Data-Driven Methods for Enhancing District Heating Network Operation

This dissertation is a collection of publications that researched the operation of district heating networks. The district heating sector will contribute significantly to the future smart energy integration of renewable energy into the power system due to its high efficiency and flexibility in converting power into heat and its thermal energy storage properties. However, reaching optimal district heating operation for both individual and overall energy systems (sector coupling) involves many complex tasks, the solution of which is of great importance. One of them is to operate the district heating network efficiently. This is done by keeping the supply temperature as low as possible without violating any restrictions (e.g. maximum flow, variation of supply temperature) while at the same time meeting the heat load. This results in lower production costs and reduced heat losses in the network. Lowering the temperature in the network also leads to better investment feasibility for power-to-heat units (e.g., heat pumps), low-temperature geothermal wells, and recycling of waste heat. Without lowering the temperature, these sources would result in lower efficiency or even be disregarded.

Hence, to enhance the operational efficiency of the district heating system, the network must be operated in an optimal mode. Many district heating networks are operated by operators who use their experience and ”scarce” data to select the set points for the supply temperature. This usually leads to suboptimal operation because the network characteristics are complex, and many variables have to be taken into account, e.g. the future heat load and time delays in the network. The objective of this thesis is to propose data-driven methods that can support operators in decision-making. In particular, mathematical data-driven models will be developed based on physical knowledge and designed for use in real-time applications to increase the efficiency of operations.

Various studies have been carried out as part of the project to investigate state-of-the-art and sub-sequentially propose new methods to highlight the importance of data-driven methods to obtain optimal operation. For instance, heat load forecasting is crucial for overall district heating operations as it gives operators insight into future consumption. This gives operators the essential information to support their decisions to minimise the costs of operating the district heating system. The more accurate the forecasts, the better the operators can make their decisions. This thesis discusses the essential features for building an accurate and robust heat load forecasting model. It is shown that the localisation of the input variables and heat load has an impact on the accuracy of the heat load forecast. In addition, new methods are presented to increase the accuracy of the current operational forecasts. The proposed methods use and extend state-of-the-art methods for hierarchical forecasting. Both temporal and spatial hierarchies of district heating are considered in order to investigate the possibilities for improving today’s state-of-the-art operational forecast. It is shown that the proposed method can improve the accuracy of current operational forecasts by about 15%. These methods will be indispensable in future decentralised district heating systems as the suggested methods both increase accuracy and make them coherent across the considered temporal and spatial aggregation levels.

Methods for temperature optimisation for the district heating network are presented and discussed. A set of controllers are used in the optimisation, supply temperature controllers and flow controllers. The supply controller ensures that the temperature is sufficient at a set of selected critical points of the network. These critical points are selected such that if the temperature is sufficiently high at the critical points, then the temperature is sufficiently high everywhere. The flow controller ensures that the flow restrictions in the system are not violated, and here the time-varying electricity prices can be taken into account. Measurements at the critical point are needed to serve as temperature feedback for the controller. Typically, measurement wells are installed in areas near a group of end-users, but these measurements come with an associated cost, and the equipment needs to be maintained to ensure reliability and high precision, i.e. temperature sensors must be finely calibrated.

This work proposes that smart meter readings can replace these measurement wells to establish the needed temperature feedback for the supply temperature controllers. Both simple and complex methods are presented. The simple method is easy to understand and can be implemented with little computational effort. On the other hand, the complex method is more robust but requires more fine-tuning. The potential savings from implemented temperature optimisation are discussed and demonstrated in a case study where the temperature is kept as low as possible. Based on real-life implementations, it is demonstrated that the precision is increased. The use of smart meters also leads to additional information about the district heating network since the network of the individual users is also taken into account. It is argued that the suggested methods give new possibilities for adaptive zonal temperature control, which could lead to further savings on heat loss and better integration of heat pumps and the use of excess heat from supermarket cooling, etc.

Finally, the potential of integrating the consumers’ heating system into the network operation through the smart operation with predictive controllers that can, for example, receive signals from the district heating system to influence its heating consumption (peak shaving) is discussed.

In summary, several new methods for data-driven optimization of district heating systems are suggested. These methods can be integrated with existing methods and lead to further savings, better energy efficiency and flexibility. The suggested methods range from simple to more complex methods, but all of the methods are intended to improve online operations.