Fusion was engaged by a State electricity organisation to apply machine learning technology to predict the demand on the WA electricity grid (SWIS) up to 7 days ahead.
The data sets consumed by this advanced technology was weather observations (temperature & humidity), substation loads (analogs) over a two-year period and connections for large generation and solar generation points. The model was adapted for focus on peaks in demand which resulted in a more accurate prediction. The Long Short-Term Memory algorithm proved most effective and resulted in a forecast of peak demand under 4% error. This machine learning technology was requested to be operationalised for the Network Operations engineers to use on a daily basis.
The Long Short-Term Memory algorithm proved effective and resulted in a forecast result within 4% of total network peak demand. This algorithm was requested to be operationalised for the Network Operations engineers to use on a daily basis.