煤自燃危险程度的判定一直是世界性难题，气体是煤火灾害检测的关键指标。煤自燃时，氧化和热解同时产生常量、微量和痕量的指标气体，红外检测技术选择性差、检测种类受限，化学传感器交叉敏感、零点漂移，气相色谱使用复杂、时效性差，这些方法均难以实现高灵敏度大量程的动态检测。可调谐半导体激光光谱分析技术（TDLAS）具有高选择性、高灵敏度、高精度及实时在线测量等优点，可同时检测多种气体，是气体检测方法的发展方向。本研究采用TDLAS技术同时高灵敏度检测煤自燃多组分气体，根据氧化气体产物的构成、浓度及其变化速率等特性，作为煤自燃早期隐患识别与预警的判据，降低误报率和漏报率，对实现煤自燃火灾的早期预测、预警、火区环境安全性判定与科学救灾决策及确保救灾安全，具有重要的实际意义。 本研究围绕煤自燃多组分气体可调谐激光光谱高灵敏度检测存在的关键科学技术问题，重点开展以下五个方面的研究： （1）针对煤自燃多组分气体近红外吸收谱线交叉混叠干扰严重的问题，研究多组分气体混合吸收谱线的精细谱带结构，揭示煤自燃多组分气体近红外波段吸收谱线的精细分布规律，分析烃类气体吸收谱线交叉干扰与自身谱线混叠干扰的现象，筛选各组分气体吸收谱线的最优中心波长，避免或降低混合气体吸收谱线的交叉混叠干扰，并确定煤自燃多组分气体的吸光度；分析TDLAS气体检测系统的噪声及其来源，研究直接吸收和二次谐波光谱信号的背景扣除及降噪方法，进而提出TDLAS气体检测系统稳定性的Allan方差评价方法，确定煤自燃多组分气体近红外波段的最小探测极限浓度。 （2）针对煤自燃近红外波段气体吸收强度弱及检测灵敏度低的问题，建立长光程高反膜Herriot池模型，分别对5 m和14.4 m的Herriot池进行仿真研究，研究入射光束多次反射的光线追迹结果及其反射光斑在镜面的分布规律，优化气体吸收池的有效光程，增强激光在光电探测器接收面的汇聚性；对调制信号输入到谐波信号输出的五个模块性能进行仿真研究，验证系统各模块的合理性；实验研究分析长光程气体吸收池的光程、光学噪声、光强稳定性等性能指标，为实验系统的搭建及参数的确定提供理论依据。 （3）针对TDLAS气体检测系统受气体温度、压强影响测量结果非线性漂移严重的问题，研究电流和温度的微小变化对激光器输出光波长、光功率及工作稳定性的影响规律；研究分析煤自燃各组分气体特征吸收谱线强度、线宽、气体分子数密度、气体分子吸收系数与环境温度的对应关系，探究二次谐波峰值与气体压力之间的内在关联关系，分析气体温度波动、压强波动分别对气体浓度测量精度的影响，进而提出温度与压强对测量浓度影响的归一化数据处理分析方法；采用混合粒子群优化支持向量机算法，建立基于SA-PSO-SVM的TDLAS气体检测系统测量误差温压补偿模型，通过与BP神经网络、SVM、PSO-SVM等算法模型的对比分析，验证该方法的检测精度及鲁棒性。 （4）针对煤自燃多组分气体近红外波段交叉混叠谱线分离难度大的问题，提出基于多重Lorentz函数模型的交叉混叠谱线分离方法及多谱线拟合算法的气体浓度反演识别方法，依据稀疏分解理论，建立描述C2H4气体吸收系数的四重Lorentz函数分解模型，从被测吸收谱线中有效分离出背景信号和待测C2H4气体气体吸收谱线，实现C2H4气体浓度的高灵敏检测；采用多谱线拟合算法，将测量分离出的WMS-2f/1f信号与仿真的WMS-2f/1f信号进行多谱线拟合，利用提取的谱线信息实现C2H4气体浓度的高精度反演识别；为有效实现煤自燃多组分气体交叉混叠谱线的分离和浓度反演提供理论依据。 （5）为了验证上述理论方法，提出TDLAS高灵敏气体检测性能的测试方法；结合煤矿井下应用环境，利用煤自燃多组分气体对TDLAS气体检测系统的探测灵敏度、稳定性、重复性、零点漂移、测量精度、不确定度、检测下限、重复性、稳定性、响应时间、示值误差等性能指标进行全面测试与分析，实验结果表明各项指标满足标准规定，实验过程系统稳定可靠，可满足矿井煤自燃多组分气体高灵敏度与高精度检测的需求。

The determination of coal spontaneous combustion risk has always been a worldwide problem, and gas is the key index of coal fire hazard detection. When coal spontaneously combusts, oxidation and pyrolysis simultaneously produce constant, trace and trace index gases. The existing infrared detection technology has poor selectivity, limits detection types, cross-sensitivity of chemical sensors, zero drift, complex use of gas chromatography and poor timelines. These methods are difficult to achieve accurate and real-time on-line detection. Tunable semiconductor laser spectroscopy has advantages of high selectivity, high sensitivity, high precision and real-time on-line measurement. It is potential of detecting several gases simultaneously, which is the future trend for the gas detection technology. In this thesis, we realize a simultaneous detection of multi-component gases for coal spontaneous combustion with high sensitivity by using the TDLAS. According to components, concentration and change rate of oxidized gas products, TDLAS is used as a criterion for early identifying and warning for hidden dangers of coal spontaneous combustion, which can reduce the false alarm rate and realize coal spontaneous combustion. The early prediction, early warning, environmental safety judgment and scientific decision-making of spontaneous combustion fire are of great practical significance. To solve key scientific and technological problems in the highly sensitive detection of multi-components gases using the laser spectroscopy, the following five aspects will be emphatically studied: (1) To solve the serious problem of cross-aliasing interference of near-infrared absorption spectra of coal spontaneous combustion multi-component gases, the fine band structures of absorption spectra of multi-component gases are determined by simulation analysis. By revealing the near-infrared absorption spectra of coal spontaneous combustion multi-component gases, we analyze that there are cross interference and aliasing interference in the absorption spectra of hydrocarbon gases. The optimum central wavelength of each component gas absorption line is selected to avoid or reduce the cross-aliasing interference of gas absorption line. The absorbance of coal spontaneous combustion multi-component gas are determined. The noise and its source of TDLAS gas detection system are analyzed. We also study how to eliminate effects of background noises in direct absorption and second harmonic spectra, and put forward the Allan variance evaluation method for the stability of TDLAS gas detection system. We determine the minimum detection limit of the multi-component gas