Development and Evaluation of a Software System for Fire Risk Prediction

Fire risks is a broad research topic that is important to many different stakeholders, from fire departments and municipalities, to homeowners. Reducing risks of fires includes not only infrastructure to extinguish flames, such as fire brigades, but also warning systems to alert of fires as early as possible, like smoke alarms. Risks can be further mitigated, by having a system to detect locations with high risk, both at present time but also predicted days ahead. It is this system and its concepts that is the core of this thesis. Stakeholders can be alerted of the high risk areas and take steps accordingly; Fire brigades can allocate more resources at a given place and time, and the general public can reduce or avoid risky behaviour, such as burning garden waste. The DYNAMIC research project has developed a model for estimating relative indoor humidity based on outdoor weather conditions. The indoor humidity can in turn be used to determine the time to flash-over after an ignition has occurred. This thesis centers around implementing and extending the usability of this model. A system of micro-services, hosted in the cloud, makes the fire risks available for stakeholders. They can add locations of interest to the system, see current and predicted risks in a heat-map imposed on a geographical map, and subscribe to be notified if specific locations have high risks. Historic weather data from the last 5 days are used to model the current risk for a location. By extending the data with weather forecasts, the predicted fire risks are also modelled. Experiments have been conducted to demonstrate that the predictions yields accurate fire risks. During February 2021, fire risks for 74 locations, spread across Norway, have been continuously collected in order to validate the system.