This paper focuses on large-scale production release planning problems in wafer fabs. Due to the complexity and dynamic characteristic of production lines, it is useful to develop simulation models to evaluate different production plans and select the optimal one. However, detailed simulation requires large computational costs, especially for large-scale problems. Therefore, we provide a multi-fidelity optimisation with ordinal transformation and optimal sampling (MO2TOS) based on the multiple population evolutionary algorithm to accelerate computational efficiency and address large-scale release planning problems for wafer fabs. In the low-fidelity approximation process of the proposed approach, we employ open queuing theory to estimate cycle times. The experimental results confirm that the low-fidelity model shows a high consistency with the high-fidelity model and the proposed multi-fidelity optimisation method is effective at solving large-scale release planning problems.