Small Sample Size and Experience-Independent Hydrate and Pipeline Leakage Identification Technique for Natural Gas Pipelines Based on Deep Forest

Hydrate blockage and pipeline leakage are two common factors that threaten the safety of natural gas pipelines. However, most of the current research focuses on nonintrusive, passive-like techniques that can only detect one of these abnormal events, with occasional attention to identification technique. This paper introduces an active method to simultaneously detect hydrate blockage and pipeline leakage using intrusive sensors, and further presents a deep forest-based classification method for two types of abnormal events, which aims to avoid the problem that the classification of traditional deep learning depends on huge number of hard-to-acquire samples. Besides, network structure and parameters in deep learning affect the classification performance, and deep forest is just a better solution to this problem. The parameter tuning experiments results of deep forest show that the classification accuracies are mostly 100% whatever in training and testing, proving that different parameter settings have little effect on the classification accuracy. The stability and portability of the classification method are tested, and it is verified that this classification method is easy to implement and has strong universality, which is expected to be applied to other types of natural gas pipeline event classification.