Predictive energy demand forecasting
The Challenge
Industrial companies participating in Ontario's Industrial Conservation Initiative (Class A consumers) pay Global Adjustment (GA) based on their percentage contribution to the top five hours of energy use in Ontario over a 12-month base period. Global Adjustment can make up to 80% of an Ontario Class A electricity consumer's annual bill.
Consumers can reduce their GA costs by anticipating the top five peak hours for the current base period and reducing their consumption accordingly. However, correctly identifying those 5 peak hours is not a trivial task. Currently, Class A consumers use forecasts made by IESO, which tend to have an error range of around 1000 MWs.
This means an average Class A customer shuts down 20 days every year to catch the 5 peak hours, effectively losing 1 to 1.5 million dollars on every extra day they shut down. The challenge was further complicated by the fact that many other forecasts were influencing action in the market and altering the amount of energy consumption as a result.
Our Solution
We pulled in weather data as well as Ontario's energy usage data from multiple sources, often in messy formats, and merged them all together. These data sets were updated at a high resolution and at an interval of every 5 minutes. The purpose was to predictively forecast when peak energy times would occur.
These challenges were overcome by using unique ensemble models generated from the data that took into account the variation caused by reactions to other forecasts as a hidden variable. We perform 8 billion calculations per day on climate data and historical demand data to predict the demand for electricity in Ontario.
We scraped web pages and pulled data in from multiple APIs on a regular basis to generate new forecasts automatically and publish them hourly. All of the forecasts are interactive and allow users to scroll through historic analysis and dates to compare previous results. These merged data sets and forecasts were then published to a live interactive dashboard.
The Impact
5 times more accurate than IESO forecasts
Our forecasting engine provides significantly more accurate predictions, reducing the error range from IESO's 1000 MWs.
Reduced annual shutdowns from 20 days to 8 hours
Class A consumers using our forecasting engine can dramatically reduce their shutdown periods, saving millions of dollars in lost production.
Protected against costly false alarms
For example, on August 21st, 2017, IESO called a peak hour 5-6pm but our system correctly predicted August 21st as a non-peak day, saving companies from unnecessary shutdowns.
Real-time interactive forecasting dashboard
Hourly updated forecasts with interactive historic analysis enable companies to make informed decisions about when to reduce consumption.