A Case Study on National Electricity Blackout of Turkey

Abstract

The necessary precautions should be taken in order to prevent service interruption during the maintenance and repairing of electricity networks. Among these measures, emergencies that may occur in the network should be foreseen, hazard scenarios should be created, and solutions should be developed. If these are not done, a blackout, which first follows the local regions and eventually results in the collapse of the national electrical network, may take place. In this study, the national blackout of Turkey that occurred on 31 March 2015 is examined. The information about Turkey’s electrical infrastructure and its energy policies was provided, as well as the reliability assessment criteria for power systems and examples of significant blackouts that occurred worldwide. The direct relation between line voltage and system frequency was provided with mathematical derivation by using real data taken from a local industrial zone. Then, a case study is presented to demonstrate this direct relation. The causes, development process, and consequences of the blackout are discussed in detail, and some recommendations are offered to increase the security of the electrical infrastructure and to prevent future occurrences while ensuring the sustainability of the system.

1. Introduction

Currently, electric energy in Turkey, as well as many other countries, has a limitation that it must be produced when it is needed and used when it is produced. This limitation has to be taken into proper consideration by central planning and the synchronous maneuver during the production, transmission, and distribution activities. The European Network of Transmission System Operators for Electricity (ENTSO-E) was established in the early 1950s. It was started by 7 members (Germany, Britain, France, Italy, Spain, Poland, and Turkey) and has since grown to 24 members in recent years. ENTSO-E countries, which are dependent on each other in terms of energy supply, aimed to create a valid uniform market model with network synchronization. Hence, the efficiency of energy production was maximized, and the utilization of transmission and distribution systems was increased. Later, the infrastructure investments of electrical transmission systems have rapidly increased since the 1970s [1].

Turkey has the largest transmission line (TL) length compared to other European countries and its installed transmission system has the fifth-highest capacity among them. The Turkey Electricity Transmission Company (TETC) is responsible and authorized for supplying the electricity demand in a timely, uninterrupted manner and is operating at high voltage levels (154 kV and 400 kV). Turkey’s total installed capacity was recorded as 44.8 GW in 2009 while it reached 104.04 GW at the end of January in 2023.

Nowadays, the electricity generation of Turkey is provided by 30.3% from hydraulic energy, 24.4% from natural gas, 21% from coal, 11% from wind power, 9.2% from solar energy, 1.6% from geothermal energy, and 2.5% from other energy sources [1]. The production system is managed from a central load dispatch (Gölbaşı) and nine branch load dispatches (Adapazarı, Çarşamba, Keban, İzmir, Gölbaşı, İkitelli, Erzurum, Çukurova, and Kepez). In addition, 400 kV transformer centers and the power plants higher than 50 MW are also monitored instantly from a central control unit.

According to 2023′s data, the transmission network of TETC consists of 73,732 km TL, 788 substations, and 11 interconnection lines with neighboring countries [2]. With the Balancing and Settlement Regulation (BSR), TETC was removed from the distribution section of the electricity sector on 3 November 2004. After that, private entrepreneurs started to take over the electricity service in this sector. With this regulation, the prices for loading and unloading from the electric network have started to be determined by market participants’ offers. All reconciliations have been made hourly since 1 December 2009. By considering the balance of the supply and demand of electric systems, a wholesale market based on competition with a new structure was established. In this competitive environment, the number of private entrepreneurs has increased in energy service [1].

The new installation of the generation and transmission section requires coordination and optimization solutions for enhancing energy efficiency and reliable operation. Any outages in those sections lead to catastrophic consequences [3].

In order to address this challenge, it is necessary to conduct a reliability evaluation of these systems. Reliability assessment is crucial, as it quantifies the capacity of a device or electrical system to deliver consistent, uninterrupted power of superior quality to users within predetermined operational parameters and within a specified timeframe [4]. Reliability assessment is to characterize the degree of operational reliability and reliability indices are the parameters of reliability characterization.

The reliability indices represent a statistical analysis of a specific attribute across an entire system, operating region, substation service territory, or feeder [5]. Most often, these reliability indices are calculated as average values derived from probability distributions. These indices gauge the frequency, duration, and severity of contingencies occurring within the network, offering valuable insights into the system’s performance [6].

Distribution level reliability utilizes two types of reliability indices: customer-based reliability indices and load-based reliability indices. Load-based indices consider the connected kilovolt–ampere (kVA) load, while customer-based indices give equal weight to each customer. Among the customer-based reliability indices, the widely used ones are the System Average Interruption Frequency Index (SAIFI) and the System Average Interruption Duration Index (SAIDI) [6].

SAIFI calculates the average frequency of sustained interruptions experienced by each customer per unit time interruptions (5 min or more of zero voltage per incident) in the distribution network. It is calculated as the annual ratio of the total number of sustained interruptions to the total number of consumers (number of sustained interruptions in a year divided by the number of consumers) [4,7]. It can be improved by reducing the number of sustained interruptions experienced by customers [5]. SAIDI calculates the total of all customer interruption durations during a year by the number of customers served. It is usually specified in customer minutes or customer hours of interruption per year. It is the ratio of the annual duration of interruptions (sustained) to the number of consumers (total duration of sustained interruptions in a year/total number of consumers). A reduction in SAIDI indicates an improvement in reliability [7]. It can be improved by reducing the number of interruptions or by reducing the duration of these interruptions [5].

In addition to SAIDI and SAIFI, the power margin (PM) is a term used by energy managers and engineers to indicate the energy system’s capacity and reliability [8]. It shows the difference between the amount of power that can be supplied and the amount that is actually required, helping to determine whether an energy grid has enough power supply to meet demand at any given time. PM is important for a more efficient and sustainable approach to energy production and consumption. If the power supply of an energy grid is not sufficient to meet demand, PM is low, which increases the risk of grid failure or system collapse. PM is a crucial measure used by managers and engineers to assess the capacity and reliability of an energy system, and it is an essential factor in ensuring efficient and sustainable energy production and consumption [9].

