Research trends and pattern analysis: the evolution of AI-based building energy estimation methods in China (2010–2024)

Artificial intelligence is now widely used to estimate building energy consumption in China, yet existing reviews have not established, for a defined Chinese corpus, how methodological composition has changed, where the evidence is drawn from, or how completely key information is reported in titles and abstracts. This paper addresses those questions through a PRISMA 2020-compliant systematic review of 185 studies published between 2010 and 2024, identified by searching Web of Science Core Collection, ScienceDirect and IEEE Xplore, screening 1,664 unique records and manually reviewing all 283 records that passed the rule-based screening stage. Four findings are reported. First, the methodological trajectory is a redistribution rather than a replacement: under the primary taxonomy, traditional machine learning falls from 60.0% of studies published in 2010–2014 to 39.7% in 2020–2024, while deep learning rises from 10.0% to 27.2% and hybrid designs remain close to one third of the literature. Second, the reported balance between families is sensitive to a taxonomic choice that reviews often leave implicit: classifying shallow feed-forward networks as deep learning rather than traditional machine learning moves the deep-learning share in 2020–2024 from 27.2% to 36.0% and the traditional share from 39.7% to 25.7%. Back-propagation/feed-forward networks are the most frequently named method in the corpus (30.8%), followed by support vector machines/regression (23.8%) and long short-term memory networks (19.5%). Third, the evidence base is geographically concentrated: of the 107 studies that name a provincial-level location, 76.6% name an eastern coastal unit, ten provincial-level units do not appear, and no industrial-building study was identified. Fourth, reporting in titles and abstracts is sparse for variables needed for deeper synthesis: 16.2% enumerate model inputs, 11.9% report a numeric accuracy value, 7.6% state a climate zone, and 1.1% name a national energy standard. The review therefore provides a measured, record-level account of methodological composition and identifies incomplete bibliographic reporting as a major practical constraint on synthesis.

Authors

Institutions

Publication Details

Journal
Ain Shams Engineering Journal
Published
2026-11-14
DOI
https://doi.org/10.1016/j.asej.2026.104509
Primary Topic
Building Energy and Comfort Optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Research trends and pattern analysis: the evolution of AI-based building energy estimation methods in China (2010–2024)

Xiangqi Yan, Di Wang, Zubair Ahmad
Ain Shams Engineering Journal
Building Energy and Comfort Optimization
article

Research trends and pattern analysis: the evolution of AI-based building energy estimation methods in China (2010–2024)

Xiangqi Yan, Di Wang, Zubair Ahmad
article en

Abstract

Artificial intelligence is now widely used to estimate building energy consumption in China, yet existing reviews have not established, for a defined Chinese corpus, how methodological composition has changed, where the evidence is drawn from, or how completely key information is reported in titles and abstracts. This paper addresses those questions through a PRISMA 2020-compliant systematic review of 185 studies published between 2010 and 2024, identified by searching Web of Science Core Collection, ScienceDirect and IEEE Xplore, screening 1,664 unique records and manually reviewing all 283 records that passed the rule-based screening stage. Four findings are reported. First, the methodological trajectory is a redistribution rather than a replacement: under the primary taxonomy, traditional machine learning falls from 60.0% of studies published in 2010–2014 to 39.7% in 2020–2024, while deep learning rises from 10.0% to 27.2% and hybrid designs remain close to one third of the literature. Second, the reported balance between families is sensitive to a taxonomic choice that reviews often leave implicit: classifying shallow feed-forward networks as deep learning rather than traditional machine learning moves the deep-learning share in 2020–2024 from 27.2% to 36.0% and the traditional share from 39.7% to 25.7%. Back-propagation/feed-forward networks are the most frequently named method in the corpus (30.8%), followed by support vector machines/regression (23.8%) and long short-term memory networks (19.5%). Third, the evidence base is geographically concentrated: of the 107 studies that name a provincial-level location, 76.6% name an eastern coastal unit, ten provincial-level units do not appear, and no industrial-building study was identified. Fourth, reporting in titles and abstracts is sparse for variables needed for deeper synthesis: 16.2% enumerate model inputs, 11.9% report a numeric accuracy value, 7.6% state a climate zone, and 1.1% name a national energy standard. The review therefore provides a measured, record-level account of methodological composition and identifies incomplete bibliographic reporting as a major practical constraint on synthesis.

Ain Shams Engineering JournalVol. 18(1)
Hunan University (CN), Yangtze University (CN)
Building Energy and Comfort Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon