Abstract: Objectives: Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) integration is widely used to provide continuous and high-precision positioning in complex urban environments. Its positioning performance depends strongly on the correct fixing of carrier-phase ambiguities. However, GNSS signals in urban canyons are frequently affected by obstruction, multipath, and non-line-of-sight propagation, resulting in degraded observation quality and reduced ambiguity-fixing reliability. Conventional ambiguity-fixing methods generally employ fixed elevation masks, signal-to-noise ratio (SNR) thresholds, observation weighting parameters, and INS uncertainty settings, and therefore have limited adaptability to rapidly changing observation environments. To address this limitation, this study proposes an adaptive optimization method for key ambiguity-fixing parameters based on Soft Actor-Critic (SAC) and a Bidirectional Gated Recurrent Unit (BiGRU), with the aim of improving ambiguity-fixing performance and solution reliability in complex urban scenarios. Methods: The adjustment of key ambiguity-fixing parameters is formulated as a Markov decision process. A continuous multi-epoch state sequence is constructed using online information on GNSS observation quality, navigation solution status, and ambiguity validation. A BiGRU network is employed to extract temporal features and dynamic variation patterns from the state sequence. The extracted features are then provided to the SAC agent, which learns a continuous parameter-adjustment policy through actor- critic optimization. According to the current observation conditions and solution status, the proposed method jointly adjusts the satellite elevation mask, SNR threshold, observation noise covariance, and INS attitude uncertainty. These parameters affect satellite selection, observation weighting, float ambiguity estimation, and ambiguity validation. Their coordinated optimization enables the ambiguity-fixing process to adapt dynamically to changes in satellite visibility and measurement quality. Results: Experiments were conducted in both open-sky and complex urban environments. The results show that the proposed method maintains positioning accuracy while improving ambiguity-fixing performance under open-sky conditions, indicating that adaptive parameter adjustment does not adversely affect the navigation solution under favorable observation conditions. In complex urban environments, compared with the conventional full ambiguity resolution scheme, the proposed method increases the ambiguity-fixing rate, correct-fix rate, and fixing success rate by 8.1, 9.1, and 3.3 percentage points, respectively, while reducing the false-fix rate by 1.0 percentage point. These results demonstrate that the proposed method simultaneously improves the availability of fixed solutions and the reliability of ambiguity fixing. Conclusions: The proposed SAC-BiGRU-based method combines the temporal feature extraction capability of BiGRU with the continuous decision-making capability of SAC, enabling state-dependent and coordinated adjustment of multiple ambiguity-fixing parameters. It addresses the limited adaptability of conventional methods based on fixed empirical parameters. The proposed method maintains stable positioning performance in open-sky scenarios and significantly improves ambiguity-fixing availability and reliability in complex urban environments. It therefore provides an effective approach for enhancing high-precision GNSS/INS integrated positioning under challenging observation conditions.