多分辨小波过程神经网络及其应用研究

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.

Research on multi-resolution wavelet process neural networks and applications

We p ropose a multiresolution wavelet p rocess neural network (MWPNN) based on wavelet multiresolu2 tion analysis theory and the p rocess neural network model. It combines their characteristics of delamination, multi2 resolution, and local learning capability. The network, making full use of the comp lementary characteristic ofwave2 let function and scale function, aswell as the ability to handle continuous input signals, can forecast comp licated nonlinear time sequences. The network’s learning algorithm is given. Using an examp le of state monitoring of ex2 haust gas temperature margins of aeroengines, forecasting was performed using the multi2resolution wavelet p rocess neural network. It was found that the p roposed network exhibited good convergence and high accuracy, p roviding an effective way for state monitoring of exhaust gas temperatures of aeroengines