煤矿安全对煤炭工业的健康持续发展至关重要, 而煤矿水灾又是煤矿事故的重大隐患, 因此煤矿水源数据的处理对于预防矿井突水事故具有重要意义。 实验在激光器的辅助下利用激光诱导荧光技术获取7种水源的数据信息, 设定激光发射功率为100 mW, 向被测水体发射波长405 nm激光, 获取实验水样210组的荧光光谱数据, 为了剔除光谱在采集过程受到的荧光背景、 检测器噪声以及功率波动等影响, 利用SG平滑、 多元散射矫正(MSC)预处理对数据进行降噪以及提高光谱特异性, 由于初始数据运算量过大并对数据压缩、 消除冗余和数据噪音, 利用主成分分析(PCA)分别对7种水样进行建模降维处理, 从而得到小数据并且保持原有信息的数据特征。 为了识别煤矿水源的突水类型, 对于降维后的数据利用粒子群算法(PSO)优化BP神经网络, PSO算法通过对新粒子的适应度值和个体极值、 群体极值适应度值的比较更新个体极值和群体极值的位置, 将最优初始权值和阈值赋予BP神经网络, 从而对待测水样的种类进行预测分析。 普通的PSO优化BP神经网络, 容易出现早熟收敛, 故在改进的PSO算法中引入变异因子来提高模型寻找更优解的可能性。 实验证明: SG, MSC以及Original三种预处理方式中, SG算法表现良好, 提高了模型的相关性。 在SG预处理的前提下, BP的决定系数 R²为0.984 5, 平均相对误差MRE 7.39%, 均方根误差为7.25%; PSO-BP的决定系数 R²为0.999 8, 平均相对误差MRE 0.17%, 均方根误差 0.08%; IPSO-BP的决定系数 R²达到0.999 9, 平均相对误差MRE和均方根误差RMSE皆为0.01%。 结果表明: 经SG预处理过后的光谱数据, 比MSC预处理效果更精确, 改进的粒子群优化算法更适用于该实验的矿井水源分类。

Coal mine safety is very important to the healthy and sustainable development of the coal industry, and the coal mine flood is a major hidden danger of coal mine accidents. Therefore, coal mine water source data processing is of great significance to prevent mine water inrush accidents. In this experiment, the laser-induced fluorescence technology was used to obtain the data information of 7 water sources. The laser power was set at 100 mW, 405 nm laser was emitted to the measured water, and 210 groups of fluorescence spectrum data of experimental water samples were obtained. n order to eliminate the influence of fluorescence background, detector noise and power fluctuation, SG smoothing and multiplicative scatter correction (MSC) preprocessing is used to reduce the noise and improve the spectral specificity of the data. Due to a large amount of initial data operation, data compression, redundancy elimination and data noise elimination, principal components analysis (PCA) is used to analyze the seven water samples Row modeling and dimensionality reduction are used to obtain small data and keep the original data characteristics. In order to identify the water inrush type of coal mine water source, particle swarm optimization (PSO) is used to optimize BP neural network for dimension reduced data. PSO algorithm updates the position of individual extremum and population extremum by comparing the fitness value of new particle with that of individual extremum and population extremum, PSO algorithm updates the position of individual extremum and population extremum by comparing the fitness value of new particle with that of individual extremum and population extremum, and endows the optimal initial weight and threshold value to BP neural network, so as to predict and analyze the types of water samples to be measured. The common PSO optimized BP neural network is prone to premature convergence, so mutation factor is introduced into the improved PSO algorithm to improve the possibility of finding a better solution. Experimental results show that the SG algorithm performs well among SG, MSC, and original preprocessing methods and improves the correlation of models. On the premise of SG pretreatment, the determination coefficient R² of BP is 0.984 5, the mean relative error MRE is 7.39%, and the root mean square error is 7.25%; the determination coefficient R² of PSO-BP is 0.999 8, the mean relative error MRE is 0.17%, the root mean square error is 0.08%; the determination coefficient R² of IPSO-BP is 0.999 9, the MRE and RMSE are 0.01%. The results show that the spectral data preprocessed by SG is more accurate than that by MSC, and the improved particle swarm optimization algorithm is more suitable for mine water source classification in this experiment.