透過您的圖書館登入
IP:3.136.154.103
  • 期刊

應用類神經網路預測台灣西南沿海地層下陷量之研究

Artificial Neural Networks for Predicting the Land Subsidence in SW Coastal Area, Taiwan

摘要


台灣西南沿海地區之地層構造較為鬆軟,故有因為土層堆積而造成緩慢地層自然沈陷之現象。然而,近年來,由於該地區土地過度開發與利用,使得區域內地層下陷的幅員與速度產生加快的現象,又因為區域內國家各項重大工程建設陸續發展、營運。因此,西南沿海地區地層下陷量之影響與危害評估,已成為一項重要的研究主題。本研究利用區域內653個TWVD2001一等水準點的高程變動速度量,應用類神經網路(Artificial Neural Networks, ANNs)建立區域內地層下陷量之預測模式,並另外選擇37個一等水準網之節點為獨立檢測點,以評估模式的預測精度。由測試結果顯示,運用20個神經元所建立的前饋式類神經網路(Feed-Forward Neural Networks, FNNs),其預測值的均方根誤差(Root Mean Square Error, RMSE)約為±5.21 mm/yr,當可有效預測台灣西南沿海區域內之地層下陷量。

並列摘要


The stratum of SW coastal area in Taiwan has lain in an uncompressed state. It has not been avoided that of naturally sinking activities caused by the land amassed phenomena. However, because of the over-developed and the overused activities upon this region these years, the range and the speed of land subsidence have been expanded and accelerated. Considering the safeties of several greatly national constructions have been developed and operated within this area, the issue of estimating the effects and damages of land subsidence has become to be a research topic of importance, recently. In this paper, 653 benchmarks established in SW Taiwan were selected from the TWVD2001 leveling networks, and the artificial neural networks (ANNs) were used to predict the land subsidence in this region. The 37 node points located in SW area of TWVD2001 were independently selected and used to check the predicted accuracy of the four designed ANNs models. Based on the checking results, it was found that the Root Mean Square Error (RMSE) of predicted errors could reach to ±5.21 mm/yr when the Feed-Forward Neural Networks (FNNs) model with 20 artificial neurons were applied. It was also showed that FNNs model could be effective in predicting the land subsidence in SW coastal area, Taiwan.

延伸閱讀