原文：
Model Reduction 模型降階 采用最佳途徑將一個(gè)高階（或非參數(shù)）模型降為更低階模型的過(guò)程。
**MV **操作變量。
Noise 噪聲 干擾真實(shí)或原始信號(hào)的某一信號(hào)或某信號(hào)的組成部分。不可測(cè)量的輸入影響了過(guò)程行為。信號(hào)的不確定性或隨機(jī)性部分。
**Non-parametric Model **非參數(shù)模型 模型不由特定的模型結(jié)構(gòu)所限制。
**Overlay ** 覆蓋 多個(gè)趨勢(shì)（響應(yīng)）鋪設(shè)在彼此的頂部或同一坐標(biāo)軸上。
**Parametric Model **參數(shù)模型 模型的結(jié)構(gòu)被定義，且由固定數(shù)目的參數(shù)來(lái)定義模型。
**PCTP ** 過(guò)程控制技術(shù)包。.
**PRBS ** 偽隨機(jī)二進(jìn)制序列。也被稱(chēng)為PRBNS，其中N是噪聲（noise），S是信號(hào)（signal）。
**Prediction **預(yù)測(cè) 模型輸入一組輸入變量后，試圖模擬真實(shí)信號(hào)響應(yīng)的過(guò)程。
**Prefilter **預(yù)濾波器 當(dāng)信號(hào)被用于模型估計(jì)練習(xí)前時(shí)，濾波器應(yīng)用于其中一個(gè)或所有信號(hào)。PVR 過(guò)程變量探索器。
QMI 質(zhì)量測(cè)量?jī)x或性能分析儀。
Ramp 一個(gè)積分過(guò)程或過(guò)程變量。
Regression 回歸 回歸是用于表征當(dāng)輸入變量變化時(shí)，哪個(gè)輸出變量產(chǎn)生變化的方式。最優(yōu)回歸是通過(guò)使用最小二乘法最小化預(yù)測(cè)輸出與實(shí)際誤差得到的。
**Residual **殘差 代表了模型擬合后輸出的未知變化。它是預(yù)測(cè)輸出與實(shí)際輸出之間的偏差。也被稱(chēng)為誤差。
**RQE ** 穩(wěn)健質(zhì)量預(yù)估器。
Sample Period 采樣周期 連續(xù)信號(hào)連續(xù)采樣周期之間的時(shí)間間隔。
Settling-time 穩(wěn)態(tài)時(shí)間 當(dāng)輸入改變時(shí)，過(guò)程/模型重新進(jìn)入穩(wěn)態(tài)所需要的時(shí)間。
Signal-to-Noise Ratio 信噪比 信號(hào)部分與噪聲部分的比例。信號(hào)中噪聲含量越多，信噪比越低。SNR = σy /σn 其中σy是信號(hào)y（也可以是x）的測(cè)量值，σn是噪聲含量。
Simulation 仿真 確定性預(yù)測(cè) 其中沒(méi)有基于歷史輸出測(cè)量值的反饋用于更新預(yù)測(cè)。
**SISO **單輸入單輸出。
**SMOC **殼牌多變量?jī)?yōu)化控制。
**State-space **狀態(tài)空間 一個(gè)用于定義系統(tǒng)的抽象數(shù)學(xué)空間。在動(dòng)態(tài)系統(tǒng)中，根據(jù)系統(tǒng)狀態(tài)系統(tǒng)可被建模為一組一階微分（差分）方程。
**Stationary **穩(wěn)定 信號(hào)到達(dá)穩(wěn)態(tài)是在一個(gè)平均值周?chē)?dòng)的。變量與非穩(wěn)定或漂移相反的，是不在固定平均值周?chē)?dòng)。
Steady-State 穩(wěn)態(tài) 當(dāng)一個(gè)過(guò)程變量不隨時(shí)間變化時(shí)的狀態(tài)。
**Step Response **階躍響應(yīng) 對(duì)自變量施加一單位階躍，獲得的因變量的響應(yīng)。
Step Testing 階躍測(cè)試 通過(guò)給輸入（自變量）一個(gè)階躍從而給過(guò)程擾動(dòng)。
Tag 位號(hào) 過(guò)程變量標(biāo)簽。偶爾也稱(chēng)為點(diǎn)。
Variance 方差 方差描述了數(shù)值與期望值（或平均值）的差異。它是標(biāo)準(zhǔn)差的平方。這里指的數(shù)值既可以是一個(gè)直接的測(cè)量值，也可以是測(cè)量值的函數(shù)。
**White Noise **白噪聲 與本身不相關(guān)（無(wú)顯著自相關(guān)系數(shù) lags>0）且完全不可預(yù)測(cè)的時(shí)間序列（或序列）。
Zero-order Hold 零階保持 當(dāng)連續(xù)信號(hào)進(jìn)行采樣時(shí)，模擬信號(hào)在一個(gè)采樣間隔點(diǎn)進(jìn)行采樣，并且一直“保持”該值直到下一個(gè)采樣時(shí)間。因?yàn)榱汶A保持的緣故，在輸入輸出系統(tǒng)的情況下當(dāng)輸入變量變化時(shí)，至少需要一個(gè)采樣周期輸出變量才會(huì)變化。
原文：
Model Reduction The process of reducing a higher-order (or non-parametric) model into a lower order model in the best way.
MV Manipulated Variable.
Noise Any signal or a component of a signal that interferes with the true or original signal. Unmeasured inputs affecting process behavior. The non-deterministic or stochastic component of a signal.
**Non-parametric Model **A model not restricted by a specific model structure.
Overlay The laying of multiple trends (responses) on top of each other or on the same plot axes.
Parametric Model A model whose structure is set and is defined by a fixed number of parameters.
**PCTP ** Process Control Technology Package.
PRBS Pseudo Random Binary Sequence. Also referred to as PRBNS with N for noise and S for signal.
