Coordination of Frequency Reserves in an Isolated Industrial Grid Equipped With Energy Storage and Dominated by Constant Power Loads

This article examines the use of interconnected synchronous system requirements for frequency containment reserves (FCR) on isolated industrial grids that use turbogenerators as main source of energy, have high penetration of wind energy, are equipped with energy storage, and have a high level of constant power loads coupled by power electronic converters. Leveraging on the recent Nordic requirements for reserves in islanded operation (FCR$_\text {I}$), we propose an expansion that allows prioritizing among various reserve providers accounting for different isolated grid conditions. The study case of a complex, isolated industrial grid is selected to test this approach. The stability of this grid is evaluated via eigenvalues and participation factors considering the detrimental effects of constant power loads. It is demonstrated that, by prioritizing the reserve allocation to the faster converter-interfaced storage devices and loads, the overall stability is increased in addition to allowing the turbogenerators to operate at a more constant load. The results are supported by computer simulations of the complex isolated grid in DIgSILENT PowerFactory and by laboratory power-hardware-in-the-loop tests which compare the performance of the proposed concept with the industry consolidated droop control. The computer simulation models developed for this article are made publicly available for reproducibility purposes.


I. INTRODUCTION
C ONSUMPTION and generation imbalances in ac power grids create frequency variations, those being counteracted by the activation of distributed power reserves in a hierarchic manner.Primary reserves perform a droop-based frequency control, responding automatically after power imbalances and limiting frequency deviations in a time scale of seconds.Secondary reserves are subordinated to the grid's automatic generation control (AGC), which brings the frequency back to its rated value within seconds to minutes after power imbalances [1].In the European context, primary reserves are named frequency containment reserves (FCR) and are coordinated nationally by transmission system operators [2].Three types of FCR have recently been defined for the Nordic synchronous area, which includes the power grids of Finland, Sweden, Norway, and eastern Denmark.These types are [3] normal operation frequency containment reserves (FCR N ), large disturbance frequency containment reserves (FCR D ), and islanded operation frequency containment reserves (FCR I ).FCR N and FCR D cooperate in the interconnected system, whereas FCR I are a simplified version of the pair FCR N /FCR D that are activated only during islanding events.The same provider can supply any of these reserves, depending on the fulfillment of technical requirements, which are evaluated in a qualification process, and the grid operating conditions.
The increased penetration of intermittent renewable energy sources (RESs) has been supported, in some countries, by power exchanges with their neighbors.Denmark's RES generation, for instance, exceeds the country's load approximately 20 % of the time, while Portugal's intermittent RESs rely on hydro power and exchanges with Spain [4].External support for ac frequency control from neighboring countries might, however, be technically limited or controlled by regulations [2].In autonomous systems and in some geographically isolated areas, on one hand, external support for ac frequency control is nonexistent.To mitigate the effects of a higher penetration of non-synchronous generation in Ireland, an increase from 0.2 GW to 1.65 GW of battery storage is expected to happen from 2020 to 2030 [5].In large interconnected systems, on the other hand, islanding scenarios are becoming more complex and proper cooperation between synchronous and non-synchronous generation is necessary.Therefore, robust coordination strategies that can easily be adapted and scaled to different grid configurations are key for maintaining frequency stability and exploiting the best characteristics of the available power reserves.Within this context, smaller scale isolated systems provide valuable insights that are, many times, transferable to country and continental-wide power grids.Relevant examples of such smaller scale systems are present in the offshore oil and gas (O&G) sector.
