Studying Energy Arbitrage

Anand KC
Anand KC
Nov 3 · 3 min read

(This is an abstract summary of the work carried out as a part of a industry project to show what I have learnt regarding energy arbitrage. The terms associated with energy arbitrage in this blog are referenced from various research and scholar’s papers as well as web sources.)

Let’s first understand what arbitrage is. Arbitrage is defined as

Buy at a particular time and place for a lower price, then go somewhere else and/or wait and sell it for a higher price.

Arbitrage using energy storage is done by charging during lower-price hours and discharging during high-price hours. Our study with regard to energy arbitrage involves batteries. Another associated term with battery and arbitrage using energy is tariff arbitrage.

Tariff arbitrage is the act of purchasing electricity from the grid when electricity is cheap and putting away it for later use when grid electricity is expensive.

There are several energy-related organizations in the market that are working on the same principles to make a profit. This might be beneficial for the organization however it puts customers on the losing side. The concern of global warming with regard to limited availability of fossil fuels has expanded the horizon to explore renewable energy. The study and research of renewable energy and its integration into arbitrage using energy is widely discussed and research area. These days’ renewable energy startups are trying to enter the big market of fossil energy and uplift the customers from the losing side. However, this too involves profit/loss in energy market to new and old players.

Before, analyzing arbitrage using energy, foremost it is important to understand the fundamental drivers that play an important role in giving rise to those phenomena that we try to discover. Such fundamental drivers that are looked after in the analysis of energy arbitrage with regard to renewable and non-renewable energy are loads, weather (temperature, wind speed, precipitation), fuel prices, surplus generation, scheduled maintenance or forced outages of the important power grid, marginal loss factor among others. Organizations like AEMO have studied these factors and built the models which forecast the spot price and demand for the day. Looking at those forecasted numbers, one can study the lower-priced hours and higher-priced hours.

Data from organizations like AEMO are publicly available which can be used to study and understand the arbitrage using energy. During my study, I got opportunities to study and analysis AEMO data to understand arbitrate using energy. This study involves analyzing lower-priced hours and higher-priced hours, detecting anomalous behavior which is a periodic collapsing phenomenon. If an anomalous behavior gets a normal value, it is an anomaly in terms of a periodicity. Detecting anomalous was a part of the study which could be used for the beneficial purpose.

The study of energy arbitrage using AEMO data required study and review of previous work in regard to energy by the researchers and scholars. This included inquiring into fundamental drivers considered by the researchers and scholars, various techniques and data used for learning, building and concluding results. The study of previous research and the nature of data given concluded us with two machine learning techniques. These two machine learning techniques are Autoregressive integrated moving average (ARIMA) and Long Short-Term Memory (LSTM).

Modeling these two machine learning techniques with AEMO data generated similar figures compared to the actual. This involves lower-price hour, higher-price hour and anomalous behavior. Both the built models were able to keep track of the phenomenon shown by the data in terms of lower-price hour, higher price hour and anomalous behavior. However, long short term memory technique appeared to be more efficient as it accounts for the non-linear aspect whereas autoregressive integrated moving average is based only upon the linear aspect from data. Due to this, the LSTM model was more accurately able to understand and take hold of patterns, trends and anomalous behavior within the data including lower price and higher price hours.

Though, a simpler model was used to study the arbitrage using energy with AEMO’s data. There is a possibility for different models and techniques that will generate more precise results. Few of such models that could be explored are multivariate, multistep, multiple outputs, hybrid among others.

Anand KC

Written by

Anand KC

Data Science Ninja

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade