Читать книгу Internet of Things in Business Transformation - Группа авторов - Страница 18

Machine Learning and Data science has gained a lot of popularity in the last decade. In the earlier sections of the chapter, we demonstrated the usage of Neural Networks for detection of leaks in a pipeline, that itself was a pre cursor of introduction of machine learning in the field of leakage detection of pipeline transfer of hydrocarbon.

Several machine learning algorithms have started being used for detection of leaks. Advanced techniques such as Neural Networks [19] and Support Vector Machines [20, 21] have already been used and proved to have given excellent results. In addition, more advanced methods employing deep learning and convolutional neural networks [22] are also being explored, in fact application of Variational Autoencoders have already been tested and used.

We shall now discuss the technique of implementation of each of these methods (Neural Network and Support vector machine based) in brief, in addition, we shall also try to implement a novel strategy to use ensemble learning algorithms to detect leakages in pipelines [23].

1 a) Neural Networks-Based Strategy for Detection of Pipeline Leak DetectionThe overall architecture of the system is already designed in the section where leakage detection using digital signal processing is explained, in the section a representative system architecture. Here, we only analyze the details of the neural network from a very computational point of view. In the paper 3-layer neural network is used with a sigmoid activation function. In the method, the error is decreased by backpropagation.

2 b) Support Vector Machines-Based Strategy for Detection of Pipeline Leak DetectionWe have already seen the use of negative pressure wave method for leak detection in earlier section. We see that in negative pressure wave method, various computational methods are used to detect the leakage from the huge dataset that contains the leakage information as well as the noise from the pipeline, pipe fittings and environment. Use of these computational methods makes the model very expensive from a computational point of view. Therefore the use of Support vector machines in conjunction with negative pressure wave architecture proposed.Figure 1.7 Use of support vector machine for pipeline leak detection.A support is used to detect extreme cases—for our case the extreme cases are Leak or No Leak case. This is depicted in Figure 1.7.The data class is separated by hyper planes that divide both the classes clearly. The hyper planes are also called support vectors.Negative pressure wave method has pressure information from two different scenarios, one is when there is no leak and the pressure profile is usual and the other is when there is a leakage and the pressure profile has disturbances, support vector machines are used for correctly classifying these data and generating leads for detection of leak.A representative figure for showing the usage of Support Vector machines for negative pressure wave method is as follows,