Cloudfridge: A Cloud-Based Control System for Commercial

Refrigeration Systems
Marco Graziano, Visible Energy, Inc.
Marco Pritoni, Western Cooling Efficiency Center, UC Davis

ABSTRACT

In California, commercial refrigeration accounts for 14% of electric energy usage,

corresponding to 9,014 GWh per year and affecting more than 110,000 commercial
establishments. A single walk-in freezer uses more energy than 5 single-family houses. Despite
more stringent standards and utility programs promoting hardware improvements, controls
systems used for commercial refrigerators are still primitive. In fact, in traditional refrigerators,
the vapor-compression cycle is controlled with a simple hysteresis controller to keep the air
cabinet temperature within a specific range. The controller does not account for system
dynamics, energy consumption, utility prices or recurrent events such as food loading schedules
and business opening hours. Also, demand response is precluded with these unsophisticated
controls. With the help advanced control theory and efficient optimization algorithms, computer-
based real-time optimization is now feasible and applicable in commercial refrigeration systems,
but its practical use to date has been limited to industrial systems requiring expensive on-premise
equipment and complex operations. This paper presents a novel hardware and software
architecture that allows advanced control algorithms for commercial refrigerators to be
developed, tested and deployed inexpensively. The aim of this new control framework is to
optimize energy consumption as a software task, utilizing the benefits of lower cost
computational resources inherent to cloud computing, minimizing on net overall energy usage of
the refrigeration system. A prototype of the proposed system has been developed and tested
under a California Energy Commission grant1.

Introduction
There are more than 12 million commercial refrigerators installed in the US (Table 1).
The vast majority of them, in particular walk-in coolers and freezers, food preparation equipment
and reach-in refrigerators still use primitive controls. Cludfridge targets primarily this large base
of unsophisticated systems.
In a traditional refrigerator, a thermostat controls the vapor-compression cycle with a
simple hysteresis scheme to keep the air cabinet temperature or in some cases the evaporator
temperature, within a specific range. The thermostat functioning parameters are set at the time of
fabrication and assembly, energy used is not measured and no factors relevant to energy
consumption dynamics are taken into consideration. With the help of advanced control theory
and efficient optimization algorithms, computer-based real-time optimization is now feasible and
applicable in commercial refrigeration systems, but its practical use to date has been limited to
industrial systems requiring expensive on-premise equipment and complex operations.
This project demonstrates a new way of operating commercial refrigerators, reducing in a
cost effective way their energy consumption while increasing their operational efficiency. The

1
Visible Energy, Inc. (2014) Cloud Based Refrigeration Control System, Final Report. California Energy Commission EISG
Program, Grant #: 57509A/1211. http://www.cloudfridge.io/

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-285

 method is deployed as an automated energy optimization service based on the unique
characteristics of each refrigerator and business. Additionally, this approach allows for insights
to be drawn across a wide range of installations.
Network latency and security have always been challenges in the implementation of
cloud-based control systems. Cloudfridge has been developed to provide secure and almost real-
time communication between the device and the remote control algorithm.

Table 1. Total installed base of refrigerators in the US in 2008 (DOE 2009a)

Total  Primary  Energy 
Commercial  Sector  Technical  Potential  by  Installed Base 
Consumption 
equipment type 
[Units]  [TWh / yr]  [TBtu / yr] 
(Supermarket) 
Display Cases  2,100,000  214  730 
Compressor Racks  140,000  373  1,273 
Condensers  140,000  50  171 
Walk‐ins  245,000  51  174 
(Non‐Supermarket) 
Walk‐in Coolers and Freezers  755,000  148  505 
Food Preparation and Service Equipment  1,516,000  55  188 
Reach‐in Refrigerators and Freezers  2,712,000  106  362 
Beverage Merchandisers  920,000  45  154 
Ice Machines  1,491,000  84  287 
Refrigerated Vending Machines  3,816,000  100  341 
Total  1,226  4,183 

