Wargame Deduction Simulation with Complex State Segmentation

We presented a novel solution to the challenges of complex decision-making and convergence difficulties faced by agents in Wargame Deduction. To address the vast state space, the pro-posed State Segmentation Algorithm divided it into two parts and employs Attention Mechanism networks and Long Short-Term Memory Networks to process different types of states separately. Furthermore, a ternary neural network was introduced to limit weight parameters to {-1, 0, +1}, resulting in reduced computational complexity and faster processing speeds. Experimental results demonstrated that the optimized network model improves decision accuracy by 3.617%.