CAJ | 학술논문

针对煤矿瓦斯涌出量的多影响因素预测问题，以多传感器的瓦斯监测系统采集处理后的数据作为样本，提出了一种自组织特征映射神经网络（Self-organizing Feature Maps，SOM）与多变量的径向基函数(Radial Basis Function，RBF)结合的组合人工神经网络的模型动态预测新方法。采用先聚类、再分类建模和预测的方法，解决了由于训练样本有限和训练样本点分散所导致的预测精度降低的问题，并通过矿井监测到的各项历史数据进行试验。结果表明，与其他预测模型相比较，该模型的预测精度更高，泛化能力更强。预测平均相对误差为2.16%，均相对变动值ARV为0.0059，均方根误差RMSE为0.1311，有效地实现了对煤矿绝对瓦斯涌出量的动态预测，有较高的实用价值。
침대매광와사용출량적다영향인소예측문제，이다전감기적와사감측계통채집처리후적수거작위양본，제출료일충자조직특정영사신경망락（Self-organizing Feature Maps，SOM）여다변량적경향기함수(Radial Basis Function，RBF)결합적조합인공신경망락적모형동태예측신방법。채용선취류、재분류건모화예측적방법，해결료유우훈련양본유한화훈련양본점분산소도치적예측정도강저적문제，병통과광정감측도적각항역사수거진행시험。결과표명，여기타예측모형상비교，해모형적예측정도경고，범화능력경강。예측평균상대오차위2.16%，균상대변동치ARV위0.0059，균방근오차RMSE위0.1311，유효지실현료대매광절대와사용출량적동태예측，유교고적실용개치。
A new model dynamic prediction method of combined artificial neural network combining self-organizing feature maps and multi-variable radial basis function is presented,which adopts collecting and processing data by multi-sensor gas monitoring system as samples,as a solution of the multi-factor prediction problem of coal mine gas emission. The modeling and prediction method are utilized as clustering firstly,and then it is utilized as classification to solve prediction accuracy loss,which is caused by the number limitation of training samples and their dispersion. The presented method is tested on the historical data monitored in the mine,and simulation results show that,the pre?sented model has a higher prediction accuracy and a better performance of generalization with average prediction er?ror 2.16%in comparison with other prediction models,and then average relation variance is 0.005 9 and root-mean-square error is 0.131 1. Therefore,it can be approved that the presented model realizes the dynamic prediction of ab?solute emission quantity of coal mine gas effectively and has a relatively high practicality.