Global Review Uncovers Stark Methodological Divides in Electricity Demand Forecasting
A comprehensive systematic review of 107 peer-reviewed studies published between 2005 and 2023 reveals profound structural and methodological divides in how electricity demand is modeled across developed and developing economies. Published in Energy Informatics by researchers Zara Omar and Liwan Liyanage, the investigation compares forecasting practices across Pakistan, Sri Lanka, and Australia. The findings demonstrate that research motivations, variable selections, and analytical tools directly mirror each nation’s economic conditions and technical infrastructure. In Australia, advanced data collection frameworks and computational resources have driven a rapid shift toward machine learning and deep learning architectures that systematically incorporate high-resolution climate variables. Conversely, power systems in Pakistan and Sri Lanka continue to rely primarily on traditional statistical and econometric techniques. These developing nations focus overwhelmingly on macroeconomic indicators, demographic shifts, and household survey data while remaining constrained by low-frequency reporting, acute data scarcity, and significant institutional challenges.
The study highlights how divergent energy consumption trajectories between developed and developing nations shape underlying analytical priorities. While annual electricity demand growth in developed economies has stabilized around 0.7 percent, demand in developing nations has surged at an annual rate of 3 percent, according to International Energy Agency benchmarks. In advanced economies, load growth is primarily propelled by pervasive digitalization, widespread electrification, and changing commercial practices alongside stringent emissions reduction mandates. In developing nations, by contrast, consumption surges are propelled by rapid population expansion, intensifying urbanization, industrial manufacturing output, and expanding service sectors. The researchers observed that while electricity supply and demand maintain relative equilibrium in industrialized systems, developing countries grapple with chronic power shortages, rolling blackouts, and transmission constraints. These operational crises make accurate predictive models critical for basic energy security, capital investment planning, infrastructure expansion, and national economic survival.
To rigorously examine these dynamics, the authors executed a systematic literature review following PRISMA guidelines using the Web of Science database. An initial search targeting English-language peer-reviewed publications containing key terms such as energy demand and electricity consumption yielded 1,638 candidate publications published over a ten-year span. This preliminary corpus showed a pronounced geographic imbalance, comprising 479 records for Australia, 399 for Pakistan, and merely 26 for Sri Lanka. Applying standardized screening criteria to evaluate time coverage from 2005 onward, methodological rigor, and topical relevance distilled this pool down to 107 qualifying studies, representing approximately 6.5 percent of initial search returns. Annual publication trajectories from 2011 to 2024 revealed an accelerating research output across all three countries, marked by publication peaks in Australia during 2015 and 2022, Sri Lanka in 2022, and Pakistan reaching its maximum output in 2024.
The empirical review for Pakistan reflects a power infrastructure under immense pressure from demographic and socio-economic transformation. Pakistan’s national population expanded dramatically from 79.98 million in 1980 to 233.738 million by 2023, while urban populations surged from 21.91 million to 77.44 million over the same period. By 2022, household consumers totaled over 29.53 million, driving substantial consumption growth across the residential, industrial, commercial, and agricultural sectors. The residential sector constitutes the largest consumer base at 46.4 percent of total national demand, followed by industry at 28.4 percent, agriculture at 11.1 percent, and commercial entities at 7.5 percent. Systemic vulnerabilities further exacerbate Pakistan’s energy crisis, characterized by significant electricity theft, administrative mismanagement, and high transmission and distribution losses, which measured 23.2 percent in 2006 to 2007 and 17.3 percent in 2011 to 2012.
Methodologically, studies examining Pakistan have historically favored econometric approaches such as autoregressive distributed lag models, vector error correction models, Johansen cointegration, ordinary least squares, and Long-range Energy Alternative Planning software. LEAP-based scenario evaluations have projected national energy demand reaching 399 terawatt-hours by 2030 under baseline conditions, which could potentially contract to 312 terawatt-hours under active conservation measures. Researchers have also established that climatic fluctuations exert immense strain on the grid, with empirical estimates indicating that a single degree Celsius rise in ambient temperature generates an additional 109.3 gigawatt-hours of electricity demand due to extensive summer cooling requirements. Since 2018, machine learning approaches such as artificial neural networks, support vector machines, random forests, and adaptive-network-based fuzzy inference systems have emerged, demonstrating superior accuracy over traditional statistical baselines when applied to household datasets like PRECON.
The energy modeling landscape in Sri Lanka presents a distinct island context characterized by a self-contained grid lacking cross-border interconnections. The island has historically relied heavily on domestic hydroelectric generation, with renewable energy sources accounting for 53.38 percent of national requirements in 2010, predominantly delivered by large-scale hydro installations alongside biomass, wind, and solar. By 2019, primary national energy supplies shifted toward oil and local bioenergy. Although Sri Lanka has attained near-universal grid electrification, elevated electricity tariffs prevent lower-income brackets from enjoying continuous, reliable power access. Research output from Sri Lanka remains comparatively scarce, representing a notable gap in international literature. Furthermore, empirical findings show a unique structural distinction: unlike Pakistan, where electricity supply expansion directly drives gross domestic product growth, studies in Sri Lanka found no such direct correlation, suggesting energy conservation policies can proceed without dampening growth.
