Hybrid System Identification and Control Techniques

Hybrid system identification and control encompass methods for modelling and regulating systems whose behaviour arises from the interaction of continuous dynamics and discrete events. Such systems are prevalent in robotics, automotive powertrains, smart grids and biological processes, where physical subsystems switch between modes or obey piecewise-defined laws. Identification techniques seek to recover both the continuous subsystem parameters and the discrete switching logic from observed data, often under uncertainty and noise. Control strategies then exploit these models to ensure stability, performance and safety, combining tools from optimal control, model predictive control and formal verification. Recent advances have emphasised data-driven approaches that balance fidelity and computational tractability, enabling real-time implementation in safety-critical environments. Key challenges include segmentation of data into operating regimes, robust parameter estimation under bounded disturbances, and the design of controllers that can cope with mode transitions without resorting to intractable combinatorial optimisation.

Research from Nature Portfolio

Recent studies have introduced a general framework for the discovery of cyber-physical systems directly from measured data. This approach simultaneously identifies the continuous dynamics of each mode and infers transition logic governing mode switches, without prior knowledge of subsystem boundaries. By formulating the estimation as an optimisation problem over both real-valued and discrete variables, the method delivers interpretable hybrid models that capture complex interactions between software and hardware components. Applications span smart grids, autonomous vehicles and manufacturing systems, where the resulting models support accurate state-trajectory prediction, performance assessment and guided redesign to meet stringent reliability requirements.

Research from all publishers

Bayesian methods have been applied to the identification of piecewise-linear mechanical oscillators, treating the number of operating regions as a model-selection problem and estimating subsystem parameters via a likelihood-free approximate Bayesian computation scheme. This strategy yields not only point estimates but also uncertainty bounds, and has been validated through numerical and experimental studies of mechanical test rigs. In the field of environmental engineering, switched Box–Jenkins models have been adapted to approximate nonlinear biological wastewater-treatment processes. An online two-stage algorithm, based on outer bounding ellipsoids, identifies both switching patterns and submodel parameters under bounded disturbances, enabling real-time reconciliation of sensor data and the design of model-based controllers. A third line of work introduces randomized algorithms for the identification of switched NARX systems, where mode-switch locations are inferred by iteratively updating a probability distribution over candidate switching sequences. The resulting method scales favourably with data size and improves segmentation accuracy on benchmark examples.

Hybrid System Identification and Control Techniques publication trend

The graph below shows the total number of articles in hybrid system identification and control techniques across all publications each year (not limited to Nature Index journals).

Technical terms

Hybrid system: A system exhibiting both continuous-time dynamics and discrete-event transitions between modes.

Piecewise-affine (PWA) model: A representation that partitions the state-space into regions, within each of which the system evolves according to an affine law.

Switched system: A subclass of hybrid systems consisting of several continuous subsystems and a logic determining active modes.

Approximate Bayesian Computation (ABC): A sampling-based inference technique that bypasses explicit likelihood evaluation by matching simulated and observed data.

Box–Jenkins model: A time-series representation combining autoregressive and moving-average components, extended here to switched or hybrid settings.

References

1. Data driven discovery of cyber physical systems. Nature Communications (2019).
2. Identification of piecewise-linear mechanical oscillators via Bayesian model selection and parameter estimation. Mechanical Systems and Signal Processing (2023).
3. An identification algorithm of switched Box-Jenkins systems in the presence of bounded disturbances: An approach for approximating complex biological wastewater treatment models. Journal of Water Process Engineering (2024).
4. A randomized method for the identification of switched NARX systems. Nonlinear Analysis Hybrid Systems (2023).