An O&G platform isolated from the continent and expected to be connected to an offshore wind farm (WF) cannot rely on external partners to balance excess or underproduction from wind power.Energy storage systems (ESSs), both centralized [6,7] and distributed [8], have been investigated in the literature as means of mitigating wind variability in offshore applications.Converter-interfaced loads (CILs) that tolerate fast power changes are also promising candidates for ac frequency control support in O&G platforms.The operation of the platform's water injection system [9] as a flexible load in conjunction with wind power has, for instance, been assessed by [10][11][12].The Nordic FCR I concept is, in this paper, expanded and applied to the operating conditions of an isolated O&G platform with traditional turbogenerators (GTs), high penetration of intermittent RES, flexible CILs, and an ESS.Despite this specific application, the concepts applied and the insights given in this paper for an offshore O&G platform are applicable to other isolated systems with multiple reserve providers, such as islands, ships, remote communities, industrial or military installations.
Power generation and loads must be properly modeled where isolated grids are connected to intermittent RES, such as wind and solar farms.Wind turbines (WTs) and solar photovoltaic (PV) panels can be considered constant power sources (CPSs), as their controllers typically operate in maximum power point tracking, in other words, tracking the optimum power for a given wind speed or solar irradiation.Storage devices connected to a common ESS dc link can also run as constant power loads (CPLs) or CPSs, and isolated grids may serve a considerable amount of CPLs, such as variable frequency drives.Constant power loads and sources are known to cause instabilities in dc microgrids [13,14] and in ac microgrids [15,16].In stability assessments of grid connected WFs [17,18], local power consumers are usually not relevant and the grid tends to be modeled as a voltage source behind an impedance and the local loads disregarded.Power electronic oriented studies on CPLs in microgrids [19,20] tend to focus on the stability of the converters, not in frequency control nor in the stability of the complete grid.It is, furthermore, common in studies of the integration of wind power and ESSs in industrial applications to simplify the total electrical load to a constant impedance load (CZL) [6,7,21,22].In the study case presented in this paper, the presence of CPLs or CPSs in the stability of an industrial isolated grid is evaluated, being the total load modeled as a mix of CPLs and CZLs, where the former dominates.
If combined with power electronic converters (PECs), CPLs and CPSs also give rise to new instability phenomena both in micro and in large interconnected grids [23,24].Remark that modern type 4 [25] WTs, solar PV farms, ESSs, and some types of loads are all connected to the grid via full-power PECs.Converter driven instabilities, therefore, should not be overlooked when integrating these equipment in isolated grids.For that reason, the stability of the study case presented in this paper is analyzed with the help of computer simulations performed with DIgSILENT PowerFactory 2020 SP2A.The impact of different contributions of the ESS and the GTs in the eigenvalues of the system is assessed.Participation factors are used to associate states with eigenvalues and help identify devices and parameters that strongly influence the system's oscillation modes.This paper, moreover, presents results obtained with real-time system (RTS) and power-hardwarein-the-loop (PHIL) tests performed in the National Smart Grid Laboratory at the Norwegian University of Science and Technology (NTNU).Such tests have proven to be efficient tools for analysis and validation of devices and their controls in isolated grids [23].
In summary, this paper expands the Nordic concept of FCR I .A subdivision of the FCR I into normal and large disturbance reserves is proposed for complex isolated grid scenarios.This expanded concept is applied, in a study case, for coordinating the power reserves of an isolated industrial grid fed by GTs, dominated by CPLs, connected to a WF, equipped with flexible CILs, and supported by an ESS.The paper demonstrates the advantages, from a frequency control perspective, of replacing slower GTs by faster converterinterfaced ESS as the main providers of primary power reserves.It also shows that this replacement causes a non-critical reduction in the damping of oscillation modes associated with frequency measurement transducers and controllers of constant power devices.These results are supported by simplified and detailed stability analyses, by computer simulations, and by PHIL tests in the laboratory, being the computer simulation models developed made publicly available for reproducibility purposes.
The paper is organized as follows: the sharing and coordination of FCR are presented in Section II, the study case is introduced in Section III, the effects of the FCR sharing on the stability are assessed in Section IV, the practical results obtained in the laboratory are shown in Section V, a discussion is made in Section VI, and, finally, the concluding remarks are listed on Section VII.

