ICT Project Field Report on the Development of a Local Smart Micro Grid in the Housing Industry

Abstract

As part of our research project “WINNER Reloaded,” we embarked on a three-year journey to oversee the establishment of a local micro smart grid in a residential community in the city of Chemnitz. Throughout this endeavor, we gained practical insights into the planning, implementation, and administration of the systems, acquiring a large amount of real-world energy data on the way. In the following chapter, we delve into the intricacies of our research project, offering a closer examination of its details and the wisdom we accrued through our experiences. Our intention is to provide a pragmatic guide that will facilitate others in embarking on the creation of their own local micro smart grid through judicious foresight and planning.

1. Introduction

A local micro smart grid is a small-scale, intelligently managed energy distribution network designed to efficiently distribute electricity within a closed, specific local area, such as a residential community or a neighborhood. The objective of the Local Smart Micro Grid (LMSG) is to ensure electric power supply and optimize the supply and distribution of electrical energy within the local grid. Therefore, it incorporates local renewable energy sources like solar panels and often integrates energy storage systems. The key features to optimize operation include real-time monitoring and control of energy production and consumption and enabling dynamic adjustments to match supply with demand. Micro smart grids promote sustainable energy use, reduce grid stress, and enhance reliability, making them a valuable component of modern, reliable, and environmentally friendly power grids.

The benefits of a local micro smart grid come with significant planning, implementation, and operation efforts. Starting with defining requirements and developing an architecture, suitable technical components must be chosen. The focus should be not solely on meeting technical requirements but also on quality attributes like interoperability and the ability to integrate into a heterogeneous system. Building a complex system as an LMSG cannot be accomplished by a single party; among others, it involves partners with backgrounds in housing management, the electrical industry, building administration, and software development. Weighing mutual interests and continuous consensus-building is essential. Lastly, the diverse components require ongoing maintenance and support beyond the initial setup, which should also be considered.

During our three-year research project “WINNER Reloaded” [1], we participated in setting up a local micro smart grid. This endeavor allowed us to shape the direction of this system and accumulate a substantial corpus of real-world energy data, which we systematically analyzed throughout the project’s duration.

In this comprehensive chapter, we share some practical experiences we gained and look into the dataset we accumulated. We aim to offer insights that go beyond the technical details of the LMSG itself.

The structure of our chapter is as follows: We start with an overview of our research project, highlighting its key objectives. Then, we delve into the technical and system architecture that underpins our LMSG, explaining how the system works in detail. In the next section, we explore the practical aspects of our project. Here, we share our accumulated experiences regarding the setup and development of system components and project administration and provide insights into the energy data we acquired throughout the project. Finally, we summarize the insights we gained and draw conclusions. Through this chapter, we hope to inspire new innovation and understanding in the realm of sustainable energy solutions.

2. Our research project “WINNER reloaded”

The research project “WINNER Reloaded” is situated within the context of promoting electromobility and establishing tenant electricity models. The project aims to realize demand-oriented charging infrastructure, tenant electricity supply, and decentralized energy generation for tenants, businesses, and other third-party users. Companies from the housing industry, the energy industry, and system installers are working together on this.

The main demonstrator, located within a new construction project in the city of Chemnitz/Germany, is at the heart of the research project. This complex consists of four residential buildings with 40 residential units, carports, and underground parking spaces. The parking spaces are equipped with charging infrastructure according to the tenants’ needs. Notable features are a public high-performance charging station and the integration of a car-sharing station with electric vehicles as an additional mobility offer in the neighborhood.

Power supply for the tenants and the private and public charging infrastructure is sourced from solar power generated on-site using photovoltaic systems. To ensure the power supply, the residential complex has its dedicated network station connected to the local medium-voltage electricity grid. The project employs an intelligent micro smart grid, a small-scale dynamic power supply network.

The University of Jena manages the decentralized data collection in the local micro smart grid, which serves for load and charging management and forms the basis for electricity billing. Particular emphasis is placed on a high temporal resolution of the collected consumption data. The data collected this way is also made available to tenants for individual real-time monitoring. Within this system and with the data, we can also calculate and simulate dynamic pricing and billing models based on factors such as time of use and internally generated energy.

