水庫集水區的治理、開發與操作,常會遭遇地表土壤沖蝕所產生的非點源污染。為了能有效防止此類災害的再次發生,並隨時監測集水區的整治情況,需要建立完備的懸浮固體濃度即時監測系統。本研究以中華民國行政院環境保護署石門水庫水質監測數據查詢資料庫中1993-2005年間的資料來進行分析,選取比電導度、溶氧、酸鹼值、濁度、溫度、採樣月份及葉綠素α,利用倒傳遞類神經網路來進行訓練及驗證,試驗結果顯示,即時推估懸浮固體濃度變化的趨勢十分良好,網路輸出與期望輸出的判定係數R2達到0.85,而均方差為23.21。將資料以移動平均濾掉人為或儀器誤差所產生的這些短週期的訊號,判定係數R2更達到0.92。本研究進一步建置倒傳遞類神經網路模式架構使其可以對懸浮固體濃度做動態即時預測,試驗結果顯示,預測懸浮固體濃度變化的趨勢符合十分良好,網路輸出與期望輸出的判定係數R2達到0.67,而均方差24.37。結果顯示,利用倒傳遞類神經網路配合其他現地數位化水質自動記錄監測儀可快速準確的推估預測懸浮固體濃度。
In the management of reservoirs, non-point source pollutions caused by surface soil erosion are frequently encountered. In order to prevent this kind of problem and monitor the reservoir at any moment, it is necessary to establish a complete real time TSS monitoring system. The data of the water quality of Shimen reservoir used in this study were provided by the Environmental Protection Administration of the Executive Yuan, R.O.C., which included electrical conductivity, dissolved oxygen, pH value, turbidity, temperature, month, and chlorophyll-α, in the period from 1993 to 2005. The method of back propagation network was used to perform the training and improving. From the results, it showed that the variation tendency of real time suspended solids concentration output estimated by using network agrees well with that in expected output (R2=0.85, MSE=23.21). When the moving average was used to reduce the short cycle signal deviation caused by man-made or device causes, the R2 can reach 0.92. When BPN was modified to predict the real time TSS, the results showed that the predicted variation tendency of TSS in network output agrees well with that in expected output (R2=0.67, MSE=24.37). It could be concluded that the method of BPN and other on site digital water-quality auto recorded monitors can be used to rapidly and accurately estimate TSS.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。