II. SHARING AND COORDINATION OF RESERVES
The stochastic nature of wind can lead to an increased number of start-stop operations and more variable load profiles for the GTs of an O&G platform connected to a WF resulting in higher wear and tear, higher nitrogen oxides (NO x ) emissions, and an overall degradation of the electric power quality and grid frequency stability [26,27].Therefore, coordination strategies that allow prioritizing CILs and ESS instead of GTs as the main source of power for fast frequency control are key to a successful integration of wind power into offshore O&G facilities.In this section, a hierarchic frequency control structure that allows such prioritization is explained.
Fig. 1 depicts the frequency control structure of the study case's autonomous grid where reserve providers play a subordinate role under a centralized power management system (PMS).The secondary frequency controller, which is a part of the PMS, is responsible for correcting steady-state errors in the ac frequency of the isolated grid.It employs a proportional and integral regulator that reacts to the frequency deviation and generates a secondary power reference, which is shared among the reserve providers.The input for the secondary power reference can be enabled by each provider that, in turn, informs the status of the input and the measured provided power to the PMS.The distributed primary frequency control is performed locally at each frequency reserve provider.
For coordinating the activation of primary frequency reserves, the Nordic concept of FCR I is expanded into normal and large disturbance reserves in this paper.For simplicity of notation, the normal operation reserves are named FCR N and the large disturbance reserves are named FCR D .These new FCR N and FCR D for isolated grids are similar to the ones for interconnected operation defined in [3].As it will be shown in this paper, the proposed FCR coordination is  reasonably flexible and can be adapted to any isolated grid with dispatchable power reserves that communicate with a centralized AGC or PMS.
The limit f N between the regions of FCR N and FCR D activation is defined in terms of the deviation of the measured ac frequency f from its nominal value f n , namely f = f − f n .FCR N are active and FCR D are inactive when the All FCR N providers shall be able to supply their allocated reserve power P N for a frequency drop of f N .Conversely, the providers should absorb P N for a frequency increase of f N .Therefore, the frequency-to-power gain of a FCR N provider is When considered for the whole system, this gain is called regulating strength in [3] and frequency bias in [1].In this paper, it will be referred to as FCR gain or as frequency-topower gain.
The value f N is set to 0.1 Hz (0.2 %) for the Nordic interconnected grid [3].However, isolated grids usually have to endure more severe relative power imbalances and resulting frequency disturbances than a country or continental wide grid.For instance, the recommended practices for the design of electrical power generation in merchant, commercial, and naval vessels [28] define a tolerance of ±3 % for the "maximum permitted departure from nominal frequency during normal operation, excluding transient and cyclic frequency variations."Hence, a value between 0.2 % and 3 % will later be adopted for f N in Section V.
The system is considered under a large disturbance when In this condition, the FCR N are saturated at P N and the FCR D are activated.The FCR D frequency-to-power gain is defined as The total frequency-to-power characteristic of the system is, therefore, a composition of the normal and large disturbance reserves as illustrated in Fig. 2. The FCR D controllers feature a dead band between ± f N which ensures that only the FCR N are activated in normal operation, as shown in Fig. 1.Notice that gains K N and K D do not necessarily have to be equal.

A. FCR Control Loop
The power contribution of the FCR providers is a function of f .Each provider measures the ac frequency f and subtracts it from the rated frequency f n , see Fig. 1.The resulting − f is fed to two proportional controllers, one for FCR N and one for FCR D .Each controller has its own gain (K), limiters (max, min), and a symmetric dead band (f dbN and f N ).The FCR N power reference ( * P N ) and the FCR D power reference ( * P D ) are summed to the secondary power reference ( * P S ), which is given to the providers by the PMS via communication link.The total power reference output ( * P ) has its own independent maximum and minimum limits.
The PMS allocates a primary reserve quota to each provider based on a security assessment, which takes into consideration the grid operational conditions and the forecasts of loads and RES.To receive a quota, the provider must be able to respond symmetrically to both positive and negative variations in f and deliver the total assigned reserve power P N (up or down) when Based on the assigned reserves, each provider calculates its gains K N and K D according to (1) and (2), respectively.Remark that the transient response of each FCR provider does not rely on a fast communication link with the PMS, as the FCR control loops are implemented locally and fast dynamic changes to the power allocation are not expected.

