Font Size: a A A

Automatic Seismic Event Pickup Using Improved Artificial Neural Network

Posted on:2013-06-23Degree:MasterType:Thesis
Country:ChinaCandidate:Y C ZhangFull Text:PDF
GTID:2230330371483934Subject:Earth Exploration and Information Technology
Abstract/Summary:PDF Full Text Request
With the development of oil and gas exploration,the technich of exploration is more and more difficult.As the basis of oil and gas detection,automaticlly pickup of seismic horizons are becoming increasingly important. The actual geological horizon andseismic profiles on the same event are corresponded to each conterparts, so the actualseismic records on the events continuous tracking is a prerequisite for obtaining avalid layer information.Seismic event pickup is a process of tracking reflection waves from the sameinterface on the seismic records, which using seismic dynamics and kinematiccharacteristics. After hard efforts of plenty scholars, many algorithms have beendeveloped in order to track the trajectory of the seismic event, such as PatternRecognition、Cross-correlation algorithm and so on. All of these algorithms couldtrack the seismic event up.Artificial intelligence is the representative of the development of digitaltechnology, which the outstanding branch is artificial neural networks. It hasself-learning, self-adaptive functions, particularly applicable to solve the problemswhich is dominated by automatically pick up.This paper introduces the current development of seismic events and artificialneural network, illustrates three kinds of events pickup theories and validates thefeasibility of BP neural network. The detailed steps are as followed:1、Determine the required BP neural network structure. The input layer、outputlayer and hidden layer are determined by the purpose of network applications. Inputsamples and initial weights all affect the network’s computing power. After thecalculation of the actual seismic records, this paper decided to use a6-3-1networkstructure, selected the adequate samples of network to learn and pickup.2、Improving BP neural network achieve optimal results. Using momentummethod, the BP algorithm could find a better solution; using self-adaptive learningrate method, the BP algorithm could shorten the training time. Combine two methodstogether, could achieve a new method which achieve the best improvement, improvethe ability of network.3、Inputting the selected actual seismic profile data which with the typicalcharacteristics of seismic waves, to determined BP neural network, then automaticallypicking up seismic events. After all above procedures, conclusions are as follows:1、Using artificial neural networks could automatically pick up the seismicevents;2、Using artificial neural networks to automatically pick up the seismic eventsmust have a sufficient number of learning samples, and the distribution and nature ofthe sample should be more uniform;3、The effect of automatically pickup can be improved by increasing the types ofcharacteristic parameters, however the conclusions are to be further studied;4、Only appropriate choice of the required accuracy in the artificial neuralnetwork structure and learning process could achieve not only improves the effectsautomatically pickup, but also reduce the learning time.
Keywords/Search Tags:Artificial neural network, Improved BP network, Seismic events pickup, SeismicHorizon pickup
PDF Full Text Request
Related items