A power-hardware-in-the-loop testbed for intelligent operation and control of low-inertia power systems

1 Motivation

To reduce greenhouse gas emissions and fossil fuel consumption in the area of energy systems, the worldwide usage of renewable energies has increased remarkably in recent years [1]. Two key aspects of this transition are the integration of a larger number of geographically dispersed small-capacity distributed generation (DG) units based on renewable energy sources (RESs) that are interfaced to the grid through power-electronic inverters (often termed as inverter-based DGs) as well as the need for advanced control and communication technologies [2–4]. The physical differences between inverter-based DGs and conventional fossil fuel-driven rotating machines imply a reduction in the overall power system inertia, thus potentially decreasing the system robustness against disturbances [5, 6]. Such power systems with a high share of inverter-based DGs and loads are therefore commonly termed as low-inertia power systems.

The new structural and dynamical properties of low-inertia power systems result in the need for novel intelligent control and operation strategies suitably tailored to such systems. An important aspect thereby is that due to these substantial changes in the system properties, the mere usage of simulation-based studies for performance demonstration and validation of novel control and operational schemes has become insufficient. In other words, the need for testing facilities to accelerate the design and evaluation process has become crucial [7].

For such purposes, hardware-in-the-loop (HiL) test environments are a well-established concept for control system design and evaluation [8]. The HiL methodology facilitates the development and test of complex systems with different applications in the industry by combining a real-time simulation platform with hardware and/or control devices under test. The power-hardware-in-the-loop (PHiL) concept is an extension of the conventional HiL approach to power (systems) applications [9]. The main modification in PHiL compared to HiL is the inclusion of actual (electric) power in the loop, which results in a physical testbed for power system components, such as inverters, synchronous generators (SGs) or protection devices, their controls, and even interconnected power systems. Thus, the addition of power together with the flexibility provided by the HiL framework puts PHiL testbeds at a highly promising position for versatile testing and validation of low-inertia power systems applications, both at an equipment level and also at a system-wide level [10]. In addition to hardware-related testing, PHiL testbeds also facilitate advanced testing of cyber-physical features in low-inertia power systems. As a consequence, several PHiL testbed facilities have been successfully established in different countries during the last decade for education and research purposes [7, 11, 12]. A comprehensive survey on such testbeds is provided in [13].

Motivated by the aforementioned advantages and importance of PHiL systems, a PHiL testbed facility for low-inertia power systems (230/400 VAC, 750 VDC, 110 kW) has been developed at the Control Systems and Network Control Technology department at the Brandenburg University of Technology (BTU), Cottbus-Senftenberg, Germany. The following are the key attributes of the PHiL testbed at BTU:

1. Functionality to operate as an AC, a DC, or a hybrid (AC and DC) low-inertia power system.

2. Flexibility to emulate a large variety of system and network configurations.

3. Deployment of control, operation, and estimation algorithms for individual components as well as interconnected systems in a rapid-prototyping fashion with the aid of real-time control units, which can be programmed directly from Matlab/Simulink.

4. Emulation of commonly observed disturbances and parameter uncertainties in low-inertia power systems, such as heterogeneous dynamics, fluctuating power flows, and cyber-physical properties.

The remainder of the paper is organized as follows. At first, we explain the characteristics and technical information of the hardware components, and then a comprehensive description of the key testing and validation capabilities, with a special focus on using custom-made control algorithms is presented. Subsequently, the testbed capabilities are illustrated via experimental results of a practically relevant use case and finally, we summarize the conclusions.

2 Composition and structure of the testbed

In this section, we introduce the electrical characteristics and technical details of the components in the PHiL testbed. The system has a total installed power of 110 kW and can be operated either as a single- and three-phase AC system (230/400 VAC) or as a DC system (±350 VDC, 750 VDC). The current setting of the test facility allows to electrically interface a grid emulator (GE) unit (50 KW) and four DG units (15 kW each) via a power line emulator cabinet,^[1] see Figure 1. The technical details of the GE unit and the DG units are summarized in Table 1 ^[2] and their characteristics are detailed next.

Figure 1:

A photograph of the PHiL testbed facility. From left to right: Grid emulator (GE) unit, power line emulation cabinet and four distributed generation (DG) units.