3. Architecture of our smart grid

The architecture of our smart grid has been based upon the requirements previously worked out in [2, 3]. It consists of a multitude of technical components that need to be independently integrated into the network and individually read and controlled. These components include multiple electricity meters, programmable logic controllers (PLCs) for capturing relevant energy flow parameters within the neighborhood, PV inverters, and electric vehicle charging stations. All components are connected via an internal local area network. A server within the internal network is available for data processing and control. An externally accessible system called the “Tenant Information Portal” exists for visualizing consumption data for the tenants.

An initial challenge in connecting such diverse components is achieving a unified integration into the overall infrastructure. Depending on the device and manufacturer, the methods for data retrieval differ significantly. In the case of our inverters and charging stations, data collection was achieved through REST requests; however, this method was not documented for either device and had to be determined through reverse engineering of their user interfaces. In contrast, the PLCs came with preinstalled software that inherently allowed data transmission at defined intervals via the Message Queuing Telemetry Transport (MQTT) protocol, although in a format predetermined by the manufacturer. To date, there has been no generally accepted IoT protocol-based standard to transmit energy measurements.

An important step in establishing a uniform infrastructure therefore is integrating heterogeneous components and selecting a common protocol and data format for the exchange of measurement and control data. In our case, we chose the MQTT protocol, designed for communication between Internet of Things (IoT) devices, through which measurement data was transmitted in an internally specified JSON-based format. To ensure data privacy and increase fault tolerance, the MQTT broker was installed on a local server within the LMSG. In this way, the raw data does not have to physically leave the local network infrastructure before it is processed further.

Data retrieval had to be performed individually for each component, necessitating the development of specialized software, also configured as services on the server. In total, there are three such components: two for data retrieval from the inverters and charging stations and one for processing and converting the measurement data received from the PLCs. Additionally, a customized version of the open-source energy management system OpenEMS [4] was deployed, serving for system status visualization and dynamic configuration. An overview of the software architecture is provided in Figure 1.

Figure 1.

An overview of the components in our local micro smart grid.

4. Purposes of the measurements and selected data points and sampling rates

In our local smart grid system, data is collected with two objectives. These objectives are the system control and the use for proof and transparency, among other things, for billing and plausibility purposes. From these two objectives, different requirements for processing location, sampling rate, and data storage are derived.

Awareness of the technical requirements for the data acquisition process developed iteratively. Initially, we collected the energy meter data only. The energy meter continuously records the active energy consumed, in the case of a household, or provided, in the case of a generation plant. The recording was done in the three-phase power supply as one total meter value across all phases.

With the awareness that the charging behavior of battery electric vehicles varies significantly with regard to the use of the individual electrical phases and that the overall local power system can be negatively influenced by unbalanced loads, we expanded the recording to include phase-accurate measured values.

Although energy and power values can be mathematically converted into each other, this conversion in real measuring systems is subject to narrow limits due to the sampling rate and data accuracy. We have therefore also gathered the instantaneous active power measured by the energy meters.

Finally, we became aware that in some constellations, the reduction of charging power by battery electric vehicles is accompanied by an increase of reactive power. In detail, this effect depends on the vehicle model and is not visible in active power measurements. Therefore, we extended the data acquisition by the phase-exact measured values of the reactive power.

The listed measurements were collected per building for tenant households and general home electricity, as well as for public and nonpublic charging infrastructure and generation facilities.

In addition to the measured values of the power grid, we directly connected to the charging infrastructure and the photovoltaic inverters and read out internal measured values and control parameters. As described, these systems do not provide an explicit software interface for integration. We created tools to access the required data via the existing web-based user interface. The creation of the tools was costly, and the implementation itself was always compromised by changes in the user interface due to software updates.

All measured values were autonomously and continuously determined by the measuring instruments or their independently running programs for interrogation. The sampling rate was uniformly set to 1 sample per 10 seconds for all measured values.

This temporal resolution appears to be very high and unnecessary for some physical quantities and use cases. On the other hand, short-term loads and fluctuations can only be seen with this high temporal resolution. In the context of the research project, there was no reason to reduce the sampling rate. We were not limited by processing speed, bandwidth, or memory in our quite small setup. For an optimized productive commercial system setup, the quantity and resolution of the measured values could be discussed again individually for each measurement.

