| 摘要: | 在本論文中使用兩組四個參數的描述因子，第一組包括：ChiA_B(e)、SpMax_EA(bo)、H7s和DLS_01，第二組包括：RSC、SMC、DHB和MPE來估計107個高分子的玻璃轉化溫度，上述描述因子是分子結構、拓撲和量子化學的相關函數。本研究中分別以多元線性迴歸分析與遺傳規劃法來建造玻璃轉化溫度的估計模式。在本研究中透過調整族群代數、族群大小、模型複雜度、基因樹深度和基因組數量等參數來獲得效能更好的遺傳規劃法估計模式，經過訓練組的訓練與驗證組的驗證測試，第一組分別得到了低度複雜度遺傳規劃法估計模式其訓練組均方根誤差為14.88K(R2=0.9247)與驗證組16.51K(R2=0.91062)，中度複雜度遺傳規劃法估計模式其訓練組均方根誤差為14.55K(R2=0.92787)與驗證組15.54K(R2=0.9203)以及高度複雜度遺傳規劃法估計模式其訓練組均方根誤差為13.81K(R2=0.923547)與驗證組13.78K(R2=0.93889)的結果，第二組分別得到了低度複雜度遺傳規劃法估計模式其訓練組均方根誤差為14.18K(R2=0.9259)與驗證組19.18K(R2=0.91812)，中度複雜度遺傳規劃法估計模式其訓練組均方根誤差為13.82K(R2=0.92952)與驗證組18.36K(R2=0.92493)以及高度複雜度遺傳規劃法估計模式其訓練組均方根誤差13.06(R2=0.93715)與驗證組15.40K(R2=0.95164)的結果。測試結果與其他現存的估計模式比較，顯示以遺傳規劃法模式估計聚苯乙烯高分子玻璃轉化温度可以得到較為準確的估計值。
In this study, two sets of four parameter description factors are used. The first set includes: RSC, SMC, DHB and MPE, and the second set includes: ChiA_B(e), SpMax_EA(bo), H7s and DLS_01 to estimate the glass transition temperature of 107 polimers. The above described factors are related functions of molecular structure, topology and quantum chemistry. In this study, multiple linear regression analysis and genetic programming were used to construct the estimating model of the glass transition temperature. In this study, we adjusted the parameters of population algebra, population size, model complexity, gene tree depth, and genome number to obtain a more efficient genetic programming estimation model. After training by the training group and verification tests by the verification group, the first group obtained the low-complexity genetic programming method estimation model ,the root mean square error of the training group was 14.88K (R2=0.9247) and the verification group was 16.51K (R2=0.91062), in the medium complexity genetic programming method, the root mean square error of the training group is 14.55K (R2=0.92787) and the verification group is 15.54K (R2=0.9203) and the high complexity genetic programming method estimates the root mean square error of the training group is 13.81K (R2=0.923547) and the verification group is 13.78K (R2=0.93889). The second group obtained the low-complexity genetic programming estimation model, the root mean square error of the training group was 14.18K (R2=0.9259) and the verification group was 19.18K (R2=0.91812) , the medium-complexity genetic programming method estimates the root mean square error of the training group is 13.82K (R2=0.92952) and the verification group 18.36K (R2=0.92493) and the high complexity genetic programming method estimates the root mean square error of the training group is 13.06K (R2=0.93715) and the verification group is 15.40K (R2=0.95164). Compared with other existing estimation models, the test results show that the genetic programming method can be used to estimate the glass transition temperature of polystyrene polymers to obtain more accurate estimates. |
