Real-time optimization in district heating: Ulm relies on a physics-based real-time Digital Twin

The article, written by M.Eng. Patrick Ruf, Head of District Heating / District Cooling, Fernwärme Ulm GmbH and Dr.-Ing. Quirin Aumann, Solutions Lead, Gradyent GmbH, was first published in the Aqua & Gas Magazine in April 2026. It has been translated from German to English by Gradyent.

To manage the growing complexity of its heating network, Fernwärme Ulm GmbH (FUG) is implementing a physics-based Digital Twin for the first time. Together with Gradyent, a real-time model is being developed that continuously simulates, analyses, and optimizes the F2 network. The approach combines hydraulic transparency, temperature-driven operational optimization, and data-driven customer monitoring to achieve a new level of efficient district heating supply.

Introducing a new operational reality for district heating

District heating is undergoing a profound transformation across Europe. Decarbonization, expanding generation portfolios, increasing cost pressure, and growing expectations for transparency are pushing established tools and operational routines to their limits. In Ulm, the gradual replacement of the steam network with hot water networks, the expansion of renewable generation sources, including biomass and waste incineration, and the construction of a large thermal storage facility are creating new operational challenges.

To manage this growing complexity while unlocking flexibility potential, Fernwärme Ulm GmbH (FUG), together with Gradyent, is optimizing its network using a real-time Digital Twin, which has been under development since 2023 and provides a physics-based representation of Ulm’s F2 network.

The Gradyent Digital Twin: Principle and optimization approach

Gradyent’s Digital Twin is a mathematical-physical real-time model that describes the entire district heating chain end-to-end, from generation and transmission network to substations and end customers. The foundation is a pipe-by-pipe thermo-hydraulic model that captures energy, mass, and momentum balances as well as pressure and heat losses, mixing processes, inertia effects, and storage behavior.

Measurements from the control room and network sensors are ingested at minute-level intervals, fed into the model, and used for forecasting. In addition, machine learning methods identify patterns, detect deviations, and improve prediction accuracy for the coming hours and days. This creates a continuously updated representation of the real system that constantly calculates optimal operating states, for example for supply temperature and flow rates.

This approach differs fundamentally from traditional static calculations. The Digital Twin optimizes end-to-end: instead of focusing on individual assets, district heating is treated as a coupled overall system. This enables supply and return temperatures, pump strategies, and storage operation to be selected in a way that reduces heat losses, avoids hydraulic bottlenecks, improves generation efficiency, and unlocks additional network capacity.

The solution is operated entirely cloud-based. This ensures reliable availability of high computing power for minute-by-minute re-optimization without modifying the existing control room technology. Where security policies require strict separation, integration is implemented via decoupled data paths: the Digital Twin provides setpoints and diagnostics outside the control room. The operator actively retrieves these and feeds them into the control system. The IT security architecture remains unchanged, while the operational benefits are fully realized.

From analysis tool to operational partner: Starting point in Ulm

Fernwärme Ulm GmbH (FUG) operates several district heating networks in Ulm. The project described here concerns the F2 network, which supplies the city center directly from the generation park, including CHP plants with biomass. The objective was to establish and operationalize three key components: continuous monitoring with precise system insight at any time, optimization of supply temperatures to reduce heat losses, and customer monitoring to identify users with elevated return temperatures, engage with them, and prioritize corrective measures.

The starting point was a specific hydraulic issue during the transition from steam to hot water operation: temperature distributions, pressure zones, and flow patterns did not match expectations. Using static tools, the root cause could not be satisfactorily identified. The first digital analysis of the F2 network created transparency and revealed hydraulic bottlenecks—some in locations different from those initially assumed. Based on these insights, the Digital Twin was established as an operational tool.

Data availability, feed-in topology, and network modeling

Data availability on the FUG side was very strong. For all relevant generation sensors, minute-level time series were available and also provided in real-time operation. Based on this, the Digital Twin continuously calculates the network and feed-in state and provides setpoints for the supply temperature at the feed-in points.

