Low-Cost Communication Interface between a Smart Meter and a Smart Inverter

Abstract

The need for a low-cost interface between the grid and small (<250 kW) renewable distributed energy resources (DERs) is growing in importance as the number of small DERs continues to grow. In this study, a system architecture was proposed to investigate paths to an affordable interconnection for small renewable DERs.Then, a low-cost communication interface between a smart meter and smart inverter was installed using a commercially available bridge device. The interface device was selected based on an assessment concluding that it would be able to support the emerging advanced metering infrastructure (AMI) network. Next, messages were passed across the experimental end-to-end communication interface to test their speed and reliability. Success was based on whether the key functions defined in the standard IEEE 2030.5 were executed or not, which include set points, disconnect/reconnect, and Volt-VAr optimization. The results of the testing provided detailed insights into the benefits and limitations of the proposed architecture. Intermittency of weather-dependent DERs (e.g., solar and wind) adversely impacts the power quality of a DER, making hourly day-ahead prediction nearly impossible. With this in mind, the investigation also considered the potential of using smart inverter functions to reduce DER’s intermittency.

1. Introduction

The growing pressure to generate more electricity from renewables and improve energy efficiency has promoted the deployment of distributed generation (DG) via distributed energy resources (DER) into electricity systems. Standard generator monitoring and control solutions employ hlremote terminal units (RTUs). At a typical cost in the range of $8000–10,000, RTUs are generally cost-prohibitive for DERs below 250 kW generating capability.

Virtual power plant (VPP) architecture is widely adopted for delivering cost-efficient integration of DER into the existing power system [1]. VPPs are formed by logically combining multiple DERs into groups that the utility can manage as a single entity. VPPs require a reliable communication interface. For example, VPP must carefully consider not only the impact on voltage and thermal limits, but also the overall stability of the grid. For example, Troester [2] warned about the risk of grid instability due to the high penetration of photovoltaic (PV) generators and discussed the grid codes that guard against the risk. The Electric Power Research Institute (EPRI) recognized that smart inverters provide value as DER penetration increases and that their flexibility will have considerable future value:

“A lesson learned from rapid DG deployment in Germany and elsewhere is to include the smart inverters on approved lists with flexibility for certain advanced functions to be enabled at a later date, as well requirements that include grid support capability before DG penetration reaches a level where it is needed” [3].

They also note the small incremental cost and encourage an early start with these capabilities as a proactive step. Smith et al. discussed the role of smart inverters in Volt/ volt-ampere reactive (VAr) Optimization programs for high penetration of PV DERs [4].

The two levels of DER integration are monitoring and control. The first level, monitoring, enables the utility to disaggregate net metering into generation and load. A smart inverter can measure the generated output power and other parameters observed at its terminals. These data, when combined with information that system operators can already access from site meters and other measurement points, allows them to see total site loads broken down into load and generation components. Knowing the load and generation components of a building or premise is essential for planning, forecasting, and grid operations.

Control is the second level of integration. It allows a grid operator to reach beyond the grid edge and change the settings of the DER to protect and optimize the overall system. The control incorporates independently owned DERs into grid management and optimization schemes. When this is not possible, grid operators have to compensate for DERs using resources on a case-by-case basis, which makes optimizing the entire system difficult. This approach paves the path to attaining the first objective of an Independent System Operator (ISO), to integrate DER into energy, ancillary services, and capacity markets [5].

The potential for leveraging the advanced metering infrastructure (AMI) network for monitoring DERs has been considered cost-effective for the utilities and has attracted recent research. McKenna et al. [6] discussed AMI’s potential for monitoring and impacting DERs at a high level. Huang et al. [7] discussed two-way communication via AMI, developed a communication model with accompanied network calculus, and validated the model in the simulation. AMI data have also been used to identify DER connectivity within a network [8].

Through this study, we aimed to achieve two objectives: (1) to investigate the use of an AMI network to monitor and control DERs, and (2) to learn whether an AMI-driven communication interface could be used to improve the power quality of DERs by reducing intermittency. In order to meet the first objective, we set out to identify opportunities and limitations associated with using the AMI network for monitoring and controlling DERs from the perspective of utility providers. For the second, we looked at the network from the viewpoint of independent DER owners, focusing on how its functionality might affect their access to energy and ancillary markets.

The role of an independent DER owner is expected to increase in time. For example, Reforming the Energy Vision (REV) program in New York State aims to produce 50% of the total energy from renewable resources by 2030 [9]. The density of independently owned renewable DERs will have to increase considerably to achieve this goal. At the current low penetration levels of DERs, the system modifications required to accommodate DER generation while maintaining the grid within safe operating limits are not very expensive. However, the cost will increase as DER penetrations continue to climb rapidly. Increasing the incremental cost of DER’s interconnection to the system drives the need to treat DERs as more active participants in managing grid conditions. IEEE 1547, a standard developed by the Institute of Electrical and Electronics Engineers (IEEE), is evolving to require DER to be able to perform grid support functions.

