Fire detection in IoT/WSN environment is a major element of early disaster-related event detection in smart cities. Recently, several firefighting strategies are presented with reasonable detection accuracy and running time for normal IoT environment. The detection accuracy and running time can be improved by using machine learning techniques. In this project, to present a practical WSN-based firefighting model and to accelerate the learning process, specifically, a reinforcement learning method called Fuzzy Q-Learning (FQL) is adopted with the end goal of enhancing the learning experience of the robot sensors (as firefighters) under the perfect and partial learning policies. In general, fuzzy techniques allow the sensors to make decisions without the need for an exhaustive map of the environment. The FQL approximates the observable configuration space allowing the robot to overcome the high dimension challenge of feature decomposition and navigation in a stochastic environment. We also formulate a sensor mode selection strategy as an optimization problem with the objective of maximizing the lifetime of the energy harvesting-enabled WSN. Simulation results show that our proposed strategy brings substantial performance gain in terms of accuracy, convergence time and efficiency, which can verify its applicability in smart cities for early firefighting operations.
