SCIENCE CHINA

Information Sciences
. PERSPECTIVE . October 2024, Vol. 67, Iss. 10, 206201:1–206201:3
[Link]

Cloud-based control systems: towards the control

architecture in cloud computing era
Yuanqing XIA1,2
1
School of Automation, Beijing Institute of Technology, Beijing 100081, China;
2
Zhongyuan University of Technology, Zhengzhou 450007, China

Received 24 December 2023/Revised 17 May 2024/Accepted 7 August 2024/Published online 24 September 2024

Era of cloud computing. After Google introduced the computing and control systems holds profound theoretical
concept of “cloud computing”, this paradigm experienced and practical significance.
rapid growth and found widespread applications across var- In 2012, the concept of cloud-based control systems
ious industries, establishing itself as one of the most favored (CCSs) was proposed [2], incorporating cloud computing
business and technological modes. According to the defi- into the traditional control systems. This innovation in-
nition provided by the National Institute of Standards and volved aggregating sensor data in the cloud, enabling online
Technology in the United States [1], “cloud computing is a system identification and modeling, and performing tasks
model for enabling ubiquitous, convenient, on-demand net- such as planning, scheduling, control, and decision-making
work access to a shared pool of configurable computing re- processes of complex systems in the cloud. In recent years,
sources”. This mode, evolved and integrated from grid com- CCSs have shown promising progress in method design, the-
puting and distributed computing, functions like a utility, oretical research, and industrial applications [3]. With the
offering business access on a pay-as-you-go basis, tailored to development and popularization of cloud computing, CCSs
resource needs. Consequently, the emergence of cloud com- will be the control architecture in the future.
puting has brought a transformative impact on the land-
Promising directions of CCSs. Now, we introduce six
scape of computing.
promising directions of CCSs as follows.
Leveraging its service-oriented characteristics, cloud com- (1) Data-driven cloud control. The improvement of com-
puting facilitates seamless expansion into diverse industries, putational ability brought by cloud computing allows for
fostering cross-sector collaboration and cost-effectiveness. the integration of more complex controlled plants and more
For instance, automotive companies such as BYD, NIO, devices into control systems. For these plants and devices,
XPeng, and Great Wall Motor have enthusiastically estab- the precise mathematical models are often challenging to es-
lished intelligent computing centers, covering areas such as tablish, due to the process complexity and involvement of
intelligent connected vehicles (ICVs), autonomous driving, numerous variables. Simultaneously, in such scenarios, we
new energy, security, and experimental prototyping. Mean- can use the sensing technologies to measure system states
while, various regions such as Beijing, Guangdong, and or outputs, containing valuable information. Then, lever-
Chongqing, are constructing industrial Internet big data aging the computational prowess of cloud computing, we
centers, to meet users’ demands for data-driven decision- can harness the vast data, analyze the characteristics of the
making, management, and innovation. These intelligent plants and devices, and establish the relationship from his-
computing centers and big data centers serve as the physical torical data to control inputs. Therefore, data-driven cloud
embodiments of cloud computing in the real world. control can be realized, holding significant importance for
Concept of cloud-based control systems (CCSs). The de- model-free control challenges in cloud environments, such as
velopment and applications of cloud computing have given large-scale multi-agent systems, ICVs platoon control with
rise to a series of emerging industries, such as ICVs, un- unknown faults, and model-varying control problems.
manned aerial vehicle (UAV) logistics, and intelligent man- (2) Model predictive cloud control. The task domain in
ufacturing. These industries bring new transformations and modern scenarios is expansive, characterized by significant
challenges for control systems, as they generate massive challenges posed by environmental disturbances and height-
amounts of data that traditional control architecture can- ened internal model uncertainties within systems. This com-
not adequately process. Control, as the core element in plexity requires advanced control algorithms to address the
these industrial scenarios, directly dictates the levels of ef- demands of complex systems with multiple constraints and
ficiency, performance, and stability. However, traditional diverse optimization objectives. In a cloud environment, if
networked control systems (NCSs) lack the capacity to pro- the precise model can be established and utilized as prior
cess vast amounts of data, suffer from computational insuf- knowledge, model predictive control (MPC) is also an ideal
ficiency, and struggle to meet real-time and high-quality de- strategy. MPC is a model-based optimal control strategy
mands. Therefore, the research on the integration of cloud that determines the future control signals for a system, by
Email: xia yuanqing@[Link]

c Science China Press 2024 [Link] [Link]