There are several ways to improve the reliability of a power system. One of them is to reduce the number of contingencies by using mechanical devices, electronic systems, and passive electrical components. For example, automatic switches and monitoring systems [6,10], communication networks [11,12], automated circuit re-closers (ACRs) [13], smart re-closers [14], optimal capacitor placement [15], transformers, and automatic manual disconnectors [4] can be used. Another way to improve reliability is to reduce the impact of contingencies by using power–electronics-based systems and devices. For example, distribution automation systems [6], distributed generators (DGs) [16], smart grid integration with distribution automation [17], islanding mode operation of micro-grids [18], and distribution flexible AC transmission systems (DFACTS) [19] can be used. These measures can minimize the values of SAIFI and SAIDI by eliminating breakdowns in the electrical network, thus the reliability of the power system increases.

Among these measures, DG is the most preferred solution to improve the reliability of power systems. For instance, the impact of ACRs and DGs on the reliability indices of electrical power systems was investigated in [13]. To conduct the study, an IEEE 34 test feeder system was modeled using MATLAB, and the Monte Carlo method was employed to determine the optimal placement of ACR and DG for the test system. Three different scenarios were carried out. The first scenario involved the installation of one ACR on the test system, the second scenario involved the installation of two ACRs, and the third scenario involved the installation of two ACRs and a 1 MW DG. The results of the study indicated that SAIFI and SAIDI were reduced by optimal placement of ACRs and DG in the test system.

There are numerous studies related to blackouts in the literature, which can be grouped into three main topics: detection of blackouts, load shedding, and restoration strategies. The studies related to the detection of blackouts are discussed in the following section.

A decision tree predictor based on three stages was proposed for estimating the size of the possible blackout [20]. New brittleness indices were obtained by using a wide area measurement system and they were computed online at each time instant using operational dynamic data. The proposed predictor was evaluated by modeling a 39-bus New England test system and the Iran 1063-bus power system in simulation case studies. According to the case results, the different sizes of a potential blackout were estimated with respect to a given operational condition of the power system. It could be used as an alarming system to activate emergency actions in the case of estimating a severe blackout.

Furthermore, an introduced screening methodology relied on sequential contingency simulation of cascading outages, incorporating a probabilistic analysis and a visualization model [21]. The expected energy not served was estimated by simulating the potential outages in the power system against severe scenarios. The Korean power system was modeled to evaluate the proposed methodology in simulation case studies. The cascading outages and severe scenarios were consecutively applied until thermal limits were reached. According to the results of the case studies, the voltage violations were alleviated or dropped below the thresholds, respectively.

A frequency simulation model was proposed by considering frequency dynamics and possible cascading events [22]. The model was based on a platform that considered the frequency dynamics and power flow distributions, and it also took into account the protection devices and supervisor control schemes. Hence, the power flow distributions were calculated jointly. The effectiveness of the proposed model was validated by modeling the IEEE 39-bus system in simulation case studies. According to the results, improving the frequency regulation performance was found to be of great significance in coping with the challenges of blackouts. The important role of frequency insecurity in causing a power blackout was also demonstrated. In addition, two indices were proposed to evaluate power system robustness [23]. These indices were employed to detect critical lines that experienced overload, resulting in cascading outages and blackouts within the power system. The initial index introduced was a linear index, while the second index relied on graph theory metrics. The performance of the proposed method was demonstrated by modeling the IEEE 118-bus test system in simulation case studies. According to the results, both indices were capable of identifying critical lines. Therefore, the proposed indices could be used as reliable metrics to apply preventive actions against possible cascading outages and blackouts.

A wavelet energy-based method was proposed to detect stable and unstable power swings for distance relays [24]. The computation involved determining the angular velocity of an equivalent machine created using the available measurements at a specific bus. Its advantages were less hardware and less time required to determine whether the resulting power oscillations were stable or unstable. To evaluate the performance of the proposed method, simulation case studies were carried out on IEEE 14-bus and 30-bus systems. According to the results, the energy computed locally assisted the tripping decision of the relay near a bus, due to the fault somewhere in the system. This criterion behaved globally and acted locally, and the performance of distance relays was improved in detecting stable and unstable power swings. Furthermore, a comprehensive false data injection (FDI) model was proposed to analyze the resilience of smart electricity grids against randomly generated sample FDI attacks [25]. To simulate and forecast the power system response following an FDI attack, a steady-state AC power flow was employed, considering the outage model. To test and analyze the proposed model, simulation case studies were carried out on the IEEE 300-bus test network by considering the load shedding and voltage instability metrics. According to the results, the FDI attack did not have a significant effect on direct overload or large-scale blackouts. However, the grid could be heavily affected by voltage fluctuations that are strongly dependent on the intensity of the FDI attack.

Using power flow tracing (PFT), an optimization model was developed that incorporates overload protection and emergency control measures. Additionally, a multi-stage optimization method, using an improved brainstorm optimization algorithm, was developed to obtain the final result [26]. The unified optimization model with power flow tracing (PFT) integrated various feasible protection and control measures. In the simulation case studies, a 5-bus system and a 29-bus system were modelled as test systems to demonstrate the effectiveness of the optimization method. Based on the outcomes of the case study, it demonstrated a reduction in load loss compared to previous studies. The optimal protection and control strategy with the proposed improved brainstorm optimization algorithm (IBSO) in multiple stages was also obtained. Three-layer IoT-based security architecture in alignment with cisco standards was proposed [27]. The system, designed to detect, warn, and respond to instability and global network blackouts, was developed using a fog-computing platform as its foundation and implemented accordingly. The proposed method was tested in real time applications by creating serious scenarios. In the case studies, it was used to diagnose and intelligently react against instability threats. The case results prompted the proposal of a three-layer architecture for the Internet of Things (IoT) based on fog computing. This architecture aimed to effectively implement a system encompassing a diagnosis, warning, and instant failure response, with detailed descriptions provided [27]. An improved cascaded structure based on the proportional integral with a fractional operator coupled with the proportional derivative was proposed [28]. The controller (1 + PD) was modified by introducing the fractional properties that improved its tracking efficiency and mitigated the frequency fluctuations taking minimal time. The proposed controller was tested under a practical load-changing scenario with the worst load. The optimal parameters of the proposed controller were extracted by deploying a dragonfly search algorithm. In simulation case studies, a power system consisting of renewable energy resources such as photovoltaic, wind, hydro, geothermal, etc., was modeled to evaluate the performance of the proposed method. According to the case results, the proposed method displayed efficient performance in response to changes in load demand under high penetration of renewable energy sources and uncertainty in the power system.