For example, ref. [10] discusses developing treatment of PV. The projected increase in responsibilities for independent DER owners is expected to be balanced by the ability to participate in the markets. The market requires the participants to predict their generation a day ahead, hour by hour. The intermittency of renewable generation is tied to the stochastic nature of weather. The weather forecast cannot provide accurate cloud coverage hour-by-hour 24 h in advance. Since the dynamics of weather models are chaotic and extremely sensitive to initial conditions (or measurements) [11], the required forecasting accuracy to continue operating the system may be out of reach in the foreseeable future.

On the other hand, average daily cloud coverage predictions are generally less uncertain. Thus, a way to achieve hour-by-hour, day-ahead power generation is to pair the generator with sufficiently large storage. Unfortunately, the cost of storage is relatively high compared with the cost of generation. There are trade-offs between storage requirements and the forecasting horizon. In addition to the power level, there are requirements on the acceptable noise or flicker; the standards for the utility power quality are governed by the information technology equipment (ITE) curve, formerly known as the consumer business equipment manufacturing (CBEMA) curve [12,13]. The improvement of power quality was identified as one of the critical roles of the microgrid [14]. While some authors questioned issues associated with intermittency from the point of view of policy and investigated whether it is an actual technical barrier or a rhetorical excuse [15], the technical community agrees that the intermittency of renewables must be carefully managed. For example, a recent publication of the National Academy of Sciences recognized forecasts that fossil fuel plants are likely to continue to be used to compensate for the fluctuations in wind and solar generation [16]; see also [17,18].

The current study was conducted using the smart inverter’s functionality for reducing storage requirements associated with day-ahead, hour-by-hour power-generation forecasting.

2. System Architecture

The investigation considered two versions of the system architecture: one for laboratory-level testing and the other for field deployment (only at the conceptual level). While the emphasis of the study was on laboratory testing, the approach had to be compatible with a practically deployable solution. The two architectures are discussed in turn.

2.1. Laboratory System-Level Architecture

Figure 1 depicts the proposed architecture. The inverter connects to ConnectDER’s bridge, which wirelessly connects to the ConnectDER collar, with integrated Mili NIC, installed on an Itron smart meter. Using the Itron radio frequency (RF) mesh network, the collar connects to the Itron internet of things (IoT) router, which in turn connects to the Itron StarFish development platform [19]. The communications protocol will consist of MODBUS RTU commands wrapped in constrained application protocol (CoAP) messages. The cost of the interface was an order of magnitude lower than RTU. The installation cost of the early version of the hardware is less than $500, with the possibility of decreasing to below $200.

CoAP is a simple, RESTful web transfer protocol designed specifically for IoT and resource-constrained equipment [20]. It is for machine-to-machine applications such as smart energy and building automation. CoAP has the advantage of being low-cost and secure. It can work with 10 KB of RAM and 100 KB of code space. At the same time, its ease of use and platform independence make CoAP more attractive to users. CoAP supports data types such as text, HTML, XML, and JSON.

Itron’s platform supports the end-to-end implementation of the CoAP protocol using the “MilliNIC” device. The MilliNIC is a radio communications module providing access to the Itron AMI network. The platform contains an inexpensive microcontroller, allowing equipment providers to create a low-cost AMI client that can be used for monitoring and control. The CoAP implementation includes gateway software as a proxy between the client and the end device. A CoAP API completes the solution, providing a complete interface for client applications.

MODBUS is a data communications protocol originally published by Modicon in 1979 for use with its programmable logic controllers (PLCs). MODBUS is ubiquitous in industrial automation, as it is an easily understandable (though not very sophisticated nor secure) protocol. More secure MODBUS implementations are conceived [21] but have yet to be widely adopted. MODBUS, however, is widely used as a standard communication protocol for grid edge DER devices such as solar inverters, storage batteries, and electric car chargers. It is so pervasive that the Sunspec Alliance [22] based many of its DER communication standards on the MODBUS data register structure.

The monitoring and control of the microgrid inverter were performed using simple MODBUS register reads and writes. The solution monitored voltage, current, real power, reactive power, power factor, and other set points exposed by the solar inverter. In addition, it curtailed power output and controlled VAR production levels to demonstrate grid optimization capabilities using DER resources.

This architecture efficiently implemented a low-cost smart-meter-to-smart-inverter interface because it leveraged existing and planned infrastructure and requires minimal additional hardware to implement. The architecture was demonstrated leveraging ConnectDER technology, which allowed testing of critical functionality of the connection. The essential functionality consisted of metering (separating energy consumption from energy generation), monitoring, and selected control (changing inverter settings, e.g., curve set according to California Rule 21 [23]).

2.2. Conceptual Architecture for Field Deployment

The wider field deployment will likely require some modification to the architecture designed for the prototype described above. The prototyping approach paid careful attention to the possible field solution. However, it recognized that the final deployment architecture would be affected by future decisions of multiple stakeholders: the utility, independent DER owners, smart-meter manufacturers, and interface developers.

