کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

A Hybrid CNN–Residual LSTM–Transformer Framework for Multi-Output SCADA Measurement Recovery in Modern Power Systems

نویسندگان :

Amir Tanoumand¹ Habib Rajabi Mashhadi² Sajjad Ahmadnia³

1- Ferdowsi University of Mashhad 2- Ferdowsi University of Mashhad 3- Ferdowsi University of Mashhad

کلمات کلیدی :

supervisory control and data acquisition (SCADA)،multi-output forecasting،short-term load prediction،hybrid stacked temporal deep network (HSTDN)،convolutional neural networks (CNN)،residual stacked long short-term memory (LSTM)

چکیده :

Accurate Supervisory Control and Data Acquisition (SCADA) readings are essential for monitoring modern power grids; however, missing samples and irregular fluctuations often reduce system observability. This paper presents a hybrid deep learning framework, termed Hybrid Stacked Temporal Deep Network (HSTDN), for SCADA data reconstruction and short-term multi-output forecasting of SCADA variables. The architecture integrates Convolutional Neural Networks (CNNs) for local feature extraction, a Residual Stacked Long Short-Term Memory (LSTM) module for temporal dependency modeling, and a lightweight Transformer to capture long-range interactions. The proposed approach is evaluated using the IEEE Reliability Test System and real SCADA measurements. Experimental results show that the HSTDN achieves a mean absolute percentage error 0.39\% across SCADA output active power, outperforming several benchmark deep learning models. In addition, the unified multi-output structure provides significantly faster inference compared to training and deploying separate models for each variable, making the proposed method suitable for real-time SCADA monitoring and data recovery applications.