Background
Recent scientific literature on advanced control strategies applied to refrigeration or
HVAC shows the potential for substantial energy savings. A well-designed optimal control
scheme, continuously maintaining a commercial refrigeration system at its optimum operation
condition, despite changing environmental conditions, will achieve an important performance
improvement, both on energy efficiency and food quality reliability (Cai et al, 2008a). Several
alternative methods for establishing a control strategy that minimizes the overall energy
consumption in the refrigeration system have been described in the literature. For instance
Jakobsen and colleagues (2001) showed how to optimize set points for operating theoretical
refrigeration systems under certain constrains. Larsen and Thybo (2002) show significant energy
saving using flexible set points. In another paper, Larsen and colleagues created an indirect
method for optimization of the energy consumption (Larsen et al, 2005) and then they applied it
to a refrigerator model, resulting in energy usage reduction up to 20%. In (Stoustrup and
Rasmussen, 2008) food quality together with energy, are used as parameters to determine an
optimal time between defrost cycles. Based on this, a new defrost-on-demand method is
proposed. The method uses a feedback loop consisting of an on-line model updating and
estimation by an Extended Kalman Filter (EKF), as well as a model-based optimization. In (Cai
et al, 2008b) a new defrost, on-demand control scheme is described resulting in 25% energy
saving from performing the defrost cycles at the estimated energy optimal points. Cai also

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-286

 discusses how a well-designed optimal control scheme, continuously maintaining a commercial
refrigeration system at its optimum operation condition, can achieve an important performance
improvement in energy efficiency and without compromising food quality. Earlier studies
proposed fuzzy logic (Becker et al, 1994)) or neural networks (Choi et al, 1998) but they have
not been studied yet with the objective of reducing overall energy consumption. A computer-
based energy management system has been installed in the world’s largest integrated nylon plant
to optimize the refrigeration systems, and it proved a substantial energy cost reduction (Cho and
Norden, 1982). Up to date this research has found little implementation in the refrigeration
market, with a few exceptions in large refrigeration systems. Cloudfridge provides the
architecture to develop these new control methods inexpensively in any type of refrigerator.

Technology Developed
Under a California Energy Commission grant, Visible Energy, Inc. developed a cloud
infrastructure and related applications able to remotely manage, control and monitor commercial
refrigerators2. The goals of the project were to prove the feasibility and develop the infrastructure
as well the hardware and software tools of a cloud-based control for commercial refrigerators.
Its practical use will allow a larger number of research institutions and companies, even
without large R&D budgets, to experiment and test novel ideas for control methods that reduce
energy consumption or improve other aspects of their products using real systems and not only
mathematical models.

System Diagram

Figure 1 depicts Cloudfridge system diagram. The refrigerators on the left side are each
equipped with a purpose-made control and communication board, which connect to the Internet
over local Wi-Fi via a facility switch to reach the Internet and the control software residing in the
cloud. Given an available Wi-Fi network, the control and communication board is the only new
piece of equipment needed for a retrofit. The system does not need a full computer on the
premises. The board is installed outside the refrigerator.
The major innovation of this system is the collection of a large amount of data, including
energy consumption, and using it in the control method, implemented as software in a cloud
environment. Each refrigerator could potentially implement a different method, customized on
its technical and environmental conditions.

Hardware

The control and communication board developed allows both data acquisition from
sensors, including power consumption, and remote control of a refrigerator’s main components,
i.e. the compressor, the evaporator fan, and the defrost element, if present. The refrigerator’s
original thermostat is bypassed and the control loop is performed purely by the software running
remotely. The infrastructure allows near real-time access to the control board in the refrigerator.
A contract manufacturer has estimated the bill of materials for the control board to be about $50
per unit (at scale3). Additional sensors can be installed to measure food temperature, door
2
Open-source software components will be released under AGPL license (www.gnu.org/licenses/agpl-3.0.html).
3
Bill on materials for 10,000 pieces

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-287

 opening and ambient temperature. In case of lost connection the controller switches back to a
traditional fixed set-points operation.

