论文部分内容阅读
针对轴流压气机系统中的分岔预测问题,基于简化的Moore-Greitzer 3阶压气机模型,分析了该系统中存在的分岔现象;利用最新发展的确定学习理论,对压气机系统随着γ参数变化出现的几种典型模态的相关系统动态进行辨识,并将所学知识保存成常值RBF神经网络以构成模式库;利用该模式库构建1组嵌入了常值RBF神经网络的动态估计器;将测试模式与估计器相比,得到1组残差,并利用动态模式识别方法的残差最小原则实现了对Pitchfork分岔的预测。
Aiming at the problem of bifurcation prediction in axial compressor system, based on the simplified Moore-Greitzer 3-stage compressor model, the bifurcation phenomenon in this system is analyzed. With the newly developed deterministic learning theory, γ parameter variation of several typical modal correlation system dynamics identification, and save the knowledge into a constant value RBF neural network to form a pattern library; using this pattern library to build a set of embedded RBF neural network dynamic One set of residuals are obtained by comparing the test pattern with the estimator, and the prediction of Pitchfork bifurcation is realized by the minimum residual error of dynamic pattern recognition method.