With the uncertainties associated with those decisions, the team proposed a tentative block diagram for a possible field solution, depicted in Figure 2. In this configuration, the ConnectDER provides two wireless interfaces: an Itron mesh connection (used to talk to the AMI network) and a Wi-Fi client (used to communicate with the on-site inverter). This adapter is required because in the demonstration configuration, the meter itself cannot assume this gateway role. As stated before, the fieldable architecture would likely differ from the one discussed here.

3. Hardware Selection and Installation

To support the SB 3.0 and the Fronius 3.8 hlresidential-style inverters, we modified the microgrid to include a split phase 240 V service and subpanel, similar to what you would see in a residence or small commercial location. Figure 3 shows the block diagram of the new microgrid. The modified microgrid features a new transformer, the 240 V residential-style panel, and two new inverters. Also installed was a residential metering socket to monitor energy flowing into and out of the 240 V subpanel, installed after the transformer (installed in this location so that transformer inefficiency, however small, would not significantly affect the data).

Together, these changes prepare the laboratory microgrid for the remaining project tasks and position the testbed for small-scale research more closely resembling residential and light commercial DER systems.

Figure 4 shows the physical connections between the smart meter and the smart inverter. Two collars were installed on the smart meter to interface with two inverters. The ConnectDER dongle connects to an inverter via a physical wire, using an RS-485 interface, and wirelessly to the Itron IoT router, which acts as a virtual network collector: it receives messages via Mili NIC and relays to the university information technology (IT) network through the Starfish development portal. The ConnectDER Wi-Fi Bridge device (the “dongle”) communicates with the SMA or Fronius inverters.

4. Results

The results consist of three subsections: performance of the communication interface, testing functionalities, and controlling intermittency.

4.1. Performance Testing of the Interface

The sequence diagram of the end-to-end message flow across the communication link that contains internet, AMI, WiFi, and RS-485, as depicted in Figure 5.

We measured the round-trip time (RTT) of the messages sent through the end-to-end communication link. Figure 6 shows the results. The subplot to the left shows the RTT data points received over time, while the subplot on the right shows the corresponding histogram. Most RTT data points were in the order of a few seconds, but occasional round trips took an order of magnitude longer.

To further enable evaluation of the network performance, one of the inverters was connected to the VPP directly, as depicted in Figure 7. In addition to the end-to-end communication link that connects the VPP to the Fronius Primo 3.8 inverter, the system employed a direct interface to the SMA Sunny Boy 3.0 inverter.

Figure 8 shows $RTT$ data associated with the direct SMA-to-VPP interface using the same visualization. $RTT$ associated with the direct interface is more than 20 times shorter, with 99% of RTTs taking ≈ 0.21 s (compared to 4.25 s associated with RTTs of the end-to-end communication link) and mean ${\mu }_{RTT}$ = 0.12 s (compared to 2.25 s of the end-to-end communication link).

Overall performance of the communication link is adequate for metering. However, this view is somewhat simplified because inverters must be scanned on a regular basis to ensure that they continue to operate with a known configuration and firmware. Figure 9 shows the information exchange associated with the SMA scan, which features 13 round trips, and the time needed for the overall scan is 13 RTTs. The models used in scanning are described in the table. On average, a scan over the communication link would take less than a minute (13 × 2.52 = 32.8 s). Nevertheless, it can be at a minute scale for more conservative estimates if the sequence of messages accounts for occasional longer delays, which can take up to ≈20 s for a single round trip.

Limiting power requires four round trips, as illustrated in Figure 10.

IEEE 2030.5 is gaining acceptance [24], and the first inverters (e.g., Enphase Energy) with built-in capability are emerging [25]. IEEE 2030.5 is a rich XML protocol that requires RESTfull web service, but it is unsuitable for a lean micro-controller and severely resource-constrained mesh network.

This apparent incompatibility can be addressed by using an edge device, as illustrated in Figure 11, which shows the interface between the inverter via an edge device and the AMI headend system. The role of the AMI headend system is typically two-fold: to acquire meter data automatically and to monitor parameters acquired from the meters (see, e.g., [25,26]).

The proposed edge device interacts with the inverter at a much faster rate. Then, it notifies the VPP via the AMI communications network only when the state changes or responds to the request. The mechanism also reduces the number of end-to-end round trips through the interface. This approach can allow reporting by exception when trigger limits are exceeded beyond the ride-through interval.

Figure 12 illustrates the concept of reporting by exception. $\Delta t_{cL}$, $\Delta t_{cU}$, and $L_{d}c$ denote the minimum time to report below the lower limit, minimum time to report above the upper limit, and the length of data capture, respectively.

4.2. Testing Functionalities

The critical testing functionalities include disconnect/reconnect, set point change, and Volt-VAr optimization. They are discussed in turn.

4.2.1. Disconnect/Reconnect Functionality

Disconnect/reconnect functionality is demonstrated in Figure 13, which shows the disconnect of the SMA inverter on command.

The disconnect was implemented using SunSpec Model 123 (refer to

Table 1. SunSpec models leverage for the implementation, with their applicability to the inverters and description.

