Neural estimation of first-order sensitivity coefficients: Application to the control of a simulated pressurized water reactor

In real, complex plants, a sensitivity analysis of the effects that variati ons in the plans inputs and design parameters have on the outputs is of gre at importance both from the point of view of productivity and of safety. To a first approximation, sensitivity analysis consists of estimating the par tial derivatives of the outputs with respect to the varied quantities. Thes e derivatives cannot be obtained on the real plant directly since the effec ts of all the involved variables are intermixed. Therefore, one has to reso rt to suitable computational models and algorithms. A new neural network approach that aims at creating a differentiable copy o f the plant is proposed. A feature of the method is that the data for netwo rk training are collected with the system in nominal operation: This repres ents, indeed, a fundamental constraint for all risky plants, for which unre strained playing is definitely nor recommended. The sensitivity coefficient s (partial derivatives) thereby obtained are applied for the regulation of the reactivity of a simulated pressurized water reactor in response to chan ges in the electric load at the power grid, so as to maintain the average t emperature of the water in the reactor core at a constant value.