Figure 1. CloudFridge system diagram.

Software

The Cloudfridge software infrastructure is composed of a back-end for 2-way device

secure communication, the sMAP Archiver and database, a highly efficient time series database
developed by UC Berkeley (Dawson-Haggerty et al. 2010), the software for control methods
execution (Workers Cluster), and an application server (Stardome), which provides the user
interface for the whole system. Cloudfridge is a distributed software system structured in four
main components, each of a certain complexity, but all capable of being distributed on a number
of servers. Figure 2 represent a map of Cloudfridge software architecture. It depicts the
computing resources, data flows and management underlying the cloud-resident control method
capabilities.
Cometa4 is the “secure switchboard” edge service5 used for data vetting and assured data
coherence across the Internet trust boundary. Cometa allows secure real-time HTTP
communication between an application server and the Amazon Redshift6 endpoints devices
connected to the refrigerators. Black arrows stand for internal data transfers separate from
Cometa. Colored arrows show specific types of data flows mediated by Cometa, most

4
Cloud infrastructure for embedded systems developed by Visible Energy Inc. http://www.cometa.io/
5
Computer for running middleware or applications that is close to the edge of the network, where the digital world
meets the real world. Edge servers are put in warehouses, distribution centers and factories, as opposed to corporate
headquarters.
6
Amazon data warehouse service

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-288

 prominently on the left side of the graphic, the red arrows representing data traffic from the
control and communication boards (Redshfit) to the sMAP Archiver and its data base(s). The
Workers Cluster runs the actual control method, using query data from the sMAP Archiver. Blue
and yellow arrows represent the command-response messages flowing between the Workers
Cluster and the endpoints. The modularity represented allows broader server distribution and
ultimately better scaling capacity.

Figure 2. CloudFridge software components.

The Stardome GUI interface includes:

1. Device Console for device creation and configuration (Figure 3)

2. Device Management, for receiving real-time data from devices, control method editing
and deployment through the Monitoring Panel (Figure 4) and the Development Console.

In the control execution environment (Astro OS, inside the Workers Cluster) each
refrigerator is associated with an independent thread of execution for its own control method
written in the Ruby language. A control loop can also invoke an external process such as Matlab
or Octave, to perform very sophisticated mathematical computations and use the results in
evaluating the optimal cooling strategy.

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-289

 Figure 3. Device Console in the Stardome application.

Figure 4. Monitoring Panel in Stardome application.

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-290

 Discussion
While the new method used in the test optimization has been achieved by writing the
software knowing the usage patterns and content, the optimization process can be easily
replicated in a class of refrigerated appliances, including higher energy usage ones, such as walk-
in boxes. The control software can be written directly in Ruby or can call external functions, for
example using advanced control tools available in Matlab.
This technology enables customizing a control sequence for each refrigerator based on
use patterns and type of application at very low cost. One can envision implementing demand
response through thermal storage only in vending machines, ice machine or freezers, while
saving energy in other refrigerators by optimizing defrost cycles. All these strategies can be
implemented with the same inexpensive hardware. Regulating set-points based on electricity
prices and anticipating compressor malfunction, thus preventing food losses, could achieve
additional operational cost savings.

Conclusions
The technology developed in this research greatly lowers the cost of implementation and
installation of state-of-the-art control methods in commercial refrigerated systems and self-
contained appliances, thereby lowering major barriers to adoption, whether through retrofit, or
by incorporation into newer refrigeration appliances and systems. Preliminary deployment test
show savings in the order of 20-50%. The technology implemented and demonstrated in the
project has a clear potential to disrupt standard practice in the commercial refrigeration industry.
It provides substantial energy savings benefits, increased operational efficiency, demand-
response capabilities, and fault detection and diagnostics at small marginal cost. Beyond the
initial objective of feasibility, the project has delivered a smart controller reference design ready
for production, and tools and functional software for deployment and commercialization of a
refrigeration energy optimization service.

