基于物理信息神经网络的电液执行器建模方法OA

Modeling electro-hydraulic actuators is challenging due to their strong nonlinearities and unobservable internal states.To address issues such as low modeling accuracy and poor generalization,we propose an improved physics-informed neural network modeling method.First,a one-dimensional convolutional neural network module is employed to extract temporal features from sensor data.Subsequently,the force balance equation derived from electro-hydraulic actuator dynamics is embedded into the loss function as a physical constraint.This mechanism compensates for the poor interpretability of pure data-driven models and accelerates convergence.Furthermore,to mitigate the interference of sensor noise on physical constraint calculations,a signal smoothing strategy based on local linear fitting is designed.The multi-condition experiments demonstrate that this method effectively balances data fitting with physical consistency.Compared with traditional models,the proposed approach significantly improves prediction accuracy and robustness under limited data conditions.