A narrative perspective of island detection methods under the lens of cyber-attack in data-driven smart grid

Abstract

As criminals and hackers are developing new methods to interfere with the operation of the power grid, the nature of grid vulnerabilities and threats is continuously evolving. The growing interest in transitioning the unidirectional power system to a bidirectional data-driven modern grid will further escalate these issues. The question of cyber security becomes essential in particular critical decisions such as island detection. The incorrect decision of island occurrence may completely disrupt the operation of a portion of the grid, causing substantial damage to electrical equipment and grid maintenance workers. Fast monitoring and accurate control of unplanned islanding detection are essential for distributed generation-based active networks for providing continuous power supply to critical loads. Considering the above aspects, this paper serves on the perspective of different island detection methods and various aspects of cyber security. The type of cyber-attacks is categorized in terms of their behavior. Key points are discussed about how, when, and in what fashion and degree it can harm all the sectors of the grid, i.e., generation, transmission, and distribution system. Finally, the impact of cyber-physical attacks on the islanding decision system is presented. The research remedies for such measures are also presented. Moreover, a comparison is being made among various island detection methods based on the extent of impact of different cyber-attacks on the operation of these methods. Some promising future solutions for cyber-secure island detection methods are also suggested.

Introduction

The sharp increase in power consumption has led to an increase in distribution generation (DG) requirements to meet the grid demand requirements and the local loads. Smart grid and DG integration are promising approaches for solving the power system demand problem. There has been a considerable increase in the installation of DG near utility distribution system over the past decade [1, 2]. The DG resources such as PV (photovoltaic) cells, wind energy conversion system or fuel cells offer several potential benefits to utility and customers in terms of economic, technical, and environmental aspects. As DG integration is unavoidable due to expanding power demand, the biggest challenge due to incorporating DG into the power system network is unintentional islanding operation. Islanding operation occurs when a part of an active distribution system, which includes DG, is separated from the rest of the network, while still being energized by a single DG or multiple DGs. Unintentional islanding impacts the personal safety of the maintenance workforce and may result in an unsynchronized reclosing action, which can also harm the DGs. Owing to the unplanned nature, it may cause damage to loads due to inadequate voltage and frequency control services provided by DG [1,2,3]. Such undesirable events could be due to tripping of circuit breakers, manual errors, interruption for maintenance services, equipment failure or network reconfiguration [4, 5]. Therefore, the IEEE std. 2003 states clearly that the disconnection of DG from utility should happen within 2 s of islanding inception [6]. Similarly, the IEC 61727 standard also states that islanding detection and DG disconnection should happen at a maximum time span of 2 s [7]. Therefore, the efficient protection and detection of these undesirable conditions are of prime importance in smart grids.

To regulate and efficiently control the electric power grids, digital measurement devices are getting the upper hand in the present time. A smart grid incorporates several computing and communication devices, control and intelligent monitoring, smart meter, and real-time communication that facilitates real-time control between various components in the power system [8]. This digitization of the power grid has led to the growth of voluminous data that are needed to be transferred to various locations of the grid. The protection of these data becoming very crucial in the data-driven intelligent power system. The amalgamation of the cyber system (communication, info-tech (IT)), security, and automated control infrastructure) and physical system in the smart grid leads to the progress of cyber-physical system, and each of system is been handled by its own regulations, protocols, and standards [9, 10]. The digital measurements reflect the accuracy and construction of improvised power system operation and control of the installed sensors, which are vulnerable to unknown events such as device malfunction and cyber-physical attacks [11]. The attacker can fabricate figures, initialize denial of services, or disrupt the communication channel of the measurement device and control center.

History clearly illustrates the exposure of industrial communication and control systems to cyber-attacks. The power outage of 9th January that left nearly all of Pakistan, i.e., approx. 200 million people without electricity [12] is a significant example. Similarly, Mumbai, on 12th October, 2020, suffered due to a small technical glitch that caused its power-transmission a blackout. In March 2019, Venezuela was left lurching in the dark without electric power for more than a week [13]. The blackout left the country with 32 million people in complete disarray [14]. At least 43 deaths were attributed to the blackout [15]. On December 23, 2015, the cyber-attack on Ukraine power grid clearly demonstrated that attackers can easily manipulate internal devices and abuse them as per their needs [16]. The Ukraine power grid cyber-attack left the country with no power for hours [17, 18]. In 2010, a cyber-attack targeted both the hardware and software of Siemens industrial control scheme and exploited its Windows operating system [19, 20]. As a result, 60% of Iran’s Bushehr Nuclear Power Plant (BNPP) remains under threat leading to a constant fear of cyber warfare [21]. Cyber spies entered the US electrical grid on 8th April 2009 that disturbed the complete software programs and power system [22]. On September 26–27, 2007, and in January 2005 a cyber-attack led to major interferences in the electric supply in the city of Rio de Janeiro, affecting more than three million people in Brazil [23]. On September 28, 2003 Italy went into a country-wide blackout except Sardinia resulted from a power failure while interconnection between power and communication network, hindering services of more than 56 million customers [16]. In the same year, between August 14–28, a power failure for 4 days was experienced in central Canada, and the North- Eastern US. This impacted over fifty million citizens, which sums up to 62,000 MW of power, and it took almost 14 days to restore from the entire event [24].

