The critical power ratio (CPR) is a safety parameter indicating the margin between the operative and the critical conditions of the coolant for each BWR fuel assembly. The CPR is normally obtained using CPR correlations derived from numerous full-scale experiments for the real FAs. As an alternative, CPR may be estimated by means of a subchannel code by increasing the fuel assembly power amplitude until the critical condition, i.e. minimum critical heat flux ratio equal 1.0, is reached. This method was applied for Transient Reload licensing Analysis (TRA) of the Swiss BWRs in the framework of the STARS (Steady-state and Transient Analysis Research for the Swiss reactors) program. The developed methodology demonstrated a reasonable accuracy in comparison to the vendor data. However, this methodology is quite computationally expensive since the critical conditions are searched by iterative method, i.e. for transient conditions, first a rough estimation of CPR value is done using eleven sub-channel simulations. After that the refined in power and time set of eleven subchannel simulations is done to estimate the final CPR value. In order to accelerate the CPR search procedure, the machine learning (ML) method was applied to predict the CPR value using only FA boundary conditions, power, transient type and moment of the fuel cycle. It was demonstrated that the developed ML-based meta-model is able to predict the CPR with good accuracy. More accurate initial guess of CPR values could help in reducing the required number of subchannel simulations by factor 2-3. In addition, the advanced iteration scheme was developed to further accelerate the CPR search method. The iteration scheme utilizes the information of the calculated CHF from the previous iteration to predict the critical conditions and hence the CPR value used as a guess for the next subchannel simulation run. The total acceleration of CPR search procedure of factor 6 was achieved. The developed methodology was demonstrated on six subsequent BWR cycles. The ML-based meta-model was permanently trained as soon as the new simulation results were available. The results of multi cycles TRA simulations were compared to the vendor results for OLMCPR (operating limit minimum CPR) for yearly reload licensing submittals. The trends for OLMCPR prediction versus the vendor data are discussed.