Consumers typically view energy price as rather stable, with slight changes introduced carefully every 6 or 12 months, never more frequently.
So, it may come as a surprise that the wholesale energy market often experiences the most violent price swings of all trading goods.
“In some geographies the price of electricity can rise by over 5000% within a few minutes and drop down just as quickly. It’s a jump from typical $20-$30 baseline per MWh to over $2000 and more. One such place is Texas in the US, “ explains David Inbar, CEO of Dejalytics based in Texas.
David Inbar, CEO of Dejalytics based in Austin, Texas
Dejalytics specializes in analytical services for energy companies, something David has been doing for over 10 years.
As he explains, the drivers behind such dramatic price spikes are a combination of 4 factors:
- Very limited ability to store electricity
- Difficulty in predicting generation and demand as more renewable sources are added to the grid
- The time lag in spinning up additional power generators to meet unexpected surges
- The fact that demand must be met continuously and reliably, regardless of the cost.
When a substantial and previously unforeseen supply/demand imbalance occurs – the energy cost can spin out of control much like Uber’s surge pricing – just much worse.
Being able to predict such events would enable spinning up additional generators, or reduce consumption, in advance. This would be hugely beneficial for everybody: energy providers, consumers and the environment since the high price periods coincide with high levels of green house gas emissions from generators.
“We are collecting all available data ranging from weather signals through the past demand to detailed pricing in hundreds of geographical locations to cover all possible indicators for a spike in demand,” describes David.
But having a wealth of data is just a starting point.
“All those signals require massive work to reshape them before they can be analyzed. There are thousands of potential hypotheses that we would like to test and it’s a classic needle in a haystack. We don’t have an army of data engineers or data scientists to do that work. In addition, we need to refresh our predictive models quite frequently because the grid itself is changing more frequently.”
Unlike traditional Machine Learning software – including current AutoML solutions, Xpanse AI is capable of building models by creating predictive variables from the raw signals stored in the database.
Ger Harte– CTO of Xpanse AI explains “Xpanse AI was built to enable rapid Predictive Modeling on raw databases. No more manual data wrangling or Feature Engineering. The machine is doing all the heavy lifting.“
With Xpanse AI technology, the Dejalytics team has been able to experiment with different forecasting targets using all the data at their disposal at a speed that was not possible before.
“We have analysed other available solutions and, for our purposes, they could not do what Xpanse AI is doing. When delivering the models, we are down from months to days or even hours. The time saving is critical for us, “ summarizes David.