多分辨小波过程神经网络及其应用研究.pdfVIP

第3卷第3期 智 能 系 统 学 报 Vo1．3 No．3 2008年6月 CAAI Transactions on Intelligent Systems Jun．2008 多分辨小波过程神经网络及其应用研究 李 洋，钟诗胜 (哈尔滨工业大学 机电工程学院，黑龙江哈尔滨 150001) 摘 要：以小波多分辨分析为理论基础，结合过程神经网络模型，建立了具有分层、多分辨和局部学习特性的多分辨 小波过程神经网络．该网络充分利用小波函数和尺度函数互补的特性，以及过程神经网络可以处理连续输入信号的 能力，能够很好地解决复杂的非线性时间序列的预测问题．给出了相应的学习算法，并以航空发动机排气温度裕度 状态监视为例，利用多分辨小波过程神经网络进行预测．结果表明，多分辨小波过程神经网络收敛速度快、精度高． 同时也为航空发动机排气温度裕度状态监视问题提供了一种有效的方法． 关键词：过程神经元；多分辨分析；多分辨小波过程神经网络；学习算法；发动机状态监测 中图分类号：TP183 文献标识码：A 文章编号2008)03—0211—05 Research on multi—resolution wavelet process neural networks and applications LI Yang，ZHONG Shi—sheng (Department of Mechanical and Electrical Engineering，Harbin Institute of Technology，Harbin 150001，China) Abstract：We propose a muhiresolution wavelet process neural network(MWPNN)based on wavelet multiresolu— tion analysis theory and the process neural network mode1．It combines their characteristics of delamination，multi— resolution，and local learning capability．The network，making full use of the complementary characteristic of wave— let function and scale function，as well as the ability to handle continuous input signals，can forecast complicated nonlinear time sequences．The network g learning algorithm is given．Using an example of state monitoring of ex— haust gas temperature margins of aeroengines，forecasting was performed using the multi—resolution wavelet process neural network．It was found that the proposed network exhibited good convergence and high accuracy，providing an effective way for state monitoring of exhaust gas temperatures of aeroengines． Keywords：process neuron；muhiresolution analysis；muhiresolution wavelet process neural network；learning algo— rithm；condition monitoring of aeroengine system