Optimizing Warehouse Efficiency

Red Cedar Technology, Inc.

The warehouse is divided into six zones and has

Introduction seven employee types. The pallets are delivered to
the warehouse by supply trucks. Servicing staff
Agent-Based Modeling (ABM) is an effective strategy
(unloaders) unload and transport the pallets to the
for simulating the emergent properties of complex
reception zone where the pallets are handled,
systems. This form of simulation models the
marked by registrars (acceptors), and then moved to
behaviors of individual agents interacting with each
the storage zone with forklift trucks. When the
other and with their world and can be used to
warehouse receives an order, a forklift picks a single
predict the outcome of discrete events, such as the
pallet from storage and transports it to a control
behavior of a warehouse. Many frameworks exist for
zone where the controlling staff matches the pallet
creating ABMs—one such program, used here, is
contents with the order. Transfer employees then
called AnyLogic [1].
take the pallet from the control zone to the dispatch
While ABM works well for modeling individual zone. At the dispatch zone, a retail truck pulls up and
behaviors to determine the resulting emergent the pallet is loaded onto the truck. The pallet is then
group properties, it is difficult to find optimal delivered to the retail store that placed the order.
warehouse layouts, configurations, and personnel Figure 1 shows a depiction of the various zones and
allocation that allow the warehouse to function as employees.
efficiently as possible. Parameters must be tuned
To summarize, the seven employee types and their
and tested to find optimal settings.
jobs are:
Unfortunately, the simulation time required to test
Unloaders - unload pallets from storage
all possible parameter combinations is prohibitive,
trucks and deliver pallets to reception zone
making this approach impractical. In this case, it is
advantageous to use a highly efficient optimization Acceptors – receive and scan in goods at
tool such as HEEDS MDO. reception area and move pallets to
placement zone
While most applications of HEEDS focus on design of
physical parts and products, HEEDS can also be Forklift truck operators – take pallets from
applied to many other forms of optimization. This placement zone and put them into storage.
paper discusses the success of HEEDS in the area of When an order comes in, they take the
optimizing warehouse efficiency, using the agent- order from storage and deliver it to the
based modeling tool AnyLogic. control zone.

Wholesale Warehouse Model Controllers – match orders with appropriate

pallets in the control zone
The agent-based traffic model is written in AnyLogic.
It utilizes the “Wholesale Warehouse” Transferers – take orders from control zone
demonstration model included with AnyLogic [2]. to dispatch zone
This simulation model uses the AnyLogic Network
Storage object set to predict the performance of a Loaders – take pallets from dispatch zone
wholesale warehouse, including employee and load on retail trucks
efficiency, zone efficiency, and storage efficiency.
Retail trucks – after pallets are loaded,
deliver to store that placed order
 Figure 1: Wholesale Warehouse model in AnyLogic with different zones shown and with arrows indicating various
warehouse employees.

overworking employees or overloading a given zone

with pallets.
When the model is executed, the warehouse is
initially empty. Supply trucks begin to come in, as Maximizing worker efficiency, storage, and retail
well as orders. The supply trucks deliver goods every truck efficiency are competing goals and will be
20-40 minutes and it takes 5-10 minutes to unload handled in a multi-objective Pareto optimization
the trucks. Orders arrive every 8-24 minutes and the using HEEDS MDO and the search method MO-
warehouse has the capacity to have 20 orders SHERPA.
queued up at a time. The retail trucks take 3-8
minutes to deliver to the stores and return, and it As mentioned previously, it takes a few days for the
takes 3-8 minutes to load the trucks. With an initially warehouse to fill and stabilize in its operations, due
empty warehouse, it takes a few days for the storage to starting with an empty warehouse. Therefore, the
shelves to become stocked and stabilize. performance of the warehouse will be judged on the
tenth day of operation. The average efficiency for
that day will be used to judge a given worker
Optimization Goal configuration.
The goal for the warehouse, as defined in this paper,
The optimization problem statement can therefore
is to maximize warehouse efficiency. This efficiency
be defined as:
is defined as a combination of employee efficiency,
retail truck efficiency, and storage efficiency, and Objectives: Maximize Worker Efficiency
translates into maximizing profits. The goal is to
determine the number of workers needed in each Maximize Retail Truck Efficiency
area, as well as the number of retail trucks required
to maximize this overall efficiency, while not Maximize Pallet Storage Efficiency
Subject To: Retail Truck Efficiency < 1.0 Starting Design

