Search

In recent years, the development of fire detectors has attracted the attention of researchers for the purpose of protecting human lives and properties from catastrophic fire disasters. However, monitoring fires is challenging due to several unique characteristics of fire sensor signals, such as the existence of temporal dependency and diverse signal patterns for different fire types, including flaming, heating, and smoldering fires. In this study, we propose a new approach for fire detection based on machine learning and optimization techniques, to monitor various types of fire by utilizing information obtained via multichannel fire sensor signals. The contribution of this study is to improve an existing fire detector by developing a new fire monitoring framework to identify fire based on support vector machine with dynamic time warping kernel function (SVM-DTWK), which considers the temporal dynamics existing in the sensor signals of different fire types. In addition, multichannel sensor signals are further considered by the SVM-DTWK with a multi-modeling framework that constructs multiple classifiers for each sensor type and effectively utilizes sensor information that is critical for the detection of fires without prior knowledge of the fire type. Using real-life fire data, the proposed approach is compared with existing fire monitoring methods and achieves superior performance in terms of both fire detection time and false alarm rate.