106
W. Xinying et al.
compared with the shallow structure neural network is that the hidden layer of the
network structure is more “deep” and can better characterize the complexity [3].
High-dimensional functions, optimized information sharing, and stronger representation. It can make up for the shortcomings of neural network applications and achieve
better results. Deep learning provides a new method for predicting the remaining life
of oil pipelines.
The deep learning network is applied to the field of pipeline residual life prediction, because deep learning has the advantage of approximating arbitrary continuous
functions with arbitrary precision. The principle of the method is to use the acoustic
emission detector to analyze certain parameters of the pipeline, such as energy and
amplitude, when the pipeline is corroded. As a training sample, the DBN of the deep
confidence network is trained, the weight is adjusted, and the model is established
[4]. Then, the corresponding oil pipeline parameters are input into the trained DBN
network model in real time, and the deep reliability network DBN judges whether
the oil pipeline is corroded, and predicts the remaining life under the existing environmental stability conditions. This method improves the reliability of the remaining
life prediction.
10.2 Basic Theory
10.2.1 Basic Ideas
Deep learning simulates the process of human brain analysis problems, and combines
and learns low-level data to form more abstract high-level feature representations
(attributes, categories, etc.) to improve the accuracy of subsequent recognition and
classification.
Suppose system S has n layers, S 1 ,…,S n , whose input is 1 and output is 0. If
output 0 is equal to input 1, that is, input 1 does not change after passing through
this system, which means input 1 passes through each layer of S i . There is no loss of
information, or the lost information is redundant. The basic idea of deep learning is to
stack multiple layers and use the output of the previous layer as the input of the next
layer. In this way, the hierarchical representation of the input information is realized.
It can be seen that this layered unsupervised feature learning is an important basis
for deep learning. It maps the feature representation of the sample in the original
space into a new feature space through layer-by-layer feature transformation, with
a large number of simple Features build complex representations that eliminate the
impact of changes in input data that are independent of learning tasks on learning
performance [5], and retain information useful for learning tasks.
Deep learning can be understood as the extension and development of neural
networks. It is a new field in machine learning research. The motivation is to establish
a neural network that simulates human brain for analysis and learning. It interprets
data according to the mechanism of human brain and forms a low-level feature.
W. Xinying et al.
compared with the shallow structure neural network is that the hidden layer of the
network structure is more “deep” and can better characterize the complexity [3].
High-dimensional functions, optimized information sharing, and stronger representation. It can make up for the shortcomings of neural network applications and achieve
better results. Deep learning provides a new method for predicting the remaining life
of oil pipelines.
The deep learning network is applied to the field of pipeline residual life prediction, because deep learning has the advantage of approximating arbitrary continuous
functions with arbitrary precision. The principle of the method is to use the acoustic
emission detector to analyze certain parameters of the pipeline, such as energy and
amplitude, when the pipeline is corroded. As a training sample, the DBN of the deep
confidence network is trained, the weight is adjusted, and the model is established
[4]. Then, the corresponding oil pipeline parameters are input into the trained DBN
network model in real time, and the deep reliability network DBN judges whether
the oil pipeline is corroded, and predicts the remaining life under the existing environmental stability conditions. This method improves the reliability of the remaining
life prediction.
10.2 Basic Theory
10.2.1 Basic Ideas
Deep learning simulates the process of human brain analysis problems, and combines
and learns low-level data to form more abstract high-level feature representations
(attributes, categories, etc.) to improve the accuracy of subsequent recognition and
classification.
Suppose system S has n layers, S 1 ,…,S n , whose input is 1 and output is 0. If
output 0 is equal to input 1, that is, input 1 does not change after passing through
this system, which means input 1 passes through each layer of S i . There is no loss of
information, or the lost information is redundant. The basic idea of deep learning is to
stack multiple layers and use the output of the previous layer as the input of the next
layer. In this way, the hierarchical representation of the input information is realized.
It can be seen that this layered unsupervised feature learning is an important basis
for deep learning. It maps the feature representation of the sample in the original
space into a new feature space through layer-by-layer feature transformation, with
a large number of simple Features build complex representations that eliminate the
impact of changes in input data that are independent of learning tasks on learning
performance [5], and retain information useful for learning tasks.
Deep learning can be understood as the extension and development of neural
networks. It is a new field in machine learning research. The motivation is to establish
a neural network that simulates human brain for analysis and learning. It interprets
data according to the mechanism of human brain and forms a low-level feature.