 Xia Y Q Sci China Inf Sci October 2024, Vol. 67, Iss. 10, 206201:2

optimizing a performance metric describing system behav- sary to study the reliable collaborative scheme with an edge
ior, while ensuring the satisfaction of constraints. Hence, disturbance observer. The above layers are interconnected
MPC has advantages such as foresight, robustness, and the via networks to achieve overall collaborative control.
ability to handle complex constraints. On the other hand, (5) Security and privacy protection. In CCSs, the phys-
MPC imposes high computational cost requirements, which ical space of the controlled system is connected to the in-
can be met by CCSs that offer ample computational re- formation space, thereby inheriting the security and privacy
sources. Therefore, model predictive cloud control, as the concerns that are inherent to cloud computing. For the se-
integration of MPC and CCSs, can effectively meet the re- curity issue, common threats include denial of service (DoS)
quirements of the mentioned challenges. and replay attacks. For instance, the Stunet malware in-
(3) Workflow-based CCSs. When deploying data-driven filtrated Iran’s nuclear enrichment facilities, accessed his-
control, MPC, and other algorithms in the cloud, how to torical operational data, and conducted replay attacks to
efficiently utilize cloud computing resources for enhancing deceive control systems. In CCSs, security threats in the
computational efficiency becomes a crucial challenge. If the information space become more severe. As for the privacy
controller is simply deployed on a cloud server, it still re- issue, the cloud platform poses a primary risk. Even if cloud
sembles closely to the research approach of NCSs, without service providers comply with privacy regulations, they still
considering the structural features and advantages specific have an interest in the data value and the capability to access
to cloud computing. Cloud computing is a network-based user data. To address these issues, researches are needed
computing model that enables users to access and share con- on security-resilient control and privacy-preserving control
figurable computing resources, such as networks, servers, based on techniques such as differential privacy and homo-
storage, and application services, on-demand, without the morphic encryption (HE) in CCSs. However, both two al-
need to understand or manage the underlying infrastruc- gorithms introduce computational disturbances, so it is im-
ture. From the structure aspect, it abstracts and consol- portant to estimate the bound of disturbance and design the
idates multiple physical computers distributed across dif- disturbance-rejection control laws. Specifically, the compu-
ferent locations into a shared virtual cloud resource pool, tation of HE is slow. Thus, the designs and analyses of cloud
encapsulating them into new logical computers as needed. control laws from the perspective of numerical computation
Therefore, cloud computing allows users to launch multiple and optimization are also significant.
virtual machines or containers in the resource pool to accom- (6) Service-oriented cloud control platform. The domain
plish complex tasks. In practical applications such as genetic of CCSs is not only a theoretical issue but also holds strong
computation and seismic wave analysis, with complex struc- practical applicability and value. For theoretical algorithms,
tures, large datasets, and high-performance requirements, their implementation is dependent on a suitable cloud con-
they are often modeled as cloud workflows with Directed trol platform. Simultaneously, a cloud control platform that
Acyclic Graph structure. Inspired by this, control missions aligns with industry ecosystems and exhibits excellent scal-
are considered as a specific type of computational workload, ability and portability is of significant importance for ad-
that can be restructured into the form of cloud workflow. vancing engineering practices. With the development and
This approach improves the utilization of cloud computing widespread adoption of cloud-native principles, container
resources and leverages the computational advantages. Con- technology has supplanted virtual machines as the main-
sequently, workflow-based CCSs are proposed [4], employing stream infrastructure in Internet production environments.
distributed numerical algorithms and distributed optimiza- For example, to conduct the control tasks with workflow
tion to establish the cloud workflows of advanced control structure, a containerized cloud control platform is designed
algorithms. and implemented in [4]. Furthermore, this platform has
been extended to the cloud-edge environment [5]. Based
(4) Cloud-network-edge-end collaborative control. For
on its advantages of high flexibility, strong reusability, and
complex, large-scale, and widely distributed systems, lever-
less resource consumption, container technology has gained
aging cloud resources is necessary to obtain sufficient com-
prominence. Concurrently, container technology is making
puting power, enhance real-time computation, and improve
inroads into control domains. The cloud control platform,
the capability for global information processing. While
implemented with a cloud-native ecosystem, allows cloud
methods like workflow-based CCS can enhance computing
controllers to be encapsulated as services, invoked on de-
efficiency, the extent of improvement is not limitless. Ad-
mand, realizing control as a service (CaaS).
ditionally, cloud computing is not a hard real-time system
and cannot guarantee strict real-time computation. Fur- In addition, resource scheduling, theoretical stability
thermore, introducing cloud computing into control systems analysis, and the integration of computation, communica-
introduces unavoidable communication delays. The con- tion, and control in CCSs are also important directions.
tingencies for occasional failures of cloud platforms must Next generation: so-called dynamic cloud and its CCSs.
also be considered. On the other hand, with the rapid de- With the advancement of technologies and the develop-
velopment of edge computing technology, computing capa- ment of various industries, whether in civil fields such as
bilities are extending to edge systems. In the long term, aviation, oceanography, fisheries, petroleum, environmental
relying solely on cloud platforms would result in resource monitoring, or in defense and military areas such as recon-
mismatch. Therefore, the cloud-network-edge-end collab- naissance, target strike, and communication relay, the tra-
orative control is required. For instance, the cloud per- ditional single-movement-body platforms struggle to meet
forms high-level decision-making tasks, the edge handles the multidimensional and all-element mission requirements.
small-scale planning tasks, and the terminal executes con- The cross-domain multi-movement-body system is an or-
trol based on planning signals. Depending on the scenarios, ganic entity composed of distinctly functional bodies op-
the cloud-edge-terminal collaborative control can be simpli- erating in different spatial domains. The cross-domain in-
fied to cloud-edge collaborative control. For example, when terconnection and collaborative cooperation among space,
a cloud controller runs, the computing process introduces air, and ground multi-movement-body systems help reduce
intrinsic disturbances and uncertainties. Thus, it is neces- the functional redundancy of heterogeneous platforms, fully
 Xia Y Q Sci China Inf Sci October 2024, Vol. 67, Iss. 10, 206201:3

