Multi-Horizon Short-Term Smart Grid Load Forecasting Using LSTM Networks for Real-Time Optimisation

Accurate short-term forecasting of electricity demand is essential for real-time smart grid optimisation and efficient energy management. This paper presents a multivariate Long Short-Term Memory (LSTM) neural network framework for short-term multi-horizon electrical load forecasting using historical smart grid operational data, historical load measurements, renewable energy variables, and environmental factors used as explanatory features for load forecasting. A comprehensive data processing pipeline is developed, incorporating time-series feature engineering, cyclical temporal encoding, chronological data partitioning, and standardisation to ensure robust model training without information leakage. Sliding-window sequence generation is employed to transform time-series data into a supervised learning format suitable for deep learning architectures. The proposed stacked LSTM model is trained to forecast short-term load dynamics at a 15-minute temporal resolution using historical energy and environmental data. The forecasting framework is evaluated using multiple performance metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). Experimental results demonstrate that the LSTM model effectively captures temporal dependencies and seasonal patterns in energy consumption, achieving high forecasting accuracy suitable for real-time decision support. The forecasting framework developed in this study serves as the predictive component of a broader smart grid optimisation system. The generated forecasts will be integrated with a multi-objective optimisation framework in subsequent work to enable real-time energy management under uncertainty and improve operational efficiency in modern smart grid environments.