Today there are many sensors in buildings and district heating/cooling systems with a high temporal resolution. Data from such sensors opens up for new AI and IoT-based solutions. Here we will describe the potential of some of such solutions for district heating systems. However, we will also touch upon how such solutions potentially make the system more vulnerable and challenging with respect to privacy and GDPR.
By Henrik Madsen, Professor and Section Head on DTU Compute
Published in Hot Cool, edition no. 1/2023 | ISSN 0904 9681 |
It will be argued that energy meter data and the use of AI in district heating provide the foundation for efficiency improvements in buildings and district heating systems. In addition, such data-driven methods give possibilities for CO2 and cost savings, better integration of wind and solar power, efficient integration of the energy systems, and more satisfied end-users due to lower costs and a better indoor climate.
One of the major problems today is that data and solutions often are linked to proprietary platforms. Consequently, it is
challenging to implement cross-system solutions and harvest synergies from systems integration. Sadly, this often hinders the possibility of obtaining large savings and efficient imple mentations.
However, this cross-system functionality can be obtained using a non-profit data hub, like the national hub for smart energy and water systems at Center Denmark. For instance, Center Denmark is successfully used for cross-system optimization in the HEAT 4.0 project (see HOTCOOL no. 8, 2022).
In the following we will start considering AI tools for individual buildings. Then we will consider the district heating networks, the plants, and conclude with remarks on district heating in relation to the energy system, the electricity/energy markets, and the society. The findings mentioned here are based on several district heating-related projects (CITIES, HEAT 4.0, FED, IDASC, ARV – please see the reference list).
Buildings and occupants
The demand for smartness, trust, transparency, and versatility in managing heating, cooling, ventilation, lighting, and access control systems for a family home, public buildings like schools, and office buildings is growing. People want a comfortable, sustainable, cost-efficient, and safe place to live and work, and that’s where sensors, AI, IoT, and automation jump in.
GDPR and privacy
Personal data are any information related to an identified or identifiable person. Only if the processing of data concerns personal data, the GDPR (General Data Protection Regulation) applies. This implies that the problem typically does not exist e.g., for a section of a district heating network, as well as for public buildings like schools and some office buildings.
Today we are able to obtain electricity consumption data with a very high temporal resolution (e.g., every 15 seconds) also for single-family buildings. Given such data, it is often relatively easy to conclude which appliance has been used at which time.
Likewise, most of us have hourly readings of water consumption online. We can see, for instance, that 2 liters have been used at 2 am, 3 am, and 5 am during a single night. This indicates frequent bathroom visits, which could lead to some privacy issues.
Frequent and real-time data for district heating consumption can be used to better control heating and ventilation systems, but also to detect absence (e.g., holiday periods). Obviously, such data can be used to identify if the build
ing calls, e.g., for a renovation like replacing the windows.
Energy performance characterization of buildings
Traditionally, energy performance characterization and labeling have mostly been based on deductive analysis, i.e., based on assumed theory for energy transfer and material properties. Today the existence of frequent meter readings and, e.g., nearby meteorological observations data opens up for evidence-based inductive analysis, i.e., data-driven methods.
Data drive TO, Tingbjerg, Copenhagen
The deductive approach used for buildings today
Today the energy performance characterization and energy labeling of buildings are based on rather simple calculations and a visit by an energy consultant. The cost of getting such a label is relatively high, around 700-1000 Euros. The methods used today are often criticized. The main problem is that two buildings, which in theory should be identical, might have a somewhat different energy performances in practice.
This well-known performance gap between predicted and actual building energy performance can be significant. Even after correcting for differences in user behavior and occupancy, the actual energy consumption can easily be 50-100 pct higher than the theoretical consumption.
Generally, the technical sources for discrepancies between the theoretical performance and the measured performance can be broken into three baskets: The design and simulation phase (limitations, inaccuracies, and assumptions in the theory used to predict the performance); the construction and commissioning phase (caused by the poor quality of workmanship and differences between assumed and actual materials, components and systems); and, the operation phase (poor-functioning of the systems and in particular the HVAC system).
Digital x-ray-based performance characterization
Frequent energy meter data opens up for new inductive or data-driven tools. The tools act as a bit like a kind of x-ray vision through the layers of the individual walls. Given this x-ray based knowledge of the performance of the individual walls, the tool can provide evidence-based information about the performance of the separate buildings. This is useful, for instance, before deciding on a possible energy renovation.
Similar AI tools and digital twin models can be used to obtain a better control of the indoor climate. This has been demonstrated, e.g., in the social housing Taastrupgaard in Høje-Taastrup Municipality, where digital tools have been used to show that many radiators were misused.
Sensors for the indoor CO2 levels, temperature, and humidity have also been installed in schools, e.g., in Hørsholm and Rudersdal, to monitor the indoor climate and obtain a better comfort and learning environment in the schools using the platform from Climify. Climify has also developed a FeedMe app, which can ensure individual and optimized control of heating and ventilation of the classrooms,lowering the return temperature, minimizing mold risk, and obtaining energy savings.