As already discussed, after data acquisition, the measured values were converted into a uniform data format and made available to interested internal and external system components by means of a message middleware. Regardless of this, we stored all the data collected in an external time series database for later review and statistical analysis.

5. Selected data analyses

This section looks at some data analyses to underpin statements made previously. The selected analyses consider three different perspectives of the overall system that build on one another. The first is to analyze the distribution and predictability of generation and consumption. This analysis results in the degrees of freedom through load shifting for operational management. However, an effective operational management requires valid measured values. So, the second analysis highlights possible effects due to different temporal sampling rates and aggregation periods, accordingly. The necessity of phase-accurate measurement is illustrated in a third analysis using the example of the different charging behavior of vehicles.

To study the degrees of self-sufficiency and the degrees of freedom of load shifting, we analyze local production and consumption. In our two demonstration sites, we have two types of power generation sources, photovoltaics and combined heat and power generation. The combined heat and power plants are operated on a heat-led basis, with electricity generation being “only” a positive side effect. Since the cogeneration units in the project are not operated in the residential structure, we consider only the photovoltaic system in the following.

The energy generation of the photovoltaic system is highly dynamic and dependent on external influencing parameters. The generation output during the course of the day, with very low disturbing influences, for example by clouds, fog, or snow, is shown in Figure 2. The photovoltaic systems in our system are mounted on four houses and vary in the number of individual PV modules and orientation. For better comparability, the energy generated is therefore standardized to the nominal generation capacity (kWp). To illustrate the seasonal influence, power curves from winter (January), spring (April), and summer (June) are shown. The seasonal influence with regard to maximum output and the start and end of generation according to the daily sunshine pattern is clearly recognizable. The shift of the generation period due to the different daylight times between summer and winter amounts to a maximum of approximately 8 hours.

Figure 2.

Daily pattern of generation power on three exemplary days in summer, spring, and winter.

The seasonal impact on the generation performance can also be seen very well in the overview of the annual course of the energy generated per day as shown in Figure 3. Compared to the peak values in summer (June, July), the plant at this location and in this fixed setup in winter (December, January) produces only approximately 5–10% of the energy.

Figure 3.

Energy production in kWh per installed generation capacity (kWp) on a daily average per month of the four photovoltaic systems in the residential system design in the project.

Systematically, generation output of the photovoltaic system depends on unchangeable technical parameters as well as seasonal and weather influences.

The most important technical parameters are in particular

• the specific type of photovoltaic modules used,

• the geographical location of the system,

• the number and orientation of the photovoltaic modules in the specific system, and

• the course of the horizon, which affects the partial or complete shading of individual modules or the entire system.

The most important seasonal and weather-related parameters are

• position of the earth in relation to the sun,

• course of the sun’s position throughout the day,

• degree and type of cloud cover throughout the day,

• rain or snowfall, both throughout the day and in the previous days,

• ambient temperature.

Based on these fixed technical parameters and the dynamic seasonal and weather-dependent parameters, the expected generation capacity for the next day can be predicted. For this purpose, there are modeling approaches that attempt to represent the system completely as a physical system. There are also forecasting models that are based on statistical methods or machine learning methods.

We experimented with a hybrid model in our project. We described the photovoltaic system physically and used this model to calculate the theoretical generation output without interfering weather influences. Based on these theoretical maximum values, we used external weather forecasts and statistical models to derive a day ahead forecast that takes the weather into account.

Whether purely physical, statistical, or hybrid, all forecasting models can provide sufficiently accurate planning values that can be used in a day-ahead planning. The selection of the appropriate model depends heavily on the available initial data. If no constructive technical information is available, physical modeling will not be successful. The same applies to statistical models if no real measured values are available. It should be noted that regardless of the quality of the forecast model, intraday real-time control is always absolutely essential, as the actual generation and consumption output always deviates from the planned values.

Unlike PV-based generation, energy use by tenants is less easy to model and predict. For this task, primarily only statistical forecast methods are available. Total energy consumption can be disaggregated into different energy consumption categories. Some main categories are

• Household operations, entertainment, and light,

• Preparation and provision of hot water,

• Provision of heating and air conditioning, and

• Battery electric vehicle (BEV) charging.

