A PINN Based Non-Recursive Control Approach for a Class of Nonlinear Systems with Mismatched Disturbances

For nonlinear system with the presence of mismatched disturbances, disturbance observation as an active intervention technology is commonly utilized, which encounters a multitude of challenges in practical scenarios. Disturbance observers often involve significant computational intensity, posing an unavoidable burden for systems with limited resources, while their performance is dependent on the accuracy and reliability of sensors. In order to address these problems, by integrating physical laws with deep learning algorithms, a novel disturbance estimation strategy by employing a physics-informed neural network-based observer (PINNO) is proposed. The accuracy and efficiency of PINNO can be enhanced via embedded physical knowledge. It reduces the dependence on large amounts of training data, thereby decreasing the model's sensitivity to data and improving its generalization capabilities under various operating conditions. The efficacy of the proposed PINNO is validated by a simple simulation.