Aug 2026· 2026 6th International Conference on Emerging Smart Technologies and Applications (eSmarTA)· pp. 1-8· 0 citations· 28 references
Abstract
Short-term load forecasting (STLF) is an essential task for reliable power system operation, economic dispatch, reserve scheduling, and grid planning. This study aims to provide an operationally realistic and interpretable comparison of five ensemble tree-based machine learning (ML) models for national electricity demand forecasting using the publicly available Panama Short-Term Electricity Load Forecasting dataset. Gradient Boosting Regressor (GBR), XGBoost, LightGBM, CatBoost, and Random Forest are evaluated using 14 predefined walk-forward train–test splits that emulate the weekly forecasting protocol of Panama’s national grid operator. A common feature set consisting of lagged demand variables, a four-week moving average, temporal indicators, calendar variables, and Tocumen temperature is used for all models. A seasonal naive baseline, statistical significance testing, COVID-period split analysis, and feature importance comparison are also included. CatBoost achieved the best average performance with an RMSE of 55.52 MWh and MAPE of 3.80%, outperforming the seasonal naive baseline, which obtained an RMSE of 78.61 MWh. However, Wilcoxon-Holm testing showed that the narrow RMSE differences among the ensemble models were not statistically significant at the 5% level. Feature importance analysis confirmed that the four-week moving average is a dominant predictor for most models. The results show that ensemble tree-based models provide accurate, robust, and interpretable STLF performance under an operationally realistic evaluation protocol.
Accurate one-hour-ahead electricity load forecasting sustains dispatch, reserve planning, and dependable power-system operation, but additional inputs do not always improve predictions. This study assesses how different parameter settings contribute to power demand forecasting. Four long short-term models with a unified architecture were trained on 48048 hourly observations and a 24-hour window. Input configurations include historical demand alone, demand with twelve weather variables, demand with two calendar indicators, or all inputs. Each configuration was trained five times and evaluated on a chronological test set. The weather-based model achieved the lowest mean errors: a mean absolute percentage error of 1.599% and a root mean square error of 26.333. The history-only model remained competitive, while calendar indicators and the all-input model offered no improvement. Recent demand therefore provides most of the useful information at this horizon, with lagged weather adding a modest signal. Careful feature selection can reduce complexity and support interpretable, dependable operational forecasts.
Jingkai Gao· Applied and Computational En...· 0 citations
Accurate short-term forecasting of electricity loads is crucial for power system operations in regions with large fluctuations in daily energy demand caused by climate variability. This study presents a next-day electricity load forecasting for Tirana, Albania, by combining temperature-based indicators, Cooling Degree Days and Heating Degree Days in a multi- input-output modeling framework. The forecasting of daily electricity load and degree day variables was carried out using several forecasting methods, including Ridge Regression, Support Vector Regression, Random Forest, Gradient Boosting and Artificial Neural Networks, over training and testing scenarios based on the daily data collected from 2020 to 2022. The forecasting framework includes short-term temporal and temperature-dependent dynamics using lagged electricity demand and degree-day indicators. The model performance was evaluated under three different seasonal scenarios: August (high cooling demand), October (neutral transition conditions), and November (high heating demand). The results indicate that under neutral conditions, all models exhibit similar performance, with an average MAPE of approximately 2.7% in October. Random Forest has the best load forecasting accuracy in August (MAPE= 4.07%), and Support Vector Regression has the best load forecasting accuracy in November (MAPE= 3.27%), followed by the tree- based ensemble methods. The results show that forecasting performance is season-dependent, with different models achieving the highest accuracy under different climatic conditions. Rather than identifying a single globally optimal model, the findings highlight the importance of regime-specific model behavior in electricity demand forecasting. This suggests that model selection should be aligned with seasonal characteristics of the load, particularly in temperature-sensitive systems. The findings provide useful information for short-term operational planning in the Albanian electricity sector.
Agresa Qosja· European Journal of Energy R...· 0 citations
The experimental findings indicate that Linear Regression (LR) model is better than the Artificial Neural Network (ANN) model because it has a small Root Mean Square Error (RMSE), which means that the underlying data set is more linear in nature and in this case, simpler models can be more effective than the more complex nonlinear models.
Shorya Mittal, N. Saxena, K. Gandhi et al.· Journal of Electrical System...· 0 citations
The empirical results indicate that tree-based ensemble models outperform deep learning approaches under the considered dataset conditions and suggest that ensemble machine learning techniques are more suitable for data-constrained environments, where deep learning models may suffer from overfitting.
Lamiaa Fares, Moad El Kharrim, Mohamed Dakkon· 0 citations
The forecasting of short-term load (STLF) is an important part of the functioning of a power system, as it aids in discovering the dispatch plans and taking the strain off. This paper researches hourly short-term electricity load forecasting of the Panama power system based on a univariate Autoregressive Integrated Moving Average (ARIMA) model as a clean statistical reference. The national demand in 20152020 was chosen as the dataset obtained in the publicly available Kaggle repository. It was tested on a 24-hour-ahead forecasting problem with a train-test split that was based on chronology. The ARIMA (2,1,2) model gained the reliability of the held-out test window of 48.88 MWh, RMSE of 60.50 MWh, and MAPE of 4.32%. Although this error rate is widely similar to classical statistical thresholds in short-horizon load forecasting papers, it is nevertheless significantly larger than the accuracy commonly found by the more advanced machine learning and deep learning models, specifically when using exogenous variables. The results hence result in ARIMA being an efficient base of clarity and benchmarking, with the development being more expressive models to route the non-linearities in electricity demand.
Three machine learning models, standalone XGBoost, Long Short-Term Memory (LSTM), and a hybrid model combining LSTM and XGBoost were systematically compared to two naive models: persistence and seasonal benchmark models to illustrate the need for selecting the most appropriate model based on data to accurately predict energy consumption in smart homes.
