从单光谱到多源融合：煤质快速检测技术发展综述

From Single-Spectral to Multi-Source Fusion: A Review of the Development of Rapid Coal Quality Detection Technology

摘要: 煤质快速检测对燃煤电厂的经济效益、设备安全与环保水平具有重要意义，传统检测方法耗时长、数据滞后，单一光谱技术又各有局限，本文旨在系统梳理煤质快速检测技术从单光谱到多源融合的发展脉络，明确技术现状与瓶颈。本文综述了瞬发伽马中子活化分析(PGNAA)、激光诱导击穿光谱(LIBS)、X射线荧光光谱(XRF)、近红外光谱(NIRS)及拉曼光谱等主流单光谱技术的原理、研究进展与固有缺陷，并归纳了多源融合在数据级、特征级与决策级三个层面的融合策略及代表性工作。分析表明，单一技术难以兼顾全指标、高精度与无辐射风险；多源融合通过信息互补可有效提升检测全面性与精度，但当前多数研究仍处于实验室阶段，在数据异质性、模型泛化能力、长期运行稳定性及可解释性方面面临共性挑战。多源融合是煤质快检的重要发展方向，未来需在样品前处理、信号处理与智能控制等方面持续突破，以推动大规模工业应用。

Abstract: Rapid coal quality detection is of great significance to the economic benefits, equipment safety and environmental protection of coal-fired power plants. Traditional detection methods are time-consuming with delayed data, while single-spectroscopy techniques each have limitations. This paper aims to systematically review the development of coal quality rapid detection technology from single spectroscopy to multi-source fusion, and to clarify the current status and bottlenecks. This paper reviews the principles, research progress and inherent defects of mainstream single-spectroscopy techniques, including prompt gamma neutron activation analysis (PGNAA), laser-induced breakdown spectroscopy (LIBS), X-ray fluorescence spectrometry (XRF), near-infrared spectroscopy (NIRS) and Raman spectroscopy. The fusion strategies and representative works of multi-source fusion at data-level, feature-level and decision-level are also summarized. The analysis shows that a single technique cannot simultaneously achieve full-index detection, high accuracy and no radiation risk. Multi-source fusion can effectively improve detection comprehensiveness and accuracy through information complementarity. However, most current studies remain at the laboratory stage and face common challenges such as data heterogeneity, model generalization ability, long-term operational stability and interpretability. Multi-source fusion is an important direction for rapid coal quality detection. Future breakthroughs in sample preparation, signal processing and intelligent control are needed to promote large-scale industrial application.