Accurate transformer oil-temperature forecasting is important for thermal-risk assessment and operational planning. However, reported gains from complex forecasting models may be affected by future information leakage, weak seasonal baselines, inconsistent target periods, and test-based model selection. This study establishes a leakage-free, target-aligned framework for direct forecasting at 6, 12, and 24 h, integrating controlled model comparison, input-utility analysis, exact temporal interpretation, and cross-dataset confirmation. Only information available at or before the forecast origin is used, and identical validation and test target periods are maintained across horizons and lookback lengths. Naive predictors, regularized autoregression, ensemble methods, XGBoost variants, deep sequence models, and linear-nonlinear hybrids are evaluated using expanding-window validation and moving-block bootstrap analysis. OT-only Ridge regression with a 72 h lookback and $\lambda = 10^{-4}$ was selected for all three horizons, achieving ETTh2 RMSEs of 3.1568, 4.0042, and 4.2920. After ETTh1-specific refitting, the corresponding RMSEs were 1.3701, 1.7532, and 2.0976. Ridge significantly outperformed the daily-seasonal baseline at 6 and 12 h, while the 24 h gain was not statistically distinguishishable. The six historical load channels provided no robust incremental value. Exact Ridge contributions showed a shift from recent thermal persistence at 6 h to dominant daily-cycle dependence at 24 h. Rapid cooling was overpredicted and rapid heating was underpredicted. Overall, increased model complexity did not provide a consistent advantage under a controlled protocol, while the combined evaluation, interpretation, and cross-dataset confirmation offer reproducible empirical guidance for transformer oil-temperature forecasting.
This study provides an in-depth comparative analysis of four state-of-the-art neural architectures, confirming that high-fidelity point forecasts and rigorously quantified uncertainty can be achieved simultaneously, providing a clear path toward more dependable PV dispatch, reserve allocation, and market participation.
Saloni Dhingra, G. Gruosso, G. Storti Gajani· Neural computing & applicati...· 0 citations
Very Short-Term Load Forecasting (VSTLF) is critical for real-time grid stability, frequency control, and economic dispatch. This study proposes a Gaussian Process Regression (GPR)-based framework for one-hour-ahead load forecasting using hourly data from January 2020 to April 2024 for Delhi, India. The model incorporates meteorological data such as temperature, humidity, and dew point with lagged load values. The research takes into account time-related dependencies and seasonal changes in order to boost the predictive power of the suggested model. Unlike deterministic neural models, GPR provides probabilistic predictions along with uncertainty quantification. Multiple kernel configurations were evaluated across datasets of increasing size (6,000–30,000 samples). The best- performing configuration (Exponential kernel, 25,000 samples) achieved a Testing RMSE of 111.43 MW, MAPE of 2.5349%, MAE of 77.99 MW, and R² of 0.9841. The evaluation highlights the model’s strength when faced with different data sizes and its capacity to deliver stable performance with little overfitting. Results demonstrate that GPR provides stable, accurate, and interpretable forecasting suitable for operational power system applications. The proposed framework presents substantial benefits regarding reliability, scalability, and adaptability for real-time implementation in contemporary smart grid settings, facilitating effective decision-making and enhanced energy management strategies. Adding uncertainty bounds to the mix bolsters operator confidence by facilitating planning that takes risk into account and management of the grid that anticipates problems.
Karan Sati, A. Yadav· Trends in Electrical Enginee...· 0 citations
Electricity demand forecasting is crucial for effective grid operation, planning, and decision-making. This study presents a comparison between classical time series models—AutoRegressive Integrated Moving Average (ARIMA) and Seasonal AutoRegressive Integrated Moving Average (SARIMA)—and a deep learning–based method, Gated Recurrent Units (GRU), for forecasting hourly electricity demand. The models are tested on a real-world dataset, enriched with weather and calendar variables to capture temporal and exogenous effects on electricity consumption. To ensure a fair and reproducible comparison, all models are trained and evaluated in a common experimental framework, including a well-defined chronological train-test split and rolling-origin (walk-forward) validation strategy. The forecasting performance is evaluated for short-term (24 h) and medium-term (168 h) horizons using standard error metrics, namely, root mean squared error, mean absolute error, and Mean Absolute Percentage Error (MAPE). The results of the experiment demonstrate that the GRU model performs better than ARIMA and SARIMA models especially for longer forecasting horizons due to its capability to learn nonlinear relationships and long-term temporal dependencies. The GRU approach gives better forecasting accuracy in the case of complex demand dynamics, but linear seasonal patterns can still be modeled by classical statistical models. The aim of this study is not to directly detect or predict system failures, nor does it depend on explicit fault or outage data. Its main contribution is instead in improving the accuracy of electricity demand forecasting, which can indirectly assist preventive grid operation and planning by reducing the uncertainty in expected load profiles.
