论文部分内容阅读
针对BP神经网络算法对煤体瓦斯渗透率预测精度低问题,筛选出影响预测精度的5个主要因素——1个宏观因素(煤层埋深)和4个微观因素(有效应力、温度、瓦斯压力、抗压强度),提出一种基于学习向量量化神经网络(LVQ)分类、混沌粒子群算法(CPSO)优化、BP神经网络预测的LVQ-CPSO-BP煤体瓦斯渗透率预测方法。从宏观上确定临界值将煤层埋深划分为2层;基于有效应力与瓦斯渗透率之间存在拐点关系,从微观上确定拐点值将有效应力划分为2段;采用LVQ将4个微观样本参数依据拐点特征进行分类识别,采用BP神经网络进行学习训练并输出预测结果,并用CPSO对BP神经网络的权值和阈值进行优化;基于样本案例对本文构建的LVQ-CPSO-BP算法进行预测结果验证,并与BP算法、GA-BP算法及PSO-BP算法预测的结果进行对比分析。结果表明:LVQ分类正确识别率较高,CPSO-BP算法预测精度较好,且优于其他3种算法。LVQ-CPSO-BP算法总体预测值与实测值吻合度高,尤其当有效应力减小时,预测精度更高。
In view of the low accuracy of BP neural network algorithm for coal gas permeability prediction, five main factors influencing prediction accuracy are selected - one macroscopic factor (depth of coal seam) and four microscopic factors (effective stress, temperature, gas pressure , Compressive strength), a LVQ-CPSO-BP coal gas permeability prediction method based on the learning vector quantization neural network (LVQ) classification, chaos particle swarm optimization (CPSO) optimization and BP neural network prediction is proposed. Based on the macroscopically determined critical value, the coal seam depth is divided into two layers. Based on the inflexion relation between effective stress and gas permeability, the effective stress is divided into two segments by microscopically determining the inflection point value. Using LVQ, the parameters of four microscopic samples According to the feature of inflection point, classification and recognition are carried out. BP neural network is used to train and output the prediction results. The weights and thresholds of BP neural network are optimized by CPSO. The prediction results of LVQ-CPSO-BP algorithm constructed in this paper are validated based on the sample case , And compared with the results predicted by BP algorithm, GA-BP algorithm and PSO-BP algorithm. The results show that the correct recognition rate of LVQ classification is high, and the prediction accuracy of CPSO-BP algorithm is better than that of the other three algorithms. The overall predicted value of LVQ-CPSO-BP algorithm is in good agreement with the measured value, especially when the effective stress decreases.