Model	Fronius	SMA	Description
1	✓	✓	Common Model (model, mfg, etc.)
11		✓	Ethernet Interface Details
12		✓	IP Networking Details
102		✓	Split Phase Inverter Measurement (I16)
111	✓		Multi-Phase Inverter Measurement (F32)
120	✓	✓	Nameplate Ratings
121	✓	✓	Basic Inverter Settings
122	✓	✓	Extended Measurement and Status
123	✓	✓	Immediate Inverter Controls
124		✓	Basic Storage Controls
126		✓	Static Volt-VAr curves
127		✓	Parameterized Hz-W params/gradients
128		✓	Dynamic reactive current params/gradients
131		✓	W-PF Curves
132		✓	V-W Curves
160	✓	✓	MPPT Extension Module

). The inverter’s response to disconnect is immediate. The response to reconnect is also immediate, but the inverter does not resume producing power until default IEEE 1547-2018 Enter Service Criteria—Delay Before Export of 5 min.

4.2.2. Set Point Change—Limiting Power

Limiting active power demonstrated the set point functionality. Figure 14 shows the data from the experiment conducted on 12 December 2021.

The set point for the SMA inverter was 16.7% of the maximum power, which translates to 500 W. Power generated by the Fronius inverter was not subjected to limits and was used as a reference. It is shown in the time domain on the lower-right subplot, rotated by 90° to map the values on the x-axis of the plot. The limited SMA power is shown in the upper-left corner; the scatter plot with Fronius active power is on the x-axis, and SMA active power is on the y-axis. The scatter indicates a high correlation between the power of the two inverters, except when the power of the SMA is limited to 500 W.

4.2.3. Volt-VAr Optimization

The Static Volt-VAr function provides one or more Volt-VAr arrays, each of which consists of Volt-VAr pairs: a set of voltage levels and their corresponding VAr levels that connect as a piecewise linear function with hysteresis. Hysteresis is not required; if not present, the falling voltage curve follows the rising curve. The Volt-VAr curve defined by CA Rule 21 was used as a starting point for demonstration.

Figure 15 shows the Volt-VAr curve defined by CA Rule 21 [23]. It is a piecewise linear curve that consists of five segments. As stated above, both abscissa and ordinate are normalized and expressed on the percent scale: the abscissa (voltage) is the percent of $V_{Ref}$ and the ordinate (reactive power) is normalized with respect to the maximum commissioned power $P_{max}$. The curve is antisymmetric relative to the coordinate ($V/V_{ref}$, $Q/P_{max}$) = (100,0) point, corresponding to $V=V_{Ref}$. It does not feature hysteresis. There is a dead zone in the middle for the voltage in ±3.3% of the reference voltage, where reactive power is set to zero. The curve is capacitive for a voltage of less than 3.3% compared with reference and changes linearly until the reactive power reaches 30% at $V=0.92V_{Ref}$. For $V\le 0.93V_{Ref}$, the curve saturates at 30%. Similarly, for $1.033V_{Ref}\le V\le 1.07V_{Ref}$, reactive power decreases linearly until it reaches −30% and then saturates. As the building voltage had a relatively small variation, less than ±3%, the curve had to be modified to show the effect.

The modification removed the dead zone segment from the volt-VAr curve defined by California Rule 21, as shown in Figure 16.

Programming and enabling optimization curves was not trivial. It was impossible without support from the inverter manufacturer. SunSpec devices are only required to implement device identification. All other models are optional. This experimentation determined that while one of the subject inverters was marketed as having SunSpec support, it does not have any support for autonomous curve control.

Requests to the Fronius support team have yet to receive responses. SMA shared the necessary information to make programming the optimization curves possible. However, the curve-programming process included unexpected steps. First, the user must have a unique installer passcode to modify the curves. Second, the physical storage that records the curve parameters is of electrically erasable programmable read-only memory (E²PROM) type. A high (≈1000) but finite number of overwrites is possible.The user can change the reference voltage $V_{Ref}$ and supply the pair of values. There was no curve by default.

As a note on topology, it is important to emphasize that both inverters have essentially identical terminal connections, often referred to as 240 split phase (see the block diagram in Figure 17), but their interpretation is different: Fronius Primo 3.8 interprets the connection as a single 240 V (which corresponds to $V_a-V_b$) and SMA Sunny Boy 3 as a two-phase system with two voltages, $V_a$ and $V_b$, with reference to neutral N. The SMA implementation of a Volt-VAr curve operates on average phase voltage, $(V_a+V_b)/2$.

The curve of Figure 16 was implemented twice, using two different reference voltages, as shown in Figure 18a, which shows the scatter plot of reactive power Q vs. voltage V. There are three groups of points: the first cluster corresponds to the data before the curve, where $Q\approx 0$; the second cluster corresponds to the modified curve with reference voltage $V_{Ref}$ = 120 V; and the third cluster corresponds to the curve with the reference voltage $V_{Ref}$ = 125 V. Figure 18b zooms into the second cluster. Figure 18c shows the same data as Figure 18a, but in the time domain; and Figure 18d shows the time-domain data associated with the close-up of the second cluster.

