If you have any further questions regarding our services or see potential for cooperation, please contact us by email or phone number given below:
© heatbeat engineering GmbH & heatbeat nrw GmbH 2017-
in the 72nd issue of our heatbeat Research Newsletter, we present two current open-access articles in which the same quantity turns out to be decisive: network heat losses. The first, in Applied Energy (University of Stuttgart, Fraunhofer IPA), asks how detailed a dispatch model must be, using the Memmingen network. The second, in Energy (TU Berlin), compares low-temperature (around 60 °C) and ultra-low-temperature networks (below 50 °C) with decentralised heat pumps for a district near Bremen.
In September, we were represented at both the Dresden District Heating Colloquium and the “Thermal Networks—Digitalization in Practice” dialogue event in Nuremberg. We gained some interesting insights from both events. We noticed how many operators view the topics of return temperature reduction and digitalization as interconnected.
In line with this, we also showcased the latest developments in our heatbeat Digital Twin at both events. These include, not least, improved settings and displays for the maximum permissible return temperatures per building station, as well as the integration of these conditions with analyses of how different consumers impact the efficiency of the overall network.
To ensure that all functions of the heatbeat Digital Twin can be easily understood on your own, we also launched our newly revised digital user manual in September, which is now accessible via the Help button in the header bar of the Digital Twin. We are currently expanding this resource week by week.
In addition, we were able to further improve the display of background maps in the Digital Twin. This now includes the display of cadastral parcels for all federal states, which facilitates useful contextualization during the planning process. And thanks to the integration of vector-based maps, higher zoom levels can now be displayed without any loss of quality. Together with the aerial images integrated last month, these improvements help round out the spatial context of the heating network, potential areas, and expansion zones.
As a further improvement, we have fundamentally revised the display and management of heat generators along with their control parameters so that they can be better integrated and controlled in the generator simulation. Additionally, other map layers in the network area now display helpful supplementary information. This includes, among other things, an expanded breakdown of capacities per pipe section (now showing effective capacity in direct comparison to the original capacity assumption before applying simultaneity effects) as well as the option to highlight pipe sections that are not utilized under design conditions.
In addition to these improvements, we were able to initiate important developments in September, which we will discuss here and in the Feature Update over the coming months.
Dispatch models for heating networks are almost always simplified – in the extreme case to a “copperplate” that places all producers and consumers on a single node. The authors separate the estimation bias (how far the model misstates operating costs) from the decision regret (the cost of the simple model's schedule re-evaluated with the detailed model). A 2×2 design (losses × topology, each yes/no) separates the two error sources.
The basis is the real industrial network in Memmingen: radial, 15 junctions, 27 metered consumers, 174 substations, 9.9 GWh annual demand, all generation at one site (CHP, boilers, 5 MW heat pump, 5 MW electrode boiler, 500 MWh storage). One year is optimised hourly as a MILP, plus 135 synthetic networks with trunk lengths of 1 to 50 km.
The copperplate underestimates operating costs by 15.1 %. The schedule matters more: blind to the roughly 1.2 GWh of network losses, it plans the heat pump too tightly – 2,063 instead of 2,309 operating hours, 77.9 instead of 87.8 % of production. Valuing the unplanned losses at the cost of the marginal unit, this schedule is 46.1 % more expensive than the reference. Losses explain 95.8 % of the gap, network topology only 0.25 %. Further detail adds little: pressure drop and pumping +1.4 %, all 174 substations under 1 % with no hydraulic violation, transport delay nothing.
The authors derive a rule of thumb: with the loss number λ = annual loss / annual demand, the cost share a copperplate misses is about λ/(1+λ) – simply the loss share of the heat fed in, known from every operator's energy balance (R² = 0.87 across 136 networks). Memmingen has λ = 0.12: predicted 11 %, computed 15 %. Below roughly 10 %, a copperplate may suffice (set, not validated). A lump-sum loss surcharge transferred from other networks is off by 23.5 percentage points on average. In a forward evaluation with lowered supply temperature (no re-optimisation), the best tested reduction of 17.5 K saves €6,220 per year (about 4.6 %, losses valued at the marginal price), and at 20 K the velocity limit is exceeded – hydraulics turn into the binding constraint. A COP gain of the heat pump is not included.