The smart grid has some potential vulnerabilities because of the enormous dependence on communication networks [25]. Steps were taken by DOE (Department of Energy’s) Cyber security Road Map for Energy Delivery System (EDS) [26], North American Electric Reliability Corporation (NERC), Critical Infrastructure Protection (CIP) standard [27], National Institute of Standard & Technology Interagency Report (NISTIR)7628 [28, 29] and National Electric Sector Cybersecurity Organization Resources (NESCOR) report [30] have discussed future grid’s safety and pliability to cyber-attacks. In these circumstances, diverse efforts are being considered as the light shed moment for the digital and the physical world.

The false trigging of the islanding detection signal, if executed by the cyber attacker can lead the smart grid to disrupt power leading to a blackout in major portion of the grid. This will result in severe fluctuation in the islanded grid's frequency and voltage, which may further reduce the power quality and hence reliability of the grid. The reverse of this condition, i.e., sending of non-occurrence of island signal by the cyber attacker even though islanding has occurred can also cause extensive damages. Therefore, it is important to analyze the impact of the cyberattack on various island detection algorithms to develop effective countermeasures for enhancing cyber-physical safety. In continuation with the discussion, this study presents a review of various island detection methods, the elements of cyber-physical systems in a smart grid, various methods of cyber-attacks, and their impact on the normal operation of island detection methods. The contributions of this narrative perspective are as per the following:

• 1. Although there are many papers on island detection methods and cyber security methods in power systems, there are no papers connecting these two. Thus, the author(s) establishes the connection between island detection methods and cyber threats in this work.

• 2. The list of objectives and requirements for cyber-physical security of island detection methods is identified for the first time.

• 3. The analysis on wide aspects of cyber threats on the functioning of different island detection methods are discussed in detail, which is not present in available literature as per the best knowledge of the author(s).

• 4. The probable methods of eliminating the fear of cyber threats for various island detection methods are also dealt with for the first time.

• 5. Finally, a research gap is identified which can motivate the researchers and engineers to think about future cyber-attack proof island detection methods.

The rest of the paper is structured in eight sections. The concept of islanding is explained in "Concept of islanding" section. Various island detection methods, with their advantages and disadvantages, are described in "Islanding detection methods" section. In "Cyber-physical security of smart grids" section, the cyber-physical security aspects of smart grids are dealt. The requirements of the cyber-physical structure of smart-grid are laid down in "Requirements of cyber-physical structure of smart grid" section. The various types of cyber-physical attack in smart grid and general defense measure is mentioned in "Types of cyber-physical-attack in smart grid and general defense measure" section. The effect of cyber threats on different island detection methods is discussed in "Effect of cyber-attack on island detection" section. "Conclusion" section concludes this paper. The future research directions are also elaborated in this section.

Concept of islanding

The islanding happens when the circuit breaker between the utility and a section of the power system containing DG and loads, opens as in Fig. 1. In this condition, the DG continues to feed the local loads. Thus, in islanding situation, a section of the grid is entirely energized by one or more DGs, while electrically detached from the main grid. This condition of islanding operation normally arises due to the following causes:

• a. Fault on the line linking the DG and the utility.

• b. In the absence of precise frequency control.

• c. Failure of the stability between generation and load in the islanded circuit.

• d. Chance opening of the electrical supply subsequently after a system let-down.

• e. The rapid change in the loads in the network distribution system.

• f. Planned downtime for maintenance activities on the grid.

• g. Physical intervention or manual errors.

• h. Natural phenomena, e.g., thunderstorms, earth quakes, volcanic eruptions, etc.

Fig. 1

Concept of islanding condition in power system

There are two basic types of islanding conditions—intentional and unintentional islanding. Intentional islanding is the process of purposely splitting the DG into self-sufficient controllable island regions, to power the local grid [31, 32]. It is performed to plan maintenance activities needed for the main utility grid. Unintentional islanding is an unexpected operation that may occur at any time due to fault or different uncertainties in the electric system [33]. This type of islanding is a risk for system security, as it may create instability to the system network like voltage and frequency unbalance, harmonics occurrence, equipment damage, electric shock, etc. [34, 35].

Islanding detection methods

The islanding detection method is broadly divided into four main sets: local scheme, remote scheme, signal processing-based technique, and intelligence-based method, as shown in Fig. 2 [4, 36, 37]. The techniques are characterized by several factors, like nondetection zone (NDZ), power quality, speed of detection in error, rate of detection of error, and its efficiency with multiple inverter cases [38]. A detailed discussion of these techniques is explained in the following sections.

Fig. 2

Classification of various islanding detection methods [4, 36, 37]

Remote islanding detection method

Remote islanding detection method utilizes the principle of communication infrastructure between utility grid and the DGs. As soon as islanding occurs, immediately a trip signal is shared with DG source, and it deliberately forces a section of the grid into a condition that will guarantee to isolate the DG systems. The remote scheme has negligible NDZ, i.e., it can work even during zero power flow (before islanding) between the main grid and the islanded grid. Additionally, it does not impact the power quality of the system, and it has high reliability and faster response time. Moreover, it works effectively with multiple DG systems. On the other side, owing to its high-priced and complex problem, its implementation on a small-scale system is not preferred [39, 40]. Power line carrier communication (PLCC) scheme, supervisory control, and data acquisition (SCADA) process, transfer-trip method, and phasor measurement unit (PMU) are the variety of schemes that falls under this detection method. A schematic of the remote islanding detection method is shown in Fig. 3, where the main utility grid and DG unit share the information using a communication channel. The summary of various remote island detection techniques is given in Table 1.