2.1 GE unit

The GE unit (50 kW) is a programmable power amplifier able to convert its three-phase 230/400 VAC power input (BTU grid) into a variable AC or DC power output. The unit comprises a three-phase four-wire AC/DC-DC/AC bidirectional converter equipped with a real-time control unit and is capable of emulating the dynamics of large-scale grids, different types of electric loads, and network disturbances.

The electrical hardware in the GE unit consists of a bidirectional AC to DC active bridge rectifier (ABR) with an LCL input filter, a common split DC link in a back-to-back configuration, and a bidirectional DC to AC converter (CONV) with an LC(L) output filter, and a real-time control unit, see Figure 2a. Both the ABR and the CONV stages consist of three-phase two-level insulated gate bipolar transistor (IGBT) bridges whose output is regulated via pulse width modulation (PWM) signals.

Figure 2:

Schematic representation of single-phase equivalent circuits corresponding to the GE and the DG units. Bold lines represent electrical connections, while dotted lines represent signal connections. The control signals provided to the insulated gate bipolar transistors (IGBT) of the active bridge rectifier (ABR) and converter (CONV) stages are denoted by u. The converter output is interfaced with the low-inertia power system at the point of common coupling (PCC).

The real-time control unit used in the GE is a powerful computer specifically designed for creating a model-based PHiL simulation platform as well as a communication infrastructure among the testbed components. This computer runs a Matlab/Simulink-compatible software, which allows generating time series and design control algorithms via standard Matlab/Simulink block models.

The control operation of the GE unit can be split into two layers. For the upper control layer, the real-time control unit processes the available voltage and current measurements at the PCC (see Figure 2a) and generates a set point signal w which is then transmitted to a digital signal processor (DSP) embedded in the power amplifier. Within this DSP, there are a set of predefined controllers for the ABR and the CONV stages that compute and generate the control signal u which is then provided to the IGBTs. Further details on the control architecture and how the set point w can be calculated are explained in the next section.

The real-time control unit can also exchange information with a main system control center for analysis, control, and performance monitoring purposes. In this regard, for synchronizing the timestamps of the measurements in different real-time control units, a precision time protocol (PTP) is employed [14]. This is necessary since – as in any real-world application – the clocks of the real-time control units are not synchronized [15]. Furthermore, direct peer-to-peer communication with other control units is also possible via a distributed communication network.

2.3 Power line emulation cabinet

This element allows us to electrically interconnect the GE and the DG units under different topologies as well as to simulate multiple types of short circuit events. The cabinet contains the π-equivalent circuits of two 0.5 km, two 10 km, and two 15 km distribution lines. The line parameter values are taken from real-world medium voltage 20 kV power lines and scaled to the laboratory setting with base values of 20 kV/20 MW for the original data and 400 V/100 kW for the laboratory, respectively. The structure of the cabinet allows connecting the power lines in a cascade manner for emulating additional power line circuits. In addition, it is possible to induce phase-to-neutral and phase-to-phase short circuits on any power line by closing the contactors K1 to K6 as shown in Figure 3.

A combination of the aforementioned components results in the current PHiL testbed, which is depicted in Figure 4.

Figure 3:

Schematic diagram of a threephase power line in the power line emulator cabinet, where a π-equivalent circuit and short circuit contactors are embedded.

3 Key testing and validation capabilities

Building upon the hardware architecture explained in Section 2, in this section, we showcase the main control, testing, and validation capabilities of the PHiL testbed. We start with the functionalities and flexibility regarding the control implementation. Then, we explain how some of the important physical and cyber-physical disturbances observed in low-inertia power systems can be emulated in the laboratory.

Figure 4:

Schematic diagram of the overall PHiL testbed architecture. Solid lines represent electrical connections while dotted lines represent signal connections.

Figure 5:

Exemplary hierarchical control architecture for the DG units and the GE unit in the PHiL testbed: (a) The control signal u provided to the IGBTs of the ABR and the CONV stages is generated by the real-time control unit, (b) the setpoint w provided to the embedded DSP is generated by the real-time control unit.

3.1 Control functionality

Often, in grid-connected power converters, a cascaded control structure is employed [16–18]. In such cases, the inner-loop control takes care of the fundamental control tasks, such as tracking of a current or voltage reference, and disturbance rejection, e.g. harmonics. The outer-loop control is then responsible for grid-related control tasks, such as the provision of ancillary services.

