as the forward area
decreases, expressed as an inverse function of the forward area.
Chold (Af , As ) Holding cost (total daily holding cost), calculated as Chold = CAf × Af + CAs ×
As .
Cshort (Af ) Shortage cost (daily shortage cost), increases as the forward area decreases.
Dt Demand (units or boxes), normally distributed Dt ∼ N (µ, σ 2 ), where µ is the
average demand and σ 2 is the variance of the demand.
Qs Replenishment quantity per time (units), related to forward area inventory and
demand.
Continued on next page
Optimization problem

Symbol Description
Frepl (Af ) Replenishment frequency (times/day), expressed as Frepl (Af ) = Arf , where r is
a constant representing the relationship between replenishment frequency and
forward area.
Trepl Replenishment time (hours per time), set as a fixed value and does not change
with area.
Tpick Picking time (hours per time), set as a fixed value and does not change with
area.
r Replenishment frequency constant, used to describe the inverse relationship
between replenishment frequency and forward area.
ϵ A small constant, used to prevent division by zero and ensure stability in
calculations.
k Forward area adjustment factor (integer), represents the number of adjustment
units (25 square meters) used in adjusting the forward area.
k1 Picking cost coefficient, representing the relationship between picking cost and
forward area.
k2 Replenishment cost coefficient, representing the relationship between replen-
ishment cost and forward area.
k3 Shortage cost coefficient, representing the relationship between shortage cost
and forward area.

Decision Variables
• Af : Forward area (square meters). This is the main decision variable that represents the area of the
forward picking zone. It is adjusted to minimize the total cost.

• k: Adjustment factor (integer). Represents the number of adjustment units (each 25 square meters)
applied to the forward area. It determines the increments or decrements in the forward area size.
• ∆A: Adjustment unit area (square meters), set as 25 square meters. It indicates the basic unit of area
adjustment in the forward area during optimization.

Objective Function
Minimize CTotal (Af ) = Cpick (Af ) + Crepl (Af ) + Chold (Af , As ) + Cshort (Af )

1. Picking Cost:
K1
Cpick (Af ) =
Af + ϵ
where K1 is a constant representing the picking cost coefficient related to the forward area size, and ϵ is
a small value to prevent division by zero.
2. Replenishment Cost:
K2 · r
Crepl (Af ) =
Af + ϵ
where K2 is the replenishment cost coefficient, r is the replenishment frequency constant, and the replen-
ishment cost increases as Af decreases.
3. Holding Cost:

Chold (Af , As ) = CAf · Af + CAs · As = CAf · Af + CAs · (A − Af )

where CAf and CAs are the holding costs per unit area for the forward and reserve areas, respectively.
4. Shortage Cost:
K3
Cshort (Af ) =
Af + ϵ
where K3 is the shortage cost coefficient, representing the impact of forward area size on shortage costs.

K1 K2 · r K3
Minimize CTotal (Af ) = + + CAf · Af + CAs · (A − Af ) +
Af + ϵ Af + ϵ Af + ϵ
Optimization problem

Constraints
X X X
Minimize tij yij (1)
i∈M j∈M ′ ,j̸=i t∈T

1. Inventory Balance Constraint:

Iit+1 = Iit + Pit − xit , ∀i ∈ M, t ∈ T (2)

Explanation: Current inventory Iit plus replenishment Pit minus demand xit equals the next period’s
inventory.
2. Initial Inventory Constraint:

Iit+1 = Iit + xit − rit , ∀i ∈ M, t ∈ T (3)

Explanation: Current inventory Iit plus demand xit minus fulfilled demand rit equals the next period’s
inventory.
3. Shortage Constraint:
xit = Ui zit − Iit , ∀i ∈ M, t ∈ T (4)
Explanation: Defines the shortage xit as the inventory upper limit Ui times the shortage indicator zit
minus current inventory Iit .
4. Inventory Upper Bound Constraint:

xit ≤ Ui zit , ∀i ∈ M, t ∈ T (5)

Explanation: Ensures the shortage does not exceed the inventory upper limit Ui times the shortage
indicator.
5. Inventory Upper Bound Constraint:

xit ≤ Ui − Iit , ∀i ∈ M, t ∈ T (6)

