Data-Driven Control of Nonlinear Dynamic Systems

Data-driven control of nonlinear dynamic systems has emerged as a transformative paradigm in control engineering, leveraging empirical data rather than detailed first-principles models. By embedding sequences of input–output measurements directly into controller synthesis, these methods bypass model identification bottlenecks and cope with uncertainty in complex environments. The field spans theoretical advances in behavioural systems theory and matrix-based parameterisations, the integration of machine learning for feature extraction, and optimisation techniques that deliver robustness guarantees. Applications range from autonomous aerial vehicles and robotic manipulators to power-grid stabilisation and neuromodulation in biological networks. This data-centric approach enables real-time adaptation where traditional modelling proves prohibitively difficult, offering new opportunities in renewable energy management, smart infrastructure and autonomous systems.

Research from Nature Portfolio

Recent studies have advanced data-driven frameworks for controlling complex networks. One seminal work introduced a data-enabled method that constructs optimal control inputs for networked systems from finite input–output datasets without requiring prior knowledge of the interconnection dynamics. The framework guarantees exact control in linear regimes and characterises performance degradation under nonlinear interactions, with case studies including power-grid frequency regulation and brain-network modulation. It also quantifies robustness to measurement noise through structured data matrices and convex optimisation, demonstrating scalability to high-dimensional networks and offering a blueprint for real-world implementation.

Research from all publishers

A novel semidefinite-programming approach has been proposed for approximate nonlinearity cancellation, leading to controllers that stabilise unknown nonlinear systems and certify robustly positively invariant sets when data are perturbed. In parallel, a unified stochastic predictive control framework employs tailored regularisation to mitigate noise effects and splits optimisation into separate stages for initial condition fitting and future performance, simplifying tuning while ensuring constraint satisfaction. Experimental work on quadcopter platforms has validated data-enabled predictive control schemes that use past trajectories to forecast future states without explicit identification, achieving reliable trajectory tracking in noisy, strongly nonlinear flight conditions.

Data-Driven Control of Nonlinear Dynamic Systems publication trend

The graph below shows the total number of articles in data-driven control of nonlinear dynamic systems across all publications each year (not limited to Nature Index journals).

Technical terms

Data-driven control: design of feedback mechanisms directly from input–output data without explicit modelling.

Nonlinear dynamic system: a system whose evolution is governed by equations that are nonlinear in their state or inputs.

Persistence of excitation: a property of input signals that ensures collected data sufficiently captures system dynamics.

Predictive control: a control strategy that optimises future control actions based on a model or data-driven representation.

Nonlinearity cancellation: a method to mitigate system nonlinearities by designing controllers that approximate their inverse behaviour.

References

1. Data-driven control of complex networks. Nature Communications (2021).
2. Learning Controllers From Data via Approximate Nonlinearity Cancellation. IEEE Transactions on Automatic Control (2023).
3. Data-driven predictive control in a stochastic setting: a unified framework. Automatica (2023).
4. Data‐enabled predictive control for quadcopters. International Journal of Robust and Nonlinear Control (2021).