這里有兩點(diǎn)值得強(qiáng)調(diào)：根據(jù)Impulse Factor的控制器行為和根據(jù)干擾配置的仿真性能。
首先，讓我們討論根據(jù)Impulse Factor的控制器行為。當(dāng)液位（POV）的Impulse Factor使用默認(rèn)值0時(shí)，結(jié)果是控制器行為變得相當(dāng)好動(dòng)和嘈雜。記住在SMOCPro理解中，Impulse Factor=0預(yù)示著當(dāng)前預(yù)測(cè)誤差取決于真實(shí)的和不會(huì)消失的擾動(dòng)。因此，這種設(shè)置使控制器預(yù)測(cè)液位估計(jì)的誤差隨著控制時(shí)域積分。因?yàn)樵诿總€(gè)控制器執(zhí)行時(shí)，噪聲貢獻(xiàn)可能會(huì)導(dǎo)致CV上下移動(dòng)，SMOCPro必須迅速響應(yīng)任何級(jí)別的錯(cuò)誤，并調(diào)整Outlet Flow OP（MV），這樣才能取消感知到的干擾。
接下來(lái)，當(dāng)我們將Impulse Factor增加到新設(shè)定值0.8，我們知道在控制器預(yù)測(cè)理解起來(lái)，這一新設(shè)定意味著只有20%的液位估計(jì)誤差隨著控制時(shí)域積分，而感知到80%的干擾將隨著時(shí)域消失。因此，Outlet Flow OP（MV）只對(duì)其中一小部分的液位誤差有反應(yīng)。因此，很清楚的是控制器將達(dá)到更平滑的行為。
我們考慮的最后一個(gè)脈沖因子是Impulse Factor = 0.95。和前面一樣，這里的控制器預(yù)測(cè)是僅有5％的當(dāng)前液位預(yù)測(cè)誤差隨著控制時(shí)域積分，而95％是轉(zhuǎn)瞬即逝的。Outlet Flow MV對(duì)液位噪聲幾乎沒(méi)有反應(yīng)，在液位階躍擾動(dòng)時(shí)的動(dòng)作也非常平滑。
其次，讓我們來(lái)測(cè)試根據(jù)干擾配置的仿真性能。通過(guò)比較仿真圖后可以顯而易見(jiàn)的是：仿真的第2組設(shè)定（干擾直接注入液位CV）與第1組設(shè)定（干擾注入U(xiǎn)NM）相比，其噪音顯著地較大。讓我們通過(guò)一個(gè)例子解釋這背后的原因。思考我們?cè)诒竟?jié)中已經(jīng)提出的例子，即注入一個(gè)標(biāo)準(zhǔn)差為1，平均值為0的噪聲信號(hào)。一方面，當(dāng)噪聲被注入U(xiǎn)NM時(shí)，顯然值為1的噪聲必須經(jīng)過(guò)模型塊進(jìn)入液位CV。因?yàn)檫@個(gè)原因，它可能不是直接通過(guò)注入噪聲到UNM來(lái)指定代表有意義的物理參數(shù)的噪聲值。為了實(shí)現(xiàn)這一點(diǎn)必須考慮到模型的參數(shù)。

原文：
There are two items worth highlighting here: controller behavior as a function of Impulse Factor and simulation performance as a function of disturbance placement.
Firstly, let us discuss the controller behavior as a function of Impulse Factor. When the default value of zero is being used for the Impulse Factor of the Level (POV) the resulting controller behavior is quite aggressive and fairly noisy. Remember that an Impulse Factor of zero translates into SMOCPro understanding that the current prediction error is due to a real and non-vanishing disturbance. Consequently, this setting makes the controller predict that the error in Level estimation is ramping along the control horizon. Because at every controller execution the noise contribution may lead the CV to move up or down, SMOCPro must quickly react to any Level error and adjusts the Outlet Flow OP (MV) so that it cancels the perceived disturbance.
Next, as we increase the Impulse Factor to a new setting of 0.8, we understand that this new setting translates into the controller predicting that only 20% of the error in the Level estimation is ramping along the control horizon and thus 80% of the perceived disturbance will vanish along the horizon. As a consequence, the Outlet Flow OP (MV) only reacts to a small part of the Level errors. Thus, it is clear that the controller achieves smoother behavior.
The last Impulse Factor under consideration here is Impulse Factor = 0.95. As in the previous case, here the controller predicts that only 5% of the current Level prediction error is ramping along the control horizon, whereas 95% is fleeting. The Outlet Flow MV barely reacts to the Level noise and moves smoothly for the Level step disturbance.
Secondly, let us examine simulation performance as a function of disturbance placement. One thing that is plainly evident by comparing the simulation figures is that the second set (disturbance injected directly into the Level CV) of simulations is significantly “noisier” as compared to the first set (disturbance injected into the UNM). Let us explain the reason behind this with an example. Consider the case that we have presented in this section, namely injecting a noise signal with a standard deviation of 1 and zero mean. On the one hand, where the noise is injected into the UNM, a noise value of 1 has to go through the model block to manifest itself in the Level CV. For this reason, it may not be as straightforward to specify noise values that represent meaningful physical parameters by injecting the noise into the UNM. To achieve this one must take into account the model parameters.

2016.5.23