论文部分内容阅读
针对基于分布式模型的传感器故障隔离方法中传统模型结构的不足,从模型结构和建模精度两个方面提出一种改进方法.首先针对传统模型结构直接将传感器输出作为模型输入,易导致故障虚警的问题,给出一种时间窗交互预测的改进结构;然后采用神经网络作为建模工具,设计了结合粒子群算法和梯度法的网络训练方法,以克服梯度法局部最优的缺陷.以B747飞机模型为诊断对象,通过分别设置攻角传感器和俯仰角速度传感器发生漂移或恒偏差故障,对比验证了所提方法的有效性.
Aiming at the deficiency of traditional model structure in sensor fault isolation method based on distributed model, an improved method is proposed from two aspects of model structure and modeling precision.Firstly, according to the traditional model structure, the sensor output is directly input as the model, The paper proposes a modified structure of time-window interaction prediction. Then, using neural network as a modeling tool, a network training method based on particle swarm optimization and gradient method is designed to overcome the local optimum defect of gradient method. B747 aircraft model as the diagnostic object, respectively, by setting the angle of attack sensor and pitch angular velocity sensor drift or constant deviation error, verify the effectiveness of the proposed method.