Dynamic Load Modelling and Optimised Battery Management Using LSTM for Enhanced Power System Flexibility

ABSTRACT This paper presents an integrated framework for enhancing energy flexibility in buildings by coupling long short‐term memory (LSTM)‐based load forecasting with an adaptive battery management system (BMS). The LSTM model predicts short‐term electricity demand using historical data and time‐based features, and the forecast is used to drive realtime battery charge and discharge actions. The control logic incorporates physical constraints, including state‐of‐charge (SoC) limits and maximum charge/discharge rates, ensuring realistic battery operation. To evaluate the effectiveness of the proposed system, comparative experiments are conducted against gated recurrent unit (GRU) and autoregressive integrated moving average (ARIMA) forecasting models, as well as a baseline rule‐based battery controller. The LSTM‐based framework achieves the lowest RMSE (19.43) and reduces battery full equivalent cycles (FEC) by 92.5% compared to the baseline, indicating significant improvements in forecasting accuracy and battery longevity. Sensitivity analysis under a +10% demand increase further validates the model's robustness. The results confirm that integrating predictive forecasting with constraint‐aware battery control can substantially enhance system‐level flexibility and operational sustainability in smart building environments.