Unit specific level 1 probabilistic safety assessment (PSA) models and results are available for the four VVER-440/213 type units of the Paks nuclear power plant, Hungary. The Paks PSA is updated annually within the framework of a living PSA program. The major purpose of the updates is to obtain a quantitative expression of safety impacts from plant modifications. Numerous modifications have been made in the past 15 years including design changes, procedural improvements, administrative measures, etc. Replacement of the earlier emergency operating procedures (EOPs) with new, symptom-oriented procedures is a recent modification. It is a complex task to determine the risk implications of this procedural change because the operators had to adopt a new philosophy (approach) to taking emergency responses and the operation of the control room crew as a team had to significantly change too. Efforts are made, over and above the annual PSA updates, to model and quantify the impact of EOP improvement. These efforts include review and modification of the accident sequence models in accordance with the new procedural requirements for emergency operations, re-definition of human failure events as post-initiator errors in the PSA model, quantification of operator reliability for the re-defined failure events, and re-quantification of accident sequences:

• The accident sequence models have been reviewed for each initiating event of the level 1 PSA. During the review the operator responses were determined on the basis of the symptom-oriented EOPs.

• Re-definition of human failure events was an embedded task within the update of the accident sequence models. The definition of operator errors was based on an initial understanding of the likely human failures that might occur when using the new procedures.

• Much attention is paid to the quantification of operator errors. In the original PSA for Paks data and insights had been used from about 200 PSA driven observations at the Paks NPP training simulator. New simulator observations were made after EOP improvement to help understand the strengths and weaknesses of the control room crews and to collect information that could possibly be useful for quantitative human reliability analysis. These observations covered 4 emergency scenarios and each of the 24 control room crews at Paks. A pre-defined taxonomy was used to collect data by an observer team of 6. This was supplemented by automatic data collection to help reproduce scenario details. Simulator insights were used in combination with expert opinion to update the earlier, decision tree based quantification method.

• The results of updated human reliability analysis (HRA) will be incorporated into the PSA model for the purpose of re-quantifying accident sequences. These results will include the final definitions of post-initiator human errors and the associated human error probabilities. PSA re-quantification will cover full power PSA with internal events, fires and flooding as well as low power and shutdown PSA with internal events.

This paper presents the analysis steps of the Paks PSA update due to EOP changes. Also, it discusses the analysis methods, data and the most important results. Novelties in using data and insights from simulator observations are highlighted.