一种新的无模型自适应控制模型参数整定方法

New determination method of parameters for model-free adaptive control

无模型自适应控制(MFAC)有四个模型参数,现有研究认为其无关联,且全程守恒。通过虚拟系统在初始时刻的状态,基于第一时刻输出值应与期望值接近的原则,采用遗传算法找到了参数关联,从而将四参数问题转化成单参数问题;并在控制过程中,依系统输出与期望差异自动改变步长因子取值,将其动态化,不但加速了初期计算,还避免了收敛时可能出现的超调或振荡;由此以参数关联和动态化为切入,改进了现有紧格式动态线性化MFAC方法,使其更为简捷和准确。某炼油单元3参数调优案例表明新方法可行,较老方法实现系统最大增益的参数调整次数由33次减少到14次,最大增益由413.4万元/年提高到542.9万元/年。

Model-free adaptive control(MFAC)has four model parameters,and existing studies consider it to be uncorrelated and conserved throughout.The relationship among 4 parameters is found out in this paper by assuming an initial state of the controlled system at the initial moment,based on the rule of the first moment output value should be close to the target value with genetic algorithm(GA),making the issue of 4 parameters to be simplified into the one of single parameter.Further,an automatic estimation method on step-length factor is suggested according to whether the absolute value of difference between output value and target value is less than a default value,thus it is regarded as a parameter with characteristic of time-variable and its maximum value is also enlarged into positive infinity from 1.The proposed two changes improve the present MFAC greatly,resulting in a faster calculation speed at the early control stage as well as avoiding overshoot and vibration at the convergence stage.A case application in a unit of oil refinery shows that the improved MFAC only needs 14 iterations for reaching the optimal target,and the system’s maximum added-benefit(MAD)can reach 5.43 million CNY/a.The number of parameter adjustments for the maximum gain of the system is reduced from 33 to 14 times,and the maximum gain is increased from 4.134 million CNY/a to 5.429 million CNY/a.