B. Role of Dead Bands in FCR
The normal operation and large disturbance reserves are properly coordinated by a dead-band block in the FCR D providers, shown in Fig. 1.The limit frequency f N is unique for the system and is known by all FCR providers.By adjusting the FCR D dead band to f N , one guarantees that the large disturbance reserves will only be activated once the FCR N are saturated.The limits * P N max and * P N min of each FCR N provider are assumed to be symmetric and their absolute values are equal to P N .Although not a desirable feature, a non negligible dead band f dbN might be necessary for a proper operation of a specific FCR N provider.In this case, calculating the gain K N with (1) leads to a reduced FCR N capacity.If the provider requires a small dead band for proper operation, the effects of the dead band can be compensated locally by calculating the frequency-to-power gain according to (3) without the intervention of the PMS.

C. Linearized Rotating Mass Model
Converter interfaced reserves have much faster responses than traditional turbogenerators [29].The benefits of these faster responses can be evaluated, initially, with the classical rotating mass model [30,31].For that, the spinning reserves of the system (the GTs) are aggregated into a single rotating mass with moment of inertia J.The turbines apply torque to increase the angular frequency ω of the rotating mass, whereas the aggregated electrical loads apply torque to reduce ω.When expressed in terms of power, the balance of torque of this simplified model is where ω = 2πf , P FCR is the power supplied by the FCR, P S is the secondary frequency control power, and P L is the total electric load power.
Let the following variables be introduced: where H is the inertia constant of the spinning reserves, ω n is the nominal angular frequency, S n is the base power of the isolated grid, p FCR is the normalized power supplied by the FCR, p is the normalized imbalance of power between load and secondary reserves, and f is the normalized deviation of the system frequency.By using the variables in (5) and assuming ω ≈ ω n , (4) can be re-written in the Laplace domain as (6), where s is the complex angular frequency.The FCR provider features a proportional controller and actuator that react to deviations in the system frequency.In the Laplace domain, the FCR power response can be represented in a simplified way as in (7), where k = Kω n /S n is the normalized FCR gain and the delay introduced by the actuator is modelled by a first-order low-pass filter (LPF) with a time constant T .From ( 6) and (7), it is possible to obtain the transfer function G(s) between p and f, as show in (8).
A constant imbalance of power p causes a steady-state variation in the system frequency that is inversely proportional to the gain k.It is important to emphasize that the inertia constant H does not influence the steady-state error.However, H, T , and k play a role in how fast and how smoothly f reaches the steady-state after a power imbalance.The damping ζ and the natural frequency of the system ω nat , which can be calculated by algebraic manipulation of (8), are a function of H, k, and T , as shown in (9).The higher the inertia constant, the more stable the system is.The longer the delay T , the more oscillatory the system is and the slower ω nat becomes.
When the primary reserve is provided by traditional GTs, the total intrinsic delay of the governor and turbine is in the order of hundreds of milliseconds [32].However, if the primary reserve is provided by an ESS, the delay drops by at least one order of magnitude [29].When T approaches zero, the transfer function between the power imbalance and the system frequency tends to a first-order low-pass filter.For the normal operation of an isolated grid, a two-fold advantage is obtained by allocating the FCR N to the ESS and leaving the GTs only with FCR D .Firstly, the GTs are allowed to operate at a more constant power which reduces the wear and tear of the mechanical parts.Secondly, the system becomes less oscillatory even without a virtual synchronous machine (VSM) scheme or derivative terms at the ESS reserves.
The FCR control strategy adopted in this paper does not include, on purpose, VSMs nor terms with the time derivative of the frequency.The reasons for that are: 1) the equivalence between frequency droop and VSMs has been pointed out by [33].2) enough physical inertia is available in the study case presented in Section III due to the GTs. 3) the time delay of the actuator controlling the primary reserve power has a considerable impact on the damping of oscillations in the system frequency.The longer the delay of the actuator, the more oscillatory the system becomes.As it will be demonstrated in this paper, the primary frequency reserves provided by a fast ESS can improve the stability of the system even without the use of VSMs.

