Research on the prediction and policy optimization model of Anhui merchants' return investment based on decision transformer

The precise prediction of investment decisions by returning Huizhou merchants in Anhui Province and the optimization of related policies are critical issues in current regional governance, given their role as the core driving force behind high-quality regional economic development. The present study proposes a novel model based on decision transformers and multimodal data fusion to predict and optimise the investment intentions of new Huizhou merchants returning to Anhui Province. Conventional models, such as LSTM, encounter challenges in accounting for nonlinear correlations and dynamic policy responses, resulting in substantial prediction errors and suboptimal resource allocation. The model under discussion successfully circumvents the limitations of traditional methods by means of the integration of spatiotemporal attention mechanisms. These mechanisms are employed for the purpose of analysing technical indicators, policy texts and social capital. The enhanced NSGA-II algorithm, when employed in conjunction with the q-power gradient update strategy, facilitates the optimisation of dynamic policies, thereby ensuring a balanced consideration of fiscal constraints, technology spillover effects, and investment risks. Empirical evidence has demonstrated that prediction errors in high-tech industries are reduced by 46%, policy response delays are decreased by 62%, and technology conversion rates reach 74.6%. This study validates the effectiveness of the "technology-policy-capital" collaborative governance framework, providing a robust solution for regional economic development and policy formulation under the Yangtze River Delta integration strategy. Future research will focus on model lightweighting, cross-regional policy simulation, and the optimization of real-time risk response mechanisms.