Received: 7 August 2025 | Revised: 20 March 2026 | Accepted: 23 June 2026
Conflicts of Interest
The authors declare that they have no conflicts of interest to this work.
Data Availability Statement
The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/saurabhshahane/electricity-load-forecasting.
Author Contribution Statement
Emrah Aslan: Conceptualization, Methodology, Software, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration. Yıldırım Özüpak: Conceptualization, Methodology, Software, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision. Feyyaz Alpsalaz: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision. Hasan Uzel: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision.
Emrah Aslan, Yıldırım Özüpak, Feyyaz Alpsalaz et al.· Artificial Intelligence and...· 0 citations
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.
Timur Lale· 2026 6th International Confe...· 0 citations
This study develops a rigorous, leakage-free forecasting framework for monthly Robusta coffee prices using historical observations from January 1975 to December 2025. A comprehensive set of explanatory variables is constructed from lagged coffee prices, moving averages, logarithmic returns, rolling volatility, and exogenous variables such as the Oceanic Niño Index (ONI), the U.S. Dollar Index, and Brent crude oil prices. To ensure methodological fairness, all predictors are generated exclusively from information available at the forecast origin, and all competing models are evaluated under a unified expanding-window walk-forward validation framework. Seven forecasting models are compared: Naïve, Exponential Smoothing (ETS), ARIMA, ARIMAX, Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). Forecasting performance is evaluated using R2, RMSE, MAE, and MAPE, while Taylor diagrams and the Diebold–Mariano test are employed to assess model agreement and differences in predictive accuracy. The results show that XGBoost achieves the highest forecasting accuracy (R2 = 0.956, RMSE = 0.264), followed closely by the Naïve (R2 = 0.954, RMSE = 0.271) and ARIMA (R2 = 0.954, RMSE = 0.270) benchmarks, whereas ARIMAX and ETS provide comparable performance and the deep learning models (LSTM and GRU) produce substantially larger prediction errors. Feature importance analysis further indicates that the first lag of coffee price is the dominant predictor, accounting for approximately 94% of the predictive gain in XGBoost. Overall, the findings demonstrate that rigorous leakage-free validation is essential for reliable forecasting research and that, for monthly Robusta coffee prices, increased model complexity does not necessarily yield superior predictive performance.
Dler H Kadir, D. Khalil, Azhin M. Khudhur· Forecasting· 0 citations
Commodity price forecasting plays a crucial role in international trade, agriculture, investment, and economic decision-making. However, accurate prediction remains challenging due to market volatility, economic uncertainty, geopolitical events, climate change, and supply chain disruptions. This study proposes a comprehensive time series forecasting framework that integrates classical statistical models, including ARIMA, SARIMA, and Exponential Smoothing, with machine learning and deep learning techniques such as Random Forest Regression, Support Vector Regression, Gradient Boosting, and Long Short-Term Memory (LSTM). The framework incorporates data preprocessing, feature engineering, trend decomposition, stationarity testing, model optimization, and rolling window validation to improve forecasting performance. Model evaluation is conducted using MAE, RMSE, MAPE, SMAPE, and R² metrics. Experimental results demonstrate that hybrid forecasting models outperform conventional statistical approaches by effectively capturing nonlinear temporal patterns and improving prediction accuracy under volatile market conditions. The proposed framework offers a scalable, interpretable, and adaptable solution for commodity price forecasting, supporting strategic decision-making, risk management, inventory optimization, and investment planning across diverse commodity sectors.
N. Karmarkar· International Journal of Com...· 0 citations