Fig. 3

A schematic of remote islanding detection scenario

Table 1 Summary of various remote island detection techniques

Local islanding detection method

This method measures variation in some system parameter, for e.g. voltage, current, impedance, frequency, phase angle, active and reactive power and harmonic distortion on DG side for detecting islanding. The measurement may be direct or indirect or both. On this basis, the local methods are further classified as, passive technique, active technique and hybrid technique. The different types of these techniques have increased rapidly over the last few years [41, 42].

Passive islanding detection technique

The passive islanding detection technique basically identifies islanding by observing passive parameters at the point of common coupling (PCC), such as current, voltage, impedance, frequency, active power, total harmonic distortion (THD), phase angle and other parameters [43]. The flow chart of the passive islanding detection process is presented in Fig. 4. It relies on a pre-set threshold to detect islanding, i.e., when the rate of alteration of a passive parameter exceeds that of a maximum limit, it indicates that the system has become islanded. Some of conventional passive detection schemes are rate of change of power (ROCOP) [44], rate of change of frequency (ROCOF) [45,46,47], phase jump detection [48, 49], rate of change of frequency over power [50, 51], (ROCOV) rate of change of voltage and change in power factor [52], change in impedance [53] voltage unbalance, over/under voltage (OV/UV), over/under frequency (OF/UF), harmonic distortion [38, 49, 54, 55] and sequence component angles [52]. Some modified, passive detection schemes are Kalman filter-based technique, auto-correlation function-based scheme, s-transform-based scheme and Fourier transform-based technique. The setting of the thresholds requires special consideration. Setting a lesser limit can cause trouble with tripping and false islanding detection. However, if the limit is too high, it will result in islanding conditions being undetected. This method cannot sense islanding if both active, as well as reactive power, are balanced in between the pre-islanded grid and the main grid. Therefore, the method suffers from large NDZ. The passive detection method has some major advantages, such as not impacting the power quality or grid operation, being cost-effective and having fast detection speed [48, 56]. A summary of various passive island detection techniques is given in Table 2.

Fig. 4

Algorithm for passive islanding detection method [43]

Table 2 Summary of various passive island detection technique

Active islanding detection technique

The active islanding detection technique deals with the grid by injecting perturbation into the system variables directly. The active islanding method involves a feedback method to find islanding through parameter changes. An external fabricated signal such as voltage, current and harmonic signal participate with power system operation and brings significant changes in system parameters during islanding, triggering the island detection signal. The effect of these injected signals is not noteworthy if the DG is not islanded from the utility grid [57, 58]. Figure 5 depicts the basic flowchart of the active islanding detection. Impedance measurement (IM) [53, 59], active frequency drift (AFD) [54, 60], frequency jump (FJ) [38], virtual inductor [61], virtual capacitor [62], (phase locked loop) PLL perturbation method, reactive power export error detection (RPEED) [5], slip mode frequency shift (SMFS) [63,64,65], Sandia frequency shift (SFS) [42, 66, 67], Sandia voltage shift (SVS) [37, 68, 69], active frequency drift with positive feedback (AFDPF) [70], high frequency signal injection [71, 72], fluctuations in active and reactive power [37, 38, 54] and negative sequence current injection [73, 74] are examples of active islanding detection. This method is more effective and accurate in comparison to the passive method of islanding detection. The active method has relatively small NDZ, but has a slow response toward error detection rate. Moreover, it requires additional electronic equipment in the system in order to inject perturbation. Furthermore, this method increases the system complexity and requires additional detection time to observe perturbation response on the power system. Since it interferes with an additional signal during the entire operation of the grid, it significantly lowers the power quality of the system and its stability. A summary of various active island detection techniques is given in Table 3.

Fig. 5

Algorithm for active islanding detection method

Table 3 Summary of various active island detection technique

Hybrid islanding detection technique

The hybrid islanding detection technique runs on the principle of active and passive islanding detection methods to remove the problem with both techniques. For example, large NDZ is the main problem in the passive detection method and power quality issue in the active method due to perturbation. Such problems can be overcome by a hybrid technique [75]. During hybrid detection of islanding, the passive technique works as the primary method, whereas the active function works as the ancillary method, as depicted in Fig. 6. It can be effectively applied to complex systems [76, 77]. The hybrid methods can improve multiple performance indices. However, the aforementioned technique escalates the system cost, along with large islanding detection time. Voltage unbalanced and frequency set point-based method [78], the average rate of voltage change and real power shift technique [79], a method based on the rate of change of reactive power and load connecting technique [80], hybrid SFS and Q-F method [81], voltage fluctuation injection [82] are some example of the hybrid islanding detection technique. Each islanding detection technique has its own pros and cons. The choice of the islanding detection method depends on the DG technology used and the complexity of the grid. The hybrid detection method with the mixture of both agile passive and a precise active technique seems to be the answer. However, this hybrid method has some unresolved issues that include accuracy, high detection time, and compatibility with multiple DGs. These restrictions can be overpowered by signal processing techniques and intelligence-based methods. A summary of various hybrid island detection techniques is given in Table 4.