In terms of analytical methodologies, Sri Lankan forecasting has evolved across two sequential phases. Earlier investigations between the 1960s and 2015 relied on econometric frameworks such as vector error correction models, autoregressive integrated moving average formulations, and multivariate probit regressions to evaluate extensive survey samples spanning 55,000 households. These analyses confirmed that residential demand is governed by household floor area, income brackets, educational attainment, and appliance usage patterns. Post-2015 research has gradually introduced artificial intelligence methods, including dynamic artificial neural networks with clustering techniques to predict next-day hourly load distributions. Notable applied machine learning studies include combining genetic algorithms with LightGBM to optimize university microgrids, achieving an observed 13.86 percent electricity cost reduction, and deploying gene expression programming to forecast wind generation at the Pawan Danavi facility with a correlation coefficient of 0.92.
Australia exhibits an entirely different modeling profile shaped by advanced power markets, dense urban populations, and aggressive clean energy transition targets. With 67 percent of its population situated in major metropolitan centers and experiencing severe per capita greenhouse emissions, Australian policymakers implemented continuous structural energy reforms starting after 1990. Traditional forecasting in Australia made extensive use of econometric panel testing, cointegration methods, and time-series modeling using National Electricity Market and Australian Energy Market Operator operational records. Key historical contributions demonstrated that hybrid seasonal autoregressive integrated moving average models incorporating minimum temperature, maximum temperature, and solar exposure achieved mean absolute percentage errors as low as 3.41 percent for weekly peak loads. Similarly, autoregressive distributed lag studies established that state-level electricity demand remains price and income inelastic, while daylight-saving difference-in-difference models captured late-afternoon demand shifts.
Over the last decade, Australian research shifted decisively toward machine learning and deep learning architectures, accounting for 62 percent of analyzed national forecasting implementations. Australian researchers have widely deployed extreme gradient boosting, support vector regression, convolutional neural networks, and long short-term memory recurrent neural networks to capture high-frequency nonlinear variations. Hybrid innovations feature prominently, such as convolutional echo state networks outperforming standard multi-layer perceptrons, and gated recurrent units integrated with bagged regression trees to predict regional market price spikes. Studies utilizing real-time smart meter streams from Western Australia verified that deep long short-term memory networks significantly surpass conventional time-series algorithms for household renewable tracking. Furthermore, comparative evaluations revealed that probabilistic deep learning frameworks achieved a 60 percent error reduction over traditional deterministic counterparts when handling complex high-dimensional load curves.
Across the entire comparative meta-analysis, the authors cataloged prominent aggregate trends across forecasting horizons, variable selection, and informatics architectures. In total, statistical models represented 60 percent of reviewed implementations, while machine learning represented 39 percent. Long-term forecasting represented the dominant temporal horizon, accounting for 56 percent of Australian and Sri Lankan studies and 45 percent of Pakistani studies, primarily driven by policy and infrastructure needs. Historical consumption data served as the most common input variable across all papers at 40 percent, followed by macroeconomic indicators at 22 percent, and climatic variables at 16 percent. However, substantial informatics hurdles persist. Developing nations remain restricted by aggregated monthly records, missing observations, and severe data siloing, whereas advanced grids confront black-box interpretability barriers and immense computational demands. To address these systemic weaknesses, the authors recommend expanding transparent hybrid models, establishing standardized benchmarks, and integrating high-resolution meteorological data with renewable generation profiles.
Subject of Research: Comparative analysis of electricity demand forecasting methodologies, variables, and horizons across developing and developed countries
Article Title: A comparative analysis of electricity demand forecasting using modelling methods, factors, and horizon: a review study of developing and developed countries
Article References: Omar, Z., & Liyanage, L. (2026). A comparative analysis of electricity demand forecasting using modelling methods, factors, and horizon: a review study of developing and developed countries. Energy Informatics, 9 (1), Article 55. https://doi.org/10.1186/s42162-025-00596-4
Keywords: electricity demand forecasting, machine learning, energy informatics, statistical modeling, smart grids, power systems, energy policy, load forecasting, comparative, analysis, electricity, demand
News Source: Denise Maddox. (November 11, 2026). Global Review Uncovers Stark Methodological Divides in Electricity Demand Forecasting. Scienmag.
We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.
Enter your email address to subscribe to this blog and receive notifications of new posts by email.
Bioengineer.org © Copyright 2023 All Rights Reserved.
Please enter your username or email address to reset your password.
Loading comments…