Rotation factor tuning 旋转因子整定

旋转因子必须在线整定，因为它依赖于不可测过程干扰和测量噪声的性质。通常，旋转因子在0.1到0.4之间就足够了。

图14至17显示了一个如何整定旋转因子的例子。在这个例子中，旋转因子为0.2时拥有最佳控制性能。

旋转因子过大将会导致预测向量在连续的控制时间间隔里从正值到负值来回波动（见图14）。这是因为预测向量旋转反映的是因变量的随机噪声。

旋转因子过小则会减慢预测校正的速度，并导致预测向量漂离希望的操作区域。如图15所示。

一个好的预测向量旋转因子整定如下图16所示。

整定旋转因子另一个有效方法是观察预测误差历史记录（见图17）。若旋转因子太小，预测误差将长时间在0的一侧漂移（低频行为）。旋转因子过大将导致预测误差高频率变化。正确旋转因子整定是一个随机预测误差的模式。

更新频率

注意当一个斜坡变量是间歇变量时，变量的更新频率将影响旋转计算。控制器将假定预测误差是从前面的周期，而不是前面的更新得出的。这可能会导致斜率调整过于剧烈。

若更新频率已知，则旋转因子应该由更新频率（周期）划分。若更新频率未知或不能被近似，那么可以使用用户计算计算在该时间所需要的旋转因子。

Figure14: Effect of a Rotation Factor that is Too Large

                                                               图14：旋转因子影响太大

Figure15: Effect of a Rotation Factor that is Too Small

                                                                  图15：旋转因子影响太小

Figure16:Effect of Rotation Factor this is about right

                                                                  图16：旋转因子影响基本正确

Figure17:Prediction Error History for choosing Rotation Factor

                                                           图17：通过预测误差选择旋转因子

附原文：

The rotation factor must be tuned online,since it depends on the nature of the unmeasured process disturbances and the measurement noise. Generally, a rotation factor between 0.1 and 0.4 is sufficient.

Figures 14 to 17 show an example of how to tune the rotation factor. For this example, a rotation factor of 0.2gives the best control performance.

A rotation factor which is too large will cause the prediction vector to "wave" back and forth from a positive to a negative value in successive control intervals (see Figure 14).This occurs because the prediction vector rotation is reacting to random noise in the dependent variable.

A rotation factor which is too small will slow down the rate at which the prediction is corrected and cause the prediction vector to drift away from the desired operating region. See Figure 15 below.

Prediction vectors for good rotation factor tuning are shown in Figure 16.

Another useful technique for tuning rotation factors is to observe the prediction error history (see Figure17). For a rotation factor that is too small, the prediction error drifts on one side of zero for prolonged periods of time (low frequency behavior). A rotation factor that is too large will result in high frequency variation in the prediction error. The correct rotation factor tuning results in a randomized prediction error pattern.

Update frequency

Note that when a ramp is also an intermittent variable, then the variable's update frequency will impact the rotation calculation. The controller will assume that the error in the prediction is from the previous cycle, rather than from the previous update. This can make the adjustment to the ramp rate too dramatic.

The rotation factor should be divided by the update frequency (in cycles) if known. If the update frequency is unknown or cannot be approximated, then user calculations could be used to calculate the rotation factor at the time it is needed.

                                                                                                  2015.9.30