Fig. 6

Algorithm for hybrid islanding detection method

Table 4 Summary of various hybrid island detection technique

Signal processing-based method

Signal processing-based methods are used to further improvise the results of passive as well as an active detection method. The signal processing method has helped many researchers in understanding the islanded mode of operation regardless of its location of control. Such methods have the properties such as conformability, cost-effectiveness, stability, and ease of alteration, which has helped the researcher in the extraction of the hidden characteristics of the measured signal to detect the islanding. Figure 7 shows the step involved in carrying out islanding detection. The signals such as voltage or current are taken from the PCC, and features are extracted with signal processing tools. The features obtained are then compared with a threshold to identify island conditions. Common signal processing classifier used in islanding detection are wavelet transform (WT) [83, 84], s-transform [85,86,87], discrete fractional Fourier transform (FT) [88], Hilbert Huang transform (HHT) [89, 90] and spectral kurtosis [91]. Over the last decade, the signal processing methods have received huge attention from researchers owing to their flexibility. A summary of various signal processing method is given in Table 5.

Fig. 7

Algorithm for signal processing-based islanding detection method

Table 5 Summary of various signal processing-based island detection technique

Intelligence-based method

Intelligence-based islanding methods are similar to that of the signal processing-based method. In the signal processing method, the input parameter signal is compared with a fixed value. The range of this fixed value is hard to determine. To obtain this threshold value automatically, an intelligence-based method is used with the signal processing method for islanding detection. Figure 8 shows an elementary flow chart diagram of the intelligence-based method with a signal processing technique. The classifier is trained from numerous simulations done beforehand. After it is properly trained, the extracted features obtained from the signal processing technique is fed to the classifier for classification of island conditions [92]. The commonly used intelligence-based method associated with signal processing-based islanding detection method includes decision tree (DT) [93,94,95], artificial neural network (ANN) [96,97,98], probabilistic neural network (PNN) [76, 99], support vector machine (SVM) [100, 101], fuzzy logic control (FL) [102,103,104], adaptive neuro-fuzzy interference system (ANFIS) [105] and random forest (RF) [106]. A summary of various intelligence-based method is given in Table 6 and the comparison between different island detection methods is made in Table 7.

Fig. 8

Algorithm for intelligence-based method islanding detection method

Table 6 Summary of various intelligence-based island detection technique

Table 7 Comparison between different island detection methods

From the above discussions, it can be observed that the island detection methods depends on the components of smart grid, especially its communication structure. Thus, the components of the smart grid must be in a healthy state for the island detection methods to work perfectly. An attack on any of these structures may cause maloperation of the islanding algorithm. Thus, it is necessary to have a knowledge of the cyber-physical security of the smart grids in order to assess the island detection algorithms in terms of cyber security.

Cyber-physical security of smart grids

Implementation of electricity transfer over grids with cyber integration has an extensive impact on power system organization. During the last few years, several researchers have comprehensively studied the smart grid, its cyber-physical system structure, security and infrastructure. Figure 9 illustrates the typical type of cyber-physical structure of a smart grid. Smart grid infrastructure of electric power system is generally categorized into generation, transmission, and distribution system [107]. Smart grid infrastructure advancement has made cyber-physical security a serious challenge for power system operation. The infrastructure generates extra issues for warranting its security. Availability, integrity and confidentiality of available data are priorities of a cyber-physical system. Cyber attackers aim and threaten these elements so as to influence the data being transferred for operation and control [108]. Cyber threat on these can affect the system component and its infrastructure. Thus, cyber security mechanism in data-driven grids is needed to prevent a rise in physical and cyber-attack.

Fig. 9

Cyber-physical structure of smart grid [107]

Physical security

Integration of power system smart devices and cyber systems in physical structure introduces new challenges, as most of the devices and systems are insufficient to security features against vulnerability. Lack of sufficient protection against cyber-attack can result in equipment damage, personal safety hazards, and grid protection device. Thus, this allows the attacker to produce high impact on the physical system. Therefore, in order to spot the attacker at physical device layering, various control devices should be installed and monitored. Energy buffer is also used to detect errors and cyber-attack at points of common coupling and buses. Further, the smart inverter can also detect cyberattacks, system anomalies and other events.

Cyber-security

Similar to physical security, cyber security of the grid needs to incorporate physical aspects. Cyber security is a component in the expansion of the smart grid [109]. The attacker may lead the system to disastrous consequences by utilizing the mechanism of the control system to lock the operator out of the system. Therefore, an extensive range of cyber security is required to prevent the attacker from acquiring access over the cyber grid network. A secured wired and wireless network must be provided for reliable connection between the control center, substation& actuator. Numerous approaches are developed to check both network communication and control devices within the system. Smart inverters, SCADA system measurement, smart meters, and WAN networks (Wide area network) are the basic databases that can be trusted to identify malicious threats.

Generation system attack

The power system is a complex enterprise and is managed by its analogous control devices; generating station is one of them. Electrical power generated is step-up so that the load generation balance can be maintained to meet the dynamic load demand. In order to maintain balance frequency and power, load–frequency control should be monitored. Load–frequency control manages turbine speed and generator output by using a primary controller while the secondary controller manages system frequency. An attacker can manipulate data in these controllers by opening and closing the circuit breaker of the generator. This instability in generating system desynchronizes the generator and ultimately hampers the stability. In addition to this, the attacker can mislead the frequency controller into performing fake frequency determination for load shedding in the generation system, which exerts significant loss and delays in the generation of power [110,111,112].

