Improving the Reliability of Decision-Support Systems for Nuclear Emergency Management by Leveraging Software Design Diversity

摘要

This paper introduces a novel method of continuous verification of simulation software used in decision-support systems for nuclear emergency management (DSNE). The proposed approach builds on methods from the field of software reliability engineering, such as N-Version Programming, Recovery Blocks, and Consensus Recovery Blocks. We introduce a new acceptance test for dispersion simulation results and a new voting scheme based on taxonomies of simulation results rather than individual simulation results. The acceptance test and the voter are used in a new scheme, which extends the Consensus Recovery Block method by a database of result taxonomies to support machine-learning. This enables the system to learn how to distinguish correct from incorrect results, with respect to the implemented numerical schemes. Considering that decision-support systems for nuclear emergency management are used in a safety-critical application context, the methods introduced in this paper help improve the reliability of the system and the trustworthiness of the simulation results used by emergency managers in the decision making process. The effectiveness of the approach has been assessed using the atmospheric dispersion forecasts of two test versions of the widely used RODOS DSNE system.