Anantha Krishna Kamath
Department of Computer Science and Design Engineering, Canara Engineering College, Mangalore, Visvesvaraya Technological University, Karnataka, India.

B. L. Rajalakshmi Samaga
Department of Electrical and Electronics Engineering, Nitte (Deemed to be University), NMAM Institute of Technology (NMAMIT), Nitte, Karnataka, India.

G. K. Dayananda
Department of Electronics and Communication Engineering, Canara Engineering College, Mangalore, Visvesvaraya Technological University, Karnataka, India.

DOI https://doi.org/10.33889/IJMEMS.2025.10.6.105

  ;

Abstract

The operation of electric power system is a continuous process which demands coordination of various entities from power generating plants to distributing substations to render uninterrupted service still sticking to quality power delivery. Electric demand depends on external factors such as temperature, humidity, social activity pattern. Power grids are becoming complex due to integration of renewable energy sources. Thus, there is a need of electric energy forecasting. Efficiency of traditional forecasting approaches are less and existing many learning and ensembled models require high computational resources. To improve accuracy of prediction, a fast and efficient processing model which is suitable for real-time applications, light weight ensemble model is proposed in this paper. This research proposes a novel stacked light weight ensemble model that integrates the prowess of various weak base learners. The final prediction of the model is further improved by using extreme gradient boosting as a meta learner, which evolutionarily learns the predictions from individual learners and gives the final load forecast. Further the temporal nature of the exogeneous variables is preserved by a unique feature fusion technique which estimates the exponentially weighted moving average of the individual variable which are then aggregated. The efficacy of this model is validated by testing it on Panama electricity load forecasting dataset and the results are explored using important regression-based metrics. The analysis shows that the proposed method can vividly forecast the electricity load using the lightweight ensemble model in terms of Root Mean Square Error (RMSE), Mean Bias Error (MBE), Mean Absolute Error (MAE) and R2 values.

Keywords- Feature fusion, Exponentially weighted moving average, Short term load forecasting, Ensemble model, Meta learner.

Citation

Kamath, A. K. Samaga, B. L. R. & Dayananda, G. K. (2025). Light Weight Stacked Ensemble Model for Electric Load Forecasting using Fused Health Index. International Journal of Mathematical, Engineering and Management Sciences, 10(6), 2268-2285. https://doi.org/10.33889/IJMEMS.2025.10.6.105.