Challenges and opportunities for the energy management of sustainable data centers in smart grids

With the increase of cloud computing and internet services, data centers are emerging to satisfy the requirement, leading to incremental energy consumption demand and emissions of green house gases. Thus, integration of renewable energies with the traditional power grid is preferred to reduce the environmental impact and increase energy efficiency, which lead to a demand of energy management strategies to coordinate the energy demand and generation. In this paper, we review the challenges for the sustainable data centers in smart grids with regards to energy management strategies, integration with renewable energies and cyber-attacks and propose possible solutions. Through the analysis of the data centers from the perspective of both smart grids level and micro-grid level, the research challenges and potential research directions in the energy management of sustainable data centers have been discussed.


Introduction
In the modern society, Internet data centers (IDC) are emerging to supply services for dataintensive and digital technologies, e.g., artificial intelligence and autonomous vehicles [1], which lead to the significant increasing power demand of IDC. IDC is composed of four parts, i.e., power equipment, cooling equipment, IT equipment, and miscellaneous components, among which, IT and cooling equipment consume about 90% of the energy of an IDC [2]. It is estimated that the energy consumption of IDC accounts for 1.3% -1.5% of total energy usage in the world [3]. Besides, the energy consumption causes a great amount of CO 2 emissions to the environment. It is reported that information and communication technologies (ICT) contribute to about 2% of the total CO 2 emissions around the globe [4] and emitted 116 million metric tons of CO 2 in 2007 [5]. Thus, new technologies and methods are expected to reduce the energy consumption, energy cost and CO 2 emissions.
One promising way to reach the goals is to integrate renewable energies with traditional power systems. In recent studies, renewable energies such as photovoltaic (PV), wind energies and hydrogen-based fuel cell systems (FCS) have attracted much research and application attention [6][7][8]. Besides the ecological efficiency of the renewable energies, they also bring many challenges, such as forecast of PV and wind energy, energy management of IDC and so on. The energy management of the IDC should take into consideration not only the coordination of various energy sources and the power demand of the IDC, but also the connection of the   Figure 1 presents the research framework of the work. The sustainable data centers receive electricity energy from the smart grids and communicate with smart grids through the two-way information exchange. Enabled by the two-way information communication between the IDC and smart grids, demand response programs can be employed to save monetary cost of the IDC. On the other hand, the two-way information exchange gives malicious attackers opportunity to disturb electricity utility service through cyber attacks. The reduction of green house emission and energy consumption of the IDC can be achieved from two aspects, i.e., integration of renewable energy to supply clean energy and development of energy saving technologies to reduce load. In the present study, we focus on the integration of renewable energy given the top of this paper. To leverage renewable energy generation and energy consumption of the IDC, proper energy management strategies need to be designed.
The rest of the paper is organized as follows. Section 2 analyzes challenges of the integration with renewable energies and the demand-side management is introduced in Section 3. In Section  3 4, we introduce the cyber security problem of IDC and finally the paper is concluded in Section 5.

Integration with renewable energies
To reduce the greenhouse gas emissions, renewable energies, such as solar energy, wind energy and hydrogen energy, are a promising solution [18]. However, these renewable energy, as primary energy, cannot be used directly to supply electricity for electrical appliances [19], thus, requires devices to convert them to electricity, for instance, PV panels, wind turbines and fuel cells.
The main advantages of the renewable energy are their zero green gas emissions and low cost. For instance, wind turbines and PV panels transfer wind and solar energy to electricity and produce no other emissions; fuel cells take hydrogen and oxygen as reactant and produce electricity and water [20]. Besides, the energy efficiency of FCS is higher than traditional diesel engines [21,22].
However, due to the characteristics of these renewable energies, there are inherent disadvantages of each renewable energy, such as incapability of energy storage and intermittent behavior due to atmospheric conditions. Thus, IDC can also be equipped with energy storage systems to remedy those shortcomings. Such energy storage unit includes batteries, ultracapacitor and hydrogen storage tanks.
The integration with renewable energies can be classified into three steps: supplementary, stand-alone and prosumer. In the first step, renewable energies work as supplementary energies of the IDC and the main energy consumption is satisfied by energy drawn from the bulk power system. With the increasing penetration of renewable energies in IDC, the generation of renewable energy can satisfy the self use of IDC and no energy is required from the bulk power system. Thus, the IDC can work as stand-alone modes [23].
If the renewable energy generation exceeds the energy requirement of an IDC, extra energy can be sold back to the bulk power system, thus, the IDC plays a role of not only an energy consumer but also a producer, thus, is called producer. It should be noticed that the third step is not exclusive of the first step [24]. When the power demand of IDC is very low and renewable energy generation exceeds the power requirement, extra electricity can be sold to other customers or utility companies; besides, the recovery heat of IDC can also be supplied to residential households.
One typical research of the renewable energy integration in IDC can be seen in References [23,25], where a hybrid power system contains PV panels, wind turbines, fuel cells and batteries are proposed to enable stand-alone operation of the IDC.

