Abstract
In this dissertation, applications of adaptive control to fermentationprocesses have been investigated. The nonlinear dynamic behaviours of thefermentation processes were on-line identified via the method of weightedmoving identification with an adjustable identification interval.Identified models were applied to control system design. While enclosingthe control system, state variables are required for feedback. However, infermentation processes, some state variables which posses physiologicalsignificance are not on-line measurable with the available analyticinstruments. For these state variables, state estimation was applied tocircumvent the inaccessibility problem. The estimated state variablesprovided the required information for feedback control. The closed-loopcontrol sytem was then established.Parameter identification is an important step for adaptive control. Inthis study, identification was based on a continuous-time model whereweighted moving identification was used. Two sets of weighting factorswere used. One was for the step of discarding old data; the other foradding new data. These weighting factors were chosen to ensure that theidentified model had a good tracking property and the estimatedparameters were convergent. On-line identification with an adjustableidentification intervall was also considered for fermentation processeswhere the dynamics changed significantly. Selection of the intervaldepended on measurement noise, modelling error, and adaptive gain. If oneof them was less than cach corresponding given value, an identificationstep was not required. If it was not the case, further estimation wascarried out. Hence, a suitable on-line identification interval wasobtained.On applications of the continuous stirred tank bioreactors (CSTBRs), theproposed adaptive control strategy provided a procedure for determining amultivariable controller. Control was carried out with manipulation ofdilution rate and substrate feeding concentration, It was possible tocontrol the system at a steady state within a limit cycle or crossing aseparatrix. The steady state might be a stable steady state or an unstableone.For fed-batch culture in glutamic acid production, on-line adaptive wascarried out. The on-line measurable state variables, such as dissolvedoxygen and carbon dioxide evolution rate, could be satisfactorilycontrolled at the desired values. By using the proposed algorithm, it waspossible to reduce operation cost and improve the productivity. For thesystem with the unmeasurable state variables, the state variables wereestimated by the indirect measurement technique. The state estimationmethods investigated were based on element balance, degree of reductance,and empirical reaction subspace. Cell mass concentration, substrateconcentration, and product concentration were determined in the p resenceof disturbances. During the period of fed-batch cultivation, substrateconcentration was controlled at a given value. The proposed method couldbe applied to industrial fermentation processes.本論文係針對非線性動態系統的醱酵程序，以可調變之識別間隔的連續式加權移動性識別法，進行系統識別以及適應控制理論的探討。對於程序上無法直接測量的狀態變數，則採用狀態估計方法，以間接測量方式估算之。並以估算的狀態值作為控制所需的訊息，完成醱酵程序回饋控制之環路。加權移動性識別法，是採用三個數據加權方式的系統識別。以其進疊代識別時，不僅可提高系統參數的收斂性。對於醱酵程序上總體誤差以及隨時間變化的系統動態等影嚮，亦可有效地予以追蹤與識別。識別間隔的線上調整，可避免了一般識別方法上不良識別情形與固定間隔識別的不當影嚮。本論文結合加權移動與線上調整識別間隔之識別技術，應用在醱酵程序的適應控制時，獲得良好之識別效果。在探討連續攪拌式醱酵程序之控制時，採用多變數控制策略，可適時調整稀釋率及基質進料濃度，使系統迅速地到達設定穩態。其中以適應控制策略進行系統橫越分界線、消除持續振盪及不穩定穩態等多變數控制，均獲得到滿意的控制效果，亦達到全面控制之要求。對於饋料批次的醱酵程序，則分別探討了線上可量測及不可量測狀態變數之線上適應控制。以可直接測量到的溶氧值或二氧化碳放出率作為被控制變數，進行控制時，可維持醱酵有利的條件。至於線上無法直接量測之菌體量、產物濃度以及殘餘基質濃度，則利用還原度均衡與經驗式反應次域兩種狀態估計方法，依據醱酵槽出口氣體成份分析，分別進行線上估算。並以估算之狀態回饋至適應控制器，達到醱酵槽之自動化目標。實驗結果發現，即使在程序操作不穩定的擾動下，經驗式反應次域法依然可提供良好的估計，因此非常適用於一般醱酵工廠的操作，控制。