A Neural Network based Path Planning Algorithm for Extinguishing Forest Fires

In this work an algorithm for automatic detection and suppression of Forest fires is proposed. The algorithm is implemented using parallel distributed model of neural network with three activation functions to determine the next consecutive moves to the cells for the actor. The algorithm uses reinforcement learning with weights determined dynamically in each iteration. The Entire forest is decomposed into grid of square cells with initial position of the Actor is assumed to be the cell 1 and the goal cell is the cell where the fire has occurred. The neural network model uses starting cell, goal cell and number of cells in each row or column and three activation functions to determine the next consecutive cells in which the robot has to travel. It uses only three movements LEFT, DIAGONAL and UP to reach the target cell. After calculating next cell, the check will be made for presence of obstacles in that cell. If there is any obstacle in that cell, then one cell from other two cells obtained using other two movements, which is free from obstacle will be chosen for next move. Then the cell number is stored in memory. This process is repeated till the next cell computed is same as the goal cell. The Actor will begin to move from start cell and reach the goal cell using the cell numbers available in the memory to extinguish Forest fire. This algorithm is designed keeping in mind only static obstacles and hence it works well for Forest environment with static obstacles. Computer simulation results show that path has been found successfully without collision with obstacles.