應用類神經網路與小波理論分析地震前地下水位波動

Applications of Artificial Neural Network and Wavelet Theory to Analyze the Fluctuation of Groundwater Level Before An Earthquake Appears

廖啟佑(Ci-Yu Liao)

指導教授 ： 陳彥璋

國立臺北科技大學/工程學院/土木與防災研究所/碩士(2005年)

https://doi.org/10.6841/NTUT.2005.00051

摘要

台灣地區由於地質複雜之特殊性，位處「歐亞大陸板塊」與「菲律賓海洋板塊」持續板塊擠壓，以致台灣每年地震活動頻繁，地震乃是大地能量釋放的自然演變過程，並非人力所能控制，經證實地震與地下水位異常變化有一定程度之相關係，加上台灣地區擁有健全地下水位觀測井網；於地震地下水研究上佔有先機，為尋求降低地震災害損失程度，極需深入探討地震地下水之相關機制。
本研究以嘉南強震觀測網中之六甲、那菝地震地下水觀測站井為研究對象，為能探討地震事件是否對地下水位產生異常行為，利用觀測的地下水位時序來進行解析，探討地震發生前後時間點之水位的變化量，而為避免地震事件受地潮、降雨等事件影響，致使小波分析之高頻異常診斷無法獲得準確之結果，故利用類神經網路將降雨、地潮及不規則訊號等影響因子之趨勢進行濾除。
將經類神經網路濾除受地潮、海潮及降雨等影響因子之地下水位序列資料，應用小波(Wavelet)分析理論，探索地下水位測站之長時程地下水水位之多分辦層結構及應用小波轉換計算地下水位時間序列之小波係數，經小波係數值之計算可評估地下水位各種交織在一起之混合訊號，分解成不同分辨層或不同頻率區塊訊號，再與Donoho和Johnstone對估測訊號所發展出來之小波收縮(Wavelet Shrinkage)方法，選取一合適的臨界值，將所得的高頻小波係數做修剪(clipping)收縮處理，再將門檻值以外之各高頻小波係數與地震發生時間做一分析整理，藉此可明顯指出地下水水位出現頻率異常之時間點，因而將有助於減災，延長避難反應時間。

關鍵字

地震 ； 類神經網路 ； 小波理論

並列摘要

Located on the Eurasia plate and the Philippine marine plate to push with the lasting, where earthquakes happen frequently, Taiwan, due to particularity with complicated geography character, was attack by earthquakes. Earthquake is a natural proceeding in which the earth releases energy. It is not the manpower that can be controlled, through verifying that relation with a certain level of education in groundwater level change and the earthquake appears. Besides, there are well network observing groundwater changes in Taiwan. For observing groundwater, we get certain advantage. Searching how to decrease the damage caused by earthquakes, we must study the correlation between groundwater and earthquakes.
Observe the monitoring station of Nabal and Liujar as the research object, at the Jarnan strong shock earthquake groundwater network in this research. In order to probe the earthquake incident, whether produce the unusual behavior to the groundwater level, make using observed of groundwater level time sequence to analyze the change amount of the groundwater level clicked, in time before and after an earthquake attack. In order to prevent the earthquake incident from being influenced by incidents, such as ground tide , rainfall ,etc. Causing the wavelet analyse obtain the accurate result, the unusually high frequency be unable diagnose. The Artificial Neural Network(ANN) is so utilizing to detrend of rainfall, ground tide, and irregular signal, etc.
Detrend of morning and evening tides, rainfall time series through kinds of ANN. Using the Wavelet theory to explore the long-time groundwater level examines the amount layer of structure separately, and uses the Wavelet transform and calculate the Wavelet coefficient, can assess various mixing interweaving kinds of the groundwater level signal by calculation of Wavelet coefficient value, resolve into and distinguish layer or different frequency block signals at differently level, and then, using Wavelet shrinkage method by Donoho and Johnstone, to estimating and examining the Wavelet coming out in development of the signal. Choose a suitable critical value, to clipping the high-frequency Wavelet coefficient. And then, using threshold to shrink wavelet coefficient, and clipping out the approximation function and the detail function of underground water level attack by just earthquake. Taking that obviously of this point out groundwater level appear frequency unusual time, therefore will contribute to reducing natural disasters, will lengthen and take refuge and reflect time.

並列關鍵字

Earthquake ； Artificial Neural Network ； Wavelet theory

參考文獻

[4] 王政偉，結合小波轉換與類神經網路辨別電力開關之切換，碩士論文，私立中原大學電機工程研究所，2003。

[18] 徐維聰，應用小波轉換與類神經網路進行電力系統故障種類之判別，碩士論文，私立中原大學電機工程研究所，2003。

[41] 楊翰宗，類神經網路應用於光纖雷射與聲波杜卜勒測速儀渠流速度剖面量測之研究，博士論文國立台灣大學生物環境系統工程學研究所，2005。

[13] 林進國，降雨和地下水位變化之關聯性分析，碩士論文，國立成功大學水利及海洋工程研究所，2003。

[14] 林穎駿，應用小波轉換與類神經網路於電力品質事故之監測與辨識，碩士論文，私立中原大學電機工程研究所，2003。