Abstract:We propose CoNSAL (Combining Neural networks and Symbolic regression for Analytical Lyapunov function) to construct analytical Lyapunov functions for nonlinear dynamic systems. This framework contains a neural Lyapunov function and a symbolic regression component, where symbolic regression is applied to distill the neural network to precise analytical forms. Our approach utilizes symbolic regression not only as a tool for translation but also as a means to uncover counterexamples. This procedure terminates when no counterexamples are found in the analytical formulation. Compared with previous results, CoNSAL directly produces an analytical form of the Lyapunov function with improved interpretability in both the learning process and the final results. We apply CoNSAL to 2-D inverted pendulum, path following, Van Der Pol Oscillator, 3-D trig dynamics, 4-D rotating wheel pendulum, 6-D 3-bus power system, and demonstrate that our algorithm successfully finds their valid Lyapunov functions. Code examples are available at <a class="link-external link-https" href="https://github.com/HaohanZou/CoNSAL" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper aims to address the problem of constructing analytical Lyapunov functions in nonlinear dynamic systems. Specifically, it proposes a framework named CoNSAL (Combining Neural Networks and Symbolic regression for Analytical Lyapunov function), which combines neural networks and symbolic regression techniques to directly generate analytical forms of Lyapunov functions. This approach not only enhances the interpretability of Lyapunov functions but also improves the verification process by discovering counterexamples through symbolic regression. ### Main Challenges 1. **Generalization Ability**: Existing learning-based methods struggle to generalize Lyapunov functions trained on specific dynamic systems and parameters to larger state spaces or slightly different dynamic systems. 2. **Verifiability**: Current formal verification methods (such as SMT solvers, mixed-integer linear programming, etc.) have high computational costs in complex systems, limiting their applicability. ### Solutions 1. **Combining Neural Networks and Symbolic Regression**: The CoNSAL framework includes a neural network learner and a symbolic regression component. The neural network is used to learn approximate forms of Lyapunov functions, while symbolic regression is used to convert these approximate forms into precise analytical forms. 2. **Efficient Counterexample Generation**: By sampling near the roots of the symbolic Lyapunov function to generate counterexamples, the framework avoids the expensive neural network verification process. 3. **Composite Neural Lyapunov Function Design**: For high-dimensional systems, composite neural Lyapunov function design is employed to improve the scalability of the algorithm. ### Experimental Validation The paper validates the effectiveness of CoNSAL through experiments on multiple nonlinear dynamic systems, including a 2D inverted pendulum, path-following problem, Van der Pol oscillator, 3D triangular dynamics, 4D rotating wheel pendulum, and a 6D three-bus power system. The experimental results demonstrate that CoNSAL successfully finds effective analytical Lyapunov functions for these systems and outperforms SMT-based methods in terms of performance and efficiency. ### Contributions 1. **First to Propose Combining Neural Networks and Symbolic Regression**: The method directly generates analytical forms of Lyapunov functions, significantly enhancing interpretability. 2. **Proposed an Efficient Counterexample Generation Paradigm**: By sampling near the roots of the symbolic Lyapunov function, the framework avoids the expensive neural network verification process. 3. **Validated the Algorithm's Effectiveness in Various Nonlinear Systems**: Demonstrated robustness and efficiency in different scenarios.

Combining Neural Networks and Symbolic Regression for Analytical Lyapunov Function Discovery

CoNSoLe: Convex Neural Symbolic Learning

Lyapunov Neural Network with Region of Attraction Search

Symbolic regression via neural networks

Learning Lyapunov Functions for Piecewise Affine Systems with Neural Network Controllers

Lyapunov-stable neural-network control

Neural Lyapunov Control of Unknown Nonlinear Systems with Stability Guarantees

Neural Lyapunov Control for Discrete-Time Systems

Deep Learning and Symbolic Regression for Discovering Parametric Equations

A Neural-Guided Dynamic Symbolic Network for Exploring Mathematical Expressions from Data

Lyapunov-stable Neural Control for State and Output Feedback: A Novel Formulation

Compositionally Verifiable Vector Neural Lyapunov Functions for Stability Analysis of Interconnected Nonlinear Systems

Koopman analysis of nonlinear systems with a neural network representation

Neural Nonparametric Stability Indicator for Self-Excited Dynamical Systems

A Neuro-Symbolic Method for Solving Differential and Functional Equations

Neural Symbolic Regression of Complex Network Dynamics

Neural Network-Based Lyapunov Stability Verification for Continuous-Time Nonlinear Systems Via Mixed Integer Linear Programming

A time-dependent symplectic network for non-convex path planning problems with linear and nonlinear dynamics

Neural-network-based observers for nonlinear systems

Observer-based Adaptive Finite-Time Neural Control for Constrained Nonlinear Systems with Actuator Saturation Compensation

Deep symbolic regression for physics guided by units constraints: toward the automated discovery of physical laws