Cylinder action system modeling method based on physical information neural network

By combining the time-series data and physical parameters of the cylinder action system with the Physical Information Neural Network (PINN), a composite loss function is constructed to optimize the model parameters, which solves the accuracy and adaptability problems of the cylinder system under multiple working conditions and achieves high-precision and strong generalization modeling effect.

CN122131585APending Publication Date: 2026-06-02DALIAN MARITIME UNIVERSITY
0 Cites 0 Cited by

Patent Information

Application Number
CN202610089395.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-01-22
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing cylinder system models cannot balance accuracy and model adaptability under multiple operating conditions. Pure data-driven methods have insufficient generalization ability, and traditional physical models are unable to cover complex nonlinear factors, resulting in limited control system performance.

Method used

By employing a Physical Information Neural Network (PINN) combined with the time-series data of the cylinder action system, a dynamic feature input layer and a physical parameter input layer are constructed. The system state variables are predicted through multi-layer nonlinear transformation, and the model parameters are optimized using a composite loss function to achieve high-precision modeling and strong generalization.

Benefits of technology

It significantly improves the prediction accuracy and generalization performance of the model, and can complete efficient modeling with only a single motion data. It is suitable for digital cylinder action loops in high-speed on/off valve control and enhances the prediction reliability in data-sparse regions.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a cylinder action system modeling method based on a physical information neural network, and the method comprises the steps: obtaining time sequence data of a cylinder action system, including a time sequence control quantity and a system state quantity; constructing a physical information neural network model provided with a dynamic feature input layer and a physical parameter input layer, and respectively receiving a time sequence control quantity and a physical parameter vector; performing nonlinear transformation on the physical parameter vector through a physical parameter coding network to obtain a coded physical feature vector; fusing the time sequence control quantity and the physical feature vector to form a joint input vector; inputting the joint input vector into a trunk neural network, and synchronously predicting a system state quantity through multilayer nonlinear transformation; constructing a composite loss function comprising a data loss item and a physical loss item; and training the model by using time sequence data, updating model parameters by minimizing a composite loss function until the model converges, and completing high-precision modeling by using only single motion data.
Need to check novelty before this filing date? Find Prior Art