III. STUDY CASE: AN ISOLATED INDUSTRIAL GRID
The study case used in this paper is based on an existing O&G platform in the North Sea.The platform operates isolated from the continent and is fed, in normal operation, by two 35.2MW aero-derivative single-cycle GTs.A techno-economical study [27] suggested that a reduction of approximately 30 % of the annual CO 2 emissions can be  achieved if the platform were connected to a 12 MW floating offshore WF.The reduction, however, relies on the installation of a centralized hybrid ESS with 4 MW proton exchange membrane (PEM) fuel cells and 6 MW PEM electrolyzers.A set of three 4 MW WTs was used by [6] as a prospective scenario for proposing a sizing methodology for the platform's hybrid ESS.This same scenario was also used by [7,34].Fig. 3 shows a single-line diagram of the study case.The average electrical load of the platform is 44 MW.Each of the 35.2 MW GTs is not able to, alone, feed the platform.Additionally, the thermal load of the industrial processes [9] requires one of the GTs to be in operation at all times.Due to safety concerns, the prospective scenario investigated by [6,7,34] assumed that two GTs would still operate simultaneously even with full production from the WF.The premise of two GTs operating at all times is kept in this paper to allow the re-use of the aforementioned scenario.
The loads of the study case are divided into three groups in Fig. 3.The first group represents the platform's water injection system, whose variable frequency drives can act as flexible loads (FLX) and provide primary frequency control reserves.For simplification purposes, the whole water injection system is grouped under a single 8.5 MVA active-front-end PEC.The second load group aggregates 26 MW of CPL.The third group gathers 11 MW of CZL.Both CPL and CZL can be changed in steps for testing the dynamic characteristics of the model.
The FLX are modelled as a single controlled current source that supplies constant power.Their grid converter (GC) operates as a dc voltage controller.On the ac side, the FLXGC regulates the reactive power exchange with the grid to zero.
The WTs are modelled in a similar way to the flexible loads (FLX).The turbines including generator and machine side converter are simplified to a single controlled current source that supplies constant power even when the dc-link voltage varies.The GCs of the WTs operate, also, as dc voltage controllers and regulate the reactive power exchange with the grid to zero.
For counteracting wind variability, a hybrid ESS is employed.A fast energy storage device (ESD) composed of a battery and a dc/dc converter provides reserves for the short term wind and load variations.The main goal of the battery is to reduce the burden of the GTs on the fast frequency control.The pair of electrolyzer and fuel cell form one single ESD.Energy is stored as hydrogen when there is wind overproduction and hydrogen is transformed into electricity when there is little production from the WF.Even though the reactive power provided by ESSs can have an important role in avoiding the loss of synchronism of nearby machines in the event of faults [35], the ESSGC of the study case keeps the reactive power exchange with the grid equal to zero.Reactive power support from the ESSGC is going to be addressed in a future work.Similarly to the GCs of the FLX and WTs, the ESSGC also runs as a dc voltage controller.The choice of setting the GC as a dc voltage regulator while the ESDs provide power to the dc link has been previously addressed by [7,36].The PowerFactory models developed by [7], which are publicly available at [37], are used as base for this paper.

A. Simplified Stability Assessment
It is interesting to assess the stability of the study case with the simplified model given by ( 6) when the FCR is shared between two GTs and the ESS.For that, the normalized p FCR is split into three components, one for the ESS and two for the GTs.The ESS component is modeled as a first-order LPF.For the GTs, two first-order LPFs are used in series representing the fuel valve and turbine delays.The model data is shown in Table I.The total power-to-frequency gain of the FCR is set to K N = 12 MW/Hz and is shared between the GTs and ESS.This gain is reasonably high when compared to the installed GT power of 88 MVA and the electric load of 44 MW.For large interconnected grids in North America, the typical gain in W/Hz is on the range of 10 % of the peak demand in W [1].
Fig. 4 shows the eigenvalues of the system obtained with MATLAB Simulink R2018a for seven different FCR sharing configurations as listed on Table II.The model and dataset is available at [38].For all FCR configurations, the eigenvalues associated with the ESS are not oscillatory and lie on the real number line.The governors are associated with the only oscillatory mode.When only the GTs are supplying FCR, this mode features a natural frequency of ω nat = 1.651 rad/s and a damping of ζ = 0.606.The more the reserves shift towards the ESS, the more the oscillatory mode moves towards the real number line.When only the ESS provides FCR, the imaginary parts of all eigenvalues are zero.

