【作者】 柴艳莉；

【导师】 钱建生；

【作者基本信息】 中国矿业大学 ， 通信与信息系统， 2011， 博士

【摘要】 近年来,煤与瓦斯突出事故频发,尤其是在一些自动化程度相当高的矿井甚至是低瓦斯矿井也开始出现突出事故,突出事故已然成为煤炭行业可持续发展的制约因素。如何更加精确地对煤与瓦斯突出进行预测预警成为亟需解决的重要问题。论文以国家自然科学基金为依托,以煤矿具体项目为实际研究背景,对煤与瓦斯突出的预测预警进行了深入研究。本文的研究工作主要包括以下四个方面:(1)综合了煤与瓦斯突出事故的预警预测理论模型。对比分析了国内外煤与瓦斯突出机理的各类假说和煤与瓦斯突出预测技术的研究现状,基于智能信息处理对煤与瓦斯突出预测预警进行了深入研究。(2)建立了粒子群优化神经网络的煤与瓦斯突出危险性区域预测模型。针对基于BP神经网络的煤与瓦斯突出危险性区域预测模型存在收敛速度慢、极易陷入局部极值等问题,结合了粒子群优化算法极强的全局搜索能力和BP算法快速的局部搜索能力,提出了结合粒子群优化算法与BP神经网络算法的煤与瓦斯突出危险性区域预测模型。实验表明,该预测模型与基于BP神经网络的预测相比,在收敛速度和泛化能力上均有很大提高。(3)建立了基于粒子群优化的PSO-SVM的瓦斯时间序列突出预测模型,对瓦斯浓度时间序列进行了e -SVR回归预测研究,基于PSO对SVM瓦斯浓度模型进行了优化,利用软阈值小波去噪法对煤矿监测监控系统中的工作面瓦斯浓度时间序列进行了去噪分析。通过调整支持向量机不敏感损失参数e对瓦斯浓度时间序列进行预测建模,分析比较了SVM参数变化对预测精度的影响,最后对预测误差进行了分析。(4)建立了瓦斯浓度突出的DFNN预警模型,通过模糊RBF神经网络对模糊控制器的隶属度函数值进行训练,DFNN模型即可表达定性知识,又能表达定量知识,具有强大的自学习能力,实现了对瓦斯突出的提前预警目的,灵活性好,综合性强,准确性高,具有较好的应用前景。最后通过Matlab仿真实验验证了模型的可行性。更多还原

【Abstract】 This paper analyzes the warning and prediction theory model based on comprehensive causation. Various hypotheses based on gas geological theory thought that coal and gas outburst is the result of joint action of crustal stress, gas pressure and coal body structure. It only gave explanation to the phenomenon, but could not work out unified and complete outburst theory. This paper establishes warning and prediction theory of serious coal and gas accident through the revelation of its causation chain and complicated mechanism.This paper makes in-depth comparison of PSO and GA (Genetic Algorithm) and integrates PSO and NN. For the insuperable problems of BP algorithm based on gradient descent like falling into local extremum and low convergence speed, it brings the PSO into the optimization of connection weight of NN and applies it to the prediction of serious coal and gas accidents. The simulation result proves that the algorithm is easy to be realized and with fast convergence speed.Making use of wavelet soft-threshold de-noising method, we de-nose according to the gas concentration time series of the working place obtained from KJ98 observation system of Luling Coal Mine. Through the adjustment of insensitive parameter e of support vector machine, we establish prediction model for gas concentration time series and analyze the influence of different parameters to prediction precision. Based on PSO, we optimize gas concentration model of support vector machine and obtain optimal pair parameters through the calculation of simulation result. Combined with prediction model of optimal parameter to gas concentration time series, we also analyze e - SVR parameter and prediction error, and makes detail comparison of the influence of training sample size to prediction precision.This paper also makes detail study of fuzzy logical system and NN, and establishes D-FNN warning model of gas outburst. The model can show qualitative knowledge and has capacity in self-learning and processing of quantitative data. It passes the verification of simulation experiment and realizes the prediction of gas outburst.更多还原