Alongside the energy consumed by the residents, there is also the energy required for the general operation of the building infrastructure, for instance, house lighting, emergency lighting, elevators, and ventilation. Each of these energy consumption categories has its own individual characteristic consumption time profile. The energy consumption for an apartment is most strongly influenced by the question of whether this apartment is currently occupied. This aspect seems trivial at first glance, but it is not only determined by the general occupancy status of the apartment, which could be retrieved from an external management system. In particular, this factor must be taken into account during typical vacation periods and holidays, when apartments are regularly unoccupied.

If the apartment is occupied, the energy consumption in the household is strongly influenced by the classification of the day. The days can be divided into the classes “Weekday,” “Saturday,” and “Sunday/Holiday”. The season naturally influences energy consumption in the household, as well. The average energy profiles for households measured in the project are shown in Figure 4.

Figure 4.

Daily pattern of average electrical power drawn by households dependening on day of the week and season.

Figure 5 shows that the time of year also has an influence on the energy consumption of households. In the winter months, energy consumption by households is visibly higher than in the summer months. The high energy consumption in December is remarkable. This is due to the Christian holidays and the associated traditions of increased baking and cooking. The energy consumption in summer, on the other hand, is noticeably lower. This is partly due to the longer daylight hours and therefore less need for artificial light and more activities outside the household, as well as the temporary absence of residents during their summer vacation.

Figure 5.

Average electrical power consumption by households on weekday and weekend.

As previously mentioned, it is absolutely essential to measure both the curves of the energy meter and the instantaneous power.

Depending on their usage, the energy meters only provide measured values of one order of magnitude—kWh. With a sampling rate of 1 measured value per 10 seconds, this order of magnitude allows very high power levels to be measured with sufficient accuracy. These are, for example, charging processes for electric vehicles. However, the typical standby power of the charging infrastructure is 1500 times lower and cannot be determined with sufficient accuracy by calculating the difference between two consecutive meter values of the energy meters (maximum charging power approx. 150 kW, standby power approx. 100 W). This finding can be clearly seen in Figure 6 in the time window between 12:30 and 12:45. The curve based on 15 min measured values shows no value. The curve based on the difference of 30 sec values shows a zero curve with individual spikes. These spikes result from the “jumping” of the read-out measured value according to the smallest possible interval.

Figure 6.

Electrical power used by high-performance charging.

One disadvantage of recording the amount of energy per unit of time compared to recording meter values for the total energy is the additional effort required to map alternative time series. For example, if an amount of energy per 5 min is determined and stored, the determination of time series with resolutions of 1 min, 60 min, or 24 h is associated with additional effort.

The aggregated recording of measured values across all phases, especially power values, has already been mentioned as unfavorable. Phase-specific measurement and, if possible, control proved to be more sensible.

Figure 7 shows the difference between the aggregated power measurement and the phase-accurate power measurement. The figure shows the three different charging processes as they were actually measured. The first charging process is a three-phase charging process with a total power of 11 kW. Accordingly, all phases L1, L2, and L3 are utilized almost equally. The second charging process is only a two-phase process with a total output of 7.2 kW, whereby phase L3 is not utilized. The third charging process only utilizes phase L1 and has a charging power of 3.6 kW.

Figure 7.

Electrical power used by standard charging in different charging modes.

However, in our setup, the control of the photovoltaic inverters and the charging infrastructure made it impossible to react to load asymmetries and take appropriate countermeasures. Future systems have to pay more attention to the necessity of balancing asymmetrical loads in the local grid and must include appropriate technical options for counteracting them.

6. Our experiences in setting up and operating a local micro smart grid

During the setup phase, a significant hurdle encountered was the integration of individual hardware components, as it demanded substantial analytical efforts and subsequent software development time for automated data access. Similarly, the operation of charging stations for the purpose of dynamic power regulation proved difficult due to the lack of a unified interface and a general high inertia of the system. At this point, we became aware that components readily available on the market may not necessarily be well-suited for integration into a smart grid setup, as off-the-shelf products do not necessarily provide robust features for automatic data access or dynamic control.

The selection of suitable components for a local micro smart grid necessitates a meticulous definition of prerequisites beforehand, which should be thoroughly scrutinized before procurement, if necessary in consultation with the manufacturer. To circumvent additional work, we strongly advise that this step be undertaken with great precision during the planning phase and that requirements be harmonized with the project partners responsible for provisioning (typically possessing extensive experience in the construction of conventional systems).

