The identification of enterprise competitive strategy and its mapping to economic growth mode provide support for intelligent decision-making and business analysis. This paper proposes a strategic analysis framework combining gated recurrent unit, graph interactive aggregation and multi-head attention to process heterogeneous enterprise data including financial indicators, competitive signals, text feedback and regional economic variables. A sliding window scheme is used to construct 13260 sets of time samples from 246 enterprises, and the time covers the first quarter of 2016 to the fourth quarter of 2024. The framework uses a dual-output structure to complete the strategic state identification and growth-driven prediction respectively, and divides the strategic state into expansion, coordination and contraction. The cross-feature attention aggregation and dynamic deduction mechanism are introduced to enhance the ability of nonlinear pattern capture and phase response. Experimental results show that the R2 of growth prediction is 0.874, the accuracy is 89.1%, and the Macro-F1 value is 0.852. Compared with RF, XGBoost and MLP baselines, the proposed framework has strong fitting stability, migration consistency and adaptation ability in enterprise scenarios.
