Conclusion

A reliable statistical model wasn’t found.

Next Steps

Two main areas of work were not explored further due to time and computing constraints:

Data Engineering

GFS Spatial resolution

The GFS data used was at a 1° spatial resolution. A higher granularity of data would provide a more accurate representation of weather effects.

Expanded Analysis Period

The initial exploratory data analysis suggested a rough seasonality to

Feature Engineering

Isolating Demand-Side Drivers

The curtailment problem could also be described from a demand-side perspective. During periods of depressed or low demand (e.g. days with low air conditioning usage), would exacerbate the curtailment quantity. In this study, a historic load from CAISO was used (though in principle, a day ahead forecast could be dropped in as well.)

Additional data engineering work could benefit the predictive capabilities of the approaches laid out in this study. Of note, historic weather forecasts at a higher resolution