Abstract:Hybrid dynamical systems are prevalent in science and engineering to express complex systems with continuous and discrete states. To learn the laws of systems, all previous methods for equation discovery in hybrid systems follow a two-stage paradigm, i.e. they first group time series into small cluster fragments and then discover equations in each fragment separately through methods in non-hybrid systems. Although effective, these methods do not fully take advantage of the commonalities in the shared dynamics of multiple fragments that are driven by the same equations. Besides, the two-stage paradigm breaks the interdependence between categorizing and representing dynamics that jointly form hybrid systems. In this paper, we reformulate the problem and propose an end-to-end learning framework, i.e. Amortized Equation Discovery (AMORE), to jointly categorize modes and discover equations characterizing the dynamics of each mode by all segments of the mode. Experiments on four hybrid and six non-hybrid systems show that our method outperforms previous methods on equation discovery, segmentation, and forecasting.

What problem does this paper attempt to address?

This paper attempts to solve the problem of discovering dynamical equations in Hybrid Dynamical Systems (HDS). Specifically, traditional methods adopt a two - stage paradigm when dealing with HDS: first, group time series into small cluster segments, and then apply equation discovery methods for non - hybrid systems in each segment respectively. This method has the following problems: 1. **Failure to fully utilize the commonalities between patterns**: Multiple segments may be driven by the same equation, but traditional methods do not fully utilize these commonalities. 2. **Breaking the interdependence between classification and representing dynamics**: The two - stage paradigm divides the learning process into two independent stages, that is, classification first and then representation, ignoring the association between them. To solve these problems, this paper redefines the problem and proposes an end - to - end learning framework - Amortized Equation Discovery (AMORE). AMORE aims to jointly classify patterns and discover the dynamical equations that describe each pattern. Through this method, AMORE can more effectively utilize the information in the data and improve the performance of equation discovery, segmentation and prediction. ### Specific improvement points - **Jointly learning pattern classification and equation discovery**: AMORE simultaneously performs pattern classification and equation discovery within a unified framework, thereby better capturing the relationships between different patterns. - **Fully utilizing the commonalities between patterns**: By learning all segments belonging to the same pattern, AMORE can more accurately discover the equation that describes this pattern. - **Modeling pattern - switching behavior**: AMORE introduces implicit class variables (mode variables) to model pattern - switching behavior, enabling the model to better adapt to complex time - series data. ### Experimental results The experimental results show that AMORE outperforms existing methods on four hybrid systems and six non - hybrid systems, especially in terms of equation discovery, segmentation and prediction. For example, on the Mass - spring Hopper data set, AMORE can accurately identify two patterns and discover dynamical equations that are almost identical to the real equations: \[ \begin{aligned} & \dot{l} = v \\ & \dot{v} = 11.03 - 10.08l \\ & \dot{l} = v \\ & \dot{v} = - 1 \end{aligned} \] In addition, AMORE also shows superior performance in multi - object hybrid systems such as the SIR disease data set, proving its wide applicability in complex scenarios. In conclusion, by proposing the AMORE framework, this paper solves the key problems of existing methods in hybrid power systems that cannot fully utilize the commonalities between patterns and break the interdependence between classification and representation, significantly improving the accuracy of equation discovery and prediction.

Amortized Equation Discovery in Hybrid Dynamical Systems

Efficient Neural Hybrid System Learning and Transition System Abstraction for Dynamical Systems

Hybrid modeling and prediction of dynamical systems

DySMHO: Data-Driven Discovery of Governing Equations for Dynamical Systems via Moving Horizon Optimization

Hybrid Persistency of Excitation in Adaptive Estimation for Hybrid Systems

A Data-Driven Hybrid Automaton Framework to Modeling Complex Dynamical Systems

Large language models for automatic equation discovery of nonlinear dynamics

Non-intrusive model combination for learning dynamical systems

Efficient Amortised Bayesian Inference for Hierarchical and Nonlinear Dynamical Systems

Finite Expression Methods for Discovering Physical Laws from Data

A Distributed Neural Hybrid System Learning Framework in Modeling Complex Dynamical Systems

Data-driven modeling of interrelated dynamical systems

Governing equation discovery based on causal graph for nonlinear dynamic systems

Unsupervised Learning of Hybrid Latent Dynamics: A Learn-to-Identify Framework

Neural Hybrid Automata: Learning Dynamics with Multiple Modes and Stochastic Transitions

Discovering governing equation in structural dynamics from acceleration-only measurements

LLM4ED: Large Language Models for Automatic Equation Discovery

Learning Governing Equations of Unobserved States in Dynamical Systems

Combining data assimilation and machine learning to build data‐driven models for unknown long time dynamics—Applications in cardiovascular modeling

Bayesian Spline Learning for Equation Discovery of Nonlinear Dynamics with Quantified Uncertainty

Automated Discovery of Continuous Dynamics from Videos