A flexible three-stage controlled islanding strategy was proposed. To identify the imminent blackout and predict its location, the approach utilized the adjusted single machine equivalent (SIME) and the onset of island-sponsored transient instability, respectively [29]. For the simulation case studies, the IEEE 10-generator 39-transport grid was modeled to evaluate the performance of the proposed method. According to the case results, the transient reliability estimation was fast and reasonably precise. An algorithm was proposed for creating an information system to evaluate the readiness of medical facilities [30]. It was based on the categorization and prioritization of facilities necessary for the resumption of electricity supply after a blackout. The preparedness of the healthcare facilities was determined by the proposed algorithm’s ability to solve a power outage. The algorithm was verified at 20 medical facilities. For the simulation case studies, the readiness of medical facilities in the Czech Republic was modeled to evaluate the performance of the proposed method. According to the case results, the proposed algorithm provided quick information about missing generators and fuel at a given time. Thanks to this system, the hospital’s crisis manager would have immediate access to this information. Studies associated with load shedding are discussed in the following section.

A hybrid algorithm based on genetic algorithms (GA) and particle swarm optimization (PSO) was proposed for under voltage load shedding [31]. The proposed method reduced the local search time of GA by using PSO. The fast voltage stability index (FVSI) was used to identify weak buses in the system, and the optimal amount of load shed was calculated to restore the system’s stability. The IEEE 30-bus test system was modeled in simulation studies to evaluate the proposed load shedding scheme. According to the results, the proposed algorithm and GA significantly improved the voltage magnitude of weak buses. The FVSI values were used to select the weak buses for load shedding. However, PSO could not prevent the voltage decline. Another adaptive under-frequency load shedding scheme was proposed [32]. The methodology relied on supervising the distance relay zone 3 decisions for all tie lines connecting different areas. Furthermore, phase measurement units were employed to initiate under-frequency load shedding actions, aiming to prevent severe split conditions that could potentially result in a complete system collapse or blackout. Simulation studies were conducted on the 400 kV Turkish transmission systems to demonstrate the effectiveness of the proposed scheme. The results clearly showed that the frequency stability problem in the power system was effectively improved, as the existing under-frequency load shedding schemes that rely only on qualitative offline analyses for severe contingency scenarios were insufficient.

For the creation of an event-based under-frequency load shedding scheme, a particle swarm optimization (PSO) approach utilizing the least-upper bound method was proposed [33]. PSO was utilized to minimize the amount of load shed while maximizing the frequency nadir without the need for power deficiency estimation. In the simulation case studies, the Kinmen offshore standalone power system (38 buses) of the Taiwan Power Company was modeled to evaluate the proposed load shedding scheme. The results showed that the proposed scheme outperformed the current utility scheme and an existing method based on estimation of power imbalance.

Furthermore, a novel under-frequency load shedding (UFLS) scheme was introduced specifically for islanding distribution networks. This scheme aimed to identify the optimal combination of loads to be shed from the network [34]. The binary evolutionary programming, binary genetic algorithm, and binary particle swarm optimization were used to determine the optimal combination of loads that needs to be shed from the islanded distribution network. In the simulation case studies, a high penetration of solar photovoltaic generation was modelled by using PSCAD/EMTDC to evaluate the proposed load shedding scheme.

Based on the outcomes of the case study, the scheme exhibited rapid response times and successfully identified the most optimal loads. These results highlighted its suitability for real-world application in an islanded distribution network.

The studies associated with the restoration strategy are given in the following section.

A robustness area technique was proposed to measure power system robustness levels, as a helper for planning power system restorations [35]. The motivation was on account of the latest blackouts in Brazil, where the local independent system operator encountered difficulties related to circuit disconnections during the restoration. In the simulation case studies, the São Paulo State power grid was modelled to evaluate the proposed technique. According to the case results, it can be used as a helper tool for the planning of power system restorations. A power system restoration planning strategy based on the combination of heuristic and discrete artificial-bee-colony optimization methods was proposed [36]. The heuristic technique was used to find an initial solution that was close to the optimal way. In addition, the scheme took into account various restoration constraints such as the availability of black start generators, maintaining a balance between load and generation, and ensuring acceptable voltage magnitudes within each island. In the simulation case studies, IEEE 39-bus, IEEE 118-bus, and 89-bus European systems were modelled to evaluate the proposed technique. According to the case results, the optimal solution for large and complex systems was achieved by using the proposed strategy. Additionally, it gave fast restoration time by means of comparison with other literature. An ultrafast grid restoration (UGR) process was proposed. The scheme was founded on three key principles: automation, machine behavior, and process simplification. In order to enable under grounding and restoration (UGR), supplementary functions had to be incorporated into the remote-control system (Supervisory Control and Data Acquisition (SCADA)) [37]. In the simulation case studies, 16.7 Hz Swiss railway power generation, 132 kV and 66 kV transmission grids, and substations of Swiss Federal Railways were modelled to evaluate the proposed technique.

Based on the case results, the scheme successfully reduced the power grid restoration time from approximately 5 h to just a few minutes. This significant reduction aimed to minimize the impact on railway customers, specifically reducing train delays and cancellations [37].

