No-fire modelling for no-fire detector

Abstract

In order to improve fire sensing functionalities through the integration of Wireless Sensor Network (NSW) and Machine Learning (ML) approaches, this project investigates the critical domain of No-Fire (NF) modelling for NF detectors. The Unified Fire Sensing idea and Siebel's NF detector concept highlight the urgent need for NF detectors. Although WSN with ML has become more and more common for fire detection systems, generic prediction tasks are the main emphasis of the current ML workflow. However, in order to meet the fire sensing tasks, the modelling accuracy of ML will be monitored. The first step in the research technique is to use an Arduino Nano BLE Sense (rev1) datalogger to gather environmental data from monitored places. Next, machine learning analysis is performed to determine the best-fitting NF models. 6 ML techniques are included in the research: Polynomial regression, Ridge regression, Lasso regression, K Nearest Neighbour regression, Elastic-net regression, and Ordinary Least Squares. Furthermore, 3 different data stacking techniques, such as Continuous, Daily, and Weekly, are used to evaluate the effect on modelling performance and accuracy. By developing ML techniques that can be utilised for modelling in NF detectors, the research findings are expected to make a substantial contribution to the field of fire detection.