Wind Velocity Neural Estimator for Small Autonomous Surface Vehicles

Abstract

Surface aquatic vehicles present a complex dynamic behavior since they operate between two different fluids, air and water, each one with a different density and viscosity. Wind and surface currents affect vehicle motion in different manners. This paper shows the development of an estimator based on an artificial neural network which is used for wind velocity estimation in autonomous surface vehicles, intended to operate in protected waters, i.e. with no water surface currents. An aquatic vehicle motion model considering surge, sway and yaw motions in the boat dynamics, as well as wind speed and direction effects on its position, is presented here. Considering the wind disturbances, an artificial neural network is used to identify wind absolute speed and direction, so the actual vehicle should not need to carry wind speed and direction sensors aboard, only a GPS and an electronic compass, as planned. A Perceptron neural network was used in order to quantify wind speed and direction based only on boat displacements and its internal operational parameters as motor thrust and rudder angle. Those conditions were compared to a reference model in order to achieve a proper network training. The network showed good results for steady-state boat operation and also satisfactory results for transient responses.