An innovative energy management system (EMS) was proposed for MVDC networks [38]. The system used a power sharing strategy to proportionally distribute power between different renewable energy source units in the voltage regulation mode. This was achieved by controlling generation units between maximum power point tracking and implementing a smart load shedding and restoration algorithm based on network parameter feedback. The proposed system was evaluated by modelling two parallel-operated PV farms, boost DC–DC converters, and a bidirectional boost converter. Based on the outcomes of the case study, EMS effectively verified the distribution network functionality of the affected MVDC (medium voltage direct current) system. It achieved this by ensuring a constant DC bus voltage, maximizing the duration of critical load supply, and ensuring the MVDC system’s readiness for black-start operations.

A parallel restoration strategy based on the A* algorithm was proposed for a power system black start [38]. The proposed strategy managed the start-up of main units with priority in the black-start units, restoration of inter-connection transmission lines, and restoration of important loads. Less important loads were restored afterwards. The strategy was evaluated by modelling IEEE 39- and IEEE 68-bus systems. According to the case results, the proposed strategy minimized the restoration time by using black-start units and tie lines from neighboring transmission system operators, creating separate islands that were finally synchronized.

The modern power system keeps advancing towards a huge complex artificial system. The interactions between electrical elements can become highly complex, to the point where even a minor disturbance can initiate a chain reaction throughout the entire system, commonly referred to as cascading [22,28]. Additionally, the recovery scenarios should be prepared by considering the worst situations. If they are not properly planned, local or national outages will take place. Maintaining the frequency of the electrical network within a specific range is crucial to ensure the uninterrupted operation of electrical elements. Otherwise, many frequency-related protection devices may be triggered and then result in severe consequences. Thus, it can be inferred that the frequency dynamics play an important role in the process of cascading failure [22,32,33,34]. The blackout that occurred in Europe on 4 November 2006 can be given as a first important example. The interconnected European electric grid was unpredictably split into three independent areas after the successive outages of overloaded lines. The frequency of Area-1 dropped quickly to about 49 Hz. It resulted in a load shedding of about 17 GW. Meanwhile, the frequency of Area-2 rapidly increased to approximately 51.4 Hz as a result of a power surplus of 10 GW. Area-3 experienced a situation of under frequency, which resulted in significant load shedding [22]. The national blackout (NB) that occurred in Turkey in 2015 can be given as a second important example. It started with frequency variations and resulted with the loss of synchronization. Therefore, frequency variations are vital for the safety of power systems and have been scrutinized for many years.

In this study, the national blackout of Turkey that occurred on 31 March 2015 was examined by considering the technical report of ENTSO-E published in 2015. It has been revealed that some operational errors and technical deficiencies were the main causes of the NB. Due to these reasons, the frequency variations occurred and then the outage was created. However, the frequency variations could not be clarified in that report. Additionally, there is no literature that examines the tragic situation of NB in a chronological manner except for this paper. Therefore, the power quality data of a local industrial zone (IZ) in Turkey was collected to further investigate this issue. According to the analyzing of these data, a direct proportional relationship between the line voltage and system frequency was observed. This relationship was demonstrated with a case study and was mathematically proved for the first time in the literature with this study. After that, the causes, chronological development process, and consequences of NB are discussed in detail and the restoration process of the national electrical network (NEN) was also analyzed. In the last part, some recommendations are also given to avoid such events in the future.

This paper is organized as follows: Section 2 presents the blackout occurrence globally, Section 3 presents the main reasons of NB and a case study, Section 4 presents the chronological development of NB, Section 5 presents the restoration process of NEN, Section 6 presents recommendations, and Section 7 presents the conclusion of this study.

2. List of Blackouts in the World

According to official records, the first mass blackout in the USA occurred on 9 November 1965. The blackout started with the tripping of breaker relays at five transmission lines in the Ontario region due to the overcurrent. This event was turned into a chain reaction within 12 min. In total, 20,000 MW of demand was disconnected from the local distribution networks for 13 h, affecting 30 million customers. It was estimated that there was an economic loss of USD 100 million [2].

A blackout in Thailand occurred on 18 March 1978. It started with a generator failure and the NEN collapsed due to a chain of failures. It took more than 9 h for the recovery of the system. In total, 1336 MW of demand was separated from the local distribution networks and 40 million customers were affected [39]. A blackout in Uttar Pradesh of India occurred on 2 January 2001. It started with a transformer failure and spread throughout the country. The recovery of the system took about 9 h. In total, 230 million people were affected. The economic loss was estimated to be USD 110 million [40].

A blackout in Italy occurred on 28 September 2003. It started with a sagging in overhead conductors due to the excessive current and then short circuits happened between each other. Hence, the electricity service was damaged and then, the electricity connection of Italy was broken with Switzerland, France, Austria, and Slovenia, respectively. The recovery of the system took about 13.5 h. In total, 24,000 MW of demand was separated from the local distribution networks and 57 million customers were affected [41,42]. A blackout in the USA and Canada occurred on 14–15 August 2003. It started with voltage fluctuation due to the excessive usage of air conditioners. Electricity service was not provided for 72 h in the USA and for 192 h in Canada. In total, 61,800 MW of demand was separated from the local distribution networks and 50 million customers were affected [43,44,45]. A blackout in Java, Bali and Indonesia occurred on 18 August 2005. It started with the failure of eight power generation systems. Electricity service was not provided for 72 h. In total, 6500 MW of demand was separated from the local distribution networks and 100 million customers were affected [40].

A large-scale blackout in Germany, France, Italy, and Spain occurred on 4 November 2006. It started with the overloading and coordination errors of power generation systems. The recovery of the system took about 2 h. In total, 14,500 MW of demand was separated from the local distribution networks and 45 million customers were affected. The economic loss was estimated to be USD 139 million. A blackout in Brazil and Paraguay occurred on 10–11 November 2009. It started with short circuit faults due to heavy rain and wind. In total, 12,000 MW of demand was separated from the local distribution networks and 87 million customers were affected [46].