concentration of coal spontaneous combustion. (2) To solve the problem of weak absorption and low sensitivity of detecting coal spontaneous combustion in near infrared band, firstly, the Herriot cell model with long optical path and high reflective film is established. Then, by carrying out numerical simulations of 5 m and 14.4 m Herriot pools via using a commercial software LightTools, we obtain the ray tracing results and the actual optical path length of multi-component gas in coal spontaneous combustion and the distribution of the reflected spot on the mirror surface. Effective optical path of gas absorption cell is optimized. The convergence of laser beam on the receiving surface of photoelectric detector is enhanced. We use the Simulink software to complete the simulation of the whole system from the modulated signal input to the harmonic signal output and verify the rationality of each module of the system. The performance of the optical path, optical noise, light intensity stability and other indicators of long optical path gas absorption cell are tested and analyzed, which provides a theoretical basis for the establishment of the experimental system and the determination of parameters. (3) The TDLAS gas detection system is suffering the nonlinear shift caused by affected by temperature and pressure. To solve this problem, we study experimentally the characteristics of six lasers and obtain the influence rule of small changes of current and temperature on laser output wavelength, optical power and working stability. The effects of surrounding temperature on the intensity and linewidth of characteristic absorption spectra, as well as the molecular density of the detected gas, are revealed. The intrinsic relationship between second harmonic peak and gas pressure is obtained. The influences of gas temperature fluctuation and pressure fluctuation on gas concentration measurement accuracy are studied and analyzed. A normalized data processing method is proposed for analyzing the influence of surrounding temperature and pressure on gas concentration. Basing on SA-PSO-SVM, we establish a temperature-pressure compensation model for the non-linear relationship of errors in the TDLAS detection system. By comparing with other algorithm models including BP neural network, SVM, PSO-SVM, we prove that this modified method can effectively improve the detection accuracy and robustness of the system. (4) The absorption lines of coal spontaneous combustion multi-component gases in the near-infrared band are seriously cross-aliased and hardly separated with each other. To solve this problem, we propose a method for separating cross-aliased absorption lines based on the multi-Lorentz function decomposition, and a method of gas concentration inversion recognition based on multi-spectral line fitting methods. According to the sparse decomposition theory, we establish a four-Lorentz function model describing the absorption coefficient of C2H4. Using this model, we effectively separate the background signal and the experimentally measure spectrum signal of C2H4, and realize a highly sensitive detection of the concentration of C2H4. Based on the multi-spectral line fitting algorithm, the measured separated WMS-2f/1f signal is fitted with the simulated WMS-2f/1f signal. A highly sensitive detection of C2H4 is realized by using the extracted spectral information. It provides a theoretical basis for separating cross-mixing spectral lines and inversion of concentration of coal spontaneous combustion multi-component gases. (5) To verify the above theoretical method, a testing method for the sensitivity of a TDLAS gas detection system is proposed. Combining with the application environment of underground coal mine, the detection sensitivity, stability, repeatability, zero drift, measurement accuracy, uncertainty, detection lower limit, repeatability, stability, response time, indication error and other performance indicators of the TDLAS detection system are tested and analyzed comprehensively. The experimental results show that the indicators meet the standard requirements, and the experimental process is stable and reliable, which proves that the proposed system can meet the needs of detections with high sensitivity and high precision for multi-component gas in underground coal spontaneous combustion.