Ensemble Machine Learning Framework for Multi-Horizon Electricity Demand Forecasting Across Indian Grid Regions

DOI

10.2991/978-94-6239-791-0_25How to use a DOI?

Keywords

Load Forecasting; Ensemble Learning; LSTM; Transformer Networks; Xgboost; Time-Series Forecasting; Smart Grid Analytics; India Electricity System

Abstract

Forecasting electricity load is an important issue in power system planning, stability and resources allocation in developing countries. This paper deals with multi-horizon forecasting across geographically diverse Indian grid regions by proposing an ensemble architecture, which merges long short-term memory (LSTM) networks with the transformer based sequence models and gradient boosting (XGBoost). The raw data includes past electricity usage, climatic data, time definitions, and socio-economic factors for five grid sectors, which were systematically featured and normalized across time. The methodology uses a chronologically-aware train-test split that keeps the temporal dependency in place, and is evaluated for 24-hour and 30-day forecasting time horizons. The ensemble mechanism is verified optimized weight updating to integrate heterogeneous models: LSTM provides 42.5% of the ensemble weight because of its excellent temporal modelling performance, while transformer (31.2%) and XGBoost (26.3%) provide attention-based representations and nonlinear feature interactions respectively. The performance of the weighted ensemble shows that it successfully reaches 0.941 of accuracy with 5.9% mean absolute error percentage (MAPE) and R2 value of 0.941, which is much higher than that of other models. The framework works especially well to represent the non-linear relationship between temperature demand and keep forecast stability throughout the seasonal transition period. This paper lays the groundwork for repeatable operational demand forecasting in complex power systems and offers real-world examples, insights and advice for the design of ensembles for energy analytics applications.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Medha Wyawahare
AU  - Arush Ambavale
AU  - Krishna Ardhapurkar
AU  - Deven Bagul
PY  - 2026
DA  - 2026/10/09
TI  - Ensemble Machine Learning Framework for Multi-Horizon Electricity Demand Forecasting Across Indian Grid Regions
BT  - Proceedings of the International Conference on Artificial Intelligence and Machine Learning for Agro-Ecosystems (ICAIML-AE 2026)
PB  - Atlantis Press
SP  - 321
EP  - 348
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-791-0_25
DO  - 10.2991/978-94-6239-791-0_25
ID  - Wyawahare2026
ER  -