Figure 19 shows an example of a full-day, static Volt-VAr optimization. Figure 19a shows that the points closely fit the programmed line, and Figure 19b shows the primary waveform in the time domain, phase voltages $V_a$, $V_b$ and their average on top; active and reactive power on the bottom. The power waveforms show that the controls are working well as long as there is enough generated power. Sunrise and sunset correspond to two cases with insufficient generated power.

In summary, static Volt-VAr optimization may require steps not included in SunSpec models, such as entering the installer’s password, and may be limited to a finite number of modifications; however, when implemented, it works remarkably well. While the number of changes to the E²PROM is finite, it is more than sufficient for any practical use case. Seasonal curve updates may be beneficial, but more frequency changes are not foreseen in the near future. Thus, the budget of 1000 4 E²PROM modifications, with four edits per year, would be sufficient for 250 years, which is more than an order of magnitude longer than the expected lifetime of the equipment.

5. VPP Implementation

The project’s focus was the development of the capabilities of individual inverters. However, considerations related to scale and generalization of the approaches to different makes and models were also considered.

Grouping aimed to support VPP. VPP does not have a commonly-agreed-upon definition; two main distinct types were identified as commercial VPP (CVPP) and technical VPP (TVPP) [1]. CVPP performs commercial aggregation with little concern regarding operations stability. In contrast, TVPP consists of DER from the same geographic location and includes the real-time influence of the local network on the DER aggregated profile as well as representing the cost. For the operating characteristics of the portfolio, we refer to [27]. The grouping performed here was primarily driven from the TVPP perspective.

IEEE 2030.5 Schema

In addition to Fronius Primo 3.8 and SMA Sunny Boy 3.0, the system monitored a Schneider ConextCL 3-phase, 18 kW inverter, which was inverting solar power of a set of panels on the university campus, but outside of the RIT-GIS microgrid. To enable consistent and coherent addressing of the inverters, IEEE 2030.5 developed a schema where the inverters have a top-down hierarchy: system, sub-transmission, substation, feeder, transformer, and service point, as shown in Figure 20.

IEEE 2030.5 requires schema in which all inverters have a universally unique identification (UUID) stored in a metadata table alongside the IEEE 2030.5 grid coordinates. These data hold information including the network address and connection type (direct transmission control protocol (TCP), SunSpec, proprietary protocol, or Starfish address). Figure 21 shows the database schema.

Individual DERs are referred to by the device UUID from within the software, providing the DER with a consistent identification even if the device’s location on the grid changes. For example, a grid location change occurs when a building is reconfigured to source power from a different feeder.

Implementation adhered to the hierarchical schema by mapping the hierarchy to the campus power network. The campus network employs two feeders, A and B, of 4.2 MW and 3.9 MW, respectively, as illustrated in Figure 22. There are PVs on circuits B1, A3, and B4.

One of the purposes of IEEE 2030.5, from a control perspective, is to control groups of DERs based on their position on the distribution network. IEEE 2030.5 commands were designed to facilitate this control. Many commands—including those that control total and reactive power—are phrased as percentages of the commissioned power. The percentages allow the VPP operator to issue and distribute fair distribution of curbing of energy generation.

Operators envision broadcasting a message to all DERs on a feeder, segment, or transformer. The idea of a broadcast is that the message is sent once, and multiple devices receive it simultaneously. The Starfish network cannot do this, so broadcast behavior has to be emulated by sending individual messages to each DER, which limits the speed at which one can control many DERs. The AMI network is based on IPv6; technologies such as IP Multicast could help solve this, but even with these, protocols on the AMI network would need to include a unicast “cleanup” semantic (to ensure every DER has received the message).

6. Controlling Intermittency

One of the key roles of microgrids is to improve the power quality of the intermittent, weather-dependent DERs, such as wind and solar [14,20]. New York independent system operator (NY-ISO) requires day-ahead, hour-by-hour forecast from the grid generators to be able to coordinate their operation [28]. PV generators can be forecasted well in the limiting cases when the cloud coverage is very low or very high, i.e., clear sky or overcast. While the ever-increasing power of machine learning suggests the promise of the improved forecasting of PV generation, research has shown that “the main origin of forecasting errors comes from the accuracy of weather prediction information” [29]. Day-ahead weather forecasting is generally very reliable, but sun-and-cloud prediction does not have hour-by-hour accuracy and resolution. Thus, some machine learning research hasfocused (and showed improvements) on a much shorter-term prediction, viz. 5-, 10-, and 15-min forecasts [30].