Limits: radial, central generation, fixed heating curve, precomputed COP (mean 2.99), deterministic hourly dispatch. Meshed networks with several producers – the case from our last issue – remain open. Memmingen is compact (pumping about 0.08 % of heat demand, travel times under one hour); large networks may behave differently. With billing meters behind mixing valves, at most the node or zone level can be validated.
The authors compare a low-temperature network (LTDH, supplying all buildings including hot water at the 60 °C level), an ultra-low-temperature network (ULTDH) that only supplies space heating of buildings with 45 °C floor heating, and fully decentralised heat pumps. Booster heat pumps (in the optimisation always air-source) cover space heating of “high-temperature” buildings and hot water of all buildings. The case is Leeste near Bremen: 15.22 GWh demand, 5,956 m trench, 2.56 MWh/(m·a); only 4.9 % of demand is at 45 °C today, so the high-temperature share (buildings needing more than 50 °C for space heating) is around 95 %. A Dymola network simulation (losses, pumping, supply temperature) feeds an hourly design and dispatch optimisation of the heat-pump-based plant. Scenarios vary renovation (25–100 %), connected demand on the same route (25–400 %), network size (25–400 %) and high-temperature share (10–100 %).
Under the base assumptions, the low-temperature network is the cheapest option in most cases studied – partly thanks to the plant's much cheaper electricity. Today it beats decentralised supply regardless of the high-temperature share; only at 0.64 MWh/(m·a) (25 % of today's demand) do decentralised heat pumps win. Below about 1.28 MWh/(m·a), the authors see an explicitly case-specific warning zone. In the current state, ULTDH competes with decentralised supply only below a 60 % high-temperature share, and with LTDH only below about 30 %; between 10 and 30 %, local conditions decide. What matters is how much heat must be boosted in the buildings: reducing the share from 90 to 10 % lowers ULTDH heat costs by 38.4 %.
Renovation lowers peaks and improves the annual COP of decentralised air heat pumps (high-temperature buildings 2.71 to 2.87), but relative network losses rise – and with them the required supply temperature, which degrades the central heat pumps. At 25 % of today's demand, losses reach up to 29.3 % in LTDH and over 65 % in ULTDH at 90 % high-temperature share. Renovation without temperature reduction worsens the network's position in the model – renovation and temperature roadmaps belong together. As pipes are resized in each scenario, existing networks may be hit harder. Conversely, from 5.12 MWh/(m·a) – at least twice the demand on the same route – both network variants beat decentralised supply at all high-temperature shares.
Context: the central-versus-decentralised result depends most on electricity prices – 20.6 ct/kWh for the plant, 33.3 ct/kWh for decentralised heat pumps. If this gap narrows, so does the network's advantage. With costs varied by ±30 to ±50 %, most cases are not clear-cut; decentralised supply is robustly preferable only at very low density and high high-temperature share. Not included: renovation costs, generators other than heat pumps (CHP or biomass would be less temperature-sensitive), meshed networks; hot water is a constant 60 °C without storage, circulation or disinfection; the coupling is one-way. Cheapest is not lowest-emission: LTDH performs worse on operational CO2 (uniform emission factor, no life-cycle assessment).
Both articles are open access and well worth reading. Together they show how central network losses are: in the dispatch model, where they explain almost the entire model error, and in strategic planning, where – alongside electricity prices – they determine supply temperature and the economics of central heat pumps as sales fall. The operator questions are: what is your network's loss share today and after renovation of the connected buildings, and which share of your customers will need more than 50 °C for space heating in the long term? In the heatbeat Digital Twin, we model networks with measured feed-in and substation data and the resulting losses, and simulate renovation, densification and temperature reduction scenarios thermo-hydraulically.
And in October, we’re looking forward to our next quarterly “Feature Update Live” webinar, highlighting the most important new features from the past three months. You can register for the webinar on October 14 at 2:00 p.m. at https://heatbeat.de/feature-update .
The next issue of our newsletter will be published on 4 November 2026.
Your heatbeat team