During the operation of the system, additional challenges related to data measurement emerged. Notably, the need for configuring a central time synchronization service became apparent to correlate timestamps of measurement data, thereby compensating for and mitigating potentially significant discrepancies that arose over extended periods, varying by device type. This measure was essential to avert potential control errors stemming from these time disparities.

As previously described, we realized the need to record phase individual measurement values rather than examining them solely in aggregates. This is of particular importance since high power applications, like charging EV or cooking, can lead to highly asymmetrical loads on individual electrical phases. Such imbalances between the phases are not recognizable in aggregates, but have to be identified. Asymmetrical loads lead to a higher load on the neutral conductor, for which it may not be designed. Consequently, it is advisable to design the data model and processing infrastructure during the planning phase with consideration for this larger volume of data. Retroactive adjustments can cause considerable work and expense.

Regarding the issue of unbalanced loads caused by electric vehicles, this is recognized in the electrical industry and various strategies have been developed to solve it. Single-phase charging electric vehicles tend to predominantly draw power from pin 1, while two-phase charging vehicles typically draw power from pin 1 and 2.

One pragmatic approach is therefore to rotate phases over several subsequent charging points. In this way, pin 1 of the charging cable is connected to phase L1 at the first charging point, to phase L2 at the next charging point, and to phase L3 at the third charging point. This straightforward approach enables a more even distribution of loads, even if only single-phase charging vehicles are used. However, this wiring method introduces the disadvantage that measured values within the measurement installation can no longer be easily attributed to the correct phase or pin. This is because the “phase L1” recorded at a charging point, in reality, could be connected to phase L2 or phase L3 within the overarching network. These intricacies must be clarified with the partner responsible for the electrical installation to prevent erroneous measurements and subsequent incorrect control operations. Beyond the case of the charging infrastructure, the consistent wiring of the electrical phase is generally an extremely important factor. Swapping or rotating the three phases within the distributed electrical system will lead to incorrect measured values and subsequently to implausible control behavior. Finding this cause is extremely difficult and time-consuming.

Finally, we would like to emphasize that a local micro smart grid is a complex system comprising numerous interdependent hardware and software components that require regular maintenance. Component failures can occur at any time and should be promptly identified and addressed. Examples of problems we encountered during the project include:

• Crashes of software due to various causes (e.g., lack of memory).

• Solar inverter hardware failures and the resulting loss of PV energy.

• Issues in the hardware of the local communication infrastructure and a resulting loss of interconnectivity and subsequently measurement data.

• Non-static system configurations caused changes, rendering devices unreachable for data exchange; for instance, queried inverters changed IP address following a reboot.

• Setup changes caused by non-static system configurations which renders devices unreachable for data exchange (e.g., queried inverters IP addresses changed following a reboot).

The variety of these errors should make it clear that LMSGs are not self-perpetuating and that appropriate monitoring of the operation should be factored in during the entire operation, beyond the initial planning and building phase. We would definitely recommend designating a special system manager for this purpose.

The large number of fallacies that we encountered in the course of the project prompted us to systematically record these in a separate article and address them in more detail [5].

7. Conclusion

The “WINNER Reloaded” project afforded us the opportunity to construct a local micro smart grid from the ground up and, in the process, accumulate a substantial amount of new experiences and data. Over the course of 3 years, we gained a profound insight into the multifaceted practical aspects of such a system. From these experiences, we aim to emphasize the divergence from conventional IT systems and emphasize the importance of proactive planning and competent administration.

Moreover, we aspire that our data analyses have provided a glimpse into load behavior across different times and situations, potentially bearing relevance for future local micro smart grid planning. We hope that our chapter will contribute to paving the way for a sustainable future.

Acknowledgments

We would like to thank all project partners of the “WINNER Reloaded” project. Without their visions, work on the project, and technical expertise in its implementation, the project and this report would not have been possible.

Conflict of interest

The authors declare no conflict of interest.

Thanks

The research project “WINNER Reloaded” was funded by the Federal Ministry for Economic Affairs and Energy of Germany under project number 01ME19001E.