Unfortunately, these short-term prediction improvements are irrelevant to the current ISO operations. In addition, these forecasting limitations will likely persist in the foreseeable future because weather is intrinsically highly nonlinear, and successful weather models have incorporated these nonlinearities. It has been known since the 1960s that these nonlinearities exhibit strong dependences on tiny variations of the initial conditions [31]. Figure 23 illustrates this sensitivity, a popular butterfly effect, by running a dynamical simulation using the original Lorenz equations

$$\begin{array}{cc}\hfill \dot{x}& =\sigma (y-x)\hfill \\ \hfill \dot{y}& =x(\rho -z)-y\hfill \\ \hfill \dot{z}& =xy-\beta z\hfill \end{array}$$

(1)

with $\sigma$ = 10 (Prandtl number), $\rho$ = 28 (Rayleigh number), and $\beta$ = 8/3 [32]. The plot shows that tiny changes in the initial conditions (≤±0.1%) give rise to significant divergence of the final trajectory, despite all solutions being bounded on the manifold. Each solution can end up anywhere on the manifold, independent from its original neighbors. Any inevitable uncertainty associated with measurements that are used to supply the models with initial conditions can significantly impact the final results. Weather forecasts typically run a large number of simulations and interpret the variety of outcomes to address this inherent problem. Since the predictions can be markedly different, the model assigns different probabilities to different predictions. This uncertainty is inherent in the weather behavior. Depending on the geographic location, during some seasons, the uncertainty is greater than during other seasons.

To examine a few practical benefits of controlling power to limit intermittency, we will consider firming renewable with storage with and without limiting the power and renewables without storage. Before we start the investigation, we need to address intermittency metrics.

6.1. Quantifying Intermittency

The standard definition of intermittency has yet to be adopted. In our prior work, we defined intermittency I as the ratio of standard deviation of the solar power $p_{PV}\left(t\right)$ and its mean, over a time interval of interest $\Delta t$

$$I(\Delta t)=\frac{\sqrt{{\langle {\left(p_{PV}-{\langle p_{PV}\rangle }_{\Delta t}\right)}^2\rangle }_{\Delta t}}}{{\langle p_{PV}\rangle }_{\Delta t}}=\frac{{\sigma }_{p_{PV}}}{{\mu }_{p_{PV}}}$$

(2)

A similar metric, based on statistics of the PV power signal $p_{PV}\left(t\right)$, was proposed in [33], who defined intermittency as the interpercentile gap of the normalized PV power over the time interval $\Delta t$

$$\begin{array}{cc}\hfill I(\Delta t)& ={interpercentile}_{5,95}\left(\frac{p_{PV}(t+\Delta t)-p_{PV}\left(t\right)}{{\langle p_{PV}\rangle }_{\Delta t}}\right)\hfill \\ \hfill & =F^{-1}\left(0.95\right)-F^{-1}\left(0.05\right)\hfill \end{array}$$

(3)

where $F^{-1}$ is the inverse cumulative distribution function (CDF) of the random variable, defined as the normalized power over the time interval $\left(p_{PV}(t+\Delta t)-p_{PV}\left(t\right)\right)/{\langle p_{PV}\rangle }_{\Delta t}$.

A large number of various additional metrics, including those based on different statistics of power samples and those based on spectral power density, were summarized in [34].

To determine which metric to use, we compared the intermittency using Equations (2) and (3), with $\Delta t$ = 1 h (because of the day-ahead hour-by-hour ISO generation scheduling), as shown in Figure 24. The comparison indicates that the two metrics are highly correlated, therefore, values in the metric as computed using Equation (3) are higher.

Figure 25 shows a close-up into time 12:00–13:00, with histograms of power distribution, with the associated kernel density estimation (KDE). The KDE was computed using the Gaussian kernel estimator implemented in SciPy [35]. Since the number of points in the interval is less than a hundred, the cumulative distribution function (CDF) estimated from the histogram may not have enough resolution to determine 5% and or 95% points. We can either approximate it to the first/last data points, as shown in Figure 24, or use KDE to extrapolate.

Due to its more straightforward form and simpler computations, we selected the metric given by Equation (2). Finally, it is convenient to have average daily intermittency, defined as the average of hourly intermittencies for the hours where generated power is greater than zero.

$${\langle I\rangle }_{day}=\frac{1}{{\sum }_h{\mathbb{I}}_{\langle P\rangle >0}\left({\langle p_{PV}\rangle }_h\right)}\sum _{h=1}^{24}I(\Delta t=1){\mathbb{I}}_{\langle P\rangle >0}\left({\langle p_{PV}\rangle }_h\right)$$

(4)

where ${\mathbb{I}}_{\langle P\rangle >0}\left({\langle p_{PV}\rangle }_h\right)$ is the indicator function which is equal to 1 when the average generated PV power ${\langle p_{PV}\rangle }_h$ for the given hour h is greater than 0 and 0 elsewhere

$${\mathbb{I}}_{\langle P\rangle >0}\left({\langle p_{PV}\rangle }_h\right)=\left\{\begin{array}{cc}1,\hfill & {\langle p_{PV}\rangle }_h>0\hfill \\ 0,\hfill & {\langle p_{PV}\rangle }_h=0\hfill \end{array}\right.$$

(5)

6.2. Firming Renewables with Storage

A technically straightforward but costly approach is to assign a storage system to an intermittent DER to smooth the intermittency. To illustrate the use of storage for smoothing renewables, Figure 26 shows simulated smoothing applied on an observed case of significant intermittency on the Fronius inverter, where the simulation employed a battery of 3.5 kWh, with the reasonable assumption for discharge of 1C and charge of C/3, where C corresponds to fully charging a battery from the two limiting states of charge within 1 h. Figure 27 shows the block diagram of the simulation. Historical PV data were combined with smart inverter power-limiting functionality and the building storage system via an AC microgrid bus.