B. Detailed Stability Assessment
The mechanical rotating mass model which results in (8) disregards many interactions between the devices of the plat- form.CPL and converter-driven instabilities are not captured by this model.To gain an insight into these complex interactions, computer simulations were performed with DIgSILENT PowerFactory 2020 SP2A, model and dataset are available at [38].Fig. 5a shows seven instances of frequency changes caused by a step load of 1.2 MW.The frequency shown is the one at the platform's main busbar measured with a phase-locked loop (PLL) [39].The FCR N sharing is different in each of the seven instances as listed on Table II.The step load is applied at t = 1 s.Within the next 4 s, the frequency reaches a new steady state at 49.9 Hz.The response when the GTs are the only FCR N providers is shown in solid red.In this case, the minimum value reached by the frequency (known as nadir) is 49.886 Hz.The responses when the FCR N are shared between GTs, battery converter (BTC), and FLX are shown in dotted gray.The dash dotted blue curve denotes the response when only BTC and FLX provide FCR N .From a frequency control perspective, the system becomes more stable as the FCR N contribution is shifted towards the BTC and FLX.These faster primary reserves improve the nadir.However, the response of the main busbar voltage becomes slightly less damped (Fig. 5b).The voltage control dynamics of the active front-end converters ESSGC and FLXGC should not be completely disregarded in a stability analysis.
When the steady state is reached in the simulations shown in Fig. 5, the eigenvalues of the system are calculated together with the participation factors of each state in each eigenvalue.The result is shown in Fig. 6.See Table II for information regarding chart legend and markers.The devices whose internal states have a participation factor higher than 50 % are named on the side of some specific eigenvalues.The real and imaginary axes are partially linear and partially  Constant damping (ζ) lines are plotted to help identifying the more oscillatory eigenvalues.The system is stable in all configurations and, as expected, all eigenvalues have negative parts.Fig. 6a shows an overview of the complex plane.The negative half of the imaginary axis is mostly omitted as oscillatory modes appear as complex conjugate eigenvalues.The eigenvalue related to the BTC and the battery main reactor (BTL) leave the real number line and reach a damping of ζ = 0.94 with a natural frequency of 22 rad/s when the reserves are exclusively supplied by the BTC and FLX.The modes related to fuel cell converter (FCC) and fuel cell main reactor are marked with FCC,FCL and the ones related to the electrolyzer converter (ELC) and electrolyzer main reactor are marked with ELC,ELL.The eigenvalues related to the wind turbines including its converters are clustered together and are marked with WTs.The modes associated with the FCC, ELC, and WTs are not affected by the FCR N sharing.They are, nevertheless, influenced by the tuning of the controllers of their PECs.However, due to brevity concerns, the influence of controller tuning in the location of these eigenvalues will not be investigated in this paper.It is worth noting, though, that FCC, ELC, WTs, and BTC operate as constant power devices.Other eigenvalues worth remarking are the ones associated with the excitation system of the turbogenerators (Excs), the ones related to the grid converter of the ESS (ESSGC), and the ones related to internal states of the turbogenerators (GTs).
Fig. 6b shows a detail of the complex plane with eigenvalues related to the frequency measurement (performed with PLLs) of the flexible loads (FLXpll), battery converter (BTCpll), and secondary frequency controller (SECpll).When the reserves are provided by the GTs only, three eigenvalues are clustered close to the point −5 + j2.25.Each of those eigenvalues is related to one frequency measurement device.Once the FCR N is shifted from the GTs to the FLX and BTC, two eigenvalues start to move towards less damped regions of the complex plane.States of the PLLs of the BTC and FLX have a participation factor above 50 % in these eigenvalues.The eigenvalue related to the secondary frequency controller PLL, however, is not affected by the sharing of FCR N .Fig. 6c presents a detailed view including the oscillation modes associated with the governors and turbogenerators (Govs,GTs).The dynamic of this mode coincides with the one observed in Fig. 4. As indicated by the gray + markers, the less the GTs provide FCR N , the more this eigenvalue moves towards the real number line.The other oscillatory eigenvalues in Fig. 6c are not affected by changes to the FCR N sharing.They are associated with internal states of the excitation systems (Excs) and turbogenerators (GTs).

