The low reproducibility is the most critical obstacle to the wide commercial application of laser-induced breakdown spectroscopy (LIBS). In this work, to improve the sample-to-sample reproducibility of LIBS real-time analysis of coal calorific value, we applied the acoustic signal generated by the laser ablation as the reference signal for normalization. First, the correlation between acoustic signal and spectral signal was analyzed and the results showed that the spectral signal from coal pellet samples with 35-ton compaction pressure had a strong correlation with acoustic signal. Second, the influences of shot-to-shot spectral intensities with different normalization methods were analyzed, including total area normalization, channel normalization, background normalization and acoustic normalization. Finally, different normalization models for calorific value were established using the support vector regression (SVR). With the acoustic normalization method, the relative standard deviation (RSD) of shot-to-shot intensities of spectral lines of C (I) 193.1 nm, C (I) 247.8 nm, H (I) 656.3 nm and N (I) 868.1 nm were significantly reduced from 10.16, 6.53, 3.79, 5.21% to 7.01, 4.40, 2.81, 3.54%, respectively. Moreover, compared with model without normalization, the average RSD of sample-to-sample measurements of validation samples based on acoustic normalization method was reduced from 6.25% to 4.16%, and the average relative error (ARE) was reduced from 4.66% to 3.28%. The results demonstrated that the acoustic normalization method can effectively reduce the fluctuations of the shot-to-shot spectral intensities and was an effective method to improve the reproducibility and accuracy of measurement of coal calorific value in LIBS.

低重现性是激光诱导击穿光谱法（LIBS）广泛商业应用的最关键障碍。在这项工作中，为了提高LIBS煤热值实时分析的样品间可重复性，我们将激光烧蚀产生的声信号用作参考信号进行归一化。首先，分析了声信号与频谱信号之间的相关性，结果表明，压实压力为35吨的煤球样品的频谱信号与声信号具有很强的相关性。其次，分析了不同归一化方法对逐次光谱强度的影响，包括总面积归一化，通道归一化，背景归一化和声学归一化。最后，使用支持向量回归（SVR）建立了不同的发热量标准化模型。使用声学归一化方法，C（I）193.1 nm，C（I）247.8 nm，H（I）656.3 nm和N（I）868.1光谱线的逐次发射强度的相对标准偏差（RSD） nm分别从10.16、6.53、3.79、5.21％显着降低至7.01、4.40、2.81、3.54％。此外，与未进行归一化的模型相比，基于声学归一化方法的验证样本的样本间平均RSD从6.25％降低至4.16％，平均相对误差（ARE）从4.66％降低至3.28 ％。