基于QSPR方法的脂肪醇化合物闪点预测

为提高脂肪醇化合物闪点预测精度,提出基于定量结构-性质关系(QSPR)原理的脂肪醇化合物闪点预测方法。应用Dragon软件计算出91种脂肪醇的分子描述符,利用遗传函数算法(GFA)从1 481个描述符中筛选出3个与脂肪醇闪点关系最密切的分子描述符。分别用多元线性回归(MLR)方法和支持向量机(SVM)方法进行建模,并采用内部验证和外部检验的方式对模型的拟合度、预测性等性能进行验证。结果表明:预测集的MLR方法和SVM方法的平均绝对误差(AAE)分别为2.870 K和2.706 K;均方根误差(RMSE)为3.451 K和3.371 K。SVM模型在精度上略优于MLR模型,而MLR模型更为简单和方便。     

Prediction of gas explosion pressures: A machine learning algorithm based on KPCA and an optimized LSSVM

A gas explosion, as a common accident in public life and industry, poses a great threat to the safety of life and property. The determination and prediction of gas explosion pressures are greatly important for safety issues and emergency rescue after an accident occurs. Compared with traditional empirical and numerical models, machine learning models are definitely a superior approach. However, the application of machine learning in gas explosion pressure prediction has not reached its full potential. In this study, a hybrid gas explosion pressure prediction model based on kernel principal component analysis (KPCA), a least square support vector machine (LSSVM), and a gray wolf optimization (GWO) algorithm is proposed. A dataset consisting of 12 influencing factors of gas explosion pressures and 317 groups of data is constructed for developing and evaluating the KPCA-GWO-LSSVM model. The results show that the correlations among the 12 influencing factors are eliminated and dimensioned down by the KPCA method, and 5 composite indicators are obtained. The proposed KPCA-GWO-LSSVM hybrid model performs well in predicting gas explosion pressures, with coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) values of 0.928, 26.234, and 12.494, respectively, for the training set; and 0.826, 25.951, and 13.964, respectively, for the test set. The proposed model outperforms the LSSVM, GWO-LSSVM, KPCA-LSSVM, beetle antennae search improved BP neural network (BAS-BPNN) models and reported empirical models. In addition, the sensitivity of influencing factors to the model is evaluated based on the constructed database, and the geometric parameters X1 and X2 of the confined structure are the most critical variables for gas explosion pressure prediction. The findings of this study can help expand the application of machine learning in gas explosion prediction and can truly benefit the treatment of gas explosion accidents.     

基于QSPR方法的活性化合物热稳定性预测研究

为形成安全、可靠、便捷的活性化合物热稳定性预测方法,快速获取活性化合物热稳定性参数,采用定量结构-性质相关性(QSPR)方法,针对38种有机过氧化物和104种硝基化合物的起始放热温度和分解热,结合遗传函数算法(GFA)和“断点原则”筛选出的分子描述符,利用遗传算法(GA)优化的BP神经网络,建立活性化合物的热稳定性GA-BP预测模型,验证分析模型的性能和应用域。研究结果表明:所建立的GA-BP模型具有良好的拟合能力、稳定性和预测能力,优于线性模型,说明活性化合物热稳定性与分子结构之间存在非线性关系;同时,得出影响活性化合物热稳定性参数的主要结构因素。     

烃类及其衍生物闪点、沸点的定量构效关系

基于定量结构一性质相关性（QSPR）原理,研究了烃类及其衍生物闪点、沸点与其分子结构间的内在定量关系。应用CODESSA软件计算384种烃类及其衍生物的分子结构描述符,建立了闪点和沸点的QSPR模型。用最佳多元线性回归（B．MLR）方法筛选得到的分子描述符建立了线性回归模型。用B-MLR方法所选择的5个描述符作为支持向量机（SVM）的输入建立了非线性模型。所有的化合物被分为训练集和测试集,对每个模型的训练集和测试集的复相关系数、交互验证系数、均方根误差等进行了计算,并用测试集对模型的预测能力进行检验,预测结果表明：预测值与实验值均符合良好,所建立的闪点模型稳健,泛化能力强,预测误差小,预测的效果令人满意,但沸点的模型预测效果有待加强。相比烃类物质的模型,加了衍生物的模型性能均有所下降。     