In standard PHiL applications, power amplifiers are used as power sources [10]. Then, the inner-loop controls are readily provided by the manufacturer, which implies that they cannot be freely adjusted by the user. Yet, in the context of low-inertia systems, this may be unsatisfactory, since already the dynamics of the inner-loop controls may impact the system stability [5, 6]. Likewise, the dynamics of frequently employed phase-locked loop (PLL) algorithms become increasingly relevant [5, 19]. Furthermore, the design of novel control schemes for low-inertia systems may require a complete redesign of the converter control structure [20, 21].

These considerations are taken into account in the system design of the four DG units in the PHiL testbed, where the operator has direct access to the reference signal provided to the IGBTs (signal u in Figure 2b). More precisely, the designer has the full flexibility to implement inner-loop controllers, e.g. following standard approaches [17, 22], or any other sort of converter control, e.g., synchronverter algorithms [20], as well as observer and estimation algorithms, such as for PLLs [17, 23]. Any such implementation can be done directly from Matlab/Simulink. This results in an efficient and easily accessible workflow since many control engineers are familiar with the Matlab/Simulink environment. For instance, to operate a DG unit in grid-feeding mode [16, 18], one can design the inner-loop current and power (or in some cases current) controllers in the converter, or for grid-forming operation, one can implement the inner-loop voltage and current controllers [16, 18, 24]. An example for the latter is schematically introduced in Figure 6.^[3] The aforementioned functionality with the DG units is a highlighting feature of the PHiL testbed.

Figure 6:

Schematic representation of a DG unit operating in grid-forming mode [16, 18, 24]. The CONV stage is connected to the PCC via a three-phase LCL filter, see also Figure 2b. In grid-forming mode, the voltage at the capacitor is regulated by using a cascaded current and voltage control (e.g., a proportional integral (PI) control in dq coordinates). The control signal u is then provided directly to the IGBTs of the CONV stage.

Recall from Section 2 that in comparison to the DG units, the GE unit is a standard power amplifier. That is, the inner-loop controls are already implemented (in the embedded DSP) and as an operator, we can specify the set points, i.e., the signal w in Figure 2a, either in terms of voltage or current. This is considered admissible, since the GE unit is intended to emulate the dynamics of a larger power system. In such a setting, a direct control of the IGBTs is not necessary, as the fast dynamics of the power electronics (including their inner-loop controls) are less relevant as in the case of individual DG units. Alternatively, the GE set points can also be calculated by using a higher-level controller, e.g., primary and secondary controllers.

The aforementioned control features for the DG units and the GE units are schematically explained in Figure 5.

4 Experimental case study

In this section, we present a detailed description of an experimental case study that demonstrates several of the main features of the PHiL testbed facility. The experimental setup emulates the behavior of a low-inertia power network with topology shown in Figure 7. In this setting, the GE emulates a larger power system, DG 1 operates as grid-supporting unit [16] and DG 2 operates as a constant power load. The motivation to choose such a configuration for the experimental case study is two-fold. On one hand we want to demonstrate that higher-level control laws can be implemented at the GE unit (see Figure 5b), e.g., a larger power network represented by the second order swing equation representing the aggregated frequency dynamics of all SGs in the power system [29, 30]. On the other hand, we want to show that one can design the inner-loop controllers of the DG units (see Figure 5a) for emulating grid-supporting and constant power load dynamics.

Figure 7:

Electrical topology of the experimental case study. The impedances Z _10 km and Z _15 km represent the π-equivalent power line circuits shown in Figure 4.

Following [29, 30], the frequency dynamics of a larger power system are emulated by the GE unit via the aggregated swing equation

where δ : R ≥ 0 → R is the phase angle, ω : R ≥ 0 → R > 0 is the rotational speed, M ∈ R > 0 is the moment of inertia of the aggregated rotor masses of the generators in the power system, ω d ∈ R > 0 is the desired synchronous speed, and D ∈ R > 0 is a damping factor. Furthermore, P m : R ≥ 0 → R > 0 and P e : R ≥ 0 → R > 0 are the mechanical power input and the electrical power output, respectively.

