Open rjbergerud opened 3 years ago
Need to sort out which of these dependencies matter, because otherwise too many feedbacks to model
{n weather stations irradiance} -> site irradiance -> solar power output -> panel_temp solar power output -> panel_temp site_irradiance -> panel_temp site_temperature -> panel_temp panel_temp -> efficiency (site irradiance, efficiency) -> solar power output
There's quite a bit of information we can make use of to get better PV power predictions given-data modelling. This however is unlikely to improve our results much, since we tend to have accurate outputs for fully-sunny days, which means our model probably has learned the relation betweeen solar irradiance and PV power pretty accurately.
What this means is that we could fit a linear model for efficiency of the panel as a function of temperature by selecting day where we're sure the irradiance value at the site is same as at stations. Knowing this relationship, which I don't think has much uncertainty associated with it, would mean that we can model site irradiance from solar power data (by taking the inverse function). That way, we're modelling the real point of uncertainty in the network of variables
{n weather stations irradiance} -> site irradiance -> solar power output