Sliding window factor graph optimization has become a key optimization framework for LiDAR-inertial simultaneous localization and mapping systems because it provides an effective balance between computational efficiency and estimation accuracy. However, existing sliding window strategies generally employ either a fixed window size or heuristic adjustment mechanisms, making them incapable of effectively adapting to dynamic changes in platform motion and environmental observation quality. Consequently, under aggressive motions or in geometrically degenerate environments, the observation distribution within the optimization window becomes imbalanced, resulting in degraded pose estimation accuracy and even optimization divergence. To address these challenges, this paper proposes an adaptive sliding window optimization method driven by the joint perception of platform motion and environmental observations. Specifically, the Chi-square statistic of inertial measurement unit pre-integration residuals is utilized to characterize the platform motion state in real time, enabling accurate detection of motion mutations such as rapid acceleration and sharp turns. Furthermore, an environmental observation quality evaluation model is established based on the Cram-r-Rao Lower Bound, where the information matrix is employed to quantitatively evaluate the pose observability and estimation accuracy of LiDAR point cloud registration. The motion-state assessment and environmental observation evaluation are further integrated to develop a joint perception-driven adaptive sliding window adjustment strategy. In addition, a multi-state marginalization mechanism is introduced to preserve historical information and maintain the consistency of the optimization model during dynamic window resizing. Real-world experimental results demonstrate that the proposed method reduces the root mean square error (RMSE) of localization by 57.5% compared with the conventional fixed sliding window strategy, effectively suppressing error accumulation and significantly improving the localization accuracy of the system.