V. EXPERIMENTAL RESULTS
In this section, the PowerFactory model used in Section IV-B is compared to a model running in a RTS [40] in a scaled-down PHIL test setup at the National Smart Grid Laboratory of NTNU, as seen in Fig. 7.The hardware under test is composed of the ESSGC, dc link capacitance, and inductive-capacitive-inductive filter.These devices are marked  with a dashed rectangle in Fig. 3 and their connections to the ac and dc grid emulator [41] are illustrated in Fig. 8.The ac grid emulator runs as a controlled voltage source and is connected to the high voltage side of ESSTR.The dc grid emulator runs as a controlled current source and feeds the dc link of the ESS with the net current coming from the ESDs.The scaling of the hardware under test is presented in Table III.
Three dynamic test cases are devised.Case 1 (GTs): FCR N provided only by the GTs.Case 2 (BTC+GTs): FCR N provided by BTC and GTs.Case 3 (BTC+FLX): FCR N provided by BTC and FLX.For all cases the total FCR N reserve is set to P N = 3 MW.A step load of 3 MW is chosen as test transient.
According to [6], this is the maximum expected load variation under normal operational conditions with 99.9 % of probability for the platform.The step load is divided between CPL and CZL proportionally to their rated values.The boundary for normal operation is set to f N = 1 Hz, or 2 % of the rated system frequency.This is larger than the 0.2 % limit required for the Nordic region [3] but lower than ±3 % defined by the recommended practices for maritime vessels [28].In all three cases, the FCR N providers operate with a 0.25 % frequency dead band and their frequency-to-power gains are compensated with (3).It is worth remarking that only the GTs are selected as FCR D providers.Nonetheless, given the choices of P N and f N , a step load of 3 MW does not activate the large disturbance reserves.
The comparison of the PowerFactory model and the PHIL setup are shown in Fig. 9.The results obtained with powerhardware-in-the-loop are marked with "PHIL" and the ones obtained with PowerFactory are marked with "SW".Initially, the system is in steady state and the ELC is absorbing power (Fig. 9i).At t = 10 s, the step load of 3 MW is applied (Fig. 9e).The initial transient caused by the step load is noticeable at the busbar voltage (Fig. 9a) and at the total power supplied by the WF (Fig. 9c).The imbalance of power of 3 MW is, initially, fully covered by the GTs (Fig. 9d) and causes the ac frequency to drop (Fig. 9b).When the FCR N are shared between BTC and GTs (Case 2), the burden on the GTs is quickly halved.For Case 3, when the FCR N are provided exclusively by the BTC and FLX, the power of the flexible loads (Fig. 9f) is reduced proportionally with the frequency deviation.This, together with the power delivered by the batteries (Fig. 9h), drives the generator power back to the level of before the step.The secondary frequency controller is activated at t = 30 s.The ELC is quickly shut-down and the FCC (Fig. 9g) slowly starts to supply power regulating the frequency back to the nominal.

