Skip to main content
Nuclear Science and Engineering Division | Artificial Intelligence and Machine Learning

Large Load Rejection Event

The management of plant dynamics in advanced reactors during operational occurrences such as large load rejection requires more than just simple P&ID control

The fast runback transient in a liquid-metal sodium reactor is an example of a control problem exhibiting strong non-linearities and having significant dynamic excitation with accompanying thermal stress issues. This transient begins with an unexpected disconnect from the grid followed by a runback of electric power to house load in preparation for a presumed re-connect. The main goal of the fast runback control strategy is to bring the system to the desired reduced power level as promptly as possible. Avoiding a reactor shutdown eliminates all the complex and time-consuming operations needed for a reactor startup from shutdown to full power conditions.


The control response is based on a feedwater flow rate leading strategy. After the disconnect and loss of external power demand, a feedwater flow rate reduction is commanded, and the feedwater control valves are appropriately adjusted. Then the turbine admission valve is promptly closed, reducing steam flow rate to the turbine. These control actions result in a decreased capability of the balance of plant to reject the thermal power produced in the primary circuit. To limit temperature, increase in the intermediate and the primary circuit, negative rod reactivity is inserted. Plant electrical power is reduced to around 5% of the rated value for powering house load.

This large multivariable constrained control problem was solved using Model-based Predictive Control (MPC). The control algorithm optimizes an objective or cost” function, which is a user-specified indicator of the desired performance of the feedback regulator. In the MPC approach, the control actions minimize a performance criterion over a prediction horizon, whose length is kept constant in time (receding horizon algorithm). The optimization problem is possibly subject to constraints on the manipulated inputs and outputs, whose future behavior is estimated according to the model of the plant. The model-based controller can predict if the proposed control maneuver will cause future violations of system constraints and will take necessary early control actions that avoid violations.

Petri nets representing state evolution of the grid (left) and the controlled dynamic behavior of the plant (right) during fast runback transient.

A Petri net represents the system desired behavior and from that finite-state machines (FSMs) describing the evolution of each involved controller were derived. The first control action to be carried out is the turbine mechanical power reduction. The turbine admission valve is closed to 20% steam mass flowrate in 10 s. At the same time the bypass valve controller, which regulates the SG pressure by adjusting the steam mass flow rate to be directed to the condenser, is switched on. This is followed by a reduction of the feedwater mass flow rate to match the electric power reduction to 5% of the nominal value.

The runback in reactor power can be achieved in about one minute. The small inherent positive reactivity associated with radial temperature profile collapse of metallic fuel requires minimal compensating control rod motion. There is a modest temperature increase in the primary circuit with a hot leg increase of 25 C.

Primary, intermediate circuit and feedwater normalized flow rates.
Reactivity contributions.
Sodium temperatures in the primary circuit.


The design of sophisticated control algorithms that avoid reactors trips and serve to limit the need for design features for bypassing energy and providing alternate heat rejection capability in the power conversion system for managing temperatures during large load imbalances.