One of the main problems of Orthogonal Time Frequency Space (OTFS) modulation is its Peak-to-Average Power Ratio (PAPR), which can cause signal distortion and deteriorate the overall system bit error rate (BER). The conventional Selective Mapping (SLM) algorithm has a very high computation complexity due to too many Heisenberg transforms required for signal branches, a Selective Mapping method of the normalized grouping (NSLM) was proposed to reduce the PAPR of OTFS system. By the linear combination of a half of signal branches, more candidate signal sets were generated. At the same time, the power normalization kept the original peak value unchanged. The μ-NSLM was then proposed by combining the NSLM with the μ-LAW companding algorithm, which further reduces the PAPR of the system. Simulation results show that the μ-NSLM algorithm outperforms the conventional SLM methods in PAPR reduction. When the Complementary Cumulative Distribution Function (CCDF) is about 10 -5 , the branch number Q=16, compared with the original signal, the conventional SLM algorithm and the NSLM algorithm, the proposed the μ-NSLM algorithm reduced the PAPR by 4.83dB, 1.67dB and 0.96dB respectively, achieving a significant improvement in PAPR reduction.