Forecast-Informed Shrinking-Tube Model Predictive Control for PV-Battery Building Energy Management

Building-integrated photovoltaic (PV) and battery energy storage (BES) systems require coordinated operation under time-varying demand, intermittent generation, battery constraints, and imperfect forecasts. This study proposes a forecast-informed shrinking-tube model predictive control (MPC) framework for PV-battery building energy management. The method progressively tightens the admissible state-of-charge (SOC) region along the prediction horizon, retaining short-term operational flexibility while guiding the battery toward a preferred terminal range. A signal-decomposition-enhanced iTransformer is used to generate multi-step load and PV forecasts, which are supplied directly to the receding-horizon controller and evaluated in an EnergyPlus–Sinergym closed-loop environment. On the held-out forecasting set, iTransformer achieved MSE values of 0.1466 and 0.0802 for the load-side HVAC and PV series, respectively. In the five-day case study, the proposed controller reduced net purchased electricity to 152.42 kWh and cumulative SOC deviation to 55.89%·h, compared with rule-based control, standard MPC, and fixed-tube MPC. When driven by learned forecasts, the controller maintained stable closed-loop performance, with 153.85 kWh of net purchased electricity and 45.20%·h of cumulative SOC deviation. These results demonstrate the potential of horizon-dependent SOC constraint tightening for PV-battery coordination under non-ideal predictive information.