應用適應性模糊類神經網路預測科學園區污水廠放流水水質

Using Adaptive fuzzy Neural Network to Predict Effluent Quality from Science Park Sewage Treatment Plant

卓宥愉(Yu-yu Cho)

指導教授 ： 張迪惠 ； 白子易

朝陽科技大學/理工學院/環境工程與管理系/碩士(2013年)

摘要

本研究採用適應性模糊類神經推論系統(Adaptive Network Based Fuzzy Inference System, ANFIS)，探討科學園區污水處理廠之出流水懸浮固體物 (Suspended Solids, SS)、化學需氧量 (Chemical Oxygen Demand, COD)的水質變化。建立科學園區污水處理廠模擬架構，並對其出流水SS、COD水質進行模擬預測，最後對模型預測值與實際值進行平均絕對百分誤差值之評析，以評估模型之預測效能。

本研究中，當預測SS時，輸入層採取三沉出流SS、二沉出流（二期）SS、三沉出流（二期）SS等3個輸入變數；輸出層採放流SS進行預測。當預測COD時，輸入層採取三沉出流BOD、二沉出流（二期）BOD、三沉出流（二期）BOD等3個輸入變數；輸出層採放流COD進行預測。預測模式以ANFIS之不同隸屬函數三角形、梯形、鐘形、高斯等隸屬函數進行個別模擬預測比較。SS模擬結果以三角形為最好、梯形次之、最後分別為鐘形以及高斯之隸屬函數；COD模擬結果以鐘形為最好、高斯次之、最後分別為梯形以及三角形之隸屬函數。

三角形隸屬函數SS模擬過程中採取2個輸入變數對1個輸出變數時，其平均絕對百分誤差值（Mean Absolute Percentage Error, MAPE）為22.55％，相關係數（Correlation Coefficient, R）為-0.527，MSE值為0.974，RMSE值為0.987；鐘形函數COD模擬過程中採取1個輸入變數對1個輸出變數時，其絕對百分誤差值（Mean Absolute Percentage Error, MAPE）為13.49％，相關係數 （Correlation Coefficient, R）為0.343，MSE 值為10.789，RMSE 值為3.285。整體而言，以ANFIS預測科學園區污水廠之SS及COD出流濃度結果顯示MAPE值大部分都為合理範圍，表示對於預測濃度及變動趨勢皆可掌握，故利用類神經網路來預測科學工業園區污水廠出流水濃度是可行的。本研究相關結果可供科學園區污水處理廠操作診斷之參考。

關鍵字

污水處理廠 ； 適應性模糊類神經推論系統 ； 平均絕對百分誤差值 ； 科學工業園區

並列摘要

In this study, adaptive network based fuzzy inference system (ANFIS) was used to explore the variation of effluent from the wastewater treatment plant of scientific park. The simulation framework for the wastewater treatment plant of scientific park was established, the effluent quality was predicted, the errors between the predicted values and the observed values were analyzed, and the model performance was evaluated.

When predicting effluent suspended solids (SS), the SS from the third clarifier, the SS from the second clarifier (second stage), and the SS from the third clarifier (second stage) were taken as the input parameters. When predicting effluent chemical oxygen demand (COD), the COD from the second clarifier, the COD from the third clarifier, and the effluent biochemical oxygen demand were taken as the input parameters. The membership functions (MFs) including triangle, trapezoidal, bell, and Gaussian MFs were adopted in prediction simulation for comparisons. When predicting SS, the performance in which the bell MFs were adopted prevailed others. When predicting COD, the performance in which the trapezoidal MFs were adopted prevailed others.

When predicting SS, the mean absolute percentage error (MAPE) was 25.24 %, the correlation coefficient (R) was 0.347 with bell MFs and three input parameters. When predicting COD, the MAPE was 19.24 %, the R was 0.895 with trapezoidal MFs and three input parameters.

並列關鍵字

wastewater treatment plant ； Scientific Park ； Mean Absolute Percentage Error (MAPE) ； Adaptive Network Based Fuzzy Inference System

參考文獻

5. 吳彬榮，「應用類神經網路預測科學園區污水廠二級出流水水質」，碩士論文，朝陽科技大學，台中（2011）。

6. 阮成隆，「以倒傳遞類神經網路及多元線性迴歸模擬建築工地對台中縣粒狀污染物之影響」，碩士論文，朝陽科技大學，台中（2010）。

10. 蘇志朋，「以倒傳遞類神經網路及多元線性回規模擬建築工地對台中市粒狀污染物之影響」，碩士論文，朝陽科技大學，台中（2010）。

19. Pai T.Y., Y.P. Tsai, H.M. Lo, C.H. Tsai and C.Y. Lin. ”Grey and neural network prediction of suspended solids and chemical oxygen demand in hospital wastewater treatment plant Effluent”, Computers & Chemical Engineering. (2007a).

20. Pai T.Y., T.J. Wan, S.T. Hsu, T.C. Chang, Y.P. Tsai, C.Y. Lin, H.C. Su and L.F. Yu, “Using fuzzy inference system to improve neural network for predicting hospital wastewater treatment plant effluent,” Computers & Chemical Engineering. (2009a).