As mentioned previously, DG unit 1 is operated as a grid-supporting unit, whose main objectives are to feed/consume a specific amount of power^[4] and the same time, support in keeping the network frequency close to the nominal value [16]. The latter objective is realized with the use of a proportional control law [16]. For the aforementioned setting, the real-time control unit in the DG unit 1 regulates the output of the ABR stage through a standard proportional resonant (PR) controller [22], while, for realizing grid-supporting mode, the output power of the CONV stage is controlled by a PI controller in dq-coordinates [16, 22]. Finally, the control of DG unit 2 is also implemented in a similar way as that of DG unit 1, where the main difference is that DG unit 2 is operated as a constant power load. The experimental configuration mentioned so far is schematically detailed in Figure 8.

Figure 8:

A schematic of the experimental configuration, where the GE unit is connected to two DG units via 15 km and 10 km power lines, see Figure 7. The GE unit emulates the frequency dynamics of a larger power system via the second order swing Equation (4.1) the DG unit 1 operates as a grid-supporting unit [16] and DG unit 2 operates as a constant power load.

At first, we operate DG unit 1 without the grid-supporting capability, i.e., the frequency regulation term depicted in the real-time control unit box of DG unit 1 in Figure 8 is inactive (K _s = 0). Then, DG unit 1 operates as a constant power source/load (depending on the power setpoint P 1 ⋆ provided to it). For simplicity, throughout the experiment, we set P 1 ⋆ = 0 W. At around 5 s, we change the power setpoint of DG unit 2 ( P 2 ⋆ in Figure 8) to −5 kW, i.e., DG unit 2 is operating as a constant power load that consumes 5 kW. The experimental result with this setting is depicted in Figure 9a, where a frequency drop from 50 Hz to 48.95 Hz can be seen. The change in frequency in correspondence to a change in active power in the network is due to the swing equation dynamics (see (4.1)) implemented at the GE unit.

Figure 9:

Frequency and active power plots for the experimental scenario. At around 5 s, the active power setpoint of the distributed generation (DG) unit 2 P 2 ⋆ is changed to −5 kW. (a) Case 1: Grid-supporting action in DG unit 1 is inactive. Observe that the frequency drops to 48.95 Hz, see the zoomed frequency plot at 60 s. Furthermore, the rate of change of frequency (RoCoF) is approximately −0.04 and note that DG unit 1 does not inject/consume any active power, see the zoomed power plot at 60 s, (b) Case 2: Grid supporting action in DG unit 1 is active. Observe that the frequency drops only to 49.05 Hz, see the zoomed frequency plot at 60 s. Furthermore, the RoCoF is approximately −0.03 and note that DG unit 1 also injects an active power of approximately 500 W, see the zoomed power plot at 60 s.

In the second phase of the experiment, we activate the grid-supporting action of DG unit 1 by choosing a positive value for the control gain K _s (see the real-time control unit block of DG unit 1 in Figure 8). We fix the active power setpoints of DG units 1 and 2 to the same value as before, i.e., P 1 ⋆ = 0 W and P 2 ⋆ = − 5 kW. The experimental result for this scenario is shown in Figure 9b, where it can be seen that the frequency drop has decreased i.e., from 48.95 Hz (without grid-supporting action, see Figure 9a) to 49.05 Hz (with grid-supporting action). Similarly, a drop in the rate of change of frequency (RoCoF) can also be observed from the mode without grid-supporting action (Figure 9a) to the mode with grid-supporting action (Figure 9b), i.e., from RoCoF ≈ − 0.04 to RoCoF ≈ − 0.03 . In principle, by choosing a higher value for the gain K _s in the grid-supporting controller (recall Figure 8), the frequency drop from the nominal value (50 Hz) and the RoCoF can be further reduced. However, choosing a wider range of values for the gain K _s and studying the steady-state and transient properties of the frequency trajectory is beyond the scope of this paper and the reader is referred to [16] for more details.

5 Conclusions

Due to the increasing challenges encountered in designing and operating low-inertia power systems, the need for flexible testing facilities is inevitable. In this paper, we have presented a PHiL testbed facility that serves as a platform for testing and validating AC and DC low-inertia power networks with different topologies, cyber-physical characteristics, and custom-made controllers among other settings. Experimental results showed that the presented testbed is suitable for studying realistic and practically relevant case studies in a flexible and versatile fashion.

As a next step, we will incorporate an SG (and its corresponding real-time control unit) into the current setting such that a wider spectrum of operation scenarios in low-inertia power systems can be emulated.