Analysis and prediction of fire water pressure in buildings based on IoT data. (November 2021)

• Record Type: Journal Article
• Authors: Hu, Jun; Wu, Jinjin; Shu, Xueming; Shen, Shifei; Ni, Xiaoyong; Yan, Jun; He, Sheng
• Is Part Of: Journal of building engineering. Volume 43(2021)
• Publication Date: 2021-11
• Subjects: Building fire safety engineering -- Indoor fire water supply system -- Water pressure prediction -- Variation of water pressure -- Long short-term memory (LSTM) method; Building -- Periodicals; 690.05
• DOI: 10.1016/j.jobe.2021.103197
• Languages: English
• ISSNs: 2352-7102

Abstract

Abstract: Indoor fire water supply system is an important guarantee for building fire safety, and the design parameters such as water flow and water pressure need to meet the requirements of fire prevention regulations. Based on the monitoring data of the Internet of Things (IoT), this paper took an office building as an example to analyze the water pressure variation trend of the building's indoor fire water supply system, and then predict it based on the historical water pressure drop rate fitting and Long short-term memory (LSTM) method. It is found that the variation of water pressure is periodic, and this trend was analyzed from the perspective of physical structure of the indoor fire water supply system. Furthermore, due to the regularity of water pressure variation, the prediction results are generally good. With historical water pressure drop rate fitting, the prediction accuracy is relatively higher, and the abnormal change points of water pressure can be discovered, but the water pressure value needs to be reset for further prediction; while the LSTM method is self-adaptive, when the frequency of water pressure monitoring is high and the amount of water pressure data is large, the LSTM is more suitable. Highlights: The water pressure variation of the office building's indoor fire water supply system is periodic. With historical water pressure drop rate fitting, the prediction result is higher. The LSTM prediction method is self-adaptive and more suitable for large amount of data.