摘要
代数神经网络算法能够克服BP神经网络易于陷入局部极小和收敛慢的问题,通过优选激励函数和采用代数算法计算权值,将复杂的非线性优化问题转化为简单的代数方程组求解问题,提高了神经网络的精度与收敛速度.在使用代数神经网络算法进行煤自燃预测的实例中,采用均值规格化数据预处理,解决了煤自燃指标气体异动对分类结果的过度扰动.实验结果表明了算法的有效性和实用性.
Algebraic neural network algorithm could overcome the problem that BP neural network may easily fall into local minimum point and converge slowly.Weights are calculated by preferred exciter functions and algebraic algorithm.The algorithm transforms the complicated nonlinear optimization problem into linear algebraic equations and the adapting speed and accuracy of the neural network are improved.In an predicting coal spontaneous Instance using the algebraic neural network algorithm for predicting coal spontaneous combustion instance,because of mean normalized data pre-processing,the problem that coal spontaneous combustion index gas transaction excessive disturbance classification results is resovled.Experimental results show the effectiveness and practicality of the algorithm.
引文
[1]肖旸,王振平,马砺等.煤自燃指标气体与特征温度的对应关系研究[J].煤炭科学技术,2008,38(6):46-51.
[2]高春学,何启林徐庄矿八层煤自燃规律研究及指标气体优选[J].能源技术与管理,2005(1):19-21.
[3]赵武,徐新斌,涂强.神经网络测评自燃危险等级模型的研究[J].煤炭工程,2010(12):85-87.
[4]Ren T X,Edwardsa J,SClaxkeb D.Adiabatic oxidation study on the propensity of pulverised coals to spontaneous combustion[J].Fuel,1999,78(14):1611-1620.
[5]张辛亥,孙久政等.基于指标气体的煤自燃预报人工神经网络专家系统研究[J].煤矿安全,2010,41(3):10-12.
[6]刘璐.煤自燃发火气体预报指标的实验研究[D].西安:西安科技大学,2007.
[7]Chung F L,Lee L.Network-growth approach to design of feedforward neural Networks[C]//IEEE Proc Control Theory Appl,1995,142(7):486-492.
[8]马斌良,黄玉美等.基于神经网络代数算法的电子罗盘的标定[J].仪器仪表学报,2008,29(11):2304-2309.
[9]张代远.零代价函数的前馈神经网络新算法[J].计算机工程与应用,2000,96(10):79-80.
[10]Ampazis N,Perantonis S J.Two highly efficient second-order algorithms for training feedforward networks[J].IEEE Trans Neural Networks,2002,13(5):1064-1073.
[11]倪骁骅,赵忠伟,孙美艳.基于代数算法神经网络在机器人逆运动学中的应用[J].制造业自动化,2009,31(8):179-183.
[12]Baldi P.Gradient Descent learning algorithm overview:General dynamical systems persp-ective[J].IEEE Trans Neural Networks,1995,6(1):182-195.
[13]Chung F L,Lee L.Network-growth approach to design of feedforward neural networks[J].IEEE Proc Control Theory Appl,1995,142(7):486-492.