CAJ | 학술논문

为了提高非线性非高斯系统故障诊断的准确性,基于改进粒子滤波方法对系统状态进行估计,将系统状态估计值和实际值之差的绝对值作为残差,当残差平滑值大于阈值时诊断故障发生,使用故障误报率和漏报率构成的平均代价作为诊断效果评价指标.对水位/温度控制系统和一维非线性单变量模型进行仿真,由系统状态方程或观测方程参数跳变模拟故障发生,结果表明,3种算法能诊断出故障的发生,改进粒子滤波算法UPF的故障诊断平均代价小于SIR和UKF,诊断效果优于后两种算法,提高了故障诊断的可靠性.
위료제고비선성비고사계통고장진단적준학성,기우개진입자려파방법대계통상태진행고계,장계통상태고계치화실제치지차적절대치작위잔차,당잔차평활치대우역치시진단고장발생,사용고장오보솔화루보솔구성적평균대개작위진단효과평개지표.대수위/온도공제계통화일유비선성단변량모형진행방진,유계통상태방정혹관측방정삼수도변모의고장발생,결과표명,3충산법능진단출고장적발생,개진입자려파산법UPF적고장진단평균대개소우SIR화UKF,진단효과우우후량충산법,제고료고장진단적가고성.
In order to improve the accuracy of fault diagnosis in non-linear and non-Gaussian system, the method based on improved particle filter estimates the system state, the absolute value of difference between system state estimation and actual value is taken as residual, the fault occurs when the smoothed residual is greater than threshold, the mean cost which is constituted by fault false alarm ratio and miss alarm ratio is taken as evaluation criterion of diagnosis effectiveness. The water level/temperature control system and one-dimensional non-linear and single-variable model are simulated, the results indicate that the three algorithms could diagnose fault occur when the fault is simulated by the change of the system state equation or observation equation parameter, the fault diagnosis mean cost of the improved particle filter Unscented Particle Filte(UPF) is less than Sampling Importance Resampling (SIR) and Unscented Kalman Filter(UKF), the diagnosis effectiveness of UPF is better than the latter two algorithms, the fault diagnosis reliability is improved.