A generalisable agent-based model of residential electricity demand for load forecasting and demand response management

Electrification and increased uptake of intermittent renewable generation challenge power systems worldwide. These challenges increase with increasing renewable generation, such as in Aotearoa New Zealand. To address these challenges, Demand Response (DR) can reduce peak loads and balance demand with intermittent supply, extending network lifetimes and reducing greenhouse gas emissions. In Aotearoa New Zealand, residential demand is the largest contributor to peak loads and a key target for DR. However, residential demand is highly influenced by human behaviour. Current electricity demand models are typically deterministic or stochastic and do not capture behavioural dynamics, the understanding of which is crucial for successful DR. This research presents an agent-based model of residential electricity demand in low-voltage networks, which is built using high-level census data and thus generalisable to regions with similar available data. The model is constructed in MATLAB R2022b with sub-models for appliance use, space heating, and water heating, and validated with real electricity demand profiles from low-voltage distribution transformers in Aotearoa New Zealand and data from appliance use in homes around the country. By incorporating realistic behaviours and their variability, this model offers a platform for testing how human behaviour influences DR strategies and impacts human outcomes. Thus, it can inform and improve the design of DR programs based on program uptake and desired outcomes, leading to decreased network costs through increased resilience and energy security, and reduced greenhouse gas emissions through better utilisation of intermittent renewable generation.