UC San Diego

The US power grid is an engineering marvel. However, it was not designed with the consideration of renewable energy, energy storage, two way electricity transfer from electric vehicles, advanced control devices, or advanced metering systems. In order to properly integrate such devices, power system planning studies in which the proposed device is simulated under yearly operation is performed on a real/existing circuit model are performed. However, the studies are extremely intensive with respect to computational and temporal resources due to: i) the size and complexity of real circuits ii) The daily and seasonal variation in load consumption and available renewable resources iii) The number of operating states of the device.

Model reduction is a common approach in big data applications to reduce the burden. However, much like the grid itself, traditional methods of circuit reduction are not designed for the growing complexity of distribution side circuits. Specifically, there are no methods in literature that consider circuits with i) unbalance in loading and generation between phases ii) unbalance in distribution line impedances between phases iii) mutual coupling between phases iv) shunt capacitance in distribution lines v) multiple voltage levels. Further, topics like reduction of forecasted generation/consumption time series or aggregation of voltage-controlled devices have not been discussed.

In this thesis, a circuit reduction technique that is specifically formulated for the complexities of real distribution feeders is introduced. The comprehensive methodology is derived from first principles to overcome all of the limitations of circuit reduction techniques listed above. Detailed algorithms of our recommended implementation are given for increase the utility to researchers.

An extensive validation is performed on several real circuit to develop a granular understanding of error sources. It is shown that for even the largest publicly available models, the method is highly accurate, with time savings of up to 99\% per simulation, while being flexible enough to handle a range of modeling or control complexities.