Unsteady aerodynamic modeling of biaxial coupled oscillation based on improved ELM

Abstract This paper presents an innovative unsteady aerodynamic modeling method based on the improved Extreme Learning Machine (ELM), which is further successfully applied into the biaxial coupled oscillation during the post stall maneuver flight. The underlying data of pitch-roll and yaw-roll coupled oscillation for the aircraft, which simulate the post stall maneuver such as Herbst maneuver and barrel roll, are obtained by the wind tunnel test equipment and then separated as training samples and testing samples. To reduce the prediction error, the cross validation method is employed. And the parameters including the connect weights between the input layer and the hidden layer, the bias and the number of the neurons in the hidden layer are selected based on the testing samples. According to the verifying samples, the comparison is accomplished among the back-propagation neural network (BPNN) method, the enhanced incremental ELM (EI-ELM) method and the improved ELM method. The simulation results demonstrate that the improved ELM method has characteristics of high precision, strong versatility and fast prediction in the unsteady aerodynamic modeling.