摘要
本文提出一類神經網路漏水估測方法，此方法是將漏水模型置於EPANET供水管網以模擬估測水壓及流量；並藉由變換不同的漏水位置與洩漏孔徑以取得類神經網路學習所需的漏水數據，且以個別管段分別進行漏水訓練。而在學習訓練的過程中，可以依據各管段的訓練收斂狀態得知是否需要新增水壓計或流量計等感測元件，以作為部署建置的決策依據。利用此分段類神經網路漏水估測，不僅可快速偵測出可能的漏水管段，且能找到漏水位置及漏損孔徑。本研究之方法較傳統擴散器漏水模型更易實現，也較過去以校正摩阻係數或用水需量等管網參數差異來定位可能漏損的方法更為直覺和易於理解；且非只找到可能的漏水區域，而是更快速地估測出確切的漏水管段、漏水位置及漏損大小；不僅可有效降低查漏設備的建置成本及可縮短查測時間，也將達成節水之實值效益。
並列摘要
A neural network algorithm is provided for water leakage estimation. In this algorithm, a leakage model is adopted to simulate the EPANET water distribution network for acquiring the measured water pressure and flow rate. By adjusting the leakage location and the leakage aperture of each fictitious pipe, we can obtain the training data for the neural network. In the training process, the decision of whether a new water sensor deployment is needed will be found according to the difficulty of training convergence. Then the neural network will be used to estimate the leak pipe, leakage position and leakage volume quickly. Our method is easier to implement than the traditional emitter-type leakage model, and is more intuitive and simpler to understand than the means of calibration parameters, such as friction coefficient or water demands. It can not only reduce the cost of leak testing equipment implementation, but also can shorten the detection and checking period effectively. The benefits of water-saving will be practically acquired.
延伸閱讀
- 陳佳瑩（2022）。應用深度學習與給水管戶籍化演算法統計分析漏水潛能〔碩士論文，國立臺灣師範大學〕。華藝線上圖書館。https://doi.org/10.6345/NTNU202200069
- Wang, Y. Y. (2006). 整合類神經網路與遺傳演算法於水庫水質管理之研究 [doctoral dissertation, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU.2006.10212
- 陳鵬安（2017）。機器學習法於水庫異重流到達時間及排砂效率預報之研究〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU201702086
- LIN, Y. P., LIEN, W. Y., CHEN, H. Y., HE, J. H., & CHOU, C. F. (2020). 應用機器學習方法於灌溉水質預測. 台灣水利, 68(1), 1-14. https://doi.org/10.6937/TWC.202003/PP_68(1).0001
- 鄭家豪（2008）。改良式倒傳遞類神經網路於水庫入流量預報之研究〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2008.01346