Business Case - Price Forecast
Price Forecast of Electricity using Machine Learning
Day ahead price forecasts of electricity are typically more complicated than demand forecasts as there are more variables to take into consideration and effects like supply vs. demand. In addition, they also require the availability of more forecasts to come up with its own.
Electricity Price Forecast for a Major Utility
The price forecast for a major utility was developed and tested against their historical data, the pricing has been removed from the graph for privacy purposes, so only dots are presented. The follow up of the forecast against the historical price is quite remarkable
Residuals Review, 2 Years Output
The residuals table show that most errors are within +/- 5% error. The major outliers were during the craziness that happen about the 15000 hours mark, where the price of fuel to generate electricity considerably increased in price. It's important to notice that while the price itself increased about 6 times in magnitude, the magnitude of the error in the forecast was only about 1 time in a few data points (when given an accurate forecast price of fuel).
These residuals graphs are quite important as we can tell if the model that was created works under all conditions given for training the algorithm. If we see that there are groups of points that are somewhat out, those can be studied in more detail to see if there are other variables to be taken into consideration.