基于IGSA-BP网络的瓦斯涌出量预测模型  

Gas emission prediction model based on IGSA-BP network

作　　者：徐耀松[1] 齐翠玉 丰胜成 Xu Yaosong;Qi Cuiyu;Feng Shengcheng(School of Electrical and Control Engineering, Liaoning Technical University, Huludao 125105, China;Wang Zhuang Coal Mine, Lu 'an Environmental Protection Energy Development Co. LTD, Changzhi 046600,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,葫芦岛125105 [2]潞安环保能源开发股份有限公司王庄煤矿,长治046600

出　　处：《电子测量与仪器学报》2019年第5期111-117,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61601212);辽宁省教育厅科学技术研究项目(LJYL014)资助

摘　　要：为提高煤矿瓦斯涌出量预测的精度和效率,提出一种基于改进的万有引力算法(IGSA)的BP神经网络IGSA-BP瓦斯涌出量预测模型。由于BP神经网络的初始权值和阈值对网络的预测精度和收敛速度有较大影响,采用改进的万有引力算法训练BP神经网络的初始权值和阈值,引入粒子群算法记忆与社会信息交流的思想,对万有引力算法(GSA)的速度与位置更新公式进行改进,采用Tent混沌映射增加GSA种群的多样性,使算法避免陷入局部极值并增强GSA的遍历搜索能力。结果表明,改进的万有引力BP神经网络预测结果的误差在0.20 m^3/min以内,与未经改进的万有引力BP神经网络和粒子群BP神经网络相比,预测精度分别提高了近5倍和10倍,说明该方法对煤矿瓦斯涌出量具有更好的预测精度和收敛速度。In order to improve the accuracy and efficiency of gas emission prediction in coal mine,a BP neural network IGSA-BP gas emission prediction model based on improved gravitation search algorithm was proposed. Because the initial weight and threshold of BP neural network have great influence on the prediction accuracy and convergence speed of the network,the improved universal gravity search algorithm is used to train the initial weight and threshold of BP neural network. This paper introduces the idea of particle swarm optimization( PSO) memory and social information exchange,improves the speed and position updating formula of GSA,and uses Tent chaotic mapping to increase the diversity of GSA population. The algorithm can avoid falling into local extremum and enhance the traversal search ability of GSA. The results show that the prediction error of the improved gravitational BP neural network is less than 0. 20 m^3/min. Compared with the unmodified gravitational search algorithm BP neural network and particle swarm BP neural network,the prediction accuracy is improved by nearly 5 times and 10 times,respectively,which shows that the method has better prediction accuracy and convergence speed for coal mine gas emission.

关 键 词：瓦斯涌出量 预测 BP神经网络 万有引力算法 反向学习机制 Tent混沌映射 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]