A data-driven method for predicting short-term electricity demand using technical indicators

Abstract

Electricity demand exhibits complex short-term behavioural and temporal dynamics that are increasingly important for operational planning in modern power systems, particularly in developing regions undergoing rapid renewable-energy expansion. This study introduces a data-driven framework that applies technical indicators, adapted from high-frequency financial time-series analysis, to extract trend, momentum and volatility features from high-resolution national electricity demand. Using one year of 15-minute data from Sri Lanka, the framework integrates engineered indicators with gradient-boosting models to identify latent demand structures that are not visible in raw load curves. The results show that momentum- and acceleration-based indicators offer the strongest operational value, with ablation tests revealing accuracy deteriorations exceeding 40% when these features are removed. The model achieved an R² of 0.846 and an overall MAPE of 6.1%, accurately capturing morning ramps, mid-day stabilisation and evening peaks. Forecast deviations during culturally driven events highlight the need for behaviour-sensitive features in developing grids. The extracted demand patterns also reveal operational windows with high potential for storage charging (mid-day) and strategic discharging (evening peaks), demonstrating applications for battery energy-storage optimisation and renewable-integration planning. By showing that finance-inspired indicators enhance both interpretability and predictive performance, this study provides a replicable methodology for grid operators seeking low-cost, data-driven tools for short-term decision support. The framework offers actionable insights for generation scheduling, reserve planning, demand-response design and the efficient utilisation of storage assets in emerging, renewables-constrained power systems.