Sensor Data Prediction for Fixed-wing Drone Based on Online Sequential Learning

Abstract

The special properties make the fixed-wing drones are widely used in the military area. Thus, the validity and the accuracy of the sensors and actuators are essential for fixed-wing drones. The sensors data prediction with a suitable learning algorithm is usually the basic step for data-driven approaches to increase the reliability of fixed-wing drones. Online learning algorithms could train the model from sensors data in real time, coinciding with the rich rough sensors data during the whole flight time. In addition, online learning algorithms based on Extreme Learning Machine, which is a light weight learning machine, show a great performance for time-series prediction. However, they have two drawbacks: overfitting models and less robust. Thus, we proposed a novel algorithm to promote the weights update strategy by updating the output weights based on statistical learning theory. And, the experiment of sensors data perdition shows the improved performance of the proposed algorithm experiment om compared with other online learning algorithms based on ELM.