Application of Multi-agent Cooperative Decision-making Model Based on LSTM and Attention Mechanism in Entrepreneurship Simulation

Artificial intelligence technology shows significant advantages in the group decision problem in the simulation software of entrepreneurship and management. In order to solve the challenges of fuzzy reliability assignment and environment non-stationarity in multi-agent cooperative decision making, a collaborative decision model based on multiagent reinforcement learning is proposed. The model innovatively integrates long and short term memory network, attention mechanism and asynchronous training strategy, and effectively solves the problem of reliability allocation and environmental non-stationarity in multi-agent systems. The results show that the algorithm stabilizes at a high reward value of 0.93 after 100 epochs, decision delay is reduced by $34.2 \%$, strategic coherence index and crisis response success rate are increased by $32.8 \%$ and $21.5 \%$, respectively, in 600 rounds of virtual business environment test. In terms of economic benefits, cross-departmental collaboration gains, net profit margin and resource utilization increased by $87.6 \%, 43.3 \%$ and $21.3 \%$, respectively. These results show that intelligent algorithms are efficient and robust in complex business decision scenarios, and provide a new technical path for the intelligent development of entrepreneurial education tools.