Date of Award

Spring 1-1-2018

Document Type

Thesis

Degree Name

Master of Science (MS)

First Advisor

Gregor P. Henze

Second Advisor

Rajagopalan Balaji

Third Advisor

Wangda Zuo

Abstract

Fifth-generation district heating and cooling (5GHDC) systems are the next generation of district systems which rely on water approaching indoor ambient temperatures (≈20-25C). The lower temperatures allow for additional heat sources to be added to the network, allowing for some buildings to become prosumers (e.g. data centers, supermarkets through refrigeration heat rejection). The energy performance of buildings served by a 5GDHC network is a strong function of the inlet temperature, which corresponds to the network supply temperature. Optimizing the network’s efficiency and its grid topology leads to an evaluation of many possible network topologies, which is numerically expensive. To be able to analyze and contrast any given district energy system layout, the impact of connecting individual buildings to a 5GDHC network must be quickly evaluated, while allowing for flexibility in deciding which buildings should be connected to the network.

Conventionally, each building on a district heating and cooling network is represented by a physical building energy model that is run in conjunction with the network simulation. Although this setup allows for running the analysis with flexibility on the building load side, it results in long-running simulations, thus limiting the ability to quickly analyze various network topologies. An alternate approach to determine the building loads is to create a surrogate model, allowing the network simulation to request the building loads from a myriad of predefined building types and building characteristics. Three reduced order modeling techniques were investigated to replace the conventional long running physical building energy models: ordinary least squares regression (linear models), random forests, and support vector machines.

Share

COinS