The target average hour-by-hour in this simulation was determined to be an unrealistic, best-case scenario (perfect forecast), based on the averaged available power of the same period.

The analysis started with the observed generated PV power. Then, it was averaged by the time interval. ($\Delta t$ = 1 h):

$$p_{target}\left(t\right)={\langle p_{PV}\left(t\right)\rangle }_{\Delta t}$$

(6)

The tentative battery power required for firming $p_{Batt}^{\prime }\left(t\right)$ was computed as the difference between generated PV and the hour-by-hour average.

$$p_{Batt}\left(t\right)=\left\{\begin{array}{cc}-p_{charg{e}_{min}},\hfill & p_B^{\prime }att\left(t\right)<-p_{charg{e}_{min}}\hfill \\ p_B^{\prime }att\left(t\right),\hfill & -p_{charg{e}_{min}}\le p_B^{\prime }att\left(t\right)\le p_{discharg{e}_{max}}\hfill \\ p_{discharg{e}_{max}},\hfill & p_{discharg{e}_{max}}<p_{Batt}^{\prime }\hfill \end{array}\right.$$

(7)

After that, the battery’s tentative state of energy was computed

$$w_{Batt}\left(t\right)=W_{Batt}\left(t_0\right)+{\int }_{t_0}^t{p}_{Batt}\left(\tau \right)d\tau$$

(8)

and battery power was set to zero for energy state below zero $w_{Batt}\left(t\right)<0$ and above total capacity $w_{Batt}\left(t\right)>W_{Bat{t}_{max}}$.

Figure 26c shows PV energy as generated and the energy associated with the firmed PV power alongside the battery energy. Figure 26d shows the intermittency with and without firming, computed by Equation (2).

Note that the power limitations of this battery would not allow this storage to perfectly remove firmed PV (see the orange trace in Figure 26a), leaving some nonzero intermittency (Figure 26d). Moreover, the total firmed energy was higher than the total PV energy in this case because firming was limited by the finite charge power.

As shown before, in Section 4.2.2, power limiting works reasonably well, as illustrated in Figure 14. We ran counterfactual simulations to explore the potential of the power limiting for reducing (1) intermittency and (2) storage requirements. However, setting power limits inevitably reduces overall energy generation, which is the nature of the trade-off: total energy vs. power quality.

Applying the power limits of $p_{P{V}_{max}}$ = 1.8 kW enables perfect firming with the same amount of storage, as shown in Figure 28 (the removed peaks are indicated in grey): after limiting, the firmed power was the same as the target power (Figure 28a) because the battery power was within its charge/discharge limits. Furthermore, the total firmed energy was equal to the total PV energy (Figure 28c), and the intermittency of the firmed power was zero (Figure 28d).

This simple example showed that limiting power on sun-and-cloud days reduces battery requirements for firming renewables. Moreover, the battery energy variations were relatively small, which enables starting with a lower state-of-charge of the battery and potentially charging during PV hours if the demand is low, without the risk of potentially reaching the zero state of charge. The main conclusion was that setting power limits enabled perfect firming of the renewables with limited available storage.

6.3. Firming Renewables without Storage

Limiting power reduces intermittency and improves power quality even in the absence of storage, as shown in Figure 29. However, this improvement reduces total energy. By limiting the peak power to $P_{P{V}_{MAX}}$ = 1.8 kW, the average daily intermittency ${\langle I(\Delta t)\rangle }_{day}$, defined by Equation (4), was reduced from 0.596 down to 0.453 (−23.9%), but this desirable reduction was accompanied by the undesirable decrease in total energy, from 11.4 kW down to 9.3 kW (−18.3%).

Figure 30 shows relative reductions in total energy and intermittency changes (compared to no limits) as the function of power limiting voltage. The two variables change similarly with $P_{P{V}_{MAX}}$, with a more relative intermittency reduction than a total energy decrease. The correlation between the two relative changes was close to unity, and the associated scatter plot was included as an inset in Figure 30.

7. Discussion

The primary purpose of this investigation was to integrate the communications between DERs to enable metering, monitoring, and control using a VPP. In deployment, this can take several forms, all sharing the over-arching requirement of smooth command and control from the field device to the utility’s control room. Enabling this communication with a mix of different equipment types, models, and manufacturers requires an implementation compatible with a set of communication standards that streamline both the messaging format (for example, is it XML or something else) and the semantics.

A variety of protocols have been used to achieve this. Many of them have been around for several years. For example, the utility communications architecture (UCA) has been in use since at least 1999, and fragments of its design are still used for inter-control center communication between neighboring utilities. UCA is difficult to use and not well suited to external (e.g., customer-owned) equipment, so a number of other standards, such as OpenADR, followed UCA, along with some open-source support projects, such as Voltron, which the U.S Department of Energy funded to facilitate wide-scale automatic and semi-automatic demand response scenarios.

