Optimized least-squares support vector machine for predicting aero-optic imaging deviation based on chaotic particle swarm optimization

Abstract

In the time-averaged flow field, aero-optic imaging deviation (AOID) and its influencing variables are connected by complicated and strongly coupled nonlinear time-dependent relationships that present many difficulties for fast calculation of AOID. To achieve real-time online AOID compensation in practice, chaotic particle swarm optimization (CPSO) and the least-squares support vector machine (LSSVM) are combined to construct a predictive AOID model for supersonic aircraft in flight. With its strong global-search capability, the CPSO algorithm is used to optimize the parameters of the LSSVM predictive model. First, a chaotic sequence is used to initialize the particle positions, thereby enhancing search diversity. The premature convergence of normal particle swarm optimization (PSO) is then countered by using CPSO. If PSO falls into a local optimum, the extreme position of the population is adjusted and the current search trajectory of the particles is disturbed so that the particles search new neighborhoods and paths, thereby increasing the probability of escaping the local optimum. The simulation results show that the predictive accuracy of the CPSO-LSSVM model exceeds that of the LSSVM model alone or the back-propagation neural network model. The CPSO-LSSVM model is effective at compensating for AOID within a limited range, and it avoids the disadvantages of traditional geometrical-optics calculations, namely that they are time-consuming, laborious, and error-prone. When the training data of LSSVM is enough, any AOID can be estimated by the CPSO-LSSVM model.