基于数据填补-机器学习的煤与瓦斯突出预测效果研究*

为解决煤与瓦斯突出事故数据集少,数据缺失严重的问题,提出将多重插补（MI）和随机森林填补(MF)应用于填补缺失参数,并将填补前和填补后的数据输入SVM,ELM,RF 3种机器学习算法进行训练,构建9种耦合模型。采用总体准确率、局部准确率、运行时间这3种指标评价模型性能。研究结果表明:采用数据填补算法后,由于训练样本增大,煤与瓦斯突出事故预测的总体准确率提高,运行时间增长;MF-RF模型的总体准确率与事故预测准确率最高,分别为97.90%和98.93%;RD-ELM模型的运行时间最短,为0.24 s;多重插补使得煤与瓦斯突出预测的总体准确率提高0.98%~1.11%,随机森林填补总体准确率提高5.13%~7.50%,随机森林填补的效果好于多重插补。     

支持向量机对多环芳烃毒性的定量构效预测

为了实现多环芳烃(PAHs)毒性的有效预测,提出应用定量构效技术对多环芳烃的空气-正辛醇分配系数(KOA)和致癌性进行预测。应用分子描述符和试验值确立构效关系,采用支持向量机算法(SVM)和人工神经网络算法(ANN)分别建立了PAHs的KOA回归预测模型和致癌性分类预测模型。利用网格划分(GS)、遗传算法(GA)、粒子群算法(PSO)对SVM进行参数寻优。应用均方误差(MSE)、拟合决定系数R2和分类准确率(Accuracy)分别对模型进行了验证与评价。结果表明,最佳回归预测模型GS-SVR的MSE为0.059 7,R2为0.913 0;最佳分类预测模型GA-SVC的Accuracy为95%。研究表明:应用SVM所建两种模型的稳定性和预测能力都优于应用ANN建立的模型;参数优化后模型的稳定性和预测能力得到了提高。     

咪唑类离子液体捕集CO_2性能的定量结构-性质相关性(QSPR)研究

CO_2是主要的温室气体,大量CO_2的存在严重影响着人类的生存环境和生态平衡,而咪唑型离子液体具有独特的气体溶解性,在CO_2的捕集分离中有很好的应用前景。基于定量结构-性质相关性(QSPR)原理,研究了咪唑类离子液体捕集CO_2的性能与其结构参数之间的内在定量关系。应用遗传算法获得与捕集量最为密切相关的一组描述符作为输入参数,随后,分别采用多元线性回归算法及支持向量机结合粒子群优化算法建立了咪唑类离子液体捕集CO_2的性能与其描述符之间的线性和非线性模型。多元线性回归算法得出训练集和测试集的复相关系数分别为0.765和0.814,支持向量机算法得出训练集和测试集的复相关系数分别为0.987和0.933。对预测模型进行了评价验证以及稳定性分析,结果表明,2种模型具有良好的稳定性能和预测能力。     

基于链式多重插补的WOA-ELM煤与瓦斯突出预测模型*

为了提高缺失数据下煤与瓦斯突出预测准确率,提出1种基于链式支持向量机多重插补（MICE_SVM）的鲸鱼优化算法（WOA）-极限学习机（ELM）预测模型,以淮南朱集矿区为例,选取5个煤与瓦斯突出影响指标作为模型特征,采用提出的MICE_SVM算法插补突出事故数据中缺失值,利用WOA优选ELM输入层权值及隐含层阈值,构建煤与瓦斯突出预测模型,将插补后数据用于WOA-ELM模型的训练与测试,并与其他模型的预测效果对比。研究结果表明:MICE_SVM插补前、后的有突出数据预测准确率分别为83.02%,90.41%,MICE_SVM显著提高了有突出预测准确率,对无突出和整体的预测准确率提高不明显;数据插补后WOA优化ELM对无突出、有突出和整体的预测准确率分别为97.94%,96.25%,96.48%,较优化前分别提高了5.79%,5.84%,5.55%,数据插补后WOA-ELM为最佳预测模型。     

A hybrid machine learning model for predicting crater width formed by explosions of natural gas pipelines

