基于确定学习理论的轴流压气机系统分岔预测

摘要

Aiming at bifurcation prediction problem in axial compression system, the bifurcation behavior of the system was analyzed based on simplified Moore-Greitzer model. Several typical patterns generated by varying in Moore-Greitzer model were identified by deterministic learning, the obtained knowledge were stored in constant RBF networks to form the pattern library finally. A dynamic estimator which was embedded in the constant RBF networks estimators was constructed using the pattern library. Comparing the set of estimators with the test pattern, a set of residual error was generated. Pitchfork bifurcation was predicted by using minimum residual of dynamical pattern recognition.%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。