摘要
神经网络具有非线性函数逼近能力,常用于非线性趋势预测。为了预测旋转机组故障特征量的非线性发展趋势,提出指数加权量子神经网络。参考传统的BP(backpropagation)神经网络及指数平滑的预测方法,将量子计算与神经网络相结合,选择误差修正学习算法,并在神经网络输入层进行指数加权,构建三层指数加权量子神经网络。该网络具有非线性逼近能力,并且能通过指数加权系数反映近期和远期历史数据对将来预测值的不同贡献程度。在将指数加权量子神经网络应用于大型旋转机组故障特征量的非线性趋势预测时,实验结果表明该网络训练的速度与预测的精度均好于传统的BP神经网络预测。
Neural network can approximate the nonlinear function and it was often applied to predict nonlinear trend. Aiming at predicting nonlinear trend of fault feature of rotating machines, the exponential-weighted quantum neural network (EQNN) was presented. Referring to the prediction methods based on traditional BP(back propagation) neural network and exponential smoothing model, the three-layer quantum neural network was constructed, which combined quantum calculation and neural network, selected the error correc- tion learning algorithm, and weighted the input data exponentially. The network has the ability of approximating the nonlinear function. And by exponential weight coefficients, it can reflect the recent data and long-dated history data have the different contribution of predicting future data. When EQNN was applied to the nonlinear trend prediction of fault feature of large rotating machines, the experiment results showed that the training speed and prediction precision of EQNN were better than the prediction results of BP neural network.
出处
《机械强度》
CAS
CSCD
北大核心
2010年第4期526-530,共5页
Journal of Mechanical Strength
基金
国家自然科学基金(50975020)
国家科技重大专项(2009ZX04014-101)
北京市属高等学校人才强教深化计划(PHR20090518)资助项目~~
关键词
旋转机组
量子神经网络
量子计算
故障特征量
趋势预测
Rotating machines
Quantum neural network
Quantum calculation
Fault feature
Trend prediction