Explanation: Ensures the shortage does not exceed the total inventory upper limit minus current inventory.
6. Time Constraint: X X
tij yij ≤ L, ∀t ∈ T (7)
i∈M j∈M ′ ,j̸=i

Explanation: Total operation time tij times the respective operation indicator yij sum must not exceed
the time limit L.
7. Replenishment Constraint: X
xit ≤ Q, ∀t ∈ T (8)
i∈M

Explanation: Total demand xit sum must not exceed the replenishment quantity Q.
8. Symmetry Constraint:
X X
yij + yji = 2zit , ∀i ∈ M ′ , t ∈ T (9)
j∈M ′ j∈M ′ ,j̸=i

Explanation: Ensures symmetry in operations, where the sum of operation indicators yij and yji equals
twice the shortage indicator.
9. Selection Constraint:
X X
yij + yji ≤ 2zit , ∀S ⊂ M, t ∈ T, k ∈ S (10)
j∈S,j̸=i j∈S

Explanation: Controls the selection constraint, ensuring chosen operations adhere to symmetry require-
ments.
10. Inventory Lower Bound Constraint:

Iit ≥ (1 − zit )rit , ∀i ∈ M, t ∈ T (11)

Explanation: Ensures inventory Iit is not less than (1 − zit ) times the replenishment rit .
Optimization problem

11. Time Constraint:

(Frepl(Af ) × Trepl ) + (Fpick(Af ) × Tpick ) ≤ T (12)

Explanation: Ensures that the total time spent on replenishment (Frepl(Af )×Trepl ) and picking (Fpick(Af )×
Tpick ) does not exceed the available working time T (e.g., 8 hours per day).

3 Conversion to Knapsack Problem

To optimize the warehouse layout and minimize the total costs, this problem can be transformed into a classical
knapsack problem. This transformation allows the use of known optimization techniques for knapsack problems
to find the optimal configuration of the forward area.
Rationale for Conversion:

1. The original problem involves adjusting the forward area (Af ) to minimize the total cost function, which
includes picking cost, replenishment cost, holding cost, and shortage cost. In the context of the knapsack
problem:
• Items: Each adjustment unit (∆Ak ) is viewed as an ”item” whose selection has a specific impact on
the total cost.
• Capacity: The total area constraint of the warehouse is analogous to the knapsack’s capacity,
requiring that the sum of the forward and reserve areas equals the total warehouse area (Af +As = A).
• Value: Each adjustment unit has an associated value (V (∆Ak )), representing its impact on the
overall cost, similar to the value of an item in the knapsack. The objective is to select the optimal
combination of these adjustment units to minimize the total cost.
• Binary Decision Variables: Binary variable xk indicates whether an adjustment unit is selected.
As in the knapsack problem, xk = 1 if the unit is included, and xk = 0 if it is not.
2. Benefits of the Conversion:
• By transforming the area adjustment problem into a knapsack problem, heuristic and metaheuristic
algorithms (such as Genetic Algorithms, Simulated Annealing, or Particle Swarm Optimization) can
be applied. These methods are particularly suitable for large-scale problems, providing near-optimal
solutions when exact solutions are computationally expensive.
• This transformation offers a structured approach to systematically explore the best area adjustment
strategies, leveraging existing optimization techniques, making the problem more manageable and
solution-oriented.

4 Knapsack problem Parameter Definition

Symbol Description
A Total warehouse area (square meters), a fixed constant.
Af Forward area (square meters), representing the ”item weight” in the knapsack
problem.
∆A Adjustment unit area (square meters), set to 25 square meters, the basic ad-
justment unit for each forward area change.
CAf Holding cost per unit area of the forward area (cost per square meter per day),
assumed to be twice that of the reserve area.
CAs Holding cost per unit area of the reserve area (cost per square meter per day).
Cpick (∆A) Impact of each area adjustment on picking cost, reflecting the value change of
the ”item” in the knapsack.
Crepl (∆A) Impact of each area adjustment on replenishment cost, reflecting the cost
change associated with area adjustment.
Chold (∆A) Impact of each adjustment on holding cost, calculated as CAf × Af + CAs ×
(A − Af ).
Cshort (∆A) Impact of each area adjustment on shortage cost, reflecting the change in short-
age cost with forward area changes.
µ Average daily demand (units), assumed stable and known.
Continued on next page
Optimization problem

Symbol Description
ρf Storage density of the forward area (units per square meter), used to calculate
the inventory capacity of the forward area.
Frepl (∆A) Replenishment frequency function (times/day), changes with forward area ad-
justments.
T Total available operation time per day (hours), a fixed value as the overall time
constraint.
Trepl Replenishment time (hours per time), a fixed value, does not change with area.
Tpick Picking time (hours per time), a fixed value, does not change with area.
V (∆A) Value of each area adjustment, representing the total cost change (objective
function to minimize).
W Knapsack capacity, corresponding to the total warehouse area constraint.

