基于差分进化交叉验证SVM的飞控系统传感器故障预测学习算法研究

Research on Sensors Fault Prognosis Learning Algorithm Based on Difference Evolution Cross Validation SVM in the Flight Control System

参考现代飞行器故障检测结构,利用支持向量回归建立了飞控系统传感器的故障预测框架。该方案包括作动器,操纵面的传感器故障诊断模块,以解决故障趋势预测的问题。指出了如何采用SVM/SVR进行权值估计。提出采用差分进化改进原有的交叉验证方式,并且对核参数寻优以减少模型误差和提高SVR模型的泛化能力。最后结合飞行控制系统实时故障仿真证实了这种预测方法的可行性。

The sensors fault prognosis scheme of the flight control system was established through Support Vector Regression （SVR） in the solution follows from the fault detection scheme of the modern fighter plane. This sensors fault diagnosis scheme which consisted of the modules of the actuator, the steering face is discussed to overcome the problem of fault trend predication. It shows how to obtain weighted esti- mates for regression by applying SVM/SVR. A method through the differential evolution to improve the original cross validation is presented. Moreover, it search optimization of kernel parameters for decrease the model errors, and improve the ability of generalization of the SVR model. And, it simulates the online faults in flight control system that proved feasibility of this prognosis algorithm.