UDC 614.842.435
The article addresses the problem of false alarm prediction in smoke fire detectors using a Long Short-Term Memory (LSTM) recurrent neural network. The objective of the study is to develop and test a model capable of distinguishing between true and false alarms based on temporal data, thereby reducing the number of false emergency responses and increasing the reliability of fire safety systems. As part of the research, an experiment was conducted, during which data were collected from a laser sensor and two fire detectors (H1 and H2). The developed LSTM model was trained on this dataset using 50-step sequences, and its accuracy was evaluated using the MAE, MSE, RMSE, and Accuracy metrics. The results demonstrated that the proposed model achieves 99,5% accuracy for H1 and 99,3% for H2, supported by low error values: MAE < 0,01, RMSE ≈ 0,05. Compared to existing methods (ensemble models, time series analysis), the proposed approach relies exclusively on the temporal dynamics of signals without requiring multisensor data, making it efficient and easily integrable into existing fire detection systems. The obtained results confirm the potential of LSTM for false alarm detection and indicate possibilities for further model improvements by incorporating additional factors (temperature, humidity, rate of smoke concentration change) and utilizing hybrid architectures (CNN + LSTM)
false alarm prediction, smoke fire detectors, lstm, neural networks, time series, deep learning, automated fire safety systems, prediction accuracy, machine learning, anomaly detection
1. Pozhary i pozharnaya bezopasnost' v 2023 godu: informacionno - analiticheskiy sbornik / avtorskiy kollektiv: Goncharenko V.S., Chechetina T.A., Sibirko V.I., Nadtochiy O.V., Polehin P.V., Kozlov A.A., Gribanov A.M. rec.: Sokolov S.V. - Balashiha: FGBU VNIIPO MChS Rossii, 2024. - 110 s.
2. Tanklevskiy L.T., Babikov I.A., Tanklevskiy A.L. Ob ocenke dostovernosti obnaruzheniya vozgoraniya // Problemy upravleniya riskami v tehnosfere. - 2021. - No 3 (59). - S. 28-33.-EDN. EDN: https://elibrary.ru/JVMNBY
3. Pireeva V.S., Rahmatullina E.F., Permyakov A.V. Vliyanie koncentracii dymovyh chastic na obnaruzhenie vozgoraniya // Tehnicheskie i tehnologicheskie sistemy: materialy XV Mezhdunarodnoy nauchnoy konferencii "TTS-24" (20-22 noyabrya 2024 goda) / FGBOU VO "KubGTU", FGK VOU VO "KVVAUL im. A.K. Serova". Krasnodar: Izdatel'skiy Dom - Yug, 2024. - S.150-153. EDN: https://elibrary.ru/WOUFQO
4. Kim Y., Heo Y., Jin B., Bae Y. Real-Time Fire Classification Models Based on Deep Learning for Building an Intelligent Multi-Sensor System. Fire - 2024, - No 7, p. 329. DOI:https://doi.org/10.3390/fire7090329 EDN: https://elibrary.ru/KEBFKC
5. Liu, Ming & Zhou, Hongbin & Ren, Yupeng & Lu, Wei. (2020). A Deep Learning Approach to Reduce False Alarms for Optical Smoke Detectors. Journal of Physics: Conference Series. 1631. 012032. DOI:https://doi.org/10.1088/1742-6596/1631/1/012032
6. Xu, Fang et al. Embedded Spatial-Temporal Convolutional Neural Network Based on Scattered Light Signals for Fire and Interferential Aerosol Classification. Sensors (Basel, Switzerland) vol. 24,3 0. 25 Jan. 2024,. DOI:https://doi.org/10.3390/s24030778
7. Al-Rayes, M. R. M., Abu-Naser, S. S. Smoke Detectors Using ANN [Elektronnyy resurs] // International Journal of Academic Engineering Research (IJAER). - 2023. - T. 7, No 10. - S. 1-9. - Rezhim dostupa: www.ijeais.org/ijaer, svobodnyy.
8. Photoelectric smoke sensing detection method for accurately identifying fire smoke: pat.CN113362560A Kitay / Zhang Hongying, Feng Yong, Zhang Guangbiao, Zhang Peng, Wang Jianjun, Lin Yuxin, Zhang Xiaoshuang, Qu Wu, He Guotang, Li Yong, zayavitel' Bengbu Ei Fire Electronics Co Ltd. - NoCN202110593735.XA, zayavl. 28.05.2021, opubl. 07.09.2021. -URL:https://patents.google.com/patent/CN113362560A.
9. National Research Council of Science and Technology.AI sensor cuts false alarms in smoke detectors [Elektronnyy resurs]//TechXplore. - 2025. - 4 fevralya. -URL:https://techXplore.com/news/2025/02-ai-sensor-false-alarms-detectors.html.
10. Rahmatullina E.F., Permyakov A.V., Hafizov I.F., Hafizov F.Sh. Prognozirovanie znacheniy pozharnogo riska s pomosch'yu mashinnogo obucheniya i analiza vremennyh ryadov // Neftegazovoe delo. - 2024. - No 5. - S. 6-20.
11. Rahmatullina E.F., Permyakov A.V. Issledovanie lozhnyh srabatyvaniy avtomaticheskih ustanovok pozharnoy signalizacii na ob'ektah s massovym prebyvaniem lyudey // Materialy 75-y nauchno - tehnicheskoy konferencii studentov, aspirantov i molodyh uchenyh UGNTU. - Ufa: Izdatel'stvo UGNTU, 2024. - S. 82. EDN: https://elibrary.ru/HNOUVO