Unloading Worker Efficiency < 1.0 The wholesale warehouse is initially configured with
24 employees and 5 retail trucks and drivers:
Reception Zone Efficiency < 1.0
Retail Trucks = 5
Acceptor Worker Efficiency < 1.0
Unloading Workers = 3
Placement Zone Efficiency < 1.0
Accepting Workers = 3
Forklift Worker Efficiency < 1.0
Forklift Workers = 7
Pallet Storage Efficiency < 1.0
Controller Workers = 3
Controller Worker Efficiency < 1.0
Transferer Workers = 5
Control Zone Efficiency < 1.0
Loading Workers =3
Transferer Worker Efficiency < 1.0
With this configuration, the warehouse operates
Dispatch Zone Efficiency < 1.0 with an average employee efficiency of 58%, pallet
Loader Worker Efficiency < 1.0 storage efficiency of 79%, and retail truck efficiency
of 86%. Figure 2 shows the metrics for this
th
By Varying: 1 < Retail Trucks < 20 configuration within AnyLogic on the 10 day, while
Figure 3 shows the status of the warehouse at the
1 < Unloading Workers < 20 th
end of the 10 day.
1 < Accepting Workers < 20

1 < Forklift Workers < 20

1 < Controller Workers < 20

1 < Transferer Workers < 20

1 < Loading Workers < 20

We have specified that only integer values be used.

The constraints imposed during the optimization
ensure the employees are not overworked and a
given zone is not overloaded with pallets. The zone
efficiencies are calculated based upon the following
capacities for each zone:

Unloading Zone Capacity = 20

Reception Zone Capacity = 20

Placement Zone Capacity = 20

Pallet Storage Capacity = 250

Control Zone Capacity = 10

Dispatch Zone Capacity = 20

Figure 2: Wholesale Warehouse performance for the baseline design on the tenth day. Average employee
efficiency: 58%; average retail truck efficiency: 86%; average pallet storage efficiency: 79%.

Figure 3: Wholesale Warehouse status at the end of the tenth day for the baseline design.
Optimization Results both pallet storage efficiency (>99%) and retail truck
efficiency (100%), however, it performs poorly in
The three objective Pareto optimization performed
terms of employee efficiency (55%). Design A,
with HEEDS MDO found designs that drastically
meanwhile, performs well in terms of all objectives:
improved the efficiency of the warehouse. In
employee efficiency = 72%, pallet storage efficiency
addition, the optimization provided a set of designs
> 99%, and retail truck efficiency > 98%. Figure 6
that were superior in at least one of the three
shows the metrics for the Design A configuration
objectives, called rank-1 designs. This allowed the th
within AnyLogic on the 10 day.
user to select the design that best met the goals of
the warehouse, seeing the tradeoff between all Table 1 compares the three designs highlighted in
objectives. Figure 4 shows a three-dimensional plot Figure 5 along with the baseline design. Design A
with the three objectives plotted, while Figure 5 improves employee efficiency by 13.8%, pallet
shows a plot view of the Pareto fronts (figures are storage efficiency by 20.1%, and retail truck
from HEEDS POST). The blue circular points in Figure efficiency by 12.5% over the baseline configuration.
4 represent all feasible designs evaluated, while the These types of efficiency improvements can
larger red square points correspond with the rank-1 correlate into large cost savings. Here, HEEDS was
designs in Figure 5. The ideal design in Figure 4 able to reduce the number of required warehouse
would be at the front, top corner of the 3D cube, employees by 6 (an accepting worker, 2 controlling
where all efficiencies are highest. workers, a loading worker, a retail truck driver, and
an unloading worker) and eliminate one required
Highlighted in Figure 5 are the three designs that
retail truck.
would be deemed optimal for a given objective.
Design A is optimal in terms of retail truck efficiency;
Design B is optimal in terms of employee efficiency; References
while Design C is optimal in terms of pallet storage
1. AnyLogic 6 Professional, XJ Technologies, 2012.
efficiency. However, clearly Design A is the top
performing design overall, when comparing all three 2. AnyLogic 6 Professional, XJ Technologies,
designs. While Design B performs very well in terms Wholesale Warehouse Example Model, 2012.
of employee efficiency (73%), it is poor in retail truck
efficiency (37%), and sub-par in pallet storage
efficiency (94%). Design C performs well in terms of

Figure 4: 3D relation plot from HEEDS POST showing the feasible designs evaluated (blue circles) and the rank-1
designs from the Pareto optimization (larger red square dots).
Figure 5: Rank-1 Pareto front depicted with multiple 2D plots in a plot view in HEEDS POST. Highlighted are the
three designs deemed optimal. Design A is selected as having the best trade-off among the three objectives.
Figure 6: Wholesale Warehouse performance for the optimized design on the tenth day (Design A from Figure 5).
Average employee efficiency:71.9%; average retail truck efficiency: 98.6%; average pallet storage efficiency:
99.3%.

Table 1: Summary of design characteristics for the baseline configuration, as well as the three deemed optimal
designs from the Pareto front of Figure 5.

Design
Baseline A B C
Employee Efficiency 0.581 0.719 0.728 0.551
Pallet Storage Efficiency 0.792 0.993 0.941 0.999
Retail Truck Efficiency 0.861 0.986 0.376 1.000
Accepting Workers 3 2 2 2
Controlling Workers 3 1 1 1
Forklift Workers 7 7 7 14
Loading Workers 3 2 2 2
Retail Trucks 5 4 11 3
Transferer Workers 5 5 5 6
Unloading Workers 3 2 3 3