Business Case - Forecasts
Better Forecasts using Machine Learning
Many kinds of forecasts are made using time series, but the problem with time series is that they don't adapt to changing conditions, while machine learning is a regression itself. Forecasts can be a lot more accurate using machine learning. For generating good results with machine learning, a deeper knowledge of the variables that affect the forecast is required.
Electricity Demand Forecast for the City of New York
Electricity forecasts are sometimes provided by the ISOs, institutions such as CAISO, ERCOT, MISO, NYISO and so on to help companies such as PG&E, Southern California Edison, Constellation Energy, Oncor and many more with their day ahead forecasts so they can organize their resources and procure energy for the day ahead market. There are also other vendors that also provide electric forecasts that serve as a second opinion and may turn out to be more accurate than the ISO forecasts. Ours is one of those forecasts, as it is based on modern machine learning methods and it is calibrated using your very own data
Residuals Review, 5 years output
The residuals table show that most errors are within +/- 5% error (about 95%, if data is counted). There are a few outliers during the 5 summers due hot weather. A better temperature location could be selected to decrease the seasonality error. Still, its precision is quite remarkable