Abstract
本研究之主要目的在於提出解決化工程序控制常遭遇的交互影響與非線性問題，以及複雜程序之最適化操作設定點的決定。在化工廠常見的極具非線性的系統為高純度蒸餾塔。高純度蒸餾塔的產品組成受塔溫影響不大，不能由塔溫精確控制。本研究發展動態變數估計器（DSVE），可以準確並迅速的估算產品組成，直接進行產品組成控制，避免因為線上分析儀的延遲時間長，而不能直接進行產品組成控制。本研究利用類神經網路發展非線性動態模式，利用此模式設計簡化型去藕合器，分別以連續攪拌反應槽與蒸餾塔的非線性模式進行去藕合控制之測試，結果証明類神經網路去藕合器可以達到去藕合之目的。化工廠中，操作變數最多之一的單元為並聯多重汽提塔之蒸餾塔（簡稱原油分餾塔），其操作變數達13個，被控制變數達7?8個。本研究利用類神經網路發展了原油分餾塔之穩態模式，並結合SQP最適方法，用來決定原油分餾塔之最適化操作變數。本研究比較了目前具有側流之高純度蒸餾塔常使用的單溫差控制與雙溫差控制之優缺點；比較三種內回流控制方式，以消除回流與側流間之相互影響；針對進料之干擾加入前饋控制，以有效的消除干擾；並發展間接模式基礎控制進行產品組成之控制，並成功的應用在實際生產的蒸餾塔。The main purpose of this study is to develop an algorithm tosolve the interaction and nonlinear problems in chemicalprocesses. The other purpose is to develop a method fordetermination of the optimal operation in a complicate process.The temperature control exhibits poor performance in a highpurity distillation column. The on-line analysis control posesdifficulties due to the presence of the long dead-time. Adynamic state variable estimator (DSVE) is proposed in thisstudy. We demonstrate the potential and versatility of NNM fordeveloping DSVE particularly useful for inferential productcomposition estimation and control in a high purity distillationcolumn. A nonlinear simplified decoupler is developed based on aneural network model(NNM). The NNM decoupler is applied to asimulated CSTR and distillation column, respectively. Theresults of the simulation study show the good capability of theNNM decoupler for multivariable systems.The crude fractionationcolumn is a complicate process. It contains 13 manipulatedvariables and 7?8 controlled variables. In this study, a crudefractionation column model is developed based on a neuralnetwork. The predicted results of neural network model match theplant operating data well. A combination of the neural networkmodel and a successive quadratic programming (SQP) algorithm isemployed to determine the optimal operating conditions of thecrude fractionation column.Many control strategies applied to ahigh purity distillation column with a sidestream areillustrated in this study. We compare the differentialtemperature control and double differential temperature control,the direct internal reflux control and indirect internal refluxcontrol.The simulation results show the excellent capability ofthe feed-forward control to disturbance rejection. An indirectmodel-base control strategy applied to product compositioncontrol of a high purity distillation column is also developed.The indirect internal reflux control, feed-forward control, andindirect model-base control strategies are successfullyimplemented to a commercial distillation column.