Transmission system attack

The electric power system is the bulk movement of generated electric power across a long distance through a transmission network to an electrical substation. Long distance transmission network requires reliable communication for control of the power grid, while voltage regulation is operated in the transmission system. Fault-sensing protective relays and circuit breakers are equipped at each end of the line to cut-off faulted conductors and overloaded lines. Protection of the transmission line from cyber-physical attack is usually so critical that it has inspired researchers to investigate the potential of attack targeting the monitoring, communication, and control of power transmission. In the meantime, vulnerabilities in transmission networks are continuously evolving. Some of the important known area of threats where a malicious attack can bypass or access control of the system are transmission line protection breaker control, substation automation mainframe, digital relay computers, communication network, etc. Manipulated control commands or false measurement injected by attacker leads to load shedding, multiple tripping and finally resulting in the massive blackout of transmission grid [113]. Interconnected power transmission grid modeled as multiple substations, where multiple measurements and complex control algorithm control are there. A minimal disturbance in operation, results in sequential loss of substation and its transmission line. An informed attacker can penetrate malicious attacks and gain full control of the substation and transmission network. False data from the attacker’s end can manipulate the voltage controller configuration, which results in voltage oscillation and voltage violation in the system [114]. The switching system subsequently steered to an unstable operation state, which can lead to frequency instabilities and cascading failure in the system [115, 116], which results in tripping of lines and circuit breakers. Communication with the help of GPS (global positioning system) satellite in a smart grid transmission system are more accurate and provides time stamp to all PMU data so that measurement can be synchronized to enhance the interconnected power system. However, the attacker can spoof the GPS signal to provide a fake time stamp, which induces error in the transmission line, resulting in miscalculation. Spoofing also changes the clock offset of PMU, which increases false alarms in the voltage stability control system, leading to a shutdown of transmission services.

Distribution system attack

The power distribution system is the stage in the delivery of electric power. The distribution substation is connected to the transmission system to manage the transmitted voltage at the distribution transformer located near customer’s premises. Smart meters have been installed in the distribution system of the smart grid for two-way communication between grid and customer for advance metering infrastructure (AMI). Advanced metering infrastructure is composed of a controller, energy display, and communication network. Communication from the grid to smart meter to the customer may occur via signaling and GPS. Bringing functionality like this to a power distribution system without incorporating security increases the probability of attack in large-scale areas like reasons of countries and cities [117]. The victims may face network blockade or blackout for an extended period of time [118].

Requirements of cyber-physical structure of smart grid

Cyber-attack prevention includes communication security, protecting sensor data, system architecture control security, access to system security control, machine learning-based algorithms to detect irregularity, cloud-based information security, communication system security, cybersecurity based on game theory approach, encryption, data privacy/integrity, filtering of bad-data, pre-estimation of bad-data, cryptographic algorithm and encryption mechanism [119]. Smart grids need to define their section of security control deriving from a standard, such as ISA (Industry Standard Architecture) [120], (National Institute of Standard and Technology) NIST SP800-53 [121], ISO 27002 (International organisation of Standardization) [122], NERC CIP (North American Electric Reliability Corporation Critical Infrastructure Protection) [123]. Figure 10 presents an overview of NIST updated smart grid infrastructure model with a composite high-level view of seven different areas of generation, operation, market, transmission, distribution, service provider, and customer [29, 29]. Each area and its sub-domain initiate technical roles and assistance for the smart grid. The generation unit produces electric energy with innovative systems and technologies in the smart grid, which involves nuclear fossil fuels, tidal force, hydroelectric, wind, and solar [124,125,126]. The deepening amalgamation of the renewable energy system is helping to improve the sustainability of these generation systems.

Fig. 10

Revised smart grid conceptual model by National Institute of Standard and Technology [29]. PMU, Phasor Measurement Unit; T-RTU, Transmission Remote Terminal Unit; TIED, Transmission Intelligent Electronic Device; ISO, Independent System Operator; RTO, Regional Transmission Organization; LFC, Load Frequency Control; ED, Economic Dispatch; UC, Unit Commitment; EMS, Energy Management Systems; OMS, Outage Management Systems; BSM, Bulk Storage Management; WAMS, Wide-Area Management System; DMS, Distributed Management System; DR, Demand Response; MDMS, Meter Data Management System; AMI, Advance Metering Infrastructure; CIS, Customer Information System; ES, Energy Storage; DG, Distributed Generation; EV, Electric Vehicle; D-RTU, Distributed Remote Terminal Unit; PLC, Programmable Logic Circuit; IED, Intelligent Electronic Device; GIS, Geographic Information System; FCT, Field Crew Tool; DS, Distributed Sensor; DDC, Distributed Data Collector

Transmission and distribution systems use smart substations, which can be remotely handled with a sensor and smart electrical devices. The PMU utilizes GPS to provide high resolution and reliable measurement in the transmission system, aiding the execution of wide-area monitoring, protection and control over high-speed communication [127]. The AMI is implemented with an intelligent meter, sensor, and sensor network in the distributed system that provides innovative two-way, real-time communication in the smart grid which monitors demand response, consumer engagement, energy management, and power quality deficiencies. The energy storage, electric vehicles, industrial area and other emerging technology on customer side continuously demands up-gradation to the generation, transmission, and distribution of electricity.