Forecasting
Load forecasting obviously calls for data-driven tools. Lately, new methods for coherent forecasting of the heating load on all relevant time horizons from, say, 15 minutes to 96 hours ahead have proven to give considerable (15 – 30 pct) improvements in the accuracy of load forecasts at some of the largest DH operators in Denmark. These forecasting improvements lead to significant economic benefits in temperature control, production planning, and participation in the electricity markets.
Forecasting of PV and thermal solar energy production also calls for data-driven approaches for several reasons. By using data-driven methods, the forecasting tool can automatically consider complex shading and the time-varying dirtiness of the panels. Some of the new forecasting methods are implemented in, e.g., HeatFor and SolarFor.
Figure 1: Differences between simulation-based and data-driven temperature optimization
Temperature optimization (TO)
Historically, methods for temperature optimization have been based on simulations using theoretical models and detailed knowledge about the network. A prerequisite for using such approaches is that the model is carefully updated with information about the physics (pipes, ground temperature, the humidity of the soil, properties of the insulation of the pipes, etc.).
First of all, this is a very time-consuming procedure, and secondly, such methods lead to suboptimal descriptions of the dynamical characteristics needed for control of the temperatures. Exactly like for the buildings mentioned above, data-driven
methods can provide significant improvements in temperature optimization, such as in zonal control of the network temperature.
Again the AI technologies implemented, for instance, in HeatTO, give a sort of x-ray vision of the thermal properties of the pipes and their surroundings. The resulting data-driven digital twin models describe the time delay, heat losses, and dynamics. According to the experiences with HeatTO, heat loss is reduced by 10 to 20 pct (see, e.g., https://enfor.dk/services/heatto/).
Access to energy meter readings from individual households has proven to give further advantages. A simple sketch of the change of the setting is shown in Figure 2. Obviously, this calls for using advanced aggregation techniques to ensure that the aggregated temperature is representative, and to respect privacy and GDPR.
The use of meter data implies that it is rather easy to operate with zonal temperatures. This new solution for using meter data and the methods for zonal temperature control leads to further savings and better options for integrating local heat pumps.
Today electricity prices are high from time to time, so the controllers have a built-in balance between reduction in heat losses and the pumping costs. Finally, it is crucial to notice that the use of data and AI methods implies that the tools are auto-calibrated continuously. This means that the system is much easier to operate and maintain.
Figure 2 – Use of meter data in temperature optimization (HeatTO)
Production and bidding optimization for DH systems
Data-driven forecasting and temperature control methods are now also used in new tools developed at DTU for production optimization in DH systems. The tools can be used for different planning problems, such as operational planning under uncertainty, optimization of bids to the day-ahead electricity market, and long-term evaluations of DH system operations. The tools are able to take advantage of the uncertainty, for instance, in the production of thermal solar heat as well as forecasts of the electricity prices on markets with varying horizons.
The general applicability and performance of the approach are evaluated based on real data from the three Danish DH systems of Brønderslev, Hillerød, and Middelfart with different characteristics. When considering bidding, the new tool reduces cost in all cases and can save up to 42.1%.
Conclusion
Development in sensor technology and the rapid development in AI and IoT have provided district heating operators with new opportunities. Using AI or data-driven models to provide information from sensors, the operations in the building, at the plants, the network, and market participation can be optimized.
The key is data-driven and auto-calibrated tools for the modern operator. Tools for coherent load forecasting are central. Knowing the future demand with reliable uncertainty intervals allows for setting the water temperature and flow optimally rather than operating with a large-than-necessary safety margin. Such state-of-the-art forecasts are also the prerequisite for smooth solutions for bidding on the electricity markets.
For the future, weather-driven society district heating is already recognized to play a central role since these systems can provide much of the needed flexibility at a low cost. Digitalization of district heating systems based on sensor data will further
strengthen the position of district heating as a sustainable and low-cost energy supply technology capable of reducing carbon emissions and contributing to climate change mitigation.
In addition, we have proven that using data-driven tools has a huge economic potential. According to the so-called Damvad Report from 2019, the potential in Denmark alone is 240 to 790 mill DKK annually with state-of-the-art data-driven methods for temperature optimization. On top of that, most of the methods for digitalization mentioned in this article will lead to considerable extra economic and operational benefits for district heating systems and their users.
For further information please contact: Henrik Madsen, hmad.dtu@gmail.com
Reference to projects
IDASC: https://issuu.com/dtudk/docs/district-heatingdigitalized?fr=sM2FiMzQ4NjgwMg
HEAT 4.0: https://dbdh.dk/wp-content/uploads/2021/04/HEAT40-AlfredHeller-Niras.pdf
CITIES: https://smart-cities-centre.org/
Flexible Energy Denmark: https://www.flexibleenergydenmark.com/
ARV: https://greendeal-arv.eu/
“Artificial intelligence in district heating” was published in Hot Cool, edition no. 1/2023
