1、翻译部分英文原文Classification technique for danger classes of coal and gas outburst in deep coal minesXueqiu He a, Wenxue Chen a,b,*, Baisheng Nie a, Ming Zhang aaDepartment of Resource and Safety Engineering, China University of Mining and Technology, Beijing 100083, ChinabDepartment of Mining and Materia
2、ls Engineering, McGill University,Canada H3A2A7Abstract: In this investigation a new classification technique based on artificial neural network (ANN) and exponent evaluation method (EEM) has been developed to classify the danger classes of coal and gas outburst in deep mines. A weight computing mod
3、el of mutual affecting factors is derived from backward algorithm of ANN (BA-ANN), which diminishes the influence of factitious factor, the environment factor and the time factor to the weight. The BA-ANN model is used for modeling the correlation between danger class and 12 affecting factors of coa
4、l and gas outburst and calculating weights of interconnection factors, which performs very well. In order to classify danger classes in a daily routine, the EEM with the well trained weights which are from BA-ANN, is performed in a deep mine. The case study shows that this new technique is useful to
5、 classify danger classes with quick and accurate computation. Moreover, the weight computing model of BA-ANN can be extended to other safety issue in different fields as well.Keywords: Coal and gas outburst; Danger classification; Weight Backward algorithm; ANN1 IntroductionCoal and gas outburst is
6、dynamic energy events which may result in the projection of fragmented coal-rock and rapid release of gases from the working face. It is driven mainly by gas pressure, which is caused by many factors such as geological factors, gases pressure and coal physical properties. Coal and gas outburst disas
7、ter is recognized worldwide as one of potentially fatal hazard to be managed during the mining in deep coal mines, which brings great threat to mine personal life and asset seriously. For these reasons, the danger classification of coal and gas outburst is significant work in daily routine. At prese
8、nt, there are a lot of approaches to classify the danger classes of coal and gas outburst. For example, the traditional analytic hierarchy process (AHP), the comprehensive evaluation (CE), the exponent evaluation method(EEM),the gray prediction approach,fuzzy comprehensive appraisal approach and art
9、ificial neural network approach.Different approaches have their own merits and draw backs. AHP, CE and EEM have simple and quick calculation merit for classification of safety issues, but the results accuracy is not enough.The ANN, the gray prediction and the fuzzy comprehensive approaches can gain
10、relatively accurate results, but the calculation speed of the ANN is slow, the gray prediction needs equal-internal data as initial data and the fuzzy comprehensive approach needs accurate weight. The quick and accurate danger classification of coal and gas outburst so that the manager can conduct c
11、orrect preventive action and production scheduling in a daily routine, is the key issue in the classification technique. Considering the accurate merit of the ANN and the simple and quick calculation merit of exponent evaluation method(EEM), a new classification technique combining BAANN model and E
12、EM to classify danger classes of coal and gas outburst is presented in this paper, which is performed in Xin-wen coal mine and gains satisfied result of danger classes.2 Artificial neural network (ANN)ANN is becoming a new discipline, as a mathematical system which can simulate the ability of biolog
13、ical neural networks by interconnecting many simple neurons. ANN is proposed upon the basis of modern neuroscience research results. The neuron accepts inputs from a single or multiple sources and produces outputs by a simple calculating process with a predetermined non-linear function.A typical thr
14、ee-layered ANN with an input layer, a hidden layer and an output layer is adopted in this study. Each layer consists of several neurons and the layers are interconnected by sets of correlation weights which are the correlation weights the research need to get. The neurons receive inputs from the ini
15、tial inputs or the interconnections and produce outputs by transformation using an adequate non-linear transfer function. A common transfer function is the sigmoid function expressed by f(x)=(1+e-x)-1, which has a characteristic of df/dx = f(x)1-f(x). The training processing of the neural network is
16、 essentially executed through a series of patterns. In order to get good result, the interconnection weights are adjusted within the input and output values. Therefore, ANN can reflect some characteristics of human brain function, for example, the ability to learn, distributed memory, fault tolerance and parallel operation. With these characteristics, ANN has been widely applied to various fields. Because of the principle of ANN has been well documente
