ISO-optimized quantile regression LSSVM-based adaptive robust MPC for forward speed control of pepper harvesters

To address the issue of significant fluctuations in drum load caused by imprecise forward speed control in spring-tooth drum pepper harvesters, this paper proposes an adaptive robust model predictive control (MPC) framework for forward speed control based on least squares support vector machine integrated with improved snake optimization and quantile regression theory (ISO-QR-LSSVM). The ISO-QR-LSSVM is employed as the dynamic model of MPC to establish the nonlinear mapping relationship between kinematic and dynamic parameters of the harvester for accurate torque prediction and forward speed control. By leveraging the capabilities of QR in uncertainty quantification and risk awareness, we design two adaptive mechanisms, including risk adaptation based on prediction interval width and online historical error compensation, as well as a robust mechanism based on quantile probability constraints. Simulation results confirm the proposed method's superiority in torque tracking accuracy, constraint satisfaction, and speed control performance. Additionally, a pepper harvesting test platform is established for further harvesting quality validation. The results show that the proposed method reduces the loss rate by 69.1%, impurity rate by 64.6%, and damage rate by 67.1% compared to the no-control scheme with constant optimal operating parameters. This effectively improves the harvesting quality and provides an effective approach for intelligent adaptive robust control of agricultural equipment under complex harvesting environments.