Prediction The response of a model to a set of inputs that tries to emulate a true signal.
Prefilter Filter applied to one or all of the signals before they are used in a model estimation exercise.
**PVR ** Process Variable Retriever.
QMI Quality Measuring Instrument or property analyzer.
Ramp An integrating process or process variable.
Regression Regression is used to characterize the manner by which an output variable changes as input variables changes. The best regression is obtained using a least-squares minimization of the output prediction version actual errors.
Residual Represents the unexplained variation of an output after fitting of a model. It is the difference between the actual output and the predicted output. Also known as error.
**RQE ** Robust Quality Estimator.
**Sample Period ** The time interval between consecutive samplings of continuous signal.
Settling-time Time taken by a process/model to become steady after an input change to it was made.
**Signal-to-Noise Ratio ** The ratio of the content of a signal to the noise content in it. The higher the noise level in a signal, the lower this ratio. SNR = σy /σn where σy is a measure of the content of the signal y (sometimes x) and σn is the content of the noise.
Simulation The deterministic prediction, where no feedback based on past measurements of
the output is used in updating the prediction.
SISO Single-Input Single-Output.
SMOC Shell Multivariable Optimizing Control.
State-space An abstract mathematical space used to define a system. In a dynamic system,
the system can be modeled as a set of first order differential (difference) equations in terms of the system states.
**Stationary **A stationary signal is one in which variation is around a mean value. The opposite of a non-stationary or drifting variable, in which variation is not around a fixed mean.
Steady-State The condition when a process variable is not changing with time.
Step Response The response of a dependent variable to a unit step change an independent variable.
**Step Testing **Perturbing a process using step changes in inputs (independents).
**Tag **A process variable label. Also referred to as point occasionally.
Variance Variance describes variability of a quantity about an expected (or mean) value. It is the square of the standard deviation. The quantity referred to can be a directly measured value or a function of measured values.
**White Noise **A time series (or sequence) that is completely unpredictable and therefore is uncorrelated with itself (no significant autocorrelation coefficients for lags > 0).
Zero-order Hold When a continuous signal is sampled, the analog signal is sampled at a sample interval and is “held” constant at that value until the next sampling instant. In the context of input-output systems, when an input change is made, the change usually takes at least one sampling interval before it shows up in the output because of zero-order hold.
2016/4/16