The smart power system (smart generation network, smart transmission network, smart sub-station and smart control center) depends on a communication system for stable and dynamic control as the system requirement is based on the information and IT security. Therefore, the communication system in smart grid has inducted numerous cyber infrastructures with the power system. Computation and information are continuously initiated and transmitted between power system and cyber systems. On the basis of this computation and measurement, optimal control policies are determined in order to handle the situation and command to coordinate actuator in the smart grid power system. Communication in these intelligent grid mainly requires SCADA system and network, or PLCC system. The fact that SCADA system work for remote site and network, therefore, are being unified with external system, creates new possibilities in power systems. This is because, SCADA system and network are more reliable, accurate, provide information in real-time, can give trend-analysis report, allow system to perform dynamic maintenance without any operation hindrance, which increases life-span of the system equipment. But on the other hand, they are extremely vulnerable to threat. As a consequence, ground of cyber-physical structure of smart grid is to conceptualize for every specific domain (generation substation, transmission substation, distribution substation, operation unit, market, service provider and consumer). Therefore, new communication standard and protocol such as DNP-3 (Distributed Network protocol) and IEC-61850 (International Electro-technical commissions) have been developed and introduced for communication in smart grid, substation and control centers in order to accommodate the integration of PMU, renewable energy and energy storage. This provides an atmosphere that involves possible digital entrance and its exposures.

Types of cyber-physical-attack in smart grid and general defense measure

This section of work offers a synopsis of various possible cyber-attacks in accordance with the cyber and the physical systems. Cyber-attack refers to a scheme to manipulate or introduce the potential vulnerabilities in the cyber system assembly of data-driven smart power grids [128, 129]. The devices such as smart inverters, battery controllers, and cloud data storage introduce vulnerabilities as the communication network connects them to a central monitoring and control station. The communication network generally has remote access to the control center for operators via a communication channel. The attacker may attack the communication system and accomplish a malicious cyber-attack on the communication protection relay. Eventually, it trips the circuit breaker, resulting in blackouts [130, 131]. Another critical threat in the smart grid is measurement-based attacks [132]. In this, the attacker manipulates the measured data and weakens the situational awareness by misleading control action, i.e., by reporting false contingencies. Further, a malicious act in the smart grid known as volumetric attack, in this kind of attack the attacker transmit a substantial amount of network-traffic to the recipient, thereby debilitating the smart grid system or cramping the network infrastructure [133]. The most common outcome of these attacks on the smart grid can be an operational failure, synchronization loss, interruption in power supply, severe financial damage, complete blackout, social welfare damage, data theft, and cascading failure [134]. It may impact the overall power system supply at all levels, including power system control, generation, transmission, and distribution.

Another method of cyber-attack includes barring the communication for data transfer. Barring the communication is accomplished by overflowing the communication channel with massive data traffic or by introducing a mechanism that triggers a communication crash. This phenomenon eventually leads to the deprivation of data to legitimate users. Another cyber-attack includes replaying/resending the same data over a long time to misdirect the legitimate users to do a particular action as per the hacker's desire [135]. Another practice of cyber threat is a malware attack, where an attacker injects malicious software to control systems to disrupt and destroy the conventional operation of the digital network by manipulating the threshold value. Another type of program attack is a web threat, where the attacker intrudes and keeps track of routing information in the web application. In this type of attack, the attacker's inclination is toward peculiar information, for example- a password or initiating a command. In a cross-site-scripting attack, multiple malicious data content is communicated in the transmission channel toward target browser; when the target network operation is set-forth, malicious data is broadcasted and exposed to vulnerabilities. Covert attack, in this type of attack the attackers uses systems specific information in order to monitor and control the transmit signal, accordingly the attacker construct a concealed and secure attack substructure with the help of systems’ precise information, which further injected in channel of measurement and actuation [136,137,138]. Thus, the following attacks are repeatedly followed by cyber-intruder for malfunctioning the communication links of data-driven smart grids.

• 1. False data injection attack

• 2. Denial-of-services attack

• 3. Replay attack

• 4. Malware-virus attack

• 5. Web-attack

• 6. XXs-attack

Since monitoring of generator unit, transmission system, substation system, smart inverter, smart meter, communication between the control center and actuator produces a huge amount of data. These data are to be kept in the cloud-based system or any remote high-security data centers; these data are more vulnerable to launch attacks. Extensive investigation on cyber threat schemes has announced grid operators establishing various defense mechanisms, divided into two stages- protection and detection.

Protection

A cyber-physical structure like a smart grid contributes a large volume of data. Protection of these data is very critical. Protection against cyber threats generally relies on the building of secured communication, conserving critical information, monitoring intrusion, use of secure software, controlling access to system and data, smart inverter, SCADA system measurement, smart meter, WAN network, physical security, cyber security, encryption, meter safety, data privacy, data integrity and mitigation of exposed vulnerabilities. Along with this, the direction of approach with technical execution (network, and physical security, encryption, meter safety, data privacy, and data integrity) instigate the system security. Mixed integer liner program optimization, game-theoretic approach, graph-based approach, state estimation approach and meter measurement approach etc. [139,140,141,142,143,144] have been proposed for false data injection type attack. Method for determining load-distribution attack is developed in [145]. Data accumulation in AMI of a smart meter is one of the primary targets for attackers [146]. Therefore, a secure distribution mechanism and key management have been proposed as the best protection against unauthorized access to smart meters [147]. Cloud-based information and communication technology is one more target for attackers. Therefore, a secure and flexible information management structure with cloud computing configuration is developed in [148]. The architecture of cloud computing consists of two parts—(a) Front-end and (b) back-end. Front-end is comprised of web servers and contains user side interface to access the computing platform, while the back-end is owned by a service provider, which includes an enormous amount of virtual machines, data storage, security-mechanism, traffic-control-mechanism server, and deploying models. For secure communication, cloud computing has four major components—(a) infrastructure services, (b) software services, (c) platform services and (d) data services. And have four main clusters for configuration—(a) general user service, (b) control and management services, (c) information service, and (d) electricity distribution service [149].

