石炭自然発火の早期検知のためのガス指標と深層学習の誤警報制御評価
False-Alarm-Controlled Evaluation of Gas Indices and Deep Learning for Early Detection of Coal Spontaneous Combustion (原題)
CINAR U
🤖 gxceed AI 要約
日本語
石炭自然発火の早期検知において、ガス指標（Graham指数）と深層学習モデルを誤警報率を揃えて公平に比較するプロトコルを提案。シミュレーションと3鉱山の実データで検証し、Graham指数は誤警報率0.20以下では火災を検知できない一方、CO閾値は全3件を検知し中央値345時間のリードタイムを達成。注意LSTMは2件を検知した。
English
Proposes a false-alarm-controlled protocol to fairly compare gas indices and deep learning for early detection of coal spontaneous combustion. In simulation and field data from three mines, Graham's index fails to detect fires at false-alarm rates ≤0.20, while a CO threshold detects all three fires with 345h median lead time; attention-LSTM detects two.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本の石炭火力や鉱山安全に関連し、AIを活用した異常検知の信頼性評価手法は、GX投資や安全管理の高度化に寄与する可能性がある。ただし、脱炭素政策との直接的な関連は薄い。
In the global GX context
Globally, this contributes to reliable early-warning systems for coal mines, relevant to safety and operational efficiency in fossil-fuel industries. The false-alarm-controlled evaluation method is a methodological advance applicable to broader industrial monitoring, but its direct link to climate disclosure or transition finance is limited.
👥 読者別の含意
🔬研究者:誤警報制御による評価プロトコルは、異常検知モデルの公平な比較に有用な方法論を提供する。
🏢実務担当者:鉱山の安全管理や監視システムの改善に応用可能な知見を提供する。
📄 Abstract（原文）
<title>Abstract</title> <p>Gas ratios such as Graham's index remain the principal field diagnostics for coal spontaneous combustion, yet their reliability is rarely assessed against monitoring periods in which no heating develops. Reported detection rates and lead times are therefore uninterpretable, because any alarm rule can be made to look prescient by alarming continuously. We propose a false-alarm-controlled protocol in which every method, index threshold or neural network, is reduced to a continuous alarm score and compared at a matched false-alarm rate. It is developed on a stochastic simulation of 210 episodes in which coal temperature is the latent state generating the gas observations, then applied unchanged to 18 field episodes from three coal mines: three confirmed fires, fifteen event-free records. In both settings Graham's index detects no fire at any threshold whose false-alarm rate is at or below 0.20. In the field its ordering inverts, exceeding 24 in seven of thirteen event-free ventilated episodes while staying below 0.50 in two sealed-area fires, because the oxygen deficiency in its denominator is near zero in ventilated roadways and above 18 behind seals. At a matched false-alarm rate a carbon-monoxide threshold detected all three fires with 345 h median lead time; attention-LSTM models detected two.</p>
🔗 Provenance — このレコードを発見したソース
- Research Square https://doi.org/10.21203/rs.3.rs-10667830/v1first seen 2026-08-31 04:37:54 · last seen 2026-09-10 04:19:39
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。