Machine-Learning Powered Energy Futures Forecasting
Our energy client had a unique position as a hybrid, focused on generation and selling supply to commercial and retail customers.
Energy use–and cost–vary from day to day, and unanticipated weather events or generation facility performance can have a substantial impact on pricing. Like the stock market, there’s always some level of mystery and risk to prediction. Inaccurate forecasting comes with a hefty price tag for energy companies.
Our client’s energy traders had no shortage of information to drive their decision-making, anchored by a predictive platform using historical data and multiple vendors who provided information on weather and generation plant availability. However, that data had to be manually consolidated and analyzed in antiquated solutions, and traders were challenged to successfully exploit it for maximum results. This impacted traders’ ability to drive real-time and day-ahead energy price trading decisions.
Our teams engineered a new Day-Ahead and Real-Time (DART) forecasting model that dramatically improved forecasting accuracy and speed. The new model is 40x more efficient and runs in minutes. It is also fully automated and is five times more accurate than the previous model.
The team also built a new enterprise-level support platform to deploy this model and future models onto the client’s newly modernized tech stack. It previously took our client days to update their old model. In a time-sensitive business like energy, that equates to days of compromised profits. Our new model was able to update frequently to keep up with the volatile market.
This enterprise-wide, model pipeline process has set a standard for reliability that is now used across the client’s multiple data science teams.
5X Forecasting Accuracy
Improved the model from a 25% margin of error to less than 5%, ensuring confidence in the model forecasts
1 Day Deployment Time
Redeployment time was reduced from several weeks to less than one day, dramatically reducing technology overhead costs
40X Run-Time Decrease
Reduced the time to run the model from an hour to less than 1 minute and removed all manual intervention points
Modernize an existing model
Dialexa’s Data Science team conducted workshops during the discovery phase and shadowed key stakeholders to understand their processes and pain points. Applying our design thinking approach, the technology assessment allowed us to analyze their current platforms, technologies, and data sources used to uncover trends. Our assessment assessed types and efficacy of the models through a more sophisticated and complete analysis of the available data.
Margins were shrinking in some markets more aggressively than others. Accuracy was more critical than ever to their bottom line, yet their existing model–which contained lots of data–underperformed.
This underlies a common data problem across industries; having a lot of data isn’t helpful or informative. Instead, knowing what data is the correct data and understanding how it should be viewed matters tremendously.
A Fortune 500 integrated energy company
- Data Scientist
- Engineer
- Engagement Manager
- Engineering Lead
- MLOps
- Platform Engineering
- Architecture
- DevOps
Let’s make it happen.
The path to your next great product, invention or software application starts here. The first step is starting the conversation. Simply fill out this form, and we’ll reach out immediately.
"*" indicates required fields
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.