Simple Models For Operational Optimisation
Description of the project
Today, even for fairly simple DH systems, operational optimisation based on transient simulations of the whole DH system are very demanding on computer simulation time. Consequently, it is very attractive if the operational costs can be minimised by use of much simpler models and with an acceptable degree of accuracy.
Work on "simple" models at an aggregated level have been in progress for the last ten years in Denmark, Finland, Germany and Sweden. Various simulation studies have been performed with promising results. In the present proposal a comparison of the results from Finland, Germany and Denmark will be made, and suitable models will be selected and applied in one DH system in each of the four countries. The difference in operational costs by applying optimisation models based on a complete description and on a simple model will be evaluated.
In order to give guidance through the "state of the art", the existing approaches, models, methods and products will be discussed systematically concerning
- Purpose and aim,
- Potentials of cost-savings,
- Mathematical methods,
- Modelling aspects, specially the representation of network dynamics,
- Data requirements,
- Availability of software products,
- Efficiency and reported problems in practical use and
- Current state of application.
Summary of the final report of the project
The purpose of this project has been to further develop and test simple models of district heating systems with respect to simulation and operational optimisation. The simple models are aggregated models of pipes and consumer installations, or artificial neural network models of district heating networks.
The neural network method works in forecasting the state of the DH network like a simulation. However, the method needs further development to take care of the time delay, especially for a step response of the supply temperature. Based on physical simulations of the DH network, the neural network model can be trained at points in the network where no measured data exist. The cost function used in the neural model can also be supported by the real time process simulator Apros. An optimum cost function was derived as a function of supply temperature and mass flow from the plant.
Automatic simplification (aggregation) of DH networks is possible for steady state as well as for dynamic simulation and optimisation of the operation of DHC and CHP systems. Aggregation of DH networks can be carried out to aggregation depths of 80-95% of the original system with very little loss of accuracy.
Aggregation to more than 90% aggregation depth could be carried out by reducing the errors of the aggregation by optimising the network parameters. It is possible to further develop the aggregation methods.
With the present computers and programme codes, the utilisation of aggregated network models are necessary for optimisation of complex DHC systems. Today, supply temperature optimisation is applicable for DHC systems with a maximum of approximately 100 elements.
Supply temperature optimisation based on simple (aggregated) models can be used to utilise the heat storage capability of the network to optimise the operation of complex DHC and CHP systems. The results are thus very promising with respect to utilising aggregated models, dynamic heat storage optimisation and demand side management for load management of DHC systems.
2002-05-14 /Benny Bøhm
Technical University of Denmark (DTU), Department of Energy Engineering, Denmark
Korea District Heating Company, Rep. of Korea
UMSICHT, Fraunhofer Institut für Umwelt-, Sicherheits- und Energietechnik, Germany
VTT Energy, Finland