Knapsack Problem Decision Variables:

• xk : Binary selection variable. xk = 1 if the adjustment unit k is selected (i.e., forward area is adjusted
by this unit), otherwise xk = 0. This variable determines whether a specific adjustment is included in the
solution.

• ∆Ak : Adjustment area (square meters) for unit k. Represents the area size of each adjustment unit
considered in the knapsack problem.
• V(∆Ak ): Value of adjustment unit ∆Ak . Represents the cost impact of selecting the adjustment unit k,
analogous to the value of an item in the knapsack.

New Objective Function

X
Minimize VTotal = V (∆Ak )
k

V (∆Ak ) = Cpick (∆Ak ) + Crepl (∆Ak ) + Chold (∆Ak ) + Cshort (∆Ak )

1. 1. Cpick (∆Ak ): Represents the impact of each area adjustment ∆Ak on the picking cost.
- Formula: Cpick (∆Ak ) = ∆Akk1+ϵ , where k1 is the picking cost coefficient, indicating the improvement in
picking efficiency with increased area.
2. 2. Crepl (∆Ak ): Represents the impact of each area adjustment on the replenishment cost.
k2 ·r
- Formula: Crepl (∆Ak ) = ∆A k +ϵ
, where k2 is the replenishment cost coefficient, and r is the replenishment
frequency constant, reflecting the relationship between replenishment frequency and area.
3. 3. Chold (∆Ak ): Represents the impact on holding costs, calculating the effect of each area adjustment on
the holding costs of the forward and reserve areas.
- Formula: Chold (∆Ak ) = CAf · ∆Ak + CAs · (A − Af ), where CAf and CAs are the holding costs per unit
area of the forward and reserve areas, respectively.

4. 4. Cshort (∆Ak ): Represents the impact of each area adjustment on the shortage cost.
- Formula: Cshort (∆Ak ) = ∆Akk3+ϵ , where k3 is the shortage cost coefficient, reflecting the impact of forward
area size on the shortage cost.

X k1 k2 · r k3

Minimize VTotal = + + CAf · ∆Ak + CAs · (A − Af ) +
∆Ak + ϵ ∆Ak + ϵ ∆Ak + ϵ
k

New Constraints
1. Total Area Constraint
Af + As = A (1)
Explanation: Ensures that the forward area Af and reserve area As sum to the total warehouse area A.
Optimization problem

2. Forward Area Adjustment Constraint:

∆Ak = 25 × k, k ∈ Z+ (2)

Explanation: Each area adjustment must be an integer multiple of 25 square meters, reflecting that each
”item” weight is a fixed adjustment unit.

3. Forward Area Lower and Upper Bounds Constraint:

25 ≤ Af < A − 25 (3)

Explanation: The forward area must be at least the basic minimum (25 square meters) and not exceed
the reserve area’s size.

4. Time Constraint:
(Frepl (∆A) · Trepl ) + (Fpick · Tpick ) ≤ T (4)
Explanation: The total time for replenishment and picking operations each day must not exceed the
available working time T .

5. Inventory Capacity Constraint:

Af · ρ f ≥ µ (5)
Explanation: The storage capacity of the forward area must meet or exceed the average daily demand µ.

6. Non-Negativity Constraint:
Af ≥ 0, As ≥ 0 (6)
Explanation: Ensures that all areas are non-negative, reflecting realistic operational conditions.

7. Integer Constraint:
k ∈ Z+ (7)
Explanation: Specifies that the adjustment factor k must be a positive integer, consistent with adjustments
being in discrete units.

8. Time Constraint:
(Frepl(Af ) × Trepl ) + (Fpick(Af ) × Tpick ) ≤ T (8)
Explanation: Ensures that the total time spent on replenishment (Frepl(Af )×Trepl ) and picking (Fpick(Af )×
Tpick ) does not exceed the available working time T (e.g., 8 hours per day).

