Building Fire Location Predictions Based on FDS and Hybrid Modelling
With the goal of addressing the difficulty of rapidly identifying the source of fire in commercial buildings, this study builds a numerical fire model based on the fire dynamics simulator (FDS) and combines it with a hybrid model to predict the location of a fire source. Different scenarios were bui...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-06-01
|
| Series: | Buildings |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-5309/15/12/2001 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850156117799206912 |
|---|---|
| author | Yanxi Cao Hongyan Ma Shun Wang Yingda Zhang |
| author_facet | Yanxi Cao Hongyan Ma Shun Wang Yingda Zhang |
| author_sort | Yanxi Cao |
| collection | DOAJ |
| description | With the goal of addressing the difficulty of rapidly identifying the source of fire in commercial buildings, this study builds a numerical fire model based on the fire dynamics simulator (FDS) and combines it with a hybrid model to predict the location of a fire source. Different scenarios were built to simulate the spatial and temporal distributions of key parameters such as temperature, smoke, and CO concentration during the fire process. Combining convolutional neural networks (CNNs) and support vector machines (SVMs) for prediction, the fire-source location prediction model with temperature, smoke, and CO concentration as feature quantities was constructed, and the hyperparameters affecting the model accuracy and generalisation were optimised by the Crested Porcupine Optimizer (CPO) algorithm. The experimental results show that the positioning error of this method under the building plane is less than 0.95 m, the mean absolute error (<i>MAE</i>) is within 0.35, and the root-mean-square error (<i>RMSE</i>) is within 0.41, which are 43% and 82% higher than the unoptimised model, respectively. The localisation accuracy of the fire-source room is 97.61%. In addition, the model’s anti-interference performance was tested under various extreme conditions. The results show that the proposed model can ensure the accurate location of a fire source and can provide information in emergencies. |
| format | Article |
| id | doaj-art-5aabde29a53e4442aaf64e61ae5c8e2f |
| institution | OA Journals |
| issn | 2075-5309 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Buildings |
| spelling | doaj-art-5aabde29a53e4442aaf64e61ae5c8e2f2025-08-20T02:24:42ZengMDPI AGBuildings2075-53092025-06-011512200110.3390/buildings15122001Building Fire Location Predictions Based on FDS and Hybrid ModellingYanxi Cao0Hongyan Ma1Shun Wang2Yingda Zhang3School of Intelligence Science and Technology, Beijing University of Civil Engineering and Architecture, Beijing 100044, ChinaSchool of Intelligence Science and Technology, Beijing University of Civil Engineering and Architecture, Beijing 100044, ChinaSchool of Intelligence Science and Technology, Beijing University of Civil Engineering and Architecture, Beijing 100044, ChinaSchool of Intelligence Science and Technology, Beijing University of Civil Engineering and Architecture, Beijing 100044, ChinaWith the goal of addressing the difficulty of rapidly identifying the source of fire in commercial buildings, this study builds a numerical fire model based on the fire dynamics simulator (FDS) and combines it with a hybrid model to predict the location of a fire source. Different scenarios were built to simulate the spatial and temporal distributions of key parameters such as temperature, smoke, and CO concentration during the fire process. Combining convolutional neural networks (CNNs) and support vector machines (SVMs) for prediction, the fire-source location prediction model with temperature, smoke, and CO concentration as feature quantities was constructed, and the hyperparameters affecting the model accuracy and generalisation were optimised by the Crested Porcupine Optimizer (CPO) algorithm. The experimental results show that the positioning error of this method under the building plane is less than 0.95 m, the mean absolute error (<i>MAE</i>) is within 0.35, and the root-mean-square error (<i>RMSE</i>) is within 0.41, which are 43% and 82% higher than the unoptimised model, respectively. The localisation accuracy of the fire-source room is 97.61%. In addition, the model’s anti-interference performance was tested under various extreme conditions. The results show that the proposed model can ensure the accurate location of a fire source and can provide information in emergencies.https://www.mdpi.com/2075-5309/15/12/2001support vector machineconvolutional neural networkcrested porcupine optimizerfire dynamics simulatorfire-source location estimation |
| spellingShingle | Yanxi Cao Hongyan Ma Shun Wang Yingda Zhang Building Fire Location Predictions Based on FDS and Hybrid Modelling Buildings support vector machine convolutional neural network crested porcupine optimizer fire dynamics simulator fire-source location estimation |
| title | Building Fire Location Predictions Based on FDS and Hybrid Modelling |
| title_full | Building Fire Location Predictions Based on FDS and Hybrid Modelling |
| title_fullStr | Building Fire Location Predictions Based on FDS and Hybrid Modelling |
| title_full_unstemmed | Building Fire Location Predictions Based on FDS and Hybrid Modelling |
| title_short | Building Fire Location Predictions Based on FDS and Hybrid Modelling |
| title_sort | building fire location predictions based on fds and hybrid modelling |
| topic | support vector machine convolutional neural network crested porcupine optimizer fire dynamics simulator fire-source location estimation |
| url | https://www.mdpi.com/2075-5309/15/12/2001 |
| work_keys_str_mv | AT yanxicao buildingfirelocationpredictionsbasedonfdsandhybridmodelling AT hongyanma buildingfirelocationpredictionsbasedonfdsandhybridmodelling AT shunwang buildingfirelocationpredictionsbasedonfdsandhybridmodelling AT yingdazhang buildingfirelocationpredictionsbasedonfdsandhybridmodelling |