- 學位論文
丙酮-甲醇-水氣液相平衡之類神經網路
The Artificial Neural Network of Vapor-Liquid Equilibria for Acetone-Methanol-Water System
摘要
本實驗室所設計蒸餾塔模擬程式，以NRTL熱力學模式計算丙酮-甲醇-水的氣液相平衡數據，相當繁瑣且費時，本研究希望利用前饋式類神經網路程式擬合此三成份系統之氣液相平衡數據，尋找最佳類神經網路模式取代NRTL熱力學模式，藉而簡化蒸餾塔模擬的計算時間。 本實驗室所設計蒸餾塔模擬程式，假設塔內壓力皆為一大氣壓進行熱力學數據計算，故首先尋找一大氣壓下氣液平衡數據之類神經網路模式。考量將來模擬程式可能會取消ㄧ大氣壓的假設，故本研究試著尋找壓力範圍為0.9到1.26個大氣壓氣液平衡數據之類神經網路模式。 本研究所求得一大氣壓下類神經網路模式的最大絕對誤差，丙酮莫耳分率為0.006504，甲醇莫耳分率為0.004261，溫度為0.025000K。0.9到1.26個大氣壓下類神經網路模式的最大絕對誤差，丙酮莫耳分率為0.001812，甲醇莫耳分率為0.001475，溫度為0.092000K。由此顯示本研究所得類神經網路模式具有不錯的實用性。
並列摘要
The distillation column simulation program developed by our laboratory calculated the acetone - methanol - water vapor-liquid equilibrium data very tedious and time consuming by NRTL thermodynamic model. The feed-forward neural network program was used to fit the vapor-liquid equilibrium data of the ternary system. It is expected that a neural network model can be found to replace the NRTL thermodynamic model, therefore to simplify the calculation of distillation column simulation. Simulation program assumed that the pressure of distillation column is 1atm. Therefore the neural network model of vapor-liquid equilibrium data at a pressure of 1atm was found firstly. Because simulation program may remove the assumption about pressure is 1 atm in the future, the study try to search the neural network model of vapor-liquid equilibrium data at the pressure of 0.9~1.26 atm. In the study, the largest absolute error of neural network model at a pressure of 1atm, the mole fraction of acetone is 0.006504, the mole fraction of methanol is 0.004261, and the temperature is 0.025000K. The largest absolute error of neural network model at the pressure of 0.9~1.26atm, the mole fraction of acetone is 0.001812, the mole fraction of methanol is 0.001475, and the temperature is 0.092000K. The research shows that the neural network model has good practicability.