[Paper Review] Electrical peak demand forecasting- A review

Shuang Dai, Fanlin Meng+3 more|arXiv (Cornell University)|Aug 3, 2021

Energy Load and Power Forecasting155 references4 citations

TL;DR

This paper presents a comprehensive review of electrical peak demand forecasting methods, categorizing 139 studies into three developmental stages and analyzing techniques like machine learning, deep learning, and clustering. It identifies key challenges such as data privacy and model interpretability, and advocates for federated learning and domain knowledge integration to improve forecast accuracy in smart grids.

ABSTRACT

The power system is undergoing rapid evolution with the roll-out of advanced metering infrastructure and local energy applications (e.g. electric vehicles) as well as the increasing penetration of intermittent renewable energy at both transmission and distribution level, which characterizes the peak load demand with stronger randomness and less predictability and therefore poses a threat to the power grid security. Since storing large quantities of electricity to satisfy load demand is neither economically nor environmentally friendly, effective peak demand management strategies and reliable peak load forecast methods become essential for optimizing the power system operations. To this end, this paper provides a timely and comprehensive overview of peak load demand forecast methods in the literature. To our best knowledge, this is the first comprehensive review on such topic. In this paper we first give a precise and unified problem definition of peak load demand forecast. Second, 139 papers on peak load forecast methods were systematically reviewed where methods were classified into different stages based on the timeline. Thirdly, a comparative analysis of peak load forecast methods are summarized and different optimizing methods to improve the forecast performance are discussed. The paper ends with a comprehensive summary of the reviewed papers and a discussion of potential future research directions.

Motivation & Objective

• To provide a unified and precise problem definition for peak load demand forecasting in modern power systems.
• To systematically review 139 peer-reviewed studies on peak load forecasting methods, organized by chronological development stages.
• To compare the performance of various forecasting models, including traditional statistical and modern machine learning approaches.
• To identify key challenges such as data privacy, model interpretability, and the impact of extreme weather on forecast accuracy.
• To propose future research directions, including federated learning and integration of domain knowledge in forecasting frameworks.

Proposed method

• The authors conducted a systematic literature review of 139 papers on peak load forecasting, classified by historical development stages: classic, advanced, and modern.
• They categorized forecasting methods based on data types (e.g., time series, climatic, economic) and model architectures (e.g., ARIMA, SVM, LSTM, XGBoost).
• The study evaluates performance-enhancing techniques such as clustering, feature engineering, and hybrid modeling to improve forecast accuracy.
• The review emphasizes the role of high-resolution smart meter data and the need for privacy-preserving methods like federated learning in multi-source data integration.
• Domain knowledge is identified as critical for interpreting clustering-based load patterns, especially at the local (e.g., community) level.
• The framework integrates stakeholder perspectives (grid operators, retailers, end-users) to align forecasting goals with real-world operational needs.

Experimental results

Research questions

• RQ1 What are the key factors that influence peak load demand, and how can they be consistently defined in a forecasting framework?
• RQ2 How have peak load forecasting methods evolved over time, and what distinguishes the classic, advanced, and modern stages of development?
• RQ3 What are the most effective machine learning and deep learning techniques for improving peak load forecast accuracy?
• RQ4 How do extreme weather events and human behavior impact forecast performance, especially at the disaggregated (e.g., community) level?
• RQ5 What are the major challenges in data privacy and multi-organization data sharing, and how can federated learning support secure forecasting?

Key findings

• Peak load forecasting is critical for grid security, cost reduction, and enabling demand response strategies across all electricity market stakeholders.
• The integration of climatic variables such as maximum/minimum temperature and humidity significantly improves forecast accuracy, especially during extreme weather events.
• Clustering-based methods show promise but heavily rely on domain knowledge for meaningful interpretation, highlighting a need for better integration of contextual information.
• At the local level (e.g., community or residential), human behavior introduces higher randomness, making disaggregated forecasting more challenging than aggregated regional forecasting.
• Data privacy and security are major concerns when combining data from multiple organizations; federated learning is identified as a key future direction for secure, privacy-preserving model training.
• Despite advances, no single model universally outperforms others across all conditions—model selection must be context-specific, considering data quality, resolution, and forecasting horizon.