Research On Time Series Data Modeling Method Combining Deep Learning And Point Process

With the rise of time series data in daily life and the Internet,time series prediction tasks have gradually been valued by people,and time series prediction tasks in different scenarios have specific value and significance.In the field of avionics equipment,using time series data to complete equipment failure prediction can not only reduce economic losses,but also avoid accidental casualties caused by equipment failure.At this stage,point processes are often used for time series prediction tasks,but the intensity function defined by traditional point processes is relatively simple,and generally can only be used for time prediction,which cannot meet more complex data scenarios and tasks.Deep learning uses neural networks to model the intensity function nonlinearly.The intensity function can be defined for different forms of time series data and prediction tasks.In addition to the completion time prediction,other relevant information can also be captured,which is just in line with the avionics failure prediction task.requirements.This paper will combine two technologies of deep learning and point process to realize the fault prediction of avionics equipment.The main work content and contributions are as follows:(1)First,this paper proposes a timing prediction model(JoAttition Hawkes Process,JoAtt-HP)that combines attention mechanism and point process for the characteristic timing data of avionics equipment.The introduction of the self-attention mechanism enables the model to process long-term historical sequences of electronic equipment in parallel,without forgetting the long-ago fault information over time,and then use the intensity function in the Hawkes process to predict the occurrence of equipment faults time,the experimental results on avionics fault time series datasets illustrate the effectiveness of the model.To illustrate the adaptability of the model,this paper also evaluates on other publicly available time series datasets.(2)Considering that the fault prediction task of avionics equipment is not limited to the prediction of failure time,but also needs to predict more fault information,this paper proposes a combination of time-series knowledge graph and point process method.Firstly,based on the fault log files of avionics equipment,the time series knowledge map in this field is constructed to form the fault time series data set AEDFI-TKG,and then a method is proposed to define the occurrence of fault facts in the time series knowledge map as a multivariate point process(Knowledge Graph Multi-Temporal Point Process,referred to as KG-MTPP),uses the score of the fault fact to adjust the intensity function,and completes the entity in the future fault event according to the intensity function,so as to complete the fault prediction of avionics equipment.