What Is State Estimation?

State estimation is a technique in control theory and signal processing concerned with inferring the internal state of a dynamical system from partial, noisy, or indirect measurements. Because sensors never provide a complete or perfectly accurate picture of a system's true condition, state estimators reconstruct the most probable internal state by combining a mathematical model of the system with incoming observations. The technique is central to feedback control, since most modern controllers require knowledge of states that cannot be measured directly.

The field draws on probability theory, linear algebra, and the theory of dynamical systems. Its development accelerated during the late 1950s and early 1960s when Rudolf Kalman introduced a recursive, optimal estimator for linear systems with Gaussian noise, a result that became foundational across engineering disciplines.

Kalman Filtering

The Kalman filter is the most widely used state estimator for linear systems operating under additive Gaussian noise. It operates in two steps: a prediction step, in which the filter propagates the estimated state forward through the system model, and an update step, in which a new measurement is used to correct that prediction. The filter weights the correction by the ratio of model uncertainty to measurement uncertainty, placing more trust in whichever source is more reliable at each time step. Because the algorithm is recursive, it does not need to store all past measurements and runs efficiently in real time. Research on dynamic state estimation in power systems demonstrates how Kalman-based methods handle the delayed and noisy sensor data found in practical grid monitoring.

Nonlinear Estimation

Most physical systems of engineering interest are nonlinear, and the standard Kalman filter does not apply directly to them. Two widely used extensions address this limitation. The extended Kalman filter (EKF) linearizes the system equations around the current state estimate, applying the standard filter to the resulting approximation. The unscented Kalman filter (UKF) instead propagates a set of deterministically chosen sample points (sigma points) through the nonlinear equations, capturing higher-order statistical information without differentiation. The IntechOpen introduction to Kalman filter applications surveys the EKF, UKF, and particle filter variants alongside their typical application domains. Particle filters take a broader Monte Carlo approach, representing the probability distribution over states as a cloud of weighted samples, which makes them applicable to highly non-Gaussian problems. Each method trades off computational cost against estimation accuracy, and the appropriate choice depends on the degree of nonlinearity and the available processing budget.

Observability and Observer Design

A prerequisite for any state estimator is that the system be observable: the state must, in principle, be recoverable from the output measurements. Observability is a structural property of the system model and is checked by examining the rank of the observability matrix formed from the system matrices. For linear systems, the Luenberger observer provides a deterministic counterpart to the Kalman filter, placing the error dynamics at chosen stable eigenvalues through a gain matrix. The MIT OpenCourseWare notes on feedback control systems treat observability and observer design together as paired concepts, reflecting how closely they are linked in practice. When a system is not fully observable, reduced-order observers can estimate only the unobservable states using the available output information.

Applications

State estimation has applications in a wide range of disciplines, including:

• Electric power grid monitoring, where real-time state estimates enable operators to detect faults and optimize dispatch
• Inertial navigation and GPS fusion for aircraft, spacecraft, and autonomous ground vehicles
• Robotics, including simultaneous localization and mapping (SLAM) for mobile platforms
• Biomedical signal processing, where physiological states such as blood glucose concentration are estimated from indirect measurements
• Industrial process control, where unmeasured temperatures, concentrations, or flow rates must be inferred to close feedback loops