海洋科学
海洋科學
해양과학
MARINE SCIENCES
2014年
11期
123-131
,共9页
尤加春%毛慧慧%段文豪%李红星
尤加春%毛慧慧%段文豪%李紅星
우가춘%모혜혜%단문호%리홍성
双相-随机介质%模糊C均值聚类(FCM)%支持向量机(SVM)%差分进化算法
雙相-隨機介質%模糊C均值聚類(FCM)%支持嚮量機(SVM)%差分進化算法
쌍상-수궤개질%모호C균치취류(FCM)%지지향량궤(SVM)%차분진화산법
two-phase and random media%fuzzy C means (FCM)%support vector machine (SVM)%evolutionary difference al-gorithm
在总结了目前海底底质分类研究的基础之上,率先提出利用计算机数值模拟技术对海底底质进行分类识别研究。相较于目前海底底质分类研究中所使用的水槽实验法,提出采用计算机数值正演技术模拟实际地震勘探中数据采集过程。在分类识别算法上,分别采用支持向量机(SVM)和模糊 C 均值聚类(FCM)算法对采集的数据进行分类,为使支持向量机分类识别率达到最大,引入差分进化算法对支持向量机中关键参数进行最优化搜索,并研究了向原始地震记录中加入10%,30%,50%的高斯白噪音时算法的稳定性。在分析了这两种算法分类识别的正确率及其各自的优缺点后,提出了海底底质分类识别的两步法,即(1)先利用模糊 C 均值聚类进行一粗糙的预测分类,在每一类中挑选聚类性较好的数据作为支持向量机的训练样本;(2)将上一步中筛选的样本作为支持向量机的训练样本,并用差分进化算法优化支持向量机分类参数,再利用训练好的支持向量机对其余数据做预测分类。鉴于计算机数值模拟的可重复性、高效快速性及本文提出的模糊C均值聚类-支持向量机方法的鲁棒性,为便于开展进一步研究,归纳总结了一套行之有效的采用计算机数值模拟技术开展海底底质分类识别研究的一般化流程。
在總結瞭目前海底底質分類研究的基礎之上,率先提齣利用計算機數值模擬技術對海底底質進行分類識彆研究。相較于目前海底底質分類研究中所使用的水槽實驗法,提齣採用計算機數值正縯技術模擬實際地震勘探中數據採集過程。在分類識彆算法上,分彆採用支持嚮量機(SVM)和模糊 C 均值聚類(FCM)算法對採集的數據進行分類,為使支持嚮量機分類識彆率達到最大,引入差分進化算法對支持嚮量機中關鍵參數進行最優化搜索,併研究瞭嚮原始地震記錄中加入10%,30%,50%的高斯白譟音時算法的穩定性。在分析瞭這兩種算法分類識彆的正確率及其各自的優缺點後,提齣瞭海底底質分類識彆的兩步法,即(1)先利用模糊 C 均值聚類進行一粗糙的預測分類,在每一類中挑選聚類性較好的數據作為支持嚮量機的訓練樣本;(2)將上一步中篩選的樣本作為支持嚮量機的訓練樣本,併用差分進化算法優化支持嚮量機分類參數,再利用訓練好的支持嚮量機對其餘數據做預測分類。鑒于計算機數值模擬的可重複性、高效快速性及本文提齣的模糊C均值聚類-支持嚮量機方法的魯棒性,為便于開展進一步研究,歸納總結瞭一套行之有效的採用計算機數值模擬技術開展海底底質分類識彆研究的一般化流程。
재총결료목전해저저질분류연구적기출지상,솔선제출이용계산궤수치모의기술대해저저질진행분류식별연구。상교우목전해저저질분류연구중소사용적수조실험법,제출채용계산궤수치정연기술모의실제지진감탐중수거채집과정。재분류식별산법상,분별채용지지향량궤(SVM)화모호 C 균치취류(FCM)산법대채집적수거진행분류,위사지지향량궤분류식별솔체도최대,인입차분진화산법대지지향량궤중관건삼수진행최우화수색,병연구료향원시지진기록중가입10%,30%,50%적고사백조음시산법적은정성。재분석료저량충산법분류식별적정학솔급기각자적우결점후,제출료해저저질분류식별적량보법,즉(1)선이용모호 C 균치취류진행일조조적예측분류,재매일류중도선취류성교호적수거작위지지향량궤적훈련양본;(2)장상일보중사선적양본작위지지향량궤적훈련양본,병용차분진화산법우화지지향량궤분류삼수,재이용훈련호적지지향량궤대기여수거주예측분류。감우계산궤수치모의적가중복성、고효쾌속성급본문제출적모호C균치취류-지지향량궤방법적로봉성,위편우개전진일보연구,귀납총결료일투행지유효적채용계산궤수치모의기술개전해저저질분류식별연구적일반화류정。
Basis on the previous research about the classification of sediment, Here, for the first time, we proposed the application of numerical simulation technology in classification of seabed sediment. Compared to the currently used flume method in the classification of sediment, we proposed the usage of computer technology to simulate the seismic acquisition process in practical exploration. In the classification and recognition algorithms, the support vector machine and fuzzy C mean clustering methods (FCM) were used to classify the data acquired. To maximize the accuracy of support vector machine (SVM), we introduced the evolutionary difference algorithm to optimize the values of the key parameters of support vector machine classification model. Then, to further study the stability of the methods, 10%, 30%and 50%of Gaussian white noise was added into the original data. After full analysis of the advantages and disadvantages of both approaches from its principle and classification accuracy, we designed a two-step classifier which combined the fuzzy C mean clustering and support vector machine. Firstly, the unsup-ervised classification algorithm, FCM, was used to classify the data preliminarily, and screened the samples with good clustering. Secondly, those selected data in step 1 were served as train samples of support vector machine model. The differential evolution algorithm was used to optimize the key parameters of support vector machine model, then the optimized support vector machine model was used to classify the remaining data. Finally, Based on the repeatable, convenient characters of the computer simulation and the relevant high accuracy and the robustness of FCM-SVM, a total solution of a classification, which will be easier, deeper, further to study the feature of reflec-tion from sediment is proposed in the article.