کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Short-Term Electric Load Forecasting Using LSTM with a Variance-Aware Loss and Chebyshev-Based Prediction Intervals

نویسندگان :

Aria Nodehi Moghadam¹ Reza Shahriari Beni² Mohammad Amin Pourmousavi³ Mahdi Aliyari Shooredeli⁴

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه صنعتی خواجه نصیرالدین طوسی 3- دانشگاه صنعتی خواجه نصیرالدین طوسی 4- دانشگاه صنعتی خواجه نصیرالدین طوسی

کلمات کلیدی :

Short-Term Load Forecasting،Chebyshev’s Theorem،Long Short-Term Memory (LSTM)،Time-Series Prediction،Variance-Aware Loss Function،Prediction Intervals

چکیده :

This paper addresses the problem of short-term energy load forecasting, key component in the efficient operation and management of electric power systems. Accurate and reliable load forecasts support effective resource planning, enhance operational stability, and reduce associated costs. In this study, historical energy consumption data are used as inputs to deep learning models to capture temporal patterns in load behavior and generate precise short-term forecasts. Due to the nonlinear and dynamic nature of energy consumption, traditional statistical approaches often struggle to achieve satisfactory performance. To address these limitations, several Long Short-Term Memory (LSTM) architectures are employed to model temporal dependencies in load time series. To further improve forecasting performance, a novel variance-aware loss function is introduced that jointly minimizes the mean prediction error and the dispersion of forecasting errors, leading to enhanced accuracy and temporal stability. In addition, prediction intervals are constructed using a Chebyshev-theorem-based approach combined with kernel density estimation, enabling the quantification of forecast uncertainty. Experimental results demonstrate that the proposed framework reduces both prediction error and error variability compared to conventional loss functions, while producing more stable and reliable forecasts. These characteristics are particularly important for short-term energy load forecasting applications in power systems operating under variable and peak load conditions.