Abstract: Aiming at the problem that the product concentration in ultrasonic natural product extraction is difficult to detect online, a soft measurement method using the least squares support vector machine (LSSVM), which is optimized by an improved fruit fly optimization algorithm (FOA), is proposed. Firstly, the improved FOA with chaotic dynamic step size (named as CDS-FOA) is obtained by combing chaos optimization and iterative step-size dynamic adjustment. This algorithm introduces the dynamic adjustment factor to update the step size dynamically, and uses chaos optimization to realize the mapping between different variables, which can effectively improve the convergence precision and convergence speed of the FOA. Then, the CDS-FOA is used to optimize the parameters of LSSVM to construct the optimal CDSFOA-LSSVM soft measurement model. Finally, the experimental data from ultrasonic quercetin extraction are used to verify the effectiveness. Results show that the proposed model not only has better learning and generalization ability, but also has good prediction accuracy, which can provide guidance for the optimization of ultrasonic natural product extraction processes.