Using Machine Learning to Predict California’s “Peak” Fossil Fuel Consumption

In our previous post on climate change, we shared our machine learning engine’s (GNY Brain) ability to predict demand for energy in California. With predictions that were 5% more accurate than those from U.S. Energy Information Administration we were emboldened to look deeper into predicting the future of the California energy market.

We wanted to know when the “peak” was going to be. “Peak” is used to describe the moment of peak consumption for a product or category which is in the process of being disrupted by a competitor or new technology. After a market’s peak it starts to contract as the competitor’s growth crowds it out. Renewable energy is crowding out fossil fuel in electricity production. We wanted to predict the exact date of the peak. It is important to note that for the purposes of this exercise we excluded energy produced out of state, and focused only on California based solar farms.

Defining peak is important because of its economic implications. As demand for fossil fuels starts to contract it will become more expensive to secure investment and competitive loans for the production and distribution of fossil fuels. This will make them less competitive and thus increases the speed of the fossil fuels decline as a basis for energy production. Basically, peak is the beginning of a major decline, and for the sake of climate change that is welcome news.

To predict “peak” we needed something more than highly accurate 24-hour predictions on energy demand. We needed to create neural nets to accurately predict:

• 24-hour future predictions of solar energy supply
• 3-year future predictions of solar supply, created by California based solar farms
• 3-year future predictions of total electricity demand

Since the government doesn’t predict supply here is our training data against actual supply data.

Here is our plot of electricity generated by California based solar farms VS demand.

We defined peak by moment that solar production consistently* produced at least 10% of the total energy consumption for the entire state of California. Based on those parameters “peak” will be approximately 2 years from now. February 7th, 2023. You can download the full excel outputs here. We also plotted it on the graph above.

In our next blog post we will share a live API feed of our next-day predictions for energy demand in California, as well as the predictions from the previous day by us and the government alongside the actual. This predictive power will be available for everyone building on our ML-powered GNY Mainchain, which is launching in Q1 2021, and will power the generation of datasets for our GNY Dataplace launching later this year.

What predictive models would you like to build with publicly available data sets? Let us know in our Telegram chat.

Note: The analysis contained in this piece took into account energy that was produced IN STATE, and not imported in from other states.

*defined as mean value over 14-day period