Controlling the supply temperature is complex, as it is directly determined by the generation temperature, with four assets feeding into the supply line simultaneously. Therefore, the feed-in topology was modeled in detail to accurately represent the mixing ratios. Uncertainties caused by missing sensors in certain components were reduced through close collaboration between Gradyent and FUG, validation using existing measurement points, and integration into the network dynamics. This made it possible to resolve inconsistencies between sensor readings and P&ID diagrams and to accurately reflect real system operation.

In addition, customer data is available via different gateways and at varying frequencies—sometimes hourly or aggregated at the end of the day. The model and its underlying infrastructure are designed to robustly handle these heterogeneous data intervals while still providing stable real-time optimization.

IT security and control room integration

FUG follows a restrictive security strategy: no inbound communication from the internet into the control room is permitted. To comply with this requirement, a decoupled intermediate layer was implemented. The Digital Twin stores its setpoints exclusively on an FUG server. From there, they are actively retrieved and processed within the control system. Gradyent has no write access to the control room, ensuring that the existing IT and OT security architecture remains unchanged.

This approach follows Gradyent’s standardized security framework, which is applied independently of the customer’s system landscape. The platform is cloud-based, providing a new level of computing power while remaining fully independent of local systems and adaptable to different operating environments. This is supported by clearly defined security commitments: ISO 27001 and ISO 9001 certified processes, segmented cloud environments for each customer, exclusively highly secure encrypted connections, and strict GDPR compliance. This enables operation of a high-performance optimization model without compromising data sovereignty or network security.

In the control room, operators continue to have access to multiple operating modes: manual operation, outdoor-temperature-based control, or the setpoints provided by the Digital Twin. The architecture ensures that, despite automated optimization, full control always remains with the operator. This setup therefore combines maximum IT security with high operational flexibility.

Project progress, commissioning, and operational stability

The project started with a kick-off in spring 2023, and platform operation began approximately one year later. Full commissioning followed during the same year as the go-live. Initially, the model representation was visualized externally to build confidence among operators. The system then began providing setpoints, which were retrieved by FUG via the SFTP layer and used within the control system. A phased approach allows model parameters to be continuously aligned with new operating conditions and increases overall system stability.

Operational results: Temperature, losses, and capacity

As a result, the supply temperature was reduced by approximately 4 °C, while the return temperature decreased by around 1 °C. Overall, this leads to a heat loss reduction of about 1–3%.

For summer operation, simulations show that volume flow—and thus pump energy consumption—can theoretically be significantly reduced through optimized temperatures, in some scenarios by up to 50%.

A key lever is customer monitoring: monthly reports on the performance of all substations enable targeted engagement with customers exhibiting unfavorable return temperatures, prioritization of operational measures, and continuous monitoring. Individual poorly performing substations can noticeably limit overall system capacity. Their systematic identification and optimization unlock additional network capacity, often without the need for physical infrastructure changes.

Collaboration and organizational aspects

The project benefited from strong data availability and clearly defined responsibilities. Close coordination between network operations and modeling accelerated the development and validation of the Digital Twin. Customer and measurement data originate from various sources and are integrated via the SFTP layer and gateways. In workshops, model assumptions, operating limits, and intervention points were refined. The following statement from management summarizes the approach:

“Gradyent works independently and in a customer-oriented way. We receive a lot of quality with little effort on our side.”

M.Eng. Patrick Ruf, Head of District Heating / District Cooling, Fernwärme Ulm GmbH

Conclusion and outlook

The introduction of a physics-based Digital Twin supported by data and machine learning enables FUG to continuously optimize its heating network in real time. The combination of thermo-hydraulic modeling, robust data integration, and forecasting logic creates a detailed representation of network behavior that supports both diagnostics and operational decision-making. The reduction of supply and return temperatures, decreased heat losses, and systematic unlocking of capacity through customer monitoring demonstrate the effectiveness of the approach.

The next step—the development of a Digital Twin for another FUG network—is already underway. FUG demonstrates how modern digital tools can be securely and practically integrated into existing control room environments, making a significant contribution to the efficiency and resilience of modern district heating systems.