VI. DISCUSSION AND FUTURE WORK
The provision of primary reserves by the faster ESS and FLX lessens the burden of the slower GTs with frequency control which leads to reduced wear and tear of the turbine governors, reduced NO x emissions, and an improved overall electric power quality in the platform.Although the higher participation of the faster reserves results in a more damped response of the frequency after sudden load changes, there is a non-critical increase in the oscillations at the main busbar voltage.Therefore, interactions between the excitation system of the GTs and the reactive power control of the ESSGC need special attention.Additionally, the tuning of the ESS and FLX frequency measurement devices needs to be carefully performed as the oscillation modes associated with them move towards less damped regions of the complex plane when the contributions of these providers increase.Nevertheless, the eigenvalue analysis and dynamic simulations performed with PowerFactory and the results obtained with the PHIL test setup indicate that the benefits, from a grid stability perspective, outweigh the disadvantages of increasing the share of converter   interfaced FCR N reserves for primary frequency control in isolated grids.
As the total allocated reserve power is maintained by the PMS, a higher participation of the battery in the primary frequency control means a higher FCR gain in the BTC and a lower gain in the governors.This results in increased damping of an oscillation mode associated with governors and GTs, whereas oscillation modes associated with the BTC and battery main reactor (BTL) become less damped.While the reduction in damping of the BTC and BTL modes is not critical, as ζ is still larger than 0.87, the increased damping of governors and GTs modes is much more expressive, from ζ = 0.64 to the real number line.Notwithstanding, there is a series of modes with ζ close to or lower than 0.5 which are associated with constant power devices as the fuel cell, electrolyzer, and WTs.Different tuning strategies for the controllers of these devices influence the location in the complex planes of these oscillation modes.However, an assessment of such tuning strategies is considered outside the scope of this paper and will be addressed in a future work.
There are a few discrepancies between the results obtained with the PHIL setup and the PowerFactory simulations.The most noticeable one is in the FCC power which is due to higher losses in the scaled-down hardware devices (ESSGC, ESSL ac , and ESSTR) that are not present in the PowerFactory simulations.Normalized resistive losses in laboratory equipment as transformers, reactors, and converters tend to be higher than the normalized losses in their full-scale counterparts.A compromise between reducing losses and matching reactance and capacitance values in pu has to be made for a scaled-down PHIL test.This topic is going to be addressed in detail in a future work by the authors.

VII. CONCLUSION
In this paper, the Nordic synchronous system concept of islanded operation frequency containment reserves (FCR I ) was expanded and subdivided into two categories.This strategy of categorized FCR I was applied to the study case of an autonomous complex industrial system which is fed by traditional GTs and by a WF, is dominated by CPLs, equipped with fast flexible CILs, and supported by an ESS.The analyses performed in this work took into consideration the detrimental effects of the CPLs and demonstrated that the overall stability of the system increased by shifting the primary power reserves from the slow GTs to the fast ESS and CILs.This also allowed the GTs of the study case to operate at a more constant power, which has the potential to reduce wear and tear in the turbine governors.While the reduction in the damping of oscillation modes associated with the CPLs and the PECs were not critical, the oscillation mode associated with the slow turbine governors were considerably damped when ESS and CIL reserves were prioritized.The results of this paper were supported by computer simulations, made publicly available, and PHIL tests.They demonstrate the versatility of the expanded FCR I concept for coordinating fast primary power reserves in autonomous grids with increased participation of non-synchronous intermittent RESs.
: D. Mota, E. Alves, and E. Tedeschi are with the Department of Electric Power Engineering, Norwegian University of Science and Technology (NTNU), Trondheim, Norway.S. D'Arco and S. Sanchez-Acevedo are with SINTEF Energy AS, Trondheim, Norway.E. Tedeschi is also with the Department of Industrial Engineering, University of Trento, Trento, Italy.

Fig. 2 :
Fig. 2: FCR N and FCR D characteristic of the system.

Fig. 3 :
Fig. 3: Single line diagram of the study case.

Fig. 4 :
Fig. 4: Eigenvalues with linearized rotating mass model for a total gain of 12 MW/Hz and different sharing of FCR N between ESS and GTs.

Fig. 5 :
Fig. 5: Voltage and frequency at the main busbar during a step load of 1.2 MW.

Fig. 6 :
Fig. 6: Complex plane with eigenvalues, for legend information see Table II.

Fig. 7 :
Fig. 7: PHIL setup at the National Smart Grid Laboratory.

Fig. 9 :
Fig. 9: Response to 3 MW step load with different sharing of frequency reserves.

TABLE I :
Simulation data for Fig.4.

TABLE II :
FCR N sharing values and chart information for Figs. 4 to 6.

TABLE III :
Scaled down PHIL and full size converter data.H equals the energy in the capacitor at rated dc voltage divided by the converter rated apparent power.