Detection

Despite efficient threat protection for the smart grid, an attacker still have control over the exposed or vulnerable component. In case of such protection failure, machine learning-based bad-data detection algorithm [150], sensor-based bad-data detection [151], filtering of bad-data detection [152], pre-estimation-based bad-data detection [153], model-based approach [154], game-theoretic approach [155], hypothesis testing detection method [156], improved Kalman filtering method [157], and federated learning-based detection mechanism [158] have been an obvious choice to provide effective security and countering against cyber threat. These methods, as mentioned above, identifies five significant types of intrusion in the system—(a) detect an attack from their signature, (b) detect the attack by deviating from the normal system behavior, (c) combination of anomaly and misused base method [159], (d) detect the attack at event management security of information system, (e) detects an attack from there communication pattern on transmission traffic. Furthermore, these methods are capable of identifying attacks that are unnoticeable.

Effect of cyber-attack on island detection

This section of the paper investigates the effect of the cyber- attack on island detection algorithms. Cyber-attack can manipulate and restrict data flowing through communication channels, which endangers island detection methods' success. Thus, a proper analysis is required, and possible solutions must be found.

False data injection attack

False data injection attack, a classification of integrity attack is the major category of cyber-attack in island detection. In this, the attacker target and modify the measurement vector and sabotage the real-time data [160]. The attacker induces false data to unstabilize the operation to disrupt the islanding detection technique. When the attacker induces false data to manipulate the threshold value of the relay setting to a different value, a false triggering happens. Even if there is no fault in the system, the attacker can inject false data, which leads to false islanding detection that could lead to load shedding and unnecessary generation rescheduling.

In the remote islanding detection method, the basic principle is communication between the main utility grid and distribution generation via the SCADA system, PLCC scheme, and the transfer trip method. As discussed in "Remote islanding detection method" section of this paper, these methods use computers, servers, and various electronic devices, which are very prone to false data injection attacks. Attackers can induce false data in any server or device to disrupt the operation, which may lead to a failure of the island detection method. Therefore, the remote island detection method requires physical security as it uses various electronic equipment and cyber security, as the communication and transmission are via server or modem, as discussed in "Cyber-physical security of smart grids" section. Various methods discussed in [152, 153, 161] can be introduced in remote techniques to detect island conditions and protect the system from cyber-attack.

Furthermore, in the case of local island detection method (passive island detection technique, active island detection technique, and hybrid island detection technique), frequency, voltage, and harmonics are monitored and injected respectively against a pre-defined threshold limit for detection. The attacker can inject false data to manipulate threshold values which could affect analyzing the parameters. This false data injection attack can be in generation, transmission, or distribution systems, which may lead to a fake island or no island detection. This leads to the failure of the local island detection method. Therefore, a graph-based approach [142], linear program operation [139], and protection method by machine learning-based bad-data detection algorithm [150] are required in local techniques. Further in-case of the signal processing island detection method and in the intelligence-based island detection method, a false data injection attack may work if the signal processing unit rapidly adapts new data format.

Denial-of-services attack

A denial-of-service attack is a data intrusion type attack, i.e., the attacker does not allow any information to pass through the transmission to receiving side. In this, a blockage mechanism is introduced so that a communication channel is obstructed between the remote and the main grid. In a smart grid distribution system, energy and information flow are bi-directional via communication links like modem, wireless point, server, and monitor control. Therefore, cyber-attack is always possible. Attacker target devices like PMU, µPMU, USB (Universal Serial Bus), Intelligent electronic devices, smart meter, main computer server, smart inverter and communication network which are connected to GPS system. Thus, when island condition occurs, the information is not transferred, leading to communication failure, and hence the DG remains unaware of the island situation. Significantly it affects the performance of the power system in the islanded grid.

As the above-mentioned attack is a data-driven type of attack and remote island detection is based on communication infrastructure, an attacker can easily obstruct the communication channel leading to the failure of the method. Therefore, a secure and flexible structure with cloud computing configuration [148] and filtering of bad data [152] protection and detection can be developed in remote techniques. Similarly, in case of local island detection methods (passive island detection technique, active island detection technique, and hybrid island detection technique), an attacker cannot affect passive and hybrid methods if they work on the traditional ROCOF method on not receiving any signal, but active methods fail to detect this denial-of-service attack as the technique deal by injecting perturbation into the grid system variable. Therefore, secure configuration [149] and pre-estimation-based bad-data detection [153] can be introduced to make it cyber-proof. For the signal processing island detection method and intelligence-based island detection method, the denial-of-service attack may be avoided if the island detection method uses the traditional ROCOF method on not receiving any signal.

Replay attack

Due to the interdependence of all islanding detection methods on wireless sensor communication channels, the attacker may use multiple identities and attack the traffic flow of the signal. As the attacker in replay attack has previous snap short of legitimate data, which can mislead the system for a longer time leading to fake or no island detection. Also, in a replay attack, the attacker gets access to the control system of the smart meter or inverter and injects a fake or modified signal into the system, causing false triggering.

