Smart Home Energy Digital Twin with Agentic Optimisation Control

Abstract

Dynamic time-of-use electricity tariffs expose households to half-hourly price volatility, yet most domestic loads and battery systems are still operated on static schedules that ignore this signal. This research presents a smart home energy digital twin that couples real half-hourly tariff data from the Octopus Energy Agile API with historical solar radiation data from OpenMeteo, a physics-informed home battery model incorporating charge and discharge efficiency and an explicit cycling degradation cost, and a deferrable appliance load. Three autonomous control strategies are evaluated within the digital twin: a static baseline controller; a forecastaware agent that applies interpretable price look-ahead rules for battery arbitrage and appliance scheduling; and a mixed-integer linear programming (MILP) optimiser that computes a costminimal 24-hour schedule subject to battery dynamics, energy-balance, and appliance-deadline constraints. A 30-day backtest over December 2024, selected as a deliberately adverse low-solar and high-volatility stress period, shows that the forecast-aware agent reduces the effective energy cost by 21.41% (£28.20) relative to the baseline, while the MILP optimiser achieves a 27.71% reduction (£36.50) and is cheaper on every one of the 30 days. An analytical derivation and a sensitivity analysis confirm that the agent’s decision thresholds are economically grounded and that its performance is robust across wide threshold ranges. Notably, the optimiser attains the lowest cost while importing slightly more grid energy, demonstrating that intelligent load shifting into low-price windows, rather than import minimisation, drives cost reduction under dynamic pricing.