基于数据融合的井下瓦斯浓度状态预测

西安科技大学 计算机科学与技术学院,陕西 西安 710054

瓦斯预测; 神经网络; 数据融合; DS证据理论

Prediction of underground gas concentration state based on data fusion
GONG Shang-fu,LI Yan-song

(College of Computer Science and Engineering,Xi'an University of Science and Technology,Xi'an 710054,China)

gas prediction; neural network; data fusion; DS evidence theory

DOI: 10.13800/j.cnki.xakjdxxb.2018.0323文章编号: 1672-9315(2018)03-0506-09

备注

针对现有方法中单独使用神经网络进行瓦斯预测时正确性得不到保障的问题,引入了数据融合的方法进行修正,期望在决策层达到提高正确率的目的 。首先,将原始数据降噪,通过状态划分以明确模型中DS证据理论所需要的识别框架; 然后,在保持数据拓扑意义不变的情况下,将数据重构为多维向量作为样本数据,利用该样本训练BP神经网络和径向基函数神经网络得到独立的2个预测模块,再选取样本训练得到分类器; 最后,结合统计误差从证据折扣的观点来构造DS证据理论中的基本概率分配函数,使得数据在融合的过程中减少冲突。结果 表明,提出的“预测-分类-融合”模型对瓦斯浓度区间判断的正确率比单独使用神经网络预测时的高。利用状态划分和数据融合可以对瓦斯数据进行有效的映射; 在构造分类器时可以结合相关性分析,以得到更多的证据体,有利于挖掘出潜在的信息,指导安全生产。

Aimed at the problem that the correctness could not be guaranteed when the neural network was used alone to predict the gas in the existing method,the data fusion method was introduced to correct the problem and the expectation was to achieve the goal of improving the accuracy at the decision-making level.Firstly,the original data was denoised,and the state identification was used to identify the identification framework needed by DS evidence theory in the model.Then,the data was reconstructed into multidimensional vectors as the sample data while maintaining the topological meaning of the data,and then the sample was used to train BP neural network and radial basis function neural network to get two independent prediction modules,and then choose the sample training to get the classifier.Finally,we combined the statistical error to construct the basic probability distribution function in DS evidence theory from the viewpoint of evidence discount for reducing conflicts in the process of convergence.The experimental results show that the proposed “prediction-classification-fusion” model has higher correctness of judging interval of gas concentration than when using neural network alone.The use of state partitioning and data fusion can effectively map the gas data.In the construction of the classifier,it can be combined with the correlation analysis in order to get more body of evidence,be conducive to mining potential information and guide safe production.