The remote island detection method uses PLCC scheme, SCADA network system and control devices etc., for secure cyber communication between generation-transmission-distribution systems. Thus, replay attackers can get easy access to the control system of SCADA, PLCC, and smart inverter etc. Therefore, the remote island detection method fails to detect replay attack disturbance. Thus, a new approach for protection, such as methods used in [140, 141, 152, 155] can be used against replay attacks. In the case of local island detection method (passive island detection technique, active island detection technique, and hybrid island detection technique) replay attackers may get detected as they use a time-shifted version of the original data, but the detection totally depends on the nature of GOOSE method. Further, signal processing island detection method and intelligence-based island detection method may work against replay attacks with the use of cloud-based information and communication technology [148], machine learning-based bad-data detection algorithm [150], sensor-based bad-data detection algorithm [151], filtering of bad-data detection [152], and pre-estimation based bad data detection [153].

A similar explanation and remedial measures follow for the malware-virus attack, web attack and XXs-attack. The remote methods fail under malware-virus attacks and web-attack. However, XXS-attack may not work on SCADA and transfer tripping. The active methods fail under malware-virus attacks and web-attack. However, XXs-attack may be avoided by using the proper disturbance injection method. The case is reversed for passive methods where it is able to handle malware-virus attacks and web-attack but not XXs-attack. Also, the signal processing and intelligence methods are unable to check XXs-attack. Hybrid methods are somewhat able to handle malware-virus attack, web-attack, and XXs-attack.

Thus, in case of false triggering of islanding, achieved through an attack, that particular islanded grid gets compromised and loses power leading to a blackout in the region. Further, if there is a real islanding situation, but the attacker formulates the outcome, and no signal is received, then the islanded portion will encounter severe fluctuation in voltage and frequency, which damages the load to an intense degree. This false islanding detection during the normal condition and also the other way round results in frequency and voltage differences between islanded grid and the main grid. For reconnecting with the main grid, frequency and voltage in this islanded grid must be synchronized accordingly with the main grid. If this synchronization is not done as per the main grid, then a huge circulating current can flow in the grid, causing extensive damage. Thus, under the real undesirable islanding scenario and false islanding scenario, the reliability and resiliency of the grid get reduced due to the cyber-attacks. Another major consequence of cyber-attack is to the maintenance team of the islanded grid, who are ignorant of the actual scenario of the islanded condition of the grid and are still in action. Thus, developing a cyber-secured island detection algorithm must be of prime importance.

The comparison of the effect of cyber-attacks on various island detection methods is given in Table 8. It can be seen from the table that the remote methods are most affected by the cyber-attacks as their principle of working is based mainly on the communication network. For active methods, a denial of services attack disrupts its functioning since the effect of introducing the disturbance injection is required for its functioning, which depends on the communication system. For false data injection, active, passive as well as hybrid methods may overcome the threat if it uses a proper virtual portable network (VPN). Application of a proper generic object-oriented substation event (GOOSE) message can obstruct replay attack to some extent for all the local island detection methods. If the passive, hybrid, signal processing, and intelligence methods works on traditional ROCOF on not receiving any signal, then it can avoid denial of services attacks to a limited extent. The signal processing techniques may rapidly adapt to new data formats and detect bad data to counter false data injection and replay attacks. The intelligence technique may use deep learning techniques to detect false data and replaying if data. Thus, as signal processing and intelligence technique are advanced in handling cyber threats, these island detection methods are least affected by cyber threats.

Table 8 Comparison of effect of cyber-attacks on various island detection methods

Conclusion

The objective of the paper is to motivate the researchers to develop an island detection method that is capable of handling cyber threats. The interval span of the survey is considered from 1990s to the present time. However, to stress on the recent methods, the majority of the research papers considered in this survey is taken after 2010. Although there are many papers on island detection methods and cyber security methods in power systems, no papers connect these two according to the authors' best knowledge. Thus, this paper tries to establish the connection between island detection methods and cyber threats. The paper analyses the island detection methods in terms of cyber security rather than reviewing any papers which deal with cyber-proof island detection methods (as there are no papers available). The objective is achieved by reviewing the prevailing islanding detection techniques and the theory, mechanism as well as control techniques of cyber-attack. After that, the effect of cyber-attack on island detection methods is further investigated. Moreover, probable measures to counter these issues in island detection methods by using machine learning techniques, computational techniques, filtering techniques, model-based approaches, etc., have also been discussed. Hence, the paper provides a narrative perspective of island detection methods under the lens of cyber-attack in a data-driven smart grid. This narrative perspective hints at opening new research gates for building cyber-secured IT-based island detection methods in the future. Some of them are mentioned as per the following:

• 1. Designing of a universal standard required for secured data communication in island detection methods.

• 2. Modifications of old protocols or introduction of new protocols for threat-free data transmission.

• 3. Utilizing advanced machine learning techniques, blockchain technology, and statistical-based new island detection methods capable of mitigating cyber-attack efficiently.

• 4. Analyzing cloud-based island detection approach possessing the ability to dynamically learn from past attacks and capable of self-healing after errors.

• 5. Revisiting smart inverters and smart meters in a smarter way to provide a tailored solution to cyber-attacks.