A Bibliometric Review and Taxonomy of Artificial Intelligence Applications in Electricity Load Forecasting

This study presents a comprehensive bibliometric review and taxonomy of 469 peer-reviewed publications on AI-based electricity load forecasting from 1997 to 2024. Through performance analysis and science mapping, the paper identifies influential authors, emerging countries, research clusters, and collaboration patterns. A three-layer taxonomy is developed to categorize the literature based on forecasting horizon (ultra-short, short, mid, long term), forecasting scale (micro, meso, macro), and model type (basic machine learning, basic and advanced deep learning, hybrid, soft computing). Results reveal that short-term forecasting at micro and meso levels dominates the field, with deep learning models - especially Convolutional Neural Network and Long Short-Term Memory - playing a central role. Hybrid models integrating optimization and decomposition techniques are gaining popularity, while ultra-short, mid-, and long-term forecasting remain underexplored. Collaboration is still regionally fragmented, but there is a noticeable shift of research influence toward South Asia. The study provides a structured overview of the intellectual landscape and highlights key gaps, including the lack of long-horizon forecasts and the need for more generalized and collaborative AI solutions across spatial levels. Research purpose: This study provides a bibliometric review and taxonomy of AI applications in electricity load forecasting. It aims to systematically classify 469 publications (1997-2024) by forecasting horizon, application scale, and model category, while identifying influential authors, countries, and emerging trends. Research motivation: Electricity load forecasting is essential for reliable system operation, market efficiency, and energy transition planning. The increasing complexity of power systems with renewables and electric vehicles requires more advanced models. Despite rapid progress, there is still a lack of structured overviews to clarify research gaps, motivating this study. Research design, approach, and method: Data was retrieved from the Scopus database using a targeted search string combining electricity load forecasting with AI-related terms. Bibliometric analysis was applied through performance indicators and science mapping (co-authorship, co-keyword networks). A three-layer taxonomy was developed covering forecasting horizon, scale, and model type. Main findings: Short-term forecasting dominates the field, especially at micro and meso levels, using deep learning models such as Convolutional Neural Network and Long Short-Term Memory. Hybrid models integrating decomposition and optimization methods are increasingly popular, while ultra-, short-, mid-, and long-term horizons remain underexplored. China leads research output, followed by India and Pakistan, signalling a regional shift toward South Asia. Practical/managerial implications: The taxonomy and mapping provide a structured reference for scholars and practitioners. Identifying gaps-such as the lack of long-term and multi-level forecasting-can guide future research. For utilities and policymakers, the findings highlight the practical relevance of deep learning and hybrid AI models in improving system reliability and energy management.