We present a unified multi-representation modeling framework for neural active reconstruction that overcomes the limitations of single-feature approaches in complex environments. Our system innovatively combines occupancy fields, signed distance functions, and neural radiance fields within a hierarchical framework to achieve high-fidelity 3D reconstruction. The proposed method features: (1) a hybrid representation architecture that dynamically integrates discrete and continuous features, and (2) an active perception mechanism that optimizes viewpoint selection based on representation uncertainty. Comprehensive evaluations on Replica and MP3D datasets demonstrate significant improvements over existing methods. This work establishes a new paradigm for active neural reconstruction by bridging traditional SLAM with modern neural implicit representations.