SolarCast-ML
GraphCast Extension focused on supporting renewable energy production
A novel machine learning model to extending GraphCast into the realm of renewable energy
Our comprehensive suite of professional services caters to a diverse clientele, ranging from homeowners to commercial developers.
Accessible
The training dataset easy is to collect using commercial-off-the-shelf equipment.
Efficient
Each lightweight model instance runs in milliseconds.
Flexible
Designed to run on GraphCast outputs, but can successfully run on any weather source with proper datapoints.
Accurate
Fully trained models are accurate to under 40W/m2 error per data point on average
Modular
You only have to run SolarCast-ML on the ERA5 datapoints that are relevant to you.
Responsible
Allows for deeper renewable energy integration into our power grid.
How to setup the model
Step-by-step details on how to predict the sun
Collect local weather data
- Use locally installed equipment to collect basic weather parameters along with solar output
- Upload data to AWS
Process the Data
- Extract relevant datapoints
- Load data into Numpy Array
Train the Model
- After testing multiple network designs, we landed one hidden layer with 32 nodes.
- Best optimizer for stability was Adam.
- Testing sets ran for 100 epochs, final training ran for 1000 epochs.
- Each input layer node corresponds to a data requirement: temperature, wind speed, dew point, humidity, solar altitude, barometric pressure, and precipitation event.
Execute the Model
- Collect weather forecast from preferred source.
- Data requirements: temperature, wind speed, dew point, precipitation, humidity, solar altitude, and barometric pressure
Join 3+ subscribers
Stay in the loop with everything you need to know.