基于相空间重构和神经网络的压气机机匣静压预测

建立了基于相空间重构和径向基神经网络的压气机机匣静压的预测模型。对试验数据进行相空间重构,分析其混沌特性,根据重构相空间的最小嵌入维数确定网络输入参数的个数,采用K均值聚类的方法优化网络的拓扑结构,根据时间序列的最大李亚普诺夫指数确定预测模型的最大有效预测步数,利用径向基神经网络的强大非线性映射能力,实现对时间序列的非线性预测,试验结果证明了该方法的有效性。该方法对压气机趋势监控具有一定的参考价值。 

Compressor Casing Static Pressure Forecasting Via Phase Space Reconstruction and Neural Network

A novel forecasting model for compressor casing wall pressure based on phase space reconstruction and radial basis function network was established.Phase space for the experimental pressure data was reconstructed, and corresponding chaotic characteristics were analyzed.Number of network input variables was determined through the computed minimum embedding dimension of the reconstructed phase space.The network's topology was optimized via K-means clustering method,and the maximum effective forecasting steps was determined by computing the largest Lyapunov exponent of the examined pressure time series.Taking advantage of the strong nonlinear mapping capability of the radial basis function neural network,nonlinear forecasting of the measured time series was realized.The forecasting results successfully validate the feasibility and effectiveness of the newly developed algorithm,which may serve as a promising trend monitoring method for axial flow compressors.