Solar generation is forecast to be the largest source of electricity generation globally by 2040.
The rise of solar is creating challenges for electricity grid operators and market participants alike. The energy generated by solar panels changes rapidly as clouds move overhead and weather forecasts struggle to predict clouds accurately.
As solar penetration increases, the uncertainty in solar electricity forecasts is causing headaches not just for those with solar assets but for anyone building a business or optimising their electricity use.
Current forecast approaches use Numerical Weather Predictions, which take several hours to compute. This leaves a “blind spot” in forecasting short time horizons from a few minutes to hours ahead. Experienced practitioners and meteorologists can fill this gap by using satellite imagery and “human intelligence” to informally improve the solar forecast. But this is not scalable as meteorologists are not available to most teams and with machine learning, it should be possible to automate these intuitions.
It needn't be like this.
Open Climate Fix manages this torrent of data with a robust data pipeline to enable our customers to rely on the best forecasts, without the need to download and wrestle with Gigabytes of satellite and weather data. The machine learning prediction is fast to run, delivering actionable information in minutes after a new satellite image is recorded, and forecasts are available through a user interface or automatically through an API.
Continually updating forecasts in real-time
High accuracy from cutting edge ML
High temporal and spatial resolution
Expected and tail forecasts
Open Climate Fix observed that the gap in short term forecasting of clouds is filled at present through meteorologists - their experience applied to satellite imagery - and real-time ground sensors. What if we applied the latest in machine learning to all the data sources available simultaneously, and developed a machine intelligence to replicate or even surpass what is available today?
NOWCASTING takes in all the data - satellite imagery, Numerical Weather Predictions, topographic and solar generation data - and applies cutting edge deep learning techniques not before seen in energy forecasting. These multimodal machine learning techniques accept data natively from different input source types without interpolation and have been only recently possible with advances in computer hardware and techniques developed in the Natural Language Processing field. Open Climate Fix has implemented these techniques and applied them to solar energy, drawing on a breadth of industry experience from Google Deepmind, NASA, wind forecasting and the transmission system operator.
Through exhaustive experimentation and optimising of architectures, Open Climate Fix has developed a best in class forecasting framework for both deterministic and probabilistic forecasts.
Open Climate Fix is building our model completely out in the open! We are actively building collaborations, working with contributors and not restricting the IP behind the model.
Why? It is becoming clear that open source software projects grow faster, are more robust and are more responsive to their users' needs. Linux, Apache, and Python are leaders in their fields. By having an open model which researchers or our clients can build on and improve, we will develop a forecast which is continually improving with the state-of-the-art.
So why buy an open service? Because NOWCASTING is much more than the model. Open Climate Fix manages and procures the huge data volumes involved; we keep the service available; we integrate the latest update from the open-source model to give the best forecast. Lastly, we can provide expert advice and tailor your use of the product.
At Open Climate Fix our mission is to make managing PV power as easy as possible. Going open source is going to help us achieve that goal.