Residential Demand Forecasting With Solar-Battery Systems: A Survey-Less Approach

Due to the push to reduce greenhouse gas emissions in the residential sector, new precincts are likely to include a higher uptake of solar and battery systems. One fundamental tool required to assess the impact and cost of solar and battery systems in residential precincts is a reliable model that can simulate electricity demand for individual homes. That model should capture the time-series complexity and the unique demographic composition of a precinct. In response, this research contributes to state-of-the-art demand forecasting by presenting a time-series, survey-less, demographically focused model formulation. The model achieves the survey-less feature by substituting household surveys for geodemographic profiling data, avoiding comprehensive surveying used in other research. This increases the usability for developers and network operators. The research applies the adaptive boost regression tree algorithm as an alternative to traditional methods, such as feed forward neural network (FFANN). The results investigate three case study precincts and show that this algorithm improves over an FFANN in aspects such as peak event estimation and daily variability, achieving an R
²
value of 0.86. Finally, the research implements an MILP battery model and demonstrates the demand model is robust to provide real value battery system design, cost, and impact estimations.