Modelling Energy Usage and Renewable Energy Generation
Electricity is a vital component of modern society. A majority of the energy produced and used in Australia comes from non-renewable sources, which presents a major problem for the future, with concerns regarding the environment and a lack of the fuel required to support the power network. Renewable energy sources such as solar and wind energy can help reduce the environmental impact of electricity generation; however, they suffer from a large amount of variability based on effects such as weather. My project was to model energy usage and renewable energy generation, in particular solar generation, to investigate the amount of generation and storage required to support the South Australian grid.
To that end, I first had to collect data for 2018. The Australian Energy Market Operator publicly publishes data on both electricity generation and usage. However, there were several problems with the data sets. First, for generation, an archive file was published each day of the year, containing spreadsheets of generation values. Only, each spreadsheet was also contained in an archive file, in a different format to the main archive files. After many hours of tedious file copying, I managed to get the data set loaded into MATLAB, the programming language I used for my project, only to find out that the data set only covered scheduled generators, which includes power generation from sources such as coal and gas, but not solar or wind generation. So, after many more hours of file copying, I finally had the full data sets for both renewable generation and usage for 2018.
From the data, I had to remove the long-term trends such as seasonal cycles. This is because it is reasonable to assume that the general trend, such as higher electricity usage in summer and winter due to air conditioning and heating, would stay approximately the same each year. The short-term fluctuations, which are largely caused by environmental conditions, were then modelled using AR(1) models; essentially, at each time, the amount of generation or usage is modelled as being some combination of previous values, added to a random component. This worked surprisingly well in fitting the data, as it captured most of the random fluctuations shown in the data.
The AR(1) models were then combined to create a model for the power transfer, i.e. the electricity that would flow into or out of a battery assuming infinite capacity and no self-discharge. Obviously, this is not realistic, so these effects were applied to the power transfer, to get a model for the current battery charge over time. This was then simulated for a range of cases, varying both the amounts of electricity generation and the battery storage. From the simulations, I determined that the optimal combination is approximately 4.2GW of solar generation, or about 80 medium-sized solar farms, connected to 5,200,000MWh of energy storage – a battery 40,000 times larger in capacity than the current largest battery in Australia. These amounts could be reduced by including alternative generation methods to the model, such as wind or geothermal which are more steady over time.
Scott Carnie-Bronca was a recipient of a 2018/19 AMSI Vacation Research Scholarship.