Demand-side management
The integration of renewable energy brings much flexibility and complexity to the demand-side management of IDC. The demand-side management can be classified into two layers: the microgrid level and the smart grids level. In the micro-grid level, an energy management system is proposed to coordinate different energy sources, energy storage systems and load whereas in the level of smart grids, various demand-response programs are proposed to reduce monetary cost.
3.1. Energy management systems and strategies 3.1.1. Energy management systems. The hybrid power system is coordinated and managed by an energy management system. Figure 2 shows the energy management system of IDC, which takes into consideration the integration of renewable energies with the bulk power system. For the standalone and prosumer modes of IDC, the bulk power system is out of consideration of the energy management system. The load is composed of main load that is used for information technology (IT) and service load that is used to guarantee the normal operation of IT. The load demand is collected by the energy management system, and then, allocated to various energy sources based on pre-determined energy management strategies. On the other hand, the available energy information is transmit to the energy management system, and then, control the load to make sure that the important load is fully supplied.
Bulk power system 3.1.2. Energy management strategies. Energy management strategies are key to the energy management system. The goal of the energy management strategies is to minimize or maximize a goal metrics, for instance, the Power Usage Effectiveness (PUE) which represents the total energy consumed by the IDC per unit of energy consumed by the main load, as well as the Carbon Usage Effectiveness (CUE), the Energy Reuse Factor (ERF) and the Green Energy Coefficient (GEC) which takes into consideration of the integration of renewable energies. Different kinds of energy management strategies for hybrid power system have been developed given different goals. These strategies are mainly rules-based and optimization based. For instance, a rule-based energy management strategy is proposed for a fuel cell hybrid power system in [20] and fuzzy logic based energy management strategy is developed for a PV hybrid power system [19]. The rule-based energy management strategies rely on the past experience and skills of experts to make rules for the regulation of energy allocation. Optimization-based energy management strategies can be classified into global optimization, e.g., dynamic programing [26], and real-time optimization, e.g., model predictive control [27,28].

Demand response
IDC is a kind of important load in power distribution system, as shown in Figure 3. Electricity power is generated by various kinds of generators, such as PV panels, wind turbines, diesel engines etc., and, then, transmit to substations and, finally, distributed to customers like residential buildings, industry and IDC by power distribution networks. On the other hand, the electricity generated by distributed renewable energy of IDC can be fed back to the bulk power system, which forms the two-way energy flow between smart grids and IDC. Besides, two-way information flow exists between utility companies and customers. The utility companies send energy usage and electricity pricing information to customers, who then schedule their demand according to demand-response programs. At the end of each time slot, energy consumption  information is collected via smart meters and sent to the utility company for adjusting electricity generation and electricity prices via advanced metering infrastructures (AMI) [29]. The main aim of demand-side management in smart grids level is to encourage customers to consume less power in peak time or to shift their energy consumption to off-peak hours [30]. Among various kinds of techniques of demand-side management, demand-response is a specific short-term tariff or program [30]. In order to engage customers in smart grids, various demand-response programs have been developed [31,32]. These demand-response programs guide customers to schedule or reschedule their consumption by dynamic prices [32]. For instance, time-of-use (TOU) and real-time pricing (RTP) programs are the most common used programs [33]. In TOU, utility companies publish day-ahead electricity prices based on prediction and then customers schedule their energy consumption at each time slot to minimize their energy cost. In RTP, the electricity is updated hourly or even every five minutes.
In order to reduce monetary cost of energy, recent research works focus on methods of demand-response at demand side. Four kinds of price-driven programs with different goals are summarized in Ref. [34], which are reducing the peak-to-average load ratio [35,36], minimizing economic cost [37,38], maximizing utility welfare of customers [39,40] and the improvement of economic cost [41,42]. However, those research directions are mainly specified for residential customers, the optimization goals of IDC should be re-defined and reformulated. Beside these four research directions, the collaboration of multiple data centers in located different geographical areas by proper resource allocation and load assignment methods can also reduce the operational cost of IDC [43]. oil and gas malware injection affect energy generation and delivery Ukraine 2015 power system false data injection more than 225k customers losing power for 1-6 hours USA 2021 oil company ransomware attack 5,500 miles of pipelines shut down