A blackout in India occurred on 30–31 July 2012. It started with the tripping of a power line due to the overloading at 132 kV and 400 kV lines. This event was turned into chain reactions and the generators were out of service due to the increase in power line voltage frequency. The recovery of the system took about 8 h. In total, 48,000 MW of demand was separated from the local distribution networks and 670 million customers were affected. It was recorded as the biggest blackout in history [47].

A blackout in Bangladesh occurred on 1 November 2014. It started with the separation of some power generation systems from the NEN and less energy import than the required amount from other countries. This situation resulted with the collapsing of the whole power system due to the decrease in power line voltage frequency. In total, 445 MW of demand was separated from the local distribution networks and 150 million customers were affected [44].

A blackout in Ukraine occurred on 23 December 2015. It was caused by a cyber-attack against the three regional energy distribution companies. The hackers breached the company’s information systems by using security vulnerabilities and then destroyed the passwords and rendering important data by using wipers and malwares. The recovery of the system took about 6 h, and 225,000 customers were affected. It was recorded as the first blackout in the world due to a cyber-attack [48].

A blackout in South Australia occurred on 28 September 2016. It started with the inability of the solar power plants to meet the demanded power due to severe weather conditions. This event was turned into a chain reaction and the whole power system collapsed due to the decrease in power line voltage frequency. In total, 1826 MW of demand was separated from the local distribution networks and 1.7 million customers were affected. It was recorded as the first blackout in the world due to renewable energy integration [49].

A blackout in Puerto Rico occurred on 12 April 2018. It was caused by the falling of a tree onto an overhead power line during the clearing of vegetation. In total, 870,000 customers were affected [50].

3. Causes of National Blackout in Turkey

In Turkey, the highest electricity consumption occurs in the summer period and the lowest electricity consumption occurs in the spring period. However, some problems have occurred at the transmission lines during the consumption of electricity due to the insufficient infrastructure. The unbalanced voltage and frequency variations are at the top of these problems. On 31 March 2015, a national blackout occurred in Turkey. It coincided with the summer period and the NEN collapsed in 3 s. Before analyzing this event, some information needs to be known, which is presented in the following section.

(a) The hydroelectric power plants located in the east region of Turkey are higher in numbers and have more power than in the west region. The installed hydroelectric power potential is approximately 32 GW in the Central and Eastern Black Sea Regions of Turkey [51] (see Figure 1). Due to spring rains, floods occurred at the dams located in the Eastern Black Sea and South and Eastern Anatolia. In order to prevent these overflows, the dam and river hydropower plants were served at full capacity. Since the energy demand was met from the east side, the hydroelectric power plants located in the west side of Turkey were taken to the offline mode. Hence, 400 kV lines that provided the power flow from the east to west were overloaded.

(b) There are eleven important TLs used for supplying electricity service from the eastern to western side of Turkey. Their total route lengths are higher than 265 km and operate at 400 kV. Before the NB, the four most important of these TLs were held offline for both minor and major maintenances. The first transmission line is the Kayabaşı-Baglum, the second and third transmission lines are the north and south sides of Gölbaşı-Kayseri, respectively, and the fourth transmission line is the Oymapınar-Ermenek. These TLs are shown in Figure 2.

At the NEN, there are 16 serial capacitor (SC) banks to compensate for the voltage drops. Before NB, they were taken out of service. Hence, electricity service from the east to west parts of Turkey was weakened by holding the important four TLs in the offline mode and taking SCs out of service [52].

(c) The length of Osmanca-Kurşunlu TL (labeled as OK in Figure 2) is 206 km. An aluminum conductor steel-reinforced (ACSR)-type conductor is used and each phase consists of three-conductor bundles throughout the entire line. According to the datasheet of the conductor, it can carry up to 2350 A under 400 kV. On the incident day, the Çankırı TL was disabled at 09:36:09.418 and the tripping values of the breaker were recorded as 1820 A under 393 kV at the incident time.

(d) In emergency situations, ACSR-type conductors can carry a higher current than its capacity for 20 to 30 min until reaching its temperature limit [52]. Hence, they provide a period of time for authorities to recover the system in case of overload situations.

(e) According to the Electricity Grid Regulation (EGR), 35.5% of the existing load should be released if the frequency of the electricity network drops to 48.4 Hz. During NB, 35.5% of the electricity network load corresponded to 4.8 GW [52]. However, the rate of load shedding could not reach that level due to the imbalance and tripping events that occurred in the NEN. The main reasons of this situation are given as follows.

• The relays and circuit breakers used for the protection of transmission lines could not perform the fast electromechanical opening/closing maneuvers.
• The stability of some power plants was damaged due to lower load shedding than the required amount.

(f) In European countries and Turkey, the frequency set point of electricity networks is 50 Hz. This frequency is controlled within a tight limit, typically within ±150 mHz. It should not go below 47.5 Hz (under frequency) and above 51.5 Hz (over frequency) [53]. According to EGRs, the power plants shall stay in service for at least 10 min if the frequency of the electricity network drops to 47.5 Hz. Unfortunately, many large thermoelectric power plants were out of service at frequencies above 47.5 Hz. The other power plants left the system and switched to the island mode. Hence, the national electricity grid collapsed in about 3 s.

As mentioned above, the biggest factor of NB is the deficiency to supply the demanded power. As a result, the voltage value decreased, and the frequency was lost. A regional collapse occurred on 19 September 2013 in an industrial zone (IZ). It is located in the southern part of Turkey with a power of 190 MW. The power data of this region related to the period of 2013–2014 were transferred to the simulation program and it was seen after the analyzing of these data that the frequency variation is directly proportional to the operating voltage value [54]. The details of this event are presented below as a case study.

