An artificial intelligence-guided decision support system for the nuclear power plant management

When an incident occurs, plant operators rely on emergency operating procedures or severe accident management guidelines to make decisions. However, the nuclear plant is a complex dynamic system (many components, various initiating events, scenarios, time-dependent variables, etc.). Therefore, the number of possible sets of failures and corresponding mitigative actions is too large to pre-plan for all of them. To deal with this complex system, an Artificial Intelligence (AI) guided decision support system is proposed. The proposed system generates a plan (an action sequence or a policy) making use of continuous monitoring of the plant status and the components health. Research in AI has led to an unconventional declarative programming approach that does not involve encoding algorithms. Instead, the problem is described, and the search for a solution is left to the computer. In declarative programming, the inputs are logical rules that describe the system configuration (e.g., whether a valve is open or closed) plus the collection of possible actions that change the configuration of its components (e.g. open/close a valve). Abnormal behavior can be observed if the system status is inconsistent with the predicted behavior. Faulty components can be identified (diagnosis) and the needed actions are autonomously computed (by searching over the actions space) to mitigate the faulty behavior. In this work, the proposed declarative programming approach is applied to find the needed actions in proof-of-concept examples: a turbine control valve (that controls steam flow to the turbine or to the condenser) drifts in the closed direction, and a feedwater control valve (that controls the feedwater flow rate to steam generators) fails to open (or fails to close).