Experiment Design Description

The objective of this experiment is to optimize warehouse layout to minimize total operational costs. The
study evaluates algorithm performance in different scales of warehouses using both exact methods and heuris-
tic/metaheuristic algorithms. The key focus is on assessing the algorithms based on total cost, computation
time, and solution stability.

Experiment Objectives
• Optimize the forward area to minimize total operational costs, including picking cost, replenishment cost,
holding cost, and shortage cost.
• Compare different algorithms’ performance in various scales to evaluate their effectiveness and efficiency
in practical applications.
Optimization problem

Experiment Scheme
1. Small-Scale Experiment Design
• Warehouse Scale: Total area set to 2000 square meters, with the initial forward area of 200 square
meters, and adjustment unit of 25 square meters.
• Exact Calculation Method:
– Use dynamic programming or branch-and-bound methods to find the exact optimal solution as a
benchmark.
• Heuristic Algorithms:
– Implement a Greedy Algorithm: Selects the locally optimal choice at each step based on immediate
cost reduction.
– Implement Hill Climbing Algorithm: Iteratively searches for better solutions within the neighborhood
of the current solution.
• Comparison Metrics:
– Total Cost Difference: Compare the total costs of exact and heuristic methods.
– Computation Time: Assess the efficiency of each method.
– Optimal Solution Deviation Percentage: Evaluate the proximity of heuristic solutions to the exact
solution.

2. Large-Scale Experiment Design

• Warehouse Scale: Total area set to 8000 square meters, with the initial forward area of 800 square
meters, and adjustment unit of 25 square meters.
• Metaheuristic Algorithms Comparison:
– Genetic Algorithm (GA): Optimizes the layout through selection, crossover, and mutation operations.
– Particle Swarm Optimization (PSO): Finds the global optimum by simulating the movement of
particles within the search space.
• Comparison Metrics:
– Final Total Cost: Compare the total costs achieved by different algorithms in large-scale problems.
– Convergence Speed: Evaluate the speed at which the algorithms reach a stable solution.
– Computation Time: Compare the efficiency of different algorithms in large-scale scenarios.
– Solution Stability: Measure the standard deviation of solutions across multiple runs to assess stability.

Parameter Settings
• Holding Costs:
– Forward Area: 2 currency units per square meter per day.
– Reserve Area: 1 currency unit per square meter per day.
• Demand Parameters:
– Small Scale: Mean demand is 10 units/day, variance is 9.
– Large Scale: Mean demand is 50 units/day, variance is 25.
• Time Parameters:
– Small Scale: Initial picking time is 12 minutes per round, replenishment time is 30 minutes per round.
– Large Scale: Initial picking time is 20 minutes per round, replenishment time is 35 minutes per round.
• Cost Coefficients:
– Small Scale: Picking cost coefficient k1 = 50, replenishment cost coefficient k2 = 20, shortage cost
coefficient k3 = 30.
– Large Scale: Picking cost coefficient k1 = 100, replenishment cost coefficient k2 = 40, shortage cost
coefficient k3 = 60.
Optimization problem

Experimental Steps
1. Initialization: Set up initial warehouse parameters, demand generation, and cost coefficients based on
the scale.

2. Exact Calculation and Heuristic Algorithms Solving: Solve the small-scale problem using dynamic
programming, branch-and-bound, greedy algorithm, and hill climbing algorithm, and compare their per-
formance.
3. Metaheuristic Algorithms Solving: For the large-scale experiment, solve the problem using genetic
algorithm and particle swarm optimization, and evaluate their effectiveness in large-scale scenarios.

4. Results Comparison and Analysis: Compare the total cost, computation time, and solution deviation
to assess the strengths and weaknesses of each algorithm.
5. Visualization: Use box plots, line charts, and other visualization tools to display the performance of the
algorithms on various metrics.

Summary
This experiment thoroughly evaluates the applicability and effectiveness of various optimization algorithms in
practical warehouse layout optimization by comparing exact and heuristic/metaheuristic algorithms’ perfor-
mance on small and large-scale scenarios. By examining total costs, computation time, and solution stability,
the study identifies the optimal algorithm configurations to guide actual warehouse operational optimization.