The demand response has evolved to enable more flexible control. There is still a requirement for a simple dial to control power (up or down). Still, many more dimensions to this control are envisioned, including the ability to generate power at multiple phase angles and program simple response curves so that DERs can autonomously alter their total real power or reactive power output based on measured parameters following an overall control strategy pushed down from the utility. Of these strategies (behavioral models as well as communication protocols), the most important are California Rule 21 [23], the Smart Energy Profile (SEP2) [36], IEEE 1547 [37], and the protocol that may eventually tie these all together—IEEE 2030.5 [38].

The IEEE 2030.5 protocol is robust and comprehensive, but unfortunately, it is also quite complex. Today, few DERs (solar inverters in particular) have the capability of supporting IEEE 2030.5 directly. To this end, we implemented a lower-level protocol—the “last inches” protocol—called SunSpec. SunSpec is an attempt to standardize communication across DERs (especially inverters), providing a standard communications protocol that can interact with any make or model. This conceptually pleasing approach still requires considerable effort to integrate different hardware components.

The system developed as part of this project was intended to behave similarly to a small subset of IEEE 2030.5 to exercise communication to the lowest level of the control pyramid—the field devices located on customer premises. SunSpec appeared to be an approach allowing software development to talk to all inverters and treat them as nearly identical resources. This empirical study identified two main difficulties:

1.

SunSpec devices are only required to implement device identification. All other models are optional. This research determined that while one of the subject inverters was marketed as having SunSpec support, it does not have any support for autonomous (e.g., California Rule 21 curve) control. This issue may be less significant as IEEE 1547-2018 certified inverters begin to proliferate for new DG sites;

2.

Device manufacturers can decide how to implement SunSpec differently, even for devices that connect to identical service points. For example, two inverters tested—Fronius Primo 3.8 and SMA SunnyBoy 3.0—are designed to attach to a single phase 240V service point with a neutral. However, the Fronius is described as a “split phase” (which means “single phase plus neutral”) device. In contrast, the SMA is described as a “multi-phase device” with two 120V inputs that are 180 degrees out of phase. The Fronius is described as a 240 V device, while the SMA is described as a 120 V device with two input channels.

A conclusion that can be drawn from this is that communication with field devices will require specialized knowledge of the “rules” of that particular manufacturer, model, and perhaps even firmware revision and configuration of that device. In an environment where equipment is customer-owned and customer-sited, this will be challenging.

Attempts to run these tests over a mockup AMI network (likely with better performance than an existing AMI network) identify problems. The backend needs its correct message interpretation to execute command and control. However, correct message interpretation requires a thorough and periodic scan of the software, registers, and configuration of the field device to identify the interpretation of responses and determine whether the software or configuration has changed due to some user action outside the purview of the utility.

These challenges underscore the necessity for follow-on work to determine the edge device’s requirements, capability, autonomy, and sophistication.

8. Conclusions

The low-cost interface based on the commercial bridge device and AMI was implemented and tested. The new solution was an order of magnitude less expensive than the traditional RTU-based solutions. The installed cost of the low-cost hardware was less than $500, with the possibility of becoming less than $200, whereas the installed cost of RTUs was in the $8000–10,000 range.

The interface was reliable for messages that did not require frequent updates. Performance testing found that the RTTover the end-to-end communication link,

$$\begin{array}{c}\hfill \mathrm{Inverter}\leftrightarrow \mathrm{ConnectDER}\mathrm{bridge}\leftrightarrow \mathrm{Itron}\mathrm{IOT}\mathrm{Router}\leftrightarrow \mathrm{Itron}\mathrm{Starfish}\leftrightarrow \mathrm{VPP},\end{array}$$

was 2.52 s on average, with 99% of RTTs less than 4.25 s, and occasional outliers of as high as 20 s. Overall, this performance is adequate for metering, but an edge device with more computational power is recommended to enable the full power of the information-rich IEEE 2030.5 standard. The inverters were organized according to the hierarchical schema specified by IEEE 2030.5. The standard is very flexible but based on the rich XML, which favors more computational power on edge over very lean controllers. The AMI network bandwidth was too narrow for full IEEE 2030.5 implementation.

SunSpec models alone were insufficient to demonstrate static Volt-VAr curve optimization. Some IEEE 1547 capabilities are not readily accessible. For example, the SMA inverter required the installer’s password to enable curve implementation. Furthermore, the knowledge of inverter configuration (i.e., two-phase voltages of the split phase or their difference) and inverter interpretation of the connection must be incorporated. Once in place, the optimization performance was excellent.

Controls enabled by a smart-meter-and-smart-inverter interface provide additional flexibility in improving power quality by voltage support functionality, enabling generation disaggregation from the load, group management, and site performance and status monitoring. The simulations on the historical PV data demonstrated that the power-limiting functionality of smart inverters was able to reduce intermittency. With the smart inverter’s power-limiting functionality, DER’s day-ahead, hour-by-hour power generation forecast required less storage.