基于万有引力的煤层底板突水预测算法

Gravitational force based coal floor water inrush prediction algorithm

传统的基于机器学习的煤层突水预测方法需要大量的训练样本进行预测模型的训练。而在矿井突水问题中,大量训练样本的获得基本上是不可能的。主要研究在突水样本数据有限的情况下提高煤层突水预测结果的准确性。结合山西省某煤矿的实际情况,提出了一种新颖的基于万有引力的煤层底板突水预测方法(Gravitational force based algorithm,GFA)。该算法采用半监督的学习方式,将万有引力公式引入到预测模型中,利用少量的突水训练样本作为引力的源点吸引测试样本进行突水安全状态的传递,进而实现突水测试样本安全性的预测。将提出的算法用于历史突水数据以及实际的煤层底板突水数据进行实验,实验结果表明,在突水训练数据有限的情况下,提出的基于万有引力的煤层底板突水预测算法可获得良好的预测效果。

In order to forecast water inrush, many experiments had been performed, and numerous methods were proposed. Most of the traditional machine learning methods need a large number of training samples to train predictive models. However, collecting a large number of training samples in coal water inrush hazard is technically almost impossible. Combining the actual situation of a real coal mine in Shanxi Province, a novel semi-supervised gravitational force based water inrush prediction algorithm was proposed. In the proposed algorithm the principle of gravitational force was employed in the predictive model. Small training samples were used as attracting sources to attract testing sample, and further to propagate the label to testing sample. Simulation results show that with limited training samples the proposed methods can achieve a good accuracy in terms of predicting. The proposed methods are further applied in solving the water-irruption prediction problem on the coal seam floor. Results from both historical data and real water inrush data demonstrate that the gravitational based water inrush prediction algorithm could be regarded as a practical and effective approach in confronting the circumstance of limited valid data. © 2015, China Coal Society. All right reserved.