Day ahead load forecasting for the modern distribution network - a Tasmanian case study

Penetration of distributed energy resources in distribution networks is predicted to increase dramatically in thenext seven years, bringing with it the opportunity for utilitiesto have a greater presence at low levels of the network. Toachieve this effectively, utilities will require accurate short termload forecasts. This paper presents a novel neural network-basedload forecasting system that applies recent advances in neuralattention mechanisms. The forecasting system is trained andassessed on ten years of historical half-hourly load, weather,and calendar data to produce a 24-hour horizon half-hourlyonline forecast. When forecasting during anomalous peak holidayperiods on a feeder that has a typical load of less than 1000kVAthe forecasting system achieves a MAPE of 7.4% and a meanerror of -15kVA. The forecasting system is implemented in aresidential battery trial and is able to successfully forecast majorpeaks with sufficient lead time and accuracy to enable the fleet ofbatteries to charge ahead of time and provide network suppor