Case Study: Blackout of an industrial zone in Turkey

The power circuit of IZ was modelled by using PSCAD. PSCAD is a time domain-based simulation software for analyzing electromagnetic transients in electrical networks. It provides a graphical Unix-based user interface to the electromagnetic transient program. This software allows the creation of new control or electrical components by using Fortran code. In the simulation model, all data were written as a text format and the value of each variable was drawn by creating a control component. These data included the phase voltage, currents, power factors, active and reactive powers, and individual and total harmonic values for each phase of the power system. Each load was representative as a combination of the current and voltage source. The electrical parameters of real power system components were transformed into a simulation counterpart. The simulation software allows the adjustment of the solution time step (dt). For a detailed analysis, dt was set to 50 microseconds, while for a long-period analysis, it was set to 2 milliseconds in this simulation case study. The duration of the simulation case study was set to 500 s [54].

The electricity grid of IZ is connected to two power plants at the rate of 154 kV by using two identical step-down transformers (154/31.5 kV, 100 MVA, ∆/∆). The main electrical network of IZ is divided into two main branches. These branches are named as ADM-1 and ADM-2 and given in Figure 3.

Each of these branches is divided into parallel subgroups (DM series). ADM-1 consists of DM5 to DM10 subgroups and ADM-2 consists of DM1 to DM3 and DM12 to DM14 subgroups. Additionally, these subgroups are divided into secondary subgroups (such as DM5A to DM5E). This is shown in Figure 4.

When the power quality data of IZ were examined, it was observed that there was a voltage drop at the C-feeder of DM5 (subgroup of ADM1) and the frequency was decreased to 32.1 Hz on 19 September 2013 at 5:00 a.m. The data of this moment corresponds to the 440th row of the voltage and current data blocks given in

Table 1. C-feeder data of DM5 [54].

No.	ADM1/DM5/C-Cur.			Freq. (Hz)	ADM1/DM5/C-Voltage			ADM1/DM5/C-PowFac.
	Ia (A)	Ib (A)	Ic (A)		Va (kV)	Vb (kV)	Vc (kV)	Pfa	Pfb	Pfc
435	7.51	7.28	7.27	50.00	17,956.70	18,052.60	18,005.80	0.98	0.98	0.98
436	6.74	6.54	6.56	50.00	17,869.30	17,982.40	17,930.40	0.98	0.98	0.98
437	8.56	8.32	8.35	50.00	17,908.80	18,017.90	17,969.80	0.98	0.98	0.98
438	8.19	7.91	7.93	50.00	18,050.60	18,140.20	18,093.60	0.98	0.98	0.98
439	8.42	8.19	8.25	50.00	18,065.00	18,155.70	18,114.70	0.98	0.98	0.98
440	5.34	5.36	5.33	32.10	11,616.30	11,676.90	11,641.60	0.64	0.63	0.63
441	2.65	2.70	2.65	22.50	8087.80	8137.20	8099.40	0.41	0.41	0.41
442	6.19	6.30	6.24	49.90	18,034.80	18,139.60	18,073.30	0.98	0.97	0.97
443	5.22	5.33	5.28	49.90	17,969.80	18,073.80	18,045.10	0.98	0.97	0.97
444	3.01	3.19	3.15	50.00	17,974.70	18,071.40	18,045.40	0.97	0.96	0.96
445	7.20	7.32	7.30	50.00	18,006.30	18,098.60	18,087.80	0.98	0.98	0.98

. Additionally, a similar problem occurred with the 441st row of the table.

The problem given above started with the connection of a large power load and the drawing of a high current from the power network of IZ. It tried to supply this high demand; however, the network could not supply this load and then a blackout started with the decrease in network frequency. This situation is explained mathematically in the following equations:

$$V\left(t\right)=V_{m }·sin\left(\omega t\right)$$

(1)

where $V\left(t\right)$ is the line voltage, $V_{m }$ is the maximum value of $V\left(t\right)$, $\omega$ is the angular velocity, and $t$ is the time constant.

$$\omega =2\pi f$$

(2)

where $f$ is the frequency of the line voltage.

By rearranging Equation (1) by using (2), the following is obtained:

$$f=\frac{1}{2\pi t} ·\arcsin (\frac{V\left(t\right)}{V_m})$$

(3)

By evaluating Equation (3), it can be seen that $f$ and $V_m$ are constant, $t$ is independent, and $V\left(t\right)$ is dependent on the variable $t$. In the assumptions to be made according to the variables $t$ and $V\left(t\right)$, $t$ will be taken into account twice. This situation can be eliminated by assuming that one of the variables is passive (with a similar approach used in the superposition theory of circuit analysis). For this purpose, $t$ will be assumed as a constant in the following assumptions and its value will be taken as $\frac{1}{2\pi }$ in Equation (3).

In this case, $f$ can be redefined as:

$$f=\arcsin (\frac{V\left(t\right)}{V_m})$$

(4)

The term $\frac{V\left(t\right)}{V_m}$ in Equation (4) is a proportional expression. Its value changes between 0 and 1 while the value of the arcsine function changes between 0 and 90 degrees. This shows that there is a direct proportion between $f$ and $V\left(t\right)$. The relation of frequencies with the voltage values can be easily extracted in rows 440 and 441 from Table 1. For example, the voltage ratio of row 440 and 439 for V_a (11,616.3 V/18,065.0 V) is 0.643. If this value is multiplied with the main frequency (50 Hz × 0.643), 32.15 Hz is obtained. This value is similar to the frequency of row 440. As another example, the voltage ratio of row 441 and 439 for V_a (8087.6 V/18,065 V) is 0.447. If this value is multiplied with the main frequency (50 Hz × 0.447), 22.35 Hz is obtained. This value is also similar to the frequency of row 441 [54]. As has been shown, the frequency drop is directly proportional to the voltage drop. This causes the deteriorating of network balance and system stability and even partial or complete collapse of the network (such as a blackout).

