Spatio-temporal modelling of solar photovoltaic adoption: an integrated neural networks and agent-based modelling approach

This paper investigates the spatio-temporal patterns of solar photovoltaic (PV) adoption, solving the ongoing need to inform the management of the distribution networks with spatially explicit estimations of PV adoption rates. This work addresses a key limitation of agent-based models (ABMs) that use rule or equation-based decision-making. It achieves this by adopting an aggregated definition of the agents using artificial neural networks (ANN) as the criteria for decision-making. This novel approachdraws from both ABM and Spatial Regression methods. It incorporates spatial and temporal dependencies as well as social dynamics that drive the adoption of PVs. Consequently, the model yields a more realistic characterisation of decision-making whilst reflecting individual behaviours for each location following the real-world layout. The model utilises the ANN’s approximation capabilities to generate knowledge from historical PV data, as well as adapt to changes in data trends. First, an autoregressive model is developed. This is then extended to capture the population heterogeneity by introducing socioeconomic variables into the agent’s decision-making. Both models are empirically validated and benchmarked against the Bass Model.

Results suggest that the model can account for the spatio-temporal and social dynamics that drive the adoption process and that the ABM and ANN integrated model has superior adaptive capabilities to the Bass model. The proposed model can estimate spatio-temporally explicit forecasts for up to five months with an accuracy of 80%. In line with the literature, results suggest that income, electricity consumption and the average household size variables yield the best results.