Adaptive Control for Nuclear Power Plant Load Changes

Abstract

The problem being considered is the control of power load changes in a nuclear power plant. In particular, the problem is that of making large and fast load changes without causing serious disturbances to the plant operation. In order to make such load changes, a control strategy is designed to coordinate the major plant inputs. However, to be viable the control must overcome several kinds of variation and uncertainty in plant behavior. Some variations are due to slow physical changes in the plant. Other variations in behavior are due to unmeasured current plant conditions and to random disturbances. These variations, together with the inadequacy of simple models to describe the plant, lead to uncertainty in the plant behavior. This paper presents means for overcoming the problems of variation and uncertainty, so that large and fast power load changes can be made.The control developed in this study makes use of a stochastic model. A low order state variable form stochastic model is used as a reduced order representation of the plant. The parameters of this model are identified from measurement data records, using a maximum likelihood identification technique. The stochastic model is then used in a state estimator which gives estimates of important unmeasured plant variables. The stochastic model is also used to adjust control constraints and to predict the plant performance during the load change for display and monitoring purposes. In the event that the plant behavior deviates from the model predictions, the model is adapted to be consistent with the plant performance. These techniques together overcome the difficulties of variation and uncertainty in the nuclear plant behavior, thus permitting large, fast load changes to be made without upsetting the plant operationsA nuclear power plant was not available for experimentation; therefore, a detailed and realistic high order simulation of a plant was used. The simulated plant included nonlinear effects, disturbances and measurement noise. The simulation was run on an analog computer which was treated as an actual plant by the digital control computer. The experiments performed in this study clearly demonstrate the feasibility of control of large and fast load changes on a nuclear power plant, and these can be made much more smoothly than the conventional control of small load changes. The experiments also demonstrate the feasibility of detecting and adapting to plant variation, and the feasibility of control in the presence of uncertainty.