URBAN FIRE SPREAD MODELLING AND SIMULATION USING CELLULAR AUTOMATON WITH EXTREME LEARNING MACHINE

Abstract

Urban fire continues to be a persistent disaster, especially with the proliferation of highly dense urban settlements. As a response, several measures were established to help mitigate the losses caused by fire including simulating the fire spread. The cellular automaton system has been widely used to simulate the complex process of fire development along with Physics-based models. A data-driven approach has been rarely employed. This paper presents the result of incorporating machine learning techniques to the existing cellular automaton based urban fire spread models. Specifically, instead of manually calculating the ignition probability of each cell in the automaton, the Extreme Learning Machine (ELM) was used to learn the ignition probability from the historical data. After building the model, its performance was evaluated using the data collected from the four fires in Basak, Lapu-Lapu City. By using a confusion matrix to compare the actual and the predicted values, the Burned Actual – Burned Predicted relationship was derived. Results suggest that the proposed method can effectively describe the development of fire, and the model accuracy is quite good (i.e., the Burned Actual - Burned Predicted relationship ranges from 78% to 83%). Lastly, the study was able to demonstrate the possibility of using a data-driven approach in creating a simple cellular automaton fire spread simulation model for urban areas. Further studies utilizing more fire incident data on with varying properties is recommended.