Abstract
To predict the variation in the displacement position of hydraulic cylinder piston rods, a neural network model is proposed to enhance the displacement control accuracy of hydraulic cylinders. The innovation of this paper is that by calculating the compressibility-induced flow loss of hydraulic fluid, mathematical models for both the internal and external leakage of hydraulic cylinders are established, identifying seven primary factors influencing piston rod displacement. Because there are many influencing factors and complex parameters different from traditional backpropagation (BP) neural network used in previous studies, this paper innovatively proposes a three-layer BP neural network ensemble model for predicting input-output characteristic curves of hydraulic cylinders. In the process of model improvement, a nonlinear adaptive decreasing weight mechanism is introduced to improve the optimization accuracy of the algorithm, facilitating the search for optimal solutions. The most reasonable weight and bias parameters are determined via the iterative training and testing of each BP neural network layer. This model enables the real-time prediction of piston rod displacement output curves after a specified time interval based on external input parameters. The predicted time is utilized to compensate for the response delays caused by directional valve switching and hydraulic fluid buffering, thereby enabling proactive displacement prediction. Validation results demonstrate that the maximum predicted displacement error is reduced to 0.5491 µm, with the model's maximum runtime being 27.82 ms. The maximum allowable time allocated for directional valve switching and fluid buffering in the hydraulic system is extended to 74.57 ms, achieving the objective of enhancing both displacement control accuracy and operational efficiency.