رحمن دشتی

Accurate short-term load forecasting (STLF) is essential for the reliable operation of power systems, electricity markets, and generation scheduling. Nevertheless, the nonlinear and uncertain nature of electric load limits the effectiveness of conventional deterministic forecasting models. To address this challenge, this paper proposes a Transformer-inspired Quantile Long Short-Term Memory (TFT-QLSTM) model for hour-ahead load forecasting. The proposed framework integrates feature gating and temporal self-attention mechanisms with a quantilebased LSTM to capture temporal dependencies while explicitly modeling forecasting uncertainty. The performance of the proposed model is evaluated using the Panama hourly electricity load dataset. A controlled comparison with a TFT-LSTM baseline demonstrates that the proposed TFT-QLSTM achieves superior forecasting accuracy, recording an RMSE of 27.73 kW and a MAPE of 1.63a paired t-test confirms that the observed improvement is statistically significant. The results indicate that quantile-based learning within a TFT-inspired architecture enhances robustness and accuracy in short-term load forecasting applications.