Cyber security
As discussed in Subsection 3.2, the two-way information communication between utility companies and IDC enables the active engagement of IDC in smart grids. However, the associated cyber security problems through cyber networks are non-negligible. The cyber security goals of smart grids include integrity, confidentiality and availability [44]. Various cyber attacks can target one or more goals of the smart grids [31], for instance, fault injection attacks can be adopted to corrupt data and break the confidentiality in cyber-physical systems [45] whereas data injection attacks can interrupt integrity of smart grids [46][47][48] and the availability goal can be unsatisfied by denial-of-service attacks [49].
In Table 1, we have summarized the cyber attacks to energy systems in last decades [7,50]. We can observe from the Table that there is increasing number of cyber attacks targeting the energy systems in recent decades and have caused severe impact. In smart grids, cyber attacks can target the power transmission systems and power distribution systems [44]. For the false data injection to power transmission networks, malicious attackers can inject false measurement data to confuse the state estimation of power system and pass the detection [51]. This kind of cyber attacks can disturb the utility services of substations in smart grids and cause severe impact. Cyber attacks that target power distribution networks mainly focus on disturbing demand-response by injecting false electricity prices [52][53][54][55]. Victims who receive the false electricity prices will schedule their consumption according, and this might lead to economic loss and even overload of the power system [52]. Furthermore, false renewable energy generation information can be injected to smart meters to misconduct the energy management behavior of customers [56]. Thus, the cyber attacks to IDC in both smart grids level, i.e., cyber attacks to power transmission networks and price-based demand-response programs, and micro-grid level, i.e., cyber attacks to renewable energy generation, should be taken into consideration to develop resilient and robust IDC.
To mitigate the impact of cyber attacks, timely detection and defending against cyber attacks are very important [31]. Recent works have addressed the cyber attacks detection and defense problems using various methods. Real-time state estimation is performed to detect bad data in power systems [57]. However, intelligent cyber attacks can easily pass the present state estimator [51], thus, many new methods have been developed, in which machine learning approaches have attracted much attention. Ref. [58] proposes an unsupervised cyber attacks detection method which is based on statistical correlation between measurements. A distributed datadriven approach is proposed to detect stealthy cyber attacks on large scale power systems [59]. For the false data injection attacks to renewable energy, machine learning methods such as long short-term memory networks (LSTM), vector autoregressive models, deep neural networks [12] and convolutional neural networks [13] have been studied.
Another research direction to mitigate the impact of cyber attacks is development of defense methods [60]. Many research works based on Markov decision process and game theory have been proposed to protect power system from cyber attacks [55,[60][61][62]. The former methods can model the decision process of cyber attackers or defenders, but fail to capture the dynamic interaction between them, whereas the game theoretic approaches can model the dynamics of the cyber attacks. In the design and operation of IDC, cyber attacks need to be taken into consideration to improve the resilience of IDC.

Conclusions and future work
In this paper, we review the recent researches of sustainable IDC as part of smart grids. Specifically, due to the rising concern of environmental problems and increasing demand of power consumption, the integration of renewable energy into the development of IDC is proposed as a promising solution. We review the challenges on the commonly used renewable energies such as solar and wind energy and highlight the necessity of hybrid power system. Together with the hybrid power system, the challenges of energy management in both micro-grid level and smart grids level have been studied. Finally, as part of smart grids, the increasing of interaction between the distributed energy generation systems of IDC and the bulk power system is vulnerable to the emerging cyber security problems in smart grids. We review the cyber attacks at power transmission and distribution levels and studied the possible methods to mitigate the impact of such cyber attacks.
In future work, several research directions can be done to reduce energy consumption and green house gases emissions of IDC, and improve the resilience of IDC to cyber attacks. Firstly, different kinds of renewable energies can be employed in the design and operation of IDC. Thus, the optimal installation strategy of renewable energies in the design of IDC, and proper energy management strategies during the operation of IDC should be developed. Then, the coordination of demand-response programs and energy management strategy to reduce monetary cost and energy consumption should be optimized. To provide long-term reliable operation of IDC, intelligent operation systems and fault diagnosis systems can be taken into consideration; and finally, the resilience of IDC can be defined and methods to improve the resilience of IDC should be studied.