<![CDATA[Transformasi sistem tenaga listrik menuju konsep smart grid menuntut kemampuan prediksi beban listrik yang akurat, adaptif, dan andal guna mendukung pengelolaan energi secara real-time. Seiring meningkatnya kompleksitas sistem akibat integrasi energi terbarukan, perangkat Internet of Things (IoT), serta dinamika konsumsi energi, pendekatan berbasis machine learning menjadi solusi yang semakin banyak diadopsi dibandingkan metode statistik konvensional. Penelitian ini bertujuan untuk memetakan perkembangan, mengklasifikasikan metode, serta menganalisis performa penerapan machine learning dalam prediksi beban listrik pada sistem smart grid melalui pendekatan Systematic Literature Review (SLR). Metode penelitian mengacu pada pedoman PRISMA 2020 dengan tahapan perumusan research question, strategi pencarian literatur, penerapan kriteria inklusi dan eksklusi, serta proses ekstraksi dan sintesis data. Pencarian dilakukan pada basis data ilmiah terindeks dengan rentang publikasi 2016–2025. Berdasarkan proses seleksi, diperoleh sembilan artikel yang memenuhi kriteria dan dianalisis menggunakan pendekatan deskriptif dan komparatif untuk mengidentifikasi tren metodologi dan perbandingan performa. Hasil kajian menunjukkan bahwa metode Long Short-Term Memory (LSTM) merupakan algoritma yang paling dominan digunakan karena kemampuannya memodelkan data deret waktu dan pola nonlinier. Tren terbaru mengarah pada model hibrida dan arsitektur transformer yang menawarkan peningkatan akurasi dan stabilitas. Evaluasi performa umumnya menggunakan metrik MAE, RMSE, dan MAPE, dengan pendekatan deep learning menunjukkan hasil lebih unggul dibandingkan metode tradisional. Tantangan utama meliputi kualitas data, kompleksitas komputasi, generalisasi antarwilayah, dan interpretabilitas model. Penelitian selanjutnya direkomendasikan untuk mengembangkan model yang lebih efisien, adaptif, dan terintegrasi dengan sistem manajemen energi berbasis IoT serta real-time analytics. The transformation of electric power systems toward the smart grid concept requires accurate, adaptive, and reliable load forecasting capabilities to support real-time energy management. As system complexity increases due to the integration of renewable energy, Internet of Things (IoT) devices, and dynamic energy consumption patterns, machine learning-based approaches are increasingly adopted compared to conventional statistical methods. This study aims to map developments, classify methods, and analyze the performance of machine learning applications in electric load forecasting within smart grid systems through a Systematic Literature Review (SLR) approach. The research method follows the PRISMA 2020 guidelines, including the formulation of research questions, literature search strategy, inclusion and exclusion criteria, and data extraction and synthesis processes. The literature search was conducted across indexed scientific databases covering publications from 2016 to 2025. Based on the selection process, nine articles met the criteria and were analyzed using descriptive and comparative approaches to identify methodological trends and performance comparisons. The findings indicate that Long Short-Term Memory (LSTM) is the most dominant algorithm due to its capability to model time-series data and nonlinear patterns. Recent trends point toward hybrid models and transformer architectures that offer improved accuracy and stability. Performance evaluation commonly employs MAE, RMSE, and MAPE metrics, with deep learning approaches generally outperforming traditional methods. Key challenges include data quality, computational complexity, cross-regional generalization, and model interpretability. Future research is recommended to develop more efficient, adaptive models integrated with IoT-based energy management systems and real-time analytics. ]]>
Copyrights © 2026
