Gas disaster can be triggered by gas concentrations exceeding standard levels, and gas concentration prediction system can reduce the occurrence of gas disaster by predicting the trend of gas concentration and alerting engineers to take necessary measures whenever needed. With the increasing use of intelligent edge devices in coal mines and the limitations of some existing systems, developing a new gas concentration prediction system for large-scale intelligent edge devices has become an important issue. This work proposes to address the issue through a novel method for predicting gas concentrations by taking full advantage of multidimensional data in an intelligent edge system. Specifically, 1) it proposed a Single hidden layer Random Weights Neural Network (SRWNN) as the prediction model, which is based on interval prediction rather than point prediction; 2) It employs a Non-dominated Sorting Genetic Algorithm II (NSGA-II) to train SRWNN; 3) To significantly reduce the time consumed during model training and facilitate real-time predictions, it proposes a distributed gas concentration prediction scheme based on an intelligent edge system; and 4) it conducts extensive experiments by using actual industrial data collected from a company to demonstrate the superior performance of the proposed method.

瓦斯浓度超过标准水平会触发瓦斯灾害，瓦斯浓度预测系统可以通过预测瓦斯浓度的趋势并提醒工程师在需要时采取必要措施来减少瓦斯灾害的发生。随着煤矿中智能边缘设备的日益使用以及现有系统的局限性，为大型智能边缘设备开发新的瓦斯浓度预测系统已经成为重要的课题。这项工作提出了一种通过利用智能边缘系统中多维数据的优势来预测气体浓度的新颖方法来解决该问题。具体而言，1）它提出了一个小号英格尔隐藏层ř andom W¯¯八分Ñeural Ñ etwork（SRWNN）作为预测模型，该模型是基于间隔预测，而不是点预测; 2）采用非支配排序遗传算法II（NSGA-II）训练SRWNN；3）为显着减少模型训练过程中的时间消耗并促进实时预测，提出了一种基于智能边缘系统的分布式气体浓度预测方案。4）通过从公司收集的实际工业数据进行了广泛的实验，以证明该方法的优越性能。