Uncertainty is the hallmark of any forecasting endeavor. No one really knows what the future holds. Often it is informative to quantify that uncertainty. There are a number of ways that we can help our clients do precisely that.
Probabilistic Weather
The TESLA Model is dependent on a weather input. However, that input does not always have to be the most accurate weather forecast available. Many weather forecast data vendors offer probabilistic temperature forecasts. We can combine these temperature forecasts with more deterministic forecasts of other weather variables to generate demand forecasts associated with different temperatures themselves associated with a different probability of exceedance.
Probabilistic Demand
Using probabilistic weather quantifies the uncertainty inherent in a given weather model, but it does not address the uncertainty inherent in the TESLA Model. We can craft systems that meticulously track the accuracy of our forecasts at every hour of the day over any desired forecast horizon. We then use those accuracy statistics to build probability distributions and extract probabilistic demand forecasts.
Probabilistic Ensemble Demand
We can also craft systems that take probabilistic forecasting to the next level. By leveraging the power of an ensemble of independent weather forecast inputs, we can generate an ensemble of independent demand forecasts. We track the accuracy of each ensemble members, and use those accuracy measures to build an ensemble of probability distributions. By strategically aggregating those distributions, we can extract probabilistic demand forecasts that quantify the uncertainty from weather model error, weather model choice, and TESLA Model error.
