DOI QR코드

DOI QR Code

Forecasting Model Design of Fire Occurrences with ARIMA Models

ARIMA모델에 기반한 화재발생 빈도 예측모델의 설계

  • Ahn, Sanghun (Department of Chemical Engineering, Myongji University) ;
  • Kang, Hoon (Department of Chemical Engineering, Myongji University) ;
  • Cho, Jaehoon (Department of Chemical Engineering, Myongji University) ;
  • Kim, Tae-Ok (Department of Chemical Engineering, Myongji University) ;
  • Shin, Dongil (Department of Chemical Engineering, Myongji University)
  • Received : 2015.03.10
  • Accepted : 2015.04.07
  • Published : 2015.04.30

Abstract

A suitable monitoring method is necessary for successful policy implementation and its evaluation, required for effective prevention of abnormal fire occurrences. To do this, there were studies for applying control charts of quality management to fire occurrence monitoring. As a result, it was proved that more fire occurs in winter and its trend moves yearly-basis with some patterns. Although it has trend, if we apply the same criteria for each time, inefficient overreacting fire prevention policy will be accomplished in winter, and deficient policy will be accomplished in summer. Thus, applying different control limits adaptively for each time would enable better forecasting and monitoring of fire occurrences. In this study, we treat fire occurrences as time series model and propose a method for configuring its coefficients with ARIMA model. Based on this, we expect to carry out advanced analysis of fire occurrences and reasonable implementation of prevention activities.

화재를 예방하기 위해서는 그에 필요한 정책과 평가가 필요하며, 화재 발생에 대한 적절한 모니터링 기법이 있어야 한다. 이를 위해 품질관리의 기법중의 하나인 관리도를 화재 모니터링에 적용하는 연구가 진행된 바 있다. 그 결과, 통계적으로 겨울에는 많은 화재가 발생한다는 것이 증명되었고, 매년마다 어떤 주기를 띠고 변화한다는 것이 발견되었다. 그럼에도 불구하고, 여름철과 겨울철에 동일한 기준(관리한계선)을 적용할 경우 상대적으로 발생건수가 과다한 겨울에는 과도한 예방활동이 진행되고, 여름에는 잠재이상이 있는 경우에도 발생건수 감소에 따른 착시현상으로 인해 활동이 부족해질 가능성이 있다. 이때, 각 계절별로 다른 관리 한계선을 적용시킨다면 합리적인 예측과 보다 효과적인 이상 패턴의 모니터링이 가능해질 수 있다. 따라서 본 연구에서는 발생건수를 대상으로 시간에 따라 변하는 시계열 모델을 사용하여 화재발생 빈도 예측 모델계수의 체계적인 설정과정을 ARIMA 모델을 기반으로 제안하였다. 이를 바탕으로 화재발생 패턴의 개선된 분석과 이에 기반한 보다 체계적인 예방활동을 진행할 수 있을 것으로 기대한다.

Keywords

References

  1. National Fire Incident Reporting System, Retrieved from http://www.dfs.dps.mo.gov/documents/nfirs-reference-guide.pdf
  2. Justin, J. P., Spatial and Temporal Patterns of Forest Fire Activity in Canada, M.S. Thesis, University of Toronto, (2001)
  3. Joo, K. D., A Study on a Real-time Detection and Monitoring System for Abnormal Fire Occurrences Based on Big Data Mining, M.S. Thesis, Myongji University, (2012)
  4. Apley, D. W. and Lee, H. C., "Robustness Comparison of Exponentially Weighted Moving-Average Charts on Autocorrelated Data and on Residuals," Journal of Quality Technology, 40(4), 428-447, (2008) https://doi.org/10.1080/00224065.2008.11917747
  5. Song, D. W., Predicting the Risk of Fire Occurrence according to the Weather Information using Statistics and Data Mining Techniques, Ph.D Thesis, Seoul National University of Science and Technology, (2014)
  6. Himmelblau, D. M., Fault Detection and Diagnosis in Chemical and Petrochemical Processes, Elsevier, pp.127-116, (1978)
  7. Box, George E. P., Time Series Analysis, Wiley, (2008)
  8. Hyndman, R. J., Forecasting Functions for Time Series and Linear Models, Retrieved from http://cran.r-project.org/web/packages/forecast/index.html, (2015)
  9. Hyndman, R. J. and Khandakar, Y., "Automatic Time Series Forecasting: The forecast package for R", Journal of Statistical Software, 27(3), 1-22, (2008)
  10. Witten, I. A. Data mining, Elsevier, pp.148, (2011)
  11. Lim, W. C., "Reliability-Based Design Optimization Using Akaike Information Criterion for Discrete Information", Transactions of the Korean Society of Mechanical Engineers A, 36(8), 921-927, (2012) https://doi.org/10.3795/KSME-A.2012.36.8.921

Cited by

  1. 미래 기상정보를 사용하지 않는 LSTM 기반의 피크시간 태양광 발전량 예측 기법 vol.24, pp.4, 2019, https://doi.org/10.7838/jsebs.2019.24.4.119