Clustering community microgrids using explainable artificial intelligence for residential households

Microgrids have become increasingly important because of the global shift toward energy decarbonisation and decentralisation. These modular networks enable communities to integrate local renewable energy sources, enhance energy security, and empower consumers to control their energy supply. Meanwhile, the spread of smart meters and IoT sensors provides high-resolution electricity usage data, driving the design of microgrids that utilise machine learning for demand-side insights. Current microgrid studies primarily focus on supply concerns, including generation, control algorithms, and economic optimisation, often overlooking the optimal selection and configuration of households based on detailed consumption behaviour. This gap is important; ignoring diverse household load profiles skews community demand and misses optimal energy-sharing opportunities. Greater load diversity reduces coincident peak demand and enhances resource usage. Conversely, homogeneous loads lead to poor energy sharing, oversized infrastructure, and less efficient demand response. We introduce an XAI-informed residential household clustering method for community microgrids (XAI-RHCCM). K-means clustering organises households by average daily energy use, identifying natural consumption types. An artificial neural network, optimised with Bayesian techniques, classifies new households into these clusters, with Shapley Additive Explanations highlighting the consumption features influencing the classification. XAI-RHCCM combines unsupervised clustering, predictive modelling, and interpretability, focusing on demand heterogeneity in microgrid configurations.