(g) A competitive environment was established in the electricity market by withdrawal of public companies from the electricity generation and distribution section. Their roles were taken over by private entrepreneurs over time. However, EGRs were disregarded by private energy suppliers, and they were acting according to their commercial interests in terms of the loading and load shedding. During NB, many of them were out of service at frequencies above 47.5 Hz and the others left the system and switched to the island mode. Hence, the NEN was not supported to recover itself.

4. Occurrence of NB

Before NB, the consumption of the west side was 22.87 GW. In total, 2.19% and 20.5% of this power was provided by Bulgaria and the eastern side of Turkey, respectively. The important four TLs given in Figure 2 were held offline for the minor and major maintenances. Therefore, electricity service was weakened from the eastern to western sides of Turkey. Before NB, the frequency variations started due to the inability to meet the power demand of the west side. The first frequency variation occurred in the Osmanca–Kurşunlu TL and it was recorded as 09:36:09:42 on 31 March 2015. The second frequency variation occurred in the Atatürk–Yeşilhisar North TL and it was recorded as 09:36:10:88. After the consecutive frequency variations given in Figure 5, the distance protection relays tripped due to ‘factual sequence events’ [52]. The parallel TLs given in Figure 2 were consecutively separated one after another from the national grid. Hence, the synchronization was lost within seconds and NB occurred at 09:36:12:53.

It was stated in EGR that the power plants should stay in service for at least 10 min when the frequency of national electricity drops from 50 Hz to the limit of 47.5 Hz. Despite this regulation, many power plants disabled themselves or switched to the island mode. As a result, electricity service was interrupted from the east to west parts of Turkey and the NEN of Turkey had lost its synchronization with the ENTSO-E system [53,54,55,56,57]. After a second, the interconnection of Turkey with Bulgaria and Greece was lost. Figure 5 shows the starting times of frequency variations and NB.

5. Restoration Process of the Power System

The restoration procedure was planned by TEIAS according to the nine regional control centers (RCCs). It was started simultaneously in each control center and the whole ring was completed in coordination with the National Control Centre (NCC). The recovery was started by obtaining electrical energy from Bulgaria and then synchronizing it with the region of Northwest Anatolia RCC. Each RCC has three or four restoration paths, and each path includes generators. The European part of Turkey started the restoration process by closing Hamitabat–Maritsa East-3 (BG) Line-2 and supplying the electrical energy from the Hamitabat and Ambarli Natural Gas Power Plants.

After 30 min from the start of the NB, the Central European system was used to power up the Thrace region of Turkey at 09:54:00 Central European Time (CET). After starting up several power plants in the European side of Turkey, some parts of the Asian side were energized at 11:11:00 (CET). At that moment, electricity service was already provided to half of the Thrace region.

The Black Sea side and east side of Turkey were synchronized at 11:30:00 (CET). Then, the east and west sides were synchronized with ENTSO-E at 16:12 (CET). At that moment, 80% of Turkey was already energized. Some frequency variations occurred during the restoration process. The re-synchronization with the Continental European (CE) area was completed gradually at 18:36 (CET) [52]. The status of frequency variations according to time is shown in Figure 6.

6. Recommendations

The above-mentioned issues led to the NB in Turkey on 31 March 2015, and caused material and other damages that were difficult to compensate. If the recommendations given below had been considered before, the NB could have been prevented.

Recommendation 1: During the frequency variations, many thermoelectric power plants were out of service without falling to frequencies below 47.5 Hz. This situation contradicts with the obligations of EGR. If authorities had paid attention to this issue, electricity service would not have been interrupted.

Recommendation 2: It was previously mentioned that regardless of whether the four TLs had been enabled or disabled, the production was reduced in the west side while increased in the east side of Turkey. Hence, electricity service from the east to west sides of Turkey was weakened before the interruption. In addition to these disadvantages, SC banks were also out of service.

With these deficiencies, the first tripping protection occurred at the Kursunlu–Osmanca TL. It is shown in Figure 5. If all SCs had been in service before NB, the current of this TL would have been on average 1570 A. Hence, the currents of all 400 kV TLs would have been lower than their nominal limits.

Recommendation 3: The relays and circuit breakers that are responsible for opening and closing the system should be monitored with smart systems. If it is necessary, an operator can intervene remotely. The information about how long it takes for the transmission lines to reach their temperature limit, in case of an overload, should be known. Thus, early system trips will be prevented.

Recommendation 4: The turbine speed of gas power plants decreases in the case of a drop in the electricity network frequency under its nominal value. The necessary precautions should be taken to prevent the turbine speed falling from its linear characteristic.

Recommendation 5: The causes of synchronous generator failures at frequencies that are higher than 47.5 Hz should be identified and eliminated as much as possible.

Recommendation 6: With the rapid spread of data centers, uninterruptible power supplies are frequently used at high-power ratings (up to 6 MVA) in recent years. They have an important role in the case of possible interruptions in the electric network and provide electricity to data centers. However, they are not used when the voltage level and frequency of the utility grid are stable.

These reserves can be used for planned electrical infrastructure works to supply electricity service and they can be used to regulate the balance between the support and supply demand.

Recommendation 7: Wind parks and photovoltaic systems have a fault-ride-through property. It signifies the ability of electrical devices to stay connected to the network and function during instances of a low voltage at the connection point, typically caused by faults. It can be used to ensure grid security and transient stability.

Recommendation 8: A good technical and organizational arrangement is required for the electricity network to prevent an outage during revisions and improvements.

