GridStor Transforms Energy Price Forecasting with TimeGPT

Summary

‍GridStor, a leader in energy storage, faced a daunting challenge: accurately forecasting electricity prices to optimize storage and sales. By leveraging Nixtla's TimeGPT, GridStor transformed its forecasting process with precise, long-term predictions and a significant reduction in operational costs. From complex models to streamlined efficiency, TimeGPT’s robust platform provided the reliable, scalable forecasts GridStor needed for smarter financial and operational decisions.

According to Brett Rudder of GridStor, “TimeGPT was a breakthrough for us, transforming our budgeting and forecasting capabilities.”

With Nixtla’s support, GridStor has turned forecasting challenges into strategic advantages.

‍
Customer Overview

‍Company headquarters location: Portland, Oregon
Company size: 11-50 employees
Industry: Energy and Utilities

‍

GridStor is a rapidly growing developer, owner and operator of battery energy storage systems (BESS). The company strategically sites projects to enable electricity from renewable resources, such as wind and solar, to charge batteries in regions where demand is highest. GridStor’s mission is to build a more resilient grid to mitigate the risks posed by climate change and severe weather events.

The company operates primarily in the U.S., with a focus on states like California and Texas, and aims to reduce carbon emissions through innovative energy storage solutions.

Brett Rudder, Senior Manager of Market Analytics, GridStor Brett Rudder has 12 years of experience in organized U.S. energy markets. His expertise spans financial and hedging products, as well as ISO/RTO market structures and operations.

‍
Challenge

GridStor faces the challenge of efficiently and accurately forecasting electricity prices based on supply and demand factors. The company needs reliable time series forecasts to determine optimal times to store and sell electricity. While the team previously used industry-standard models such as ARIMA regressions and production cost models, these methods were either too specific or costly to maintain, with some AWS instances running up to $20,000 per month. In addition, accuracy was lacking, with forecasts often differing from actual results by 25 percent or more. GridStor requires long-term forecasts to support the financing and construction of battery storage infrastructure, with predictions needed on both a monthly and annual basis.

‍
Solution

GridStor experimented with Nixtla's open source solution for a few months before switching over to its commercial offering TimeGPT, which provided GridStor with a robust and flexible solution to accurately forecast electricity prices. By integrating supply and demand fundamentals into TimeGPT, GridStor was able to generate highly accurate time series forecasts that signal the best times to store or sell energy. The long-horizon capabilities of TimeGPT allowed GridStor to generate reliable monthly forecasts, significantly improving their ability to make informed financial decisions, coordinate outages, and hedge electricity prices. TimeGPT’s ease of use also reduced GridStor’s reliance on costly and complex production cost models.

‍
Implementation

GridStor implemented TimeGPT by integrating its API into the team's daily forecasting pipeline, running it on VMs which meant very little IT support was needed. The company primarily focuses on forecasting at the state level, using data from sources like ERCOT and YES Energy to inform its models. GridStor’s forecasting team, consisting of four engineers, relies on TimeGPT for hourly forecasts up to one month in advance and is exploring additional forecasting opportunities for longer-term predictions.

Rudder says, "The TimeGPT documentation and tutorials are really good and helped get us up and going quickly. Now we just hit the API key and get a new forecast in two seconds."

Impact

Since adopting TimeGPT, GridStor has achieved significant improvements in forecasting accuracy, particularly for month-ahead hourly forecasts, which were previously difficult to generate reliably. The ability to chain monthly forecasts together has also opened up opportunities for quarterly and yearly predictions. This has provided GridStor with the data necessary to make more informed decisions about energy storage, hedging opportunities and operational planning. By reducing the need for expensive AWS instances and other models, TimeGPT has also helped GridStor lower operational costs by 50%.

‍
Testimonial

‍
"TimeGPT, especially with the longer horizon, was a big breakthrough for us because that's where we do our budgeting and answer to our board about what the forecast is going to be. I recommend TimeGPT to other people constantly."
— Brett Rudder, Senior Manager of Market Analytics, GridStor