What is a machine learning?
Machine learning is a kind of application of artificial intelligence, it is based on the concept: machine should be given the data access and I am learning a specific task, without a clear programming, machine learning system by calculation of learning from data and forecast, when dealing with large data sets, it is particularly useful, because it can detect mode and predict (in some cases, the recommended) as a result, its ability to learn from the data and construction experience, in contrast to the deterministic method, which requires the user instructions and human knowledge,
Machine learning and earth science
Paradigm SeisEarth democracy in neural network association
Machine learning application of earth science data have been used for more than 25 years, along with the growth of oil technical data, they are now has become a practical need,
In the available computing systems, inspired by biological neural network of artificial neural network (ANN) is the most commonly used in oil and gas industry, used to handle the increasing seismic and well data, artificial neural network calculation is the study of adaptive node network, the network learning based on data exposure and experience to perform a task, usually without any specific task the rules of programming, they aim to way to categorize information similar to that of the human brain, for example, they can be taught to recognize the image and categorize them according to their different element, the
Emerson is the implementation of advanced, mature and reliable machine learning solutions pioneer,
Our technology based on machine learning can from a large number of different types of data is described in the underground, this allows the user to:
Describe and explain the existing results
Predict what will happen to
Suggestions for the decision of risk management and
This strategy allows the machine learning and prediction analysis is applied to the exploration, development and production optimization,
Machine learning methods can be divided into supervised learning and unsupervised learning,
Unsupervised learning
Input data not mark (soft); Goal is to find the data points of similarity, so that the similar data points together, through the derivation that exist in the input data to prepare the model structure, can be systematically through mathematical processes to reduce redundancy, or through similarity to organize data, some algorithms including clustering, dimensionality of association rule learning,
Emerson unsupervised machine learning example
Self-organizing map (SOM) gaussian process
Back propagation neural network - K means clustering analysis
Ascending hierarchical clustering self-growth neural networks
Dynamic clustering of principal component analysis (PCA)
Mixed classification based on graphic multiresolution clustering
Supervised learning
Input data with known label (hard data), model is prepared through the training process of need to predict, and the prediction error correcting, training process continues until the accuracy of the model to achieve the training data, include classification and regression,
Emerson's overseeing learning example
Neural network "collection" or democratic neural network association (DNNA) convolutional neural networks
Deep learning
In recent years, the amount of data and the number of surge, the development of the GPU and hardware falling prices led to the further study of advanced AI algorithms, there are already a new machine learning technique can handle big data: deep learning,
Deep learning is a kind of machine learning algorithms:
Using cascade multi-level nonlinear feature extraction and conversion processing units, each successive layer using the previous output as input,
Learning supervision (for example, classification) and/or unsupervised (pattern analysis, for example),
Learning and different levels of abstraction, the multiple levels of the corresponding said
Emerson using machine learning solutions
First pick seismic data transformation filter (Echos)
Well logging and seismic multi-attribute classification (Geolog SeisFacies)
Classification of rock types (SeisEarth, both)
Seismic multi-attribute classification (SeisFacies)
Prestack seismic data interpretation
Seismic facies classification (Stratimagic)
Electric facies classification (Geolog Facimage)
Neural network seismic inversion (both)
Uncertainty analysis and history matching (RMS, Tempest, SKUA - GOCAD)
Well logging and seismic data clustering (SKUA - GOCAD)
Data reduction (SeisFacies)