基于L-M梯度迭代算法的煤质发热量预测模型  

A prediction model for coal calorific value based on L-M gradient iterative algorithm

作　　者：黄奎 王充实 王林立 窦有权 张冬练 曾文慧 王嘉垠 曾勇 HUANG Kui;WANG Chongshi;WANG Linli;DOU Youquan;ZHANG Donglian;ZENG Wenhui;WANG Jiayin;ZENG Yong(Guoneng Nanjing Coal Quality Supervision and Inspection Corporation Ltd.,Nanjing 210031,China;Guoneng Daduhe New Energy Investment Co.,Ltd.,Chengdu 610096,China)

机构地区：[1]国能南京煤炭质量监督检验有限公司,南京210031 [2]国能大渡河新能源投资有限公司,成都610096

出　　处：《科技导报》2023年第20期106-112,共7页Science & Technology Review

基　　金：国家能源投资集团有限责任公司科技项目(GJNY-22-90)。

摘　　要：选取150家火电企业的煤质检测数据,通过分析泛在煤质化验数据信息,构建L-M算法下的煤质发热量的预测模型。实验结果表明:(1)煤质化验数据中仅碳(Cd)、灰分(Aad)与发热量(Qgr,ad)的线性关系较为显著,相关系数R2为0.8768和0.6880;(2)主成分分析法挖掘出影响煤质发热量的主成分特征值、特征矩阵及得分,实现了由六维矩阵降至四维矩阵的降维效果,增强了神经网络在训练过程中收敛的稳定性;(3)基于L-M算法下,改进的BP神经网络预测模型(LMBP)的训练集系数Rt、验证集系数Ra和测试集系数Rm分别为0.9957、0.9942和0.9963,总体系数为0.9931,同时通过待测20组数据进一步验证了LMBP预测模型可靠,预测精度较高,更符合实际预测需求。The coal quality test data of about 150 thermal power enterprises are selected,and the coal calorific value prediction model based on L-M algorithm is built by analyzing the ubiquitous coal quality test data information.The experimental results show that:(1)the linear relationships between calorific value and carbon,ash are significant respectively,and the correlation coefficients are 0.8768 and 0.6880;(2)principal component analysis(PCA)method excavates the information of the principal component eigenvalue,characteristic matrix and score,which could realize the dimension reduction effect from six-dimensional matrix to four-dimensional matrix,and enhance the convergence stability of neural network in the training process;(3)based on the L-M algorithm,the training coefficient,verification coefficient and test coefficient of the Levenberg-Marquardt back propagation neural network prediction method(LMBP)model are 0.9957,0.9942 and 0.9963 respectively,and the overall coefficient is 0.9931.Through the further verification of 20 groups of data to be tested,the LMBP prediction model is more reliable,the prediction accuracy is higher,and better to meet the actual forecast demand.

关 键 词：煤质化验数据 主成分分析法 L-M算法 训练集系数 

分 类 号：TQ533.4[化学工程—煤化学工程] TP18[自动化与计算机技术—控制理论与控制工程]