발간년도 : [2018]
| 논문정보 |
|
| 논문명(한글) |
[Vol.13, No.6] An Unsupervised Real-time Anomaly Detection of Furnace System based on HTM |
|
| 논문투고자 |
Ky Vannroath, Jong-Min Bae, Hyun-Syug Kang |
|
| 논문내용 |
The immediate recognition of abnormal temperatures in the furnace system is very important for timely disaster control. Conventional abnormal temperature detection techniques typically use pre-programmed thresholds to identify abnormal patterns from a stream of temperature data in a statistical manner. Therefore, it is not suitable for use in a heating furnace system in which abnormal patterns must be detected in real time. Hierarchical Temporal Memory (HTM) is one of the machine learning models with excellent ability to analyze patterns of stream data. This model is a new machine intelligence technique that uses unsupervised continuous learning using Cortical Learning Algorithm (CLA), which is a time-based learning algorithm, and Spare Distributed Representation (SDR), which stores temporal and spatial patterns. This HTM, which is distinguished from existing deep learning or artificial neural network model, is capable of real-time continuous prediction from input of stream data and can be used to recognize abnormal states in various fields. HTM can be used as a robust model to detect abnormalities in successive temperature data in a heating furnace system because the furnace system continues to generate temperature stream data with a certain pattern over time. Here we developed an HTM-based, unsupervised, real-time anomaly detection system for heating furnaces. Performance evaluation shows that the suggested system can perform excellent abnormal real-time detection. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
