Derivative Failure of Compressor Station Analysis Based on Hawkes Process

In the circumstance of practical production, derivative failures is an important part of system failures in the compressor station systems. In order to prevent the occurrence of subsequent faults, we would like to analyze the trigger pattern between the failures. However, the existing methods rarely consider the time-stamp of the failures and find the rule how the trigger patterns change over time. Therefore, we intend to analyze the characteristics of the failure trigger pattern over time. After learning the law of trigger patterns of the derivative failures, which change with time-lapse, we can take precautionary measures to reduce the occurrence of derivative failures, which will bring economic and security benefits. Thus, in this paper, we utilize the Hawkes process to model and analyze the cause and trigger pattern. We label the sequences of fault events on nearly 90 compressor stations and use the sequences of failure events to derive failure cause analysis. We devise a non-parametric Hawkes process model and introduce two sparse regularizers to select the salient non-parametric basis function and circumvent the over-fitting. In the experiment, our proposed model outperforms two Hawkes process and two prior Poisson process baselines with intensities that follow the exponential distribution or Weibull distribution. We conduct a detailed analysis of the self-excitation modes and the other-excitation modes of the derivative faults, and based on the stable and unstable trigger pattern of the failures, we put forward some suggestions for the actual production process of the compressor stations.