Abstract:The Fokker-Plank-Kolmogorov (FPK) equation is an idealized model representing many stochastic systems commonly encountered in the analysis of stochastic structures as well as many other applications. Its solution thus provides an invaluable insight into the performance of many engineering systems. Despite its great importance, the solution of the FPK equation is still extremely challenging. For systems of practical significance, the FPK equation is usually high dimensional, rendering most of the numerical methods ineffective. In this respect, the present work introduces the FPK-DP Net as a physics-informed network that encodes the physical insights, i.e. the governing constrained differential equations emanated out of physical laws, into a deep neural network. FPK-DP Net is a mesh-free learning method that can solve the density evolution of stochastic dynamics subjected to additive white Gaussian noise without any prior simulation data and can be used as an efficient surrogate model afterward. FPK-DP Net uses the dimension-reduced FPK equation. Therefore, it can be used to address high-dimensional practical problems as well. To demonstrate the potential applicability of the proposed framework, and to study its accuracy and efficacy, numerical implementations on five different benchmark problems are investigated.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to address the problem of solving the Fokker-Planck-Kolmogorov (FPK) equation in high-dimensional stochastic dynamical systems. Specifically, the paper proposes a deep learning-based method (referred to as FPK-DP Net) to overcome the challenges faced by traditional numerical methods in handling high-dimensional FPK equations. #### Main Objectives: 1. **Solving the FPK Equation**: The paper proposes a Physics-Informed Neural Network (PINN) framework based on deep learning, called FPK-DP Net, for solving the FPK equation in high-dimensional stochastic dynamical systems. 2. **Handling High-Dimensional Problems**: Traditional numerical methods face severe dimensionality issues when dealing with high-dimensional FPK equations. By introducing a reduced-dimensional form of the FPK equation, this method effectively addresses high-dimensional problems. 3. **Unsupervised Learning**: This method does not rely on any prior simulation data but directly extracts constraints from physical laws and encodes them into the deep neural network. 4. **Efficient Surrogate Model**: FPK-DP Net can serve as an efficient surrogate model for subsequent rapid simulations and analyses. ### Research Background and Motivation: - **Importance**: The FPK equation is of significant importance in many engineering and scientific problems, especially in the dynamic analysis of stochastic structures. - **Existing Challenges**: For high-dimensional systems in practical applications, traditional numerical methods often fail to effectively solve the FPK equation. - **Advantages of Deep Learning**: In recent years, with the development of deep learning technology, machine learning methods have been widely applied to solve complex physical problems, particularly the application of Physics-Informed Neural Networks (PINNs). ### Method Overview: 1. **Reduced-Dimensional FPK Equation**: By decomposing stochastic excitations and using the Probability Density Evolution Method (PDEM), the high-dimensional FPK equation is transformed into a series of low-dimensional equations. 2. **Deep Learning Network**: A deep neural network (DNN) is used to approximate the probability density function and ensure it meets physical constraints. 3. **Loss Function Design**: A loss function based on the residual form of the FPK equation is defined, and the derivatives of the network output are computed through Automatic Differentiation (AD). ### Experimental Validation: - **Benchmark Problems**: The effectiveness and accuracy of the proposed method are validated through five different benchmark problems. - **Performance Evaluation**: The superior performance of FPK-DP Net in network training and prediction is demonstrated. ### Conclusion: This paper proposes a novel deep learning method that can effectively solve the FPK equation for high-dimensional stochastic dynamical systems, providing a new solution for practical engineering problems.

Solution of FPK Equation for Stochastic Dynamics Subjected to Additive Gaussian Noise via Deep Learning Approach

Stochastic averaging of n-dimensional non-linear dynamical systems subject to non-Gaussian wide-band random excitations

Physics-Informed Solution of The Stationary Fokker-Plank Equation for a Class of Nonlinear Dynamical Systems: An Evaluation Study

Solving Fokker-Planck Equation Using Deep Learning

A Novel Method for Response Probability Density of Nonlinear Stochastic Dynamic Systems

Physically Guided Deep Learning Solver for Time-Dependent Fokker–Planck Equation

Statistical Solution to SDEs with $$\alpha $$-Stable Lévy Noise Via Deep Neural Network

Solving Inverse Stochastic Problems from Discrete Particle Observations Using the Fokker-Planck Equation and Physics-informed Neural Networks

Learning Dynamics from Noisy Measurements using Deep Learning with a Runge-Kutta Constraint

Artificial Neural Network Solver for Time-Dependent Fokker-Planck Equations

Deep Hidden Physics Models: Deep Learning of Nonlinear Partial Differential Equations

Neural Network-Based Analytical Solver for Fokker–Planck Equation

Evolutionary probability density reconstruction of stochastic dynamic responses based on physics-aided deep learning

Deep learning based numerical approximation algorithms for stochastic partial differential equations and high-dimensional nonlinear filtering problems

A deep learning solution approach for high-dimensional random differential equations

Solving Fokker–Planck Equations Using Deep KD-tree with a Small Amount of Data

Deep Latent Regularity Network for Modeling Stochastic Partial Differential Equations

Filtered Partial Differential Equations: a robust surrogate constraint in physics-informed deep learning framework

Learning effective stochastic differential equations from microscopic simulations: Linking stochastic numerics to deep learning.

Enriched Physics-informed Neural Networks for Dynamic Poisson-Nernst-Planck Systems

Deep learning numerical methods for high-dimensional fully nonlinear PIDEs and coupled FBSDEs with jumps