基于极限学习机的矿井突水水源快速识别模型

A rapid identification model of mine water inrush based on extreme learning machine

在煤矿突水灾害防治过程中,需要快速准确地识别出突水水源类型。激光诱导荧光技术具有灵敏度高和快速监测的特点,利用该技术获取水样的荧光光谱。光谱经卷积平滑预处理和主成分分析提取特征信息后,采用极限学习机算法建立多元分类学习模型。确定隐含层激励函数为Sigmoid函数,并通过交叉验证法确定最优隐含层节点个数。从训练网络的平均时间、训练和测试的平均分类准确率和标准差方面,与BP和SVM传统分类算法进行了性能比较。结果表明:在训练集和测试集上的平均分类准确率方面,该模型与传统分类模型基本一致,但该模型分类准确率的标准差最小,说明其具有较稳定的分类性能;在训练模型学习时间方面,该模型能够大幅度降低分类学习时间,说明其具备快速识别突水水源性能。

In the process of disaster prevention of coal mine water inrush,it is necessary to quickly and accurately identify the types of water sources.The technology of laser induced fluorescence has the characteristics of high sensitivity,rapid and accurate for monitoring,and it also obtains the fluorescence spectra of water samples.After preprocessing spectra with Savitzky-Golay algorithm and feature extraction with principal component analysis,the multi-classification learning model is established by the extreme learning machine algorithm.The Sigmoid function is determined as hidden layer activation function,and the optimal number of hidden layer nodes is determined through the cross validation method.From the average time of training network,the average accuracy of classification and the standard deviation of accuracies,the performance is compared with the conventional classification algorithms such as BP and SVM. The results show that the model is consistent with the conventional classification model on the average accuracy of classification in the training and test set.While the standard deviation of accuracies is minimum,it shows that the model has the stable performance of classification.When training the model,the learning time is greatly reduced.Therefore,the model is more suitable for the rapid and accurate classification of water inrush sources.