Despite the existence of industry models for estimating the crater width formed by the explosion of natural gas pipelines, their applicability is still limited since the complex formation mechanisms. In this work, a novel hybrid model was developed to predict crater width formed by explosions of natural gas pipelines, using artificial neural networks (ANN) as the fundamental predictor. Based on the historical accident records, the proposed hybrid model was trained by the pipeline parameter, the operating condition, the installation parameter, and the crater width. A novel nature-inspired optimization algorithm, i.e., the Lévy-Weighted Quantum particle swarm optimization (LWQPSO) algorithm, was proposed to optimize the ANN model's parameters. Three machine learning models were developed for comparative reasons to predict the crater width. The use of precision and error analysis indicators assesses prediction performance. The results show that the proposed hybrid model (LWQPSO-ANN) has high prediction accuracy and stability, which outperforms QPSO-ANN-based benchmark hybrid models and the model without an optimizer (Support Vector Machine, SVM). The parameter sensitivities of the proposed algorithm, including the maximum number of iterations, population size and contraction-expansion coefficient, were determined. The proposed hybrid model is expected to support the quantitative risk assessment (QRA), Right-of-Way (ROW) definition and the inherently safer design of the underground parallel pipelines.     

Development of machine learning based prediction models for hazardous properties of chemical mixtures

Lower flammability limit (LFL), upper flammability limit (UFL), auto-ignition temperature (AIT) and flash point (FP) are crucial hazardous properties for fire and explosion hazards assessment and consequence analysis. In this study, a comprehensive prediction model set was constructed by using expanded chemical mixture databases of chemical mixture hazardous properties. Machine learning based gradient boosting quantitative structure-property relationship (GB-QSPR) method is implemented for the first time to improve the model performance and prediction accuracy. The result shows that all developed models have significantly higher accuracy than other regular QSPR models, with the 5-fold cross-validation RMSE of LFL, UFL, AIT, and FP models being 1.06, 1.14, 1.08, and 1.17, respectively. All developed QSPR models can be used to estimate reliable chemical mixture hazardous properties and provide useful guidance in chemical mixture hazard assessment and consequence analysis.     

QSPR在混合物典型燃爆特性预测中的研究进展

为全面了解定量结构-性质关系（QSPR）方法在混合物燃爆特性预测中的研究现状，展望其发展趋势，综述其在混合物闪点、爆炸极限与自燃温度预测中的国内外研究进展，分析预测目标参数的选择、数据收集、描述符计算和筛选以及模型建立和验证等方面的不足与研究方向。结果表明:QSPR在混合物燃爆特性预测中尚处于起步阶段，当前研究的首要限制是混合物燃爆特性参数实验数据样本不足，关键点及难点是混合物结构的准确表征，未来研究应关注的重点是大量数据源统一的数据样本的获取方法、非加和性混合物分子描述符的计算方法以及机器学习等非线性建模方法。     

基于GA-ELM浆体管道输送临界流速预测模型研究

针对浆体管道输送临界流速预测难度大、精确度低等技术难题,提出了基于极限学习机(ELM)的临界流速预测模型,用训练集对模型进行训练,以验证集预测值的均方误差作为适应度函数,利用遗传算法(GA)对ELM模型参数进行优化,应用优化得到的ELM模型对预测集进行预测。以某矿山为例,模型参数优化结果如下:隐含层节点数L为400,输入权值ai、偏置向量bi最优组合下预测结果适应度为0.0201。采用优化的ELM模型对预测集进行预测,预测结果的最大相对误差x=3.96%,平均相对误差y=1.58%,对比BP神经网络(x=12.95%)和SVM模型(x=3.19%),表明ELM模型更加精确、高效。     

基于QPSO-RBF的瓦斯涌出量预测模型

为了提高径向基(RBF)网络预测瓦斯涌出量的泛化能力,提出QPSO-RBF模型。该模型采用量子粒子群(QPSO)算法优化RBF网络隐层基函数中心、扩展系数以及输出权等初始参数,将网络参数编码为QPSO学习算法中的粒子个体,在全局空间中搜索最优适应值参数。其中,RBF网络选取5-3-1的精简结构,采用5个变量作为影响因子预测瓦斯涌出量。结果表明,经QPSO优化后的RBF网络模型预测结果稳定且唯一,其泛化指标平均相对变动值(ARV)为0.012 2。与PSO-RBF、RBF模型预测结果比较,QPSO-RBF模型的泛化能力和网络训练速度优于前2种;预测精度约为PSO-RBF模型的1.5倍、RBF模型的4倍。