<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE ArticleSet PUBLIC "-//NLM//DTD PubMed 2.7//EN" "https://dtd.nlm.nih.gov/ncbi/pubmed/in/PubMed.dtd">
<ArticleSet>
<Article>
<Journal>
				<PublisherName>Research Institute of Petroleum Industry (RIPI)</PublisherName>
				<JournalTitle>Journal of Petroleum Science and Technology</JournalTitle>
				<Issn>2251-659X</Issn>
				<Volume></Volume>
				<Issue>Articles in Press</Issue>
				<PubDate PubStatus="epublish">
					<Year>2026</Year>
					<Month>09</Month>
					<Day>30</Day>
				</PubDate>
			</Journal>
<ArticleTitle>A Hybrid Mechanistic–Data-Driven Approach for Predicting Flowing Bottom-Hole Pressure in Multiphase Oil Wells</ArticleTitle>
<VernacularTitle></VernacularTitle>
			<FirstPage></FirstPage>
			<LastPage></LastPage>
			<ELocationID EIdType="pii">1583</ELocationID>
			
<ELocationID EIdType="doi">10.22078/jpst.2026.5859.2003</ELocationID>
			
			<Language>EN</Language>
<AuthorList>
<Author>
					<FirstName>Alireza</FirstName>
					<LastName>Ebrahimzadeh</LastName>
<Affiliation>Department of Petroleum and Geoenergy Engineering, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran.</Affiliation>

</Author>
<Author>
					<FirstName>Ehsan</FirstName>
					<LastName>Khamehchi</LastName>
<Affiliation>Department of Petroleum and Geoenergy Engineering, Amirkabir University of Technology</Affiliation>

</Author>
<Author>
					<FirstName>Pezhman</FirstName>
					<LastName>Nazeri</LastName>
<Affiliation>Department of Petroleum and Geoenergy Engineering, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran.</Affiliation>

</Author>
</AuthorList>
				<PublicationType>Journal Article</PublicationType>
			<History>
				<PubDate PubStatus="received">
					<Year>2026</Year>
					<Month>07</Month>
					<Day>13</Day>
				</PubDate>
			</History>
		    <Abstract>Flowing bottom-hole pressure (FBHP) is a key quantity for the design, optimization, and surveillance of oil wells, yet its direct and continuous measurement is costly and often unavailable. Hybrids of a mechanistic pressure traverse and machine learning are established, but published studies report only that the combination outperforms its parts: which part produces the gain, and whether that gain survives a change of field, have not been quantified. This study addresses both on 1621 field records from seven published sources (1111 wells, 386 of them deviated), released with the paper. A Beggs–Brill traverse is marched from the measured wellhead pressure to the reference depth, and its output reaches a gradient-boosting model by two routes — as an input feature, and as an anchor whose residual is learned — crossed in a 2 × 2 ablation at fixed features and hyperparameters. Under leakage-safe well-level cross-validation the two routes prove largely redundant: of the 119 psia by which the hybrid improves on a data-driven baseline in root-mean-square error, 101 psia is delivered by the estimate entering as a single feature and a further 16 psia by the residual structure; the best configuration reaches R² = 0.957 and a mean relative error of 5.0%. Under leave-one-source-out validation the two routes separate: 10.3% mean relative error for the feature route against 8.9% for the anchor route. The untrained traverse itself remains at 7.7% out of source, so the learned correction is field-specific rather than universal, and the practical claim is limited accordingly.</Abstract>
		<ObjectList>
			<Object Type="keyword">
			<Param Name="value">Flowing bottom-hole pressure, Multiphase flow, Pressure traverse, Machine learning, Hybrid mechanistic&amp;ndash</Param>
			</Object>
			<Object Type="keyword">
			<Param Name="value">Data-Driven Model</Param>
			</Object>
		</ObjectList>
</Article>
</ArticleSet>
