Learning-based probabilistic load forecasting with post-hoc and in-model uncertainty

Probabilistic net load forecasting using sensor-driven data and dual-coefficient network

With the widespread deployment of sensor networks in modern energy systems, large-scale time-series data from distributed sources (e.g., load demand, photovoltaic generation, and wind power) provide new opportunities for intelligent system perception and predictive modeling. Accurate and uncertainty-aware net load modeling plays a critical role in supporting fault diagnosis, risk assessment, and fault-tolerant control. To overcome the deficiencies of current approaches in representing uncertainty, distribution shift, and quantile crossing, the present study develops a novel probabilistic framework for net load forecasting. It employs Dual-coefficient network (DUCET) to model both input-space and output-space distribution characteristics, thereby enhancing robustness against distribution shifts. Furthermore, a quantile loss–driven Informer architecture is adopted for modeling temporal dependencies. It can also generate probabilistic forecasts through conditional quantiles. To resolve the quantile crossing problem commonly observed in quantile regression, a quantile reconstruction strategy (QRS) is introduced, which reconstructs a coherent predictive distribution using kernel density estimation and Gaussian approximation. Experimental results on real-world net load data from Austria show that the proposed probabilistic framework surpasses benchmark methods. Additional ablation experiments confirm the validity of DUCET and QRS in improving forecasting accuracy and reliability.

Multi-Horizon Transformer Oil-Temperature Forecasting: Temporal Dependence, Load-Variable Utility, and Model Complexity

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.

Efficacy of historical context and exogenous features on deep learning for cooling load forecasting in chilled water plants.

Accurate day-ahead cooling load forecasting is a time series forecasting problem that is essential for optimizing the scheduling and energy efficiency of chilled water plants (CWPs). While previous studies have explored different forecasting models and input features, the combined effects of key factors such as historical context length (including regressor lags), sampling resolution, and exogenous inputs remain insufficiently examined for real industrial data. This study examines, under varying regressors and data granularity, the amount of historical context that provides stable and reliable forecasts with low computational cost. The experimental results show that (i) weekly scale look-back windows (LBW) provide an optimal balance between accuracy and computational cost for day-ahead horizons, (ii) hourly inputs often match or exceed finer-resolution accuracy and reveal distinct compute scaling: deep-learning (DL) models remain comparatively efficient as resolution increases, whereas the machine learning (ML) model, XGBoost, grows steeply, (iii) incorporating weather and calendar features enhances prediction accuracy around weekends and holidays. The DL model NHiTS, when integrating exogenous features and a 7-day LBW, improves forecasting accuracy by 50.8%, outperforming all other deep learning, machine learning, and statistical forecasting models by achieving a MASE of 1.88, compared to 3.82 for the baseline. These findings provide practical insights for smart-building and industrial applications, guiding practitioners such as building-energy engineers and facility managers in selecting optimal historical data lengths and parameter combinations to improve forecasting accuracy.

An AI-Based Decision-Support Pipeline for Day-Ahead Photovoltaic Forecasting

Reliable photovoltaic (PV) forecasts are needed for low-carbon energy systems, but newly deployed sites often have short, imperfect records. This makes standard day-ahead forecasting difficult: persistence and physical baselines can be sensitive to calibration and timestamp alignment, while single machine-learning models may capture only one structure in the data and overstate skill under non-temporal validation. We study this problem at a United Kingdom charging-station site, where PV forecast errors affect charging availability, storage scheduling, and downstream control. Using measured inverter output and publicly available meteorological inputs, we develop a deployment-oriented environmental-AI pipeline for day-ahead hourly PV forecasting. The pipeline corrects timestamp conventions, constructs leakage-safe solar-geometry and clearness-index features, adds short-term atmospheric context, and combines complementary predictors through validation-learned stacking. Against smart persistence, a clear-sky baseline that adjusts recent PV output using expected clear-sky irradiance, the best ensemble reduces daylight normalised RMSE by about 32% under random day-blocked evaluation and 9% under the stricter rolling-origin protocol. It also reduces daylight RMSE relative to the strongest individual machine-learning baseline by 6.6% and 6.4%, respectively. The results show that physics-aware stacking can support PV forecasts from limited site data, but its value depends on model class, evaluation protocol, and deployment context.

A Machine Learning-Based Probabilistic Electricity Load Forecasting Method under Extreme Weather Conditions

In the energy transition of the world, the models to be used in power load prediction should be capable of delivering predictions that are not only accurate but also have a reasonable measure of uncertainty. The increase in the number of extreme weather events has caused the behavior of the loads to be nonlinear and unpredictable and this has restricted the effectiveness of the traditional deterministic forecasting approach in grid dispatching as well as warning of risk. To address pattern identification, data sparsity, and uncertainty under extreme weather, this paper develops an integrated probabilistic forecasting framework with three linked stages: extreme-weather load identification, TimeGAN-based sample augmentation, and conformal quantile forecasting. The method first builds a high-confidence extreme-weather load sample repository, then augments scarce extreme-weather sequences, and finally provides calibrated prediction intervals for short-term load forecasting. Experimental results show that the proposed method improves forecasting accuracy under extreme-weather conditions, with MAPE reduced across all five tested models after data augmentation; for example, ARIMA decreases from 12.37% to 8.65% and iTransformer decreases from 6.12% to 5.28%. The conformal quantile forecasting model also achieves 97.62% empirical coverage under the nominal 95% prediction interval, indicating improved prediction-interval reliability.

LSTM–GRU Hybrid Deep Learning Model for Short-Term Load Forecasting in Smart Grid Systems