| 摘要: | 地震是重大的天然災害之一，往往造成數以千計的人民及財物的損傷。根據資料統計，全球每年地震發生的次數約為500萬次，而規模5級以上的地震次數約2000次。2016年2月台灣南部發生高雄美濃地震，是繼921大地震後最嚴重的地震，造成傷亡非常嚴重。地震發生時，若能爭取到數秒的時間做出立即應變，就能避免造成生命危險，因此，地震預警的需求已經成為不可忽視的重要議題。
地震預警除了必須快速地傳遞地震訊息之外，更需要精確又快速的演算法協助判別地震。目前美國、加拿大、日本、台灣皆投入地震預警相關的研究，然而以往地震預警的方法如：類神經網路、多樣本中位數差異檢定 (Kruskal-Wallis test)、傅立葉轉換、小波轉換、向量支持機 (Support vector machine) 等等，皆為複雜度較高的演算法，若應用於普及的裝置如智慧型手機、平板電腦、IoT裝置等等，可能導致運算量過高而無法發揮及時性。現今的地震預警系統期望能使用大量裝置形成地震網絡以增加回報的可靠度，因此計算複雜度過高的演算法不利於一般設備上實現。
本論文以輕量化地震預測演算法的角度出發，希望藉由降低地震預測演算法之複雜度加速判別地震，並且利用大量的真實地震事件作為分析的樣本，驗證演算法達到精確及快速的判別。
;Earthquake is one of the major natural disasters, which could kill or injure thousands of people and cause huge property loss. According to the statistics, the number of earthquake events is about five million times per year and two thousands of them exceed magnitude 5. If we can win few seconds before the earthquake comes, it may save lots of lives and reduce economic losses. The earthquake early warning becomes an issue that cannot be ignored.
Earthquake early warning system (EEWS) needs rapid transmission of seismic information. Moreover, it requires accurate and fast algorithm to support the detection of earthquakes. In the past decades, progress has been made to invest the EEWS in countries where earthquake occurs frequently. For example, the United States of America, Canada, Japan and Taiwan have participated in doing the researches of EEWS. The earthquake warning detection methods such as: Artificial Neural Networks, Kruskal-wallis test, Fourier transform, Wavelet transform, Support Vector Machine are potential algorithms with high complexity. Nowadays, the EEWS is expected to use a large number of devices to form an earthquake detection network to increase the reliability. However, the algorithms with high computation complexity are not conducive to be implemented on general devices, such as smart phones, tablets or IoT-devices.
In this thesis, we aim to reduce the complexity of the seismic algorithm. To accomplish it, we use a large number of real earthquake events as the analysis samples to verify the algorithm and improve accuracy. |
