ML4Heat - Tools for the optimized operation of existing district heating networks based on machine learning methods

The increasing trend towards smart metering in district heating networks (DHNs) makes the networks more transparent than they have ever been. Previously the consumer side was autonomous, and the network was optimized based on the experience of a few plant operators. The global objective of the ML4Heat project is the development of methods and software tools to optimize the operation of existing DHNs from an energetic and economic point of view. Meter, controller, and plant data is evaluated using machine learning processes on three levels:1. Individual substations: Tools for performance monitoring and optimizated control of substations are developed and implemented. These range from detecting and where possible repairing anomalous behaviour to minimizing return flow temperatures without the use of additional sensor.2. Strand optimization: Methods are being developed which, based on the measurement data from the district heating transfer stations, can quickly identify anomalous behaviour of subsections (strands), such as high heat losses. For this purpose, machine learning processes are used in combination with basic physical equations.3. Network optimization: First, methods have been developed to predict the energy demand for the entire district heating network more precisely than before. This uses only readily available data and allows for the optimization of the power plants under consideration of heat losses, supply delays and individual consumer patterns.