The technical arrangements are given as follows:

• Adequate infrastructure and maintenance: A reliable electrical system requires sufficient generation capacity, transmission lines, distribution networks, and substations. The power system can meet the demand and maintain the stability during normal and abnormal operating conditions [58]. Regular maintenance of equipment is very important in ensuring the reliable operation of the electrical system and it prevents equipment failures and reduces maintenance activities [59].
• System protection: System protection means the use of protection devices such as protective relays and circuit breakers to isolate the system from the faults and prevent cascading failures in the power system. These devices can detect faults very fast and isolate the faulty equipment [59].
• Power quality improvement: The use of power electronics-based devices, such as APF, STATCOM, and DVR, can help improve the power quality and stability of the electrical system. These devices can reduce voltage fluctuations, voltage regulation, and harmonic and reactive power compensations [60].
• The organizational arrangements are given as follows:
• Adequate staffing and effective communication: An electrical power system requires skilled technical staff to operate and maintain the equipment and infrastructure. Hence, the power system can be operated and maintained properly. Effective communication between the system operators and maintenance staff increases the response time to emergency situations, allowing for prompt resolution [60].
• Periodic training and development: The knowledge of technical staff and operators can be improved by continuous training and development. Hence, the power system operates safely, and they interfere in the system very efficiently during emergency situations [20].

Recommendation 9: To eliminate errors in electrical network management in the future, several steps can be taken. One of the main steps is to implement intelligent networks for self-controlling and automatically balancing energy systems and allowing for efficient energy transfer and distribution. The other main step is that the technical apparatus of intelligent networks includes digital control systems that solve various problems of artificial intelligence, which can help in optimizing the operational management of maintenance works [61]. The other steps are given as follows:

• Online monitoring of electrical networks to detect any potential issues in real-time [62].
• Integrating AC and DC relays to detect and isolate faults for preventing any further damage to the network [62].
• Providing network operators with instantaneous online data, network information, and modelling through distribution management systems and supervisory control and data acquisition systems to help improve overall operational efficiency of the network and rapidly locate and resolve faults in the network [62].
• Using renewable systems for electricity generation in remote areas with the inclusion of electricity storage in hybrid systems to improve reliability of the electricity supply [63].
• Establishing a maintenance program that achieves an optimal balance between the cost of maintenance and enhancing reliability, ensuring the distribution network operates safely and efficiently while preserving its functionality [64].
• Reducing the values of SAIFI and SAIDI, which are important indicators used to measure the reliability of an electrical network. This can be achieved by taking appropriate measures, such as optimal placement of automatic circuit re-closers and distributed generators, as demonstrated in above details of this study. By reducing the frequency and duration of power outages, the reliability of the power system can be increased, which ultimately benefits all stakeholders involved.
• Enhancing the skills of network managers through training programs in problem-solving, analytical thinking, cloud management, project management, network monitoring, and access management. This will ensure that they have the necessary skills to handle the complex and intertwined relationships between various components of the network and can plan ahead, spot potential issues, and take preventive measures.
• Renewable systems are also useful in generating electricity in remote areas, and hybrid systems that include electricity storage can provide high reliability of the electricity supply.
• By employing distribution management systems and SCADA systems, network operators can access real-time online data and modeling, leading to enhanced operational efficiency of the network and swift identification and resolution of faults.

By implementing these steps and developing these skills, network managers can improve the efficiency, reliability, and quality of electrical network management.

Recommendation 10: Generally, power systems are designed to operate with an additional capacity or margin that provides a safety corridor against unexpected variations in the power demand. In the event of a sharp increase in the power consumption, PM is an important factor in maintaining a stable and reliable electrical grid. If the power margin increases, it means that the power system has more capacity to meet power demand, which reduces the risk of power outages. It is particularly critical during the most heavily utilized hours of the grid (usually in the afternoon). If the power margin is too low, demand increases, or system failures can cause power outages [65]. To mitigate these risks and protect the power system, various measures can be taken. These are given as follows:

• Advanced forecasting and modeling techniques can predict the future power demand and it can enable the network operators to adjust the power margin accordingly [66].
• The capacity and reliability of the system is increased by upgrading and expanding the power grid infrastructure [67].
• The energy storage systems can be used to store the excess power during low demand periods and release it during high-power demand periods to increase the system reliability. Hence, they help balance the grid and reduce the requirement of an additional power generation capacity [68].
• Electric load shedding is conducted to maintain the generation power margin at the nominated level when the power demand is higher than electric supply. This prevents widespread system collapse when a fault occurs [69].
• The use of distributed energy resources such as solar panels and small wind turbines can be used to reduce the power demand during peak hours [70].
• Power electronics-based devices such as an active power filter, static synchronous compensator, and dynamic voltage restorer can also be used to help the stabilizing of the power system, to improve the power quality, to reduce energy waste, and to minimize the risk of power outages caused by voltage sags, harmonics, and other power quality issues [71].
• The integration of grid modernization technologies, such as smart grids and advanced metering infrastructure, increases the efficiency and reliability of the power system by managing the power flows [72].
• The power grid infrastructure should have regular maintenance and inspections to ensure that it is in good condition, and it is able to operate at full capacity if it is needed [73].

7. Conclusions

If the stability of a power system is changed from normal conditions to abnormal conditions (such as an overload, a generator outage, TL tripping, etc.), the system dynamics (such as frequency and voltage instability) must be quickly controlled in time to prevent the cascading events that might lead to a blackout. Additionally, the recovery scenarios are required to protect the NEN from these situations.

This national blackout of Turkey occurred on 31 March 2015. It started with frequency variations and resulted with the collapse of the NEN. Similar load conditions were experienced on 30 March 2015. However, the system survived due to that the generation pattern was different. One important point is that the authorities did not realize this situation.

In this study, NB of Turkey was analyzed in detail. To obtain a better understanding of the generation, transmission, and distribution regulations in Turkey, the electrical infrastructure and political visions were summarized. Then, the most important blackouts were summarized. After that, a case study related to the regional collapse of IZ was given. In this case, a relationship between the operating voltage and frequency variation was proven mathematically by using the power quality data of IZ. Hence, it is the first time that the relationship between the operating voltage and grid frequency has been shown in the literature. Then, the occurrence and restoration processes of NB was analyzed. At the end of this study, some recommendations were given for a sustainable and resilient power delivery system.