Recommendations and Future Work

A clear next step for this technology is to develop different advanced controls methods
based on sensors data collected by the system. Using the online platform computation could be
performed in real-time in the cloud, not constrained by limited hardware resources of a
traditional refrigerator. One example of an energy optimized control strategy is called model
predictive control (Rossiter, 2003; Wang, 2009; Boyd 2012), which continuously adjusts the
refrigerator working parameters to optimize a function of energy, peak demand, cost, and food
quality. Machine learning algorithms can be used during operation of the system to automatically
detect usage patterns that can be incorporated in the advanced control method. While preliminary
testing has been done in the lab and at one location in the field, large-scale testing and energy
savings verification is essential to prove the technology for commercial adoption.

References
Becker M., Oestreich M., Hasse, H. and Litz L. (1994). Fuzzy control for temperature and
humidity in refrigeration systems. Proc. 3rd IEEE Conference on Control Applications,
Glasgow, UK, 1994.

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-291

 Boyd S., Hovgaard T., Larsen L., Jørgensen J. (2012). Nonconvex Model Predictive Control for
Commercial Refrigeration. International Journal of Control, 2012.

Cai, J. (2007). Control of refrigeration systems for trade-off between energy consumption and
food quality loss. Ph.D Thesis. Automation and Control Department of Electronic Systems,
Aalborg University, Denmark.

Cai, J., J.B. Jensen, S. Skogestad, J.Stroustrup (2008a). On the trade-off between energy
consumption and food quality loss in supermarket refrigeration systems. 2008 American
Control Conference. Seattle 2008.

Cai, J., J. Stoustrup, B.D. Rasmussen (2008b). An active defrost scheme with a balanced energy
consumption and food quality loss in supermarket refrigeration systems. Proc. 17th World
Congress, The International Federation of Automatic Control. Korea 2008.

CEC (2006). California Commercial End-Use Survey. California Energy Commission. 2006.

Cho, C. H. and Norden, N. (1982). Computer optimization of refrigeration systems in a

textileplant: a case history. Automatica, 18(6), pp. 675– 683.

Choi, B.J., S.W. Han, and S.K. Hong (1998). Refrigerator temperature control using fuzzy logic
and neural network. Proc. IEEE International Symposium on Industrial Electronics ISIE ’98,
Pretoria, South Africa.

CEC (2006). California Commercial End Use Survey. California Energy Commission. 2006.

Dawson-Haggerty S., Jiang X., Tolle G., Ortiz J, and Culler D.(2010). sMAP – a Simple
Measurement and Actuation Profile for Physical Information, Sensys 2010
http://www.cs.berkeley.edu/~stevedh/pubs/sensys10smap.pdf

Jakobsen, A., B.D. Rasmussen, M.J. Skovrup and J. Fredsted (2001). Development of energy
optimal capacity control in refrigeration systems. International Refrigeration Conference.
Purdue.

Larsen, L.S. and C. Thybo (2002). Potential energy savings in refrigeration systems using
optimal setpoints. Preprint version, Aalborg University.

Larsen, L.S. (2005). Model based control of refrigeration systems. PhD thesis, Department of
Control Engineering, Aalborg University, Denmark, 2005.

Rossiter, J. A. (2003). Modelbased Predictive Control: A Practical Approach: CRC Press.

Stoustrup J. and Rasmussen B.(2008). An Active Defrost Scheme with a Balanced Energy
Consumption and Food Quality Loss in Supermarket Refrigeration Systems. Proceedings of
the 17th World Congress The International Federation of Automatic Control Seoul, Korea,
2008.

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-292

 Wang, L. (2009). Model Predictive Control System and Design Implementation using
MATLAB.

©2014 ACEEE Summer Study on Energy Efficiency in Buildings 11-293