学术文献库

Expansion planning problems refer to the monetary and unit investment needed for energy production or storage. An inherent element in these problems is the element of stochasticity in various aspects, such as the generation output of the units, climate change or frequency and duration of grid outages. Especially for the latter one, outage modeling is crucial to be carefully considered when designing systems with distributed generation at their core, such as microgrids. In most studies so far, a single statistical distribution is used, such as a Poisson Process. However, by taking a closer look at the real outage data provided by the state of NY, it is observed that the outages do not seem to come from the same distribution. In some years, there is a huge spike in the average duration per outage and this is because of catastrophic events. Therefore, in this study we propose and test an alternative modeling for outage events. This alternative scheme will be based on the premise that outages can be broadly classified into two categories: regular and severe. Under this taxonomy, it can still be assumed that each type of events follows a Poisson Process but outages, in general, follow a Poisson Process which is truly a superposition of these two types. A reinforcement learning approach is used to solve the expansion planning problem and real location-specific data are used. The results verify our initial hypothesis and show that the optimization results are significantly affected by the outage modeling. To sum up, modeling accurately the grid outage events and measuring directly the reliability performance of an energy system during catastrophic failures could provide invaluable tools and insights that could therefore be used for the best possible preparation for this type of outages.