2002-04-012024-05-14https://scholars.lib.ntu.edu.tw/handle/123456789/660418摘要：本研究將藉由收集土石流發生之雨量資料、氣象資訊、臨界條件、發生時間、流速、流量、衝擊力等各項資料，將之實際運用於不同之類神經網路中，以評估不同之類神經網路（包含前饋式與回饋式兩大類）於土石流警戒模式之可行性。並評判其準確度及實用的效果，且對土石流預警的時效性所能掌握的精確性及穩定性。並進一步完成土石流災害嚴重地區相關資料之蒐集、調查與彙整。 最後綜合上述，並以類神經網路強大的處理信息能力、優越的非線性映射能力以及高度的容錯能力，架構一以土石流預警為目的之類神經網路模式，且判斷模式的架構及操作方式是否能達到預期所需，並針對模式建立時所遇到的困難及模式本身所缺乏亦或是不足的部份逐一進行探討及改善。使其可實際應用於土石流危險區域，以利事前防災、減災方式，以減少災害發生。並提供決策者即時、充足且可靠之資訊，以為其訂定決策之參考依據。<br> Abstract: In this study, the data which might cause of debris flow, such as weather conditions, critical conditions, time, velocity, discharge, will be collected and applied to build the neural network model. Several artificial neural networks including feedforward recurrent and forward networks will be evaluated by those collected data to determine the model that can be used for forecasting debris flow. The performance of the model will be evaluated by the applicability, accuracy and stability. Neural network is composed of a large number of interconnected processing units that can deal with mass information. It has good ability of functional mapping between input and output and provide an optimal function approximation. Thus the neural network can be used to construct a warning system of debris flow. In addition, the developed model can provide real-time useful information for the decision makers.類神經網路前饋式回饋式即時回饋學習演算法Artificial Neural NetworkFeedforwardFeedbackReal Time Recurrent Learning