Figure 1 (Color online) Space-air-ground-ocean cross-domain collaborative system.

leverage the system robustness and complementary capabil- made up of surface and underwater vehicles. These dynamic
ities, and achieve the integrated utilization of various units. clouds can be centrally scheduled and managed by the static
Due to these requirements from the control aspect, cloud cloud. The different dynamic clouds can interact with each
computing may evolve into a dynamic cloud called by the other, and when necessary, they can obtain various service
author [6], combined with the cross-domain collaborative supports from the ground static cloud. In many scenar-
control systems based on multiple clouds. Research on the ios, each dynamic cloud needs to operate independently, en-
so-called dynamic cloud and its CCSs is poised to play a abling autonomous decision and control.
crucial role, possibly serving as one of the architectures for Conclusion. In the era of cloud computing, CCSs emerge
the next generation of cloud computing and control sys- as an effective control solution to meet the demands of com-
tems. Meanwhile, the introduction of CCSs also provides plex systems. Leveraging services such as big data, cloud
a new perspective for addressing the emerging space-air- computing, and cloud storage, coupled with advanced con-
ground-ocean issues. As shown in Figure 1, in the space-air- trol methods, CCSs efficiently facilitate the modeling, plan-
ground-ocean cross-domain collaborative system, the space ning, scheduling, control, and decision-making processes of
cloud, air cloud, ground cloud, and ocean cloud could not complex systems. Currently, the research on CCSs is in its
only form individual sub-CCSs but also interconnect, in- early stages, with numerous meaningful questions awaiting
tercommunicate, and interoperate through the integrated exploration. Furthermore, the integration of the so-called
space-air-ground-ocean network. This system consists of dynamic cloud and its CCSs with the space-air-ground-
five layers, which are the hardware, data, service, mission, ocean cross-domain collaborative system expands the re-
and plan layers. The different dynamic clouds have en- search scopes of CCSs, which will play a crucial role in both
hanced the coverage, resource integration, and centralized the national economy and defense sectors.
management of cross-domain environments, which broadens
Acknowledgements This work was supported by the
the detection and perception scope, as well as the execution National Key Basic Research Program (Grant No.
scenarios of cross-domain collaborative systems. There are 2021YFC2202600).
five critical issues in dynamic cloud and its CCSs, which
are cluster management of dynamic cloud topology, space- References
air-ground-ocean dynamic environment collaborative per- 1 Mell P, Grance T. The NIST definition of cloud comput-
ception, multi-dimensional heterogeneous movement-body ing. National Institute of Standards and Technology, Re-
port 800-145, 2011
collaborative cloud control, cross-domain collaborative sys-
2 Xia Y. From networked control systems to cloud control
tem security, and dynamic-cloud-based intelligent decision-
systems. In: Proceedings of the 31st Chinese Control Con-
making. ference, 2012. 5878–5883
In this system, the so-called static cloud is also encom- 3 Xia Y. Cloud control systems. IEEE/CAA J Autom Sin,
passed. Under specific conditions, the static cloud pos- 2015, 2: 134–142
sesses the capability to manage all dynamic clouds within 4 Gao R, Li Q, Dai L, et al. Workflow-based fast data-driven
the system. It primarily handles tasks with high complex- predictive control with disturbance observer in cloud-edge
ity requirements, leveraging functionalities such as human- collaborative architecture. IEEE Trans Autom Sci Eng,
machine interface, data analysis, and digital twins. Typi- 2024, 21: 2816–2840
5 Shan C, Gao R, Han Q, et al. KCES: a workflow container-
cally, the static cloud is managed by the large-scale com-
ization scheduling scheme under cloud-edge collaboration
puting centers on the ground. The dynamic clouds consist
framework. arXiv:2401.01217, 2024
of a series of homogeneous movement-bodies, including the 6 Xia Y, Xie C, Gao H, et al. Cross-domain cooperative
space cloud composed of spacecraft, the air cloud consisting control and intelligent decision-making of multi-dynamic
of aerial vehicles, the ground cloud formed by the ground agents in space-air-ground integrated network environment.
platforms such as autonomous vehicles, and the ocean cloud Contr Decis, 2023, 38: 1176–1199